Independent scholarship · Open accessThe monthly digest

Full manuscript · browser edition

Differential Topology in Higher Categories: Mapping Objects, Derived Intersections, Spaces of Choices, and Cobordism

Continuous, single-column reading. Equations, tables, proofs, references and appendices stay in document order.

Contents · 144 sections

image

About this text. This monograph is a higher-categorical treatment of the territory mapped by Morris W. Hirsch’s Differential Topology (Graduate Texts in Mathematics 33, Springer, 1976): manifolds and maps, function spaces, transversality, vector bundles and tubular neighbourhoods, degrees and intersection numbers, Morse theory, cobordism, isotopy, and the classification of surfaces. It is an independent work. It reproduces none of that book’s text or proofs verbatim; it is not affiliated with or endorsed by its publisher. gives a concordance so that the two can be read side by side.

Status of results. Every numbered statement carries a tag: classical (standard differential topology, attributed), literature (established in the categorical, homotopical or derived-geometric literature, attributed), reformulation (a known result restated in a different language, with the restatement proved or sketched here), observation (an elementary remark we have not found stated in this form), or programme (a conjecture or a direction, not a result). No cited theorem is claimed as new.

Authorship and assistance. The text was drafted and typeset with a large language model (Anthropic’s Claude) under the direction of the submitting author, who is responsible for it. Generative AI assisted with drafting, literature recall, mathematical consistency checks, computation, and typesetting. The numerical examples and a number of finite structural claims are recomputed by the accompanying script verify-differential-topology.py ([app:computations]). Proofs and bibliographic details have not been independently refereed; corrections are welcome.

Open access. Rights remain with the author.

Preface

Hirsch’s Differential Topology [27] is a book about what can be done to a smooth map by moving it a little. Its method is uniform across nine chapters. One topologizes the set Cr(M,N)C^r(M,N) of maps; one shows that the maps with a desirable property (immersions, embeddings, maps transverse to a submanifold, Morse functions) form an open and dense, or at least residual, subset; one defines an invariant using a map in that subset; and one shows that the invariant does not depend on which map was used, because any two such maps are joined by a path that is itself generic one dimension up. Degree, intersection number, Euler characteristic, the cobordism class of a preimage, and the diffeomorphism type of a surface are all obtained in this way.

Each step of that method has a counterpart in the language of higher categories, and the counterparts are by now theorems in the literature and no longer analogies. The set of maps is the object of global points of a mapping object in a topos of smooth sets, whose homotopy type is its shape; the classification theorems of the subject (for vector bundles, for immersions, for manifolds up to cobordism) compute such shapes. Transversality is the condition under which a fibre product of manifolds exists and is correct; the category of manifolds has a universal completion in which every fibre product is correct, the ∞\infty-category of derived manifolds, and the genericity theorems say that every object of the completion can be deformed back into the original category. Independence of choices is the statement that a space of choices (of tubular neighbourhoods, collars, metrics, embeddings in R∞\mathbb{R}^\infty) is contractible, which is what makes a construction functorial. Invariants are traces. Decompositions by Morse functions are presentations by generators and relations of cobordism categories.

This book follows Hirsch’s order and states each of his main theorems in that language, with attribution. It is not a replacement for his book, and it does not reprove his analytic theorems. The inverse function theorem, partitions of unity, the Morse–Sard theorem, and the existence and uniqueness of solutions of ordinary differential equations are used as stated. What changes is the form of the conclusions, and the last chapter says plainly how much that change is worth. In brief: of the three volumes in this series so far, this is the one in which the higher-categorical statements carry the most real content, because the subject is where a good part of higher category theory came from; and it is also the one in which the gap between a formal statement and its proof is widest, because the proofs are analysis.

A companion volume, Parametrized Neural Networks and Their Representations in Differential Topology, applies the results of this book to smooth families of maps given by neural network architectures. It was written to be read after this one and cites it by chapter.

Introduction

What is assumed

The reader should know the definition of a smooth manifold and of a smooth map, the inverse function theorem, and the language of categories, functors, limits and adjunctions. Sheaves on a site, simplicial sets, and the idea of an ∞\infty-category as a category with spaces of morphisms are used from [ch:manifolds] on; [app:toolkit] collects the definitions and fixes the notation. Spectra appear in Tubular Neighbourhoods, Collars, and Spaces of Choices, Degree, Intersection Numbers, and the Euler Characteristic, Cobordism and are recalled where they enter. No result from stable homotopy theory beyond the statement of the Pontryagin–Thom theorem is needed to follow the text.

How the book is organized

The four parts follow the logical order of the source. Part I builds the ambient categories and treats the two genericity mechanisms, function spaces and transversality. Part II treats bundles and the invariants that come from them. Part III treats the three ways of taking a manifold apart or putting manifolds in families: Morse functions, cobordisms, isotopies; and ends with surfaces, where all three can be carried out completely. Part IV is an assessment. shows the dependencies.

Dependencies among the chapters. Chapters 1–4 build the categories and the genericity theorems; 5–7 treat bundles, choices and numerical invariants; 8–11 treat Morse functions, cobordism, isotopy and surfaces.

A dictionary

The table lists, for each of the recurring notions of the source, the categorical object that this book uses in its place, and the chapter in which the translation is made.

In the classical text Here Ch.
the set C∞(M,N)C^\infty(M,N) with a topology the mapping object [M,N][M,N] in smooth sets and its shape [ch:function]
the chain rule functoriality of TT and of Weil functors [ch:tangent]
a map transverse to a submanifold a fibre product that the embedding into derived manifolds preserves [ch:transversality]
“by a small perturbation we may assume” every derived manifold is cobordant to a manifold [ch:transversality], [ch:cobordism]
isomorphism classes of vector bundles homotopy classes of maps to a classifying space; concordance [ch:bundles]
“unique up to isotopy” a contractible space of choices [ch:tubular]
Euler characteristic, Lefschetz number trace of an endomorphism of a dualizable object [ch:degree]
passing a critical point a homotopy pushout along a sphere [ch:morse]
cobordism group π0\pi_0 of a cobordism category; a homotopy group of a Thom spectrum [ch:cobordism]
isotopy extension a restriction map between embedding spaces is a fibration [ch:isotopy]
classification of surfaces generators and relations for the two-dimensional cobordism category [ch:surfaces]

Conventions

Manifolds are smooth (C∞C^\infty), Hausdorff, second countable, and finite-dimensional; different components may have different dimensions unless a dimension is named. A closed manifold is compact without boundary. Hirsch works with CrC^r maps for 1≤r≤ω1\le r\le\omega and proves that nothing is lost by passing to C∞C^\infty ([thm:smoothing]); we use the finite-differentiability classes only where a theorem needs them. “Space” means ∞\infty-groupoid, that is, a homotopy type; the ∞\infty-category of spaces is written S\mathcal{S}. Citations of the form “Hirsch 3.2.7” mean Chapter 3, Section 2, item 7 of the source.

How to read the tags

A statement tagged classical or literature is someone else’s theorem, with a reference; a proof is given only when it is short or when its form matters for what follows. A statement tagged reformulation is a restatement, and the text says what it restates. Statements tagged observation are elementary; we have not found them written down in the form given and make no claim of depth. Statements tagged programme are not results.

Spaces of Smooth Things

Manifolds and Smooth Maps

Hirsch’s first chapter defines manifolds, maps, tangent bundles, embeddings and manifolds with boundary, and proves two theorems that the rest of his book uses constantly: the preimage of a regular value is a submanifold, and a compact manifold embeds in a Euclidean space. This chapter records the category those definitions produce, says exactly what is wrong with it as a category, and introduces the four enlargements that repair it. Each enlargement repairs a different defect, and it is worth seeing at the outset which is which.

The category of manifolds

Definition 1.1. Mfd\mathsf{Mfd} is the category whose objects are smooth manifolds (Hausdorff, second countable, finite-dimensional, without boundary) and whose morphisms are smooth maps. Cart⊂Mfd\mathsf{Cart}\subset\mathsf{Mfd} is the full subcategory on the objects Rn\mathbb{R}^n, n≥0n\ge0, and Open⊂Mfd\mathsf{Open}\subset\mathsf{Mfd} the full subcategory on open subsets of the Rn\mathbb{R}^n.

The source works throughout with a differentiability class 1≤r≤ω1\le r\le\omega. The following theorem is the reason one may fix r=∞r=\infty.

Theorem 1.2 (Smoothing of differential structures). [classical; Whitney; Hirsch 2.2.9, 2.2.10] Let 1≤r<∞1\le r<\infty. Every CrC^r manifold is CrC^r diffeomorphic to a C∞C^\infty manifold, and two C∞C^\infty manifolds that are CrC^r diffeomorphic are C∞C^\infty diffeomorphic.

So the groupoids of CrC^r manifolds and CrC^r diffeomorphisms have the same sets of isomorphism classes for all r≥1r\ge1. For r=0r=0 the statement fails in both directions: there are topological manifolds with no smooth structure and topological manifolds with several (Kervaire, Milnor). The theorem is proved by approximation in the strong topology ([ch:function]), and it is the first instance of the method described in the Preface.

What the category lacks

Mfd\mathsf{Mfd} has finite products and countable coproducts, and little else. Three fibre products show the possibilities. In each, the cospan is Rk→fR←0pt\mathbb{R}^k\xrightarrow{f}\mathbb{R}\xleftarrow{0}\mathrm{pt}, so the fibre product, if it exists, is “the zero set of ff”.

Example 1.3 (Three zero sets). (a) f(x,y)=x2+y2−1f(x,y)=x^2+y^2-1. The value 00 is regular, the zero set is the circle, and the circle is the fibre product in Mfd\mathsf{Mfd}.

(b) f(x)=x2f(x)=x^2. A cone over the cospan is a manifold SS with a smooth u ⁣:S→Ru\colon S\to\mathbb{R} such that u2=0u^2=0; then u=0u=0. So the fibre product exists in Mfd\mathsf{Mfd} and is a point. It is the wrong answer: the tangent functor does not preserve it (the fibre product of the tangent maps is {(x,v):x2=0, 2xv=0}\{(x,v):x^2=0,\ 2xv=0\}, a line), and it is unstable, since the zero set of x2−εx^2-\varepsilon is two points or none.

(c) f(x,y)=xyf(x,y)=xy. The fibre product does not exist in Mfd\mathsf{Mfd} (1.4).

Proposition 1.4. [classical; ] No manifold represents the functor S↦{u∈C∞(S,R2):u1u2=0}S\mapsto\{u\in C^\infty(S,\mathbb{R}^2): u_1u_2=0\} on Mfd\mathsf{Mfd}.

Proof. Suppose PP does, with universal map j ⁣:P→R2j\colon P\to\mathbb{R}^2. Evaluating on S=ptS=\mathrm{pt} shows that jj is a bijection onto the cross X={xy=0}X=\{xy=0\}. Let p0=j−1(0)p_0=j^{-1}(0) and let dd be the dimension of PP at p0p_0. The two axes are smooth curves in XX through 00, so they lift to smooth injective curves a,b ⁣:R→Pa,b\colon\mathbb{R}\to P through p0p_0 with a(R)∩b(R)={p0}a(\mathbb{R})\cap b(\mathbb{R})=\{p_0\}; hence d≥1d\ge1. If d=1d=1, an injective continuous map between 11-manifolds is open, so a(R)a(\mathbb{R}) and b(R)b(\mathbb{R}) are both neighbourhoods of p0p_0 and meet in more than a point. If d≥2d\ge2, then jj has rank at most 11 everywhere, since XX has empty interior, and by the constant rank theorem jj fails to be injective on an open dense set where its rank is locally constant. ◻

There are also no mapping objects: outside degenerate cases there is no manifold NMN^M with Mfd(S×M,N)≅Mfd(S,NM)\mathsf{Mfd}(S\times M,N)\cong\mathsf{Mfd}(S,N^M). And quotients by even very good equivalence relations leave the category.

Smooth sets and diffeological spaces

The first repair is to keep the test objects and enlarge the class of things tested.

Definition 1.5. A family {Ui→U}\{U_i\to U\} in Cart\mathsf{Cart} is a covering if it is an open cover all of whose finite non-empty intersections are diffeomorphic to Euclidean spaces. A smooth set is a sheaf of sets on Cart\mathsf{Cart} for this coverage; SmSet=Sh(Cart)\mathsf{SmSet}=\mathsf{Sh}(\mathsf{Cart}). For a smooth set XX, the elements of X(Rn)X(\mathbb{R}^n) are its plots of dimension nn. A smooth set is concrete, or a diffeological space, if each plot is determined by the function Rn→X(pt)\mathbb{R}^n\to X(\mathrm{pt}) it induces on points; Dflg⊂SmSet\mathsf{Dflg}\subset\mathsf{SmSet} is the full subcategory of these.

Theorem 1.6. [classical; Souriau, Iglesias-Zemmour [29]; Baez–Hoffnung [8]]

  1. The functor Mfd→SmSet\mathsf{Mfd}\to\mathsf{SmSet}, M↦C∞(−,M)M\mapsto C^\infty(-,M), is fully faithful and takes values in Dflg\mathsf{Dflg}.

  2. It preserves every limit that exists in Mfd\mathsf{Mfd}, and it sends an open cover of a manifold to a colimit diagram.

  3. SmSet\mathsf{SmSet} is a Grothendieck topos and Dflg\mathsf{Dflg} a quasitopos. Both are complete, cocomplete and locally cartesian closed. The mapping object has plots [X,Y](Rn)=Hom(Rn×X,Y)[X,Y](\mathbb{R}^n)=\mathop{\mathrm{Hom}}(\mathbb{R}^n\times X,Y).

Proof of (i). A natural transformation α ⁣:C∞(−,M)→C∞(−,N)\alpha\colon C^\infty(-,M)\to C^\infty(-,N) gives a function f=αpt ⁣:M→Nf=\alpha_{\mathrm{pt}}\colon M\to N, and naturality with respect to the inclusions of points shows that α\alpha is composition with ff. For every chart φ ⁣:Rm→M\varphi\colon\mathbb{R}^m\to M the composite fφ=α(φ)f\varphi=\alpha(\varphi) is smooth, so ff is smooth. ◻

By (ii), smooth sets inherit the fibre products of 1.3(a) and (b) unchanged, and in case (c) they supply one: the cross, with the plots being the smooth maps into R2\mathbb{R}^2 that land in it. By (iii) they supply the mapping object, and its plots are exactly what the source calls a smooth family of maps. is about this object. What smooth sets do not repair is (b). A sheaf on Cart\mathsf{Cart} is tested only by reduced objects, and the equation u2=0u^2=0 has the same solutions as u=0u=0 in every C∞(Rn)C^\infty(\mathbb{R}^n).

Function algebras

The second repair goes the other way and enlarges the class of test objects, by admitting algebras of functions that are not algebras of functions on a manifold.

Definition 1.7. A C∞C^\infty-ring is a functor A ⁣:Cart→SetA\colon\mathsf{Cart}\to\mathsf{Set} that preserves finite products. Its underlying set is A(R)A(\mathbb{R}); each smooth h ⁣:Rn→Rh\colon\mathbb{R}^n\to\mathbb{R} gives an nn-ary operation A(h) ⁣:A(R)n→A(R)A(h)\colon A(\mathbb{R})^n\to A(\mathbb{R}). The category of these is written C∞RingC^\infty\mathsf{Ring}. For a manifold MM, the C∞C^\infty-ring C∞(M)C^\infty(M) is Rn↦C∞(M,Rn)\mathbb{R}^n\mapsto C^\infty(M,\mathbb{R}^n).

Taking hh to be addition and multiplication shows that a C∞C^\infty-ring has an underlying commutative R\mathbb{R}-algebra. The free C∞C^\infty-ring on nn generators is C∞(Rn)C^\infty(\mathbb{R}^n), and quotients by ideals of the underlying algebra are again C∞C^\infty-rings.

Theorem 1.8. [literature; Moerdijk–Reyes [39], Ch. I, Thm. 2.8] The functor Mfd→C∞Ringop\mathsf{Mfd}\to C^\infty\mathsf{Ring}^{\mathrm{op}}, M↦C∞(M)M\mapsto C^\infty(M), is fully faithful, and it sends transverse fibre products (1.12) to pushouts: if f ⁣:M→Nf\colon M\to N and g ⁣:P→Ng\colon P\to N are transverse then C∞(M×NP)≅C∞(M)⊗C∞(N)∞C∞(P)C^\infty(M\times_NP)\cong C^\infty(M)\otimes^\infty_{C^\infty(N)}C^\infty(P), the pushout in C∞RingC^\infty\mathsf{Ring}.

In 1.3(b) the pushout is C∞(R)/(x2)C^\infty(\mathbb{R})/(x^2). By Hadamard’s lemma every smooth function is g(x)=g(0)+g′(0)x+x2h(x)g(x)=g(0)+g'(0)x+x^2h(x), so this ring is R[ε]/(ε2)\mathbb{R}[\varepsilon]/(\varepsilon^2), the dual numbers. Its “spectrum” has one point and a one-dimensional tangent space, which is what the tangent functor predicted. The ring has dimension 22, as does C∞(R)/(x2−ε)≅R×RC^\infty(\mathbb{R})/(x^2-\varepsilon)\cong\mathbb{R}\times\mathbb{R} for ε>0\varepsilon>0. So C∞C^\infty-rings repair (b) on one side. For ε<0\varepsilon<0 the function x2−εx^2-\varepsilon is invertible and the quotient is the zero ring; no algebra of real functions will make the length constant. What is constant for all ε\varepsilon is the count with signs, (+1)+(−1)=0(+1)+(-1)=0, and that is a statement about cobordism (Sard’s Theorem and Transversality, Cobordism).

The third and fourth repairs are homotopical. The third, the ∞\infty-topos of smooth ∞\infty-groupoids, is needed as soon as the objects being parametrized have automorphisms, as vector bundles and manifolds themselves do; it is introduced next. The fourth, derived manifolds, replaces the pushout of 1.8 by a homotopy pushout, and is the subject of [ch:transversality]. shows which fibre product is seen where; the fourth row is the simplest case in which the derived answer differs from all the others.

Four fibre products Rk×Rpt\mathbb{R}^k\times_{\mathbb{R}}\mathrm{pt} in four categories.
zero set of in Mfd\mathsf{Mfd} in SmSet\mathsf{SmSet} in C∞RingopC^\infty\mathsf{Ring}^{\mathrm{op}} in DMfd\mathsf{DMfd}
x2+y2−1x^2+y^2-1 circle circle circle circle
x2x^2 point point R[ε]/(ε2)\mathbb{R}[\varepsilon]/(\varepsilon^2) R[ε]/(ε2)\mathbb{R}[\varepsilon]/(\varepsilon^2)
xyxy none cross C∞(R2)/(xy)C^\infty(\mathbb{R}^2)/(xy) C∞(R2)/(xy)C^\infty(\mathbb{R}^2)/(xy)
0 ⁣:pt→R0\colon\mathrm{pt}\to\mathbb{R} point point point point of dimension −1-1

Smooth ∞\infty-groupoids and shape

Definition 1.9. H=Sh∞(Cart)\mathbf{H}=\mathsf{Sh}_\infty(\mathsf{Cart}) is the ∞\infty-topos of sheaves of spaces on Cart\mathsf{Cart}; its objects are smooth ∞\infty-groupoids, or smooth stacks. The constant-sheaf functor S→H\mathcal{S}\to\mathbf{H} has a left adjoint Π ⁣:H→S\Pi\colon\mathbf{H}\to\mathcal{S}, the shape. Two points of X(pt)X(\mathrm{pt}) are concordant if they are the restrictions to 00 and 11 of a point of X(R)X(\mathbb{R}).

Smooth sets are the 00-truncated objects of H\mathbf{H}. The left adjoint exists because the objects of the site are contractible; there is a formula for it.

Theorem 1.10. [literature; see [7], [45]] For X∈HX\in\mathbf{H}, the shape is the geometric realization of the simplicial space obtained by evaluating on the extended simplices Δek={t∈Rk+1:∑ti=1}≅Rk\Delta^k_e=\{t\in\mathbb{R}^{k+1}:\sum t_i=1\}\cong\mathbb{R}^k: ΠX ≃ colim[k]∈ΔopX(Δek).\Pi X\ \simeq\ \mathop{\mathrm{colim}}_{[k]\in\Delta^{\mathrm{op}}}X(\Delta^k_e). For a manifold MM, ΠM\Pi M is the homotopy type of the underlying space of MM, and π0ΠX\pi_0\Pi X is the set of concordance classes of points of XX whenever XX is a smooth set.

For a manifold, the simplicial set k↦C∞(Δek,M)k\mapsto C^\infty(\Delta^k_e,M) is the smooth singular complex, and the statement that it has the homotopy type of MM is the smooth approximation theorem (Hirsch 2.2.6) in its relative form. The functor Π\Pi is the device by which this book passes from smooth objects to homotopy types. Nearly every classification theorem in later chapters has the form: for a certain smooth ∞\infty-groupoid XX of geometric structures, ΠX\Pi X is such-and-such a space.

Regular values and the fibre products that exist

Definition 1.11. Smooth maps f ⁣:M→Nf\colon M\to N and g ⁣:P→Ng\colon P\to N are transverse, written f⋔gf\pitchfork g, if dfx(TxM)+dgy(TyP)=TzNdf_x(T_xM)+dg_y(T_yP)=T_zN whenever f(x)=g(y)=zf(x)=g(y)=z. If gg is the inclusion of a submanifold AA one writes f⋔Af\pitchfork A. A point z∈Nz\in N is a regular value of ff if f⋔{z}f\pitchfork\{z\}.

Theorem 1.12 (Transverse fibre products). [classical; Hirsch 1.3.2, 1.3.3] If f⋔gf\pitchfork g then M×NP={(x,y):f(x)=g(y)}M\times_NP=\{(x,y):f(x)=g(y)\} is a submanifold of M×PM\times P of dimension dim⁡M+dim⁡P−dim⁡N\dim M+\dim P-\dim N. It is the fibre product in Mfd\mathsf{Mfd}, and its tangent bundle is TM×TNTPTM\times_{TN}TP.

Proof. By 1.5, f⋔gf\pitchfork g if and only if f×g⋔ΔNf\times g\pitchfork\Delta_N, so it suffices to treat the preimage W=h−1(A)W=h^{-1}(A) of a submanifold A⊂NA\subset N of codimension kk under a map h⋔Ah\pitchfork A. Locally AA is the zero set of a submersion ϕ ⁣:V→Rk\phi\colon V\to\mathbb{R}^k, and transversality says that ϕh\phi h is a submersion at the points of WW. The implicit function theorem then gives charts in which WW is a coordinate subspace of codimension kk and TxW=ker⁡d(ϕh)x=dhx−1(Th(x)A)T_xW=\ker d(\phi h)_x=dh_x^{-1}(T_{h(x)}A). A smooth map into M×PM\times P with image in an embedded submanifold is smooth into the submanifold, which gives the universal property. ◻

The theorem identifies a class of cospans whose limits exist, are preserved by TT, and by 1.8 are preserved by passage to function algebras. shows that this class is large (every cospan is arbitrarily close to one in it) and that it determines a universal completion.

Example 1.13. Let f ⁣:Rn×n→Symnf\colon\mathbb{R}^{n\times n}\to\mathsf{Sym}_n, f(A)=ATAf(A)=A^{\mathsf T}A. At A∈O(n)A\in\mathrm{O}(n) the differential H↦ATH+HTAH\mapsto A^{\mathsf T}H+H^{\mathsf T}A is onto: H=12ASH=\tfrac12AS maps to SS. So O(n)=f−1(I)\mathrm{O}(n)=f^{-1}(I) is a manifold of dimension n2−12n(n+1)=12n(n−1)n^2-\tfrac12n(n+1)=\tfrac12n(n-1). The script checks the rank numerically for n≤4n\le4.

Embeddings and idempotents

Theorem 1.14 (Whitney). [classical; Hirsch 1.3.5, 2.2.14] Every manifold of dimension nn admits an embedding into R2n+1\mathbb{R}^{2n+1} with closed image.

Combined with the tubular neighbourhood theorem ([thm:tubular]), this says that every manifold is a smooth retract of an open subset of a Euclidean space. The converse also holds, and together they give a description of Mfd\mathsf{Mfd} in which charts and atlases do not appear.

Theorem 1.15. [classical; attributed to Lawvere] Mfd\mathsf{Mfd} is the idempotent completion of Open\mathsf{Open}: every idempotent in Mfd\mathsf{Mfd} splits, and every manifold is a retract of an object of Open\mathsf{Open}.

Proof. The second statement was just explained. For the first, let e ⁣:M→Me\colon M\to M be smooth with e2=ee^2=e; we show that e(M)=Fix(e)e(M)=\mathop{\mathrm{Fix}}(e) is a submanifold (1.2 asks for the details). At a fixed point xx the differential dexde_x is an idempotent matrix, so its rank equals its trace, which is continuous and integer-valued along Fix(e)\mathop{\mathrm{Fix}}(e); let it be kk near xx. Rank is lower semicontinuous, so rkdey≥k\mathop{\mathrm{rk}}de_y\ge k for yy near xx, and dee(y)dey=deyde_{e(y)}de_y=de_y gives rkdey≤rkdee(y)=k\mathop{\mathrm{rk}}de_y\le\mathop{\mathrm{rk}}de_{e(y)}=k. So ee has constant rank kk near xx, and its image near xx is a kk-dimensional submanifold. ◻

The theorem explains two features of what follows. A product-preserving functor on Cart\mathsf{Cart}, such as a C∞C^\infty-ring, has a canonical value on every manifold, because the manifold is a retract of an open subset of some Rq\mathbb{R}^q. And the universal completion of [thm:CS] has to be required to be idempotent complete.

Boundaries

Hirsch allows manifolds with boundary from the start, since cobordisms need them. A manifold with boundary is modelled on the half-space R+n={xn≥0}\mathbb{R}^n_+=\{x_n\ge0\}; its boundary ∂M\partial M is a manifold of dimension n−1n-1. The category Mfd∂\mathsf{Mfd}^\partial is less well behaved than Mfd\mathsf{Mfd}: the product of two objects has corners, and ∂\partial is not a functor on all smooth maps. The notion that makes the regular value theorem work is the following.

Definition 1.16. A submanifold A⊂MA\subset M of a manifold with boundary is neat if ∂A=A∩∂M\partial A=A\cap\partial M and AA is not tangent to ∂M\partial M at any point of ∂A\partial A.

Theorem 1.17. [classical; Hirsch 1.4.2] Let f ⁣:M→Nf\colon M\to N be smooth, with ∂N=∅\partial N=\emptyset, and let zz be a regular value of both ff and f∣∂Mf|_{\partial M}. Then f−1(z)f^{-1}(z) is a neat submanifold of MM with boundary (f∣∂M)−1(z)(f|_{\partial M})^{-1}(z).

In smooth sets the half-line is the subobject [0,∞)⊂R[0,\infty)\subset\mathbb{R}, and a function on it is smooth in the sheaf sense exactly when it is the restriction of a smooth function on R\mathbb{R} [31, §24]. So manifolds with boundary embed fully faithfully in Dflg\mathsf{Dflg} as well. The boundary is recovered from the object, as the set of points at which it is not locally Euclidean, although not functorially.

Notes

Diffeological spaces are due to Souriau; the standard reference is [29], and the identification with concrete sheaves is in [8]. C∞C^\infty-rings come from Lawvere’s programme for synthetic differential geometry; see [39], [34] and [30]. The ∞\infty-topos H\mathbf{H} and its shape are developed in [45]; the formula of 1.10 is proved in [7] and in work of Pavlov and of Bunk. Manifolds with corners as a category are treated by Melrose and by Joyce.

Exercises

Exercise 1.1. Verify the second row of 1.1: the fibre product of x↦x2x\mapsto x^2 and 0 ⁣:pt→R0\colon\mathrm{pt}\to\mathbb{R} is a point in Mfd\mathsf{Mfd} and in SmSet\mathsf{SmSet}, and is corepresented by R[ε]/(ε2)\mathbb{R}[\varepsilon]/(\varepsilon^2) in C∞RingC^\infty\mathsf{Ring}. Show that a map of C∞C^\infty-rings C∞(M)→R[ε]/(ε2)C^\infty(M)\to\mathbb{R}[\varepsilon]/(\varepsilon^2) is a tangent vector to MM.

Exercise 1.2. Complete the proof of 1.15: if e ⁣:M→Me\colon M\to M is a smooth idempotent then Fix(e)\mathop{\mathrm{Fix}}(e) is a closed submanifold of MM and e ⁣:M→Fix(e)e\colon M\to\mathop{\mathrm{Fix}}(e) is smooth. Give an example in which Fix(e)\mathop{\mathrm{Fix}}(e) has components of different dimensions.

Exercise 1.3. Carry out 1.13 in detail, and show that TIO(n)T_I\mathrm{O}(n) is the space of skew-symmetric matrices.

Exercise 1.4. Let M,NM,N be manifolds. Show that the smooth set [M,N][M,N] is concrete, that its points are the smooth maps M→NM\to N, and that Hom(S,[M,N])≅Hom(S×M,N)\mathop{\mathrm{Hom}}(S,[M,N])\cong\mathop{\mathrm{Hom}}(S\times M,N) for every smooth set SS.

Exercise 1.5. Show that f ⁣:M→Nf\colon M\to N and g ⁣:P→Ng\colon P\to N are transverse if and only if f×g ⁣:M×P→N×Nf\times g\colon M\times P\to N\times N is transverse to the diagonal ΔN\Delta_N.

The Tangent Functor and Jets

The source introduces the tangent bundle in a few pages and the bundle of jets in a few more, and uses both as tools. Here they are treated as structure on the category. The chain rule is the statement that TT is a functor; the symmetry of second derivatives is the statement that a certain algebra is commutative; the higher chain rule is the statement that jets compose. All product-preserving functors of this kind are classified by finite-dimensional algebras, and the axioms those functors satisfy can be isolated and imposed on other categories.

The tangent functor

On Cart\mathsf{Cart} put T(Rn)=Rn×RnT(\mathbb{R}^n)=\mathbb{R}^n\times\mathbb{R}^n and Tf(x,v)=(f(x),Dfxv)Tf(x,v)=(f(x),Df_x v). The chain rule D(gf)x=Dgf(x)DfxD(g f)_x=Dg_{f(x)}Df_x is the statement T(gf)=Tg TfT(gf)=Tg\,Tf.

Proposition 1.1. [classical; ] TT extends uniquely to a functor T ⁣:Mfd→MfdT\colon\mathsf{Mfd}\to\mathsf{Mfd} with a natural transformation p ⁣:T⇒idp\colon T\Rightarrow\mathrm{id} such that TT sends open embeddings to open embeddings and T(U)=pM−1(U)T(U)=p_M^{-1}(U) for U⊂MU\subset M open. It preserves finite products and transverse fibre products, and each pM ⁣:TM→Mp_M\colon TM\to M is a vector bundle.

Proof. A manifold is the colimit in Mfd\mathsf{Mfd} of the Čech diagram of an atlas, and the diagram consists of open embeddings between open subsets of Rn\mathbb{R}^n, on which TT is defined by the formula. The colimit of the image diagram exists and is TMTM. Products are preserved by the formula; transverse fibre products by [thm:transpull]. ◻

Weil algebras

Definition 1.2. A Weil algebra is a finite-dimensional commutative R\mathbb{R}-algebra A=R⊕nA=\mathbb{R}\oplus\mathfrak n with n\mathfrak n a nilpotent ideal. It is a C∞C^\infty-ring in a unique way: for smooth hh and a=a0+na=a_0+n one puts h(a)=∑αDαh(a0) nα/α!h(a)=\sum_\alpha D^\alpha h(a_0)\,n^\alpha/\alpha!, a finite sum. The Weil functor of AA is TAM=C∞Ring(C∞(M),A),T_AM=C^\infty\mathsf{Ring}\bigl(C^\infty(M),A\bigr), the set of AA-points of MM, with the manifold structure for which TA(Rn)=AnT_A(\mathbb{R}^n)=A^n.

For A=RA=\mathbb{R} this is the identity functor (the R\mathbb{R}-points of MM are its points). For the dual numbers W=R[ε]/(ε2)W=\mathbb{R}[\varepsilon]/(\varepsilon^2) it is TT, by [exer:xsq]. In coordinates a map ff acts on AA-points by Taylor expansion in the nilpotent directions: f(x+vε)=f(x)+Dfx(v) εf(x+v\varepsilon)=f(x)+Df_x(v)\,\varepsilon.

Theorem 1.3 (Classification of product-preserving bundle functors). [literature; Eck, Kainz–Michor, Luciano; [32], §§35–36]

  1. For every Weil algebra AA, TAT_A is a functor Mfd→Mfd\mathsf{Mfd}\to\mathsf{Mfd} that preserves finite products, with a natural fibre bundle projection TA⇒idT_A\Rightarrow\mathrm{id}.

  2. Every product-preserving bundle functor on Mfd\mathsf{Mfd} is naturally isomorphic to TAT_A for a Weil algebra AA, unique up to isomorphism.

  3. Natural transformations TA⇒TBT_A\Rightarrow T_B correspond bijectively to algebra homomorphisms A→BA\to B.

  4. TA∘TB≅TA⊗BT_A\circ T_B\cong T_{A\otimes B}, and TA×idTB≅TA×RBT_A\times_{\mathrm{id}}T_B\cong T_{A\times_{\mathbb{R}}B}.

By (iii) and (iv), every natural operation on iterated tangent bundles is a homomorphism of small algebras. lists the ones that matter.

The structure maps of the tangent functor as homomorphisms of Weil algebras. Here W=R[ε]/(ε2)W=\mathbb{R}[\varepsilon]/(\varepsilon^2), W⊗W=R[ε,η]/(ε2,η2)W\otimes W=\mathbb{R}[\varepsilon,\eta]/(\varepsilon^2,\eta^2) and W×RW=R[ε1,ε2]/(εiεj)W\times_\mathbb{R}W=\mathbb{R}[\varepsilon_1,\varepsilon_2]/(\varepsilon_i\varepsilon_j).
natural transformation algebra homomorphism in coordinates
projection p ⁣:T⇒idp\colon T\Rightarrow\mathrm{id} W→RW\to\mathbb{R}, ε↦0\varepsilon\mapsto0 (x,v)↦x(x,v)\mapsto x
zero section 0 ⁣:id⇒T0\colon\mathrm{id}\Rightarrow T R→W\mathbb{R}\to W x↦(x,0)x\mapsto(x,0)
addition + ⁣:T×idT⇒T+\colon T\times_{\mathrm{id}}T\Rightarrow T W×RW→WW\times_\mathbb{R}W\to W, ε1,ε2↦ε\varepsilon_1,\varepsilon_2\mapsto\varepsilon (x,v1,v2)↦(x,v1+v2)(x,v_1,v_2)\mapsto(x,v_1+v_2)
vertical lift ℓ ⁣:T⇒TT\ell\colon T\Rightarrow TT W→W⊗WW\to W\otimes W, ε↦εη\varepsilon\mapsto\varepsilon\eta (x,v)↦(x,0,0,v)(x,v)\mapsto(x,0,0,v)
flip c ⁣:TT⇒TTc\colon TT\Rightarrow TT W⊗W→W⊗WW\otimes W\to W\otimes W, ε↔η\varepsilon\leftrightarrow\eta (x,v,w,a)↦(x,w,v,a)(x,v,w,a)\mapsto(x,w,v,a)

Proposition 1.4 (Symmetry of second derivatives). [reformulation; cf. [34], [32]] The naturality of the flip cc with respect to a map ff is the equality D2fx(v,w)=D2fx(w,v)D^2f_x(v,w)=D^2f_x(w,v). It holds because W⊗WW\otimes W is commutative.

Proof. Expanding in W⊗WW\otimes W, f(x+vε+wη+aεη)=f(x)+Dfx(v)ε+Dfx(w)η+(Dfx(a)+D2fx(v,w))εη,f(x+v\varepsilon+w\eta+a\varepsilon\eta)=f(x)+Df_x(v)\varepsilon+Df_x(w)\eta+ \bigl(Df_x(a)+D^2f_x(v,w)\bigr)\varepsilon\eta , which is the formula for TTfTTf. Exchanging ε\varepsilon and η\eta exchanges vv and ww on the left; on the right it leaves the coefficient of εη=ηε\varepsilon\eta=\eta\varepsilon in place. So TTf∘c=c∘TTfTTf\circ c=c\circ TTf if and only if D2fx(v,w)=D2fx(w,v)D^2f_x(v,w)=D^2f_x(w,v), and the exchange is an algebra automorphism, hence natural by 1.3(iii). ◻

The script evaluates polynomials in R[ε,η]/(ε2,η2)\mathbb{R}[\varepsilon,\eta]/(\varepsilon^2,\eta^2) and confirms the expansion and its symmetry, and checks the chain rule in WW against finite differences.

Tangent categories

The five maps of 1.1 and the equations among them can be abstracted.

Definition 1.5. A tangent category is a category X\mathcal X with an endofunctor TT and natural transformations p,0,+,ℓ,cp,0,+,\ell,c as in 1.1, such that the fibre powers of each pMp_M exist and are preserved by TT; (p,0,+)(p,0,+) makes each TMTM a commutative monoid over MM; ℓ\ell and cc are additive, c2=idc^2=\mathrm{id}, c ℓ=ℓc\,\ell=\ell, and the usual coherence equations hold; and the vertical lift is universal, in the sense that TM×MTM→TTMTM\times_MTM\to TTM, (a,b)↦T(+)(ℓ(a),0(b))(a,b)\mapsto T(+)\bigl(\ell(a),0(b)\bigr), is the equalizer of T(p)T(p) and 0 p p0\,p\,p [10, Def. 2.3].

The definition is Rosický’s [44], revived and developed by Cockett and Cruttwell. The last axiom says that a second-order tangent vector that is vertical is the same thing as a pair of first-order ones. It is what makes it possible to define the Lie bracket of vector fields, connections, and solutions of differential equations in any tangent category.

Theorem 1.6. [literature; Leung [35]] Let Weil1\mathsf{Weil}_1 be the monoidal category of Weil algebras generated from WW by ⊗\otimes and ×R\times_\mathbb{R}. A tangent structure on X\mathcal X is the same thing as a strong monoidal functor Weil1→End(X)\mathsf{Weil}_1\to\mathsf{End}(\mathcal X) that preserves the pullbacks expressing ×R\times_\mathbb{R} and the universality of ℓ\ell.

So 1.1 is the definition, read backwards. Examples of tangent categories include Mfd\mathsf{Mfd}; C∞RingopC^\infty\mathsf{Ring}^{\mathrm{op}} and the category of affine schemes, with TT corepresented by A↦A[ε]A\mapsto A[\varepsilon] and Kähler differentials; every model of synthetic differential geometry; and cartesian differential categories, which axiomatize Cart\mathsf{Cart} itself [10]. The definition has an ∞\infty-categorical form, in which the Goodwillie calculus of functors is an example: the tangent bundle of an ∞\infty-category is its fibrewise stabilization [5].

Infinitesimals as objects

In C∞RingopC^\infty\mathsf{Ring}^{\mathrm{op}} the algebra WW is an object, D=Spec⁡WD=\operatorname{Spec}W, and TAMT_AM is the set of maps Spec⁡A→M\operatorname{Spec}A\to M. One would like TMTM to be the mapping object MDM^D.

Theorem 1.7. [literature; Dubuc [17]; Kock [34]; Moerdijk–Reyes [39]] Let Z\mathcal Z be the Cahiers topos, the topos of sheaves on the category of objects Rn×Spec⁡A\mathbb{R}^n\times\operatorname{Spec}A with AA a Weil algebra. Then Mfd→Z\mathsf{Mfd}\to\mathcal Z is fully faithful and preserves transverse fibre products; RD≅R×R\mathbb{R}^D\cong\mathbb{R}\times\mathbb{R} (the Kock–Lawvere axiom); and (TAM)≅MSpec⁡A(T_AM)\cong M^{\operatorname{Spec}A} for every manifold MM and Weil algebra AA.

In Z\mathcal Z the tangent functor is an exponential, so it preserves all limits. The functor (−)D(-)^D also has a right adjoint there, which is a statement about how small DD is. None of this is available in SmSet\mathsf{SmSet}. The restriction of DD to Cart\mathsf{Cart} is the terminal object, because a nilpotent element of C∞(Rn)C^\infty(\mathbb{R}^n) is zero; a sheaf on Cart\mathsf{Cart} cannot tell DD from a point. This is the same defect as in [ex:three](b), and it is the reason that tangent functors on diffeological spaces are a subject with several inequivalent answers.

Jets

Definition 1.8. Two maps f,g ⁣:M→Nf,g\colon M\to N have the same rr-jet at xx if f(x)=g(x)f(x)=g(x) and their partial derivatives up to order rr agree at xx in some, hence any, charts. Jr(M,N)J^r(M,N) is the set of rr-jets of maps M→NM\to N at all points of MM; it is a manifold and a fibre bundle over M×NM\times N by source and target. A map ff has an rr-jet prolongation jrf ⁣:M→Jr(M,N)j^rf\colon M\to J^r(M,N), a section of the source projection.

The fibre of Jr(Rm,Rn)J^r(\mathbb{R}^m,\mathbb{R}^n) over (0,0)(0,0) is the space of polynomial maps Rm→Rn\mathbb{R}^m\to\mathbb{R}^n of degree at most rr without constant term, of dimension n((m+rr)−1)n\bigl(\binom{m+r}{r}-1\bigr). For r=1r=1, J1(M,N)=Hom(TM,TN)J^1(M,N)=\mathop{\mathrm{Hom}}(TM,TN) as a bundle over M×NM\times N.

Proposition 1.9 (Jets compose). [classical; ] For each rr there is a category Jr\mathcal J^r whose objects are pairs (M,x)(M,x) and whose morphisms (M,x)→(N,y)(M,x)\to(N,y) are the rr-jets at xx with target yy. Taking the rr-jet is a functor from germs of smooth maps to Jr\mathcal J^r, and truncation gives functors Jr→Jr−1\mathcal J^r\to\mathcal J^{r-1}. In coordinates, composition in Jr\mathcal J^r is composition of polynomials followed by truncation at degree rr. The automorphism group GmrG^r_m of (Rm,0)(\mathbb{R}^m,0) in Jr\mathcal J^r is a Lie group, Gm1=GLm(R)G^1_m=\mathop{\mathrm{GL}}_m(\mathbb{R}), and the kernel of Gmr→Gm1G^r_m\to G^1_m is contractible.

Proof. Functoriality is the higher chain rule: the derivatives of order at most rr of gfg f at xx are polynomials in those of ff at xx and of gg at f(x)f(x) (Faà di Bruno), so the rr-jet of gfgf depends only on the rr-jets, and associativity is inherited from composition of maps. The kernel consists of polynomial maps x+(terms of degree 2,…,r)x+(\text{terms of degree }2,\dots,r) under truncated composition; scaling the higher terms to zero contracts it. ◻

For one variable and r=2r=2 the composition rule is (a1,a2)∘(b1,b2)=(a1b1,  a1b2+a2b12)(a_1,a_2)\circ(b_1,b_2)=(a_1b_1,\;a_1b_2+a_2b_1^2) for jets a1t+12a2t2a_1t+\tfrac12a_2t^2, which is (gf)′′=g′f′′+g′′(f′)2(gf)''=g'f''+g''(f')^2. The script checks associativity of this and of the three-variable analogue. The last sentence of the proposition has a consequence that is used without comment throughout the subject: a structure defined by higher jets carries the same homotopical information as one defined by first jets, so the characteristic classes of a manifold are those of its tangent bundle.

Jets fit the framework of the previous sections: the functor N↦J0r(Rm,N)N\mapsto J^r_0(\mathbb{R}^m,N) of rr-jets at the origin is the Weil functor of the algebra R[x1,…,xm]/mr+1\mathbb{R}[x_1,\dots,x_m]/\mathfrak m^{r+1}.

Two derivatives of a composite

The tangent functor is covariant and pushes vectors forward. Its transpose pulls covectors back, and since T∗T^* is not a functor on Mfd\mathsf{Mfd} the transpose needs a different kind of category to live in.

Definition 1.10. Lens\mathsf{Lens} is the category whose objects are pairs (X,X′)(X,X') of Euclidean spaces and whose morphisms (X,X′)→(Y,Y′)(X,X')\to(Y,Y') are pairs (f,f♯)(f,f^\sharp) of smooth maps f ⁣:X→Yf\colon X\to Y and f♯ ⁣:X×Y′→X′f^\sharp\colon X\times Y'\to X', composed by (g,g♯)∘(f,f♯)=(gf, (x,z′)↦f♯(x,g♯(f(x),z′))).(g,g^\sharp)\circ(f,f^\sharp)=\bigl(gf,\ (x,z')\mapsto f^\sharp(x,g^\sharp(f(x),z'))\bigr). The reverse derivative of f ⁣:Rm→Rnf\colon\mathbb{R}^m\to\mathbb{R}^n is R[f] ⁣:Rm×Rn→RmR[f]\colon\mathbb{R}^m\times\mathbb{R}^n\to\mathbb{R}^m, R[f](x,w)=(Dfx)TwR[f](x,w)=(Df_x)^{\mathsf T}w.

Proposition 1.11 (Reverse chain rule). [literature; Fong–Spivak–Tuyéras [20]; Cockett et al. [11]] f↦(f,R[f])f\mapsto(f,R[f]) is a product-preserving functor Cart→Lens\mathsf{Cart}\to\mathsf{Lens}.

Proof. R[gf](x,w)=(Dgf(x)Dfx)Tw=(Dfx)T(Dgf(x))Tw=R[f](x,R[g](f(x),w))R[gf](x,w)=(Dg_{f(x)}Df_x)^{\mathsf T}w=(Df_x)^{\mathsf T}(Dg_{f(x)})^{\mathsf T}w=R[f](x,R[g](f(x),w)), which is the composition law of Lens\mathsf{Lens}. ◻

On manifolds the same structure is the correspondence T∗M←f∗T∗N→T∗NT^*M\leftarrow f^*T^*N\to T^*N. The forward functor computes a directional derivative of a composite at the cost of one evaluation; the reverse functor computes the gradient of a scalar-valued composite at the cost of one evaluation and one backward pass. In the companion volume the reverse functor is the backpropagation algorithm, and 1.11 is the reason it is compositional.

Notes

Weil introduced his algebras in 1953 as “points proches”. The classification theorem is due independently to Eck, to Kainz and Michor, and to Luciano; [32] is the standard account and treats natural operations systematically. Tangent categories are surveyed in [10]. For synthetic differential geometry see [34] and [39]; the Cahiers topos is Dubuc’s [17]. Jet bundles as a comonad on bundles, with differential operators as its coKleisli maps, are a further piece of structure that this book does not use.

Exercises

Exercise 1.1. Show that T(M×N)≅TM×TNT(M\times N)\cong TM\times TN naturally, and that if zz is a regular value of f ⁣:M→Nf\colon M\to N then T(f−1(z))=(Tf)−1(0z)T\bigl(f^{-1}(z)\bigr)=(Tf)^{-1}(0_z), where 0z0_z is the zero vector at zz.

Exercise 1.2. Let A=R[x]/(x3)A=\mathbb{R}[x]/(x^3) and B=W⊗WB=W\otimes W. Show that TAMT_AM is the bundle of 22-jets at 00 of curves in MM, of fibre dimension 2dim⁡M2\dim M, and that the homomorphism A→BA\to B, x↦ε+ηx\mapsto\varepsilon+\eta, induces an embedding TAM→TTMT_AM\to TTM whose image is the fixed set of the flip intersected with {T(p)=pT}\{T(p)=p_T\}.

Exercise 1.3. Derive the composition law for 22-jets of maps (Rm,0)→(Rn,0)→(Rk,0)(\mathbb{R}^m,0)\to(\mathbb{R}^n,0)\to(\mathbb{R}^k,0) in terms of the linear and quadratic parts, and verify associativity directly.

Exercise 1.4. Verify the last three rows of 1.1: compute the natural transformations induced by the stated algebra homomorphisms in the coordinates (x,v,w,a)(x,v,w,a) on TTRnTT\mathbb{R}^n.

Exercise 1.5. Show that composition in Lens\mathsf{Lens} is associative, and that the reverse derivative of a composite of kk maps Rn0→Rn1→⋯→Rnk\mathbb{R}^{n_0}\to\mathbb{R}^{n_1}\to\dots\to\mathbb{R}^{n_k} with nk=1n_k=1 can be evaluated with one forward pass storing the intermediate values and kk transposed Jacobian–vector products.

Function Spaces

The second chapter of the source puts two topologies on the set Cr(M,N)C^r(M,N), proves that the important classes of maps are open in the stronger one, and proves approximation theorems by convolution and partitions of unity. This is the analytic engine of the book. The mapping object [M,N][M,N] of [thm:embed] is a different kind of structure on the same set, and it is important to be clear about what each is for. The strong topology is for proving that properties are generic. The mapping object is for parametrizing, and its shape is the homotopy type of the space of maps. The two are compatible but not interchangeable.

Two topologies

Fix 0≤r<∞0\le r<\infty. For f∈Cr(M,N)f\in C^r(M,N), charts (φ,U)(\varphi,U) on MM and (ψ,V)(\psi,V) on NN, a compact K⊂UK\subset U with f(K)⊂Vf(K)\subset V, and ε>0\varepsilon>0, let Nr(f;φ,ψ,K,ε)\mathcal N^r(f;\varphi,\psi,K,\varepsilon) be the set of g∈Cr(M,N)g\in C^r(M,N) with g(K)⊂Vg(K)\subset V and ∥Dk(ψgφ−1)(x)−Dk(ψfφ−1)(x)∥<εfor x∈φ(K), 0≤k≤r.\bigl\|D^k(\psi g\varphi^{-1})(x)-D^k(\psi f\varphi^{-1})(x)\bigr\|<\varepsilon \qquad\text{for }x\in\varphi(K),\ 0\le k\le r.

Definition 1.1. The weak (or compact-open CrC^r) topology on Cr(M,N)C^r(M,N) is generated by the sets Nr(f;φ,ψ,K,ε)\mathcal N^r(f;\varphi,\psi,K,\varepsilon). The strong (or Whitney) topology has as a base the sets ⋂iNr(f;φi,ψi,Ki,εi)\bigcap_i\mathcal N^r(f;\varphi_i,\psi_i,K_i,\varepsilon_i) over locally finite families of charts (φi,Ui)(\varphi_i,U_i) of MM, with arbitrary εi>0\varepsilon_i>0. On C∞(M,N)C^\infty(M,N) each topology is the union of the ones induced from CrC^r, r<∞r<\infty. The spaces are written CWr(M,N)C^r_W(M,N) and CSr(M,N)C^r_S(M,N).

A weak neighbourhood controls a map and its derivatives on a compact set. A strong neighbourhood controls them everywhere, with a tolerance that may shrink to zero at infinity as fast as one likes. If MM is compact the two agree. If MM is not compact the strong topology is not metrizable and not even first countable, but it is the one in which the theorems below are true.

Theorem 1.2 (Open conditions). [classical; Hirsch 2.1.1–2.1.6] For r≥1r\ge1 the sets of immersions, submersions, embeddings, closed embeddings and diffeomorphisms are open in CSr(M,N)C^r_S(M,N). For r≥0r\ge0 the set of proper maps is open in CSr(M,N)C^r_S(M,N).

Example 1.3. Let β≥0\beta\ge0 be a bump function on R\mathbb{R} supported in [−1,1][-1,1], with maximum at 00 and max⁡β′=1\max\beta'=1, and put gn(x)=x−2β(x−n)g_n(x)=x-2\beta(x-n). Then gng_n agrees with the identity on [−n+1,n−1][-n+1,n-1], so gn→idg_n\to\mathrm{id} in CW∞(R,R)C^\infty_W(\mathbb{R},\mathbb{R}); but gn′=1−2β′(x−n)g_n'=1-2\beta'(x-n) takes the value −1-1, so no gng_n is a diffeomorphism or even an immersion. The set of diffeomorphisms of R\mathbb{R} is not weakly open. In the strong topology gng_n does not converge to the identity: the neighbourhood {g:∣g(x)−x∣<β(0)e−x2}\{g:|g(x)-x|<\beta(0)e^{-x^2}\} contains none of them. The script confirms both statements numerically.

Theorem 1.4 (Approximation). [classical; Hirsch 2.2.6] For 0≤r<s≤∞0\le r<s\le\infty, Cs(M,N)C^s(M,N) is dense in CSr(M,N)C^r_S(M,N). If ff is already CsC^s on a neighbourhood of a closed set AA, the approximating maps may be taken to agree with ff on AA.

Since diffeomorphisms are open (1.2) and smooth maps are dense, a CrC^r diffeomorphism can be approximated by a C∞C^\infty one once source and target carry compatible smooth structures; this is the main step in [thm:smoothing]. The same two theorems give the first comparison of homotopy types.

Corollary 1.5. [classical; Hirsch 5.1.5] Every continuous map between manifolds is homotopic to a smooth map, and two smooth maps that are continuously homotopic are smoothly homotopic.

The mapping object

By [thm:embed] the smooth set [M,N][M,N] has as plots the smooth maps F ⁣:Rn×M→NF\colon\mathbb{R}^n\times M\to N. A plot is a family s↦Fss\mapsto F_s of maps, smooth jointly in the parameter and the variable. Nothing is said about neighbourhoods. A topology can be recovered from the plots, as for any smooth set.

Definition 1.6. A set AA of points of a smooth set XX is DD-open if p−1(A)⊂Rnp^{-1}(A)\subset\mathbb{R}^n is open for every plot pp of XX. A subobject of XX is DD-open if its set of points is and it contains every plot whose points lie in it.

Theorem 1.7. [literature; Christensen–Sinnamon–Wu [15]] On C∞(M,N)C^\infty(M,N) the DD-topology of the mapping object [M,N][M,N] is finer than the weak topology and coarser than the strong topology. If MM is compact all three coincide.

The first inequality says that a smooth family is continuous into the weak topology, which is easy. The second says that a DD-open set is strongly open, which is not. For non-compact MM the second is not an equality: a plot R→[M,N]\mathbb{R}\to[M,N] is in particular a path, and paths in the strong topology are very restricted.

Proposition 1.8. [classical; ] Let MM be non-compact. If t↦ftt\mapsto f_t is a continuous path in CS0(M,N)C^0_S(M,N), there is a compact K⊂MK\subset M such that ft=f0f_t=f_0 outside KK for all t∈[0,1]t\in[0,1].

So the path components of CS∞(M,N)C^\infty_S(M,N) are far smaller than homotopy classes (1.2), and the strong topology does not compute the homotopy type of anything one wants. It is a device for stating that a set of maps is large.

The homotopy type of a space of maps

The homotopy type attached to the mapping object is its shape ([def:shape]). By [thm:shape] it is the realization of the simplicial set k↦C∞(Δek×M,N)k\mapsto C^\infty(\Delta^k_e\times M,N), whose vertices are smooth maps, whose edges are smooth homotopies, and so on.

Theorem 1.9 (Shape of mapping objects). [literature; Berwick-Evans–Boavida de Brito–Pavlov [7], Thm. 1.1] For a manifold MM and any F∈HF\in\mathbf{H} the natural map Π[M,F] ⟶ Map(ΠM,ΠF)\Pi[M,F]\ \longrightarrow\ \mathop{\mathrm{Map}}(\Pi M,\Pi F) is an equivalence of spaces. In particular Π[M,N]≃Map(M,N)\Pi[M,N]\simeq\mathop{\mathrm{Map}}(M,N), the space of continuous maps, for manifolds MM and NN.

For F=NF=N a manifold and MM compact this is classical, and it is a parametrized form of 1.5. In general it is a recent theorem, and it will be used in [ch:bundles] with FF a stack of bundles. On π0\pi_0 it says that concordance classes of sections of FF over MM are homotopy classes of maps from MM to the space ΠF\Pi F. The hypothesis that MM is a manifold cannot be dropped; what is used is that MM has partitions of unity and is locally contractible, which is where the analysis of this chapter enters.

Remark 1.10. Concordance is an equivalence relation on the points of any smooth set, but transitivity is not automatic: two plots R→X\mathbb{R}\to X that agree at a point need not glue to a plot. One reparametrizes so that each is constant near the ends (1.4). The same device makes the smooth singular complex a Kan complex.

Subobjects cut out by open conditions

Definition 1.11. Imm(M,N)\mathop{\mathrm{Imm}}(M,N), Emb(M,N)\mathop{\mathrm{Emb}}(M,N) and Diff(M)\mathop{\mathrm{Diff}}(M) are the subobjects of [M,N][M,N] (of [M,M][M,M]) whose plots are the families F ⁣:Rn×M→NF\colon\mathbb{R}^n\times M\to N such that every FsF_s is an immersion, an embedding, a diffeomorphism.

For non-compact MM this definition of Emb\mathop{\mathrm{Emb}} is the weakest reasonable one, and there are others (requiring (s,x)↦(s,Fs(x))(s,x)\mapsto(s,F_s(x)) to be an embedding, or requiring compact support). For compact MM they agree.

Proposition 1.12. [reformulation; cf. Hirsch 2.1.1–2.1.6] If MM is compact, Imm(M,N)\mathop{\mathrm{Imm}}(M,N), Emb(M,N)\mathop{\mathrm{Emb}}(M,N) and Diff(M)\mathop{\mathrm{Diff}}(M) are DD-open subobjects of the mapping object. That is: in a smooth family of maps from a compact manifold, the set of parameters at which the map is an immersion (an embedding, a diffeomorphism) is open.

Proof. A plot FF induces a continuous map Rn→CW∞(M,N)\mathbb{R}^n\to C^\infty_W(M,N), since FF and its derivatives in the MM directions are uniformly continuous on compact subsets of Rn×M\mathbb{R}^n\times M. For compact MM the weak and strong topologies agree, and the three sets are open by 1.2. ◻

The reformulation is modest, but it is the form in which the companion volume uses 1.2: the parameters of a network at which it is an embedding of a compact data manifold form an open set. It is also the form that generalizes. A DD-open subobject of a mapping object has the same local structure as the mapping object, and so inherits whatever infinite-dimensional manifold structure the latter has [31].

Genericity

Theorem 1.13 (Baire property). [classical; Hirsch 2.4.4] Every weakly closed subset of CSr(M,N)C^r_S(M,N) is a Baire space in the strong topology: a countable intersection of open dense subsets is dense.

A subset is residual if it contains a countable intersection of open dense sets. The theorems of the next chapter assert that certain sets are residual, and 1.13 is what turns such an assertion into one about density. The proof realizes Cr(M,N)C^r(M,N), by the prolongation jr ⁣:Cr(M,N)→C0(M,Jr(M,N))j^r\colon C^r(M,N)\to C^0(M,J^r(M,N)), as a weakly closed set of sections of the jet bundle; the strong CrC^r topology is by definition the one induced from the strong C0C^0 topology on sections, and the Baire property is proved there by a diagonal argument.

Remark 1.14 (What the topos does not see). The DD-topology is defined from the plots, so openness is a notion internal to smooth sets. Residuality is not. A residual set need not be DD-open, and a set meeting every plot in a residual set of parameters need not be residual. We know of no formulation of the Baire property of CS∞(M,N)C^\infty_S(M,N) in terms of the sheaf [M,N][M,N] alone. What the sheaf does see is the parametric form of genericity ([thm:parametric]): in a family rich enough to be a submersion onto its target, almost every member is generic. In applications that statement is usually the one wanted.

Jets and partial differential relations

The prolongation is a map of smooth sets jr ⁣:[M,N]→Γ(Jr(M,N))j^r\colon[M,N]\to\Gamma(J^r(M,N)) into the smooth set of sections of the jet bundle over MM. It is a monomorphism, and its image consists of the holonomic sections. A condition on maps that is defined by a subset R⊂Jr(M,N)\mathcal R\subset J^r(M,N), such as being an immersion (the 11-jet is injective) or a Morse function (the 11-jet meets the zero locus only where the 22-jet is non-degenerate), gives a pullback square image in which the right-hand side involves no differential equation: Γ(R)\Gamma(\mathcal R) is the object of formal solutions. If R\mathcal R is open, the left-hand subobject is DD-open when MM is compact, by the argument of 1.12. Whether the top map induces an equivalence of shapes is the question answered, for immersions, by the Smale–Hirsch theorem ([thm:SH]).

Notes

The weak and strong topologies and their properties are developed in Hirsch’s Chapter 2 and, for C∞C^\infty and jets, in Golubitsky–Guillemin and Mather. The convenient setting of [31] gives C∞(M,N)C^\infty(M,N) the structure of an infinite-dimensional manifold whose smooth curves are the smooth families with locally compact support in the parameter; 1.8 is the reason for that restriction. The comparison of the DD-topology with the classical ones is [15]. is called the smooth Oka principle by Sati and Schreiber; earlier special cases are due to Madsen and Weiss.

Exercises

Exercise 1.1. Verify the claims of 1.3. Show also that Diff(R)\mathop{\mathrm{Diff}}(\mathbb{R}) is not closed in CW∞(R,R)C^\infty_W(\mathbb{R},\mathbb{R}).

Exercise 1.2. Prove 1.8 for M=N=RM=N=\mathbb{R}. Deduce that the path component of ff in CS0(R,R)C^0_S(\mathbb{R},\mathbb{R}) is the set of maps that agree with ff outside a compact set.

Exercise 1.3. Show that a plot of [M,N][M,N] is continuous as a map into CW1(M,N)C^1_W(M,N), and give an example of a smooth family F ⁣:R×R→R2F\colon\mathbb{R}\times\mathbb{R}\to\mathbb{R}^2 of maps for which {s:Fs is an embedding}\{s:F_s\text{ is an embedding}\} is not open.

Exercise 1.4. Let XX be a smooth set. Show that concordance is an equivalence relation on X(pt)X(\mathrm{pt}).

Exercise 1.5. A function f ⁣:R→Rf\colon\mathbb{R}\to\mathbb{R} is Morse if f′′(x)≠0f''(x)\ne0 wherever f′(x)=0f'(x)=0. Show that the set of Morse functions is dense but not open in CW∞(R,R)C^\infty_W(\mathbb{R},\mathbb{R}).

Sard’s Theorem and Transversality

This is the chapter on which the source turns. Its results say that the transverse cospans of [thm:transpull] are dense among all cospans, and that the same holds in families. The proofs reduce, by one piece of linear algebra, to a single analytic theorem. The higher-categorical counterpart is a universal property: there is a smallest enlargement of Mfd\mathsf{Mfd} in which all fibre products exist and the transverse ones are unchanged, and the genericity theorems say that every object of the enlargement is cobordant to an object of Mfd\mathsf{Mfd}.

The Morse–Sard theorem

Theorem 1.1 (Morse–Sard). [classical; Morse, Sard, Brown; Hirsch 3.1.3] Let f ⁣:M→Nf\colon M\to N be of class CrC^r with r>max⁡{0,dim⁡M−dim⁡N}r>\max\{0,\dim M-\dim N\}. Then the set of critical values of ff has measure zero in NN, and the set of regular values is residual.

The proof is an induction on dim⁡M\dim M by slicing with Fubini’s theorem, with the differentiability hypothesis used to control the image of the set where many derivatives vanish. Nothing in this book adds to it. The theorem is the source of every statement of the form “for almost all” below.

Corollary 1.2. [classical; Hirsch 3.1.4] A compact manifold with non-empty boundary admits no smooth retraction onto its boundary.

Proof. Let ρ ⁣:M→∂M\rho\colon M\to\partial M be one and yy a regular value of ρ\rho; it is automatically a regular value of ρ∣∂M=id\rho|_{\partial M}=\mathrm{id}. By [thm:neat], ρ−1(y)\rho^{-1}(y) is a compact neat 11-dimensional submanifold with boundary ρ−1(y)∩∂M={y}\rho^{-1}(y)\cap\partial M=\{y\}. A compact 11-manifold has an even number of boundary points. ◻

The argument is the prototype for the whole subject: a count that would have to be odd is the boundary of something, hence even. Brouwer’s fixed point theorem follows (1.5).

The transversality theorems

Theorem 1.3 (Transversality). [classical; Thom; Hirsch 3.2.1] Let A⊂NA\subset N be a submanifold and 1≤r≤∞1\le r\le\infty. The set of f∈Cr(M,N)f\in C^r(M,N) with f⋔Af\pitchfork A is residual, hence dense, in CSr(M,N)C^r_S(M,N); it is open if AA is closed.

The proof uses a parametric statement, which is the form we shall use most, and the parametric statement rests on a lemma worth isolating.

Lemma 1.4 (Fibrewise transversality). [classical; ] Let F ⁣:V×M→NF\colon V\times M\to N be smooth with F⋔AF\pitchfork A, let W=F−1(A)W=F^{-1}(A) and let π ⁣:W→V\pi\colon W\to V be the projection. Then Fv=F(v,−)F_v=F(v,-) is transverse to AA if and only if vv is a regular value of π\pi.

Proof. Fix (v,x)∈W(v,x)\in W, z=F(v,x)z=F(v,x), and let L ⁣:TvV⊕TxM→Q=TzN/TzAL\colon T_vV\oplus T_xM\to Q=T_zN/T_zA be the differential of FF followed by the quotient. By hypothesis LL is onto, and T(v,x)W=ker⁡LT_{(v,x)}W=\ker L. Write LML_M for the restriction of LL to TxMT_xM; then Fv⋔AF_v\pitchfork A at xx means that LML_M is onto, and dπd\pi is the projection ker⁡L→TvV\ker L\to T_vV.

If LML_M is onto and a∈TvVa\in T_vV, choose bb with LM(b)=L(a,0)L_M(b)=L(a,0); then (a,−b)∈ker⁡L(a,-b)\in\ker L projects to aa. Conversely, if the projection is onto and q=L(a,b)∈Qq=L(a,b)\in Q, choose (a,b′)∈ker⁡L(a,b')\in\ker L; then q=L(a,b)−L(a,b′)=LM(b−b′)q=L(a,b)-L(a,b')=L_M(b-b'). ◻

Theorem 1.5 (Parametric transversality). [classical; Hirsch 3.2.7] Let F ⁣:V×M→NF\colon V\times M\to N be of class CrC^r with r>max⁡{0,dim⁡M+dim⁡A−dim⁡N}r>\max\{0,\dim M+\dim A-\dim N\} and F⋔AF\pitchfork A. Then the set of v∈Vv\in V with Fv⋔AF_v\pitchfork A is residual in VV, and its complement has measure zero.

Proof. Apply 1.1 to π ⁣:W→V\pi\colon W\to V, noting dim⁡W−dim⁡V=dim⁡M+dim⁡A−dim⁡N\dim W-\dim V=\dim M+\dim A-\dim N, and use 1.4. ◻

The lemma has a reading that will recur. The manifold WW is the fibre product of the whole family with AA, and it is always a manifold. The individual fibre products Fv−1(A)F_v^{-1}(A) are its fibres over VV. So a non-transverse member Fv0−1(A)F_{v_0}^{-1}(A) of the family is a singular fibre of a map between manifolds, and nearby regular fibres are its replacements. If FF is merely a plot of [M,N][M,N], with no transversality assumed, nothing can be said; the hypothesis F⋔AF\pitchfork A is what makes a family “rich enough” in the sense of [rem:nobaire]. It holds in particular when FF is a submersion, for instance F(v,x)=f(x)+vF(v,x)=f(x)+v with N=V=RnN=V=\mathbb{R}^n (1.1).

Theorem 1.6 (Jet transversality). [classical; Thom; Hirsch 3.2.8, 3.2.9] Let A⊂Jr(M,N)A\subset J^r(M,N) be a submanifold and r<s≤∞r<s\le\infty. The set of f∈Cs(M,N)f\in C^s(M,N) with jrf⋔Aj^rf\pitchfork A is residual in CSs(M,N)C^s_S(M,N), and open if AA is closed.

Dimension counts

A map transverse to a submanifold of codimension greater than dim⁡M\dim M misses it. So genericity theorems become existence theorems as soon as one can compute a codimension.

Proposition 1.7. [classical; ] The m×nm\times n real matrices of rank ρ\rho form a submanifold of Rm×n\mathbb{R}^{m\times n} of dimension ρ(m+n−ρ)\rho(m+n-\rho), that is, of codimension (m−ρ)(n−ρ)(m-\rho)(n-\rho).

Corollary 1.8. [classical; Whitney; Hirsch 2.2.12, 2.2.13, 6.1.2] Let dim⁡M=m\dim M=m, dim⁡N=n\dim N=n, with MM compact.

  1. If n≥2mn\ge2m, immersions are dense in C∞(M,N)C^\infty(M,N).

  2. If n≥2m+1n\ge2m+1, embeddings are dense in C∞(M,N)C^\infty(M,N).

  3. Morse functions are open and dense in C∞(M,R)C^\infty(M,\mathbb{R}).

Proof. (a) In J1(M,N)J^1(M,N) the jets of rank ρ<m\rho<m form submanifolds of codimension (m−ρ)(n−ρ)≥n−m+1>m(m-\rho)(n-\rho)\ge n-m+1>m. A map whose 11-jet prolongation is transverse to all of them misses them. (b) For an immersion ff, injectivity is the statement that f×ff\times f restricted to M×M∖ΔM\times M\setminus\Delta misses ΔN\Delta_N, of codimension n>2mn>2m; one applies a multijet form of 1.6. (c) A function is Morse when dfdf, as a section of T∗MT^*M, is transverse to the zero section, a closed submanifold of codimension mm. ◻

The script checks 1.7 by computing the rank of the differential of a local defining map for several (m,n,ρ)(m,n,\rho), and checks (b) in a small case by sampling linear projections.

Clean and transverse

Transversality is a sufficient condition for a fibre product in Mfd\mathsf{Mfd} to be correct. It is useful to separate the two things it provides.

Definition 1.9. Maps f ⁣:M→Nf\colon M\to N and g ⁣:P→Ng\colon P\to N intersect cleanly if Z=M×NPZ=M\times_NP is a submanifold of M×PM\times P and TZ=TM×TNTPTZ=TM\times_{TN}TP. The excess bundle is the vector bundle E=coker⁡(TM⊕TP→TN)E=\operatorname{coker}(TM\oplus TP\to TN) over ZZ, of rank e=dim⁡N−dim⁡M−dim⁡P+dim⁡Ze=\dim N-\dim M-\dim P+\dim Z.

Proposition 1.10. [classical; ] If ff and gg intersect cleanly then ZZ is the fibre product in Mfd\mathsf{Mfd} and TT preserves it. They are transverse if and only if they intersect cleanly with e=0e=0. If M,P⊂NM,P\subset N are compact submanifolds meeting cleanly in ZZ, then for every sufficiently small generic perturbation M′M' of MM the transverse intersection M′∩PM'\cap P is cobordant to the zero set of a generic section of EE over ZZ.

So a clean intersection has the right tangent spaces and the wrong dimension, by ee, and the excess bundle measures exactly what a perturbation will cut away. A submanifold meets itself cleanly with EE its normal bundle; the diagonal of M×MM\times M meets itself cleanly with E=TME=TM. (b) is not clean. In each case there is an object that records the correct answer before any perturbation is made.

Derived manifolds

By [thm:MR] the function algebra of a transverse fibre product is a pushout of C∞C^\infty-rings. The derived fibre product is defined by the homotopy pushout.

Definition 1.11. A simplicial C∞C^\infty-ring is a product-preserving functor Cart→S\mathsf{Cart}\to\mathcal{S}; these form an ∞\infty-category sC∞RingsC^\infty\mathsf{Ring}, with homotopy pushouts written ⊗L\otimes^{\mathbb L}. Each has homotopy groups πk\pi_k, with π0\pi_0 a C∞C^\infty-ring and πk\pi_k a π0\pi_0-module.

Theorem 1.12 (Universal property of derived manifolds). [literature; Carchedi–Steffens [14]; Spivak [48]] There is an ∞\infty-category DMfd\mathsf{DMfd} that has finite limits and is idempotent complete, with a functor i ⁣:Mfd→DMfdi\colon\mathsf{Mfd}\to\mathsf{DMfd} that preserves transverse fibre products and the terminal object, and that is universal with these properties: for every idempotent complete ∞\infty-category C\mathcal C with finite limits, composition with ii is an equivalence between finite-limit-preserving functors DMfd→C\mathsf{DMfd}\to\mathcal C and functors Mfd→C\mathsf{Mfd}\to\mathcal C preserving transverse fibre products and the terminal object. Moreover DMfd\mathsf{DMfd} is equivalent to the opposite of the ∞\infty-category of finitely presented simplicial C∞C^\infty-rings, with i(M)=C∞(M)i(M)=C^\infty(M), and ii is fully faithful.

The theorem says that nothing was chosen. Whoever wants all fibre products of manifolds, and wants the transverse ones to stay as they are, is led to this ∞\infty-category and no other. The role of idempotent completeness was explained after [thm:karoubi]. That the answer is an ∞\infty-category and not a category is forced: the last row of [tab:four] has no 11-categorical solution.

Locally every derived fibre product of manifolds is a derived zero locus, and those can be written down.

Proposition 1.13 (Koszul model). [literature; [48], [14]] Let E→ME\to M be a vector bundle of rank kk with a section ss, and let Z(s)=M×s,E,0hMZ(s)=M\times^h_{s,E,0}M be the derived zero locus. Its simplicial C∞C^\infty-ring corresponds to the commutative differential graded algebra (Γ(Λ∙E∨), ιs):⋯→Γ(Λ2E∨)→ ιs Γ(E∨)→ ιs C∞(M),\bigl(\Gamma(\Lambda^\bullet E^\vee),\ \iota_s\bigr):\qquad \cdots\to\Gamma(\Lambda^2E^\vee)\xrightarrow{\ \iota_s\ }\Gamma(E^\vee)\xrightarrow{\ \iota_s\ }C^\infty(M), so π0=C∞(M)/(s)\pi_0=C^\infty(M)/(s). If ss is transverse to the zero section then πk=0\pi_k=0 for k>0k>0 and Z(s)=i(s−1(0))Z(s)=i\bigl(s^{-1}(0)\bigr). For maps f ⁣:M→Nf\colon M\to N and g ⁣:P→Ng\colon P\to N, the derived fibre product M×NhPM\times^h_NP is the derived zero locus of a section of (f×g)∗TN(f\times g)^*TN over a neighbourhood of M×NPM\times_NP in M×PM\times P.

The number vdimZ(s)=dim⁡M−k\mathop{\mathrm{vdim}}Z(s)=\dim M-k is the virtual dimension; for a derived fibre product it is dim⁡M+dim⁡P−dim⁡N\dim M+\dim P-\dim N whether or not the maps are transverse.

Example 1.14. (a) M=RM=\mathbb{R}, EE trivial of rank 11, s=x2s=x^2. The complex is C∞(R)→x2C∞(R)C^\infty(\mathbb{R})\xrightarrow{x^2}C^\infty(\mathbb{R}), which is injective, so π1=0\pi_1=0 and Z(s)Z(s) is the classical object R[ε]/(ε2)\mathbb{R}[\varepsilon]/(\varepsilon^2) of virtual dimension 00: not a manifold, and not derived either.

(b) M=ptM=\mathrm{pt}, E=RE=\mathbb{R}, s=0s=0. The complex is R→0R\mathbb{R}\xrightarrow{0}\mathbb{R}; π0=π1=R\pi_0=\pi_1=\mathbb{R}. This is the point of virtual dimension −1-1 in [tab:four].

(c) s=0s=0 on any E→ME\to M: the algebra is Γ(Λ∙E∨)\Gamma(\Lambda^\bullet E^\vee) with zero differential, of virtual dimension dim⁡M−k\dim M-k. It is the derived self-intersection of the zero section, and by the tubular neighbourhood theorem the derived self-intersection of any submanifold with normal bundle EE.

(d) The diagonal: M×M×MhMM\times^h_{M\times M}M has functions Γ(Λ∙T∗M)\Gamma(\Lambda^\bullet T^*M), the differential forms placed in homological degrees, and virtual dimension 00. Its “number of points” will turn out to be the Euler characteristic ([ch:degree]).

Every object of DMfd\mathsf{DMfd} has a tangent complex at each point. For X=M×NhPX=M\times^h_NP at (x,y)(x,y) over zz it is TX=[TxM⊕TyP→ df−dg TzN]\mathbb T_X=\bigl[T_xM\oplus T_yP\xrightarrow{\ df-dg\ }T_zN\bigr] in degrees 00 and 11, with H0H^0 the fibre product of tangent spaces and H1H^1 the fibre of the excess bundle. Its Euler characteristic is the virtual dimension.

Theorem 1.15 (Transversality, derived form). [reformulation; cf. [48], [14]] For smooth maps f ⁣:M→Nf\colon M\to N and g ⁣:P→Ng\colon P\to N, the derived fibre product M×NhPM\times^h_NP lies in the essential image of i ⁣:Mfd→DMfdi\colon\mathsf{Mfd}\to\mathsf{DMfd} if and only if f⋔gf\pitchfork g.

Proof. If f⋔gf\pitchfork g, then ii preserves the fibre product by 1.12. Conversely, if M×NhP≃i(Z)M\times^h_NP\simeq i(Z) for a manifold ZZ, its tangent complex is concentrated in degree 00 at every point, so H1=TzN/(df TxM+dg TyP)H^1=T_zN/(df\,T_xM+dg\,T_yP) vanishes at every point of the fibre product, which is transversality. ◻

Compare the situation in Mfd\mathsf{Mfd}. There a fibre product may fail to exist, or exist and be wrong, or exist and be right without one’s knowing which. In DMfd\mathsf{DMfd} it always exists and is always right, and it is a manifold in exactly the transverse case.

Genericity as a statement about derived manifolds

Theorem 1.16. [literature; Spivak [48]] Compact derived manifolds that are locally derived zero loci have a cobordism theory, and ii induces an isomorphism from the cobordism ring of manifolds to it. In particular a compact X=M×NhPX=M\times^h_NP has a class [X][X] in the cobordism group Nd\mathfrak N_d of dimension d=dim⁡M+dim⁡P−dim⁡Nd=\dim M+\dim P-\dim N ([ch:cobordism]), represented by M×NP′M\times_NP' for any g′⋔fg'\pitchfork f sufficiently close to gg.

Surjectivity is the transversality theorem: perturb. Injectivity, and the independence of the perturbation, is 1.4 applied to a path of perturbations: the fibre product of the path with AA is a manifold with boundary whose two ends are the two candidates. Putting Theorem 1.12, Theorem 1.15, Theorem 1.16 together gives a statement of the method of the source that makes no mention of perturbation until the end.

Principle 1.17 (Generic position). [reformulation; cf. Hirsch, Ch. 3; [48]] Every finite diagram of manifolds has a limit in DMfd\mathsf{DMfd}. If the limit is compact it has a well-defined cobordism class. The class is represented by the honest limit of any sufficiently generic perturbation of the diagram, and two such representatives are cobordant by the limit of a generic path between the perturbations.

Example 1.18. The zero set of x2−εx^2-\varepsilon on R\mathbb{R}, as an oriented 00-manifold, is {+ε}\{+\sqrt\varepsilon\} with sign ++ and {−ε}\{-\sqrt\varepsilon\} with sign −- for ε>0\varepsilon>0, and empty for ε<0\varepsilon<0. Both have class 0∈Ω0=Z0\in\Omega_0=\mathbb{Z}, and the cobordism between them is the parabola {x2=ε}\{x^2=\varepsilon\} in the (ε,x)(\varepsilon,x)-plane, a manifold by 1.4. The derived zero set at ε=0\varepsilon=0 is the dual numbers, with class 00. Its length as an algebra is 22, which is the count of complex roots; the cobordism class and the length are different invariants and only the first is stable under real perturbation.

An object of negative virtual dimension, such as 1.14(b), has class zero for trivial reasons and is nevertheless not empty. The cobordism class is the part of a derived manifold that perturbation can reach, and it is the only part the classical theory uses.

Notes

Transversality is Thom’s; the parametric form and its use as the organizing principle are the substance of Hirsch’s Chapter 3 and of Abraham’s work. Derived manifolds were introduced by Spivak, after Lurie’s sketch in his work on derived algebraic geometry; variants are due to Borisov and Noel and to Joyce, and the universal property is [14], building on [37]. Clean intersections are Bott’s. The excess bundle and the associated formula are Quillen’s in cobordism and Fulton–MacPherson’s in algebraic geometry.

Exercises

Exercise 1.1. Let f ⁣:M→Rnf\colon M\to\mathbb{R}^n be smooth and A⊂RnA\subset\mathbb{R}^n a submanifold. Show that f+vf+v is transverse to AA for almost every v∈Rnv\in\mathbb{R}^n. Deduce that almost every translate of a curve in R3\mathbb{R}^3 misses a given curve.

Exercise 1.2. Prove 1.7.

Exercise 1.3. Compute the homotopy groups of R⊗C∞(R)LR\mathbb{R}\otimes^{\mathbb L}_{C^\infty(\mathbb{R})}\mathbb{R}, where both maps are evaluation at 00, and of C∞(R)⊗C∞(R)LRC^\infty(\mathbb{R})\otimes^{\mathbb L}_{C^\infty(\mathbb{R})}\mathbb{R}, where the first map is x↦x2x\mapsto x^2. Compare with 1.14.

Exercise 1.4. Let U⊂RmU\subset\mathbb{R}^m be open and f ⁣:U→Rf\colon U\to\mathbb{R} smooth. Show that fa(x)=f(x)+⟨a,x⟩f_a(x)=f(x)+\langle a,x\rangle is a Morse function for almost every a∈Rma\in\mathbb{R}^m.

Exercise 1.5. Deduce from 1.2 that every continuous map Dn→DnD^n\to D^n has a fixed point.

Bundles and Invariants

Vector Bundles and Classifying Maps

The first half of Hirsch’s fourth chapter proves that a vector bundle over B×IB\times I is a product in the II direction, deduces that homotopic maps pull back isomorphic bundles, and classifies bundles over an nn-manifold by homotopy classes of maps to a Grassmannian. The three statements are one statement about a stack: it is concordance-invariant, so its sections are classified by maps to its shape. The same theorem, applied to other stacks, gives de Rham’s theorem and the classification of principal bundles, and in later chapters the classification of immersions and of manifolds up to cobordism.

The stack of vector bundles

Definition 1.1. For a manifold MM, Vectk(M)\mathsf{Vect}_k(M) is the groupoid of real vector bundles of rank kk over MM and their isomorphisms. Pullback makes M↦Vectk(M)M\mapsto\mathsf{Vect}_k(M) a functor Mfdop→Gpd\mathsf{Mfd}^{\mathrm{op}}\to\mathsf{Gpd}.

Proposition 1.2. [classical; ] Vectk\mathsf{Vect}_k is a stack: bundles and their isomorphisms glue along open covers. Its restriction to Cart\mathsf{Cart} is equivalent to U↦B C∞(U,GLk)U\mapsto B\,C^\infty(U,\mathop{\mathrm{GL}}_k), the groupoid with one object and automorphisms the smooth maps U→GLk(R)U\to\mathop{\mathrm{GL}}_k(\mathbb{R}). Hence Vectk≃BGLk\mathsf{Vect}_k\simeq B\mathop{\mathrm{GL}}_k, the delooping in H\mathbf H of the sheaf of groups represented by GLk(R)\mathop{\mathrm{GL}}_k(\mathbb{R}), and Vectk(M) ≃ H(M,BGLk).\mathsf{Vect}_k(M)\ \simeq\ \mathbf H(M,B\mathop{\mathrm{GL}}_k).

Proof. Descent for bundles is the construction of a bundle from transition functions satisfying the cocycle condition. Over a Euclidean space every bundle is trivial (by 1.3 below, since Rn\mathbb{R}^n is contractible), which identifies the restriction. An object of H\mathbf H is determined by its restriction to Cart\mathsf{Cart}. ◻

So “a vector bundle is a map to BGLkB\mathop{\mathrm{GL}}_k” is true as stated, in H\mathbf H, with no homotopy classes taken; a cocycle (gij)(g_{ij}) on a cover is the same map written out on the Čech nerve of the cover. The object BGLkB\mathop{\mathrm{GL}}_k is not a smooth set. Its value on a point is the groupoid of kk-dimensional vector spaces, and it is the first place in this book where the ∞\infty-topos of [def:shape], and not just its 00-truncated part, is needed.

Homotopy invariance

Theorem 1.3. [classical; Hirsch 4.1.5, 4.2.4] Every vector bundle over M×IM\times I is isomorphic to (ξ∣M×0)×I(\xi|_{M\times0})\times I. Consequently, if f0,f1 ⁣:M→Nf_0,f_1\colon M\to N are homotopic and ξ\xi is a bundle over NN, then f0∗ξ≅f1∗ξf_0^*\xi\cong f_1^*\xi.

The proof chooses a connection, or in Hirsch’s version covers M×IM\times I by product neighbourhoods on which the bundle is trivial and patches with a partition of unity; either way a bundle over M×IM\times I is transported along the II direction. In the language of [def:shape], two bundles over MM are concordant if they are the ends of a bundle over M×RM\times\mathbb{R}, and the theorem says that concordant bundles are isomorphic. The converse is trivial.

Classification

Let Grk(Rs)\mathop{\mathrm{Gr}}_k(\mathbb{R}^s) be the Grassmannian of kk-planes in Rs\mathbb{R}^s, a compact manifold of dimension k(s−k)k(s-k), with tautological bundle γk\gamma_k.

Theorem 1.4 (Classification of vector bundles). [classical; Hirsch 4.3.4] Let ξ\xi be a vector bundle of rank kk over a manifold MM of dimension nn. If s≥k+ns\ge k+n there is a map f ⁣:M→Grk(Rs)f\colon M\to\mathop{\mathrm{Gr}}_k(\mathbb{R}^s) with f∗γk≅ξf^*\gamma_k\cong\xi, and if s>k+ns>k+n it is unique up to homotopy. So for s>k+ns>k+n, π0Vectk(M) ≅ [M,Grk(Rs)].\pi_0\mathsf{Vect}_k(M)\ \cong\ [M,\mathop{\mathrm{Gr}}_k(\mathbb{R}^s)].

Hirsch proves existence by embedding the bundle in a trivial one, which is a transversality argument, and uniqueness by the same argument one dimension up. The dimension restrictions are the connectivity of the inclusion Grk(Rs)→Grk(R∞)=BO(k)\mathop{\mathrm{Gr}}_k(\mathbb{R}^s)\to\mathop{\mathrm{Gr}}_k(\mathbb{R}^\infty)=B\mathrm{O}(k).

Theorem 1.5 (Concordance classification). [literature; [7]; cf. [thm:oka]] For every F∈HF\in\mathbf H and every manifold MM, the set of concordance classes of sections of FF over MM is in natural bijection with [M,ΠF][M,\Pi F], and more precisely Π[M,F]≃Map(M,ΠF)\Pi[M,F]\simeq\mathop{\mathrm{Map}}(M,\Pi F).

For F=BGLkF=B\mathop{\mathrm{GL}}_k the shape is the classifying space of the topological group GLk(R)\mathop{\mathrm{GL}}_k(\mathbb{R}), which is BO(k)B\mathrm{O}(k) by Gram–Schmidt ([exer:gram]). Concordance classes are isomorphism classes by 1.3. So 1.5 gives π0Vectk(M)≅[M,BO(k)]\pi_0\mathsf{Vect}_k(M)\cong[M,B\mathrm{O}(k)], which is 1.4 with s=∞s=\infty; and it gives more, namely the homotopy type Map(M,BO(k))\mathop{\mathrm{Map}}(M,B\mathrm{O}(k)) for the space of all rank-kk bundles over MM, whose fundamental group at ξ\xi is the group of components of the gauge group of ξ\xi.

lists other instances. In each row the left-hand side is a geometric structure that can be pulled back, the right-hand side is a space, and the theorem says that deformation classes of the structure on MM are homotopy classes of maps into the space.

Instances of 1.5. The union in the last row is over diffeomorphism classes of closed dd-manifolds.
FF (sections over MM) shape ΠF\Pi F classical name
maps to a manifold NN NN smooth approximation ([cor:homotopy])
vector bundles of rank kk BO(k)B\mathrm{O}(k) 1.4
principal GG-bundles, GG a Lie group BGBG classification of principal bundles
vector bundles with connection BO(k)B\mathrm{O}(k) connections form a contractible space
closed differential pp-forms K(R,p)K(\mathbb{R},p) de Rham’s theorem
all differential pp-forms pt\mathrm{pt} forms are contractible
bundles of closed dd-manifolds ∐[W]BDiff(W)\coprod_{[W]}B\mathop{\mathrm{Diff}}(W) classification of manifold bundles

The fifth row deserves a comment. Two closed pp-forms on MM are concordant exactly when they are cohomologous (1.4 for p=1p=1), so the theorem says HdRp(M)≅[M,K(R,p)]H^p_{\mathrm{dR}}(M)\cong[M,K(\mathbb{R},p)], which is de Rham’s theorem. The sheaf of closed forms is not homotopy invariant; its concordance classes are.

Bundles as idempotents

Theorem 1.6. [classical; Swan; Hirsch 4.3.3] Every vector bundle ξ\xi over a manifold MM has a complement: a bundle η\eta with ξ⊕η\xi\oplus\eta trivial. The functor of sections is an equivalence from vector bundles over MM to finitely generated projective C∞(M)C^\infty(M)-modules, taking ⊕\oplus and ⊗\otimes to ⊕\oplus and ⊗C∞(M)\otimes_{C^\infty(M)}.

So the category of vector bundles over MM is the idempotent completion of the category of trivial bundles, just as Mfd\mathsf{Mfd} is the idempotent completion of Open\mathsf{Open} ([thm:karoubi]), and for the same reason: finite dimension bounds the number of charts needed. A complement exhibits ξ\xi as the image of an idempotent endomorphism ee of a trivial bundle M×RsM\times\mathbb{R}^s, that is, of a smooth map from MM to the manifold of idempotent s×ss\times s matrices of rank kk. Projecting an idempotent to the orthogonal projection with the same image identifies that manifold, up to homotopy, with Grk(Rs)\mathop{\mathrm{Gr}}_k(\mathbb{R}^s). A classifying map and a complement are the same datum.

Example 1.7. Over the circle RP1=R/πZ\mathbb{RP}^1=\mathbb{R}/\pi\mathbb{Z} the matrix e(θ)=(cos⁡2θcos⁡θsin⁡θcos⁡θsin⁡θsin⁡2θ)e(\theta)=\bigl(\begin{smallmatrix}\cos^2\theta&\cos\theta\sin\theta\\\cos\theta\sin\theta&\sin^2\theta \end{smallmatrix}\bigr) is an idempotent of rank 11; its image is the tautological line bundle. A unit vector in the image transported once around the circle returns to its negative, so the bundle is the Möbius band. Over SnS^n the idempotent e(x)=I−xxTe(x)=I-xx^{\mathsf T} on Rn+1\mathbb{R}^{n+1} has image TSnTS^n and complement the trivial normal line. The script checks both.

Orientations and tangential structures

A bundle is orientable if its transition functions can be chosen in GLk+(R)\mathop{\mathrm{GL}}_k^+(\mathbb{R}); an orientation is such a choice up to equivalence. In terms of classifying maps, an orientation is a lift of M→BO(k)M\to B\mathrm{O}(k) through the double cover BSO(k)→BO(k)B\mathrm{SO}(k)\to B\mathrm{O}(k), and the obstruction to its existence is the class w1(ξ)∈H1(M;Z/2)w_1(\xi)\in H^1(M;\mathbb{Z}/2) pulled back from the generator of H1(BO(k);Z/2)H^1(B\mathrm{O}(k);\mathbb{Z}/2). A manifold is orientable if its tangent bundle is.

Definition 1.8. A tangential structure is a map of spaces θ ⁣:B→BO(d)\theta\colon B\to B\mathrm{O}(d). A θ\theta-structure on a dd-manifold MM is a lift of the classifying map τM ⁣:M→BO(d)\tau_M\colon M\to B\mathrm{O}(d) of TMTM through θ\theta.

Orientations (B=BSO(d)B=B\mathrm{SO}(d)), framings (B=ptB=\mathrm{pt}), spin structures, almost complex structures, and maps to a fixed background space XX (B=BO(d)×XB=B\mathrm{O}(d)\times X) are all of this form. Because θ\theta-structures pull back along open embeddings and are defined by a lifting problem in spaces, the stack of dd-manifolds with θ\theta-structure is again an object to which 1.5 applies; this is how the tangential structures of [ch:cobordism] enter.

Theorem 1.9. [classical; Hirsch 4.4.7] A closed non-orientable nn-manifold does not embed in a simply connected (n+1)(n+1)-manifold. In particular RP2n\mathbb{RP}^{2n} does not embed in R2n+1\mathbb{R}^{2n+1}.

Proof. Let M⊂NM\subset N be such an embedding with normal line bundle ν\nu. A loop in NN transverse to MM meets it in an even number of points, since the count modulo 22 is a homotopy invariant ([thm:intersection]) and the loop is null-homotopic. A loop in MM along which ν\nu is non-trivial can be pushed off MM to a loop meeting MM once. So ν\nu is trivial, and TM⊕ν≅TN∣MTM\oplus\nu\cong TN|_M gives w1(TM)=w1(TN∣M)=0w_1(TM)=w_1(TN|_M)=0, as NN is simply connected. ◻

The tangent bundle as a map

The tangent bundle of an mm-manifold is a point τM\tau_M of Map(M,BO(m))\mathop{\mathrm{Map}}(M,B\mathrm{O}(m)). For a submanifold M⊂Rm+kM\subset\mathbb{R}^{m+k} the classifying map can be taken to be the Gauss map x↦TxM∈Grm(Rm+k)x\mapsto T_xM\in\mathop{\mathrm{Gr}}_m(\mathbb{R}^{m+k}), and the normal bundle is classified by x↦(TxM)⊥x\mapsto(T_xM)^\perp. The normal bundle depends on the embedding, but its class in [M,BO][M,B\mathrm{O}], where BO=colimkBO(k)B\mathrm{O}=\mathop{\mathrm{colim}}_kB\mathrm{O}(k), does not; this stable normal bundle νM\nu_M is inverse to τM\tau_M under direct sum, and [ch:tubular] shows that it is well defined as a point of a contractible space of choices. Characteristic classes of MM are the pullbacks along τM\tau_M or νM\nu_M of classes in H∗(BO;Z/2)=Z/2[w1,w2,… ]H^*(B\mathrm{O};\mathbb{Z}/2)=\mathbb{Z}/2[w_1,w_2,\dots]. Every invariant of manifolds used in Degree, Intersection Numbers, and the Euler Characteristic, Cobordism is obtained from τM\tau_M, as a consequence of the remark after [prop:jets].

Notes

Steenrod’s book is the classical source for classification by maps to Grassmannians; Hirsch’s treatment by transversality is his own. The stack-theoretic formulation is standard; see [45]. The equivalence with projective modules is Swan’s, after Serre. has a long prehistory: Brown representability, Gromov’s sheaf-theoretic h-principle, and Madsen and Weiss’s use of concordance in the proof of the Mumford conjecture.

Exercises

Exercise 1.1. Show that the tautological line bundle over RP1\mathbb{RP}^1 is not trivial, and that its total space is the open Möbius band.

Exercise 1.2. Let e ⁣:M→Rs×se\colon M\to\mathbb{R}^{s\times s} be smooth with e2=ee^2=e and constant rank kk. Show that the image of ee is a vector bundle whose module of sections is e C∞(M)se\,C^\infty(M)^s, a direct summand of a free module.

Exercise 1.3. Show that TSn⊕RTS^n\oplus\mathbb{R} is trivial, and conclude that the composite Sn→BO(n)→BOS^n\to B\mathrm{O}(n)\to B\mathrm{O} classifying the stable tangent bundle is null-homotopic.

Exercise 1.4. Show that two closed 11-forms on MM are concordant (are the restrictions to M×{0}M\times\{0\} and M×{1}M\times\{1\} of a closed 11-form on M×RM\times\mathbb{R}) if and only if they differ by an exact form.

Exercise 1.5. Show that isomorphism classes of line bundles over MM correspond to H1(M;Z/2)=Hom(π1M,Z/2)H^1(M;\mathbb{Z}/2)=\mathop{\mathrm{Hom}}(\pi_1M,\mathbb{Z}/2), and count them for the torus and for the Klein bottle. Which one is the determinant of the tangent bundle?

Tubular Neighbourhoods, Collars, and Spaces of Choices

The second half of Hirsch’s fourth chapter proves that a submanifold has a tubular neighbourhood, that the boundary of a manifold has a collar, and that each is unique up to isotopy. The uniqueness statements are what allow later constructions (connected sums, gluing of cobordisms, the Pontryagin–Thom collapse) to be called well defined. “Unique up to isotopy” says that a space of choices is connected. In every case in this chapter the space of choices is in fact contractible, and that stronger statement is what makes the constructions functorial.

Tubular neighbourhoods

Definition 1.1. Let M⊂NM\subset N be a submanifold, both without boundary. A tubular neighbourhood of MM in NN is a vector bundle ξ=(p,E,M)\xi=(p,E,M) with an embedding φ ⁣:E→N\varphi\colon E\to N onto an open set that restricts to the identity on the zero section. The normal bundle is ν=ν(M,N)=TN∣M/TM\nu=\nu(M,N)=TN|_M/TM.

The differential of φ\varphi along the zero section identifies ξ\xi with ν\nu, so one may take ξ=ν\xi=\nu.

Theorem 1.2 (Tubular neighbourhoods). [classical; Hirsch 4.5.2, 4.5.3] Every submanifold M⊂NM\subset N has a tubular neighbourhood, and any two are isotopic relative to MM.

The existence proof embeds NN in a Euclidean space, uses the normal exponential map (x,v)↦x+v(x,v)\mapsto x+v there, which is a local diffeomorphism along the zero section by the inverse function theorem, and retracts back to NN. For the unit circle in the plane the map is (θ,t)↦(1+t)(cos⁡θ,sin⁡θ)(\theta,t)\mapsto(1+t)(\cos\theta,\sin\theta), with Jacobian determinant 1+t1+t; it is an embedding for t>−1t>-1. In general the radius is limited by the focal points of MM (1.3).

Theorem 1.3. [literature; folklore; see [23]] The space Tub(M,N)\mathop{\mathrm{Tub}}(M,N) of tubular neighbourhoods φ ⁣:ν→N\varphi\colon\nu\to N whose differential along the zero section induces the identity of ν\nu is contractible.

Idea of proof. Given two, first shrink one fibrewise until its image lies in the image of the other; this is a canonical path. Then ψ=φ0−1φ1\psi=\varphi_0^{-1}\varphi_1 is an embedding of ν\nu in itself, fixing the zero section, with fibre derivative the identity there, and ψt(v)=ψ(tv)/t\psi_t(v)=\psi(tv)/t for 0<t≤10<t\le1 extends smoothly to t=0t=0 by the identity. Both steps depend continuously on parameters. ◻

With boundary, the right notion is that of a neat submanifold ([def:neat]), and the theorems hold with tubular neighbourhoods that restrict to tubular neighbourhoods in the boundary (Hirsch 4.6.3, 4.6.4).

Collars and gluing

Theorem 1.4 (Collars). [classical; Brown; Hirsch 4.6.1] For a manifold MM with boundary there is an embedding ∂M×[0,∞)→M\partial M\times[0,\infty)\to M extending the identity of ∂M×0\partial M\times0. Any two are isotopic, and the space of collars is contractible.

Theorem 1.5 (Gluing). [classical; Hirsch 8.2.1, 8.2.3] Let P,QP,Q be manifolds with boundary and f ⁣:∂Q→∂Pf\colon\partial Q\to\partial P a diffeomorphism. The space P∪fQP\cup_fQ has a smooth structure restricting to the given ones on PP and QQ; any two such structures are diffeomorphic by a diffeomorphism supported near the seam; and isotopic gluing maps give diffeomorphic results.

The smooth structure is built by choosing collars on both sides and declaring the union of the two collars to be a product neighbourhood of the seam. So composition of cobordisms is defined only after a choice, and the choice is from a contractible space. A category whose composition depends on contractible choices is an ∞\infty-category presented in the usual way, and this is the elementary reason that cobordism categories are ∞\infty-categories ([ch:cobordism]). The dependence on the isotopy class of ff is real: gluing two discs along a diffeomorphism of Sn−1S^{n-1} not isotopic to one that extends over the disc can give an exotic sphere (1.5).

Contractible spaces of choices

Principle 1.6 (Choices). [reformulation; cf. ] Let p ⁣:E→Bp\colon\mathcal E\to\mathcal B be a map of spaces whose fibres are contractible. Then pp is an equivalence, and every functor on E\mathcal E factors essentially uniquely through B\mathcal B. A construction on objects bb that requires a choice cc from a contractible space Eb\mathcal E_b is therefore as good as a construction on the objects alone.

lists the cases used in this book. The first four are convex or affine and need no argument. The fifth and sixth are Theorem 1.3, Theorem 1.4. The seventh is Whitney’s theorem with parameters.

Contractible spaces of choices.
choice what it makes well defined
inner product on a vector bundle reduction BGLk≃BO(k)B\mathop{\mathrm{GL}}_k\simeq B\mathrm{O}(k); sphere and disc bundles
Riemannian metric on MM exponential maps, gradients
connection on a bundle parallel transport, [thm:bundlehomotopy]
partition of unity subordinate to a cover gluing of local constructions
tubular neighbourhood of M⊂NM\subset N collapse maps, normal data
collar of ∂M\partial M gluing, composition of cobordisms
embedding M→R∞M\to\mathbb{R}^\infty stable normal bundle; a model of BDiff(M)B\mathop{\mathrm{Diff}}(M)

Theorem 1.7. [classical; Whitney] For every manifold MM the space Emb(M,R∞)=colimqEmb(M,Rq)\mathop{\mathrm{Emb}}(M,\mathbb{R}^\infty)=\mathop{\mathrm{colim}}_q\mathop{\mathrm{Emb}}(M,\mathbb{R}^q) is weakly contractible.

It is non-empty by [thm:whitney]. Contractibility is a two-step linear homotopy: move a family of embeddings into the odd coordinates, then slide it to a fixed embedding placed in the even coordinates (1.2). The group Diff(M)\mathop{\mathrm{Diff}}(M) acts freely on Emb(M,R∞)\mathop{\mathrm{Emb}}(M,\mathbb{R}^\infty), and the quotient, the space of submanifolds of R∞\mathbb{R}^\infty diffeomorphic to MM, is therefore a model for the classifying space BDiff(M)B\mathop{\mathrm{Diff}}(M). A map B→BDiff(M)B\to B\mathop{\mathrm{Diff}}(M) is a family of submanifolds of R∞\mathbb{R}^\infty parametrized by BB, that is, a bundle of manifolds with fibre MM; this is the last row of [tab:oka].

The collapse map and Thom spaces

Definition 1.8. The Thom space Th(ξ)\mathop{\mathrm{Th}}(\xi) of a vector bundle ξ\xi over a compact base is the one-point compactification of its total space; equivalently the quotient D(ξ)/S(ξ)D(\xi)/S(\xi) of the unit disc bundle by the unit sphere bundle, for any inner product.

Proposition 1.9. [classical; ] Th(ξ×η)≅Th(ξ)∧Th(η)\mathop{\mathrm{Th}}(\xi\times\eta)\cong\mathop{\mathrm{Th}}(\xi)\wedge\mathop{\mathrm{Th}}(\eta); Th(ξ⊕Rj)≅ΣjTh(ξ)\mathop{\mathrm{Th}}(\xi\oplus\mathbb{R}^j)\cong\Sigma^j\mathop{\mathrm{Th}}(\xi); the Thom space of the trivial bundle M×RkM\times\mathbb{R}^k is ΣkM+\Sigma^kM_+; and the Thom space of the tautological line bundle over RPn\mathbb{RP}^n is RPn+1\mathbb{RP}^{n+1}.

Let M⊂Rn+kM\subset\mathbb{R}^{n+k} be a closed submanifold of dimension nn with a tubular neighbourhood φ ⁣:ν→U\varphi\colon\nu\to U. The Pontryagin–Thom collapse is the map c ⁣:Sn+k=Rn+k∪{∞}⟶U+≅Th(ν)c\colon S^{n+k}=\mathbb{R}^{n+k}\cup\{\infty\}\longrightarrow U^+\cong\mathop{\mathrm{Th}}(\nu) that is φ−1\varphi^{-1} on UU and sends everything else to the point at infinity. More generally an embedding M⊂NM\subset N of closed manifolds gives N+→Th(ν(M,N))N_+\to\mathop{\mathrm{Th}}(\nu(M,N)). By 1.3 and 1.6 the collapse is a well-defined point of a mapping space, not merely a homotopy class. It is a map in the wrong direction, from the ambient manifold to (a twisted suspension of) the submanifold, and it is the geometric source of all wrong-way maps in cohomology.

The Thom space depends on the embedding only through a suspension. By 1.7 the desuspended suspension spectrum M−TM=Σ−(n+k)Σ∞Th(ν)M^{-TM}=\Sigma^{-(n+k)}\Sigma^\infty\mathop{\mathrm{Th}}(\nu) is well defined; it is the Thom spectrum of the stable normal bundle νM=−TM\nu_M=-TM.

Theorem 1.10 (Atiyah duality). [classical; Atiyah [4]; Milnor–Spanier] For a closed manifold MM the suspension spectrum Σ+∞M\Sigma^\infty_+M is dualizable in the stable homotopy category, with dual M−TMM^{-TM}. The coevaluation S→Σ+∞M∧M−TM\mathbb S\to\Sigma^\infty_+M\wedge M^{-TM} is the collapse Sn+k→Th(ν)S^{n+k}\to\mathop{\mathrm{Th}}(\nu) followed by the diagonal Th(ν)→M+∧Th(ν)\mathop{\mathrm{Th}}(\nu)\to M_+\wedge\mathop{\mathrm{Th}}(\nu). The evaluation is the collapse for the diagonal embedding M→M×MM\to M\times M, whose normal bundle is TMTM.

Dualizability and its use are recalled in [app:toolkit]. The theorem is the homotopy-theoretic form of Poincaré duality, and [ch:degree] uses it to compute Euler characteristics.

The normal bundle without choices

There is a short exact sequence 0→TM→TN∣M→ν→00\to TM\to TN|_M\to\nu\to0, so the normal bundle is the cofibre of the map of tangent bundles; in the notation of [ch:transversality] it is the degree-11 part of the relative tangent complex of M→NM\to N. The tubular neighbourhood theorem says that the germ of NN along MM is diffeomorphic to the germ of ν\nu along its zero section. There is a canonical object behind the non-canonical diffeomorphism.

Proposition 1.11 (Deformation to the normal bundle). [classical; ] There is a manifold D(M,N)\mathcal D(M,N) with a submersion to R\mathbb{R} whose fibre over t≠0t\ne0 is NN and whose fibre over 00 is ν(M,N)\nu(M,N), containing M×RM\times\mathbb{R} as a closed submanifold. It is functorial for maps of pairs. A tubular neighbourhood is a trivialization of this family near M×RM\times\mathbb{R}.

For the diagonal M⊂M×MM\subset M\times M this is Connes’s tangent groupoid, interpolating between the pair groupoid of MM and the tangent bundle. The proposition illustrates a general pattern: where the classical text makes a choice and proves uniqueness up to isotopy, there is usually a canonical object in a larger category of which the choices are trivializations.

By [ex:derived](c) the derived self-intersection M×NhMM\times^h_NM depends only on the normal bundle: its functions are Γ(Λ∙ν∨)\Gamma(\Lambda^\bullet\nu^\vee). Its virtual dimension is 2dim⁡M−dim⁡N2\dim M-\dim N, and its cobordism class is that of the zero set of a generic section of ν\nu ([prop:clean]).

Notes

Hirsch’s proofs of uniqueness of tubular neighbourhoods and collars give isotopies depending continuously on the data, which is most of the way to contractibility; the statements with parameters are spelled out in [23] and in the literature on cobordism categories [22]. The collapse map is Pontryagin’s for framed submanifolds and Thom’s in general. For Thom spectra of virtual bundles and their multiplicative properties see [1]. The deformation to the normal cone is due to Fulton and MacPherson in algebraic geometry; the smooth version is in Connes’s and Higson’s work on index theory.

Exercises

Exercise 1.1. Show that the space of inner products on a finite-dimensional real vector space is convex, and that O(k)→GLk(R)\mathrm{O}(k)\to\mathop{\mathrm{GL}}_k(\mathbb{R}) is a homotopy equivalence. Deduce BGLk(R)≃BO(k)B\mathop{\mathrm{GL}}_k(\mathbb{R})\simeq B\mathrm{O}(k).

Exercise 1.2. Prove 1.7 for π0\pi_0: any two embeddings M→R∞M\to\mathbb{R}^\infty are isotopic through embeddings. Explain why the straight-line homotopy between two embeddings in Rq\mathbb{R}^q need not be an isotopy.

Exercise 1.3. For a closed plane curve with curvature κ\kappa and inward unit normal nn, show that (s,t)↦γ(s)+t n(s)(s,t)\mapsto\gamma(s)+t\,n(s) has Jacobian determinant 1−tκ(s)1-t\kappa(s). Find the largest rr such that the map is a local diffeomorphism for ∣t∣<r|t|<r when the curve is the ellipse with semi-axes a>ba>b.

Exercise 1.4. Prove 1.9.

Exercise 1.5. Let ff be a diffeomorphism of Sn−1S^{n-1} and Σf=Dn∪fDn\Sigma_f=D^n\cup_fD^n. Show that Σf\Sigma_f is homeomorphic to SnS^n, and diffeomorphic to SnS^n if ff extends to a diffeomorphism of DnD^n.

Degree, Intersection Numbers, and the Euler Characteristic

Hirsch’s fifth chapter is where the method of the Preface first produces numbers. Each invariant is a signed count of points in a transverse preimage, and each proof of invariance exhibits two such counts as the two ends of a compact oriented 11-manifold. In the terms of [ch:transversality], each invariant is the class in Ω0=Z\Omega_0=\mathbb{Z} (or N0=Z/2\mathfrak N_0=\mathbb{Z}/2) of a compact derived manifold of virtual dimension zero. This chapter records the invariants in that form and then gives the second, independent description of the Euler characteristic and the Lefschetz number as traces.

Degree

Definition 1.1. Let M,NM,N be closed oriented nn-manifolds with NN connected, f ⁣:M→Nf\colon M\to N smooth, and yy a regular value. The degree is deg⁡f=∑x∈f−1(y)sign⁡det⁡dfx\deg f=\sum_{x\in f^{-1}(y)}\operatorname{sign}\det df_x. Without orientations the mod 22 degree is #f−1(y)\#f^{-1}(y) modulo 22.

Theorem 1.2. [classical; Brouwer; Hirsch 5.1.6, 5.1.8]

  1. deg⁡f\deg f does not depend on the regular value, and homotopic maps have the same degree.

  2. If M=∂WM=\partial W with WW compact oriented and ff extends to WW, then deg⁡f=0\deg f=0.

  3. If WW is connected and N=SnN=S^n, the converse of (b) holds: a map ∂W→Sn\partial W\to S^n of degree zero extends to WW.

Proof of (a) and (b). Let F ⁣:W→NF\colon W\to N extend ff, and let yy be a regular value of FF and of ff ([thm:sard]). By [thm:neat], F−1(y)F^{-1}(y) is a compact oriented 11-manifold with boundary f−1(y)f^{-1}(y), and the boundary orientation is the sign in the definition. The signed number of boundary points of a compact oriented 11-manifold is zero. Homotopy invariance is the case W=M×IW=M\times I. Independence of yy follows because any two points of a connected manifold are exchanged by a diffeomorphism isotopic to the identity ([ch:isotopy]). ◻

For example z↦zkz\mapsto z^k on the circle has degree kk, and the antipodal map of SnS^n has degree (−1)n+1(-1)^{n+1} (1.1). The script computes the first both by counting preimages and as a winding-number integral.

Theorem 1.3 (Hopf). [classical; Hopf; Hirsch 5.1.10] Let MM be a closed connected nn-manifold. If MM is oriented, deg⁡ ⁣:[M,Sn]→Z\deg\colon[M,S^n]\to\mathbb{Z} is a bijection. If MM is not orientable, the mod 22 degree [M,Sn]→Z/2[M,S^n]\to\mathbb{Z}/2 is a bijection.

Hopf’s theorem as a classification theorem

There are two ways to see 1.3 as an instance of something general, and they are dual.

The first is geometric. For a closed manifold MM let Fk(M)\mathcal F_k(M) be the set of closed submanifolds A⊂MA\subset M of codimension kk equipped with a framing of the normal bundle ν(A,M)\nu(A,M). For any kk, taking the preimage of a regular value gives a bijection [M,Sk] ≅ Fk(M) / framed cobordism in M×I,[M,S^k]\ \cong\ \mathcal F_k(M)\,/\,\text{framed cobordism in }M\times I, with inverse the collapse M→Th(ν(A,M))≅ΣkA+→SkM\to\mathop{\mathrm{Th}}(\nu(A,M))\cong\Sigma^kA_+\to S^k of [ch:tubular]. This is the Pontryagin–Thom construction inside MM [42]. For k=nk=n a framed submanifold is a finite set of points with signs, and a framed cobordism cancels pairs of opposite sign; if MM is connected and oriented the classes are counted by the total sign, and if MM is non-orientable a point can be carried around an orientation-reversing loop and return with the opposite sign, so only the parity remains.

The second is homotopical. The map Sn→K(Z,n)S^n\to K(\mathbb{Z},n) inducing an isomorphism on πn\pi_n is (n+1)(n+1)-connected, so [M,Sn]→[M,K(Z,n)]=Hn(M;Z)[M,S^n]\to[M,K(\mathbb{Z},n)]=H^n(M;\mathbb{Z}) is a bijection for dim⁡M≤n\dim M\le n; and Hn(M;Z)H^n(M;\mathbb{Z}) is Z\mathbb{Z} or Z/2\mathbb{Z}/2 according to orientability. The two readings are identified by Poincaré duality, Hn(M;Z)≅H0(M;Zw)H^n(M;\mathbb{Z})\cong H_0(M;\mathbb{Z}^{w}) with twisted coefficients, a class in the latter being a finite set of points with local orientations.

Intersection numbers

Theorem 1.4. [classical; Hirsch 5.2.1] Let f ⁣:M→Nf\colon M\to N be a map of oriented manifolds and A⊂NA\subset N a closed oriented submanifold, with MM compact and dim⁡M+dim⁡A=dim⁡N\dim M+\dim A=\dim N. If f⋔Af\pitchfork A, put #(f,A)=∑x∈f−1(A)±1\#(f,A)=\sum_{x\in f^{-1}(A)}\pm1, the sign comparing dfx(TxM)⊕Tf(x)Adf_x(T_xM)\oplus T_{f(x)}A with Tf(x)NT_{f(x)}N. Then homotopic maps transverse to AA have the same intersection number. Without orientations the same holds modulo 22.

So #(f,A)\#(f,A) is defined for every ff, by perturbing. In the language of [ch:transversality] no perturbation is needed to define it: #(f,A)=[M×NhA]∈Ω0=Z,\#(f,A)=\bigl[M\times^h_NA\bigr]\in\Omega_0=\mathbb{Z}, the cobordism class of the derived fibre product, which is compact of virtual dimension zero ([thm:spivak]). Degree is the case A={y}A=\{y\}.

Example 1.5. On the torus R2/Z2\mathbb{R}^2/\mathbb{Z}^2 the closed curve of slope class (a,b)(a,b) and the curve of class (c,d)(c,d), in general position, meet in ∣ad−bc∣|ad-bc| points all of the same sign, and the intersection number is ad−bcad-bc. The script counts the intersection points for several pairs.

The Euler number

Definition 1.6. Let ξ\xi be an oriented vector bundle of rank nn over a closed oriented nn-manifold MM, with zero section ZZ. Its Euler number is χ(ξ)=#(Z,Z)\chi(\xi)=\#(Z,Z), the signed number of zeros of a section transverse to ZZ. The Euler characteristic of MM is χ(M)=χ(TM)\chi(M)=\chi(TM).

Theorem 1.7. [classical; Hopf; Hirsch 5.2.2, 5.2.3, 5.2.5, 5.2.10]

  1. If ξ\xi has a nowhere-zero section then χ(ξ)=0\chi(\xi)=0; if MM is connected the converse holds.

  2. χ(ξ)=0\chi(\xi)=0 if nn is odd.

  3. χ(Sn)=1+(−1)n\chi(S^n)=1+(-1)^n, and χ(M×N)=χ(M)χ(N)\chi(M\times N)=\chi(M)\chi(N).

  4. (Poincaré–Hopf) For a vector field on MM with isolated zeros, the sum of the indices is χ(M)\chi(M).

The Euler number is the class of the derived self-intersection of the zero section, which by [ex:derived](c) is the derived manifold with functions Γ(Λ∙ξ∨)\Gamma(\Lambda^\bullet\xi^\vee). Part (b) is the observation that s↦−ss\mapsto-s is an automorphism of this object that multiplies its orientation by (−1)n(-1)^n. The converse in (a) is an instance of 1.2(c): the zeros of a section can be gathered into a disc by an isotopy, and the section on the boundary sphere then has degree zero and extends without zeros.

Traces

The second description starts from a definition that makes no mention of manifolds ([app:toolkit]). In a symmetric monoidal category an object XX is dualizable if there are X∨X^\vee, a coevaluation η ⁣:1→X⊗X∨\eta\colon\mathbf 1\to X\otimes X^\vee and an evaluation ϵ ⁣:X∨⊗X→1\epsilon\colon X^\vee\otimes X\to\mathbf 1 satisfying the triangle identities. The trace of an endomorphism ff of such an XX is the composite tr(f) ⁣: 1→ η X⊗X∨→ f⊗1 X⊗X∨→ swap X∨⊗X→ ϵ 1,\mathop{\mathrm{tr}}(f)\colon\ \mathbf 1\xrightarrow{\ \eta\ }X\otimes X^\vee\xrightarrow{\ f\otimes1\ }X\otimes X^\vee \xrightarrow{\ \text{swap}\ }X^\vee\otimes X\xrightarrow{\ \epsilon\ }\mathbf 1 , an element of the commutative monoid End⁡(1)\operatorname{End}(\mathbf 1), and χ(X)=tr(idX)\chi(X)=\mathop{\mathrm{tr}}(\mathrm{id}_X).

Proposition 1.8. [classical; Dold–Puppe [16]] tr(fg)=tr(gf)\mathop{\mathrm{tr}}(fg)=\mathop{\mathrm{tr}}(gf); tr(f⊗g)=tr(f)tr(g)\mathop{\mathrm{tr}}(f\otimes g)=\mathop{\mathrm{tr}}(f)\mathop{\mathrm{tr}}(g); and symmetric monoidal functors preserve dualizable objects and traces. In finite-dimensional vector spaces tr\mathop{\mathrm{tr}} is the trace of a matrix. In graded vector spaces with the Koszul sign in the symmetry it is the alternating sum of the traces on the graded pieces.

Theorem 1.9 (Lefschetz–Hopf, after Dold and Puppe). [classical; [16]] Let MM be a closed manifold and f ⁣:M→Mf\colon M\to M a map. The trace of Σ+∞f\Sigma^\infty_+f in the stable homotopy category is an integer, an element of π0S=Z\pi_0\mathbb S=\mathbb{Z}, and:

  1. it equals the Lefschetz number L(f)=∑i(−1)itr(f∗∣Hi(M;Q))L(f)=\sum_i(-1)^i\mathop{\mathrm{tr}}\bigl(f_*\mid H_i(M;\mathbb Q)\bigr);

  2. it equals the intersection number of the graph of ff with the diagonal in M×MM\times M; if the fixed points of ff are non-degenerate, this is ∑f(x)=xsign⁡det⁡(1−dfx)\sum_{f(x)=x}\operatorname{sign}\det(1-df_x).

In particular χ(M)=∑i(−1)idim⁡Hi(M;Q)=#(Δ,Δ)=χ(TM)\chi(M)=\sum_i(-1)^i\dim H_i(M;\mathbb Q)=\#(\Delta,\Delta)=\chi(TM).

Sketch. (i) Rational homology is a symmetric monoidal functor from finite spectra to graded vector spaces, by the Künneth theorem; apply 1.8. (ii) By [thm:atiyah] the coevaluation is the collapse onto a tubular neighbourhood of M⊂Rn+kM\subset\mathbb{R}^{n+k} and the evaluation is the collapse onto a tubular neighbourhood of the diagonal. The composite defining the trace is a map Sn+k→Sn+kS^{n+k}\to S^{n+k} whose preimage of the base point of the last collapse is the set of fixed points of ff, and the local degree at a non-degenerate fixed point is the sign of det⁡(1−dfx)\det(1-df_x). ◻

So the equality of the two definitions of the Euler characteristic, which the source reaches at the end of its treatment of Morse theory (Hirsch 6.3.8), is the statement that a trace can be computed either after applying a monoidal functor or before. The multiplicativity χ(M×N)=χ(M)χ(N)\chi(M\times N)=\chi(M)\chi(N) is tr(f⊗g)=tr(f)tr(g)\mathop{\mathrm{tr}}(f\otimes g)=\mathop{\mathrm{tr}}(f)\mathop{\mathrm{tr}}(g). Additivity over pushouts, χ(A∪B)=χ(A)+χ(B)−χ(A∩B)\chi(A\cup B)=\chi(A)+\chi(B)-\chi(A\cap B), is a property of traces in stable settings [38], and gives the formula χ(M)=∑(−1)kνk\chi(M)=\sum(-1)^k\nu_k in terms of the numbers of critical points of a Morse function ([ch:morse]).

Example 1.10. A matrix A∈GL2(Z)A\in\mathop{\mathrm{GL}}_2(\mathbb{Z}) acts on the torus. On H0,H1,H2H_0,H_1,H_2 its traces are 11, trA\mathop{\mathrm{tr}}A, det⁡A\det A, so L(A)=1−trA+det⁡A=det⁡(1−A)L(A)=1-\mathop{\mathrm{tr}}A+\det A=\det(1-A). The fixed points are the solutions of (A−1)x∈Z2(A-1)x\in\mathbb{Z}^2 modulo Z2\mathbb{Z}^2; there are ∣det⁡(A−1)∣|\det(A-1)| of them and each has index sign⁡det⁡(1−A)\operatorname{sign}\det(1-A). The script enumerates them for several AA.

The diagonal, three ways

The self-intersection of the diagonal has now appeared in three forms.

Proposition 1.11. [reformulation; cf. Hirsch 5.2, 6.3.8; [16]] For a closed manifold MM and f ⁣:M→Mf\colon M\to M, the following integers are equal:

  1. the signed number of fixed points of a generic perturbation of ff;

  2. the class in Ω0\Omega_0 of the derived fixed-point object M×M×MhMM\times^h_{M\times M}M, the fibre product of the graph of ff and the diagonal;

  3. the trace of Σ+∞f\Sigma^\infty_+f.

For f=idf=\mathrm{id} the object in (ii) has functions the differential forms Γ(Λ∙T∗M)\Gamma(\Lambda^\bullet T^*M) in homological degrees, and the integer is χ(M)\chi(M).

Form (i) is the definition in the source; it requires a choice and a proof of independence. Form (ii) requires no choice but lives in DMfd\mathsf{DMfd}. Form (iii) requires no manifold: it makes sense for any dualizable object, and it is the one that generalizes, to parametrized families, to equivariant settings, and to bicategories, where the Lefschetz number acquires a refinement due to Reidemeister [43].

The derived fixed-point object of the identity is the smooth counterpart of the derived loop space of algebraic geometry. That its functions are differential forms is the Hochschild–Kostant–Rosenberg theorem in this setting, and the Euler characteristic is then the “dimension” of MM in the sense in which the dimension of a vector space is the trace of its identity map.

Example 1.12 (Linking numbers). For disjoint oriented closed curves K,L⊂R3K,L\subset\mathbb{R}^3 the linking number is the degree of the Gauss map K×L→S2K\times L\to S^2, (x,y)↦(y−x)/∣y−x∣(x,y)\mapsto(y-x)/|y-x|. It equals the intersection number of LL with any oriented surface bounded by KK, and it is given by Gauss’s integral. For the Hopf link it is ±1\pm1, and the script evaluates the integral numerically.

Notes

The treatment of degree by regular values and one-dimensional cobordisms is Pontryagin’s and is the subject of Milnor’s Topology from the Differentiable Viewpoint. The trace in a symmetric monoidal category and its application to fixed point theory are due to Dold and Puppe; see also [38] for additivity and [43] for an exposition and the bicategorical refinement. Derived loop spaces and their relation to differential forms are due to Ben-Zvi and Nadler and to Toën and Vezzosi in the algebraic setting.

Exercises

Exercise 1.1. Show that z↦zkz\mapsto z^k has degree kk on S1S^1 and that the antipodal map of SnS^n has degree (−1)n+1(-1)^{n+1}. Deduce that every vector field on S2mS^{2m} has a zero.

Exercise 1.2. Verify 1.10: for A∈GL2(Z)A\in\mathop{\mathrm{GL}}_2(\mathbb{Z}) with det⁡(A−1)≠0\det(A-1)\ne0, the induced map of the torus has exactly ∣det⁡(A−1)∣|\det(A-1)| fixed points, each of index sign⁡det⁡(1−A)\operatorname{sign}\det(1-A).

Exercise 1.3. Prove tr(fg)=tr(gf)\mathop{\mathrm{tr}}(fg)=\mathop{\mathrm{tr}}(gf) and tr(f⊗g)=tr(f)tr(g)\mathop{\mathrm{tr}}(f\otimes g)=\mathop{\mathrm{tr}}(f)\mathop{\mathrm{tr}}(g) from the definition, and show that in graded vector spaces with the Koszul sign rule the trace of a degree-preserving endomorphism is the alternating sum of the traces of its components.

Exercise 1.4. Compute χ(Sn)\chi(S^n) from a vector field with two zeros, and again as ∑(−1)idim⁡Hi\sum(-1)^i\dim H_i.

Exercise 1.5. Let KK be the unit circle in the xyxy-plane and LL the circle of radius 11 centred at (1,0,0)(1,0,0) in the xzxz-plane. Compute the linking number as the intersection number of LL with the disc bounded by KK.

Decompositions and Moduli

Morse Theory

A Morse function on a manifold is a generic point of the mapping object [M,R][M,\mathbb{R}], and Hirsch’s sixth chapter shows how much of the manifold such a point determines: a decomposition into handles, a cell structure, inequalities for the Betti numbers, the Euler characteristic. In categorical terms a Morse function presents the homotopy type of MM as an iterated pushout, and presents MM itself as a composite of elementary cobordisms. This chapter states the classical theorems in that form, and then turns to what the source does not treat: the structure on the set of all Morse functions, which is where the presentations become canonical.

Morse functions

Definition 1.1. A critical point pp of f ⁣:M→Rf\colon M\to\mathbb{R} is non-degenerate if the Hessian of ff at pp is a non-degenerate quadratic form; its index is the number of negative eigenvalues. ff is a Morse function if all its critical points are non-degenerate. Its type is (ν0,…,νn)(\nu_0,\dots,\nu_n), with νk\nu_k the number of critical points of index kk.

Theorem 1.2. [classical; Morse; Hirsch 6.1.1, 6.1.2] Near a non-degenerate critical point of index kk there are coordinates in which f=f(p)−x12−⋯−xk2+xk+12+⋯+xn2f=f(p)-x_1^2-\dots-x_k^2+x_{k+1}^2+\dots+x_n^2. Morse functions form an open dense subset of CS∞(M,R)C^\infty_S(M,\mathbb{R}).

Density was derived in [cor:counts](c) from jet transversality: ff is Morse when dfdf is transverse to the zero section of T∗MT^*M.

Example 1.3. On the torus with coordinates (θ,φ)(\theta,\varphi) let f=(R+rcos⁡φ)cos⁡θf=(R+r\cos\varphi)\cos\theta, R>r>0R>r>0, the height function of a torus standing on its edge. The critical points are the four points with θ,φ∈{0,π}\theta,\varphi\in\{0,\pi\}, and the Hessian there is diagonal with entries −(R+rcos⁡φ)cos⁡θ-(R+r\cos\varphi)\cos\theta and −rcos⁡φcos⁡θ-r\cos\varphi\cos\theta. The indices are 2,1,1,02,1,1,0 at (0,0),(0,π),(π,π),(π,0)(0,0),(0,\pi),(\pi,\pi),(\pi,0), with critical values R+r, R−r, −(R−r), −(R+r)R+r,\ R-r,\ -(R-r),\ -(R+r). The type is (1,2,1)(1,2,1).

Regular intervals

Theorem 1.4 (Regular interval). [classical; Hirsch 6.2.2] Let f ⁣:M→[a,b]f\colon M\to[a,b] be smooth on a compact manifold with boundary, with no critical points and f(∂M)={a,b}f(\partial M)=\{a,b\}. Then there is a diffeomorphism M≅f−1(a)×[a,b]M\cong f^{-1}(a)\times[a,b] commuting with the maps to [a,b][a,b].

The proof integrates the vector field gradf/∣gradf∣2\mathop{\mathrm{grad}}f/|\mathop{\mathrm{grad}}f|^2, along which ff increases at unit speed. The theorem is the one-parameter case of a statement about families.

Theorem 1.5 (Ehresmann). [classical; ] A proper submersion p ⁣:E→Bp\colon E\to B is a locally trivial fibre bundle.

In the language of [ch:bundles]: let Md\mathcal M_d be the stack on Cart\mathsf{Cart} whose sections over UU are the proper submersions over UU with dd-dimensional fibres. Ehresmann’s theorem says that Md\mathcal M_d is locally constant in the strongest sense: every section over a Euclidean space is a product. Its shape is ∐[F]BDiff(F)\coprod_{[F]}B\mathop{\mathrm{Diff}}(F), and a proper submersion over BB is classified by a map from BB to this space ([tab:oka]). Compact smooth manifolds have no local moduli, only monodromy. The regular interval theorem is the case of a contractible base.

Passing a critical point

Call a Morse function f ⁣:M→[a,b]f\colon M\to[a,b] on a compact manifold admissible if ∂M=f−1(a)∪f−1(b)\partial M=f^{-1}(a)\cup f^{-1}(b) and a,ba,b are regular values.

Theorem 1.6 (Attaching a cell). [classical; Hirsch 6.3.1–6.3.3] Let f ⁣:M→[a,b]f\colon M\to[a,b] be an admissible Morse function with exactly one critical point, of index kk. Then there is a kk-cell ek⊂M∖f−1(b)e^k\subset M\setminus f^{-1}(b), attached to f−1(a)f^{-1}(a) along its boundary, such that f−1(a)∪ekf^{-1}(a)\cup e^k is a deformation retract of MM. Smoothly, MM is obtained from f−1(a)×[0,1]f^{-1}(a)\times[0,1] by attaching a handle Dk×Dn−kD^k\times D^{n-k} along an embedding of Sk−1×Dn−kS^{k-1}\times D^{n-k} in f−1(a)×{1}f^{-1}(a)\times\{1\}.

For a proper function f ⁣:M→Rf\colon M\to\mathbb{R} bounded below, write Mc=f−1(−∞,c]M^c=f^{-1}(-\infty,c]. The theorem and 1.4 give the following form, which is the one that composes.

Corollary 1.7. [reformulation; cf. Hirsch 6.2.2, 6.3.1] If [c,c′][c,c'] contains no critical value, Mc→Mc′M^c\to M^{c'} is a homotopy equivalence. If [c−ε,c+ε][c-\varepsilon, c+\varepsilon] contains a single critical value, attained at one critical point of index kk, there is a homotopy pushout square of spaces image and hence a cofibre sequence Σ+∞Mc−ε→Σ+∞Mc+ε→Sk\Sigma^\infty_+M^{c-\varepsilon}\to\Sigma^\infty_+M^{c+\varepsilon}\to\mathbb S^k of spectra.

The local statement behind the theorem is about the quadratic form q(x,y)=−∣x∣2+∣y∣2q(x,y)=-|x|^2+|y|^2 on Rk×Rn−k\mathbb{R}^k\times\mathbb{R}^{n-k}: the set {q≤−ε}\{q\le-\varepsilon\} together with the disc {y=0,∣x∣2≤ε}\{y=0,|x|^2\le\varepsilon\} is a deformation retract of {q≤ε}\{q\le\varepsilon\}, by a deformation that shrinks yy and does not move xx (1.4).

Theorem 1.8. [classical; Hirsch 6.4.1–6.4.3, 6.2.4] A compact manifold with a Morse function of type (ν0,…,νn)(\nu_0,\dots,\nu_n) has the homotopy type of a finite CW complex with νk\nu_k cells of dimension kk. Every nn-manifold has the homotopy type of a CW complex of dimension at most nn. A closed manifold with a Morse function having exactly two critical points is homeomorphic to a sphere.

So the shape of a compact manifold is a finite space, a compact object of S\mathcal{S}, and its suspension spectrum is a finite spectrum. Finite spectra are dualizable; this is the abstract half of [thm:atiyah], the concrete half being the identification of the dual.

Inequalities

Theorem 1.9 (Morse inequalities). [classical; Morse; Hirsch 6.3.4–6.3.8] Let ff be a Morse function of type (νk)(\nu_k) on a closed manifold MM, and βk=dim⁡Hk(M;F)\beta_k=\dim H_k(M;F) for a field FF. Then ∑kνktk−∑kβktk=(1+t) Q(t)\sum_k\nu_kt^k-\sum_k\beta_kt^k=(1+t)\,Q(t) for a polynomial QQ with non-negative integer coefficients. In particular νk≥βk\nu_k\ge\beta_k, and χ(M)=∑(−1)kνk=∑(−1)kβk\chi(M)=\sum(-1)^k\nu_k=\sum(-1)^k\beta_k; and χ(M)=0\chi(M)=0 if dim⁡M\dim M is odd.

Proof. By 1.7, crossing a critical point of index kk changes the homology of the sublevel set in one of two ways: either βk\beta_k increases by one, or βk−1\beta_{k-1} decreases by one. Let qk−1q_{k-1} be the number of critical points of index kk of the second kind. Each of them kills a class created earlier by a critical point of index k−1k-1 of the first kind. So νk=βk+qk+qk−1\nu_k=\beta_k+q_k+q_{k-1}, which is the identity with Q=∑qktkQ=\sum q_kt^k. Setting t=−1t=-1 gives the Euler characteristic, and replacing ff by −f-f exchanges νk\nu_k and νn−k\nu_{n-k}. ◻

The equality χ(TM)=∑(−1)kνk\chi(TM)=\sum(-1)^k\nu_k also follows from [thm:euler](d) applied to the gradient, whose zero at a critical point of index kk has index (−1)k(-1)^k. For the torus of 1.3, 1+2t+t21+2t+t^2 is the Poincaré polynomial and Q=0Q=0. On RPn\mathbb{RP}^n there is a Morse function with one critical point of each index, which is optimal for F=Z/2F=\mathbb{Z}/2 and not for F=QF=\mathbb Q (1.3).

The sublevel functor

The proof just given used more than the statement records. The assignment c↦Mcc\mapsto M^c is a functor from the ordered set R\mathbb{R} to manifolds with boundary and embeddings, and its composite with the shape, Sf ⁣:(R,≤)→S,c↦Π(Mc),S_f\colon(\mathbb{R},\le)\to\mathcal{S},\qquad c\mapsto\Pi(M^c), is the sublevel functor of ff.

Proposition 1.10. [reformulation; cf. 1.7; persistence, [51]] For a Morse function ff on a closed manifold, SfS_f is locally constant on the complement of the critical values, and at a critical value it changes by a pushout along a sphere for each critical point. The functor Hk(Sf;F) ⁣:(R,≤)→VectFH_k(S_f;F)\colon(\mathbb{R},\le)\to\mathsf{Vect}_F is a direct sum of interval modules; the number of infinite intervals is βk\beta_k, and the number of finite intervals is the coefficient qkq_k of QQ in 1.9.

So the polynomial QQ counts pairs of critical points that cancel in homology, and the multiset of intervals, the barcode, refines the Morse inequalities to a statement about which pairs and at which levels. The barcode is a continuous function of ff in the uniform norm, which the type is not. The script computes the degree-zero barcode of a function on a circle by a union–find pass and checks that the number of finite intervals is ν0−1\nu_0-1.

Flow categories

A Morse function with a generic metric determines more than a filtration.

Theorem 1.11. [literature; Cohen–Jones–Segal [13]] Let ff be a Morse function on a closed manifold MM with a Morse–Smale gradient. There is a topological category Cf\mathcal C_f whose objects are the critical points and whose morphism space from pp to qq is the compactified space of gradient flow lines from pp to qq, a manifold with corners of dimension ind⁡p−ind⁡q−1\operatorname{ind}p-\operatorname{ind}q-1, with composition the inclusion of broken flow lines. Its classifying space is homeomorphic to MM.

The Morse complex is what remains of Cf\mathcal C_f when one keeps only the zero-dimensional morphism spaces and counts their points; the higher-dimensional ones carry the attaching maps of the cells. The transversality condition on the metric is [thm:transversality] applied to the stable and unstable manifolds, and the corner structure is where manifolds with corners ([ch:manifolds]) become unavoidable.

Families of functions

The set of Morse functions is open and dense in [M,R][M,\mathbb{R}] but it is not a good parameter space: two Morse functions of different types cannot be joined by a path of Morse functions. Jet transversality in one parameter says what a generic path looks like.

Theorem 1.12 (Cerf). [classical; Cerf [12]] Any two Morse functions on a closed manifold are joined by a path ftf_t that is Morse except at finitely many times, at each of which exactly one of the following occurs: two critical points have the same value; or there is a single degenerate critical point, with local form x13−(t−t0)x1−x22−⋯−xk+12+xk+22+⋯+xn2x_1^3-(t-t_0)x_1-x_2^2-\dots-x_{k+1}^2+x_{k+2}^2+\dots+x_n^2, at which a pair of critical points of indices kk and k+1k+1 is born or dies.

The count behind the theorem: in the space of 22-jets, the jets of degenerate critical points have codimension n+1n+1, so a one-parameter family, of total dimension n+1n+1, meets them at isolated points. On the side of handle decompositions the two events are the reordering or sliding of handles and the cancellation of a pair. So the theorem says that any two handle decompositions of a manifold are related by these moves. It is a presentation by generators and relations, one categorical level up from 1.6.

The space of functions with only Morse and birth–death singularities is connected by Cerf’s theorem, but it is still not contractible. Igusa found the decoration that makes it so.

Theorem 1.13. [literature; Igusa [28]; Lurie [36]; Eliashberg–Mishachev [19]] The space of framed functions on a compact manifold MM (functions with only Morse and birth–death singularities, with a framing of the negative eigenspace of the Hessian at each critical point) is contractible.

Igusa proved that the space is (dim⁡M−1)(\dim M-1)-connected; contractibility is due to Lurie and, by geometric methods, to Eliashberg and Mishachev. By [prin:choices] the theorem says that a framed function is a choice of the harmless kind, so that the handle decomposition it induces is as good as canonical. That is the form in which Morse theory enters the proof of the cobordism hypothesis ([ch:cobordism]).

Notes

Milnor’s Morse Theory and Lectures on the h-Cobordism Theorem are the classical accounts; Hirsch’s chapter follows the same lines with more attention to the differentiability class. Persistence modules and their decomposition are in [51] and in work of Crawley-Boevey. The flow category is the starting point of Floer homotopy theory. Cerf’s theorem is the one-parameter beginning of a hierarchy governed by the classification of singularities of functions; Igusa’s work on higher parameters and the notion of framed function are from the 1980s.

Exercises

Exercise 1.1. Verify 1.3.

Exercise 1.2. Deduce from the polynomial identity of 1.9 the inequalities νk−νk−1+⋯±ν0≥βk−βk−1+⋯±β0\nu_k-\nu_{k-1}+\dots\pm\nu_0\ge\beta_k-\beta_{k-1}+\dots\pm\beta_0 for every kk. Show that a Morse function on a closed connected surface of genus gg has at least 2g+22g+2 critical points.

Exercise 1.3. Let λ0<⋯<λn\lambda_0<\dots<\lambda_n and f([x])=∑λixi2/∑xi2f([x])=\sum\lambda_ix_i^2/\sum x_i^2 on RPn\mathbb{RP}^n. Show that ff is a Morse function with one critical point of each index 0,…,n0,\dots,n. Find QQ for F=Z/2F=\mathbb{Z}/2 and for F=QF=\mathbb Q.

Exercise 1.4. For q(x,y)=−∣x∣2+∣y∣2q(x,y)=-|x|^2+|y|^2 on Rk×Rn−k\mathbb{R}^k\times\mathbb{R}^{n-k} and ε>0\varepsilon>0, construct a deformation retraction of {q≤ε}\{q\le\varepsilon\} onto {q≤−ε}∪{y=0, ∣x∣2≤ε}\{q\le-\varepsilon\}\cup\{y=0,\ |x|^2\le\varepsilon\}.

Exercise 1.5. For ft(x)=x3−txf_t(x)=x^3-tx on R\mathbb{R}, find the critical points and their indices for each tt, and describe the sublevel functor of ftf_t restricted to a large interval [−L,L][-L,L] before and after t=0t=0.

Cobordism

In the source, cobordism occupies a short chapter: the groups are defined, the Pontryagin–Thom construction is carried out, and Thom’s theorem identifying cobordism groups with homotopy groups of Thom spaces is proved in the range the earlier chapters allow. It is the place where the three tools of the book (classifying maps, tubular neighbourhoods, transversality) are used together. It is also the place from which the higher-categorical view of manifolds grew. This chapter states the classical results and then the two modern ones that extend them: the homotopy type of the cobordism category, and the cobordism hypothesis.

Cobordism groups

Definition 1.1. Closed nn-manifolds M0,M1M_0,M_1 are cobordant if there is a compact (n+1)(n+1)-manifold WW with ∂W≅M0⊔M1\partial W\cong M_0\sqcup M_1. The set of classes is Nn\mathfrak N_n. For oriented manifolds one requires WW oriented with ∂W≅M1⊔−M0\partial W\cong M_1\sqcup-M_0, and the set of classes is Ωn\Omega_n.

Theorem 1.2. [classical; Thom; Hirsch 7.1.1] Disjoint union makes Nn\mathfrak N_n and Ωn\Omega_n abelian groups, and product makes N∗\mathfrak N_* and Ω∗\Omega_* graded rings, commutative in the graded sense. Every element of N∗\mathfrak N_* has order 22.

Transitivity of the relation uses [thm:gluing]; the inverse of [M][M] is [−M][-M], with M×IM\times I as the null-cobordism of M⊔−MM\sqcup-M. In low dimensions N0=Z/2\mathfrak N_0=\mathbb{Z}/2, Ω0=Z\Omega_0=\mathbb{Z}, N1=Ω1=0\mathfrak N_1=\Omega_1=0, and N2=Z/2\mathfrak N_2=\mathbb{Z}/2, Ω2=0\Omega_2=0 by the classification of surfaces ([ch:surfaces]).

The Pontryagin–Thom theorem

Let M⊂Rn+kM\subset\mathbb{R}^{n+k} be a closed nn-manifold. The collapse of [ch:tubular], followed by the map of Thom spaces induced by the classifying map of the normal bundle ([ch:bundles]), is a map Sn+k⟶Th(ν)⟶Th(γk),S^{n+k}\longrightarrow\mathop{\mathrm{Th}}(\nu)\longrightarrow\mathop{\mathrm{Th}}(\gamma_k), where γk\gamma_k is the tautological bundle over Grk(Rs)\mathop{\mathrm{Gr}}_k(\mathbb{R}^s) for ss large.

Theorem 1.3 (Pontryagin–Thom). [classical; Pontryagin [42]; Thom [49]; Hirsch 7.2.3] For f ⁣:Sn+k→Th(γk)f\colon S^{n+k}\to\mathop{\mathrm{Th}}(\gamma_k) transverse to the zero section, f↦[f−1(Grk(Rs))]f\mapsto[f^{-1}(\mathop{\mathrm{Gr}}_k(\mathbb{R}^s))] defines a homomorphism πn+k(Th(γk→Grk(Rs)))→Nn\pi_{n+k}\bigl(\mathop{\mathrm{Th}}(\gamma_k\to\mathop{\mathrm{Gr}}_k(\mathbb{R}^s))\bigr)\to\mathfrak N_n that is surjective if k>nk>n and s≥k+ns\ge k+n and bijective if k>n+1k>n+1 and s≥k+n+1s\ge k+n+1; the construction above is its inverse. Stably, Nn≅πn(MO),Ωn≅πn(MSO),Ωnfr≅πn(S),\mathfrak N_n\cong\pi_n(MO),\qquad\Omega_n\cong\pi_n(MSO),\qquad\Omega^{\mathrm{fr}}_n\cong\pi_n(\mathbb S), and in general Ωnθ≅πn(Mθ)\Omega^\theta_n\cong\pi_n(M\theta) for a stable tangential structure θ ⁣:B→BO\theta\colon B\to B\mathrm{O}, where MθM\theta is the Thom spectrum of θ\theta.

Structure of the proof. Well-definedness: the embedding and the tubular neighbourhood are contractible choices (Theorem 1.7, Theorem 1.3), and a cobordism embedded in Rn+k×I\mathbb{R}^{n+k}\times I collapses to a homotopy. Inverse: a map f ⁣:Sn+k→Th(γk)f\colon S^{n+k}\to\mathop{\mathrm{Th}}(\gamma_k) can be made transverse to the zero section ([thm:transversality]), and M=f−1(Grk)M=f^{-1}(\mathop{\mathrm{Gr}}_k) is a closed nn-manifold with normal bundle pulled back from γk\gamma_k; a homotopy made transverse gives a cobordism ([lem:fibre]). The dimension restrictions come from Theorem 1.14, Theorem 1.4. ◻

The theorem is [prin:generic] for the cospan Sn+k→Th(γk)←GrkS^{n+k}\to\mathop{\mathrm{Th}}(\gamma_k)\leftarrow\mathop{\mathrm{Gr}}_k: every map to a Thom space has a derived preimage of the zero section, which has a cobordism class, and the class determines the map up to homotopy.

Thom’s computation

Theorem 1.4 (Thom). [classical; [49]] MOMO is a wedge of suspensions of the Eilenberg–MacLane spectrum HZ/2H\mathbb{Z}/2, and N∗≅Z/2 [xi: i≥2, i≠2j−1],∣xi∣=i,\mathfrak N_*\cong\mathbb{Z}/2\,[x_i:\ i\ge2,\ i\ne2^j-1],\qquad|x_i|=i, with x2ix_{2i} the class of RP2i\mathbb{RP}^{2i}. A closed manifold is a boundary if and only if all its Stiefel–Whitney numbers vanish.

So dim⁡Z/2Nn\dim_{\mathbb{Z}/2}\mathfrak N_n is the number of partitions of nn into parts not of the form 2j−12^j-1: for n=0,…,10n=0,\dots,10 these are 1,0,1,0,2,1,3,1,5,3,81,0,1,0,2,1,3,1,5,3,8. For a partition n=i1+⋯+irn=i_1+\dots+i_r the Stiefel–Whitney number is ⟨wi1⋯wir(TM),[M]⟩∈Z/2\langle w_{i_1}\cdots w_{i_r}(TM),[M]\rangle\in\mathbb{Z}/2. It vanishes on a boundary M=∂WM=\partial W because TW∣M=TM⊕RTW|_M=TM\oplus\mathbb{R}, so the class being evaluated is restricted from WW, and the fundamental class of a boundary maps to zero in the homology of WW.

Example 1.5. The total Stiefel–Whitney class of RPn\mathbb{RP}^n is (1+a)n+1(1+a)^{n+1} with aa the generator of H1H^1. For RP2\mathbb{RP}^2, w=1+a+a2w=1+a+a^2, and w2[RP2]=w12[RP2]=1w_2[\mathbb{RP}^2]=w_1^2[\mathbb{RP}^2]=1: not a boundary. For RP3\mathbb{RP}^3, w=(1+a)4=1w=(1+a)^4=1: all numbers vanish and RP3\mathbb{RP}^3 bounds. In dimension 44 the numbers (w14,w12w2,w22,w1w3,w4)(w_1^4,w_1^2w_2,w_2^2,w_1w_3,w_4) are (1,0,0,0,1)(1,0,0,0,1) for RP4\mathbb{RP}^4 and (0,0,1,0,1)(0,0,1,0,1) for RP2×RP2\mathbb{RP}^2\times\mathbb{RP}^2. The two manifolds are independent in N4=(Z/2)2\mathfrak N_4=(\mathbb{Z}/2)^2; they are x4x_4 and x22x_2^2. The script computes these numbers and the ranks above.

Theorem 1.6. [classical; Thom; Hirsch 7.2.4] Ωn\Omega_n is finite if nn is not divisible by 44, and Ω4k\Omega_{4k} has rank equal to the number of partitions of kk. Rationally, Ω∗⊗Q=Q[CP2,CP4,… ]\Omega_*\otimes\mathbb Q=\mathbb Q[\mathbb{CP}^2,\mathbb{CP}^4,\dots].

Thom spectra as colimits

The spectra MOMO, MSOMSO, MθM\theta were defined by Thom spaces of bundles over Grassmannians. There is a description in which no bundle appears.

Theorem 1.7. [literature; Ando–Blumberg–Gepner–Hopkins–Rezk [1]] Let Pic⁡(S)\operatorname{Pic}(\mathbb S) be the ∞\infty-groupoid of invertible spectra. A stable vector bundle on a space XX determines a map X→BO→JPic⁡(S)X\to B\mathrm{O}\xrightarrow{J}\operatorname{Pic}(\mathbb S), sending a point to the one-point compactification of the fibre, and its Thom spectrum is the colimit of the composite functor X→Pic⁡(S)⊂SpX\to\operatorname{Pic}(\mathbb S)\subset\mathsf{Sp}. In particular MO=colim(BO→Sp)MO=\mathop{\mathrm{colim}}(B\mathrm{O}\to\mathsf{Sp}) is the homotopy quotient of the sphere spectrum by the stable orthogonal group, and it is an E∞E_\infty-ring because JJ is an infinite loop map. For an E∞E_\infty-ring RR, E∞E_\infty-ring maps MO→RMO\to R correspond to null-homotopies of BO→BGL1(R)B\mathrm{O}\to B\mathop{\mathrm{GL}}_1(R) as infinite loop maps.

For a trivial bundle the diagram is constant and the colimit is Σ+∞X\Sigma^\infty_+X, which is [prop:thom]. The description makes the Thom isomorphism a formal consequence of an orientation, a trivialization of the diagram after smashing with RR. And it gives a reading of 1.3: framed cobordism is the sphere spectrum, and unoriented cobordism is what is left when one divides by all changes of framing.

Cobordism categories

The groups Nn\mathfrak N_n remember that a cobordism exists. The cobordism category remembers the cobordisms.

Definition 1.8. Cobd\mathsf{Cob}_d is the ∞\infty-category whose objects are closed (d−1)(d-1)-manifolds and whose space of morphisms from M0M_0 to M1M_1 is ∐[W]BDiff(W rel ∂W)\coprod_{[W]}B\mathop{\mathrm{Diff}}(W\text{ rel }\partial W), the union over diffeomorphism classes of cobordisms. Composition is gluing ([thm:gluing]). Disjoint union makes it symmetric monoidal. With a tangential structure θ ⁣:B→BO(d)\theta\colon B\to B\mathrm{O}(d) on the cobordisms one writes Cobdθ\mathsf{Cob}^\theta_d.

A concrete model takes objects to be closed submanifolds of R∞\mathbb{R}^\infty and morphisms to be submanifolds of [a0,a1]×R∞[a_0,a_1]\times\mathbb{R}^\infty that are products near the ends; by [thm:embcontract] the embeddings add nothing, and by [thm:collar] neither do the product conditions. The set of components of the classifying space BCobdB\mathsf{Cob}_d is the set of objects modulo the relation generated by morphisms, which is Nd−1\mathfrak N_{d-1}.

Theorem 1.9 (Galatius–Madsen–Tillmann–Weiss). [literature; [22]] Let MTO(d)MTO(d) be the Thom spectrum of the virtual bundle −γd-\gamma_d over BO(d)B\mathrm{O}(d). Then BCobd ≃ Ω∞−1MTO(d),B\mathsf{Cob}_d\ \simeq\ \Omega^{\infty-1}MTO(d), and likewise BCobdθ≃Ω∞−1MTθ(d)B\mathsf{Cob}^\theta_d\simeq\Omega^{\infty-1}MT\theta(d) for every tangential structure θ\theta.

On π0\pi_0 this is a form of 1.3. The proof is an application of the concordance classification ([thm:concordance]): the cobordism category is the shape of a sheaf whose sections over UU are families, parametrized by UU, of dd-dimensional submanifolds of R×R∞\mathbb{R}\times\mathbb{R}^\infty that are proper over R\mathbb{R}, and the shape of that sheaf is computed by a parametrized Pontryagin–Thom construction. So the theorem is the Pontryagin–Thom theorem for the whole category of cobordisms and not only for its set of components. For oriented surfaces, combined with Harer’s stability theorem, it gives the Madsen–Weiss theorem ([thm:MW]).

Field theories and the cobordism hypothesis

Definition 1.10. A dd-dimensional topological field theory with values in a symmetric monoidal ∞\infty-category C\mathcal C is a symmetric monoidal functor Z ⁣:Cobd→CZ\colon\mathsf{Cob}_d\to\mathcal C.

An invertible theory, one whose values are invertible objects and morphisms, factors through the groupoid completion of Cobd\mathsf{Cob}_d, so by 1.9 invertible theories are classified by maps out of the spectrum ΣMTO(d)\Sigma MTO(d). For general theories one needs to know Cobd\mathsf{Cob}_d itself, by generators and relations. Morse theory provides them: by [thm:handle] every cobordism is a composite of handle attachments, and by Theorem 1.12, Theorem 1.13 the relations among such composites are known and the choices involved are contractible. The cleanest statement is obtained after extending the category downwards, allowing manifolds with corners of all codimensions, to a symmetric monoidal (∞,d)(\infty,d)-category Bordd\mathsf{Bord}_d.

Theorem 1.11 (Cobordism hypothesis). [literature; conjectured by Baez–Dolan [6]; Lurie [36]; see also [24]] Let C\mathcal C be a symmetric monoidal (∞,d)(\infty,d)-category. Evaluation at the point is an equivalence between symmetric monoidal functors Borddfr→C\mathsf{Bord}^{\mathrm{fr}}_d\to\mathcal C, from the framed bordism category, and the ∞\infty-groupoid of fully dualizable objects of C\mathcal C.

Lurie’s paper gives a detailed outline of a proof; the case d=2d=2 is worked out by generators and relations in [47], and a proof in a geometric formulation has been given by Grady and Pavlov. For d=1d=1 the statement is that the oriented one-dimensional cobordism category is the free symmetric monoidal ∞\infty-category on a dualizable object. The value of the theory on the circle is then the trace of the identity, χ(X)\chi(X) in the sense of [ch:degree]; for C=Sp\mathcal C=\mathsf{Sp} and X=Σ+∞MX=\Sigma^\infty_+M it is the Euler characteristic of MM. The Euler characteristic of a closed manifold is the partition function, on the circle, of the one-dimensional field theory it generates.

Derived cobordism

By [thm:spivak] the cobordism ring of compact derived manifolds is N∗\mathfrak N_* again. The two enlargements of this chapter are therefore complementary. Passing from the groups to the category Cobd\mathsf{Cob}_d adds information: the homotopy types of diffeomorphism groups. Passing from manifolds to derived manifolds adds none at the level of cobordism classes, and that is its use: fibre products, fixed-point sets and zero loci have classes without being put in general position first.

Notes

Stong’s Notes on Cobordism Theory is the standard reference for the classical computations. The identification of Thom spectra with colimits is from [1]. Cobordism categories as topological categories go back to Segal’s definition of conformal field theory and to Tillmann; [22] determined their homotopy type. The cobordism hypothesis was formulated in [6]; besides [36], see Ayala and Francis for an approach through factorization homology.

Exercises

Exercise 1.1. Let WW be a compact odd-dimensional manifold with boundary. Show that χ(∂W)=2χ(W)\chi(\partial W)=2\chi(W). Deduce that the Euler characteristic modulo 22 is a ring homomorphism N∗→Z/2\mathfrak N_*\to\mathbb{Z}/2, and that RP2m\mathbb{RP}^{2m} is not a boundary.

Exercise 1.2. Compute dim⁡Nn\dim\mathfrak N_n for n≤10n\le10 from 1.4.

Exercise 1.3. Verify the Stiefel–Whitney numbers in 1.5.

Exercise 1.4. Show that N1=0\mathfrak N_1=0 and that the Klein bottle is a boundary. Using the classification of surfaces, conclude that N2=Z/2\mathfrak N_2=\mathbb{Z}/2, generated by RP2\mathbb{RP}^2.

Exercise 1.5. Show directly that the Pontryagin–Thom collapse of a single point of Rk\mathbb{R}^k with the standard framing is a map Sk→SkS^k\to S^k of degree 11, and that reversing one framing vector gives degree −1-1. Conclude Ω0fr≅Z\Omega^{\mathrm{fr}}_0\cong\mathbb{Z}.

Isotopy

Hirsch’s eighth chapter is about paths in spaces of embeddings: an isotopy of a compact submanifold extends to an isotopy of the ambient manifold; embeddings of discs are unique up to isotopy; manifolds glued by isotopic maps are diffeomorphic. Each statement is about π0\pi_0 or about lifting paths. Each has a form that concerns the whole homotopy type of a space of embeddings, and those forms are where the subject went after 1976. This chapter states them and ends with the functor calculus that organizes spaces of embeddings in high codimension. The companion volume uses this chapter more than any other.

Isotopies and vector fields

Definition 1.1. An isotopy of embeddings of VV in MM is a plot I→Emb(V,M)I\to\mathop{\mathrm{Emb}}(V,M): a smooth map F ⁣:V×I→MF\colon V\times I\to M with every FtF_t an embedding. A diffeotopy of MM is an isotopy F ⁣:M×I→MF\colon M\times I\to M of diffeomorphisms with F0=idF_0=\mathrm{id}. Embeddings are isotopic if joined by an isotopy, and ambient isotopic if one is carried to the other by the end of a diffeotopy.

So isotopy is concordance of points of the smooth set Emb(V,M)\mathop{\mathrm{Emb}}(V,M) in the sense of [def:shape]. (The word concordance is used differently in the classical literature on embeddings, for embeddings of V×IV\times I in M×IM\times I that need not preserve levels; we avoid that usage.)

Theorem 1.2. [classical; Hirsch 8.1.1] A time-dependent vector field GG on MM of bounded velocity, for instance one with compact support, generates a unique diffeotopy FF with ∂tFt(x)=G(Ft(x),t)\partial_tF_t(x)=G(F_t(x),t). Every diffeotopy with compact support arises in this way from its velocity field.

So paths in Diffc(M)\mathop{\mathrm{Diff}}_c(M) starting at the identity are the same thing as paths in the vector space Xc(M)\mathfrak X_c(M) of compactly supported vector fields, which is the statement that Diffc(M)\mathop{\mathrm{Diff}}_c(M) is a Lie group with that Lie algebra [31]. The hypothesis on the velocity cannot be dropped: the field x2∂xx^2\partial_x on R\mathbb{R} has solutions x0/(1−x0t)x_0/(1-x_0t), which leave every compact set in finite time.

Isotopy extension

Theorem 1.3 (Isotopy extension). [classical; Thom, Palais, Cerf; Hirsch 8.1.3] Let V⊂MV\subset M be a compact submanifold and F ⁣:V×I→MF\colon V\times I\to M an isotopy with image in M∖∂MM\setminus\partial M. Then FF extends to a diffeotopy of MM with compact support.

The proof extends the velocity field of FF, which is defined along the track of the isotopy in M×IM\times I, to a compactly supported field on M×IM\times I using a tubular neighbourhood of the track, and integrates. The extension is a contractible choice. Carrying parameters through the argument gives the stronger statement.

Theorem 1.4 (Restriction is a fibration). [classical; Palais [41]; Cerf; Lima] Let V⊂MV\subset M be a compact submanifold. The restriction map Diffc(M)→Emb(V,M)\mathop{\mathrm{Diff}}_c(M)\to\mathop{\mathrm{Emb}}(V,M) is a locally trivial fibre bundle with fibre the group Diffc(M rel V)\mathop{\mathrm{Diff}}_c(M\text{ rel }V) of diffeomorphisms fixing VV. More generally, for compact V⊂V′V\subset V' the restriction Emb(V′,M)→Emb(V,M)\mathop{\mathrm{Emb}}(V',M)\to\mathop{\mathrm{Emb}}(V,M) is a fibration.

is the path-lifting property of this map; 1.4 says that all homotopies lift. The consequence is a long exact sequence of homotopy groups. For VV a point it reads ⋯→π1Diff(M)→π1(M)→π0Diff(M rel x)→π0Diff(M)→π0(M),\cdots\to\pi_1\mathop{\mathrm{Diff}}(M)\to\pi_1(M)\to\pi_0\mathop{\mathrm{Diff}}(M\text{ rel }x)\to\pi_0\mathop{\mathrm{Diff}}(M)\to\pi_0(M), and the middle map sends a loop to the diffeomorphism that pushes the base point around it. For compact VV the quotient of Emb(V,M)\mathop{\mathrm{Emb}}(V,M) by Diff(V)\mathop{\mathrm{Diff}}(V) is the space of submanifolds of MM diffeomorphic to VV, and the theorem says that the group Diffc(M)\mathop{\mathrm{Diff}}_c(M) acts on it with orbits that are unions of components.

Compactness is essential. Let Ft ⁣:R→R3F_t\colon\mathbb{R}\to\mathbb{R}^3 be a long knot whose knotted part sits at distance 1/(1−t)1/(1-t) along the line, for t<1t<1, and let F1F_1 be the straight line. This is an isotopy in the sense of 1.1, smooth because on every compact set it is eventually constant, from a knotted line to an unknotted one; it cannot be ambient (1.4). In the language of [ch:function], it is a plot of the mapping object that is not a path in the strong topology ([prop:strongpath]).

Embeddings of discs

Theorem 1.5 (Discs). [classical; Cerf, Palais; Hirsch 8.3.1 on π0\pi_0] For an nn-manifold MM, the map Emb(Dn,M)→Fr(TM)\mathop{\mathrm{Emb}}(D^n,M)\to\mathop{\mathrm{Fr}}(TM), f↦(f(0),df0)f\mapsto(f(0),df_0), to the frame bundle is a homotopy equivalence. More generally Emb(Dk,M)\mathop{\mathrm{Emb}}(D^k,M) is equivalent to the bundle of kk-frames in TMTM, and Emb(∐jDn,M)\mathop{\mathrm{Emb}}(\coprod_jD^n,M) to the pullback of Fr(TM)j\mathop{\mathrm{Fr}}(TM)^j to the configuration space Confj(M)\mathop{\mathrm{Conf}}_j(M).

The proof is the Alexander-type shrinking of [thm:tubcontract]: ft(x)=f(tx)f_t(x)=f(tx) rescaled deforms an embedding to its linear part (1.3). On components this is Hirsch’s theorem that two embeddings of DkD^k in a connected MM are isotopic, provided, when k=nk=n and MM is orientable, they induce the same orientation: the bundle of kk-frames is connected for k<nk<n, and Fr(TM)\mathop{\mathrm{Fr}}(TM) has one or two components according as MM is non-orientable or orientable. With [thm:gluing] it makes the connected sum of connected oriented manifolds well defined.

For M=RnM=\mathbb{R}^n the theorem identifies the spaces of embeddings of several discs in one: Emb(∐jRn,Rn) ≃ Confj(Rn)×O(n)j.\mathop{\mathrm{Emb}}\bigl(\textstyle\coprod_j\mathbb{R}^n,\mathbb{R}^n\bigr)\ \simeq\ \mathop{\mathrm{Conf}}_j(\mathbb{R}^n)\times\mathrm{O}(n)^j . These spaces, with composition of embeddings, form the framed little nn-discs operad; restricting to embeddings that are rectilinear, or equivalently equipped with a trivialization of the derivative, gives the operad EnE_n. Write Discn\mathsf{Disc}_n for the symmetric monoidal ∞\infty-category of finite disjoint unions of copies of Rn\mathbb{R}^n and their embeddings. Every nn-manifold MM defines a presheaf EM=Emb(−,M)E_M=\mathop{\mathrm{Emb}}(-,M) on it, and 1.5 says that this presheaf is assembled from the configuration spaces of MM and its tangent bundle.

Diffeomorphisms of spheres

Theorem 1.6. [classical; Hirsch 8.3.3; Smale; Hatcher; Cerf; Kervaire–Milnor] The inclusion O(n+1)→Diff(Sn)\mathrm{O}(n+1)\to\mathop{\mathrm{Diff}}(S^n) is a homotopy equivalence for n=1,2,3n=1,2,3. For n≥5n\ge5 the group π0Diff+(Sn)\pi_0\mathop{\mathrm{Diff}}^+(S^n) is isomorphic to the group Θn+1\Theta_{n+1} of exotic (n+1)(n+1)-spheres, and is in general not trivial.

Hirsch proves the statement on π0\pi_0 for n=1n=1: every diffeomorphism of the circle is isotopic to the identity or to a reflection. The case n=2n=2 is Smale’s, n=3n=3 is Hatcher’s, and for n=4n=4 the inclusion is known not to be an equivalence. The contrast with 1.5 is instructive. Embeddings of a disc into a manifold are governed by linear algebra in every dimension; diffeomorphisms of a disc fixing the boundary are not, and the difference is the subject of pseudoisotopy theory.

Immersions

For immersions the passage from differential conditions to homotopy theory is complete.

Theorem 1.7 (Smale–Hirsch). [classical; Smale [46]; Hirsch [26]] Let dim⁡V<dim⁡M\dim V<\dim M, or dim⁡V=dim⁡M\dim V=\dim M with VV having no closed component. Then the differential Imm(V,M)⟶Mono⁡(TV,TM)\mathop{\mathrm{Imm}}(V,M)\longrightarrow\operatorname{Mono}(TV,TM) to the space of fibrewise injective bundle maps is a weak homotopy equivalence.

In the diagram at the end of [ch:function], this says that for the open relation R⊂J1(V,M)\mathcal R\subset J^1(V,M) of injective 11-jets the inclusion of holonomic sections into all sections induces an equivalence of shapes. Gromov’s formulation explains why. The presheaf U↦Imm(U,M)U\mapsto\mathop{\mathrm{Imm}}(U,M) on the open subsets of VV takes unions of open sets along handles to homotopy pullbacks, by a restriction theorem like 1.4 that holds for immersions without any compactness hypothesis on the complement; so it is a homotopy sheaf, determined by its values on discs; and on a disc Imm(Dm,M)≃Emb(Dm,M)\mathop{\mathrm{Imm}}(D^m,M)\simeq\mathop{\mathrm{Emb}}(D^m,M) is the bundle of mm-frames by 1.5. A homotopy sheaf with those values is the sheaf of sections of a bundle, namely Mono⁡(TV,TM)\operatorname{Mono}(TV,TM).

Corollary 1.8. [classical; Whitney–Graustein; Smale] Regular homotopy classes of immersions S1→R2S^1\to\mathbb{R}^2 are classified by the rotation number, an integer. All immersions S2→R3S^2\to\mathbb{R}^3 are regularly homotopic; in particular the sphere can be turned inside out.

Proof. Up to homotopy Mono⁡(TS1,TR2)\operatorname{Mono}(TS^1,T\mathbb{R}^2) is the free loop space of the circle of unit vectors, whose components are indexed by degree. For the sphere the relevant group is π2\pi_2 of the space of 22-frames in R3\mathbb{R}^3, which is π2(SO(3))=0\pi_2(\mathrm{SO}(3))=0. ◻

The rotation numbers of the circle, the figure eight and the limaçon r=1+2cos⁡θr=1+2\cos\theta are 11, 00 and 22; the script computes them by accumulating the angle of the tangent.

Embeddings, codimension, and calculus

Embeddings do not satisfy an h-principle. Whether two embeddings are isotopic depends on the codimension, and the two extremes are classical.

Theorem 1.9. [classical; Whitney; Wu; Haefliger] Let VV be a compact mm-manifold. If n≥2m+2n\ge2m+2, any two embeddings V→RnV\to\mathbb{R}^n are isotopic. If VV is connected, m≥2m\ge2 and n=2m+1n=2m+1, the same holds.

Proof of the first statement. Join the embeddings by the straight-line homotopy and regard it as a family F ⁣:V×I→RnF\colon V\times I\to\mathbb{R}^n. By the multijet transversality theorem one may perturb FF, fixing the ends, so that the set {(x,y,t):x≠y, Ft(x)=Ft(y)}\{(x,y,t):x\ne y,\ F_t(x)=F_t(y)\}, of expected dimension 2m+1−n<02m+1-n<0, is empty, and so that the set where dFtdF_t drops rank, of expected dimension (m+1)−(n−m+1)<0(m+1)-(n-m+1)<0, is empty. Then every FtF_t is an embedding. ◻

This is the argument of [cor:counts] with one more parameter. In codimension 22 nothing of the kind holds: π0Emb(S1,R3)\pi_0\mathop{\mathrm{Emb}}(S^1,\mathbb{R}^3) is the set of knot types. Between the extremes there is a systematic theory.

Theorem 1.10 (Manifold calculus). [literature; Weiss [50]; Goodwillie–Weiss [25]; Goodwillie–Klein [21]] For a presheaf of spaces FF on the open subsets of an mm-manifold VV that takes isotopy equivalences to equivalences, let TkF(U)T_kF(U) be the homotopy limit of F(W)F(W) over the open W⊂UW\subset U diffeomorphic to a disjoint union of at most kk discs. Then TkFT_kF is the universal approximation to FF by a presheaf that is a homotopy sheaf for covers in which every set of at most kk points lies in some member. For F=Emb(−,M)F=\mathop{\mathrm{Emb}}(-,M) with dim⁡M=n\dim M=n: T1F=Imm(−,M)T_1F=\mathop{\mathrm{Imm}}(-,M), and the map Emb(V,M)→TkEmb(V,M)\mathop{\mathrm{Emb}}(V,M)\to T_k\mathop{\mathrm{Emb}}(V,M) is (k(n−m−2)+1−m)(k(n-m-2)+1-m)-connected. In particular, if n−m≥3n-m\ge3 then Emb(V,M)≃holim⁡kTkEmb(V,M)\mathop{\mathrm{Emb}}(V,M)\simeq\operatorname{holim}_kT_k\mathop{\mathrm{Emb}}(V,M).

So the Smale–Hirsch theorem is the linear approximation, and in codimension at least three the higher approximations converge. The limit has a description that does not mention VV’s open sets.

Theorem 1.11. [literature; Boavida de Brito–Weiss [9]] holim⁡kTkEmb(V,M)\operatorname{holim}_kT_k\mathop{\mathrm{Emb}}(V,M) is the space of derived maps EV→EME_V\to E_M between the presheaves on discs defined by VV and MM.

In codimension at least three, then, the space of embeddings of VV in MM is determined by the configuration spaces of the two manifolds, with their tangential data and the maps among them: by how finitely many points sit in each. In codimension two the tower does not converge to the embedding space, but its stages are still invariants of knots, related to those of finite type.

Factorization homology

Presheaves on discs classify embeddings; copresheaves on discs give invariants of manifolds. Let AA be an algebra over the operad of 1.5 in a symmetric monoidal ∞\infty-category, that is, a symmetric monoidal functor on Discn\mathsf{Disc}_n. Its factorization homology over an nn-manifold MM is the colimit ∫MA=colimA(U)\int_MA=\mathop{\mathrm{colim}}A(U) over the discs UU embedded in MM.

Theorem 1.12. [literature; Ayala–Francis [3]; Lurie] M↦∫MAM\mapsto\int_MA is symmetric monoidal and satisfies excision: if M=M−∪M0×RM+M=M_-\cup_{M_0\times\mathbb{R}}M_+ is a decomposition along a collared hypersurface, then ∫MA≃∫M−A⊗∫M0×RA∫M+A\int_MA\simeq\int_{M_-}A\otimes_{\int_{M_0\times\mathbb{R}}A}\int_{M_+}A. These properties characterize it.

Excision is where Theorem 1.4, Theorem 1.5 enter: a collar of M0M_0 is what makes ∫M0×RA\int_{M_0\times\mathbb{R}}A an algebra and the two sides modules over it. For n=1n=1 and M=S1M=S^1 the invariant is Hochschild homology, which for the algebra of functions on a manifold returns the differential forms of [prop:three].

Notes

The fibration theorem is due independently to Palais and to Cerf, with a short proof by Lima. For the homotopy types of diffeomorphism groups see Smale, Hatcher, and the survey literature on pseudoisotopy; the failure in dimension four is due to Watanabe. Gromov’s Partial Differential Relations and the book of Eliashberg and Mishachev treat the h-principle. Manifold calculus is surveyed by Munson; the configuration-space description is [9]. Factorization homology is developed in [3] and in Lurie’s Higher Algebra.

Exercises

Exercise 1.1. Find the diffeotopy of R2\mathbb{R}^2 generated by G(x,t)=(−x2,x1)G(x,t)=(-x_2,x_1), and the velocity field of the diffeotopy Ft(x)=etxF_t(x)=e^{t}x. Show that x2∂xx^2\partial_x does not generate a diffeotopy of R\mathbb{R}.

Exercise 1.2. Compute the rotation numbers of the unit circle, the figure eight (sin⁡t,sin⁡tcos⁡t)(\sin t,\sin t\cos t) and the limaçon r=1+2cos⁡θr=1+2\cos\theta. Which pairs are regularly homotopic?

Exercise 1.3. Show that Emb(Dn,Rn)→Rn×GLn(R)\mathop{\mathrm{Emb}}(D^n,\mathbb{R}^n)\to\mathbb{R}^n\times\mathop{\mathrm{GL}}_n(\mathbb{R}), f↦(f(0),df0)f\mapsto(f(0),df_0), is a homotopy equivalence.

Exercise 1.4. Show that the isotopy of long knots described after 1.4 is smooth, and that it is not induced by a diffeotopy of R3\mathbb{R}^3 if the knot is non-trivial.

Exercise 1.5. Show from 1.9 that any two embeddings of a circle in R4\mathbb{R}^4 are isotopic, and explain where the proof fails in R3\mathbb{R}^3.

Surfaces

The source ends by classifying compact surfaces, using everything before it: Morse functions to cut a surface into pieces, isotopy and gluing to recognise the pieces and put them back, the Euler characteristic and orientability to tell the results apart. It is the one dimension above 11 in which the programme of the book can be carried to the end. The higher-categorical statements of the previous three chapters can also be carried to the end in dimension two, and this chapter does so. The classification theorem turns out to be the statement about components; the statements about higher homotopy are theorems of Earle and Eells and of Madsen and Weiss.

The classification

Let Σg\Sigma_g be the closed orientable surface of genus gg, the sphere with gg handles, and NhN_h the sphere with h≥1h\ge1 crosscaps.

Theorem 1.1 (Classification of surfaces). [classical; Hirsch 9.3.5, 9.3.10, 9.3.11] Two compact connected surfaces are diffeomorphic if and only if they have the same Euler characteristic, the same number of boundary components, and are both orientable or both non-orientable. Every closed connected surface is diffeomorphic to exactly one Σg\Sigma_g, with χ=2−2g\chi=2-2g, or NhN_h, with χ=2−h\chi=2-h. Removing bb open discs lowers χ\chi by bb.

Hirsch’s proof takes a Morse function, orders its critical points, and inducts on the number of saddles. The pieces between consecutive critical levels are recognised by two lemmas: a connected surface carrying a Morse function with critical points only of index 00 and 22 and boundary a level set is a disc or an annulus (Hirsch 9.2), and a connected surface with a single critical point, a saddle, is a disc with two holes if it is orientable and a Möbius band with one hole if it is not (Hirsch 9.3.1–9.3.3; 1.4). The reassembly uses Theorem 1.5, Theorem 1.5, Theorem 1.6: gluing along circles depends only on the isotopy class of the gluing map, and a diffeomorphism of the circle is isotopic to the identity or a reflection.

The monoid of surfaces

Connected sum is well defined on connected surfaces up to diffeomorphism, by [thm:discs] and the fact that every surface admits a diffeomorphism reversing the orientation of a small disc up to isotopy when it is non-orientable, and that Σg\Sigma_g admits an orientation-reversing diffeomorphism. Write T=Σ1T=\Sigma_1 and P=RP2P=\mathbb{RP}^2.

Corollary 1.2. [classical; Hirsch 9.1.7] Under connected sum, the diffeomorphism classes of closed connected surfaces form the commutative monoid with unit S2S^2, generators TT and PP, and the single relation T#P=P#P#PT\#P=P\#P\#P.

A surface is often presented by a word: a polygon with 2n2n sides labelled by nn letters, each occurring twice, with exponents giving the direction of identification. The quotient has one face and nn edges, so χ=V−n+1\chi=V-n+1 with VV the number of vertex classes, and it is orientable exactly when every letter occurs once with each exponent. The script computes VV by a union–find pass and identifies aba−1b−1aba^{-1}b^{-1} as the torus, abababab as the projective plane, abab−1abab^{-1} as the Klein bottle, and so on (1.1).

The classification as a presentation

Let 2Cob2\mathsf{Cob} be the symmetric monoidal category whose objects are finite disjoint unions of circles and whose morphisms are diffeomorphism classes of compact oriented cobordisms: the homotopy category of Cob2or\mathsf{Cob}^{\mathrm{or}}_2 of [def:cobcat]. By [thm:handle] every cobordism is a composite of the four elementary ones given by a single critical point: the disc as a cobordism ∅→S1\emptyset\to S^1 (index 00) and S1→∅S^1\to\emptyset (index 22), and the pair of pants as a cobordism from two circles to one and from one to two (index 11).

Theorem 1.3. [classical; folklore; Abrams [2]; Kock [33]] 2Cob2\mathsf{Cob} is generated as a symmetric monoidal category by η ⁣:∅→S1,ϵ ⁣:S1→∅,μ ⁣:S1⊔S1→S1,δ ⁣:S1→S1⊔S1,\eta\colon\emptyset\to S^1,\qquad\epsilon\colon S^1\to\emptyset,\qquad \mu\colon S^1\sqcup S^1\to S^1,\qquad\delta\colon S^1\to S^1\sqcup S^1, subject to the relations: (μ,η)(\mu,\eta) is a commutative monoid, (δ,ϵ)(\delta,\epsilon) is a cocommutative comonoid, and (μ⊔id)(id⊔δ)=δμ=(id⊔μ)(δ⊔id)(\mu\sqcup\mathrm{id})(\mathrm{id}\sqcup\delta)=\delta\mu=(\mathrm{id}\sqcup\mu)(\delta\sqcup\mathrm{id}).

Where the classification enters. Generation is Morse theory. Using the relations, every connected composite of generators from mm circles to nn circles can be brought to the normal form δn−1(μδ)gμm−1\delta^{n-1}(\mu\delta)^g\mu^{m-1}, with gg the number of times a δ\delta is followed by a μ\mu that rejoins its two outputs. Two cobordisms with the same normal form are equal in the presented category. Conversely, two connected oriented cobordisms with the same mm, nn and gg are diffeomorphic by 1.1, since gg is the genus and m+nm+n the number of boundary components. So the functor from the presented category to 2Cob2\mathsf{Cob} is faithful as well as full. ◻

The classification theorem is thus exactly the completeness of the relations. The same structure is visible one level up: by [thm:cerf], two Morse functions on the same cobordism are joined by a path with isolated births, deaths and crossings, and each such event is one of the relations of the theorem.

Two-dimensional field theories

A commutative Frobenius algebra over a field kk is a finite-dimensional commutative algebra AA with a linear form ε ⁣:A→k\varepsilon\colon A\to k such that (a,b)↦ε(ab)(a,b)\mapsto\varepsilon(ab) is non-degenerate.

Theorem 1.4. [classical; Dijkgraaf; Abrams [2]; [33]] The category of symmetric monoidal functors 2Cob→Vectk2\mathsf{Cob}\to\mathsf{Vect}_k is equivalent to the category of commutative Frobenius algebras over kk and their isomorphisms, by Z↦Z(S1)Z\mapsto Z(S^1). The comultiplication is dual to the multiplication under the pairing, and for the closed surface of genus gg, Z(Σg)=ε(wg),w=μδ(1)=∑ieiei#,Z(\Sigma_g)=\varepsilon(w^g),\qquad w=\mu\delta(1)=\sum_ie_ie_i^\#, where (ei)(e_i) is a basis and (ei#)(e_i^\#) the dual basis for the pairing.

Example 1.5. (a) A=k[x]/(x2)A=k[x]/(x^2) with ε(1)=0\varepsilon(1)=0, ε(x)=1\varepsilon(x)=1: the cohomology of the sphere. The dual basis of (1,x)(1,x) is (x,1)(x,1), so w=2xw=2x and Z(S2)=0Z(S^2)=0, Z(T2)=2Z(T^2)=2, Z(Σg)=0Z(\Sigma_g)=0 for g≥2g\ge2. The handle element is the Euler class.

(b) A=C[G]A=\mathbb{C}[G] for a finite abelian group GG of order nn, with ε(g)=δg,e/n\varepsilon(g)=\delta_{g,e}/n. The dual basis of (g)(g) is (n g−1)(n\,g^{-1}), so w=n2ew=n^2e and Z(Σg)=ε(n2ge)=n2g−1Z(\Sigma_g)=\varepsilon(n^{2g}e)=n^{2g-1}. This is #Hom(π1Σg,G)/#G\#\mathop{\mathrm{Hom}}(\pi_1\Sigma_g,G)/\#G, the number of principal GG-bundles on Σg\Sigma_g counted with weight one over the order of the automorphism group.

In every case Z(T2)=dim⁡AZ(T^2)=\dim A: the torus is the trace of the identity of Z(S1)Z(S^1), as in [thm:CH] for d=1d=1. The script checks the Frobenius relation and these values from the structure constants.

Extended theories

A field theory in the sense above assigns nothing to a point, and circles cannot be cut. Allowing 11-manifolds with boundary and surfaces with corners gives a symmetric monoidal bicategory Bord2\mathsf{Bord}_2.

Theorem 1.6. [literature; Schommer-Pries [47]] The oriented bordism bicategory Bord2or\mathsf{Bord}^{\mathrm{or}}_2 has a finite presentation by generators and relations, obtained from the classification of the generic singularities of maps from surfaces to the plane and of one-parameter families of them. Consequently, extended oriented two-dimensional field theories with values in the bicategory of algebras, bimodules and bimodule maps are classified by separable symmetric Frobenius algebras.

The proof is the method of 1.3 with R\mathbb{R} replaced by R2\mathbb{R}^2: a generic map to the plane has folds and cusps where a Morse function has critical points, and the jet transversality theorem in one and two parameters supplies the relations. For framed theories the theorem confirms [thm:CH] in dimension two: the fully dualizable algebras are the separable ones.

Moduli

By [thm:ehresmann] the stack of bundles of closed connected surfaces has shape M2 ≃ ∐g≥0BDiff(Σg) ⊔ ∐h≥1BDiff(Nh),\mathcal M_2\ \simeq\ \coprod_{g\ge0}B\mathop{\mathrm{Diff}}(\Sigma_g)\ \sqcup\ \coprod_{h\ge1}B\mathop{\mathrm{Diff}}(N_h), and 1.1 is the computation of π0M2\pi_0\mathcal M_2. The next question is the homotopy type of each component.

Theorem 1.7. [classical; Smale; Earle–Eells [18]; Gramain] The identity component of Diff(Σg)\mathop{\mathrm{Diff}}(\Sigma_g) is contractible for g≥2g\ge2, is homotopy equivalent to the torus for g=1g=1, and to SO(3)\mathrm{SO}(3) for g=0g=0. The identity component of Diff(Nh)\mathop{\mathrm{Diff}}(N_h) is contractible for h≥3h\ge3, a circle for the Klein bottle, and SO(3)\mathrm{SO}(3) for RP2\mathbb{RP}^2.

So for g≥2g\ge2 the component BDiff+(Σg)B\mathop{\mathrm{Diff}}^+(\Sigma_g) is the classifying space BΓgB\Gamma_g of the mapping class group Γg=π0Diff+(Σg)\Gamma_g=\pi_0\mathop{\mathrm{Diff}}^+(\Sigma_g): a surface bundle is determined by its monodromy. It is also the shape of the moduli stack of Riemann surfaces of genus gg, since the space of complex structures on a surface is contractible (a choice in the sense of [prin:choices]).

Theorem 1.8 (Madsen–Weiss). [literature; [40]; Harer; [22]] The cohomology of BΓgB\Gamma_g is independent of gg in a range of degrees growing linearly with gg, and the stable cohomology is that of a component of Ω∞MTSO(2)\Omega^\infty MTSO(2). Rationally, H∗(BΓ∞;Q)≅Q[κ1,κ2,… ],∣κi∣=2i.H^*(B\Gamma_\infty;\mathbb Q)\cong\mathbb Q[\kappa_1,\kappa_2,\dots],\qquad|\kappa_i|=2i .

The second statement is [thm:GMTW] for oriented surfaces together with the stability theorem; the classes κi\kappa_i are the characteristic classes of surface bundles defined by Mumford, Morita and Miller, and the theorem was Mumford’s conjecture.

What the classification is the π0\pi_0 of

collects the chapter. In each row the first entry is a statement of the source or a direct consequence, and the second is the statement about a space or a category of which it is the shadow on components.

Surfaces: classical statements and the homotopy types behind them.
classical statement of which it is π0\pi_0 or the homotopy category
closed surfaces are Σg\Sigma_g, NhN_h M2≃∐BDiff\mathcal M_2\simeq\coprod B\mathop{\mathrm{Diff}}; 1.7
N2=Z/2\mathfrak N_2=\mathbb{Z}/2, Ω2=0\Omega_2=0 BCob2≃Ω∞−1MTO(2)B\mathsf{Cob}_2\simeq\Omega^{\infty-1}MTO(2); [thm:GMTW]
generators and relations for 2Cob2\mathsf{Cob} presentation of Bord2\mathsf{Bord}_2; Theorem 1.6, Theorem 1.11
a diffeomorphism of S1S^1 is isotopic to an isometry Diff(S1)≃O(2)\mathop{\mathrm{Diff}}(S^1)\simeq\mathrm{O}(2)
connected sum is well defined Emb(D2,M)≃Fr(TM)\mathop{\mathrm{Emb}}(D^2,M)\simeq\mathop{\mathrm{Fr}}(TM); [thm:discs]
stable range for H∗(BΓg)H_*(B\Gamma_g) Z×BΓ∞+≃Ω∞MTSO(2)\mathbb{Z}\times B\Gamma_\infty^+\simeq\Omega^\infty MTSO(2); 1.8

Notes

The classification of surfaces is from the nineteenth century (Möbius, Jordan, Dyck); proofs by Morse theory are in Hirsch and in Gramain’s notes. The equivalence between two-dimensional field theories and Frobenius algebras is folklore from the late 1980s; [33] is a careful account. The unoriented analogue is due to Turaev and Turner. For mapping class groups see Farb and Margalit; for the Madsen–Weiss theorem and its descendants, Galatius and Randal-Williams.

Exercises

Exercise 1.1. Identify the closed surfaces given by the words aba−1b−1aba^{-1}b^{-1}, abababab, abab−1abab^{-1}, aabbaabb, abca−1b−1c−1abca^{-1}b^{-1}c^{-1} and abcabcabcabc.

Exercise 1.2. Deduce T#P≅P#P#PT\#P\cong P\#P\#P from 1.1, using χ(M#N)=χ(M)+χ(N)−2\chi(M\#N)=\chi(M)+\chi(N)-2.

Exercise 1.3. Verify 1.5, and show that for a semisimple commutative Frobenius algebra A=⨁ik eiA=\bigoplus_ik\,e_i with ε(ei)=θi\varepsilon(e_i)=\theta_i one has Z(Σg)=∑iθi1−gZ(\Sigma_g)=\sum_i\theta_i^{1-g}.

Exercise 1.4. Let WW be a compact connected surface with a Morse function f ⁣:W→[a,b]f\colon W\to[a,b], ∂W=f−1{a,b}\partial W=f^{-1}\{a,b\}, having exactly one critical point, of index 11. Show that χ(W)=−1\chi(W)=-1, and that WW is a disc with two holes if it is orientable and a Möbius band with one hole if it is not.

Exercise 1.5. Show from the relations of 1.3 that Z(T2)=dim⁡Z(S1)Z(T^2)=\dim Z(S^1) for every two-dimensional field theory with values in vector spaces.

Assessment

What the Language Did

The two earlier volumes of this series ended with an assessment of how much of their higher-categorical vocabulary was doing work. The same question is asked here, and the answer is different in both directions: more of the vocabulary is essential, and less of the mathematics is touched by it.

Where the proofs live

Principle 1.1 (Four inputs). [observation] The theorems of the source cited in this book are derived there from four analytic facts and the topology of function spaces: the inverse function theorem; the existence of partitions of unity; the Morse–Sard theorem; and the existence, uniqueness and smooth dependence of solutions of ordinary differential equations. The first gives [thm:transpull] and tubular neighbourhoods. The second gives approximation, collars, and every passage from local to global. The third gives transversality and so every genericity statement. The fourth gives the regular interval theorem, isotopy extension, and the handle description of a critical point.

None of the four is proved, shortened, or explained by anything in this book. The categorical statements begin where the four inputs end. characterizes derived manifolds, but that the transverse fibre products exist is the inverse function theorem, and that they are dense is Sard’s theorem. classifies concordance classes, but its proof uses partitions of unity in an essential way. says that restriction is a fibration, and the lifting is done by integrating a vector field. What the language provides is the form of the conclusions, and the form matters for what can be done next.

The source as a book about components

The main theorems of the source as statements about components.
theorem of the source the statement of which it is the shadow on π0\pi_0
homotopic maps pull back isomorphic bundles; classification (4.2.4, 4.3.4) Π[M,BGLk]≃Map(M,BO(k))\Pi[M,B\mathop{\mathrm{GL}}_k]\simeq\mathop{\mathrm{Map}}(M,B\mathrm{O}(k)) ([thm:concordance])
tubular neighbourhoods and collars are unique up to isotopy (4.5.3, 4.6) the spaces of them are contractible (Theorem 1.3, Theorem 1.4)
Hopf’s theorem (5.1.10) Sn→K(Z,n)S^n\to K(\mathbb{Z},n) is (n+1)(n+1)-connected; Pontryagin–Thom in MM
invariance of intersection numbers (5.2.1) derived fibre products have cobordism classes ([thm:spivak])
χ\chi by vector fields equals χ\chi by homology (6.3.8) a trace computed before and after a monoidal functor ([thm:DP])
regular interval theorem (6.2.2) families of compact manifolds are classified by ∐BDiff\coprod B\mathop{\mathrm{Diff}} ([thm:ehresmann])
Thom’s theorem (7.2.3) BCobd≃Ω∞−1MTO(d)B\mathsf{Cob}_d\simeq\Omega^{\infty-1}MTO(d) ([thm:GMTW])
isotopy extension (8.1.3) Diffc(M)→Emb(V,M)\mathop{\mathrm{Diff}}_c(M)\to\mathop{\mathrm{Emb}}(V,M) is a fibre bundle ([thm:palais])
embeddings of discs are isotopic (8.3.1) Emb(Dn,M)≃Fr(TM)\mathop{\mathrm{Emb}}(D^n,M)\simeq\mathop{\mathrm{Fr}}(TM) ([thm:discs])
classification of surfaces (9.3.11) M2≃∐BDiff\mathcal M_2\simeq\coprod B\mathop{\mathrm{Diff}}, with Theorem 1.7, Theorem 1.8

is the summary of the book. Read down the left column and one has Hirsch’s table of contents; read down the right and one has a good part of geometric topology since. In every row the left-hand statement is what one can prove with one parameter of genericity (a path between two choices) and the right-hand statement is what one gets by carrying all parameters. The source’s method, described in the Preface, already contains the idea: an invariant is independent of choices because two choices are joined by a generic path. The higher statement is that the space of choices is contractible because spheres of choices bound generic discs.

What is higher here

In the volume on game theory the ∞\infty-categorical structure was largely decorative, and in the volume on information theory it was concentrated in one place (the bar construction behind entropy). Here it is needed in at least four places where no 11-categorical substitute exists.

First, derived fibre products. The last row of [tab:four] has no solution in any ordinary category of ringed spaces, and [thm:CS] characterizes the ∞\infty-category that solves it.

Second, moduli. The stack BGLkB\mathop{\mathrm{GL}}_k and the stack of manifold bundles are not sheaves of sets, and their shapes BO(k)B\mathrm{O}(k) and ∐BDiff(W)\coprod B\mathop{\mathrm{Diff}}(W) are not discrete.

Third, composition of cobordisms, which is associative only up to the contractible choices of [ch:tubular]; the cobordism category is an ∞\infty-category in an unavoidable way, and its homotopy type is the content of [thm:GMTW].

Fourth, embedding spaces, whose description in Theorem 1.10, Theorem 1.11 is by homotopy limits over categories of discs.

In each case the higher statement was a theorem of someone else. This book contributes arrangement: the observation that Hirsch’s nine chapters are, in order, the π0\pi_0 of these theories, with the dictionary made explicit and the status of each translation marked.

What the language did not do

It proved no theorem of differential topology. It did not make the analytic inputs of 1.1 easier. It did not internalize genericity: the Baire property of the strong topology has no formulation in terms of the sheaf [M,N][M,N] that we know of ([rem:nobaire]), and the density theorems are stated here exactly as in the source.

It is blind, at the level of generality used here, to the phenomena that distinguish the smooth category from the topological one. Exotic spheres appear in [thm:spheres] as π0Diff(Sn)\pi_0\mathop{\mathrm{Diff}}(S^n) and nowhere else; nothing in the categorical formalism predicts them. The same formalism applied to topological or piecewise-linear manifolds gives the same tables with different entries, and the differences among the entries are the hard part of the subject.

And several of the statements quoted are much harder than the theorems they generalize. One should not come away thinking that [thm:GMTW] is a reformulation of Thom’s theorem. It contains Thom’s theorem as its statement about components, and its proof uses new ideas.

Directions

Open Problem 1.2 (Internal genericity). [programme] Find a property of a subobject A⊂[M,N]A\subset[M,N] of the mapping object, stated in terms of plots, that is implied by residuality in the strong topology, implies the conclusion of the parametric transversality theorem for every sufficiently rich plot, and is closed under countable intersections.

Open Problem 1.3 (Synthetic transversality). [programme] In a model of synthetic differential geometry, or in differential cohesive homotopy type theory, the transverse fibre products are those preserved by passage to tangent complexes ([thm:derivedtrans]). Give a synthetic proof of [lem:fibre] and isolate the axiom that plays the role of Sard’s theorem.

Open Problem 1.4 (Families). [programme] concerns a single diagram. For a family of diagrams parametrized by a manifold BB, the limits form a derived manifold over BB, and the points of BB where the fibre is not a manifold form a closed set of positive codimension in good cases. Describe the stratification of BB by the type of the singular fibre in terms of the tangent complex, in the generality of [thm:CS].

The third problem is the one the companion volume takes up in a special case. A neural network architecture is a smooth family of maps parametrized by its weights; the decision boundaries, level sets and fixed-point sets of the network are fibres of fibre products over the parameter space; and the questions asked about them in practice (when is the boundary a manifold, how does its topology change in training, what is the set of parameters fitting the data) are questions about that stratification. The theorems of this book that are used there are Theorem 1.12, Theorem 1.5, Lemma 1.4, the openness results of [ch:function], Morse theory in the form of Corollary 1.7, Theorem 1.12, and the isotopy theory of [ch:isotopy].

Categorical Toolkit

This appendix fixes the meaning of the categorical terms used in the text. It is a list of definitions with references and proves nothing.

Spaces and ∞\infty-categories

A space is an ∞\infty-groupoid: a homotopy type, presented for instance by a Kan complex or a CW complex. Spaces form an ∞\infty-category S\mathcal{S}. An ∞\infty-category has objects and, for each pair of objects, a space of morphisms, with composition associative up to coherent homotopy; quasi-categories and complete Segal spaces are equivalent models. An ordinary category is an ∞\infty-category whose mapping spaces are discrete. Limits and colimits in an ∞\infty-category are homotopy limits and colimits in any model; in particular a pushout in S\mathcal{S} is a homotopy pushout. The classifying space BCB\mathcal C of an ∞\infty-category is the space obtained by inverting all morphisms. A category in which composition is defined after a choice from a contractible space, such as the cobordism category, presents an ∞\infty-category.

Sheaves on Euclidean spaces

Cart\mathsf{Cart} is the category of the spaces Rn\mathbb{R}^n and smooth maps, with the coverage of [def:smset]. A sheaf of sets on it is a smooth set; a sheaf of spaces (a functor Cartop→S\mathsf{Cart}^{\mathrm{op}}\to\mathcal{S} satisfying descent for covers) is a smooth ∞\infty-groupoid. The ∞\infty-topos H\mathbf H of these has a chain of adjoint functors Π ⊣ Disc⁡ ⊣ Γ ⊣ coDisc⁡ ⁣:H⇄S,\Pi\ \dashv\ \operatorname{Disc}\ \dashv\ \Gamma\ \dashv\ \operatorname{coDisc}\colon\quad \mathbf H\rightleftarrows\mathcal{S}, where Γ\Gamma is evaluation at the point, Disc⁡\operatorname{Disc} is the constant sheaf, and Π\Pi is the shape [45]. For X∈HX\in\mathbf H and a manifold MM, a section of XX over MM is a map M→XM\to X in H\mathbf H; the mapping object [M,X][M,X] has sections over UU the sections of XX over U×MU\times M. The delooping BGBG of a sheaf of groups GG is the sheaf of groupoids with one object and automorphisms GG, sheafified; maps M→BGM\to BG are principal GG-bundles.

Dualizable objects and traces

Let (C,⊗,1)(\mathcal C,\otimes,\mathbf 1) be a symmetric monoidal category or ∞\infty-category. An object XX is dualizable if there are an object X∨X^\vee and maps η ⁣:1→X⊗X∨\eta\colon\mathbf 1\to X\otimes X^\vee, ϵ ⁣:X∨⊗X→1\epsilon\colon X^\vee\otimes X\to\mathbf 1 such that the composites X→X⊗X∨⊗X→XX\to X\otimes X^\vee\otimes X\to X and X∨→X∨⊗X⊗X∨→X∨X^\vee\to X^\vee\otimes X\otimes X^\vee\to X^\vee are identities. The trace of f ⁣:X→Xf\colon X\to X is ϵ∘swap∘(f⊗1)∘η∈End⁡(1)\epsilon\circ\text{swap}\circ(f\otimes1)\circ\eta\in \operatorname{End}(\mathbf 1), and the Euler characteristic, or dimension, of XX is tr(idX)\mathop{\mathrm{tr}}(\mathrm{id}_X) [16, 43]. In vector spaces the dualizable objects are the finite-dimensional ones; in spectra they are the finite spectra. An object of a symmetric monoidal (∞,d)(\infty,d)-category is fully dualizable if it is dualizable and the evaluation and coevaluation maps have adjoints, which have adjoints, and so on through dimension d−1d-1 [36].

Spectra and Thom spectra

The ∞\infty-category Sp\mathsf{Sp} of spectra is the stabilization of pointed spaces; it is symmetric monoidal under the smash product, with unit the sphere spectrum S\mathbb S, and πnS\pi_n\mathbb S is the nn-th stable homotopy group of spheres. There are adjoint functors Σ+∞ ⁣:S→Sp\Sigma^\infty_+\colon\mathcal{S}\to\mathsf{Sp} and Ω∞ ⁣:Sp→S\Omega^\infty\colon\mathsf{Sp}\to\mathcal{S}, and Ω∞−1E\Omega^{\infty-1}E means Ω∞ΣE\Omega^\infty\Sigma E. The Thom spectrum of a stable vector bundle ξ\xi over XX has kk-th space the Thom space of the rank-kk part of ξ\xi over the corresponding piece of XX; equivalently it is the colimit of [thm:ABGHR]. MOMO is the Thom spectrum of the universal stable bundle over BOB\mathrm{O}, and MTO(d)MTO(d) that of the virtual bundle −γd-\gamma_d over BO(d)B\mathrm{O}(d).

C∞C^\infty-rings

A C∞C^\infty-ring is a product-preserving functor Cart→Set\mathsf{Cart}\to\mathsf{Set} ([def:cinf]); a simplicial C∞C^\infty-ring is a product-preserving functor Cart→S\mathsf{Cart}\to\mathcal{S} ([def:scr]). The free C∞C^\infty-ring on nn generators is C∞(Rn)C^\infty(\mathbb{R}^n). A C∞C^\infty-ring is finitely presented if it is C∞(Rn)/(f1,…,fk)C^\infty(\mathbb{R}^n)/(f_1,\dots,f_k); a simplicial one is finitely presented if it is a finite colimit of free ones, and the opposite of the ∞\infty-category of these is DMfd\mathsf{DMfd} ([thm:CS]). Hadamard’s lemma, that a smooth function vanishing on {x1=0}\{x_1=0\} is divisible by x1x_1, is what makes quotients of C∞C^\infty-rings by coordinate functions behave as in algebra.

Notation

Mfd\mathsf{Mfd}, Cart\mathsf{Cart}, Open\mathsf{Open} manifolds; Euclidean spaces; open subsets of Euclidean spaces
SmSet\mathsf{SmSet}, Dflg\mathsf{Dflg}, H\mathbf H smooth sets; diffeological spaces; smooth ∞\infty-groupoids
C∞RingC^\infty\mathsf{Ring}, DMfd\mathsf{DMfd} C∞C^\infty-rings; derived manifolds
[M,N][M,N], Π\Pi mapping object; shape
Emb\mathop{\mathrm{Emb}}, Imm\mathop{\mathrm{Imm}}, Diff\mathop{\mathrm{Diff}} embeddings, immersions, diffeomorphisms (as smooth sets or their shapes)
f⋔gf\pitchfork g, M×NhPM\times^h_NP transverse; derived fibre product
Th(ξ)\mathop{\mathrm{Th}}(\xi), M−TMM^{-TM} Thom space; Thom spectrum of the stable normal bundle
Nn\mathfrak N_n, Ωn\Omega_n, Cobd\mathsf{Cob}_d, Bordd\mathsf{Bord}_d cobordism groups; cobordism categories
Σg\Sigma_g, NhN_h closed surfaces, orientable of genus gg and non-orientable with hh crosscaps

Concordance with Hirsch

The table lists the sections of M. W. Hirsch, Differential Topology (Graduate Texts in Mathematics 33), with the place in this book where their content is treated and the form it takes there. Section numbers are those of the source; “2.1” means Chapter 2, Section 1.

Source Here Form
1.1–1.2 Differential structures; maps and tangent bundles [ch:manifolds], [ch:tangent] the category Mfd\mathsf{Mfd} and its four enlargements; TT as a Weil functor
1.3 Embeddings and immersions §§1.6–1.7 transverse fibre products; Mfd\mathsf{Mfd} as an idempotent completion
1.4 Manifolds with boundary §1.8 neat submanifolds; the half-line as a smooth set
2.1 The weak and strong topologies §§3.1, 3.4 open conditions as DD-open subobjects of [M,N][M,N]
2.2 Approximations §3.1 density; smoothing of structures ([thm:smoothing])
2.3 Approximations on ∂\partial-manifolds and pairs §1.8 not treated separately
2.4 Jets and the Baire property §§2.5, 3.5, 3.6 jets compose; genericity is not internal to the topos
2.5 Analytic approximations — not treated
3.1 The Morse–Sard theorem §4.1 stated; the single analytic source of genericity
3.2 Transversality §§4.2–4.6 fibrewise lemma; derived manifolds; cobordism classes of derived fibre products
4.1–4.3 Vector bundles; classification [ch:bundles] the stack BGLkB\mathop{\mathrm{GL}}_k; concordance classification; bundles as idempotents
4.4 Oriented vector bundles §5.5 tangential structures
4.5 Tubular neighbourhoods §6.1 contractible space of choices
4.6 Collars and tubular neighbourhoods of neat submanifolds §6.2 gluing; why cobordism categories are ∞\infty-categories
4.7 Analytic differential structures — not treated
5.1 Degrees of maps §§7.1–7.2 Hopf’s theorem as Pontryagin–Thom in MM and as a Postnikov statement
5.2 Intersection numbers and the Euler characteristic §§7.3–7.6 classes of derived intersections; traces; the diagonal
5.3 Historical remarks — —
6.1 Morse functions §8.1 generic points of [M,R][M,\mathbb{R}]
6.2 Differential equations and regular level surfaces §8.2 Ehresmann; families have no local moduli
6.3 Passing critical levels and attaching cells §§8.3–8.5 homotopy pushouts; the sublevel functor; barcodes
6.4 CW-complexes §8.3 compact manifolds are finite spaces
7.1 Cobordism and transversality §9.1 cobordism groups
7.2 The Thom homomorphism §§9.2–9.5 Thom spectra as colimits; the cobordism category
8.1 Extending isotopies §§10.1–10.2 restriction is a fibration
8.2 Gluing manifolds together §6.2 contractible choices
8.3 Isotopies of disks §§10.3–10.4 Emb(Dn,M)≃Fr(TM)\mathop{\mathrm{Emb}}(D^n,M)\simeq\mathop{\mathrm{Fr}}(TM); the little discs operad
9.1 Models of surfaces §11.2 the monoid of surfaces
9.2 Characterization of the disk §11.1 used in the classification
9.3 The classification of compact surfaces §§11.1–11.7 presentation of the cobordism category; field theories; moduli

Topics treated here that are not in the source: Weil algebras and tangent categories ([ch:tangent]); the mapping object and its shape ([ch:function]); derived manifolds ([ch:transversality]); Atiyah duality and traces (Tubular Neighbourhoods, Collars, and Spaces of Choices, Degree, Intersection Numbers, and the Euler Characteristic); flow categories, Cerf theory and framed functions ([ch:morse]); the cobordism category and the cobordism hypothesis ([ch:cobordism]); the Smale–Hirsch theorem, manifold calculus and factorization homology ([ch:isotopy]); field theories and moduli of surfaces ([ch:surfaces]).

Computations

The script verify-differential-topology.py that accompanies this book uses only the Python standard library and runs in a few seconds. It performs 71 checks, each printing PASS or FAIL, and exits with status zero if and only if all pass. The checks recompute the numerical examples in the text and test finite or numerical instances of structural statements. They are evidence that the examples are stated correctly; they do not prove the theorems.

Chapter What is checked
[ch:manifolds] the differential of A↦ATAA\mapsto A^{\mathsf T}A is onto at orthogonal matrices, n=2,3,4n=2,3,4 ([ex:On]); signed zero counts of x2−εx^2-\varepsilon; a smooth idempotent has rank equal to trace along its fixed set ([thm:karoubi]); the stereographic transition is an involution
[ch:tangent] the chain rule in dual numbers against finite differences, in one and two variables; the expansion in R[ε,η]/(ε2,η2)\mathbb{R}[\varepsilon,\eta]/(\varepsilon^2,\eta^2) and its symmetry ([prop:flip]); the vertical lift; associativity of composition of 22-jets in one and several variables ([prop:jets]); the reverse chain rule and the transpose relation ([prop:reverse])
[ch:function] [ex:weakfail]: gng_n equals the identity on [−n+1,n−1][-n+1,n-1], has derivative −1-1 somewhere, and stays outside a strong neighbourhood; a degenerate critical point of x−β(x−n)x-\beta(x-n) ([exer:morseweak])
[ch:transversality] the dimension ρ(m+n−ρ)\rho(m+n-\rho) of the tangent space to the rank-ρ\rho matrices in four cases ([prop:rank]); the linear algebra of [lem:fibre] on random instances, including non-transverse ones; the family x2−vx^2-v; the Koszul complex of x2x^2 in a polynomial model; the Euler characteristic of the tangent complex; injectivity of random projections of a circle from R5\mathbb{R}^5 to R3\mathbb{R}^3 ([cor:counts])
[ch:bundles] [ex:taut]: the idempotent of the tautological line and its monodromy; the idempotent of TS2TS^2; the count of line bundles on the torus and Klein bottle ([exer:lines])
[ch:tubular] the Jacobian 1+t1+t of the normal exponential map of the circle; the focal radius b2/ab^2/a of the ellipse ([exer:focal]); convexity of inner products; the two linear homotopies of the shift trick and the failure of the naive one ([exer:shift])
[ch:degree] the degree of zkz^k by preimages and by winding integral; the degree of the antipodal map; intersection numbers on the torus ([ex:torus]); fixed points and Lefschetz numbers of six toral automorphisms ([ex:lefschetz]); Poincaré–Hopf indices on S2S^2 and S3S^3; Euler characteristics and mod 22 Betti numbers of triangulations of S2S^2, RP2\mathbb{RP}^2 and the torus; the Gauss linking integral ([ex:linking])
[ch:morse] the critical points and indices of [ex:torusheight]; the Morse polynomial; the birth–death family; the degree-zero barcode of random functions on a circle ([prop:persist]); [exer:rpn]; the deformation of [exer:model]
[ch:cobordism] the ranks of Nn\mathfrak N_n for n≤10n\le10; the Stiefel–Whitney numbers of [ex:sw]; the arithmetic of [exer:chimod2]; the degree of the collapse of a framed point ([exer:framed0])
[ch:isotopy] a time-dependent flow on R2\mathbb{R}^2 is injective with positive Jacobian; the rotation flow; blow-up of x2∂xx^2\partial_x; the rotation numbers of [cor:WG]; the shrinking of [exer:discRn]
[ch:surfaces] vertex counts, Euler characteristics and orientability for seven polygon words ([exer:words]); T#PT\#P and P#P#PP\#P\#P; the Frobenius algebras of [ex:frob]: handle elements, Z(Σg)Z(\Sigma_g) for g≤3g\le3, the Frobenius relation, and Z(T2)=dim⁡AZ(T^2)=\dim A

Numerical tolerances are stated in the script. The random seed is fixed, so the output is reproducible.

Solutions to the Exercises

Solutions to all exercises follow, by chapter. Where a solution relies on a computation, the computation is also carried out by verify-differential-topology.py.

Chapter 1

1.1. In Mfd\mathsf{Mfd}: a cone is (S,u)(S,u) with u ⁣:S→Ru\colon S\to\mathbb{R} smooth and u2=0u^2=0, so u=0u=0 and the cone is the unique map S→ptS\to\mathrm{pt}. The same argument applies to every S=RnS=\mathbb{R}^n, so the sheaf-theoretic fibre product has exactly one plot of each dimension and is the terminal smooth set. In C∞RingC^\infty\mathsf{Ring} the pushout of C∞(R)←x↦x2C∞(R)→ev0RC^\infty(\mathbb{R})\xleftarrow{x\mapsto x^2}C^\infty(\mathbb{R})\xrightarrow{\mathrm{ev}_0}\mathbb{R} is C∞(R)/(x2)C^\infty(\mathbb{R})/(x^2), and Hadamard’s lemma g=g(0)+g′(0)x+x2hg=g(0)+g'(0)x+x^2h shows that g↦g(0)+g′(0)εg\mapsto g(0)+g'(0)\varepsilon is an isomorphism onto R[ε]/(ε2)\mathbb{R}[\varepsilon]/(\varepsilon^2). A C∞C^\infty-ring map ϕ ⁣:C∞(M)→R[ε]\phi\colon C^\infty(M)\to\mathbb{R}[\varepsilon] has the form ϕ(g)=ϕ0(g)+ϕ1(g)ε\phi(g)=\phi_0(g)+\phi_1(g)\varepsilon. The component ϕ0\phi_0 is an R\mathbb{R}-algebra map to R\mathbb{R}, hence evaluation at a point pp (every character of C∞(M)C^\infty(M) is a point evaluation, MM being second countable). Multiplicativity of ϕ\phi gives ϕ1(gh)=g(p)ϕ1(h)+h(p)ϕ1(g)\phi_1(gh)=g(p)\phi_1(h)+h(p)\phi_1(g), so ϕ1\phi_1 is a derivation at pp, a tangent vector. Conversely a tangent vector vv at pp gives g↦g(p)+v(g)εg\mapsto g(p)+v(g)\varepsilon, which respects all smooth operations by the chain rule.

1.2. The proof in the text shows that each x∈Fix(e)x\in\mathop{\mathrm{Fix}}(e) has a neighbourhood UU on which ee has constant rank k=trdexk=\mathop{\mathrm{tr}}de_x. By the constant rank theorem there are charts near xx and e(x)=xe(x)=x in which ee is (u,v)↦(u,0)(u,v)\mapsto(u,0) with u∈Rku\in\mathbb{R}^k; shrinking UU, the set e(U)e(U) is a kk-dimensional embedded submanifold near xx, and Fix(e)∩U⊂e(U)\mathop{\mathrm{Fix}}(e)\cap U\subset e(U). Conversely e(U)∩U′⊂Fix(e)e(U)\cap U'\subset\mathop{\mathrm{Fix}}(e) for a smaller neighbourhood U′U', because e(e(y))=e(y)e(e(y))=e(y). So Fix(e)\mathop{\mathrm{Fix}}(e) is an embedded submanifold near xx; it is closed as the equalizer of ee and id\mathrm{id} in a Hausdorff space. Smoothness of ee into Fix(e)\mathop{\mathrm{Fix}}(e) follows since the submanifold is embedded. For the example take M=R⊔R2M=\mathbb{R}\sqcup\mathbb{R}^2 and ee the map which is the identity on R\mathbb{R} and (x,y)↦(0,0)(x,y)\mapsto(0,0) on R2\mathbb{R}^2; then Fix(e)=R⊔{0}\mathop{\mathrm{Fix}}(e)=\mathbb{R}\sqcup\{0\}. On a connected MM the set Fix(e)=e(M)\mathop{\mathrm{Fix}}(e)=e(M) is connected, so the dimension is constant.

1.3. f(A)=ATAf(A)=A^{\mathsf T}A is polynomial, hence smooth, with values in the vector space Symn\mathsf{Sym}_n of dimension 12n(n+1)\frac12n(n+1). Its differential at AA is dfA(H)=ATH+HTAdf_A(H)=A^{\mathsf T}H+H^{\mathsf T}A. If ATA=IA^{\mathsf T}A=I and SS is symmetric, then H=12ASH=\frac12AS gives 12ATAS+12STATA=S\frac12A^{\mathsf T}AS+\frac12S^{\mathsf T}A^{\mathsf T}A=S. So II is a regular value and dim⁡O(n)=n2−12n(n+1)=12n(n−1)\dim\mathrm{O}(n)=n^2-\frac12n(n+1)=\frac12n(n-1). The tangent space at II is ker⁡dfI={H:H+HT=0}\ker df_I=\{H:H+H^{\mathsf T}=0\}.

1.4. A plot of [M,N][M,N] of dimension nn is by definition a smooth map F ⁣:Rn×M→NF\colon\mathbb{R}^n\times M\to N, and the function it induces on points is s↦F(s,−)s\mapsto F(s,-), which determines FF; so [M,N][M,N] is concrete and its points are Hom(pt×M,N)=C∞(M,N)\mathop{\mathrm{Hom}}(\mathrm{pt}\times M,N)=C^\infty(M,N). For the adjunction, when S=RnS=\mathbb{R}^n the bijection is the definition. A general smooth set is a colimit of representables, S=colimiRniS=\mathop{\mathrm{colim}}_i\mathbb{R}^{n_i}; the functor −×M-\times M preserves colimits because SmSet\mathsf{SmSet} is a topos, so Hom(S×M,N)=lim⁡iHom(Rni×M,N)=lim⁡i[M,N](Rni)=Hom(S,[M,N])\mathop{\mathrm{Hom}}(S\times M,N)=\lim_i\mathop{\mathrm{Hom}}(\mathbb{R}^{n_i}\times M,N)=\lim_i[M,N](\mathbb{R}^{n_i})=\mathop{\mathrm{Hom}}(S,[M,N]).

1.5. Let f(x)=g(y)=zf(x)=g(y)=z. The tangent space to ΔN\Delta_N at (z,z)(z,z) is {(w,w)}\{(w,w)\}. Transversality of f×gf\times g to ΔN\Delta_N at (x,y)(x,y) says that every (a,b)∈TzN⊕TzN(a,b)\in T_zN\oplus T_zN is (dfxu+w, dgyv+w)(df_xu+w,\,dg_yv+w) for some u,v,wu,v,w. Subtracting, this holds if and only if every a−ba-b is of the form dfxu−dgyvdf_xu-dg_yv, that is, if and only if dfx(TxM)+dgy(TyP)=TzNdf_x(T_xM)+dg_y(T_yP)=T_zN: given such a decomposition of a−ba-b, set w=a−dfxuw=a-df_xu.

Chapter 2

2.1. For open subsets of Euclidean spaces, T(U×V)=U×V×Rm×Rn≅TU×TVT(U\times V)=U\times V\times\mathbb{R}^m\times\mathbb{R}^n\cong TU\times TV by reordering coordinates, and the reordering commutes with the maps T(f×g)T(f\times g) and Tf×TgTf\times Tg since D(f×g)=Df⊕DgD(f\times g)=Df\oplus Dg. The isomorphism is therefore natural and glues over atlases. For the second statement, Z=f−1(z)Z=f^{-1}(z) is a submanifold with TxZ=ker⁡dfxT_xZ=\ker df_x by [thm:transpull]. A vector (x,v)∈TM(x,v)\in TM satisfies Tf(x,v)=0zTf(x,v)=0_z exactly when f(x)=zf(x)=z and dfxv=0df_xv=0, that is, when (x,v)∈TZ(x,v)\in TZ. One checks that 0z0_z is a regular value of TfTf: in coordinates D(Tf)(x,v)(ξ,ν)=(Dfxξ, D2fx(v,ξ)+Dfxν)D(Tf)_{(x,v)}(\xi,\nu)=(Df_x\xi,\,D^2f_x(v,\xi)+Df_x\nu), which is onto when DfxDf_x is. So the equality is one of submanifolds.

2.2. An AA-point of Rn\mathbb{R}^n is x+v t+12a t2x+v\,t+\tfrac12a\,t^2 with t3=0t^3=0, the 22-jet at 00 of the curve t↦x+vt+12at2t\mapsto x+vt+\frac12at^2; a map ff acts by Taylor expansion, which is the composition of 22-jets, so TAMT_AM is the bundle of 22-jets of curves and its fibre has dimension 2n2n. The map x↦ε+ηx\mapsto \varepsilon+\eta is a homomorphism since (ε+η)3=3ε2η+3εη2=0(\varepsilon+\eta)^3=3\varepsilon^2\eta+3\varepsilon\eta^2=0, and it is injective: 1, ε+η, (ε+η)2=2εη1,\ \varepsilon+\eta,\ (\varepsilon+\eta)^2=2\varepsilon\eta are independent. On points it sends x+vt+12at2x+vt+\frac12at^2 to x+vε+vη+aεηx+v\varepsilon+v\eta+a\varepsilon\eta, that is, to (x,v,v,a)(x,v,v,a) in the coordinates of [tab:weil]. The fixed set of the flip is {v=w}\{v=w\}, and T(p)T(p) and pTp_T are the two projections (x,v,w,a)↦(x,v)(x,v,w,a)\mapsto(x,v) and (x,w)(x,w), so the condition T(p)=pTT(p)=p_T is again v=wv=w. The image is exactly {(x,v,v,a)}\{(x,v,v,a)\}, of fibre dimension 2n2n, and the map is an embedding.

2.3. Write f(x)=Ax+12B(x,x)+…f(x)=Ax+\frac12B(x,x)+\dots and g(y)=Cy+12E(y,y)+…g(y)=Cy+\frac12E(y,y)+\dots with B,EB,E symmetric bilinear. Then g(f(x))=CAx+12(C B(x,x)+E(Ax,Ax))+O(∣x∣3)g(f(x))=CAx+\frac12\bigl(C\,B(x,x)+E(Ax,Ax)\bigr)+O(|x|^3), so (C,E)∘(A,B)=(CA, C∘B+E∘(A×A)).(C,E)\circ(A,B)=\bigl(CA,\ C\circ B+E\circ(A\times A)\bigr). For a third jet (F,G)(F,G), (F,G)∘((C,E)∘(A,B))=(FCA, FCB+FE(A×A)+G(CA×CA))(F,G)\circ\bigl((C,E)\circ(A,B)\bigr)=\bigl(FCA,\ FCB+FE(A\times A)+G(CA\times CA)\bigr) and ((F,G)∘(C,E))∘(A,B)=(FCA, FCB+(FE+G(C×C))(A×A))\bigl((F,G)\circ(C,E)\bigr)\circ(A,B)=\bigl(FCA,\ FCB+(FE+G(C\times C))(A\times A)\bigr), which agree.

2.4. A point of TTRn=TW⊗WRnTT\mathbb{R}^n=T_{W\otimes W}\mathbb{R}^n is x+vε+wη+aεηx+v\varepsilon+w\eta+a\varepsilon\eta. A homomorphism ϕ ⁣:A→B\phi\colon A\to B acts on AA-points by applying ϕ\phi to the coefficients. Addition: a point of TW×RWT_{W\times_\mathbb{R}W} is x+v1ε1+v2ε2x+v_1\varepsilon_1+v_2\varepsilon_2, and εi↦ε\varepsilon_i\mapsto\varepsilon sends it to x+(v1+v2)εx+(v_1+v_2)\varepsilon. Vertical lift: x+vε↦x+vεηx+v\varepsilon\mapsto x+v\varepsilon\eta, which is (x,0,0,v)(x,0,0,v). Flip: exchanging ε\varepsilon and η\eta sends (x,v,w,a)(x,v,w,a) to (x,w,v,a)(x,w,v,a). Each map is natural because it is induced by a homomorphism ([thm:weil](iii)); directly, naturality of ℓ\ell is the identity f(x+vεη)=f(x)+Dfx(v)εηf(x+v\varepsilon\eta)=f(x)+Df_x(v)\varepsilon\eta, since (εη)2=0(\varepsilon\eta)^2=0.

2.5. Let (f,f♯)(f,f^\sharp), (g,g♯)(g,g^\sharp), (h,h♯)(h,h^\sharp) be composable. Both bracketings have forward part hgfhgf and backward part (x,u′)↦f♯(x,g♯(f(x),h♯(gf(x),u′)))(x,u')\mapsto f^\sharp\bigl(x,g^\sharp(f(x),h^\sharp(gf(x),u'))\bigr), as one sees by unfolding the definition twice; so composition is associative. For F=fk⋯f1F=f_k\cdots f_1 with scalar output, [prop:reverse] gives R[F](x,1)=R[f1](x0,R[f2](x1,…R[fk](xk−1,1)))R[F](x,1)=R[f_1](x_0,R[f_2](x_1,\dots R[f_k](x_{k-1},1))) with x0=xx_0=x and xi=fi(xi−1)x_i=f_i(x_{i-1}). The forward pass computes and stores x0,…,xk−1x_0,\dots,x_{k-1}; the backward pass applies the transposed Jacobians (Dfi)xi−1T(Df_i)_{x_{i-1}}^{\mathsf T} to a single covector, for i=k,…,1i=k,\dots,1. The result is the gradient of FF at xx.

Chapter 3

3.1. The support of β(⋅−n)\beta(\cdot-n) is [n−1,n+1][n-1,n+1], so gn=idg_n=\mathrm{id} on [−n+1,n−1][-n+1,n-1] and every weak neighbourhood of id\mathrm{id}, which constrains maps on a compact set only, contains gng_n for large nn. Since β′\beta' attains its maximum 11 at some x∗x^*, gn′(n+x∗)=−1<0g_n'(n+x^*)=-1<0 while gn′=1g_n'=1 outside the support, so gng_n is not monotone. For the strong topology, ∣gn(n)−n∣=2β(0)>β(0)e−n2|g_n(n)-n|=2\beta(0)>\beta(0)e^{-n^2}, so no gng_n lies in the stated strong neighbourhood of the identity. Not closed: hn(x)=x/nh_n(x)=x/n are diffeomorphisms converging weakly to the constant map 00.

3.2. Fix t0t_0. We claim there are δ>0\delta>0 and a compact KK with ft=ft0f_t=f_{t_0} off KK for ∣t−t0∣<δ|t-t_0|<\delta. If not, for each nn there are tnt_n with ∣tn−t0∣<1/n|t_n-t_0|<1/n and xn∉[−n,n]x_n\notin[-n,n] with ftn(xn)≠ft0(xn)f_{t_n}(x_n)\ne f_{t_0}(x_n); passing to a subsequence, the xnx_n are distinct and tend to infinity. Choose a continuous ε ⁣:R→(0,∞)\varepsilon\colon\mathbb{R}\to(0,\infty) with ε(xn)<∣ftn(xn)−ft0(xn)∣\varepsilon(x_n)<|f_{t_n}(x_n)-f_{t_0}(x_n)| for all nn. The strong neighbourhood {g:∣g−ft0∣<ε}\{g:|g-f_{t_0}|<\varepsilon\} contains no ftnf_{t_n}, contradicting continuity at t0t_0. Cover [0,1][0,1] by finitely many such intervals and take the union of the compact sets. Conversely, if g=fg=f off a compact KK, the straight-line path f+t(g−f)f+t(g-f) is strongly continuous: it is constant off KK, and on KK uniform convergence suffices. So the path component of ff is as stated.

3.3. Let F ⁣:Rn×M→NF\colon\mathbb{R}^n\times M\to N be smooth and s0∈Rns_0\in\mathbb{R}^n. For a compact K⊂MK\subset M in a chart, FF and its first derivative in the MM-direction are uniformly continuous on Bˉ(s0,1)×K\bar B(s_0,1)\times K, so sup⁡x∈K(∣Fs(x)−Fs0(x)∣+∥DxFs(x)−DxFs0(x)∥)→0\sup_{x\in K}\bigl(|F_s(x)-F_{s_0}(x)|+\|D_xF_s(x)-D_xF_{s_0}(x)\|\bigr)\to0 as s→s0s\to s_0; this is convergence in CW1C^1_W. For the example let Fs(x)=(γs(x),0)F_s(x)=(\gamma_s(x),0) with γs(x)=x−2β(x−1/s)\gamma_s(x)=x-2\beta(x-1/s) for s>0s>0 and γs(x)=x\gamma_s(x)=x for s≤0s\le0, with β\beta as in [ex:weakfail]. This is smooth in (s,x)(s,x) because for ∣x∣<R|x|<R and 0<s<1/(R+1)0<s<1/(R+1) the bump is not seen. F0F_0 is an embedding and FsF_s is not injective for small s>0s>0 ([ex:weakfail]), so the set of parameters giving embeddings is not open at s=0s=0.

3.4. Reflexivity: the constant plot R→pt→X\mathbb{R}\to\mathrm{pt}\to X. Symmetry: compose with t↦1−tt\mapsto1-t. Transitivity: let p,q ⁣:R→Xp,q\colon\mathbb{R}\to X be plots with p(1)=q(0)p(1)=q(0). Choose a smooth non-decreasing λ ⁣:R→R\lambda\colon\mathbb{R}\to\mathbb{R} with λ(t)=0\lambda(t)=0 for t≤13t\le\frac13 and λ(t)=1\lambda(t)=1 for t≥23t\ge\frac23. Then pλp\lambda is a plot, constant equal to p(1)p(1) on (23,∞)(\frac23,\infty), and qλ(⋅−1)q\lambda(\cdot-1) is a plot, constant equal to q(0)q(0) on (−∞,43)(-\infty,\frac43). On the overlap (23,43)(\frac23,\frac43) of the open sets (−∞,43)(-\infty,\frac43) and (23,∞)(\frac23,\infty), which cover R\mathbb{R}, the two plots restrict to the same constant plot. By the sheaf condition (after pulling back along diffeomorphisms of the intervals with R\mathbb{R}) they glue to a plot r ⁣:R→Xr\colon\mathbb{R}\to X with r(0)=p(0)r(0)=p(0) and r(2)=q(1)r(2)=q(1); reparametrize by t↦2tt\mapsto2t.

3.5. Dense: given ff and a weak neighbourhood, which constrains ff on some [−R,R][-R,R], the functions fa(x)=f(x)+axf_a(x)=f(x)+ax are Morse for almost every aa, because faf_a is Morse exactly when −a-a is a regular value of f′f', and Sard’s theorem applies to f′f'; and fa→ff_a\to f on [−R,R][-R,R] in every CrC^r norm as a→0a\to0. Not open: f(x)=xf(x)=x is Morse. With β\beta as in [ex:weakfail], let gn(x)=x−β(x−n)g_n(x)=x-\beta(x-n). At the point x∗x^* where β′\beta' attains its maximum 11 we have β′′(x∗)=0\beta''(x^*)=0, so gn′(n+x∗)=0g_n'(n+x^*)=0 and gn′′(n+x∗)=0g_n''(n+x^*)=0: a degenerate critical point. And gn→fg_n\to f weakly.

Chapter 4

4.1. Let F ⁣:Rn×M→RnF\colon\mathbb{R}^n\times M\to\mathbb{R}^n, F(v,x)=f(x)+vF(v,x)=f(x)+v. Its differential in the vv-direction is the identity, so FF is a submersion and in particular F⋔AF\pitchfork A. By [thm:parametric], Fv=f+vF_v=f+v is transverse to AA for almost every vv. If MM and AA are curves in R3\mathbb{R}^3 then dim⁡M+dim⁡A<3\dim M+\dim A<3, and a map of a 11-manifold transverse to a 11-dimensional submanifold of R3\mathbb{R}^3 cannot meet it: at a common point the images of the two tangent spaces would have to span R3\mathbb{R}^3.

4.2. Let X0X_0 have rank ρ\rho. After multiplying on both sides by invertible matrices, which are diffeomorphisms of Rm×n\mathbb{R}^{m\times n} preserving rank, we may take X0=(Iρ000)X_0=\bigl(\begin{smallmatrix}I_\rho&0\\0&0 \end{smallmatrix}\bigr). Write a nearby matrix in blocks X=(PQRS)X=\bigl(\begin{smallmatrix}P&Q\\R&S\end{smallmatrix}\bigr) with PP invertible. Row and column operations give rkX=ρ+rk(S−RP−1Q)\mathop{\mathrm{rk}}X=\rho+\mathop{\mathrm{rk}}(S-RP^{-1}Q). So near X0X_0 the set of rank-ρ\rho matrices is the zero set of Φ(X)=S−RP−1Q\Phi(X)=S-RP^{-1}Q, a map to R(m−ρ)×(n−ρ)\mathbb{R}^{(m-\rho)\times(n-\rho)} whose differential in the SS-direction is the identity. It is a submersion, and the zero set is a submanifold of codimension (m−ρ)(n−ρ)(m-\rho)(n-\rho).

4.3. Resolve R\mathbb{R} as a C∞(R)C^\infty(\mathbb{R})-module by the Koszul complex C∞(R)→xC∞(R)C^\infty(\mathbb{R})\xrightarrow{x}C^\infty(\mathbb{R}); it is exact on the left because xx is not a zero divisor, and the cokernel is R\mathbb{R} by Hadamard’s lemma. This resolution is also a cofibrant model as a simplicial C∞C^\infty-ring (it is free on one generator in degree 11 over C∞(R)C^\infty(\mathbb{R})). Tensoring with R\mathbb{R} along evaluation at 00 gives R→0R\mathbb{R}\xrightarrow{0}\mathbb{R}: the homotopy groups are R\mathbb{R} in degrees 00 and 11, as in [ex:derived](b). For the second, resolve R\mathbb{R} in the same way and tensor with C∞(R)C^\infty(\mathbb{R}) along x↦x2x\mapsto x^2: the complex is C∞(R)→x2C∞(R)C^\infty(\mathbb{R})\xrightarrow{x^2}C^\infty(\mathbb{R}). If x2g=0x^2g=0 then g=0g=0 away from 00 and so everywhere; the map is injective, π1=0\pi_1=0, and π0=C∞(R)/(x2)=R[ε]/(ε2)\pi_0=C^\infty(\mathbb{R})/(x^2)=\mathbb{R}[\varepsilon]/(\varepsilon^2), as in [ex:derived](a).

4.4. The critical points of faf_a are the solutions of ∇f(x)=−a\nabla f(x)=-a, and the Hessian of faf_a is that of ff, which is the Jacobian of the map ∇f ⁣:U→Rm\nabla f\colon U\to\mathbb{R}^m. So faf_a is Morse if and only if −a-a is a regular value of ∇f\nabla f. By [thm:sard], applied to the smooth map ∇f\nabla f between manifolds of the same dimension, almost every aa qualifies.

4.5. Suppose g ⁣:Dn→Dng\colon D^n\to D^n is continuous without fixed points. Then ∣g(x)−x∣≥2δ>0|g(x)-x|\ge2\delta>0 by compactness. By [thm:approx] there is a smooth h ⁣:Dn→Rnh\colon D^n\to\mathbb{R}^n with ∣h−g∣<δ|h-g|<\delta; then h1=h/(1+δ)h_1=h/(1+\delta) maps DnD^n to itself, is smooth, and has ∣h1−g∣<2δ|h_1-g|<2\delta, so h1h_1 has no fixed point. Let ρ(x)\rho(x) be the point where the ray from h1(x)h_1(x) through xx leaves DnD^n: ρ(x)=x+t(x) (x−h1(x))\rho(x)=x+t(x)\,(x-h_1(x)) with t(x)≥0t(x)\ge0 the larger root of ∣x+t(x−h1(x))∣2=1|x+t(x-h_1(x))|^2=1. The discriminant of this quadratic is positive because x≠h1(x)x\ne h_1(x) and ∣x∣≤1|x|\le1, so tt is smooth, and t=0t=0 when ∣x∣=1|x|=1. Thus ρ\rho is a smooth retraction of DnD^n onto Sn−1S^{n-1}, contradicting [cor:noretract].

Chapter 5

5.1. The fibre over the line ℓθ\ell_\theta spanned by u(θ)=(cos⁡θ,sin⁡θ)u(\theta)=(\cos\theta,\sin\theta) is ℓθ\ell_\theta itself. A nowhere-zero section would be s(θ)=λ(θ)u(θ)s(\theta)=\lambda(\theta)u(\theta) with λ\lambda continuous and non-zero on [0,π][0,\pi], and since u(π)=−u(0)u(\pi)=-u(0) while s(π)=s(0)s(\pi)=s(0) we need λ(π)=−λ(0)\lambda(\pi)=-\lambda(0), so λ\lambda changes sign; contradiction. The total space is {(θ,t)}=[0,π]×R\{(\theta,t)\}=[0,\pi]\times\mathbb{R} with (0,t)(0,t) glued to (π,−t)(\pi,-t), the open Möbius band.

5.2. Since ee has constant rank, near any point one can choose kk columns of ee that are independent; they are smooth sections of the image spanning it, so the image ξ\xi is a subbundle of M×RsM\times\mathbb{R}^s, with complement the image of 1−e1-e. A section of M×RsM\times\mathbb{R}^s is an element v∈C∞(M)sv\in C^\infty(M)^s; it lies in ξ\xi if and only if ev=vev=v, so Γ(ξ)=e C∞(M)s\Gamma(\xi)=e\,C^\infty(M)^s, and C∞(M)s=e C∞(M)s⊕(1−e) C∞(M)sC^\infty(M)^s=e\,C^\infty(M)^s\oplus(1-e)\,C^\infty(M)^s.

5.3. TSn={(x,v):∣x∣=1, ⟨x,v⟩=0}TS^n=\{(x,v):|x|=1,\ \langle x,v\rangle=0\} and the normal bundle of Sn⊂Rn+1S^n\subset\mathbb{R}^{n+1} is trivialized by the section x↦xx\mapsto x. So TSn⊕R≅TSn⊕ν=TRn+1∣SnTS^n\oplus\mathbb{R}\cong TS^n\oplus\nu=T\mathbb{R}^{n+1}|_{S^n}, which is trivial. The map BO(n)→BO(n+1)B\mathrm{O}(n)\to B\mathrm{O}(n+1) classifies ξ↦ξ⊕R\xi\mapsto\xi\oplus\mathbb{R}, so the composite Sn→BO(n)→BO(n+1)S^n\to B\mathrm{O}(n)\to B\mathrm{O}(n+1) classifies a trivial bundle and is null-homotopic by [thm:classify]; hence so is the composite to BOB\mathrm{O}.

5.4. If ω1−ω0=dh\omega_1-\omega_0=dh, let λ ⁣:R→[0,1]\lambda\colon\mathbb{R}\to[0,1] be smooth with λ(0)=0\lambda(0)=0, λ(1)=1\lambda(1)=1, and put Ω=ω0+d(λ(t)h)\Omega=\omega_0+d(\lambda(t)h) on M×RM\times\mathbb{R}. It is closed and restricts to ω0\omega_0 and ω1\omega_1. Conversely, let Ω=αt+gt dt\Omega=\alpha_t+g_t\,dt be closed on M×RM\times\mathbb{R}, with αt\alpha_t a path of 11-forms on MM and gtg_t of functions. The dtdt-component of dΩ=0d\Omega=0 reads ∂tαt=dMgt\partial_t\alpha_t=d_Mg_t, so α1−α0=d∫01gt dt\alpha_1-\alpha_0=d\int_0^1g_t\,dt.

5.5. A line bundle has transition functions in GL1(R)\mathop{\mathrm{GL}}_1(\mathbb{R}), homotopy equivalent to Z/2\mathbb{Z}/2, so BO(1)=RP∞=K(Z/2,1)B\mathrm{O}(1)=\mathbb{RP}^\infty=K(\mathbb{Z}/2,1) and [M,K(Z/2,1)]=H1(M;Z/2)=Hom(π1M,Z/2)[M,K(\mathbb{Z}/2,1)]=H^1(M;\mathbb{Z}/2)=\mathop{\mathrm{Hom}}(\pi_1M,\mathbb{Z}/2) for connected MM. For the torus π1=Z2\pi_1=\mathbb{Z}^2 and there are 44 homomorphisms. For the Klein bottle π1=⟨a,b∣abab−1⟩\pi_1=\langle a,b\mid abab^{-1}\rangle has abelianization Z⊕Z/2\mathbb{Z}\oplus\mathbb{Z}/2, and again Hom(−,Z/2)=(Z/2)2\mathop{\mathrm{Hom}}(-,\mathbb{Z}/2)=(\mathbb{Z}/2)^2 has 44 elements. The determinant of the tangent bundle is trivial for the torus. For the Klein bottle, presented as the square with aa the pair of sides glued by a reflection and bb the pair glued by a translation, the orientation reverses along the loop that crosses the reflected sides, so det⁡T\det T is the line bundle with w1w_1 non-zero on that loop and zero on the other.

Chapter 6

6.1. If g0,g1g_0,g_1 are positive definite symmetric bilinear forms then so is (1−t)g0+tg1(1-t)g_0+tg_1, since (1−t)g0(v,v)+tg1(v,v)>0(1-t)g_0(v,v)+tg_1(v,v)>0 for v≠0v\ne0. Every A∈GLk(R)A\in\mathop{\mathrm{GL}}_k(\mathbb{R}) has a unique polar decomposition A=QPA=QP with QQ orthogonal and P=(ATA)1/2P=(A^{\mathsf T}A)^{1/2} positive definite symmetric, depending continuously on AA. The path At=Q((1−t)P+tI)A_t=Q\bigl((1-t)P+tI\bigr) stays in GLk\mathop{\mathrm{GL}}_k by convexity, fixes O(k)\mathrm{O}(k), and ends in O(k)\mathrm{O}(k); so O(k)\mathrm{O}(k) is a deformation retract. A homotopy equivalence of topological groups that is a homomorphism induces an equivalence of classifying spaces.

6.2. Let e0,e1 ⁣:M→Rq⊂R∞e_0,e_1\colon M\to\mathbb{R}^q\subset\mathbb{R}^\infty be embeddings. Let S,S′ ⁣:R∞→R∞S,S'\colon\mathbb{R}^\infty\to\mathbb{R}^\infty be the linear injections placing a vector in the odd, respectively even, coordinates. The linear map (1−t)id+tS(1-t)\mathrm{id}+tS is injective for t∈[0,1]t\in[0,1]: an element of its kernel would be an eigenvector of SS with eigenvalue −(1−t)/t≤0-(1-t)/t\le0, and SS has none (if Sx=λxSx=\lambda x with λ≠0,1\lambda\ne0,1 then x2=0x_2=0, x3=x2/λ=0x_3=x_2/\lambda=0, and so on, and x1=λx1x_1=\lambda x_1 forces x1=0x_1=0; and SS is injective). An injective linear map composed with an embedding of a compact manifold is an embedding, so (1−t)e1+tSe1(1-t)e_1+tSe_1 is an isotopy from e1e_1 to Se1Se_1; similarly from e0e_0 to S′e0S'e_0. Finally (1−t)Se1+tS′e0(1-t)Se_1+tS'e_0 is an embedding for every tt: for t<1t<1 the odd coordinates are (1−t)e1(1-t)e_1, an injective immersion, and for t>0t>0 the even coordinates are te0te_0. For non-compact MM one uses closed embeddings and the same formulas. The naive homotopy can fail: for e1=−e0e_1=-e_0 the path (1−2t)e0(1-2t)e_0 is constant at t=12t=\frac12.

6.3. Parametrize by arc length, with γ′=T\gamma'=T and n′=−κTn'=-\kappa T (Frenet, with nn the inward normal and κ>0\kappa>0 for a convex curve traversed counterclockwise). Then ∂s=(1−tκ)T\partial_s=(1-t\kappa)T and ∂t=n\partial_t=n, so the determinant is 1−tκ(s)1-t\kappa(s) up to the sign of the frame. The map is a local diffeomorphism where tκ(s)≠1t\kappa(s)\ne1, hence for ∣t∣<1/κmax⁡|t|<1/\kappa_{\max}. For the ellipse (acos⁡u,bsin⁡u)(a\cos u,b\sin u) the curvature is ab/(a2sin⁡2u+b2cos⁡2u)3/2ab/(a^2\sin^2u+b^2\cos^2u)^{3/2}, with maximum a/b2a/b^2 at the ends of the major axis. So r=b2/ar=b^2/a. (Global injectivity may fail earlier in general; for the ellipse it does not on the inward side before t=b2/at=b^2/a, and it never fails on the outward side.)

6.4. The one-point compactification of a product of locally compact spaces is the smash product of the compactifications, which gives the first identity; the second and third are the cases η=Rj\eta=\mathbb{R}^j over a point, where Th=Sj\mathop{\mathrm{Th}}=S^j, and ξ=0\xi=0, where Th=M+\mathop{\mathrm{Th}}=M_+. For the last, RPn+1∖{[0:⋯:0:1]}\mathbb{RP}^{n+1}\setminus\{[0:\dots:0:1]\} is the total space of the tautological bundle over RPn\mathbb{RP}^n by [x:t]↦([x], t x/∣x∣2)[x:t]\mapsto([x],\ t\,x/|x|^2), where t x/∣x∣2t\,x/|x|^2 is a vector in the line [x][x] that is unchanged by (x,t)↦(λx,λt)(x,t)\mapsto(\lambda x,\lambda t); this is a diffeomorphism, and the remaining point is the point at infinity.

6.5. Extend ff radially: F(rx)=rf(x)F(rx)=rf(x) for x∈Sn−1x\in S^{n-1}, 0≤r≤10\le r\le1. This is a homeomorphism of DnD^n (the Alexander trick), smooth away from the origin. The map Σf=Dn∪fDn→Dn∪idDn=Sn\Sigma_f=D^n\cup_fD^n\to D^n\cup_{\mathrm{id}}D^n=S^n that is the identity on the first disc and FF on the second is well defined and bijective, continuous, and hence a homeomorphism. If ff extends to a diffeomorphism GG of DnD^n, the same construction with GG in place of FF is smooth on each disc and compatible with the gluing; it is a diffeomorphism for the smooth structures of [thm:gluing] after adjusting GG near the boundary to respect collars, which is possible by uniqueness of collars ([thm:collar]).

Chapter 7

7.1. For k≠0k\ne0 every point of S1S^1 is a regular value of z↦zkz\mapsto z^k, with ∣k∣|k| preimages, at each of which the derivative θ↦kθ\theta\mapsto k\theta has the sign of kk; so the degree is kk. (For k=0k=0 the map is constant.) The antipodal map is the restriction of −1-1 on Rn+1\mathbb{R}^{n+1}, a composite of n+1n+1 reflections in coordinate hyperplanes, each of which restricts to a diffeomorphism of SnS^n of degree −1-1 (one preimage, orientation reversed). So its degree is (−1)n+1(-1)^{n+1}. If vv is a nowhere-zero vector field on SnS^n, normalized to unit length, then H(x,t)=cos⁡(πt) x+sin⁡(πt) v(x)H(x,t)=\cos(\pi t)\,x+\sin(\pi t)\,v(x) is a homotopy on SnS^n from the identity to the antipodal map, so 1=(−1)n+11=(-1)^{n+1} and nn is odd.

7.2. A point [x][x] is fixed if and only if (A−1)x∈Z2(A-1)x\in\mathbb{Z}^2. The map A−1A-1 is an injective endomorphism of Z2\mathbb{Z}^2 with image of index ∣det⁡(A−1)∣|\det(A-1)|, so the solutions x∈(A−1)−1Z2x\in(A-1)^{-1}\mathbb{Z}^2 form a lattice containing Z2\mathbb{Z}^2 with index ∣det⁡(A−1)∣|\det(A-1)|; modulo Z2\mathbb{Z}^2 there are that many. The differential of the torus map at any point is AA, so each fixed point is non-degenerate with index sign⁡det⁡(1−A)\operatorname{sign}\det(1-A). The sum is det⁡(1−A)=1−trA+det⁡A\det(1-A)=1-\mathop{\mathrm{tr}}A+\det A, which is L(A)L(A) since AA acts on H1=Z2H_1=\mathbb{Z}^2 by AA and on H2=Λ2H1H_2=\Lambda^2H_1 by det⁡A\det A.

7.3. Cyclicity: by the triangle identities, the dual map g∨ ⁣:Y∨→X∨g^\vee\colon Y^\vee\to X^\vee of g ⁣:X→Yg\colon X\to Y satisfies (g⊗1)ηX=(1⊗g∨)ηY(g\otimes1)\eta_X=(1\otimes g^\vee)\eta_Y and ϵY(1⊗g)=ϵX(g∨⊗1)\epsilon_Y(1\otimes g)=\epsilon_X(g^\vee\otimes1). For f ⁣:Y→Xf\colon Y\to X one slides gg around the loop: tr(gf)=ϵYσ(gf⊗1)ηY=ϵYσ(g⊗1)(f⊗1)ηY\mathop{\mathrm{tr}}(gf)=\epsilon_Y\sigma(gf\otimes1)\eta_Y=\epsilon_Y\sigma(g\otimes1)(f\otimes1)\eta_Y, and moving gg through ϵ\epsilon as g∨g^\vee and then through η\eta back as gg on the other side yields ϵXσ(fg⊗1)ηX=tr(fg)\epsilon_X\sigma(fg\otimes1)\eta_X=\mathop{\mathrm{tr}}(fg). Multiplicativity: X⊗YX\otimes Y is dualizable with dual Y∨⊗X∨Y^\vee\otimes X^\vee and structure maps built from those of XX and YY; the trace diagram for f⊗gf\otimes g is the tensor product of the two trace diagrams after rearranging by symmetries, and End⁡(1)\operatorname{End}(\mathbf 1) is commutative. For a graded vector space V=⨁ViV=\bigoplus V_i the coevaluation is 1↦∑ej⊗ej1\mapsto\sum e_j\otimes e^j over a homogeneous basis, and the symmetry sends ej⊗eje_j\otimes e^j to (−1)∣ej∣ ∣ej∣ej⊗ej=(−1)∣ej∣ej⊗ej(-1)^{|e_j|\,|e^j|}e^j\otimes e_j=(-1)^{|e_j|}e^j\otimes e_j, since ∣ej∣=−∣ej∣|e^j|=-|e_j|. Evaluating gives tr(f)=∑j(−1)∣ej∣⟨ej,fej⟩=∑i(−1)itr(f∣Vi)\mathop{\mathrm{tr}}(f)=\sum_j(-1)^{|e_j|}\langle e^j,fe_j\rangle=\sum_i(-1)^i\mathop{\mathrm{tr}}(f|_{V_i}).

7.4. On Sn⊂Rn+1S^n\subset\mathbb{R}^{n+1} let v(x)=en+1−⟨x,en+1⟩xv(x)=e_{n+1}-\langle x,e_{n+1}\rangle x, the projection of the constant field en+1e_{n+1}. It vanishes at the poles ±en+1\pm e_{n+1}. Near the south pole, in the chart y↦(y,−1−∣y∣2)y\mapsto(y,-\sqrt{1-|y|^2}), vv is y↦yy\mapsto y to first order, of index +1+1. Near the north pole it is y↦−yy\mapsto-y, of index (−1)n(-1)^n. So χ(Sn)=1+(−1)n\chi(S^n)=1+(-1)^n. Homologically Hi(Sn)H_i(S^n) is Q\mathbb Q for i=0,ni=0,n and zero otherwise, giving the same.

7.5. The disc D={x2+y2≤1, z=0}D=\{x^2+y^2\le1,\ z=0\} is bounded by KK. The circle LL is {(1+cos⁡t,0,sin⁡t)}\{(1+\cos t,0,\sin t)\}; it meets the plane z=0z=0 at t=0t=0, the point (2,0,0)(2,0,0) outside DD, and at t=πt=\pi, the point (0,0,0)(0,0,0) inside DD. So LL meets DD once, transversally (its tangent there is (0,0,−1)(0,0,-1)), and the linking number is ±1\pm1, the sign depending on the orientations: with DD oriented by the normal e3e_3 and LL by increasing tt, it is −1-1.

Chapter 8

8.1. fθ=−(R+rcos⁡φ)sin⁡θf_\theta=-(R+r\cos\varphi)\sin\theta and fφ=−rsin⁡φcos⁡θf_\varphi=-r\sin\varphi\cos\theta. Since R+rcos⁡φ>0R+r\cos\varphi>0, fθ=0f_\theta=0 forces sin⁡θ=0\sin\theta=0; then cos⁡θ=±1\cos\theta=\pm1 and fφ=0f_\varphi=0 forces sin⁡φ=0\sin\varphi=0. The second derivatives are fθθ=−(R+rcos⁡φ)cos⁡θf_{\theta\theta}=-(R+r\cos\varphi)\cos\theta, fφφ=−rcos⁡φcos⁡θf_{\varphi\varphi}=-r\cos\varphi\cos\theta, fθφ=rsin⁡φsin⁡θ=0f_{\theta\varphi}=r\sin\varphi\sin\theta=0 at the critical points. At (0,0)(0,0): (−(R+r),−r)(-(R+r),-r), index 22. At (0,π)(0,\pi): (−(R−r),+r)(-(R-r),+r), index 11. At (π,π)(\pi,\pi): (R−r,−r)(R-r,-r), index 11. At (π,0)(\pi,0): (R+r,r)(R+r,r), index 00. All are non-degenerate because R>r>0R>r>0.

8.2. Write M(t)=∑νktkM(t)=\sum\nu_kt^k, P(t)=∑βktkP(t)=\sum\beta_kt^k, so M−P=(1+t)QM-P=(1+t)Q. Dividing by 1+t1+t as a formal power series, Q(t)=(M(t)−P(t))∑j(−1)jtjQ(t)=(M(t)-P(t))\sum_j(-1)^jt^j, whose coefficient of tkt^k is ∑i≤k(−1)k−i(νi−βi)\sum_{i\le k}(-1)^{k-i}(\nu_i-\beta_i). Non-negativity of the coefficients of QQ is exactly the stated inequality. For a closed connected orientable surface of genus gg, with F=QF=\mathbb Q, β=(1,2g,1)\beta=(1,2g,1), so ∑νk≥∑βk=2g+2\sum\nu_k\ge\sum\beta_k=2g+2. For a non-orientable surface NhN_h use F=Z/2F=\mathbb{Z}/2: β=(1,h,1)\beta=(1,h,1) and the bound is h+2h+2.

8.3. In the chart xi=1x_i=1 with coordinates (xj)j≠i(x_j)_{j\ne i}, f=(λi+∑j≠iλjxj2)/(1+∑j≠ixj2)=λi+∑j≠i(λj−λi)xj2+O(∣x∣4)f=\bigl(\lambda_i+\sum_{j\ne i}\lambda_jx_j^2\bigr)/\bigl(1+\sum_{j\ne i}x_j^2\bigr) =\lambda_i+\sum_{j\ne i}(\lambda_j-\lambda_i)x_j^2+O(|x|^4). So the origin of each chart is a non-degenerate critical point of index #{j:λj<λi}=i\#\{j:\lambda_j<\lambda_i\}=i. There are no others: a critical point of ff on RPn\mathbb{RP}^n lifts to a critical point of the quadratic form on SnS^n, an eigenvector of diag⁡(λi)\operatorname{diag}(\lambda_i), and the eigenvalues are distinct. So νk=1\nu_k=1 for 0≤k≤n0\le k\le n. With F=Z/2F=\mathbb{Z}/2, βk=1\beta_k=1 for all kk and Q=0Q=0. With F=QF=\mathbb Q, β0=1\beta_0=1, βn=1\beta_n=1 if nn is odd, and the others vanish; for nn even M−P=t+⋯+tn=(1+t)(t+t3+⋯+tn−1)M-P=t+\dots+t^n=(1+t)(t+t^3+\dots+t^{n-1}) and for nn odd M−P=t+⋯+tn−1=(1+t)(t+t3+⋯+tn−2)M-P=t+\dots+t^{n-1}=(1+t)(t+t^3+\dots+t^{n-2}).

8.4. Define μ(x,y)=min⁡(∣y∣,max⁡(∣x∣2−ε,0))\mu(x,y)=\min\bigl(|y|,\sqrt{\max(|x|^2-\varepsilon,0)}\bigr), a continuous function, and let rt(x,y)=(x, yt)r_t(x,y)=(x,\ y_t) where yt=yy_t=y if y=0y=0 and otherwise yt=((1−t)∣y∣+tμ(x,y)) y/∣y∣y_t=\bigl((1-t)|y|+t\mu(x,y)\bigr)\,y/|y|. Then rtr_t is continuous, r0=idr_0=\mathrm{id}, and ∣yt∣≤∣y∣|y_t|\le|y|, so q(rt(x,y))≤q(x,y)≤εq(r_t(x,y))\le q(x,y)\le\varepsilon and the deformation stays in {q≤ε}\{q\le\varepsilon\}. At t=1t=1: if ∣x∣2≤ε|x|^2\le\varepsilon then μ=0\mu=0 and r1(x,y)=(x,0)r_1(x,y)=(x,0), a point of the disc; if ∣x∣2>ε|x|^2>\varepsilon then ∣y1∣2=min⁡(∣y∣2,∣x∣2−ε)|y_1|^2=\min(|y|^2,|x|^2-\varepsilon) and q(r1(x,y))≤−εq(r_1(x,y))\le-\varepsilon. Points of the target are fixed for all tt: on the disc y=0y=0, and on {q≤−ε}\{q\le-\varepsilon\} we have ∣y∣2≤∣x∣2−ε|y|^2\le|x|^2-\varepsilon so μ=∣y∣\mu=|y|.

8.5. ft′=3x2−tf_t'=3x^2-t and ft′′=6xf_t''=6x. For t<0t<0 there are no critical points. For t=0t=0 there is one, at 00, degenerate. For t>0t>0 there are two: x=−t/3x=-\sqrt{t/3} with f′′<0f''<0, a local maximum (index 11) with value 2t3t/3\frac{2t}{3}\sqrt{t/3}, and x=+t/3x=+\sqrt{t/3} with f′′>0f''>0, a local minimum (index 00) with value −2t3t/3-\frac{2t}{3}\sqrt{t/3}. On [−L,L][-L,L] with LL large, ftf_t is increasing near the ends. For t<0t<0 the sublevel sets are intervals [−L,c′][-L,c'], all contractible: the functor is constant after the minimum ft(−L)f_t(-L). For t>0t>0 a second component is born at the level of the local minimum and merges with the first at the level of the local maximum: the degree-zero barcode acquires one finite interval, of length 4t3t/3\frac{4t}{3}\sqrt{t/3}, which shrinks to nothing as t→0+t\to0^+.

Chapter 9

9.1. Let DW=W∪∂WWDW=W\cup_{\partial W}W be the double, a closed manifold of odd dimension, so χ(DW)=0\chi(DW)=0 ([thm:morseineq]). By additivity, χ(DW)=2χ(W)−χ(∂W)\chi(DW)=2\chi(W)-\chi(\partial W). Hence χ(∂W)=2χ(W)\chi(\partial W)=2\chi(W) is even. If dim⁡W\dim W is even, ∂W\partial W is closed of odd dimension and χ(∂W)=0\chi(\partial W)=0. So χ\chi mod 22 vanishes on boundaries; it is additive under disjoint union and multiplicative under products ([thm:euler]), hence a ring homomorphism on N∗\mathfrak N_*. χ(RP2m)=1\chi(\mathbb{RP}^{2m})=1, as RP2m\mathbb{RP}^{2m} has one cell in each dimension 0,…,2m0,\dots,2m.

9.2. The allowed parts are 2,4,5,6,8,9,10,…2,4,5,6,8,9,10,\dots (excluding 1,3,71,3,7). Partitions: n=0n=0: the empty one; n=1n=1: none; n=2n=2: 22; n=3n=3: none; n=4n=4: 44, 2+22{+}2; n=5n=5: 55; n=6n=6: 66, 4+24{+}2, 2+2+22{+}2{+}2; n=7n=7: 5+25{+}2; n=8n=8: 88, 6+26{+}2, 4+44{+}4, 4+2+24{+}2{+}2, 2+2+2+22{+}2{+}2{+}2; n=9n=9: 99, 5+45{+}4, 5+2+25{+}2{+}2; n=10n=10: 1010, 8+28{+}2, 6+46{+}4, 6+2+26{+}2{+}2, 5+55{+}5, 4+4+24{+}4{+}2, 4+2+2+24{+}2{+}2{+}2, 2+2+2+2+22{+}2{+}2{+}2{+}2. The counts are 1,0,1,0,2,1,3,1,5,3,81,0,1,0,2,1,3,1,5,3,8.

9.3. w(RPn)=(1+a)n+1w(\mathbb{RP}^n)=(1+a)^{n+1} with an+1=0a^{n+1}=0, coefficients modulo 22. n=2n=2: (1+a)3=1+a+a2(1+a)^3=1+a+a^2, so w1=aw_1=a, w2=a2w_2=a^2, and w12=w2=a2w_1^2=w_2=a^2 evaluates to 11. n=3n=3: (1+a)4=1+a4=1(1+a)^4=1+a^4=1. n=4n=4: (1+a)5=1+a+a4(1+a)^5=1+a+a^4 since (52),(53)\binom52,\binom53 are even; so w1=aw_1=a, w2=w3=0w_2=w_3=0, w4=a4w_4=a^4, giving (w14,w12w2,w22,w1w3,w4)=(1,0,0,0,1)(w_1^4,w_1^2w_2,w_2^2,w_1w_3,w_4)=(1,0,0,0,1). For RP2×RP2\mathbb{RP}^2\times\mathbb{RP}^2, w=(1+a+a2)(1+b+b2)w=(1+a+a^2)(1+b+b^2): w1=a+bw_1=a+b, w2=a2+ab+b2w_2=a^2+ab+b^2, w3=a2b+ab2w_3=a^2b+ab^2, w4=a2b2w_4=a^2b^2, and the numbers are the coefficients of a2b2a^2b^2 modulo 22: w14=(a+b)4w_1^4=(a+b)^4 gives 6≡06\equiv0; w12w2w_1^2w_2 gives 1+2+1≡01+2+1\equiv0; w22w_2^2 gives 2+1≡12+1\equiv1; w1w3w_1w_3 gives 2≡02\equiv0; w4w_4 gives 11.

9.4. A closed 11-manifold is a finite union of circles, each the boundary of a disc. The Klein bottle is the boundary of the solid Klein bottle, the non-trivial D2D^2-bundle over S1S^1 (the mapping torus of a reflection of D2D^2). By [thm:classification] a closed connected surface is Σg\Sigma_g or NhN_h. Σg\Sigma_g bounds a handlebody. NhN_h is a connected sum of hh copies of RP2\mathbb{RP}^2, and connected sum is cobordant to disjoint union (attach a 11-handle to the product cobordism), so [Nh]=h[RP2][N_h]=h[\mathbb{RP}^2]; this is zero for hh even since every element has order 22, and equal to [RP2]≠0[\mathbb{RP}^2]\ne0 for hh odd by [exer:chimod2].

9.5. A tubular neighbourhood of the origin in Rk\mathbb{R}^k with the standard framing is the open unit ball BB with the identification B≅RkB\cong\mathbb{R}^k, x↦x/(1−∣x∣2)x\mapsto x/(1-|x|^2), which is the identity to first order at 00. The collapse Sk=Rk∪∞→Rk∪∞=SkS^k=\mathbb{R}^k\cup\infty\to\mathbb{R}^k\cup\infty=S^k is this diffeomorphism on BB and constant at ∞\infty outside. The origin is a regular value with exactly one preimage, at which the differential is the identity; so the degree is 11. Reversing one framing vector composes the identification with a reflection, and the degree becomes −1-1. A framed 00-manifold in Rk\mathbb{R}^k is a finite set of points with signs, the collapse has degree the sum of the signs, and by [thm:hopf] the degree classifies maps Sk→SkS^k\to S^k; so Ω0fr≅πk(Sk)≅Z\Omega^{\mathrm{fr}}_0\cong\pi_k(S^k)\cong\mathbb{Z}.

Chapter 10

10.1. The equation x˙=(−x2,x1)\dot x=(-x_2,x_1) is linear with solution Ft(x)=(x1cos⁡t−x2sin⁡t, x1sin⁡t+x2cos⁡t)F_t(x)=(x_1\cos t-x_2\sin t,\ x_1\sin t+x_2\cos t), rotation by angle tt. For Ft(x)=etxF_t(x)=e^tx, ∂tFt(x)=etx=Ft(x)\partial_tF_t(x)=e^tx=F_t(x), so G(y,t)=yG(y,t)=y. For x˙=x2\dot x=x^2, the solution through x0>0x_0>0 is x0/(1−x0t)x_0/(1-x_0t), which is unbounded as t→1/x0t\to1/x_0; so for every t>0t>0 the time-tt map is undefined at x0≥1/tx_0\ge1/t, and there is no diffeotopy.

10.2. The rotation number is the degree of t↦γ′(t)/∣γ′(t)∣t\mapsto\gamma'(t)/|\gamma'(t)|. Circle: the tangent turns once; 11. Figure eight: γ′=(cos⁡t,cos⁡2t)\gamma'=(\cos t,\cos2t); the curve has the symmetry t↦t+πt\mapsto t+\pi, γ↦(−x,y)\gamma\mapsto(-x,y), under which the contributions of the two lobes to the total turning cancel; 00. Limaçon: in complex notation γ=(1+2cos⁡θ)eiθ=eiθ+e2iθ+1\gamma=(1+2\cos\theta)e^{i\theta}=e^{i\theta}+e^{2i\theta}+1, so γ′=ieiθ(1+2eiθ)\gamma'=ie^{i\theta}(1+2e^{i\theta}); the factor eiθe^{i\theta} has winding number 11 and 1+2eiθ1+2e^{i\theta} winds once around 00 since ∣2eiθ∣>1|2e^{i\theta}|>1; total 22. By [cor:WG] no two of the three are regularly homotopic.

10.3. A homotopy inverse sends (p,A)(p,A) to the embedding x↦p+Axx\mapsto p+Ax. One composite is the identity. For the other, given an embedding ff define, for 0<t≤10<t\le1, ft(x)=f(0)+(f(tx)−f(0))/tf_t(x)=f(0)+\bigl(f(tx)-f(0)\bigr)/t, and f0(x)=f(0)+df0(x)f_0(x)=f(0)+df_0(x). Each ftf_t is an embedding (a rescaling of the restriction of ff to a smaller disc), f1=ff_1=f, and by Taylor’s theorem ft(x)=f(0)+df0(x)+t r(x,t)f_t(x)=f(0)+df_0(x)+t\,r(x,t) with rr smooth, so the family is smooth down to t=0t=0 and depends continuously on ff. It preserves (f(0),df0)(f(0),df_0) throughout. So the space of embeddings deformation retracts to the affine ones.

10.4. Parametrize so that Ft(x)=(x,0,0)F_t(x)=(x,0,0) unless ∣x−1/(1−t)∣<1|x-1/(1-t)|<1, where a fixed knotted arc is inserted, for t<1t<1; put F1(x)=(x,0,0)F_1(x)=(x,0,0). On the open set {(x,t):x<R}\{(x,t):x<R\} we have Ft(x)=(x,0,0)F_t(x)=(x,0,0) as soon as 1/(1−t)>R+11/(1-t)>R+1, so FF is smooth near every point with t=1t=1; it is smooth elsewhere by construction, and each FtF_t is an embedding. If a diffeotopy Φt\Phi_t of R3\mathbb{R}^3 satisfied Φ1∘F0=F1\Phi_1\circ F_0=F_1, then Φ1\Phi_1 would be a diffeomorphism of R3\mathbb{R}^3 carrying the knotted line to the straight line, hence a homeomorphism of the complements. The complement of the straight line has fundamental group Z\mathbb{Z}, and the complement of a long knot has the group of the corresponding closed knot, which is not abelian if the knot is non-trivial.

10.5. Here m=1m=1 and n=4=2m+2n=4=2m+2, so [thm:whitneyiso] applies. In R3\mathbb{R}^3 the dimension count for double points of a one-parameter family gives 2m+1−n=02m+1-n=0: a generic path of maps S1→R3S^1\to\mathbb{R}^3 has isolated times at which the curve crosses itself transversally, and these crossings are exactly the events that change the knot type. They cannot be removed by perturbation because a zero-dimensional transverse intersection is stable.

Chapter 11

11.1. For a word of length 2n2n, χ=V−n+1\chi=V-n+1, and the surface is orientable if and only if each letter occurs with both exponents. aba−1b−1aba^{-1}b^{-1}: all four corners are identified, V=1V=1, χ=0\chi=0, orientable: the torus. abababab: V=2V=2, χ=1\chi=1, non-orientable: RP2\mathbb{RP}^2. abab−1abab^{-1}: V=1V=1, χ=0\chi=0, non-orientable: the Klein bottle. aabbaabb: V=1V=1, χ=0\chi=0, non-orientable: the Klein bottle again, as P#PP\#P. abca−1b−1c−1abca^{-1}b^{-1}c^{-1}: V=2V=2, χ=2−3+1=0\chi=2-3+1=0, orientable: the torus. abcabcabcabc: V=3V=3, χ=3−3+1=1\chi=3-3+1=1, non-orientable: RP2\mathbb{RP}^2. (The vertex counts are those returned by the script.)

11.2. χ(M#N)=χ(M)+χ(N)−2\chi(M\#N)=\chi(M)+\chi(N)-2 because a disc is removed from each (lowering each χ\chi by 11) and the results are glued along a circle, of Euler characteristic 00. So χ(T#P)=0+1−2=−1\chi(T\#P)=0+1-2=-1 and χ(P#P#P)=1+1+1−4=−1\chi(P\#P\#P)=1+1+1-4=-1. Both are closed, connected and non-orientable (a connected sum is orientable only if both summands are). By [thm:classification] they are diffeomorphic, to N3N_3.

11.3. (a) The pairing matrix in the basis (1,x)(1,x) is (0110)\bigl(\begin{smallmatrix}0&1\\1&0\end{smallmatrix}\bigr), so 1#=x1^\#=x and x#=1x^\#=1, and w=1⋅x+x⋅1=2xw=1\cdot x+x\cdot1=2x. Then ε(w0)=ε(1)=0\varepsilon(w^0)=\varepsilon(1)=0, ε(w)=2\varepsilon(w)=2, and w2=4x2=0w^2=4x^2=0. (b) ε(g⋅h)=δgh,e/n\varepsilon(g\cdot h)=\delta_{gh,e}/n, so g#=n g−1g^\#=n\,g^{-1} and w=∑gg⋅ng−1=n2ew=\sum_gg\cdot ng^{-1}=n^2e; hence ε(wg)=n2gε(e)=n2g−1\varepsilon(w^g)=n^{2g}\varepsilon(e)=n^{2g-1}. A homomorphism π1Σg→G\pi_1\Sigma_g\to G to an abelian group is a homomorphism from H1=Z2gH_1=\mathbb{Z}^{2g}, and there are n2gn^{2g} of them; dividing by nn gives the same number. Semisimple case: eiej=δijeie_ie_j=\delta_{ij}e_i, so the pairing is diagonal with entries θi\theta_i, the dual basis is ei#=ei/θie_i^\#=e_i/\theta_i, and w=∑iei/θiw=\sum_ie_i/\theta_i. Then wg=∑iei/θigw^g=\sum_ie_i/\theta_i^g and ε(wg)=∑iθi1−g\varepsilon(w^g)=\sum_i\theta_i^{1-g}.

11.4. By [cor:pushout], WW is obtained up to homotopy from f−1(a)f^{-1}(a), a union of circles with χ=0\chi=0, by attaching a 11-cell; so χ(W)=−1\chi(W)=-1. Both f−1(a)f^{-1}(a) and f−1(b)f^{-1}(b) are non-empty: the minimum and maximum of ff are not attained at the only critical point, a saddle, so they are attained on the boundary. Hence WW has b≥2b\ge2 boundary components. If WW is orientable, −1=2−2g−b-1=2-2g-b gives (g,b)=(0,3)(g,b)=(0,3), as (1,1)(1,1) is excluded: a sphere with three holes. If not, −1=2−h−b-1=2-h-b with h≥1h\ge1, b≥2b\ge2 gives (h,b)=(1,2)(h,b)=(1,2): a projective plane with two holes, which is a Möbius band with one hole.

11.5. The torus is the composite ∅→ηS1→δS1⊔S1→μS1→ϵ∅\emptyset\xrightarrow{\eta}S^1\xrightarrow{\delta}S^1\sqcup S^1\xrightarrow{\mu}S^1\xrightarrow{\epsilon} \emptyset. Let β=ϵμ\beta=\epsilon\mu, the pairing, and γ=δη\gamma=\delta\eta, the copairing. The Frobenius relation and the unit and counit laws give the zigzag identity (β⊔id)(id⊔γ)=id(\beta\sqcup\mathrm{id})(\mathrm{id}\sqcup\gamma)=\mathrm{id}: indeed (ϵ⊔id)(μ⊔id)(id⊔δ)(id⊔η)=(ϵ⊔id) δ μ (id⊔η)=id(\epsilon\sqcup\mathrm{id})(\mu\sqcup\mathrm{id})(\mathrm{id}\sqcup\delta)(\mathrm{id}\sqcup\eta) =(\epsilon\sqcup\mathrm{id})\,\delta\,\mu\,(\mathrm{id}\sqcup\eta)=\mathrm{id}. So Z(γ)Z(\gamma) and Z(β)Z(\beta) exhibit A=Z(S1)A=Z(S^1) as its own dual, with Z(γ)Z(\gamma) the coevaluation and Z(β)Z(\beta) the evaluation, and by commutativity Z(T2)=Z(β)Z(γ)=tr(idA)=dim⁡AZ(T^2)=Z(\beta)Z(\gamma)=\mathop{\mathrm{tr}}(\mathrm{id}_A)=\dim A.

References

References

  1. M. Ando, A. J. Blumberg, D. Gepner, M. J. Hopkins, C. Rezk, An ∞\infty-categorical approach to RR-line bundles, RR-module Thom spectra, and twisted RR-homology, J. Topol. 7 (2014), 869–893.

  2. L. Abrams, Two-dimensional topological quantum field theories and Frobenius algebras, J. Knot Theory Ramifications 5 (1996), 569–587.

  3. D. Ayala, J. Francis, Factorization homology of topological manifolds, J. Topol. 8 (2015), 1045–1084.

  4. M. F. Atiyah, Thom complexes, Proc. London Math. Soc. (3) 11 (1961), 291–310.

  5. K. Bauer, M. Burke, M. Ching, Tangent ∞\infty-categories and Goodwillie calculus, preprint, arXiv:2101.07819.

  6. J. C. Baez, J. Dolan, Higher-dimensional algebra and topological quantum field theory, J. Math. Phys. 36 (1995), 6073–6105.

  7. D. Berwick-Evans, P. Boavida de Brito, D. Pavlov, Classifying spaces of infinity-sheaves, Algebr. Geom. Topol. 24 (2024), 4891–4937.

  8. J. C. Baez, A. E. Hoffnung, Convenient categories of smooth spaces, Trans. Amer. Math. Soc. 363 (2011), 5789–5825.

  9. P. Boavida de Brito, M. Weiss, Manifold calculus and homotopy sheaves, Homology Homotopy Appl. 15 (2013), 361–383.

  10. J. R. B. Cockett, G. S. H. Cruttwell, Differential structure, tangent structure, and SDG, Appl. Categ. Structures 22 (2014), 331–417.

  11. R. Cockett, G. Cruttwell, J. Gallagher, J.-S. P. Lemay, B. MacAdam, G. Plotkin, D. Pronk, Reverse derivative categories, in Computer Science Logic 2020, LIPIcs 152, 18:1–18:16.

  12. J. Cerf, La stratification naturelle des espaces de fonctions différentiables réelles et le théorème de la pseudo-isotopie, Publ. Math. IHÉS 39 (1970), 5–173.

  13. R. L. Cohen, J. D. S. Jones, G. B. Segal, Morse theory and classifying spaces, preprint, 1995.

  14. D. Carchedi, P. Steffens, On the universal property of derived manifolds, preprint, arXiv:1905.06195.

  15. J. D. Christensen, G. Sinnamon, E. Wu, The DD-topology for diffeological spaces, Pacific J. Math. 272 (2014), 87–110.

  16. A. Dold, D. Puppe, Duality, trace, and transfer, in Proceedings of the International Conference on Geometric Topology (Warsaw, 1978), PWN, 1980, 81–102.

  17. E. J. Dubuc, Sur les modèles de la géométrie différentielle synthétique, Cahiers Topologie Géom. Différentielle 20 (1979), 231–279.

  18. C. J. Earle, J. Eells, A fibre bundle description of Teichmüller theory, J. Differential Geometry 3 (1969), 19–43.

  19. Y. M. Eliashberg, N. M. Mishachev, The space of framed functions is contractible, in Essays in Mathematics and its Applications, Springer, 2012, 81–109.

  20. B. Fong, D. I. Spivak, R. Tuyéras, Backprop as functor: a compositional perspective on supervised learning, in LICS 2019, IEEE, 2019.

  21. T. G. Goodwillie, J. R. Klein, Multiple disjunction for spaces of smooth embeddings, J. Topol. 8 (2015), 651–674.

  22. S. Galatius, I. Madsen, U. Tillmann, M. Weiss, The homotopy type of the cobordism category, Acta Math. 202 (2009), 195–239.

  23. V. Godin, Higher string topology operations, preprint, arXiv:0711.4859.

  24. D. Grady, D. Pavlov, The geometric cobordism hypothesis, preprint, arXiv:2111.01095.

  25. T. G. Goodwillie, M. Weiss, Embeddings from the point of view of immersion theory, Part II, Geom. Topol. 3 (1999), 103–118.

  26. M. W. Hirsch, Immersions of manifolds, Trans. Amer. Math. Soc. 93 (1959), 242–276.

  27. M. W. Hirsch, Differential Topology, Graduate Texts in Mathematics 33, Springer, 1976.

  28. K. Igusa, The space of framed functions, Trans. Amer. Math. Soc. 301 (1987), 431–477.

  29. P. Iglesias-Zemmour, Diffeology, Mathematical Surveys and Monographs 185, American Mathematical Society, 2013.

  30. D. Joyce, Algebraic Geometry over C∞C^\infty-rings, Mem. Amer. Math. Soc. 260 (2019), no. 1256.

  31. A. Kriegl, P. W. Michor, The Convenient Setting of Global Analysis, Mathematical Surveys and Monographs 53, American Mathematical Society, 1997.

  32. I. Kolář, P. W. Michor, J. Slovák, Natural Operations in Differential Geometry, Springer, 1993.

  33. J. Kock, Frobenius Algebras and 2D Topological Quantum Field Theories, London Math. Soc. Student Texts 59, Cambridge University Press, 2004.

  34. A. Kock, Synthetic Differential Geometry, 2nd ed., London Math. Soc. Lecture Note Series 333, Cambridge University Press, 2006.

  35. P. Leung, Classifying tangent structures using Weil algebras, Theory Appl. Categ. 32 (2017), 286–337.

  36. J. Lurie, On the classification of topological field theories, in Current Developments in Mathematics 2008, International Press, 2009, 129–280.

  37. A. W. Macpherson, The universal property of derived geometry, preprint, arXiv:1701.08359.

  38. J. P. May, The additivity of traces in triangulated categories, Adv. Math. 163 (2001), 34–73.

  39. I. Moerdijk, G. E. Reyes, Models for Smooth Infinitesimal Analysis, Springer, 1991.

  40. I. Madsen, M. Weiss, The stable moduli space of Riemann surfaces: Mumford’s conjecture, Ann. of Math. (2) 165 (2007), 843–941.

  41. R. S. Palais, Local triviality of the restriction map for embeddings, Comment. Math. Helv. 34 (1960), 305–312.

  42. L. S. Pontryagin, Smooth manifolds and their applications in homotopy theory, Trudy Mat. Inst. Steklov 45 (1955); English transl. in Amer. Math. Soc. Transl. (2) 11 (1959).

  43. K. Ponto, M. Shulman, Traces in symmetric monoidal categories, Expo. Math. 32 (2014), 248–273.

  44. J. Rosický, Abstract tangent functors, Diagrammes 12 (1984), JR1–JR11.

  45. U. Schreiber, Differential cohomology in a cohesive infinity-topos, preprint, arXiv:1310.7930.

  46. S. Smale, The classification of immersions of spheres in Euclidean spaces, Ann. of Math. (2) 69 (1959), 327–344.

  47. C. J. Schommer-Pries, The classification of two-dimensional extended topological field theories, Ph.D. thesis, University of California, Berkeley, 2009; arXiv:1112.1000.

  48. D. I. Spivak, Derived smooth manifolds, Duke Math. J. 153 (2010), 55–128.

  49. R. Thom, Quelques propriétés globales des variétés différentiables, Comment. Math. Helv. 28 (1954), 17–86.

  50. M. Weiss, Embeddings from the point of view of immersion theory, Part I, Geom. Topol. 3 (1999), 67–101.

  51. A. Zomorodian, G. Carlsson, Computing persistent homology, Discrete Comput. Geom. 33 (2005), 249–274.

Complete flattened LaTeX source
\documentclass[11pt,openany]{book}
\usepackage[T1]{fontenc}

\usepackage[utf8]{inputenc}

\usepackage[sc,osf]{mathpazo}

\linespread{1.05}
\usepackage[scaled=0.9]{helvet}

\usepackage[paperwidth=7in,paperheight=10in,inner=0.95in,outer=0.85in,top=0.95in,bottom=1.0in,headsep=14pt]{geometry}

\usepackage{amsmath,amssymb,amsthm,mathtools,stmaryrd}

\usepackage{microtype}

\usepackage{xcolor}

\definecolor{accent}{HTML}{7A1F2B}
\definecolor{ink2}{HTML}{4A5560}
\definecolor{fwd}{HTML}{1F4E79}
\definecolor{bwd}{HTML}{A8432C}
\usepackage{tikz,tikz-cd}

\usetikzlibrary{arrows.meta,positioning,calc,decorations.markings,shapes.geometric,backgrounds,fit}
\usepackage{booktabs,array,tabularx,longtable}

\usepackage{enumitem}

\setlist{itemsep=2pt,topsep=4pt}
\usepackage{titlesec}

\usepackage{fancyhdr}

\usepackage[font=small,labelfont={bf,sf},margin=1.5em]{caption}

\usepackage[colorlinks=true,linkcolor=accent,citecolor=accent,urlcolor=accent,
  pdftitle={Differential Topology in Higher Categories},
  pdfsubject={Differential topology in higher categorical semantics},
  pdfkeywords={differential topology, transversality, derived manifolds, smooth sets, cobordism, Morse theory, isotopy, mapping spaces}]{hyperref}

\usepackage[nameinlink,capitalise,noabbrev]{cleveref}


\titleformat{\chapter}[display]{\normalfont\Large\scshape\color{accent}}{\chaptertitlename\ \thechapter}{4pt}{\normalfont\huge\bfseries\color{black}}
\titlespacing*{\chapter}{0pt}{10pt}{26pt}
\titleformat{\section}{\normalfont\large\bfseries}{\thesection}{0.7em}{}
\titleformat{\subsection}{\normalfont\normalsize\bfseries\itshape}{\thesubsection}{0.7em}{}
\titleformat{\part}[display]{\normalfont\centering\Large\scshape\color{accent}}{\partname\ \thepart}{12pt}{\normalfont\Huge\bfseries\color{black}}
\pagestyle{fancy}\fancyhf{}
\fancyhead[LE]{\small\thepage\quad\textsc{\nouppercase{\leftmark}}}
\fancyhead[RO]{\small\textsc{\nouppercase{\rightmark}}\quad\thepage}
\renewcommand{\headrulewidth}{0pt}
\renewcommand{\chaptermark}[1]{\markboth{#1}{#1}}
\renewcommand{\sectionmark}[1]{\markright{\thesection\ \ #1}}
\fancypagestyle{plain}{\fancyhf{}\fancyfoot[C]{\small\thepage}}
\setlength{\headheight}{14pt}
\setcounter{tocdepth}{1}
\setcounter{secnumdepth}{2}
\raggedbottom
\emergencystretch=2em
\tolerance=1200
\hbadness=2000

\theoremstyle{plain}
\newtheorem{theorem}{Theorem}[chapter]
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{principle}[theorem]{Principle}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\newtheorem{construction}[theorem]{Construction}
\newtheorem{problem}[theorem]{Open Problem}
\newtheorem{exercise}{Exercise}[chapter]
\theoremstyle{remark}
\newtheorem{remark}[theorem]{Remark}
\crefname{principle}{Principle}{Principles}
\crefname{construction}{Construction}{Constructions}
\crefname{conjecture}{Conjecture}{Conjectures}
\crefname{exercise}{Exercise}{Exercises}
\crefname{problem}{Open Problem}{Open Problems}

\newcommand{\stag}[2]{{\footnotesize\sffamily\upshape\color{#1}[#2]}\enspace\ignorespaces}
\newcommand{\Scl}[1][]{\stag{ink2}{classical\if\relax\detokenize{#1}\relax\else; #1\fi}}
\newcommand{\Sli}[1][]{\stag{fwd}{literature\if\relax\detokenize{#1}\relax\else; #1\fi}}
\newcommand{\Srf}[1][]{\stag{accent}{reformulation\if\relax\detokenize{#1}\relax\else; cf.\ #1\fi}}
\newcommand{\Sob}{\stag{accent}{observation}}
\newcommand{\Spg}{\stag{bwd}{programme}}

\newcommand{\cat}[1]{\mathsf{#1}}
\newcommand{\Set}{\cat{Set}}
\newcommand{\FinSet}{\cat{FinSet}}
\newcommand{\Meas}{\cat{Meas}}
\newcommand{\Stoch}{\cat{Stoch}}
\newcommand{\FinStoch}{\cat{FinStoch}}
\newcommand{\BorelStoch}{\cat{BorelStoch}}
\newcommand{\FinProb}{\cat{FinProb}}
\newcommand{\FinStat}{\cat{FinStat}}
\newcommand{\Krn}{\cat{Krn}}
\newcommand{\Hilb}{\cat{Hilb}}
\newcommand{\Ban}{\cat{Ban}}
\newcommand{\Src}{\cat{Src}}
\newcommand{\Chan}{\cat{Chan}}
\newcommand{\Dyn}{\cat{Dyn}}
\newcommand{\Cat}{\cat{Cat}}
\newcommand{\CC}{\mathcal{C}}
\newcommand{\FF}{\mathcal{F}}
\newcommand{\GG}{\mathcal{G}}
\newcommand{\II}{\mathcal{I}}
\newcommand{\Sinf}{\mathcal{S}}
\newcommand{\Qlaw}{\mathcal{Q}}
\newcommand{\Fun}{\mathcal{F}}
\newcommand{\Dist}{\Delta}
\newcommand{\Giry}{\mathcal{G}}
\newcommand{\R}{\mathbb{R}}
\newcommand{\Rbar}{\overline{\mathbb{R}}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\Fp}{\mathbb{F}_p}
\newcommand{\E}{\mathbb{E}}
\newcommand{\Prob}{\mathbb{P}}
\DeclareMathOperator*{\argmax}{arg\,max}
\DeclareMathOperator*{\colim}{colim}
\DeclareMathOperator{\Hom}{Hom}
\DeclareMathOperator{\Ext}{Ext}
\DeclareMathOperator{\Fix}{Fix}
\DeclareMathOperator{\supp}{supp}
\DeclareMathOperator{\TV}{TV}
\DeclareMathOperator{\rk}{rk}
\newcommand{\id}{\mathrm{id}}
\newcommand{\op}{\mathrm{op}}
\newcommand{\adj}{\dashv}
\newcommand{\iso}{\cong}
\newcommand{\st}{\mid}
\newcommand{\cpy}{\mathrm{copy}}
\newcommand{\del}{\mathrm{del}}
\newcommand{\KL}[2]{D(#1\,\|\,#2)}
\newcommand{\hb}{h}
\newcommand{\AZ}{A^{\Z}}
\newcommand{\BZ}{B^{\Z}}
\newcommand{\term}[1]{\textbf{\itshape #1}}

\newcommand{\Mfd}{\cat{Mfd}}
\newcommand{\Cart}{\cat{Cart}}
\newcommand{\SmSet}{\cat{SmSet}}
\newcommand{\Dflg}{\cat{Dflg}}
\newcommand{\CinfRing}{C^\infty\cat{Ring}}
\newcommand{\DMfd}{\cat{DMfd}}
\newcommand{\Spc}{\mathcal{S}}
\newcommand{\Sp}{\cat{Sp}}
\newcommand{\Cob}{\cat{Cob}}
\newcommand{\Bord}{\cat{Bord}}
\newcommand{\Vect}{\cat{Vect}}
\newcommand{\Top}{\cat{Top}}
\newcommand{\Para}{\cat{Para}}
\DeclareMathOperator{\Diff}{Diff}
\DeclareMathOperator{\Emb}{Emb}
\DeclareMathOperator{\Imm}{Imm}
\DeclareMathOperator{\Map}{Map}
\DeclareMathOperator{\Sing}{Sing}
\DeclareMathOperator{\Fr}{Fr}
\DeclareMathOperator{\Th}{Th}
\DeclareMathOperator{\Gr}{Gr}
\DeclareMathOperator{\GL}{GL}
\DeclareMathOperator{\ind}{ind}
\DeclareMathOperator{\codim}{codim}
\DeclareMathOperator{\vdim}{vdim}
\DeclareMathOperator{\tr}{tr}
\DeclareMathOperator{\grad}{grad}
\DeclareMathOperator{\Tub}{Tub}
\DeclareMathOperator{\Conf}{Conf}
\DeclareMathOperator{\diag}{diag}
\DeclareMathOperator{\relu}{ReLU}
\newcommand{\Or}{\mathrm{O}}
\newcommand{\SO}{\mathrm{SO}}
\newcommand{\RP}{\mathbb{RP}}
\newcommand{\CP}{\mathbb{CP}}
\newcommand{\C}{\mathbb{C}}
\newcommand{\trans}{\pitchfork}
\newcommand{\pt}{\mathrm{pt}}
\newcommand{\loss}{\mathcal{L}}

\tikzset{
  ogbox/.style={draw, thick, rounded corners=2pt, minimum width=1.25cm, minimum height=1.05cm, fill=black!4},
  fw/.style={-{Stealth[length=5pt]}, thick, fwd},
  bw/.style={-{Stealth[length=5pt]}, thick, bwd, densely dashed},
  dec/.style={circle, fill, inner sep=1.7pt},
  nat/.style={circle, draw, thick, fill=white, inner sep=1.7pt},
  leaf/.style={inner sep=2pt, font=\small},
  lab/.style={font=\small, inner sep=1.5pt},
  infoset/.style={densely dashed, thick, accent},
  st/.style={circle, draw, thick, fill=white, inner sep=1.6pt},
  ty/.style={rectangle, draw, thick, fill=black!8, inner sep=2.6pt},
}
\newcommand{\sol}[1]{\par\medskip\noindent\textbf{#1.}\enspace\ignorespaces}
\newcommand{\bookauthor}{Anonymous}
\begin{document}
\frontmatter
\begin{titlepage}
\thispagestyle{empty}
\vspace*{1.1in}
\noindent{\Large\scshape\color{accent} Higher Categorical Semantics}\par
\vspace{0.9in}
\noindent{\fontsize{30}{38}\selectfont\bfseries\raggedright Differential Topology\\ in Higher Categories\par}
\vspace{0.35in}
\noindent{\Large\itshape Mapping objects, derived intersections,\\[2pt] spaces of choices, and cobordism}\par
\vfill
\noindent\begin{tikzpicture}[x=1cm,y=1cm]
  \draw[thick,fwd] (-0.2,0) .. controls (1.2,1.5) and (2.4,-1.5) .. (3.8,0.2);
  \draw[thick,bwd] (-0.2,0.25) -- (3.8,-0.1);
  \node[dec] at (0.07,0.23) {};
  \node[dec] at (1.83,0.07) {};
  \node[dec] at (3.63,-0.08) {};
  \node[lab] at (4.6,0.05) {$=$};
  \node[lab] (P) at (5.6,0.7) {$M\times_N P$};
  \node[lab] (M) at (5.6,-0.7) {$M$};
  \node[lab] (Q) at (7.8,0.7) {$P$};
  \node[lab] (N) at (7.8,-0.7) {$N$};
  \draw[fw,black] (P) -- (M);
  \draw[fw,black] (P) -- (Q);
  \draw[fw,black] (M) -- (N);
  \draw[fw,black] (Q) -- (N);
  \draw[thick] (6.0,0.25) -- (6.25,0.25) -- (6.25,0.45);
\end{tikzpicture}\par
\vspace{0.5in}
\noindent{\large \bookauthor}\par
\vspace{4pt}
\noindent{\small 2026}
\end{titlepage}

\thispagestyle{empty}
\vspace*{\fill}
{\small
\noindent\textbf{About this text.}
This monograph is a higher-categorical treatment of the territory mapped by Morris~W.~Hirsch's
\emph{Differential Topology} (Graduate Texts in Mathematics~33, Springer, 1976): manifolds and maps, function
spaces, transversality, vector bundles and tubular neighbourhoods, degrees and intersection numbers, Morse
theory, cobordism, isotopy, and the classification of surfaces. It is an independent work. It reproduces none
of that book's text or proofs verbatim; it is not affiliated with or endorsed by its publisher.
\Cref{app:concordance} gives a concordance so that the two can be read side by side.

\medskip
\noindent\textbf{Status of results.}
Every numbered statement carries a tag: \textsf{classical} (standard differential topology, attributed),
\textsf{literature} (established in the categorical, homotopical or derived-geometric literature, attributed),
\textsf{reformulation} (a known result restated in a different language, with the restatement proved or
sketched here), \textsf{observation} (an elementary remark we have not found stated in this form), or
\textsf{programme} (a conjecture or a direction, not a result). No cited theorem is claimed as new.

\medskip
\noindent\textbf{Authorship and assistance.}
The text was drafted and typeset with a large language model (Anthropic's Claude) under the direction of
the submitting author, who is responsible for it. Generative AI assisted with drafting, literature recall,
mathematical consistency checks, computation, and typesetting. The numerical examples and a number of finite
structural claims are recomputed by the accompanying script \texttt{verify-differential-topology.py} (\cref{app:computations}).
Proofs and bibliographic details have not been independently refereed; corrections are welcome.

\medskip
\noindent Open access. Rights remain with the author.
}
\clearpage

\tableofcontents

\chapter{Preface}

Hirsch's \emph{Differential Topology} \cite{Hir76} is a book about what can be done to a smooth map by moving it a
little. Its method is uniform across nine chapters. One topologizes the set $C^r(M,N)$ of maps; one shows
that the maps with a desirable property (immersions, embeddings, maps transverse to a submanifold, Morse
functions) form an open and dense, or at least residual, subset; one defines an invariant using a map in
that subset; and one shows that the invariant does not depend on which map was used, because any two such
maps are joined by a path that is itself generic one dimension up. Degree, intersection number, Euler
characteristic, the cobordism class of a preimage, and the diffeomorphism type of a surface are all obtained
in this way.

Each step of that method has a counterpart in the language of higher categories, and the counterparts are
by now theorems in the literature and no longer analogies. The set of maps is the object of global points of
a mapping object in a topos of smooth sets, whose homotopy type is its \emph{shape}; the classification
theorems of the subject (for vector bundles, for immersions, for manifolds up to cobordism) compute such
shapes. Transversality is the condition under which a fibre product of manifolds exists and is correct; the
category of manifolds has a universal completion in which every fibre product is correct, the
$\infty$-category of derived manifolds, and the genericity theorems say that every object of the completion
can be deformed back into the original category. Independence of choices is the statement that a space of
choices (of tubular neighbourhoods, collars, metrics, embeddings in $\R^\infty$) is contractible, which is
what makes a construction functorial. Invariants are traces. Decompositions by Morse functions are
presentations by generators and relations of cobordism categories.

This book follows Hirsch's order and states each of his main theorems in that language, with attribution. It
is not a replacement for his book, and it does not reprove his analytic theorems. The inverse function
theorem, partitions of unity, the Morse--Sard theorem, and the existence and uniqueness of solutions of
ordinary differential equations are used as stated. What changes is the form of the conclusions, and the last
chapter says plainly how much that change is worth. In brief: of the three volumes in this series so far,
this is the one in which the higher-categorical statements carry the most real content, because the
subject is where a good part of higher category theory came from; and it is also the one in which the
gap between a formal statement and its proof is widest, because the proofs are analysis.

A companion volume, \emph{Parametrized Neural Networks and Their Representations in Differential Topology},
applies the results of this book to smooth families of maps given by neural network architectures. It was
written to be read after this one and cites it by chapter.

\chapter{Introduction}

\section*{What is assumed}

The reader should know the definition of a smooth manifold and of a smooth map, the inverse function
theorem, and the language of categories, functors, limits and adjunctions. Sheaves on a site, simplicial
sets, and the idea of an $\infty$-category as a category with spaces of morphisms are used from
\cref{ch:manifolds} on; \cref{app:toolkit} collects the definitions and fixes the notation. Spectra appear in
\cref{ch:tubular,ch:degree,ch:cobordism} and are recalled where they enter. No result from stable homotopy
theory beyond the statement of the Pontryagin--Thom theorem is needed to follow the text.

\section*{How the book is organized}

The four parts follow the logical order of the source. Part~I builds the ambient categories and treats the
two genericity mechanisms, function spaces and transversality. Part~II treats bundles and the invariants
that come from them. Part~III treats the three ways of taking a manifold apart or putting manifolds in
families: Morse functions, cobordisms, isotopies; and ends with surfaces, where all three can be carried out
completely. Part~IV is an assessment. \Cref{fig:plan} shows the dependencies.

\begin{figure}[ht]
\centering
\begin{tikzpicture}[x=1cm,y=1cm,every node/.style={font=\small}]
  \node[ogbox] (c1) at (0,0) {1};
  \node[ogbox] (c2) at (2.3,0.9) {2};
  \node[ogbox] (c3) at (2.3,-0.9) {3};
  \node[ogbox] (c4) at (4.6,0) {4};
  \node[ogbox] (c5) at (6.9,0.9) {5};
  \node[ogbox] (c6) at (9.2,0.9) {6};
  \node[ogbox] (c7) at (11.5,0.9) {7};
  \node[ogbox] (c8) at (6.9,-0.9) {8};
  \node[ogbox] (c9) at (9.2,-0.9) {9};
  \node[ogbox] (c10) at (11.5,-0.9) {10};
  \node[ogbox] (c11) at (11.5,-2.7) {11};
  \draw[fw] (c1) -- (c2); \draw[fw] (c1) -- (c3);
  \draw[fw] (c2) -- (c4); \draw[fw] (c3) -- (c4);
  \draw[fw] (c4) -- (c5); \draw[fw] (c5) -- (c6); \draw[fw] (c6) -- (c7);
  \draw[fw] (c4) -- (c8); \draw[fw] (c8) -- (c9); \draw[fw] (c6) -- (c9);
  \draw[fw] (c6) -- (c10); \draw[fw] (c9) -- (c11); \draw[fw] (c10) -- (c11);
  \draw[fw] (c8) |- (c11);
\end{tikzpicture}
\caption{Dependencies among the chapters. Chapters 1--4 build the categories and the genericity theorems;
5--7 treat bundles, choices and numerical invariants; 8--11 treat Morse functions, cobordism, isotopy and
surfaces.}\label{fig:plan}
\end{figure}

\section*{A dictionary}

The table lists, for each of the recurring notions of the source, the categorical object that this book
uses in its place, and the chapter in which the translation is made.

\begin{center}\small
\begin{tabular}{@{}p{0.36\textwidth}p{0.47\textwidth}l@{}}
\toprule
In the classical text & Here & Ch.\\
\midrule
the set $C^\infty(M,N)$ with a topology & the mapping object $[M,N]$ in smooth sets and its shape & \ref{ch:function}\\
the chain rule & functoriality of $T$ and of Weil functors & \ref{ch:tangent}\\
a map transverse to a submanifold & a fibre product that the embedding into derived manifolds preserves & \ref{ch:transversality}\\
``by a small perturbation we may assume'' & every derived manifold is cobordant to a manifold & \ref{ch:transversality}, \ref{ch:cobordism}\\
isomorphism classes of vector bundles & homotopy classes of maps to a classifying space; concordance & \ref{ch:bundles}\\
``unique up to isotopy'' & a contractible space of choices & \ref{ch:tubular}\\
Euler characteristic, Lefschetz number & trace of an endomorphism of a dualizable object & \ref{ch:degree}\\
passing a critical point & a homotopy pushout along a sphere & \ref{ch:morse}\\
cobordism group & $\pi_0$ of a cobordism category; a homotopy group of a Thom spectrum & \ref{ch:cobordism}\\
isotopy extension & a restriction map between embedding spaces is a fibration & \ref{ch:isotopy}\\
classification of surfaces & generators and relations for the two-dimensional cobordism category & \ref{ch:surfaces}\\
\bottomrule
\end{tabular}
\end{center}

\section*{Conventions}

Manifolds are smooth ($C^\infty$), Hausdorff, second countable, and finite-dimensional; different components
may have different dimensions unless a dimension is named. A \emph{closed} manifold is compact without
boundary. Hirsch works with $C^r$ maps for $1\le r\le\omega$ and proves that nothing is lost by passing to
$C^\infty$ (\cref{thm:smoothing}); we use the finite-differentiability classes only where a theorem needs
them. ``Space'' means $\infty$-groupoid, that is, a homotopy type; the $\infty$-category of spaces is written
$\Spc$. Citations of the form ``Hirsch 3.2.7'' mean Chapter~3, Section~2, item~7 of the source.

\section*{How to read the tags}

A statement tagged \textsf{classical} or \textsf{literature} is someone else's theorem, with a reference; a
proof is given only when it is short or when its form matters for what follows. A statement tagged
\textsf{reformulation} is a restatement, and the text says what it restates. Statements tagged
\textsf{observation} are elementary; we have not found them written down in the form given and make no claim
of depth. Statements tagged \textsf{programme} are not results.
\mainmatter
\part{Spaces of Smooth Things}
\chapter{Manifolds and Smooth Maps}\label{ch:manifolds}

Hirsch's first chapter defines manifolds, maps, tangent bundles, embeddings and manifolds with boundary, and
proves two theorems that the rest of his book uses constantly: the preimage of a regular value is a
submanifold, and a compact manifold embeds in a Euclidean space. This chapter records the category those
definitions produce, says exactly what is wrong with it as a category, and introduces the four enlargements
that repair it. Each enlargement repairs a different defect, and it is worth seeing at the outset which
is which.

\section{The category of manifolds}

\begin{definition}\label{def:mfd}
$\Mfd$ is the category whose objects are smooth manifolds (Hausdorff, second countable, finite-dimensional,
without boundary) and whose morphisms are smooth maps. $\Cart\subset\Mfd$ is the full subcategory on the
objects $\R^n$, $n\ge0$, and $\cat{Open}\subset\Mfd$ the full subcategory on open subsets of the $\R^n$.
\end{definition}

The source works throughout with a differentiability class $1\le r\le\omega$. The following theorem is
the reason one may fix $r=\infty$.

\begin{theorem}[Smoothing of differential structures]\label{thm:smoothing}
\Scl[Whitney; Hirsch 2.2.9, 2.2.10]
Let $1\le r<\infty$. Every $C^r$ manifold is $C^r$ diffeomorphic to a $C^\infty$ manifold, and two $C^\infty$
manifolds that are $C^r$ diffeomorphic are $C^\infty$ diffeomorphic.
\end{theorem}

So the groupoids of $C^r$ manifolds and $C^r$ diffeomorphisms have the same sets of isomorphism classes for
all $r\ge1$. For $r=0$ the statement fails in both directions: there are topological manifolds with no smooth
structure and topological manifolds with several (Kervaire, Milnor). The theorem is proved by approximation
in the strong topology (\cref{ch:function}), and it is the first instance of the method described in the
Preface.

\section{What the category lacks}

$\Mfd$ has finite products and countable coproducts, and little else. Three
fibre products show the possibilities. In each, the cospan is $\R^k\xrightarrow{f}\R\xleftarrow{0}\pt$, so
the fibre product, if it exists, is ``the zero set of $f$''.

\begin{example}[Three zero sets]\label{ex:three}
(a) $f(x,y)=x^2+y^2-1$. The value $0$ is regular, the zero set is the circle, and the circle is the fibre
product in $\Mfd$.

(b) $f(x)=x^2$. A cone over the cospan is a manifold $S$ with a smooth $u\colon S\to\R$ such that $u^2=0$;
then $u=0$. So the fibre product exists in $\Mfd$ and is a point. It is the wrong answer: the tangent functor
does not preserve it (the fibre product of the tangent maps is $\{(x,v):x^2=0,\ 2xv=0\}$, a line), and it
is unstable, since the zero set of $x^2-\varepsilon$ is two points or none.

(c) $f(x,y)=xy$. The fibre product does not exist in $\Mfd$ (\cref{prop:cross}).
\end{example}

\begin{proposition}\label{prop:cross}\Scl
No manifold represents the functor $S\mapsto\{u\in C^\infty(S,\R^2): u_1u_2=0\}$ on $\Mfd$.
\end{proposition}

\begin{proof}
Suppose $P$ does, with universal map $j\colon P\to\R^2$. Evaluating on $S=\pt$ shows that $j$ is a bijection
onto the cross $X=\{xy=0\}$. Let $p_0=j^{-1}(0)$ and let $d$ be the dimension of $P$ at $p_0$. The two axes
are smooth curves in $X$ through $0$, so they lift to smooth injective curves $a,b\colon\R\to P$ through $p_0$
with $a(\R)\cap b(\R)=\{p_0\}$; hence $d\ge1$. If $d=1$, an injective continuous map between $1$-manifolds is
open, so $a(\R)$ and $b(\R)$ are both neighbourhoods of $p_0$ and meet in more than a point. If $d\ge2$, then
$j$ has rank at most $1$ everywhere, since $X$ has empty interior, and by the constant rank theorem $j$ fails
to be injective on an open dense set where its rank is locally constant.
\end{proof}

There are also no mapping objects: outside degenerate cases there is no manifold $N^M$ with
$\Mfd(S\times M,N)\cong\Mfd(S,N^M)$. And quotients by even very good equivalence relations leave the category.

\section{Smooth sets and diffeological spaces}

The first repair is to keep the test objects and enlarge the class of things tested.

\begin{definition}\label{def:smset}
A family $\{U_i\to U\}$ in $\Cart$ is a covering if it is an open cover all of whose finite non-empty
intersections are diffeomorphic to Euclidean spaces. A \term{smooth set} is a sheaf of sets on $\Cart$ for
this coverage; $\SmSet=\cat{Sh}(\Cart)$. For a smooth set $X$, the elements of $X(\R^n)$ are its \term{plots}
of dimension~$n$. A smooth set is \term{concrete}, or a \term{diffeological space}, if each plot is determined
by the function $\R^n\to X(\pt)$ it induces on points; $\Dflg\subset\SmSet$ is the full subcategory of these.
\end{definition}

\begin{theorem}\label{thm:embed}\Scl[Souriau, Iglesias-Zemmour \cite{IZ13}; Baez--Hoffnung \cite{BH11}]
\begin{enumerate}[label=(\roman*)]
\item The functor $\Mfd\to\SmSet$, $M\mapsto C^\infty(-,M)$, is fully faithful and takes values in $\Dflg$.
\item It preserves every limit that exists in $\Mfd$, and it sends an open cover of a manifold to a colimit
diagram.
\item $\SmSet$ is a Grothendieck topos and $\Dflg$ a quasitopos. Both are complete, cocomplete and locally
cartesian closed. The mapping object has plots
$[X,Y](\R^n)=\Hom(\R^n\times X,Y)$.
\end{enumerate}
\end{theorem}

\begin{proof}[Proof of (i)]
A natural transformation $\alpha\colon C^\infty(-,M)\to C^\infty(-,N)$ gives a function $f=\alpha_{\pt}\colon
M\to N$, and naturality with respect to the inclusions of points shows that $\alpha$ is composition with $f$.
For every chart $\varphi\colon\R^m\to M$ the composite $f\varphi=\alpha(\varphi)$ is smooth, so $f$ is smooth.
\end{proof}

By (ii), smooth sets inherit the fibre products of \cref{ex:three}(a) and (b) unchanged, and in case (c) they
supply one: the cross, with the plots being the smooth maps into $\R^2$ that land in it. By (iii) they supply
the mapping object, and its plots are exactly what the source calls a smooth family of maps. \Cref{ch:function}
is about this object. What smooth sets do not repair is (b). A sheaf on $\Cart$ is tested only by reduced
objects, and the equation $u^2=0$ has the same solutions as $u=0$ in every $C^\infty(\R^n)$.

\section{Function algebras}

The second repair goes the other way and enlarges the class of test objects, by admitting algebras of
functions that are not algebras of functions on a manifold.

\begin{definition}\label{def:cinf}
A \term{$C^\infty$-ring} is a functor $A\colon\Cart\to\Set$ that preserves finite products. Its underlying set
is $A(\R)$; each smooth $h\colon\R^n\to\R$ gives an $n$-ary operation $A(h)\colon A(\R)^n\to A(\R)$. The
category of these is written $\CinfRing$. For a manifold $M$, the $C^\infty$-ring $C^\infty(M)$ is
$\R^n\mapsto C^\infty(M,\R^n)$.
\end{definition}

Taking $h$ to be addition and multiplication shows that a $C^\infty$-ring has an underlying commutative
$\R$-algebra. The free $C^\infty$-ring on $n$ generators is $C^\infty(\R^n)$, and quotients by ideals of the
underlying algebra are again $C^\infty$-rings.

\begin{theorem}\label{thm:MR}\Sli[Moerdijk--Reyes \cite{MR91}, Ch.~I, Thm.~2.8]
The functor $\Mfd\to\CinfRing^{\op}$, $M\mapsto C^\infty(M)$, is fully faithful, and it sends transverse fibre
products (\cref{thm:transpull}) to pushouts: if $f\colon M\to N$ and $g\colon P\to N$ are transverse then
$C^\infty(M\times_NP)\cong C^\infty(M)\otimes^\infty_{C^\infty(N)}C^\infty(P)$, the pushout in $\CinfRing$.
\end{theorem}

In \cref{ex:three}(b) the pushout is $C^\infty(\R)/(x^2)$. By Hadamard's lemma every smooth function is
$g(x)=g(0)+g'(0)x+x^2h(x)$, so this ring is $\R[\varepsilon]/(\varepsilon^2)$, the dual numbers. Its
``spectrum'' has one point and a one-dimensional tangent space, which is what the tangent functor predicted.
The ring has dimension $2$, as does $C^\infty(\R)/(x^2-\varepsilon)\cong\R\times\R$ for $\varepsilon>0$. So
$C^\infty$-rings repair (b) on one side. For $\varepsilon<0$ the function $x^2-\varepsilon$ is invertible and
the quotient is the zero ring; no algebra of real functions will make the length constant. What is constant
for all $\varepsilon$ is the count with signs, $(+1)+(-1)=0$, and that is a statement about cobordism
(\cref{ch:transversality,ch:cobordism}).

The third and fourth repairs are homotopical. The third, the $\infty$-topos of smooth $\infty$-groupoids,
is needed as soon as the objects being parametrized have automorphisms, as vector bundles and manifolds
themselves do; it is introduced next. The fourth, derived manifolds, replaces the pushout of
\cref{thm:MR} by a homotopy pushout, and is the subject of \cref{ch:transversality}. \Cref{tab:four} shows
which fibre product is seen where; the fourth row is the simplest case in which the derived answer differs
from all the others.

\begin{table}[ht]\centering\small
\begin{tabular}{@{}lllll@{}}
\toprule
zero set of & in $\Mfd$ & in $\SmSet$ & in $\CinfRing^{\op}$ & in $\DMfd$\\
\midrule
$x^2+y^2-1$ & circle & circle & circle & circle\\
$x^2$ & point & point & $\R[\varepsilon]/(\varepsilon^2)$ & $\R[\varepsilon]/(\varepsilon^2)$\\
$xy$ & none & cross & $C^\infty(\R^2)/(xy)$ & $C^\infty(\R^2)/(xy)$\\
$0\colon\pt\to\R$ & point & point & point & point of dimension $-1$\\
\bottomrule
\end{tabular}
\caption{Four fibre products $\R^k\times_{\R}\pt$ in four categories.}\label{tab:four}
\end{table}

\section{Smooth \texorpdfstring{$\infty$}{infinity}-groupoids and shape}

\begin{definition}\label{def:shape}
$\mathbf{H}=\cat{Sh}_\infty(\Cart)$ is the $\infty$-topos of sheaves of spaces on $\Cart$; its objects are
\term{smooth $\infty$-groupoids}, or smooth stacks. The constant-sheaf functor $\Spc\to\mathbf{H}$ has a left
adjoint $\Pi\colon\mathbf{H}\to\Spc$, the \term{shape}. Two points of $X(\pt)$ are \term{concordant} if they
are the restrictions to $0$ and $1$ of a point of $X(\R)$.
\end{definition}

Smooth sets are the $0$-truncated objects of $\mathbf{H}$. The left adjoint exists because the objects of the
site are contractible; there is a formula for it.

\begin{theorem}\label{thm:shape}\Sli[see \cite{BEBP24}, \cite{Sch13}]
For $X\in\mathbf{H}$, the shape is the geometric realization of the simplicial space obtained by evaluating
on the extended simplices $\Delta^k_e=\{t\in\R^{k+1}:\sum t_i=1\}\cong\R^k$:
\[\Pi X\ \simeq\ \colim_{[k]\in\Delta^{\op}}X(\Delta^k_e).\]
For a manifold $M$, $\Pi M$ is the homotopy type of the underlying space of $M$, and $\pi_0\Pi X$ is the set of
concordance classes of points of $X$ whenever $X$ is a smooth set.
\end{theorem}

For a manifold, the simplicial set $k\mapsto C^\infty(\Delta^k_e,M)$ is the smooth singular complex, and the
statement that it has the homotopy type of $M$ is the smooth approximation theorem (Hirsch 2.2.6) in its
relative form. The functor $\Pi$ is the device by which this book passes from smooth objects to homotopy
types. Nearly every classification theorem in later chapters has the form: for a certain smooth
$\infty$-groupoid $X$ of geometric structures, $\Pi X$ is such-and-such a space.

\section{Regular values and the fibre products that exist}

\begin{definition}\label{def:transverse}
Smooth maps $f\colon M\to N$ and $g\colon P\to N$ are \term{transverse}, written $f\trans g$, if
$df_x(T_xM)+dg_y(T_yP)=T_zN$ whenever $f(x)=g(y)=z$. If $g$ is the inclusion of a submanifold $A$ one writes
$f\trans A$. A point $z\in N$ is a \term{regular value} of $f$ if $f\trans\{z\}$.
\end{definition}

\begin{theorem}[Transverse fibre products]\label{thm:transpull}\Scl[Hirsch 1.3.2, 1.3.3]
If $f\trans g$ then $M\times_NP=\{(x,y):f(x)=g(y)\}$ is a submanifold of $M\times P$ of dimension
$\dim M+\dim P-\dim N$. It is the fibre product in $\Mfd$, and its tangent bundle is
$TM\times_{TN}TP$.
\end{theorem}

\begin{proof}
By \cref{exer:diag}, $f\trans g$ if and only if $f\times g\trans\Delta_N$, so it suffices to treat the
preimage $W=h^{-1}(A)$ of a submanifold $A\subset N$ of codimension $k$ under a map $h\trans A$. Locally $A$ is
the zero set of a submersion $\phi\colon V\to\R^k$, and transversality says that $\phi h$ is a submersion at
the points of $W$. The implicit function theorem then gives charts in which $W$ is a coordinate subspace of
codimension $k$ and $T_xW=\ker d(\phi h)_x=dh_x^{-1}(T_{h(x)}A)$. A smooth map into $M\times P$ with image in an
embedded submanifold is smooth into the submanifold, which gives the universal property.
\end{proof}

The theorem identifies a class of cospans whose limits exist, are preserved by $T$, and by \cref{thm:MR} are
preserved by passage to function algebras. \Cref{ch:transversality} shows that this class is large (every
cospan is arbitrarily close to one in it) and that it determines a universal completion.

\begin{example}\label{ex:On}
Let $f\colon\R^{n\times n}\to\cat{Sym}_n$, $f(A)=A^{\mathsf T}A$. At $A\in\Or(n)$ the differential
$H\mapsto A^{\mathsf T}H+H^{\mathsf T}A$ is onto: $H=\tfrac12AS$ maps to $S$. So $\Or(n)=f^{-1}(I)$ is a
manifold of dimension $n^2-\tfrac12n(n+1)=\tfrac12n(n-1)$. The script checks the rank numerically for
$n\le4$.
\end{example}

\section{Embeddings and idempotents}

\begin{theorem}[Whitney]\label{thm:whitney}\Scl[Hirsch 1.3.5, 2.2.14]
Every manifold of dimension $n$ admits an embedding into $\R^{2n+1}$ with closed image.
\end{theorem}

Combined with the tubular neighbourhood theorem (\cref{thm:tubular}), this says that every manifold is a
smooth retract of an open subset of a Euclidean space. The converse also holds, and together they give a
description of $\Mfd$ in which charts and atlases do not appear.

\begin{theorem}\label{thm:karoubi}\Scl[attributed to Lawvere]
$\Mfd$ is the idempotent completion of $\cat{Open}$: every idempotent in $\Mfd$ splits, and every manifold is
a retract of an object of $\cat{Open}$.
\end{theorem}

\begin{proof}
The second statement was just explained. For the first, let $e\colon M\to M$ be smooth with $e^2=e$; we show
that $e(M)=\Fix(e)$ is a submanifold (\cref{exer:idem} asks for the details). At a fixed point $x$ the
differential $de_x$ is an idempotent matrix, so its rank equals its trace, which is continuous and
integer-valued along $\Fix(e)$; let it be $k$ near $x$. Rank is lower semicontinuous, so $\rk de_y\ge k$ for
$y$ near $x$, and $de_{e(y)}de_y=de_y$ gives $\rk de_y\le\rk de_{e(y)}=k$. So $e$ has constant rank $k$ near
$x$, and its image near $x$ is a $k$-dimensional submanifold.
\end{proof}

The theorem explains two features of what follows. A product-preserving functor on $\Cart$, such as a
$C^\infty$-ring, has a canonical value on every manifold, because the manifold is a retract of an open subset
of some $\R^q$. And the universal completion of \cref{thm:CS} has to be required to be idempotent complete.

\section{Boundaries}

Hirsch allows manifolds with boundary from the start, since cobordisms need them. A manifold with boundary is
modelled on the half-space $\R^n_+=\{x_n\ge0\}$; its boundary $\partial M$ is a manifold of dimension $n-1$.
The category $\Mfd^\partial$ is less well behaved than $\Mfd$: the product of two objects has corners, and
$\partial$ is not a functor on all smooth maps. The notion that makes the regular value theorem work is the
following.

\begin{definition}\label{def:neat}
A submanifold $A\subset M$ of a manifold with boundary is \term{neat} if $\partial A=A\cap\partial M$ and $A$
is not tangent to $\partial M$ at any point of $\partial A$.
\end{definition}

\begin{theorem}\label{thm:neat}\Scl[Hirsch 1.4.2]
Let $f\colon M\to N$ be smooth, with $\partial N=\emptyset$, and let $z$ be a regular value of both $f$ and
$f|_{\partial M}$. Then $f^{-1}(z)$ is a neat submanifold of $M$ with boundary
$(f|_{\partial M})^{-1}(z)$.
\end{theorem}

In smooth sets the half-line is the subobject $[0,\infty)\subset\R$, and a function on it is smooth in the
sheaf sense exactly when it is the restriction of a smooth function on $\R$ \cite[\S24]{KM97}. So
manifolds with boundary embed fully faithfully in $\Dflg$ as well. The boundary is recovered from
the object, as the set of points at which it is not locally Euclidean, although not functorially.

\section*{Notes}

Diffeological spaces are due to Souriau; the standard reference is \cite{IZ13}, and the identification with
concrete sheaves is in \cite{BH11}. $C^\infty$-rings come from Lawvere's programme for synthetic differential
geometry; see \cite{MR91}, \cite{Koc06} and \cite{Joy19}. The $\infty$-topos $\mathbf{H}$ and its shape are
developed in \cite{Sch13}; the formula of \cref{thm:shape} is proved in \cite{BEBP24} and in work of
Pavlov and of Bunk. Manifolds with corners as a category are treated by Melrose and by Joyce.

\section*{Exercises}

\begin{exercise}\label{exer:xsq}
Verify the second row of \cref{tab:four}: the fibre product of $x\mapsto x^2$ and $0\colon\pt\to\R$ is a
point in $\Mfd$ and in $\SmSet$, and is corepresented by $\R[\varepsilon]/(\varepsilon^2)$ in $\CinfRing$.
Show that a map of $C^\infty$-rings $C^\infty(M)\to\R[\varepsilon]/(\varepsilon^2)$ is a tangent vector to
$M$.
\end{exercise}

\begin{exercise}\label{exer:idem}
Complete the proof of \cref{thm:karoubi}: if $e\colon M\to M$ is a smooth idempotent then $\Fix(e)$ is a
closed submanifold of $M$ and $e\colon M\to\Fix(e)$ is smooth. Give an example in which $\Fix(e)$ has
components of different dimensions.
\end{exercise}

\begin{exercise}\label{exer:On}
Carry out \cref{ex:On} in detail, and show that $T_I\Or(n)$ is the space of skew-symmetric matrices.
\end{exercise}

\begin{exercise}\label{exer:exp}
Let $M,N$ be manifolds. Show that the smooth set $[M,N]$ is concrete, that its points are the smooth maps
$M\to N$, and that $\Hom(S,[M,N])\cong\Hom(S\times M,N)$ for every smooth set $S$.
\end{exercise}

\begin{exercise}\label{exer:diag}
Show that $f\colon M\to N$ and $g\colon P\to N$ are transverse if and only if
$f\times g\colon M\times P\to N\times N$ is transverse to the diagonal $\Delta_N$.
\end{exercise}
\chapter{The Tangent Functor and Jets}\label{ch:tangent}

The source introduces the tangent bundle in a few pages and the bundle of jets in a few more, and uses both
as tools. Here they are treated as structure on the category. The chain rule is the statement that $T$ is a
functor; the symmetry of second derivatives is the statement that a certain algebra is commutative; the
higher chain rule is the statement that jets compose. All product-preserving functors of this kind are
classified by finite-dimensional algebras, and the axioms those functors satisfy can be isolated and
imposed on other categories.

\section{The tangent functor}

On $\Cart$ put $T(\R^n)=\R^n\times\R^n$ and $Tf(x,v)=(f(x),Df_x v)$. The chain rule
$D(g f)_x=Dg_{f(x)}Df_x$ is the statement $T(gf)=Tg\,Tf$.

\begin{proposition}\label{prop:T}\Scl
$T$ extends uniquely to a functor $T\colon\Mfd\to\Mfd$ with a natural transformation $p\colon T\Rightarrow\id$
such that $T$ sends open embeddings to open embeddings and $T(U)=p_M^{-1}(U)$ for $U\subset M$ open. It
preserves finite products and transverse fibre products, and each $p_M\colon TM\to M$ is a vector bundle.
\end{proposition}

\begin{proof}
A manifold is the colimit in $\Mfd$ of the \v{C}ech diagram of an atlas, and the diagram consists of open
embeddings between open subsets of $\R^n$, on which $T$ is defined by the formula. The colimit of the
image diagram exists and is $TM$. Products are preserved by the formula; transverse fibre products by
\cref{thm:transpull}.
\end{proof}

\section{Weil algebras}

\begin{definition}\label{def:weil}
A \term{Weil algebra} is a finite-dimensional commutative $\R$-algebra $A=\R\oplus\mathfrak n$ with
$\mathfrak n$ a nilpotent ideal. It is a $C^\infty$-ring in a unique way: for smooth $h$ and $a=a_0+n$ one
puts $h(a)=\sum_\alpha D^\alpha h(a_0)\,n^\alpha/\alpha!$, a finite sum. The \term{Weil functor} of $A$ is
\[T_AM=\CinfRing\bigl(C^\infty(M),A\bigr),\]
the set of $A$-points of $M$, with the manifold structure for which $T_A(\R^n)=A^n$.
\end{definition}

For $A=\R$ this is the identity functor (the $\R$-points of $M$ are its points). For the dual numbers
$W=\R[\varepsilon]/(\varepsilon^2)$ it is $T$, by \cref{exer:xsq}. In coordinates a map $f$ acts on
$A$-points by Taylor expansion in the nilpotent directions:
$f(x+v\varepsilon)=f(x)+Df_x(v)\,\varepsilon$.

\begin{theorem}[Classification of product-preserving bundle functors]\label{thm:weil}
\Sli[Eck, Kainz--Michor, Luciano; \cite{KMS93}, \S\S35--36]
\begin{enumerate}[label=(\roman*)]
\item For every Weil algebra $A$, $T_A$ is a functor $\Mfd\to\Mfd$ that preserves finite products, with a
natural fibre bundle projection $T_A\Rightarrow\id$.
\item Every product-preserving bundle functor on $\Mfd$ is naturally isomorphic to $T_A$ for a Weil algebra
$A$, unique up to isomorphism.
\item Natural transformations $T_A\Rightarrow T_B$ correspond bijectively to algebra homomorphisms $A\to B$.
\item $T_A\circ T_B\cong T_{A\otimes B}$, and $T_A\times_{\id}T_B\cong T_{A\times_{\R}B}$.
\end{enumerate}
\end{theorem}

By (iii) and (iv), every natural operation on iterated tangent bundles is a homomorphism of small
algebras. \Cref{tab:weil} lists the ones that matter.

\begin{table}[ht]\centering\small
\begin{tabular}{@{}lll@{}}
\toprule
natural transformation & algebra homomorphism & in coordinates\\
\midrule
projection $p\colon T\Rightarrow\id$ & $W\to\R$, $\varepsilon\mapsto0$ & $(x,v)\mapsto x$\\
zero section $0\colon\id\Rightarrow T$ & $\R\to W$ & $x\mapsto(x,0)$\\
addition $+\colon T\times_{\id}T\Rightarrow T$ & $W\times_\R W\to W$, $\varepsilon_1,\varepsilon_2\mapsto\varepsilon$ & $(x,v_1,v_2)\mapsto(x,v_1+v_2)$\\
vertical lift $\ell\colon T\Rightarrow TT$ & $W\to W\otimes W$, $\varepsilon\mapsto\varepsilon\eta$ & $(x,v)\mapsto(x,0,0,v)$\\
flip $c\colon TT\Rightarrow TT$ & $W\otimes W\to W\otimes W$, $\varepsilon\leftrightarrow\eta$ & $(x,v,w,a)\mapsto(x,w,v,a)$\\
\bottomrule
\end{tabular}
\caption{The structure maps of the tangent functor as homomorphisms of Weil algebras. Here
$W=\R[\varepsilon]/(\varepsilon^2)$, $W\otimes W=\R[\varepsilon,\eta]/(\varepsilon^2,\eta^2)$ and
$W\times_\R W=\R[\varepsilon_1,\varepsilon_2]/(\varepsilon_i\varepsilon_j)$.}\label{tab:weil}
\end{table}

\begin{proposition}[Symmetry of second derivatives]\label{prop:flip}\Srf[\cite{Koc06}, \cite{KMS93}]
The naturality of the flip $c$ with respect to a map $f$ is the equality $D^2f_x(v,w)=D^2f_x(w,v)$. It holds
because $W\otimes W$ is commutative.
\end{proposition}

\begin{proof}
Expanding in $W\otimes W$,
\[f(x+v\varepsilon+w\eta+a\varepsilon\eta)=f(x)+Df_x(v)\varepsilon+Df_x(w)\eta+
\bigl(Df_x(a)+D^2f_x(v,w)\bigr)\varepsilon\eta ,\]
which is the formula for $TTf$. Exchanging $\varepsilon$ and $\eta$ exchanges $v$ and $w$ on the left;
on the right it leaves the coefficient of $\varepsilon\eta=\eta\varepsilon$ in place. So $TTf\circ c=c\circ
TTf$ if and only if $D^2f_x(v,w)=D^2f_x(w,v)$, and the exchange is an algebra automorphism, hence natural by
\cref{thm:weil}(iii).
\end{proof}

The script evaluates polynomials in $\R[\varepsilon,\eta]/(\varepsilon^2,\eta^2)$ and confirms the expansion
and its symmetry, and checks the chain rule in $W$ against finite differences.

\section{Tangent categories}

The five maps of \cref{tab:weil} and the equations among them can be abstracted.

\begin{definition}\label{def:tancat}
A \term{tangent category} is a category $\mathcal X$ with an endofunctor $T$ and natural transformations
$p,0,+,\ell,c$ as in \cref{tab:weil}, such that the fibre powers of each $p_M$ exist and are preserved by
$T$; $(p,0,+)$ makes each $TM$ a commutative monoid over $M$; $\ell$ and $c$ are additive, $c^2=\id$,
$c\,\ell=\ell$, and the usual coherence equations hold; and the vertical lift is universal, in the sense
that
$TM\times_MTM\to TTM$, $(a,b)\mapsto T(+)\bigl(\ell(a),0(b)\bigr)$, is the equalizer of $T(p)$ and
$0\,p\,p$ \cite[Def.~2.3]{CC14}.
\end{definition}

The definition is Rosick\'y's \cite{Ros84}, revived and developed by Cockett and Cruttwell. The last axiom
says that a second-order tangent vector that is vertical is the same thing as a pair of first-order ones. It
is what makes it possible to define the Lie bracket of vector fields, connections, and solutions of
differential equations in any tangent category.

\begin{theorem}\label{thm:leung}\Sli[Leung \cite{Leu17}]
Let $\cat{Weil}_1$ be the monoidal category of Weil algebras generated from $W$ by $\otimes$ and
$\times_\R$. A tangent structure on $\mathcal X$ is the same thing as a strong monoidal functor
$\cat{Weil}_1\to\cat{End}(\mathcal X)$ that preserves the pullbacks expressing $\times_\R$ and the
universality of $\ell$.
\end{theorem}

So \cref{tab:weil} is the definition, read backwards. Examples of tangent categories include $\Mfd$;
$\CinfRing^{\op}$ and the category of affine schemes, with $T$ corepresented by $A\mapsto A[\varepsilon]$ and
Kähler differentials; every model of synthetic differential geometry; and cartesian differential categories,
which axiomatize $\Cart$ itself \cite{CC14}. The definition has an $\infty$-categorical form, in which the
Goodwillie calculus of functors is an example: the tangent bundle of an $\infty$-category is its fibrewise
stabilization \cite{BBC21}.

\section{Infinitesimals as objects}

In $\CinfRing^{\op}$ the algebra $W$ is an object, $D=\operatorname{Spec}W$, and
$T_AM$ is the set of maps $\operatorname{Spec}A\to M$. One would like $TM$ to be the mapping object $M^D$.

\begin{theorem}\label{thm:sdg}\Sli[Dubuc \cite{Dub79}; Kock \cite{Koc06}; Moerdijk--Reyes \cite{MR91}]
Let $\mathcal Z$ be the Cahiers topos, the topos of sheaves on the category of objects
$\R^n\times\operatorname{Spec}A$ with $A$ a Weil algebra. Then $\Mfd\to\mathcal Z$ is fully faithful and
preserves transverse fibre products; $\R^D\cong\R\times\R$ (the Kock--Lawvere axiom); and
$(T_AM)\cong M^{\operatorname{Spec}A}$ for every manifold $M$ and Weil algebra $A$.
\end{theorem}

In $\mathcal Z$ the tangent functor is an exponential, so it preserves all limits. The functor $(-)^D$ also
has a right adjoint there, which is a statement about how small $D$ is. None of this is available in
$\SmSet$. The restriction of $D$ to $\Cart$ is the terminal object, because a nilpotent element of
$C^\infty(\R^n)$ is zero; a sheaf on $\Cart$ cannot tell $D$ from a point. This is the same defect as in
\cref{ex:three}(b), and it is the reason that tangent functors on diffeological spaces are a subject with
several inequivalent answers.

\section{Jets}

\begin{definition}\label{def:jets}
Two maps $f,g\colon M\to N$ have the same \term{$r$-jet} at $x$ if $f(x)=g(x)$ and their partial derivatives
up to order $r$ agree at $x$ in some, hence any, charts. $J^r(M,N)$ is the set of $r$-jets of maps $M\to N$ at
all points of $M$; it is a manifold and a fibre bundle over $M\times N$ by source and target. A map $f$ has an
\term{$r$-jet prolongation} $j^rf\colon M\to J^r(M,N)$, a section of the source projection.
\end{definition}

The fibre of $J^r(\R^m,\R^n)$ over $(0,0)$ is the space of polynomial maps $\R^m\to\R^n$ of degree at most $r$
without constant term, of dimension $n\bigl(\binom{m+r}{r}-1\bigr)$. For $r=1$,
$J^1(M,N)=\Hom(TM,TN)$ as a bundle over $M\times N$.

\begin{proposition}[Jets compose]\label{prop:jets}\Scl
For each $r$ there is a category $\mathcal J^r$ whose objects are pairs $(M,x)$ and whose morphisms
$(M,x)\to(N,y)$ are the $r$-jets at $x$ with target $y$. Taking the $r$-jet is a functor from germs of
smooth maps to $\mathcal J^r$, and truncation gives functors $\mathcal J^r\to\mathcal J^{r-1}$. In
coordinates, composition in $\mathcal J^r$ is composition of polynomials followed by truncation at degree $r$.
The automorphism group $G^r_m$ of $(\R^m,0)$ in $\mathcal J^r$ is a Lie group, $G^1_m=\GL_m(\R)$, and the
kernel of $G^r_m\to G^1_m$ is contractible.
\end{proposition}

\begin{proof}
Functoriality is the higher chain rule: the derivatives of order at most $r$ of $g f$ at $x$ are
polynomials in those of $f$ at $x$ and of $g$ at $f(x)$ (Faà di Bruno), so the $r$-jet of $gf$ depends only
on the $r$-jets, and associativity is inherited from composition of maps. The kernel consists of polynomial
maps $x+(\text{terms of degree }2,\dots,r)$ under truncated composition; scaling the higher terms to zero
contracts it.
\end{proof}

For one variable and $r=2$ the composition rule is
$(a_1,a_2)\circ(b_1,b_2)=(a_1b_1,\;a_1b_2+a_2b_1^2)$ for jets $a_1t+\tfrac12a_2t^2$, which is
$(gf)''=g'f''+g''(f')^2$. The script checks associativity of this and of the three-variable analogue. The last
sentence of the proposition has a consequence that is used without comment throughout the subject: a
structure defined by higher jets carries the same homotopical information as one defined by first jets, so
the characteristic classes of a manifold are those of its tangent bundle.

Jets fit the framework of the previous sections: the functor $N\mapsto J^r_0(\R^m,N)$ of $r$-jets at the origin
is the Weil functor of the algebra $\R[x_1,\dots,x_m]/\mathfrak m^{r+1}$.

\section{Two derivatives of a composite}

The tangent functor is covariant and pushes vectors forward. Its transpose pulls covectors back, and
since $T^*$ is not a functor on $\Mfd$ the transpose needs a different kind of category to live in.

\begin{definition}\label{def:lens}
$\cat{Lens}$ is the category whose objects are pairs $(X,X')$ of Euclidean spaces and whose morphisms
$(X,X')\to(Y,Y')$ are pairs $(f,f^\sharp)$ of smooth maps $f\colon X\to Y$ and
$f^\sharp\colon X\times Y'\to X'$, composed by
\[(g,g^\sharp)\circ(f,f^\sharp)=\bigl(gf,\ (x,z')\mapsto f^\sharp(x,g^\sharp(f(x),z'))\bigr).\]
The \term{reverse derivative} of $f\colon\R^m\to\R^n$ is
$R[f]\colon\R^m\times\R^n\to\R^m$, $R[f](x,w)=(Df_x)^{\mathsf T}w$.
\end{definition}

\begin{proposition}[Reverse chain rule]\label{prop:reverse}
\Sli[Fong--Spivak--Tuy\'eras \cite{FST19}; Cockett et al.\ \cite{CCG20}]
$f\mapsto(f,R[f])$ is a product-preserving functor $\Cart\to\cat{Lens}$.
\end{proposition}

\begin{proof}
$R[gf](x,w)=(Dg_{f(x)}Df_x)^{\mathsf T}w=(Df_x)^{\mathsf T}(Dg_{f(x)})^{\mathsf T}w=R[f](x,R[g](f(x),w))$,
which is the composition law of $\cat{Lens}$.
\end{proof}

On manifolds the same structure is the correspondence $T^*M\leftarrow f^*T^*N\to T^*N$. The forward functor
computes a directional derivative of a composite at the cost of one evaluation; the reverse functor computes
the gradient of a scalar-valued composite at the cost of one evaluation and one backward pass. In the
companion volume the reverse functor is the backpropagation algorithm, and \cref{prop:reverse} is the reason
it is compositional.

\section*{Notes}

Weil introduced his algebras in 1953 as ``points proches''. The classification theorem is due independently
to Eck, to Kainz and Michor, and to Luciano; \cite{KMS93} is the standard account and treats natural
operations systematically. Tangent categories are surveyed in \cite{CC14}. For synthetic differential
geometry see \cite{Koc06} and \cite{MR91}; the Cahiers topos is Dubuc's \cite{Dub79}. Jet bundles as a comonad
on bundles, with differential operators as its coKleisli maps, are a further piece of structure that this
book does not use.

\section*{Exercises}

\begin{exercise}\label{exer:Tprod}
Show that $T(M\times N)\cong TM\times TN$ naturally, and that if $z$ is a regular value of $f\colon M\to N$
then $T\bigl(f^{-1}(z)\bigr)=(Tf)^{-1}(0_z)$, where $0_z$ is the zero vector at $z$.
\end{exercise}

\begin{exercise}\label{exer:weil2}
Let $A=\R[x]/(x^3)$ and $B=W\otimes W$. Show that $T_AM$ is the bundle of $2$-jets at $0$ of curves in $M$, of
fibre dimension $2\dim M$, and that the homomorphism $A\to B$, $x\mapsto\varepsilon+\eta$, induces an
embedding $T_AM\to TTM$ whose image is the fixed set of the flip intersected with $\{T(p)=p_T\}$.
\end{exercise}

\begin{exercise}\label{exer:jet2}
Derive the composition law for $2$-jets of maps $(\R^m,0)\to(\R^n,0)\to(\R^k,0)$ in terms of the linear and
quadratic parts, and verify associativity directly.
\end{exercise}

\begin{exercise}\label{exer:structure}
Verify the last three rows of \cref{tab:weil}: compute the natural transformations induced by the stated
algebra homomorphisms in the coordinates $(x,v,w,a)$ on $TT\R^n$.
\end{exercise}

\begin{exercise}\label{exer:lens}
Show that composition in $\cat{Lens}$ is associative, and that the reverse derivative of a composite of $k$
maps $\R^{n_0}\to\R^{n_1}\to\dots\to\R^{n_k}$ with $n_k=1$ can be evaluated with one forward pass storing
the intermediate values and $k$ transposed Jacobian--vector products.
\end{exercise}
\chapter{Function Spaces}\label{ch:function}

The second chapter of the source puts two topologies on the set $C^r(M,N)$, proves that the important classes
of maps are open in the stronger one, and proves approximation theorems by convolution and partitions of
unity. This is the analytic engine of the book. The mapping object $[M,N]$ of \cref{thm:embed} is a different
kind of structure on the same set, and it is important to be clear about what each is for. The strong
topology is for proving that properties are generic. The mapping object is for parametrizing, and its shape
is the homotopy type of the space of maps. The two are compatible but not interchangeable.

\section{Two topologies}

Fix $0\le r<\infty$. For $f\in C^r(M,N)$, charts $(\varphi,U)$ on $M$ and $(\psi,V)$ on $N$, a compact
$K\subset U$ with $f(K)\subset V$, and $\varepsilon>0$, let $\mathcal N^r(f;\varphi,\psi,K,\varepsilon)$ be the
set of $g\in C^r(M,N)$ with $g(K)\subset V$ and
\[\bigl\|D^k(\psi g\varphi^{-1})(x)-D^k(\psi f\varphi^{-1})(x)\bigr\|<\varepsilon
\qquad\text{for }x\in\varphi(K),\ 0\le k\le r.\]

\begin{definition}\label{def:topologies}
The \term{weak} (or compact-open $C^r$) topology on $C^r(M,N)$ is generated by the sets
$\mathcal N^r(f;\varphi,\psi,K,\varepsilon)$. The \term{strong} (or Whitney) topology has as a base the sets
$\bigcap_i\mathcal N^r(f;\varphi_i,\psi_i,K_i,\varepsilon_i)$ over locally finite families of charts
$(\varphi_i,U_i)$ of $M$, with arbitrary $\varepsilon_i>0$. On $C^\infty(M,N)$ each topology is the union of
the ones induced from $C^r$, $r<\infty$. The spaces are written $C^r_W(M,N)$ and $C^r_S(M,N)$.
\end{definition}

A weak neighbourhood controls a map and its derivatives on a compact set. A strong neighbourhood controls them
everywhere, with a tolerance that may shrink to zero at infinity as fast as one likes. If $M$ is compact the
two agree. If $M$ is not compact the strong topology is not metrizable and not even first countable, but it
is the one in which the theorems below are true.

\begin{theorem}[Open conditions]\label{thm:open}\Scl[Hirsch 2.1.1--2.1.6]
For $r\ge1$ the sets of immersions, submersions, embeddings, closed embeddings and diffeomorphisms are open
in $C^r_S(M,N)$. For $r\ge0$ the set of proper maps is open in $C^r_S(M,N)$.
\end{theorem}

\begin{example}\label{ex:weakfail}
Let $\beta\ge0$ be a bump function on $\R$ supported in $[-1,1]$, with maximum at $0$ and $\max\beta'=1$, and put
$g_n(x)=x-2\beta(x-n)$. Then $g_n$ agrees with the identity on $[-n+1,n-1]$, so $g_n\to\id$ in $C^\infty_W(\R,\R)$;
but $g_n'=1-2\beta'(x-n)$ takes the value $-1$, so no $g_n$ is a diffeomorphism or even an immersion. The
set of diffeomorphisms of $\R$ is not weakly open. In the strong topology $g_n$ does not converge to the
identity: the neighbourhood $\{g:|g(x)-x|<\beta(0)e^{-x^2}\}$ contains none of them. The script confirms both
statements numerically.
\end{example}

\begin{theorem}[Approximation]\label{thm:approx}\Scl[Hirsch 2.2.6]
For $0\le r<s\le\infty$, $C^s(M,N)$ is dense in $C^r_S(M,N)$. If $f$ is already $C^s$ on a neighbourhood of a
closed set $A$, the approximating maps may be taken to agree with $f$ on $A$.
\end{theorem}

Since diffeomorphisms are open (\cref{thm:open}) and smooth maps are dense, a $C^r$ diffeomorphism can be
approximated by a $C^\infty$ one once source and target carry compatible smooth structures; this is the main
step in \cref{thm:smoothing}. The same two theorems give the first comparison of homotopy types.

\begin{corollary}\label{cor:homotopy}\Scl[Hirsch 5.1.5]
Every continuous map between manifolds is homotopic to a smooth map, and two smooth maps that are
continuously homotopic are smoothly homotopic.
\end{corollary}

\section{The mapping object}

By \cref{thm:embed} the smooth set $[M,N]$ has as plots the smooth maps $F\colon\R^n\times M\to N$. A plot is
a family $s\mapsto F_s$ of maps, smooth jointly in the parameter and the variable. Nothing is said about
neighbourhoods. A topology can be recovered from the plots, as for any smooth set.

\begin{definition}\label{def:Dtop}
A set $A$ of points of a smooth set $X$ is \term{$D$-open} if $p^{-1}(A)\subset\R^n$ is open for every plot
$p$ of $X$. A subobject of $X$ is $D$-open if its set of points is and it contains every plot whose points
lie in it.
\end{definition}

\begin{theorem}\label{thm:CSW}\Sli[Christensen--Sinnamon--Wu \cite{CSW14}]
On $C^\infty(M,N)$ the $D$-topology of the mapping object $[M,N]$ is finer than the weak topology and coarser
than the strong topology. If $M$ is compact all three coincide.
\end{theorem}

The first inequality says that a smooth family is continuous into the weak topology, which is easy. The
second says that a $D$-open set is strongly open, which is not. For non-compact $M$ the second is not an
equality: a plot $\R\to[M,N]$ is in particular a path, and paths in the strong topology are very restricted.

\begin{proposition}\label{prop:strongpath}\Scl
Let $M$ be non-compact. If $t\mapsto f_t$ is a continuous path in $C^0_S(M,N)$, there is a compact $K\subset M$
such that $f_t=f_0$ outside $K$ for all $t\in[0,1]$.
\end{proposition}

So the path components of $C^\infty_S(M,N)$ are far smaller than homotopy classes
(\cref{exer:strongpath}), and the strong topology does not compute the homotopy type of anything one wants.
It is a device for stating that a set of maps is large.

\section{The homotopy type of a space of maps}

The homotopy type attached to the mapping object is its shape (\cref{def:shape}). By \cref{thm:shape} it is
the realization of the simplicial set $k\mapsto C^\infty(\Delta^k_e\times M,N)$, whose vertices are smooth
maps, whose edges are smooth homotopies, and so on.

\begin{theorem}[Shape of mapping objects]\label{thm:oka}\Sli[Berwick-Evans--Boavida de Brito--Pavlov
\cite{BEBP24}, Thm.~1.1]
For a manifold $M$ and any $F\in\mathbf{H}$ the natural map
\[\Pi[M,F]\ \longrightarrow\ \Map(\Pi M,\Pi F)\]
is an equivalence of spaces. In particular $\Pi[M,N]\simeq\Map(M,N)$, the space of continuous maps, for
manifolds $M$ and $N$.
\end{theorem}

For $F=N$ a manifold and $M$ compact this is classical, and it is a parametrized form of
\cref{cor:homotopy}. In general it is a recent theorem, and it will be used in \cref{ch:bundles} with $F$ a
stack of bundles. On $\pi_0$ it says that concordance classes of sections of $F$ over $M$ are homotopy
classes of maps from $M$ to the space $\Pi F$. The hypothesis that $M$ is a manifold cannot be dropped;
what is used is that $M$ has partitions of unity and is locally contractible, which is where the analysis
of this chapter enters.

\begin{remark}\label{rem:sing}
Concordance is an equivalence relation on the points of any smooth set, but transitivity is not
automatic: two plots $\R\to X$ that agree at a point need not glue to a plot. One reparametrizes so that
each is constant near the ends (\cref{exer:sitting}). The same device makes the smooth singular complex a
Kan complex.
\end{remark}

\section{Subobjects cut out by open conditions}

\begin{definition}\label{def:emb}
$\Imm(M,N)$, $\Emb(M,N)$ and $\Diff(M)$ are the subobjects of $[M,N]$ (of $[M,M]$) whose plots are the
families $F\colon\R^n\times M\to N$ such that every $F_s$ is an immersion, an embedding, a diffeomorphism.
\end{definition}

For non-compact $M$ this definition of $\Emb$ is the weakest reasonable one, and there are others (requiring
$(s,x)\mapsto(s,F_s(x))$ to be an embedding, or requiring compact support). For compact $M$ they agree.

\begin{proposition}\label{prop:Dopen}\Srf[Hirsch 2.1.1--2.1.6]
If $M$ is compact, $\Imm(M,N)$, $\Emb(M,N)$ and $\Diff(M)$ are $D$-open subobjects of the mapping object.
That is: in a smooth family of maps from a compact manifold, the set of parameters at which the map is an
immersion (an embedding, a diffeomorphism) is open.
\end{proposition}

\begin{proof}
A plot $F$ induces a continuous map $\R^n\to C^\infty_W(M,N)$, since $F$ and its derivatives in the $M$
directions are uniformly continuous on compact subsets of $\R^n\times M$. For compact $M$ the weak and strong
topologies agree, and the three sets are open by \cref{thm:open}.
\end{proof}

The reformulation is modest, but it is the form in which the companion volume uses \cref{thm:open}: the
parameters of a network at which it is an embedding of a compact data manifold form an open set. It is also
the form that generalizes. A $D$-open subobject of a mapping object has the same local structure as the
mapping object, and so inherits whatever infinite-dimensional manifold structure the latter has
\cite{KM97}.

\section{Genericity}

\begin{theorem}[Baire property]\label{thm:baire}\Scl[Hirsch 2.4.4]
Every weakly closed subset of $C^r_S(M,N)$ is a Baire space in the strong topology: a countable
intersection of open dense subsets is dense.
\end{theorem}

A subset is \term{residual} if it contains a countable intersection of open dense sets. The theorems of the
next chapter assert that certain sets are residual, and \cref{thm:baire} is what turns such an assertion
into one about density. The proof realizes $C^r(M,N)$, by the prolongation
$j^r\colon C^r(M,N)\to C^0(M,J^r(M,N))$, as a weakly closed set of sections of the jet bundle; the strong
$C^r$ topology is by definition the one induced from the strong $C^0$ topology on sections, and the Baire
property is proved there by a diagonal argument.

\begin{remark}[What the topos does not see]\label{rem:nobaire}
The $D$-topology is defined from the plots, so openness is a notion internal to smooth sets.
Residuality is not. A residual set need not be $D$-open, and a set meeting every plot in a residual set
of parameters need not be residual. We know of no formulation of the Baire property of
$C^\infty_S(M,N)$ in terms of the sheaf $[M,N]$ alone. What the sheaf does see is the parametric form of
genericity (\cref{thm:parametric}): in a family rich enough to be a submersion onto its target, almost every
member is generic. In applications that statement is usually the one wanted.
\end{remark}

\section{Jets and partial differential relations}

The prolongation is a map of smooth sets $j^r\colon[M,N]\to\Gamma(J^r(M,N))$ into the smooth set of sections
of the jet bundle over $M$. It is a monomorphism, and its image consists of the \term{holonomic} sections. A
condition on maps that is defined by a subset $\mathcal R\subset J^r(M,N)$, such as being an immersion (the
$1$-jet is injective) or a Morse function (the $1$-jet meets the zero locus only where the $2$-jet is
non-degenerate), gives a pullback square
\[\begin{tikzcd}
{[M,N]_{\mathcal R}}\arrow[r]\arrow[d] & \Gamma(\mathcal R)\arrow[d]\\
{[M,N]}\arrow[r,"j^r"] & \Gamma(J^r(M,N))
\end{tikzcd}\]
in which the right-hand side involves no differential equation: $\Gamma(\mathcal R)$ is the object of
\term{formal solutions}. If $\mathcal R$ is open, the left-hand subobject is $D$-open when $M$ is compact, by
the argument of \cref{prop:Dopen}. Whether the top map induces an equivalence of shapes is the question
answered, for immersions, by the Smale--Hirsch theorem (\cref{thm:SH}).

\section*{Notes}

The weak and strong topologies and their properties are developed in Hirsch's Chapter~2 and, for
$C^\infty$ and jets, in Golubitsky--Guillemin and Mather. The convenient setting of \cite{KM97} gives
$C^\infty(M,N)$ the structure of an infinite-dimensional manifold whose smooth curves are the smooth families
with locally compact support in the parameter; \cref{prop:strongpath} is the reason for that restriction.
The comparison of the $D$-topology with the classical ones is \cite{CSW14}. \Cref{thm:oka} is called the
smooth Oka principle by Sati and Schreiber; earlier special cases are due to Madsen and Weiss.

\section*{Exercises}

\begin{exercise}\label{exer:weakopen}
Verify the claims of \cref{ex:weakfail}. Show also that $\Diff(\R)$ is not closed in $C^\infty_W(\R,\R)$.
\end{exercise}

\begin{exercise}\label{exer:strongpath}
Prove \cref{prop:strongpath} for $M=N=\R$. Deduce that the path component of $f$ in $C^0_S(\R,\R)$ is the set
of maps that agree with $f$ outside a compact set.
\end{exercise}

\begin{exercise}\label{exer:Dopen}
Show that a plot of $[M,N]$ is continuous as a map into $C^1_W(M,N)$, and give an example of a smooth family
$F\colon\R\times\R\to\R^2$ of maps for which $\{s:F_s\text{ is an embedding}\}$ is not open.
\end{exercise}

\begin{exercise}\label{exer:sitting}
Let $X$ be a smooth set. Show that concordance is an equivalence relation on $X(\pt)$.
\end{exercise}

\begin{exercise}\label{exer:morseweak}
A function $f\colon\R\to\R$ is Morse if $f''(x)\ne0$ wherever $f'(x)=0$. Show that the set of Morse functions
is dense but not open in $C^\infty_W(\R,\R)$.
\end{exercise}
\chapter{Sard's Theorem and Transversality}\label{ch:transversality}

This is the chapter on which the source turns. Its results say that the transverse cospans of
\cref{thm:transpull} are dense among all cospans, and that the same holds in families. The proofs reduce, by
one piece of linear algebra, to a single analytic theorem. The higher-categorical counterpart is a universal
property: there is a smallest enlargement of $\Mfd$ in which all fibre products exist and the transverse ones
are unchanged, and the genericity theorems say that every object of the enlargement is cobordant to an
object of $\Mfd$.

\section{The Morse--Sard theorem}

\begin{theorem}[Morse--Sard]\label{thm:sard}\Scl[Morse, Sard, Brown; Hirsch 3.1.3]
Let $f\colon M\to N$ be of class $C^r$ with $r>\max\{0,\dim M-\dim N\}$. Then the set of critical values of $f$
has measure zero in $N$, and the set of regular values is residual.
\end{theorem}

The proof is an induction on $\dim M$ by slicing with Fubini's theorem, with the differentiability hypothesis
used to control the image of the set where many derivatives vanish. Nothing in this book adds to it. The
theorem is the source of every statement of the form ``for almost all'' below.

\begin{corollary}\label{cor:noretract}\Scl[Hirsch 3.1.4]
A compact manifold with non-empty boundary admits no smooth retraction onto its boundary.
\end{corollary}

\begin{proof}
Let $\rho\colon M\to\partial M$ be one and $y$ a regular value of $\rho$; it is automatically a regular
value of $\rho|_{\partial M}=\id$. By \cref{thm:neat}, $\rho^{-1}(y)$ is a compact neat $1$-dimensional
submanifold with boundary $\rho^{-1}(y)\cap\partial M=\{y\}$. A compact $1$-manifold has an even number of
boundary points.
\end{proof}

The argument is the prototype for the whole subject: a count that would have to be odd is the boundary of
something, hence even. Brouwer's fixed point theorem follows (\cref{exer:brouwer}).

\section{The transversality theorems}

\begin{theorem}[Transversality]\label{thm:transversality}\Scl[Thom; Hirsch 3.2.1]
Let $A\subset N$ be a submanifold and $1\le r\le\infty$. The set of $f\in C^r(M,N)$ with $f\trans A$ is
residual, hence dense, in $C^r_S(M,N)$; it is open if $A$ is closed.
\end{theorem}

The proof uses a parametric statement, which is the form we shall use most, and the parametric statement
rests on a lemma worth isolating.

\begin{lemma}[Fibrewise transversality]\label{lem:fibre}\Scl
Let $F\colon V\times M\to N$ be smooth with $F\trans A$, let $W=F^{-1}(A)$ and let $\pi\colon W\to V$ be the
projection. Then $F_v=F(v,-)$ is transverse to $A$ if and only if $v$ is a regular value of $\pi$.
\end{lemma}

\begin{proof}
Fix $(v,x)\in W$, $z=F(v,x)$, and let $L\colon T_vV\oplus T_xM\to Q=T_zN/T_zA$ be the differential of $F$
followed by the quotient. By hypothesis $L$ is onto, and $T_{(v,x)}W=\ker L$. Write $L_M$ for the
restriction of $L$ to $T_xM$; then $F_v\trans A$ at $x$ means that $L_M$ is onto, and $d\pi$ is the
projection $\ker L\to T_vV$.

If $L_M$ is onto and $a\in T_vV$, choose $b$ with $L_M(b)=L(a,0)$; then $(a,-b)\in\ker L$ projects to $a$.
Conversely, if the projection is onto and $q=L(a,b)\in Q$, choose $(a,b')\in\ker L$; then
$q=L(a,b)-L(a,b')=L_M(b-b')$.
\end{proof}

\begin{theorem}[Parametric transversality]\label{thm:parametric}\Scl[Hirsch 3.2.7]
Let $F\colon V\times M\to N$ be of class $C^r$ with $r>\max\{0,\dim M+\dim A-\dim N\}$ and $F\trans A$. Then
the set of $v\in V$ with $F_v\trans A$ is residual in $V$, and its complement has measure zero.
\end{theorem}

\begin{proof}
Apply \cref{thm:sard} to $\pi\colon W\to V$, noting $\dim W-\dim V=\dim M+\dim A-\dim N$, and use
\cref{lem:fibre}.
\end{proof}

The lemma has a reading that will recur. The manifold $W$ is the fibre product of the whole family with $A$,
and it is always a manifold. The individual fibre products $F_v^{-1}(A)$ are its fibres over $V$. So a
non-transverse member $F_{v_0}^{-1}(A)$ of the family is a singular fibre of a map between manifolds, and
nearby regular fibres are its replacements. If $F$ is merely a plot of $[M,N]$, with no transversality
assumed, nothing can be said; the hypothesis $F\trans A$ is what makes a family ``rich enough'' in the sense
of \cref{rem:nobaire}. It holds in particular when $F$ is a submersion, for instance $F(v,x)=f(x)+v$ with
$N=V=\R^n$ (\cref{exer:translate}).

\begin{theorem}[Jet transversality]\label{thm:jettrans}\Scl[Thom; Hirsch 3.2.8, 3.2.9]
Let $A\subset J^r(M,N)$ be a submanifold and $r<s\le\infty$. The set of $f\in C^s(M,N)$ with $j^rf\trans A$ is
residual in $C^s_S(M,N)$, and open if $A$ is closed.
\end{theorem}

\section{Dimension counts}

A map transverse to a submanifold of codimension greater than $\dim M$ misses it. So genericity theorems
become existence theorems as soon as one can compute a codimension.

\begin{proposition}\label{prop:rank}\Scl
The $m\times n$ real matrices of rank $\rho$ form a submanifold of $\R^{m\times n}$ of dimension
$\rho(m+n-\rho)$, that is, of codimension $(m-\rho)(n-\rho)$.
\end{proposition}

\begin{corollary}\label{cor:counts}\Scl[Whitney; Hirsch 2.2.12, 2.2.13, 6.1.2]
Let $\dim M=m$, $\dim N=n$, with $M$ compact.
\begin{enumerate}[label=(\alph*)]
\item If $n\ge2m$, immersions are dense in $C^\infty(M,N)$.
\item If $n\ge2m+1$, embeddings are dense in $C^\infty(M,N)$.
\item Morse functions are open and dense in $C^\infty(M,\R)$.
\end{enumerate}
\end{corollary}

\begin{proof}
(a) In $J^1(M,N)$ the jets of rank $\rho<m$ form submanifolds of codimension $(m-\rho)(n-\rho)\ge n-m+1>m$. A
map whose $1$-jet prolongation is transverse to all of them misses them. (b) For an immersion $f$,
injectivity is the statement that $f\times f$ restricted to $M\times M\setminus\Delta$ misses $\Delta_N$, of
codimension $n>2m$; one applies a multijet form of \cref{thm:jettrans}. (c) A function is Morse when $df$,
as a section of $T^*M$, is transverse to the zero section, a closed submanifold of codimension $m$.
\end{proof}

The script checks \cref{prop:rank} by computing the rank of the differential of a local defining map for
several $(m,n,\rho)$, and checks (b) in a small case by sampling linear projections.

\section{Clean and transverse}

Transversality is a sufficient condition for a fibre product in $\Mfd$ to be correct. It is useful to
separate the two things it provides.

\begin{definition}\label{def:clean}
Maps $f\colon M\to N$ and $g\colon P\to N$ intersect \term{cleanly} if $Z=M\times_NP$ is a submanifold of
$M\times P$ and $TZ=TM\times_{TN}TP$. The \term{excess bundle} is the vector bundle
$E=\operatorname{coker}(TM\oplus TP\to TN)$ over $Z$, of rank $e=\dim N-\dim M-\dim P+\dim Z$.
\end{definition}

\begin{proposition}\label{prop:clean}\Scl
If $f$ and $g$ intersect cleanly then $Z$ is the fibre product in $\Mfd$ and $T$ preserves it. They are
transverse if and only if they intersect cleanly with $e=0$. If $M,P\subset N$ are compact submanifolds
meeting cleanly in $Z$, then for every sufficiently small generic perturbation $M'$ of $M$ the transverse
intersection $M'\cap P$ is cobordant to the zero set of a generic section of $E$ over $Z$.
\end{proposition}

So a clean intersection has the right tangent spaces and the wrong dimension, by $e$, and the excess bundle
measures exactly what a perturbation will cut away. A submanifold meets itself cleanly with $E$ its normal
bundle; the diagonal of $M\times M$ meets itself cleanly with $E=TM$. \Cref{ex:three}(b) is not clean. In
each case there is an object that records the correct answer before any perturbation is made.

\section{Derived manifolds}

By \cref{thm:MR} the function algebra of a transverse fibre product is a pushout of $C^\infty$-rings. The
derived fibre product is defined by the homotopy pushout.

\begin{definition}\label{def:scr}
A \term{simplicial $C^\infty$-ring} is a product-preserving functor $\Cart\to\Spc$; these form an
$\infty$-category $s\CinfRing$, with homotopy pushouts written $\otimes^{\mathbb L}$. Each has homotopy groups
$\pi_k$, with $\pi_0$ a $C^\infty$-ring and $\pi_k$ a $\pi_0$-module.
\end{definition}

\begin{theorem}[Universal property of derived manifolds]\label{thm:CS}
\Sli[Carchedi--Steffens \cite{CS19}; Spivak \cite{Spi10}]
There is an $\infty$-category $\DMfd$ that has finite limits and is idempotent complete, with a functor
$i\colon\Mfd\to\DMfd$ that preserves transverse fibre products and the terminal object, and that is universal
with these properties: for every idempotent complete $\infty$-category $\mathcal C$ with finite limits,
composition with $i$ is an equivalence between finite-limit-preserving functors $\DMfd\to\mathcal C$ and
functors $\Mfd\to\mathcal C$ preserving transverse fibre products and the terminal object. Moreover $\DMfd$
is equivalent to the opposite of the $\infty$-category of finitely presented simplicial $C^\infty$-rings, with
$i(M)=C^\infty(M)$, and $i$ is fully faithful.
\end{theorem}

The theorem says that nothing was chosen. Whoever wants all fibre products of manifolds, and wants the
transverse ones to stay as they are, is led to this $\infty$-category and no other. The role of idempotent
completeness was explained after \cref{thm:karoubi}. That the answer is an $\infty$-category and not a
category is forced: the last row of \cref{tab:four} has no $1$-categorical solution.

Locally every derived fibre product of manifolds is a derived zero locus, and those can be written down.

\begin{proposition}[Koszul model]\label{prop:koszul}\Sli[\cite{Spi10}, \cite{CS19}]
Let $E\to M$ be a vector bundle of rank $k$ with a section $s$, and let
$Z(s)=M\times^h_{s,E,0}M$ be the derived zero locus. Its simplicial $C^\infty$-ring corresponds to the
commutative differential graded algebra
\[\bigl(\Gamma(\Lambda^\bullet E^\vee),\ \iota_s\bigr):\qquad
\cdots\to\Gamma(\Lambda^2E^\vee)\xrightarrow{\ \iota_s\ }\Gamma(E^\vee)\xrightarrow{\ \iota_s\ }C^\infty(M),\]
so $\pi_0=C^\infty(M)/(s)$. If $s$ is transverse to the zero section then $\pi_k=0$ for $k>0$ and
$Z(s)=i\bigl(s^{-1}(0)\bigr)$. For maps $f\colon M\to N$ and $g\colon P\to N$, the derived fibre product
$M\times^h_NP$ is the derived zero locus of a section of $(f\times g)^*TN$ over a neighbourhood of
$M\times_NP$ in $M\times P$.
\end{proposition}

The number $\vdim Z(s)=\dim M-k$ is the \term{virtual dimension}; for a derived fibre product it is
$\dim M+\dim P-\dim N$ whether or not the maps are transverse.

\begin{example}\label{ex:derived}
(a) $M=\R$, $E$ trivial of rank $1$, $s=x^2$. The complex is $C^\infty(\R)\xrightarrow{x^2}C^\infty(\R)$, which
is injective, so $\pi_1=0$ and $Z(s)$ is the classical object $\R[\varepsilon]/(\varepsilon^2)$ of virtual
dimension $0$: not a manifold, and not derived either.

(b) $M=\pt$, $E=\R$, $s=0$. The complex is $\R\xrightarrow{0}\R$; $\pi_0=\pi_1=\R$. This is the point of
virtual dimension $-1$ in \cref{tab:four}.

(c) $s=0$ on any $E\to M$: the algebra is $\Gamma(\Lambda^\bullet E^\vee)$ with zero differential, of virtual
dimension $\dim M-k$. It is the derived self-intersection of the zero section, and by the tubular
neighbourhood theorem the derived self-intersection of any submanifold with normal bundle $E$.

(d) The diagonal: $M\times^h_{M\times M}M$ has functions $\Gamma(\Lambda^\bullet T^*M)$, the differential forms
placed in homological degrees, and virtual dimension $0$. Its ``number of points'' will turn out to be the
Euler characteristic (\cref{ch:degree}).
\end{example}

Every object of $\DMfd$ has a tangent complex at each point. For $X=M\times^h_NP$ at $(x,y)$ over $z$ it is
\[\mathbb T_X=\bigl[T_xM\oplus T_yP\xrightarrow{\ df-dg\ }T_zN\bigr]\]
in degrees $0$ and $1$, with $H^0$ the fibre product of tangent spaces and $H^1$ the fibre of the excess
bundle. Its Euler characteristic is the virtual dimension.

\begin{theorem}[Transversality, derived form]\label{thm:derivedtrans}\Srf[\cite{Spi10}, \cite{CS19}]
For smooth maps $f\colon M\to N$ and $g\colon P\to N$, the derived fibre product $M\times^h_NP$ lies in the
essential image of $i\colon\Mfd\to\DMfd$ if and only if $f\trans g$.
\end{theorem}

\begin{proof}
If $f\trans g$, then $i$ preserves the fibre product by \cref{thm:CS}. Conversely, if
$M\times^h_NP\simeq i(Z)$ for a manifold $Z$, its tangent complex is concentrated in degree $0$ at every
point, so $H^1=T_zN/(df\,T_xM+dg\,T_yP)$ vanishes at every point of the fibre product, which is
transversality.
\end{proof}

Compare the situation in $\Mfd$. There a fibre product may fail to exist, or exist and be wrong, or exist
and be right without one's knowing which. In $\DMfd$ it always exists and is always right, and it is a
manifold in exactly the transverse case.

\section{Genericity as a statement about derived manifolds}

\begin{theorem}\label{thm:spivak}\Sli[Spivak \cite{Spi10}]
Compact derived manifolds that are locally derived zero loci have a cobordism theory, and
$i$ induces an isomorphism from the cobordism ring of manifolds to it. In particular a compact
$X=M\times^h_NP$ has a class $[X]$ in the cobordism group $\mathfrak N_d$ of dimension
$d=\dim M+\dim P-\dim N$ (\cref{ch:cobordism}), represented by $M\times_NP'$ for any $g'\trans f$ sufficiently
close to $g$.
\end{theorem}

Surjectivity is the transversality theorem: perturb. Injectivity, and the independence of the perturbation,
is \cref{lem:fibre} applied to a path of perturbations: the fibre product of the path with $A$ is a manifold
with boundary whose two ends are the two candidates. Putting \cref{thm:CS,thm:derivedtrans,thm:spivak}
together gives a statement of the method of the source that makes no mention of perturbation until the
end.

\begin{principle}[Generic position]\label{prin:generic}\Srf[Hirsch, Ch.~3; \cite{Spi10}]
Every finite diagram of manifolds has a limit in $\DMfd$. If the limit is compact it has a well-defined
cobordism class. The class is represented by the honest limit of any sufficiently generic perturbation of
the diagram, and two such representatives are cobordant by the limit of a generic path between the
perturbations.
\end{principle}

\begin{example}\label{ex:xsqagain}
The zero set of $x^2-\varepsilon$ on $\R$, as an oriented $0$-manifold, is $\{+\sqrt\varepsilon\}$ with sign
$+$ and $\{-\sqrt\varepsilon\}$ with sign $-$ for $\varepsilon>0$, and empty for $\varepsilon<0$. Both have
class $0\in\Omega_0=\Z$, and the cobordism between them is the parabola $\{x^2=\varepsilon\}$ in the
$(\varepsilon,x)$-plane, a manifold by \cref{lem:fibre}. The derived zero set at $\varepsilon=0$ is the dual
numbers, with class $0$. Its length as an algebra is $2$, which is the count of complex roots; the
cobordism class and the length are different invariants and only the first is stable under real
perturbation.
\end{example}

An object of negative virtual dimension, such as \cref{ex:derived}(b), has class zero for trivial reasons
and is nevertheless not empty. The cobordism class is the part of a derived manifold that perturbation can
reach, and it is the only part the classical theory uses.

\section*{Notes}

Transversality is Thom's; the parametric form and its use as the organizing principle are the substance of
Hirsch's Chapter~3 and of Abraham's work. Derived manifolds were introduced by Spivak, after Lurie's sketch
in his work on derived algebraic geometry; variants are due to Borisov and Noel and to Joyce, and the
universal property is \cite{CS19}, building on \cite{Mac17}. Clean intersections are Bott's. The excess
bundle and the associated formula are Quillen's in cobordism and Fulton--MacPherson's in algebraic
geometry.

\section*{Exercises}

\begin{exercise}\label{exer:translate}
Let $f\colon M\to\R^n$ be smooth and $A\subset\R^n$ a submanifold. Show that $f+v$ is transverse to $A$ for
almost every $v\in\R^n$. Deduce that almost every translate of a curve in $\R^3$ misses a given curve.
\end{exercise}

\begin{exercise}\label{exer:rank}
Prove \cref{prop:rank}.
\end{exercise}

\begin{exercise}\label{exer:tor}
Compute the homotopy groups of $\R\otimes^{\mathbb L}_{C^\infty(\R)}\R$, where both maps are evaluation at
$0$, and of $C^\infty(\R)\otimes^{\mathbb L}_{C^\infty(\R)}\R$, where the first map is $x\mapsto x^2$. Compare
with \cref{ex:derived}.
\end{exercise}

\begin{exercise}\label{exer:morsegeneric}
Let $U\subset\R^m$ be open and $f\colon U\to\R$ smooth. Show that $f_a(x)=f(x)+\langle a,x\rangle$ is a Morse
function for almost every $a\in\R^m$.
\end{exercise}

\begin{exercise}\label{exer:brouwer}
Deduce from \cref{cor:noretract} that every continuous map $D^n\to D^n$ has a fixed point.
\end{exercise}
\part{Bundles and Invariants}
\chapter{Vector Bundles and Classifying Maps}\label{ch:bundles}

The first half of Hirsch's fourth chapter proves that a vector bundle over $B\times I$ is a product in the
$I$ direction, deduces that homotopic maps pull back isomorphic bundles, and classifies bundles over an
$n$-manifold by homotopy classes of maps to a Grassmannian. The three statements are one statement about a
stack: it is concordance-invariant, so its sections are classified by maps to its shape. The same theorem,
applied to other stacks, gives de Rham's theorem and the classification of principal bundles, and in
later chapters the classification of immersions and of manifolds up to cobordism.

\section{The stack of vector bundles}

\begin{definition}\label{def:vect}
For a manifold $M$, $\Vect_k(M)$ is the groupoid of real vector bundles of rank $k$ over $M$ and their
isomorphisms. Pullback makes $M\mapsto\Vect_k(M)$ a functor $\Mfd^{\op}\to\cat{Gpd}$.
\end{definition}

\begin{proposition}\label{prop:stack}\Scl
$\Vect_k$ is a stack: bundles and their isomorphisms glue along open covers. Its restriction to $\Cart$ is
equivalent to $U\mapsto B\,C^\infty(U,\GL_k)$, the groupoid with one object and automorphisms the smooth
maps $U\to\GL_k(\R)$. Hence $\Vect_k\simeq B\GL_k$, the delooping in $\mathbf H$ of the sheaf of groups
represented by $\GL_k(\R)$, and
\[\Vect_k(M)\ \simeq\ \mathbf H(M,B\GL_k).\]
\end{proposition}

\begin{proof}
Descent for bundles is the construction of a bundle from transition functions satisfying the cocycle
condition. Over a Euclidean space every bundle is trivial (by \cref{thm:bundlehomotopy} below, since
$\R^n$ is contractible), which identifies the restriction. An object of $\mathbf H$ is determined by its
restriction to $\Cart$.
\end{proof}

So ``a vector bundle is a map to $B\GL_k$'' is true as stated, in $\mathbf H$, with no homotopy classes
taken; a cocycle $(g_{ij})$ on a cover is the same map written out on the \v{C}ech nerve of the cover. The
object $B\GL_k$ is not a smooth set. Its value on a point is the groupoid of $k$-dimensional vector spaces,
and it is the first place in this book where the $\infty$-topos of \cref{def:shape}, and not just its
$0$-truncated part, is needed.

\section{Homotopy invariance}

\begin{theorem}\label{thm:bundlehomotopy}\Scl[Hirsch 4.1.5, 4.2.4]
Every vector bundle over $M\times I$ is isomorphic to $(\xi|_{M\times0})\times I$. Consequently, if
$f_0,f_1\colon M\to N$ are homotopic and $\xi$ is a bundle over $N$, then $f_0^*\xi\cong f_1^*\xi$.
\end{theorem}

The proof chooses a connection, or in Hirsch's version covers $M\times I$ by product neighbourhoods on which
the bundle is trivial and patches with a partition of unity; either way a bundle over $M\times I$ is
transported along the $I$ direction. In the language of \cref{def:shape}, two bundles over $M$ are concordant
if they are the ends of a bundle over $M\times\R$, and the theorem says that concordant bundles are
isomorphic. The converse is trivial.

\section{Classification}

Let $\Gr_k(\R^s)$ be the Grassmannian of $k$-planes in $\R^s$, a compact manifold of dimension $k(s-k)$, with
tautological bundle $\gamma_k$.

\begin{theorem}[Classification of vector bundles]\label{thm:classify}\Scl[Hirsch 4.3.4]
Let $\xi$ be a vector bundle of rank $k$ over a manifold $M$ of dimension $n$. If $s\ge k+n$ there is a map
$f\colon M\to\Gr_k(\R^s)$ with $f^*\gamma_k\cong\xi$, and if $s>k+n$ it is unique up to homotopy. So for
$s>k+n$,
\[\pi_0\Vect_k(M)\ \cong\ [M,\Gr_k(\R^s)].\]
\end{theorem}

Hirsch proves existence by embedding the bundle in a trivial one, which is a transversality argument, and
uniqueness by the same argument one dimension up. The dimension restrictions are the connectivity of the
inclusion $\Gr_k(\R^s)\to\Gr_k(\R^\infty)=B\Or(k)$.

\begin{theorem}[Concordance classification]\label{thm:concordance}\Sli[\cite{BEBP24}; cf.\ \cref{thm:oka}]
For every $F\in\mathbf H$ and every manifold $M$, the set of concordance classes of sections of $F$ over $M$
is in natural bijection with $[M,\Pi F]$, and more precisely $\Pi[M,F]\simeq\Map(M,\Pi F)$.
\end{theorem}

For $F=B\GL_k$ the shape is the classifying space of the topological group $\GL_k(\R)$, which is
$B\Or(k)$ by Gram--Schmidt (\cref{exer:gram}). Concordance classes are isomorphism classes by
\cref{thm:bundlehomotopy}. So \cref{thm:concordance} gives $\pi_0\Vect_k(M)\cong[M,B\Or(k)]$, which is
\cref{thm:classify} with $s=\infty$; and it gives more, namely the homotopy type
$\Map(M,B\Or(k))$ for the space of all rank-$k$ bundles over $M$, whose fundamental group at $\xi$ is the
group of components of the gauge group of $\xi$.

\Cref{tab:oka} lists other instances. In each row the left-hand side is a geometric structure that can be
pulled back, the right-hand side is a space, and the theorem says that deformation classes of the
structure on $M$ are homotopy classes of maps into the space.

\begin{table}[ht]\centering\small
\begin{tabular}{@{}p{0.34\textwidth}p{0.2\textwidth}p{0.36\textwidth}@{}}
\toprule
$F$ (sections over $M$) & shape $\Pi F$ & classical name\\
\midrule
maps to a manifold $N$ & $N$ & smooth approximation (\cref{cor:homotopy})\\
vector bundles of rank $k$ & $B\Or(k)$ & \cref{thm:classify}\\
principal $G$-bundles, $G$ a Lie group & $BG$ & classification of principal bundles\\
vector bundles with connection & $B\Or(k)$ & connections form a contractible space\\
closed differential $p$-forms & $K(\R,p)$ & de Rham's theorem\\
all differential $p$-forms & $\pt$ & forms are contractible\\
bundles of closed $d$-manifolds & $\coprod_{[W]}B\Diff(W)$ & classification of manifold bundles\\
\bottomrule
\end{tabular}
\caption{Instances of \cref{thm:concordance}. The union in the last row is over diffeomorphism classes of
closed $d$-manifolds.}\label{tab:oka}
\end{table}

The fifth row deserves a comment. Two closed $p$-forms on $M$ are concordant exactly when they are
cohomologous (\cref{exer:forms} for $p=1$), so the theorem says $H^p_{\mathrm{dR}}(M)\cong[M,K(\R,p)]$, which is
de Rham's theorem. The sheaf of closed forms is not homotopy invariant; its concordance classes are.

\section{Bundles as idempotents}

\begin{theorem}\label{thm:swan}\Scl[Swan; Hirsch 4.3.3]
Every vector bundle $\xi$ over a manifold $M$ has a complement: a bundle $\eta$ with $\xi\oplus\eta$ trivial.
The functor of sections is an equivalence from vector bundles over $M$ to finitely generated projective
$C^\infty(M)$-modules, taking $\oplus$ and $\otimes$ to $\oplus$ and $\otimes_{C^\infty(M)}$.
\end{theorem}

So the category of vector bundles over $M$ is the idempotent completion of the category of trivial bundles,
just as $\Mfd$ is the idempotent completion of $\cat{Open}$ (\cref{thm:karoubi}), and for the same reason:
finite dimension bounds the number of charts needed. A complement exhibits $\xi$ as the image of an
idempotent endomorphism $e$ of a trivial bundle $M\times\R^s$, that is, of a smooth map from $M$ to the
manifold of idempotent $s\times s$ matrices of rank $k$. Projecting an idempotent to the orthogonal
projection with the same image identifies that manifold, up to homotopy, with $\Gr_k(\R^s)$. A classifying
map and a complement are the same datum.

\begin{example}\label{ex:taut}
Over the circle $\RP^1=\R/\pi\Z$ the matrix
$e(\theta)=\bigl(\begin{smallmatrix}\cos^2\theta&\cos\theta\sin\theta\\\cos\theta\sin\theta&\sin^2\theta
\end{smallmatrix}\bigr)$ is an idempotent of rank $1$; its image is the tautological line bundle. A unit
vector in the image transported once around the circle returns to its negative, so the bundle is the Möbius
band. Over $S^n$ the idempotent $e(x)=I-xx^{\mathsf T}$ on $\R^{n+1}$ has image $TS^n$ and complement the
trivial normal line. The script checks both.
\end{example}

\section{Orientations and tangential structures}

A bundle is \term{orientable} if its transition functions can be chosen in $\GL_k^+(\R)$; an
\term{orientation} is such a choice up to equivalence. In terms of classifying maps, an orientation is a
lift of $M\to B\Or(k)$ through the double cover $B\SO(k)\to B\Or(k)$, and the obstruction to its existence is
the class $w_1(\xi)\in H^1(M;\Z/2)$ pulled back from the generator of $H^1(B\Or(k);\Z/2)$. A manifold is
orientable if its tangent bundle is.

\begin{definition}\label{def:theta}
A \term{tangential structure} is a map of spaces $\theta\colon B\to B\Or(d)$. A \term{$\theta$-structure} on a
$d$-manifold $M$ is a lift of the classifying map $\tau_M\colon M\to B\Or(d)$ of $TM$ through $\theta$.
\end{definition}

Orientations ($B=B\SO(d)$), framings ($B=\pt$), spin structures, almost complex structures, and maps to a
fixed background space $X$ ($B=B\Or(d)\times X$) are all of this form. Because $\theta$-structures pull back
along open embeddings and are defined by a lifting problem in spaces, the stack of $d$-manifolds with
$\theta$-structure is again an object to which \cref{thm:concordance} applies; this is how the tangential
structures of \cref{ch:cobordism} enter.

\begin{theorem}\label{thm:nonorient}\Scl[Hirsch 4.4.7]
A closed non-orientable $n$-manifold does not embed in a simply connected $(n+1)$-manifold. In particular
$\RP^{2n}$ does not embed in $\R^{2n+1}$.
\end{theorem}

\begin{proof}
Let $M\subset N$ be such an embedding with normal line bundle $\nu$. A loop in $N$ transverse to $M$ meets
it in an even number of points, since the count modulo $2$ is a homotopy invariant
(\cref{thm:intersection}) and the loop is null-homotopic. A loop in $M$ along which $\nu$ is non-trivial
can be pushed off $M$ to a loop meeting $M$ once. So $\nu$ is trivial, and
$TM\oplus\nu\cong TN|_M$ gives $w_1(TM)=w_1(TN|_M)=0$, as $N$ is simply connected.
\end{proof}

\section{The tangent bundle as a map}

The tangent bundle of an $m$-manifold is a point $\tau_M$ of $\Map(M,B\Or(m))$. For a submanifold
$M\subset\R^{m+k}$ the classifying map can be taken to be the Gauss map $x\mapsto T_xM\in\Gr_m(\R^{m+k})$, and
the normal bundle is classified by $x\mapsto(T_xM)^\perp$. The normal bundle depends on the embedding, but
its class in $[M,B\Or]$, where $B\Or=\colim_kB\Or(k)$, does not; this \term{stable normal bundle} $\nu_M$ is
inverse to $\tau_M$ under direct sum, and \cref{ch:tubular} shows that it is well defined as a point of a
contractible space of choices. Characteristic classes of $M$ are the pullbacks along $\tau_M$ or $\nu_M$
of classes in $H^*(B\Or;\Z/2)=\Z/2[w_1,w_2,\dots]$. Every invariant of manifolds used in
\cref{ch:degree,ch:cobordism} is obtained from $\tau_M$, as a consequence of the remark after
\cref{prop:jets}.

\section*{Notes}

Steenrod's book is the classical source for classification by maps to Grassmannians; Hirsch's treatment by
transversality is his own. The stack-theoretic formulation is standard; see \cite{Sch13}. The equivalence
with projective modules is Swan's, after Serre. \Cref{thm:concordance} has a long prehistory: Brown
representability, Gromov's sheaf-theoretic h-principle, and Madsen and Weiss's use of concordance in the
proof of the Mumford conjecture.

\section*{Exercises}

\begin{exercise}\label{exer:mobius}
Show that the tautological line bundle over $\RP^1$ is not trivial, and that its total space is the open
Möbius band.
\end{exercise}

\begin{exercise}\label{exer:swan}
Let $e\colon M\to\R^{s\times s}$ be smooth with $e^2=e$ and constant rank $k$. Show that the image of $e$ is a
vector bundle whose module of sections is $e\,C^\infty(M)^s$, a direct summand of a free module.
\end{exercise}

\begin{exercise}\label{exer:TS}
Show that $TS^n\oplus\R$ is trivial, and conclude that the composite $S^n\to B\Or(n)\to B\Or$ classifying the
stable tangent bundle is null-homotopic.
\end{exercise}

\begin{exercise}\label{exer:forms}
Show that two closed $1$-forms on $M$ are concordant (are the restrictions to $M\times\{0\}$ and
$M\times\{1\}$ of a closed $1$-form on $M\times\R$) if and only if they differ by an exact form.
\end{exercise}

\begin{exercise}\label{exer:lines}
Show that isomorphism classes of line bundles over $M$ correspond to $H^1(M;\Z/2)=\Hom(\pi_1M,\Z/2)$, and
count them for the torus and for the Klein bottle. Which one is the determinant of the tangent bundle?
\end{exercise}
\chapter{Tubular Neighbourhoods, Collars, and Spaces of Choices}\label{ch:tubular}

The second half of Hirsch's fourth chapter proves that a submanifold has a tubular neighbourhood, that the
boundary of a manifold has a collar, and that each is unique up to isotopy. The uniqueness statements are
what allow later constructions (connected sums, gluing of cobordisms, the Pontryagin--Thom collapse) to be
called well defined. ``Unique up to isotopy'' says that a space of choices is connected. In every case in
this chapter the space of choices is in fact contractible, and that stronger statement is what makes the
constructions functorial.

\section{Tubular neighbourhoods}

\begin{definition}\label{def:tub}
Let $M\subset N$ be a submanifold, both without boundary. A \term{tubular neighbourhood} of $M$ in $N$ is a
vector bundle $\xi=(p,E,M)$ with an embedding $\varphi\colon E\to N$ onto an open set that restricts to the
identity on the zero section. The \term{normal bundle} is $\nu=\nu(M,N)=TN|_M/TM$.
\end{definition}

The differential of $\varphi$ along the zero section identifies $\xi$ with $\nu$, so one may take $\xi=\nu$.

\begin{theorem}[Tubular neighbourhoods]\label{thm:tubular}\Scl[Hirsch 4.5.2, 4.5.3]
Every submanifold $M\subset N$ has a tubular neighbourhood, and any two are isotopic relative to $M$.
\end{theorem}

The existence proof embeds $N$ in a Euclidean space, uses the normal exponential map
$(x,v)\mapsto x+v$ there, which is a local diffeomorphism along the zero section by the inverse function
theorem, and retracts back to $N$. For the unit circle in the plane the map is
$(\theta,t)\mapsto(1+t)(\cos\theta,\sin\theta)$, with Jacobian determinant $1+t$; it is an embedding for
$t>-1$. In general the radius is limited by the focal points of $M$ (\cref{exer:focal}).

\begin{theorem}\label{thm:tubcontract}\Sli[folklore; see \cite{God07}]
The space $\Tub(M,N)$ of tubular neighbourhoods $\varphi\colon\nu\to N$ whose differential along the zero
section induces the identity of $\nu$ is contractible.
\end{theorem}

\begin{proof}[Idea of proof]
Given two, first shrink one fibrewise until its image lies in the image of the other; this is a canonical
path. Then $\psi=\varphi_0^{-1}\varphi_1$ is an embedding of $\nu$ in itself, fixing the zero section, with
fibre derivative the identity there, and $\psi_t(v)=\psi(tv)/t$ for $0<t\le1$ extends smoothly to $t=0$ by the
identity. Both steps depend continuously on parameters.
\end{proof}

With boundary, the right notion is that of a neat submanifold (\cref{def:neat}), and the theorems hold with
tubular neighbourhoods that restrict to tubular neighbourhoods in the boundary (Hirsch 4.6.3, 4.6.4).

\section{Collars and gluing}

\begin{theorem}[Collars]\label{thm:collar}\Scl[Brown; Hirsch 4.6.1]
For a manifold $M$ with boundary there is an embedding $\partial M\times[0,\infty)\to M$ extending the
identity of $\partial M\times0$. Any two are isotopic, and the space of collars is contractible.
\end{theorem}

\begin{theorem}[Gluing]\label{thm:gluing}\Scl[Hirsch 8.2.1, 8.2.3]
Let $P,Q$ be manifolds with boundary and $f\colon\partial Q\to\partial P$ a diffeomorphism. The space
$P\cup_fQ$ has a smooth structure restricting to the given ones on $P$ and $Q$; any two such structures are
diffeomorphic by a diffeomorphism supported near the seam; and isotopic gluing maps give diffeomorphic
results.
\end{theorem}

The smooth structure is built by choosing collars on both sides and declaring the union of the two collars
to be a product neighbourhood of the seam. So composition of cobordisms is defined only after a choice, and
the choice is from a contractible space. A category whose composition depends on contractible choices is
an $\infty$-category presented in the usual way, and this is the elementary reason that cobordism categories
are $\infty$-categories (\cref{ch:cobordism}). The dependence on the isotopy class of $f$ is real: gluing two
discs along a diffeomorphism of $S^{n-1}$ not isotopic to one that extends over the disc can give an exotic
sphere (\cref{exer:alexander}).

\section{Contractible spaces of choices}

\begin{principle}[Choices]\label{prin:choices}\Srf
Let $p\colon\mathcal E\to\mathcal B$ be a map of spaces whose fibres are contractible. Then $p$ is an
equivalence, and every functor on $\mathcal E$ factors essentially uniquely through $\mathcal B$. A
construction on objects $b$ that requires a choice $c$ from a contractible space $\mathcal E_b$ is
therefore as good as a construction on the objects alone.
\end{principle}

\Cref{tab:choices} lists the cases used in this book. The first four are convex or affine and need no
argument. The fifth and sixth are \cref{thm:tubcontract,thm:collar}. The seventh is Whitney's theorem with
parameters.

\begin{table}[ht]\centering\small
\begin{tabular}{@{}p{0.42\textwidth}p{0.46\textwidth}@{}}
\toprule
choice & what it makes well defined\\
\midrule
inner product on a vector bundle & reduction $B\GL_k\simeq B\Or(k)$; sphere and disc bundles\\
Riemannian metric on $M$ & exponential maps, gradients\\
connection on a bundle & parallel transport, \cref{thm:bundlehomotopy}\\
partition of unity subordinate to a cover & gluing of local constructions\\
tubular neighbourhood of $M\subset N$ & collapse maps, normal data\\
collar of $\partial M$ & gluing, composition of cobordisms\\
embedding $M\to\R^\infty$ & stable normal bundle; a model of $B\Diff(M)$\\
\bottomrule
\end{tabular}
\caption{Contractible spaces of choices.}\label{tab:choices}
\end{table}

\begin{theorem}\label{thm:embcontract}\Scl[Whitney]
For every manifold $M$ the space $\Emb(M,\R^\infty)=\colim_q\Emb(M,\R^q)$ is weakly contractible.
\end{theorem}

It is non-empty by \cref{thm:whitney}. Contractibility is a two-step linear homotopy: move a family of
embeddings into the odd coordinates, then slide it to a fixed embedding placed in the even coordinates
(\cref{exer:shift}). The group $\Diff(M)$ acts freely on $\Emb(M,\R^\infty)$, and the quotient, the space of
submanifolds of $\R^\infty$ diffeomorphic to $M$, is therefore a model for the classifying space
$B\Diff(M)$. A map $B\to B\Diff(M)$ is a family of submanifolds of $\R^\infty$ parametrized by $B$, that is, a
bundle of manifolds with fibre $M$; this is the last row of \cref{tab:oka}.

\section{The collapse map and Thom spaces}

\begin{definition}\label{def:thom}
The \term{Thom space} $\Th(\xi)$ of a vector bundle $\xi$ over a compact base is the one-point
compactification of its total space; equivalently the quotient $D(\xi)/S(\xi)$ of the unit disc bundle by
the unit sphere bundle, for any inner product.
\end{definition}

\begin{proposition}\label{prop:thom}\Scl
$\Th(\xi\times\eta)\cong\Th(\xi)\wedge\Th(\eta)$; $\Th(\xi\oplus\R^j)\cong\Sigma^j\Th(\xi)$; the Thom space of
the trivial bundle $M\times\R^k$ is $\Sigma^kM_+$; and the Thom space of the tautological line bundle over
$\RP^n$ is $\RP^{n+1}$.
\end{proposition}

Let $M\subset\R^{n+k}$ be a closed submanifold of dimension $n$ with a tubular neighbourhood
$\varphi\colon\nu\to U$. The \term{Pontryagin--Thom collapse} is the map
\[c\colon S^{n+k}=\R^{n+k}\cup\{\infty\}\longrightarrow U^+\cong\Th(\nu)\]
that is $\varphi^{-1}$ on $U$ and sends everything else to the point at infinity. More generally an
embedding $M\subset N$ of closed manifolds gives $N_+\to\Th(\nu(M,N))$. By \cref{thm:tubcontract} and
\cref{prin:choices} the collapse is a well-defined point of a mapping space, not merely a homotopy
class. It is a map in the wrong direction, from the ambient manifold to (a twisted suspension of) the
submanifold, and it is the geometric source of all wrong-way maps in cohomology.

The Thom space depends on the embedding only through a suspension. By \cref{thm:embcontract} the
desuspended suspension spectrum
\[M^{-TM}=\Sigma^{-(n+k)}\Sigma^\infty\Th(\nu)\]
is well defined; it is the Thom spectrum of the stable normal bundle $\nu_M=-TM$.

\begin{theorem}[Atiyah duality]\label{thm:atiyah}\Scl[Atiyah \cite{Ati61}; Milnor--Spanier]
For a closed manifold $M$ the suspension spectrum $\Sigma^\infty_+M$ is dualizable in the stable homotopy
category, with dual $M^{-TM}$. The coevaluation $\mathbb S\to\Sigma^\infty_+M\wedge M^{-TM}$ is the collapse
$S^{n+k}\to\Th(\nu)$ followed by the diagonal $\Th(\nu)\to M_+\wedge\Th(\nu)$. The evaluation is the collapse
for the diagonal embedding $M\to M\times M$, whose normal bundle is $TM$.
\end{theorem}

Dualizability and its use are recalled in \cref{app:toolkit}. The theorem is the homotopy-theoretic form of
Poincaré duality, and \cref{ch:degree} uses it to compute Euler characteristics.

\section{The normal bundle without choices}

There is a short exact sequence $0\to TM\to TN|_M\to\nu\to0$, so the normal bundle is the cofibre of the
map of tangent bundles; in the notation of \cref{ch:transversality} it is the degree-$1$ part of the
relative tangent complex of $M\to N$. The tubular neighbourhood theorem says that the germ of $N
$ along $M$ is diffeomorphic to the germ of $\nu$ along its zero section. There is a canonical object behind
the non-canonical diffeomorphism.

\begin{proposition}[Deformation to the normal bundle]\label{prop:deformation}\Scl
There is a manifold $\mathcal D(M,N)$ with a submersion to $\R$ whose fibre over $t\ne0$ is $N$ and whose
fibre over $0$ is $\nu(M,N)$, containing $M\times\R$ as a closed submanifold. It is functorial for maps of
pairs. A tubular neighbourhood is a trivialization of this family near $M\times\R$.
\end{proposition}

For the diagonal $M\subset M\times M$ this is Connes's tangent groupoid, interpolating between the pair
groupoid of $M$ and the tangent bundle. The proposition illustrates a general pattern: where the classical
text makes a choice and proves uniqueness up to isotopy, there is usually a canonical object in a larger
category of which the choices are trivializations.

By \cref{ex:derived}(c) the derived self-intersection $M\times^h_NM$ depends only on the normal bundle: its
functions are $\Gamma(\Lambda^\bullet\nu^\vee)$. Its virtual dimension is $2\dim M-\dim N$, and its cobordism
class is that of the zero set of a generic section of $\nu$ (\cref{prop:clean}).

\section*{Notes}

Hirsch's proofs of uniqueness of tubular neighbourhoods and collars give isotopies depending continuously on
the data, which is most of the way to contractibility; the statements with parameters are spelled out in
\cite{God07} and in the literature on cobordism categories \cite{GMTW09}. The collapse map is Pontryagin's
for framed submanifolds and Thom's in general. For Thom spectra of virtual bundles and their multiplicative
properties see \cite{ABGHR14}. The deformation to the normal cone is due to Fulton and MacPherson in
algebraic geometry; the smooth version is in Connes's and Higson's work on index theory.

\section*{Exercises}

\begin{exercise}\label{exer:gram}
Show that the space of inner products on a finite-dimensional real vector space is convex, and that
$\Or(k)\to\GL_k(\R)$ is a homotopy equivalence. Deduce $B\GL_k(\R)\simeq B\Or(k)$.
\end{exercise}

\begin{exercise}\label{exer:shift}
Prove \cref{thm:embcontract} for $\pi_0$: any two embeddings $M\to\R^\infty$ are isotopic through embeddings.
Explain why the straight-line homotopy between two embeddings in $\R^q$ need not be an isotopy.
\end{exercise}

\begin{exercise}\label{exer:focal}
For a closed plane curve with curvature $\kappa$ and inward unit normal $n$, show that
$(s,t)\mapsto\gamma(s)+t\,n(s)$ has Jacobian determinant $1-t\kappa(s)$. Find the largest $r$ such that the
map is a local diffeomorphism for $|t|<r$ when the curve is the ellipse with semi-axes $a>b$.
\end{exercise}

\begin{exercise}\label{exer:thom}
Prove \cref{prop:thom}.
\end{exercise}

\begin{exercise}\label{exer:alexander}
Let $f$ be a diffeomorphism of $S^{n-1}$ and $\Sigma_f=D^n\cup_fD^n$. Show that $\Sigma_f$ is homeomorphic to
$S^n$, and diffeomorphic to $S^n$ if $f$ extends to a diffeomorphism of $D^n$.
\end{exercise}
\chapter{Degree, Intersection Numbers, and the Euler Characteristic}\label{ch:degree}

Hirsch's fifth chapter is where the method of the Preface first produces numbers. Each invariant is a signed
count of points in a transverse preimage, and each proof of invariance exhibits two such counts as the two
ends of a compact oriented $1$-manifold. In the terms of \cref{ch:transversality}, each invariant is the
class in $\Omega_0=\Z$ (or $\mathfrak N_0=\Z/2$) of a compact derived manifold of virtual dimension zero. This
chapter records the invariants in that form and then gives the second, independent description of the Euler
characteristic and the Lefschetz number as traces.

\section{Degree}

\begin{definition}\label{def:degree}
Let $M,N$ be closed oriented $n$-manifolds with $N$ connected, $f\colon M\to N$ smooth, and $y$ a regular value.
The \term{degree} is $\deg f=\sum_{x\in f^{-1}(y)}\operatorname{sign}\det df_x$. Without orientations the
\term{mod $2$ degree} is $\#f^{-1}(y)$ modulo $2$.
\end{definition}

\begin{theorem}\label{thm:degree}\Scl[Brouwer; Hirsch 5.1.6, 5.1.8]
\begin{enumerate}[label=(\alph*)]
\item $\deg f$ does not depend on the regular value, and homotopic maps have the same degree.
\item If $M=\partial W$ with $W$ compact oriented and $f$ extends to $W$, then $\deg f=0$.
\item If $W$ is connected and $N=S^n$, the converse of (b) holds: a map $\partial W\to S^n$ of degree zero
extends to $W$.
\end{enumerate}
\end{theorem}

\begin{proof}[Proof of (a) and (b)]
Let $F\colon W\to N$ extend $f$, and let $y$ be a regular value of $F$ and of $f$ (\cref{thm:sard}). By
\cref{thm:neat}, $F^{-1}(y)$ is a compact oriented $1$-manifold with boundary $f^{-1}(y)$, and the boundary
orientation is the sign in the definition. The signed number of boundary points of a compact oriented
$1$-manifold is zero. Homotopy invariance is the case $W=M\times I$. Independence of $y$ follows because any
two points of a connected manifold are exchanged by a diffeomorphism isotopic to the identity
(\cref{ch:isotopy}).
\end{proof}

For example $z\mapsto z^k$ on the circle has degree $k$, and the antipodal map of $S^n$ has degree
$(-1)^{n+1}$ (\cref{exer:degzk}). The script computes the first both by counting preimages and as a
winding-number integral.

\begin{theorem}[Hopf]\label{thm:hopf}\Scl[Hopf; Hirsch 5.1.10]
Let $M$ be a closed connected $n$-manifold. If $M$ is oriented, $\deg\colon[M,S^n]\to\Z$ is a bijection. If
$M$ is not orientable, the mod $2$ degree $[M,S^n]\to\Z/2$ is a bijection.
\end{theorem}

\section{Hopf's theorem as a classification theorem}

There are two ways to see \cref{thm:hopf} as an instance of something general, and they are dual.

The first is geometric. For a closed manifold $M$ let $\mathcal F_k(M)$ be the set of closed submanifolds
$A\subset M$ of codimension $k$ equipped with a framing of the normal bundle $\nu(A,M)$. For any $k$, taking
the preimage of a regular value gives a bijection
\[[M,S^k]\ \cong\ \mathcal F_k(M)\,/\,\text{framed cobordism in }M\times I,\]
with inverse the collapse $M\to\Th(\nu(A,M))\cong\Sigma^kA_+\to S^k$ of \cref{ch:tubular}. This is the
Pontryagin--Thom construction inside $M$ \cite{Pon55}. For $k=n$ a framed submanifold is a finite set of points
with signs, and a framed cobordism cancels pairs of opposite sign; if $M$ is connected and oriented the
classes are counted by the total sign, and if $M$ is non-orientable a point can be carried around an
orientation-reversing loop and return with the opposite sign, so only the parity remains.

The second is homotopical. The map $S^n\to K(\Z,n)$ inducing an isomorphism on $\pi_n$ is $(n+1)$-connected,
so $[M,S^n]\to[M,K(\Z,n)]=H^n(M;\Z)$ is a bijection for $\dim M\le n$; and $H^n(M;\Z)$ is $\Z$ or $\Z/2$
according to orientability. The two readings are identified by Poincaré duality,
$H^n(M;\Z)\cong H_0(M;\Z^{w})$ with twisted coefficients, a class in the latter being a finite set of points
with local orientations.

\section{Intersection numbers}

\begin{theorem}\label{thm:intersection}\Scl[Hirsch 5.2.1]
Let $f\colon M\to N$ be a map of oriented manifolds and $A\subset N$ a closed oriented submanifold, with $M$
compact and $\dim M+\dim A=\dim N$. If $f\trans A$, put $\#(f,A)=\sum_{x\in f^{-1}(A)}\pm1$, the sign comparing
$df_x(T_xM)\oplus T_{f(x)}A$ with $T_{f(x)}N$. Then homotopic maps transverse to $A$ have the same
intersection number. Without orientations the same holds modulo $2$.
\end{theorem}

So $\#(f,A)$ is defined for every $f$, by perturbing. In the language of \cref{ch:transversality} no
perturbation is needed to define it:
\[\#(f,A)=\bigl[M\times^h_NA\bigr]\in\Omega_0=\Z,\]
the cobordism class of the derived fibre product, which is compact of virtual dimension zero
(\cref{thm:spivak}). Degree is the case $A=\{y\}$.

\begin{example}\label{ex:torus}
On the torus $\R^2/\Z^2$ the closed curve of slope class $(a,b)$ and the curve of class $(c,d)$, in general
position, meet in $|ad-bc|$ points all of the same sign, and the intersection number is $ad-bc$. The
script counts the intersection points for several pairs.
\end{example}

\section{The Euler number}

\begin{definition}\label{def:euler}
Let $\xi$ be an oriented vector bundle of rank $n$ over a closed oriented $n$-manifold $M$, with zero section
$Z$. Its \term{Euler number} is $\chi(\xi)=\#(Z,Z)$, the signed number of zeros of a section transverse to
$Z$. The Euler characteristic of $M$ is $\chi(M)=\chi(TM)$.
\end{definition}

\begin{theorem}\label{thm:euler}\Scl[Hopf; Hirsch 5.2.2, 5.2.3, 5.2.5, 5.2.10]
\begin{enumerate}[label=(\alph*)]
\item If $\xi$ has a nowhere-zero section then $\chi(\xi)=0$; if $M$ is connected the converse holds.
\item $\chi(\xi)=0$ if $n$ is odd.
\item $\chi(S^n)=1+(-1)^n$, and $\chi(M\times N)=\chi(M)\chi(N)$.
\item (Poincaré--Hopf) For a vector field on $M$ with isolated zeros, the sum of the indices is $\chi(M)$.
\end{enumerate}
\end{theorem}

The Euler number is the class of the derived self-intersection of the zero section, which by
\cref{ex:derived}(c) is the derived manifold with functions $\Gamma(\Lambda^\bullet\xi^\vee)$. Part (b) is
the observation that $s\mapsto-s$ is an automorphism of this object that multiplies its orientation by
$(-1)^n$. The converse in (a) is an instance of \cref{thm:degree}(c): the zeros of a section can be gathered
into a disc by an isotopy, and the section on the boundary sphere then has degree zero and extends without
zeros.

\section{Traces}

The second description starts from a definition that makes no mention of manifolds
(\cref{app:toolkit}). In a symmetric monoidal category an object $X$ is \term{dualizable} if there are
$X^\vee$, a coevaluation $\eta\colon\mathbf 1\to X\otimes X^\vee$ and an evaluation
$\epsilon\colon X^\vee\otimes X\to\mathbf 1$ satisfying the triangle identities. The \term{trace} of an
endomorphism $f$ of such an $X$ is the composite
\[\tr(f)\colon\ \mathbf 1\xrightarrow{\ \eta\ }X\otimes X^\vee\xrightarrow{\ f\otimes1\ }X\otimes X^\vee
\xrightarrow{\ \text{swap}\ }X^\vee\otimes X\xrightarrow{\ \epsilon\ }\mathbf 1 ,\]
an element of the commutative monoid $\operatorname{End}(\mathbf 1)$, and $\chi(X)=\tr(\id_X)$.

\begin{proposition}\label{prop:trace}\Scl[Dold--Puppe \cite{DP80}]
$\tr(fg)=\tr(gf)$; $\tr(f\otimes g)=\tr(f)\tr(g)$; and symmetric monoidal functors preserve dualizable
objects and traces. In finite-dimensional vector spaces $\tr$ is the trace of a matrix. In graded vector
spaces with the Koszul sign in the symmetry it is the alternating sum of the traces on the graded pieces.
\end{proposition}

\begin{theorem}[Lefschetz--Hopf, after Dold and Puppe]\label{thm:DP}\Scl[\cite{DP80}]
Let $M$ be a closed manifold and $f\colon M\to M$ a map. The trace of $\Sigma^\infty_+f$ in the stable homotopy
category is an integer, an element of $\pi_0\mathbb S=\Z$, and:
\begin{enumerate}[label=(\roman*)]
\item it equals the Lefschetz number $L(f)=\sum_i(-1)^i\tr\bigl(f_*\mid H_i(M;\mathbb Q)\bigr)$;
\item it equals the intersection number of the graph of $f$ with the diagonal in $M\times M$; if the fixed
points of $f$ are non-degenerate, this is $\sum_{f(x)=x}\operatorname{sign}\det(1-df_x)$.
\end{enumerate}
In particular $\chi(M)=\sum_i(-1)^i\dim H_i(M;\mathbb Q)=\#(\Delta,\Delta)=\chi(TM)$.
\end{theorem}

\begin{proof}[Sketch]
(i) Rational homology is a symmetric monoidal functor from finite spectra to graded vector spaces, by the
Künneth theorem; apply \cref{prop:trace}. (ii) By \cref{thm:atiyah} the coevaluation is the collapse onto a
tubular neighbourhood of $M\subset\R^{n+k}$ and the evaluation is the collapse onto a tubular neighbourhood of
the diagonal. The composite defining the trace is a map $S^{n+k}\to S^{n+k}$ whose preimage of the base point
of the last collapse is the set of fixed points of $f$, and the local degree at a non-degenerate fixed point
is the sign of $\det(1-df_x)$.
\end{proof}

So the equality of the two definitions of the Euler characteristic, which the source reaches at the end of
its treatment of Morse theory (Hirsch 6.3.8), is the statement that a trace can be computed either after
applying a monoidal functor or before. The multiplicativity $\chi(M\times N)=\chi(M)\chi(N)$ is
$\tr(f\otimes g)=\tr(f)\tr(g)$. Additivity over pushouts, $\chi(A\cup B)=\chi(A)+\chi(B)-\chi(A\cap B)$, is a
property of traces in stable settings \cite{May01}, and gives the formula
$\chi(M)=\sum(-1)^k\nu_k$ in terms of the numbers of critical points of a Morse function
(\cref{ch:morse}).

\begin{example}\label{ex:lefschetz}
A matrix $A\in\GL_2(\Z)$ acts on the torus. On $H_0,H_1,H_2$ its traces are $1$, $\tr A$, $\det A$, so
$L(A)=1-\tr A+\det A=\det(1-A)$. The fixed points are the solutions of $(A-1)x\in\Z^2$ modulo $\Z^2$;
there are $|\det(A-1)|$ of them and each has index $\operatorname{sign}\det(1-A)$. The script enumerates
them for several $A$.
\end{example}

\section{The diagonal, three ways}

The self-intersection of the diagonal has now appeared in three forms.

\begin{proposition}\label{prop:three}\Srf[Hirsch 5.2, 6.3.8; \cite{DP80}]
For a closed manifold $M$ and $f\colon M\to M$, the following integers are equal:
\begin{enumerate}[label=(\roman*)]
\item the signed number of fixed points of a generic perturbation of $f$;
\item the class in $\Omega_0$ of the derived fixed-point object
$M\times^h_{M\times M}M$, the fibre product of the graph of $f$ and the diagonal;
\item the trace of $\Sigma^\infty_+f$.
\end{enumerate}
For $f=\id$ the object in (ii) has functions the differential forms $\Gamma(\Lambda^\bullet T^*M)$ in
homological degrees, and the integer is $\chi(M)$.
\end{proposition}

Form (i) is the definition in the source; it requires a choice and a proof of independence. Form (ii)
requires no choice but lives in $\DMfd$. Form (iii) requires no manifold: it makes sense for any dualizable
object, and it is the one that generalizes, to parametrized families, to equivariant settings, and to
bicategories, where the Lefschetz number acquires a refinement due to Reidemeister \cite{PS14}.

The derived fixed-point object of the identity is the smooth counterpart of the derived loop space of
algebraic geometry. That its functions are differential forms is the Hochschild--Kostant--Rosenberg
theorem in this setting, and the Euler characteristic is then the ``dimension'' of $M$ in the sense in
which the dimension of a vector space is the trace of its identity map.

\begin{example}[Linking numbers]\label{ex:linking}
For disjoint oriented closed curves $K,L\subset\R^3$ the linking number is the degree of the Gauss map
$K\times L\to S^2$, $(x,y)\mapsto(y-x)/|y-x|$. It equals the intersection number of $L$ with any oriented
surface bounded by $K$, and it is given by Gauss's integral. For the Hopf link it is $\pm1$, and the
script evaluates the integral numerically.
\end{example}

\section*{Notes}

The treatment of degree by regular values and one-dimensional cobordisms is Pontryagin's and is the subject
of Milnor's \emph{Topology from the Differentiable Viewpoint}. The trace in a symmetric monoidal category
and its application to fixed point theory are due to Dold and Puppe; see also \cite{May01} for additivity
and \cite{PS14} for an exposition and the bicategorical refinement. Derived loop spaces and their relation
to differential forms are due to Ben-Zvi and Nadler and to Toën and Vezzosi in the algebraic setting.

\section*{Exercises}

\begin{exercise}\label{exer:degzk}
Show that $z\mapsto z^k$ has degree $k$ on $S^1$ and that the antipodal map of $S^n$ has degree $(-1)^{n+1}$.
Deduce that every vector field on $S^{2m}$ has a zero.
\end{exercise}

\begin{exercise}\label{exer:lefschetz}
Verify \cref{ex:lefschetz}: for $A\in\GL_2(\Z)$ with $\det(A-1)\ne0$, the induced map of the torus has
exactly $|\det(A-1)|$ fixed points, each of index $\operatorname{sign}\det(1-A)$.
\end{exercise}

\begin{exercise}\label{exer:trace}
Prove $\tr(fg)=\tr(gf)$ and $\tr(f\otimes g)=\tr(f)\tr(g)$ from the definition, and show that in graded vector
spaces with the Koszul sign rule the trace of a degree-preserving endomorphism is the alternating sum of the
traces of its components.
\end{exercise}

\begin{exercise}\label{exer:chiSn}
Compute $\chi(S^n)$ from a vector field with two zeros, and again as $\sum(-1)^i\dim H_i$.
\end{exercise}

\begin{exercise}\label{exer:linking}
Let $K$ be the unit circle in the $xy$-plane and $L$ the circle of radius $1$ centred at $(1,0,0)$ in the
$xz$-plane. Compute the linking number as the intersection number of $L$ with the disc bounded by $K$.
\end{exercise}
\part{Decompositions and Moduli}
\chapter{Morse Theory}\label{ch:morse}

A Morse function on a manifold is a generic point of the mapping object $[M,\R]$, and Hirsch's sixth chapter
shows how much of the manifold such a point determines: a decomposition into handles, a cell structure,
inequalities for the Betti numbers, the Euler characteristic. In categorical terms a Morse function presents
the homotopy type of $M$ as an iterated pushout, and presents $M$ itself as a composite of elementary
cobordisms. This chapter states the classical theorems in that form, and then turns to what the source does
not treat: the structure on the set of all Morse functions, which is where the presentations become
canonical.

\section{Morse functions}

\begin{definition}\label{def:morse}
A critical point $p$ of $f\colon M\to\R$ is \term{non-degenerate} if the Hessian of $f$ at $p$ is a
non-degenerate quadratic form; its \term{index} is the number of negative eigenvalues. $f$ is a \term{Morse
function} if all its critical points are non-degenerate. Its \term{type} is $(\nu_0,\dots,\nu_n)$, with
$\nu_k$ the number of critical points of index $k$.
\end{definition}

\begin{theorem}\label{thm:morselemma}\Scl[Morse; Hirsch 6.1.1, 6.1.2]
Near a non-degenerate critical point of index $k$ there are coordinates in which
$f=f(p)-x_1^2-\dots-x_k^2+x_{k+1}^2+\dots+x_n^2$. Morse functions form an open dense subset of
$C^\infty_S(M,\R)$.
\end{theorem}

Density was derived in \cref{cor:counts}(c) from jet transversality: $f$ is Morse when $df$ is transverse
to the zero section of $T^*M$.

\begin{example}\label{ex:torusheight}
On the torus with coordinates $(\theta,\varphi)$ let $f=(R+r\cos\varphi)\cos\theta$, $R>r>0$, the height
function of a torus standing on its edge. The critical points are the four points with
$\theta,\varphi\in\{0,\pi\}$, and the Hessian there is diagonal with entries
$-(R+r\cos\varphi)\cos\theta$ and $-r\cos\varphi\cos\theta$. The indices are $2,1,1,0$ at
$(0,0),(0,\pi),(\pi,\pi),(\pi,0)$, with critical values $R+r,\ R-r,\ -(R-r),\ -(R+r)$. The type is $(1,2,1)$.
\end{example}

\section{Regular intervals}

\begin{theorem}[Regular interval]\label{thm:regint}\Scl[Hirsch 6.2.2]
Let $f\colon M\to[a,b]$ be smooth on a compact manifold with boundary, with no critical points and
$f(\partial M)=\{a,b\}$. Then there is a diffeomorphism $M\cong f^{-1}(a)\times[a,b]$ commuting with the
maps to $[a,b]$.
\end{theorem}

The proof integrates the vector field $\grad f/|\grad f|^2$, along which $f$ increases at unit speed. The
theorem is the one-parameter case of a statement about families.

\begin{theorem}[Ehresmann]\label{thm:ehresmann}\Scl
A proper submersion $p\colon E\to B$ is a locally trivial fibre bundle.
\end{theorem}

In the language of \cref{ch:bundles}: let $\mathcal M_d$ be the stack on $\Cart$ whose sections over $U$ are
the proper submersions over $U$ with $d$-dimensional fibres. Ehresmann's theorem says that $\mathcal M_d$ is
locally constant in the strongest sense: every section over a Euclidean space is a product. Its shape is
$\coprod_{[F]}B\Diff(F)$, and a proper submersion over $B$ is classified by a map from $B$ to this space
(\cref{tab:oka}). Compact smooth manifolds have no local moduli, only monodromy. The regular interval theorem
is the case of a contractible base.

\section{Passing a critical point}

Call a Morse function $f\colon M\to[a,b]$ on a compact manifold \term{admissible} if
$\partial M=f^{-1}(a)\cup f^{-1}(b)$ and $a,b$ are regular values.

\begin{theorem}[Attaching a cell]\label{thm:handle}\Scl[Hirsch 6.3.1--6.3.3]
Let $f\colon M\to[a,b]$ be an admissible Morse function with exactly one critical point, of index $k$. Then
there is a $k$-cell $e^k\subset M\setminus f^{-1}(b)$, attached to $f^{-1}(a)$ along its boundary, such that
$f^{-1}(a)\cup e^k$ is a deformation retract of $M$. Smoothly, $M$ is obtained from
$f^{-1}(a)\times[0,1]$ by attaching a handle $D^k\times D^{n-k}$ along an embedding of
$S^{k-1}\times D^{n-k}$ in $f^{-1}(a)\times\{1\}$.
\end{theorem}

For a proper function $f\colon M\to\R$ bounded below, write $M^c=f^{-1}(-\infty,c]$. The theorem and
\cref{thm:regint} give the following form, which is the one that composes.

\begin{corollary}\label{cor:pushout}\Srf[Hirsch 6.2.2, 6.3.1]
If $[c,c']$ contains no critical value, $M^c\to M^{c'}$ is a homotopy equivalence. If $[c-\varepsilon,
c+\varepsilon]$ contains a single critical value, attained at one critical point of index $k$, there is a
homotopy pushout square of spaces
\[\begin{tikzcd}
S^{k-1}\arrow[r]\arrow[d] & M^{c-\varepsilon}\arrow[d]\\
D^k\arrow[r] & M^{c+\varepsilon}
\end{tikzcd}\]
and hence a cofibre sequence $\Sigma^\infty_+M^{c-\varepsilon}\to\Sigma^\infty_+M^{c+\varepsilon}\to\mathbb
S^k$ of spectra.
\end{corollary}

The local statement behind the theorem is about the quadratic form $q(x,y)=-|x|^2+|y|^2$ on
$\R^k\times\R^{n-k}$: the set $\{q\le-\varepsilon\}$ together with the disc $\{y=0,|x|^2\le\varepsilon\}$
is a deformation retract of $\{q\le\varepsilon\}$, by a deformation that shrinks $y$ and does not move $x$
(\cref{exer:model}).

\begin{theorem}\label{thm:cw}\Scl[Hirsch 6.4.1--6.4.3, 6.2.4]
A compact manifold with a Morse function of type $(\nu_0,\dots,\nu_n)$ has the homotopy type of a finite CW
complex with $\nu_k$ cells of dimension $k$. Every $n$-manifold has the homotopy type of a CW complex of
dimension at most $n$. A closed manifold with a Morse function having exactly two critical points is
homeomorphic to a sphere.
\end{theorem}

So the shape of a compact manifold is a finite space, a compact object of $\Spc$, and its suspension
spectrum is a finite spectrum. Finite spectra are dualizable; this is the abstract half of
\cref{thm:atiyah}, the concrete half being the identification of the dual.

\section{Inequalities}

\begin{theorem}[Morse inequalities]\label{thm:morseineq}\Scl[Morse; Hirsch 6.3.4--6.3.8]
Let $f$ be a Morse function of type $(\nu_k)$ on a closed manifold $M$, and $\beta_k=\dim H_k(M;F)$ for a field
$F$. Then
\[\sum_k\nu_kt^k-\sum_k\beta_kt^k=(1+t)\,Q(t)\]
for a polynomial $Q$ with non-negative integer coefficients. In particular $\nu_k\ge\beta_k$, and
$\chi(M)=\sum(-1)^k\nu_k=\sum(-1)^k\beta_k$; and $\chi(M)=0$ if $\dim M$ is odd.
\end{theorem}

\begin{proof}
By \cref{cor:pushout}, crossing a critical point of index $k$ changes the homology of the sublevel set in one
of two ways: either $\beta_k$ increases by one, or $\beta_{k-1}$ decreases by one. Let $q_{k-1}$ be the
number of critical points of index $k$ of the second kind. Each of them kills a class created earlier by a
critical point of index $k-1$ of the first kind. So $\nu_k=\beta_k+q_k+q_{k-1}$, which is the identity with
$Q=\sum q_kt^k$. Setting $t=-1$ gives the Euler characteristic, and replacing $f$ by $-f$ exchanges $\nu_k$
and $\nu_{n-k}$.
\end{proof}

The equality $\chi(TM)=\sum(-1)^k\nu_k$ also follows from \cref{thm:euler}(d) applied to the gradient, whose
zero at a critical point of index $k$ has index $(-1)^k$. For the torus of \cref{ex:torusheight},
$1+2t+t^2$ is the Poincaré polynomial and $Q=0$. On $\RP^n$ there is a Morse function with one critical
point of each index, which is optimal for $F=\Z/2$ and not for $F=\mathbb Q$ (\cref{exer:rpn}).

\section{The sublevel functor}

The proof just given used more than the statement records. The assignment $c\mapsto M^c$ is a functor from
the ordered set $\R$ to manifolds with boundary and embeddings, and its composite with the shape,
\[S_f\colon(\R,\le)\to\Spc,\qquad c\mapsto\Pi(M^c),\]
is the \term{sublevel functor} of $f$.

\begin{proposition}\label{prop:persist}\Srf[\cref{cor:pushout}; persistence, \cite{ZC05}]
For a Morse function $f$ on a closed manifold, $S_f$ is locally constant on the complement of the critical
values, and at a critical value it changes by a pushout along a sphere for each critical point. The functor
$H_k(S_f;F)\colon(\R,\le)\to\cat{Vect}_F$ is a direct sum of interval modules; the number of infinite
intervals is $\beta_k$, and the number of finite intervals is the coefficient $q_k$ of $Q$ in
\cref{thm:morseineq}.
\end{proposition}

So the polynomial $Q$ counts pairs of critical points that cancel in homology, and the multiset of intervals,
the \term{barcode}, refines the Morse inequalities to a statement about which pairs and at which levels.
The barcode is a continuous function of $f$ in the uniform norm, which the type is not. The script
computes the degree-zero barcode of a function on a circle by a union--find pass and checks that the number
of finite intervals is $\nu_0-1$.

\section{Flow categories}

A Morse function with a generic metric determines more than a filtration.

\begin{theorem}\label{thm:CJS}\Sli[Cohen--Jones--Segal \cite{CJS95}]
Let $f$ be a Morse function on a closed manifold $M$ with a Morse--Smale gradient. There is a topological
category $\mathcal C_f$ whose objects are the critical points and whose morphism space from $p$ to $q$ is the
compactified space of gradient flow lines from $p$ to $q$, a manifold with corners of dimension
$\operatorname{ind}p-\operatorname{ind}q-1$, with composition the inclusion of broken flow lines. Its
classifying space is homeomorphic to $M$.
\end{theorem}

The Morse complex is what remains of $\mathcal C_f$ when one keeps only the zero-dimensional morphism spaces
and counts their points; the higher-dimensional ones carry the attaching maps of the cells. The transversality
condition on the metric is \cref{thm:transversality} applied to the stable and unstable manifolds, and the
corner structure is where manifolds with corners (\cref{ch:manifolds}) become unavoidable.

\section{Families of functions}

The set of Morse functions is open and dense in $[M,\R]$ but it is not a good parameter space: two Morse
functions of different types cannot be joined by a path of Morse functions. Jet transversality in one
parameter says what a generic path looks like.

\begin{theorem}[Cerf]\label{thm:cerf}\Scl[Cerf \cite{Cer70}]
Any two Morse functions on a closed manifold are joined by a path $f_t$ that is Morse except at finitely
many times, at each of which exactly one of the following occurs: two critical points have the same value;
or there is a single degenerate critical point, with local form
$x_1^3-(t-t_0)x_1-x_2^2-\dots-x_{k+1}^2+x_{k+2}^2+\dots+x_n^2$, at which a pair of critical points of indices
$k$ and $k+1$ is born or dies.
\end{theorem}

The count behind the theorem: in the space of $2$-jets, the jets of degenerate critical points have
codimension $n+1$, so a one-parameter family, of total dimension $n+1$, meets them at isolated points. On the
side of handle decompositions the two events are the reordering or sliding of handles and the cancellation
of a pair. So the theorem says that any two handle decompositions of a manifold are related by these moves.
It is a presentation by generators and relations, one categorical level up from \cref{thm:handle}.

The space of functions with only Morse and birth--death singularities is connected by Cerf's theorem,
but it is still not contractible. Igusa found the decoration that makes it so.

\begin{theorem}\label{thm:framed}\Sli[Igusa \cite{Igu87}; Lurie \cite{Lur09}; Eliashberg--Mishachev \cite{EM12}]
The space of framed functions on a compact manifold $M$ (functions with only Morse and birth--death
singularities, with a framing of the negative eigenspace of the Hessian at each critical point) is
contractible.
\end{theorem}

Igusa proved that the space is $(\dim M-1)$-connected; contractibility is due to Lurie and, by geometric
methods, to Eliashberg and Mishachev. By \cref{prin:choices} the theorem says that a framed function is a
choice of the harmless kind, so that the handle decomposition it induces is as good as canonical. That is the
form in which Morse theory enters the proof of the cobordism hypothesis (\cref{ch:cobordism}).

\section*{Notes}

Milnor's \emph{Morse Theory} and \emph{Lectures on the h-Cobordism Theorem} are the classical accounts;
Hirsch's chapter follows the same lines with more attention to the differentiability class. Persistence
modules and their decomposition are in \cite{ZC05} and in work of Crawley-Boevey. The flow category is the
starting point of Floer homotopy theory. Cerf's theorem is the one-parameter beginning of a hierarchy governed by the classification of
singularities of functions;
Igusa's work on higher parameters and the notion of framed function are from the 1980s.

\section*{Exercises}

\begin{exercise}\label{exer:torus}
Verify \cref{ex:torusheight}.
\end{exercise}

\begin{exercise}\label{exer:morsepoly}
Deduce from the polynomial identity of \cref{thm:morseineq} the inequalities
$\nu_k-\nu_{k-1}+\dots\pm\nu_0\ge\beta_k-\beta_{k-1}+\dots\pm\beta_0$ for every $k$. Show that a Morse function
on a closed connected surface of genus $g$ has at least $2g+2$ critical points.
\end{exercise}

\begin{exercise}\label{exer:rpn}
Let $\lambda_0<\dots<\lambda_n$ and $f([x])=\sum\lambda_ix_i^2/\sum x_i^2$ on $\RP^n$. Show that $f$ is a
Morse function with one critical point of each index $0,\dots,n$. Find $Q$ for $F=\Z/2$ and for
$F=\mathbb Q$.
\end{exercise}

\begin{exercise}\label{exer:model}
For $q(x,y)=-|x|^2+|y|^2$ on $\R^k\times\R^{n-k}$ and $\varepsilon>0$, construct a deformation retraction of
$\{q\le\varepsilon\}$ onto $\{q\le-\varepsilon\}\cup\{y=0,\ |x|^2\le\varepsilon\}$.
\end{exercise}

\begin{exercise}\label{exer:birth}
For $f_t(x)=x^3-tx$ on $\R$, find the critical points and their indices for each $t$, and describe the
sublevel functor of $f_t$ restricted to a large interval $[-L,L]$ before and after $t=0$.
\end{exercise}
\chapter{Cobordism}\label{ch:cobordism}

In the source, cobordism occupies a short chapter: the groups are defined, the Pontryagin--Thom construction
is carried out, and Thom's theorem identifying cobordism groups with homotopy groups of Thom spaces is proved
in the range the earlier chapters allow. It is the place where the three tools of the book (classifying maps,
tubular neighbourhoods, transversality) are used together. It is also the place from which the
higher-categorical view of manifolds grew. This chapter states the classical results and then the two
modern ones that extend them: the homotopy type of the cobordism category, and the cobordism hypothesis.

\section{Cobordism groups}

\begin{definition}\label{def:cob}
Closed $n$-manifolds $M_0,M_1$ are \term{cobordant} if there is a compact $(n+1)$-manifold $W$ with
$\partial W\cong M_0\sqcup M_1$. The set of classes is $\mathfrak N_n$. For oriented manifolds one requires
$W$ oriented with $\partial W\cong M_1\sqcup-M_0$, and the set of classes is $\Omega_n$.
\end{definition}

\begin{theorem}\label{thm:cobring}\Scl[Thom; Hirsch 7.1.1]
Disjoint union makes $\mathfrak N_n$ and $\Omega_n$ abelian groups, and product makes
$\mathfrak N_*$ and $\Omega_*$ graded rings, commutative in the graded sense. Every element of
$\mathfrak N_*$ has order $2$.
\end{theorem}

Transitivity of the relation uses \cref{thm:gluing}; the inverse of $[M]$ is $[-M]$, with $M\times I$ as the
null-cobordism of $M\sqcup-M$. In low dimensions $\mathfrak N_0=\Z/2$, $\Omega_0=\Z$,
$\mathfrak N_1=\Omega_1=0$, and $\mathfrak N_2=\Z/2$, $\Omega_2=0$ by the classification of surfaces
(\cref{ch:surfaces}).

\section{The Pontryagin--Thom theorem}

Let $M\subset\R^{n+k}$ be a closed $n$-manifold. The collapse of \cref{ch:tubular}, followed by the map of
Thom spaces induced by the classifying map of the normal bundle (\cref{ch:bundles}), is a map
\[S^{n+k}\longrightarrow\Th(\nu)\longrightarrow\Th(\gamma_k),\]
where $\gamma_k$ is the tautological bundle over $\Gr_k(\R^s)$ for $s$ large.

\begin{theorem}[Pontryagin--Thom]\label{thm:PT}\Scl[Pontryagin \cite{Pon55}; Thom \cite{Tho54}; Hirsch 7.2.3]
For $f\colon S^{n+k}\to\Th(\gamma_k)$ transverse to the zero section, $f\mapsto[f^{-1}(\Gr_k(\R^s))]$ defines a
homomorphism $\pi_{n+k}\bigl(\Th(\gamma_k\to\Gr_k(\R^s))\bigr)\to\mathfrak N_n$ that is surjective if $k>n$ and
$s\ge k+n$ and bijective if $k>n+1$ and $s\ge k+n+1$; the construction above is its inverse. Stably,
\[\mathfrak N_n\cong\pi_n(MO),\qquad\Omega_n\cong\pi_n(MSO),\qquad\Omega^{\mathrm{fr}}_n\cong\pi_n(\mathbb S),\]
and in general $\Omega^\theta_n\cong\pi_n(M\theta)$ for a stable tangential structure $\theta\colon B\to B\Or$,
where $M\theta$ is the Thom spectrum of $\theta$.
\end{theorem}

\begin{proof}[Structure of the proof]
Well-definedness: the embedding and the tubular neighbourhood are contractible choices
(\cref{thm:embcontract,thm:tubcontract}), and a cobordism embedded in $\R^{n+k}\times I$ collapses to a
homotopy. Inverse: a map $f\colon S^{n+k}\to\Th(\gamma_k)$ can be made transverse to the zero section
(\cref{thm:transversality}), and $M=f^{-1}(\Gr_k)$ is a closed $n$-manifold with normal bundle pulled back from
$\gamma_k$; a homotopy made transverse gives a cobordism (\cref{lem:fibre}). The dimension restrictions come
from \cref{thm:whitney,thm:classify}.
\end{proof}

The theorem is \cref{prin:generic} for the cospan $S^{n+k}\to\Th(\gamma_k)\leftarrow\Gr_k$: every map to a Thom
space has a derived preimage of the zero section, which has a cobordism class, and the class determines the
map up to homotopy.

\section{Thom's computation}

\begin{theorem}[Thom]\label{thm:thom}\Scl[\cite{Tho54}]
$MO$ is a wedge of suspensions of the Eilenberg--MacLane spectrum $H\Z/2$, and
\[\mathfrak N_*\cong\Z/2\,[x_i:\ i\ge2,\ i\ne2^j-1],\qquad|x_i|=i,\]
with $x_{2i}$ the class of $\RP^{2i}$. A closed manifold is a boundary if and only if all its Stiefel--Whitney
numbers vanish.
\end{theorem}

So $\dim_{\Z/2}\mathfrak N_n$ is the number of partitions of $n$ into parts not of the form $2^j-1$: for
$n=0,\dots,10$ these are $1,0,1,0,2,1,3,1,5,3,8$. For a partition $n=i_1+\dots+i_r$ the
\term{Stiefel--Whitney number} is $\langle w_{i_1}\cdots w_{i_r}(TM),[M]\rangle\in\Z/2$. It vanishes on a
boundary $M=\partial W$ because $TW|_M=TM\oplus\R$, so the class being evaluated is restricted from $W$, and
the fundamental class of a boundary maps to zero in the homology of $W$.

\begin{example}\label{ex:sw}
The total Stiefel--Whitney class of $\RP^n$ is $(1+a)^{n+1}$ with $a$ the generator of $H^1$. For $\RP^2$,
$w=1+a+a^2$, and $w_2[\RP^2]=w_1^2[\RP^2]=1$: not a boundary. For $\RP^3$, $w=(1+a)^4=1$: all numbers vanish
and $\RP^3$ bounds. In dimension $4$ the numbers $(w_1^4,w_1^2w_2,w_2^2,w_1w_3,w_4)$ are $(1,0,0,0,1)$ for
$\RP^4$ and $(0,0,1,0,1)$ for $\RP^2\times\RP^2$. The two manifolds are independent in
$\mathfrak N_4=(\Z/2)^2$; they are $x_4$ and $x_2^2$. The script computes these numbers and the ranks above.
\end{example}

\begin{theorem}\label{thm:oriented}\Scl[Thom; Hirsch 7.2.4]
$\Omega_n$ is finite if $n$ is not divisible by $4$, and $\Omega_{4k}$ has rank equal to the number of
partitions of $k$. Rationally, $\Omega_*\otimes\mathbb Q=\mathbb Q[\CP^2,\CP^4,\dots]$.
\end{theorem}

\section{Thom spectra as colimits}

The spectra $MO$, $MSO$, $M\theta$ were defined by Thom spaces of bundles over Grassmannians. There is a
description in which no bundle appears.

\begin{theorem}\label{thm:ABGHR}\Sli[Ando--Blumberg--Gepner--Hopkins--Rezk \cite{ABGHR14}]
Let $\operatorname{Pic}(\mathbb S)$ be the $\infty$-groupoid of invertible spectra. A stable vector bundle on a
space $X$ determines a map $X\to B\Or\xrightarrow{J}\operatorname{Pic}(\mathbb S)$, sending a point to the
one-point compactification of the fibre, and its Thom spectrum is the colimit of the composite functor
$X\to\operatorname{Pic}(\mathbb S)\subset\Sp$. In particular $MO=\colim(B\Or\to\Sp)$ is the homotopy
quotient of the sphere spectrum by the stable orthogonal group, and it is an $E_\infty$-ring because $J$ is
an infinite loop map. For an $E_\infty$-ring $R$, $E_\infty$-ring maps $MO\to R$ correspond to
null-homotopies of $B\Or\to B\GL_1(R)$ as infinite loop maps.
\end{theorem}

For a trivial bundle the diagram is constant and the colimit is $\Sigma^\infty_+X$, which is
\cref{prop:thom}. The description makes the Thom isomorphism a formal consequence of an orientation, a
trivialization of the diagram after smashing with $R$. And it gives a reading of \cref{thm:PT}: framed
cobordism is the sphere spectrum, and unoriented cobordism is what is left when one divides by all changes
of framing.

\section{Cobordism categories}

The groups $\mathfrak N_n$ remember that a cobordism exists. The cobordism category remembers the
cobordisms.

\begin{definition}\label{def:cobcat}
$\Cob_d$ is the $\infty$-category whose objects are closed $(d-1)$-manifolds and whose space of morphisms
from $M_0$ to $M_1$ is $\coprod_{[W]}B\Diff(W\text{ rel }\partial W)$, the union over diffeomorphism classes
of cobordisms. Composition is gluing (\cref{thm:gluing}). Disjoint union makes it symmetric monoidal. With a
tangential structure $\theta\colon B\to B\Or(d)$ on the cobordisms one writes $\Cob^\theta_d$.
\end{definition}

A concrete model takes objects to be closed submanifolds of $\R^\infty$ and morphisms to be submanifolds of
$[a_0,a_1]\times\R^\infty$ that are products near the ends; by \cref{thm:embcontract} the embeddings add
nothing, and by \cref{thm:collar} neither do the product conditions. The set of components of the
classifying space $B\Cob_d$ is the set of objects modulo the relation generated by morphisms, which is
$\mathfrak N_{d-1}$.

\begin{theorem}[Galatius--Madsen--Tillmann--Weiss]\label{thm:GMTW}\Sli[\cite{GMTW09}]
Let $MTO(d)$ be the Thom spectrum of the virtual bundle $-\gamma_d$ over $B\Or(d)$. Then
\[B\Cob_d\ \simeq\ \Omega^{\infty-1}MTO(d),\]
and likewise $B\Cob^\theta_d\simeq\Omega^{\infty-1}MT\theta(d)$ for every tangential structure $\theta$.
\end{theorem}

On $\pi_0$ this is a form of \cref{thm:PT}. The proof is an application of the concordance classification
(\cref{thm:concordance}): the cobordism category is the shape of a sheaf whose sections over $U$ are
families, parametrized by $U$, of $d$-dimensional submanifolds of $\R\times\R^\infty$ that are proper over
$\R$, and the shape of that sheaf is computed by a parametrized Pontryagin--Thom construction. So the
theorem is the Pontryagin--Thom theorem for the whole category of cobordisms and not only for its set of
components. For oriented surfaces, combined with Harer's stability theorem, it gives the Madsen--Weiss
theorem (\cref{thm:MW}).

\section{Field theories and the cobordism hypothesis}

\begin{definition}\label{def:tft}
A $d$-dimensional \term{topological field theory} with values in a symmetric monoidal $\infty$-category
$\mathcal C$ is a symmetric monoidal functor $Z\colon\Cob_d\to\mathcal C$.
\end{definition}

An invertible theory, one whose values are invertible objects and morphisms, factors through the
groupoid completion of $\Cob_d$, so by \cref{thm:GMTW} invertible theories are classified by maps out of
the spectrum $\Sigma MTO(d)$. For general theories one needs to know $\Cob_d$ itself, by generators and
relations. Morse theory provides them: by \cref{thm:handle} every cobordism is a composite of handle
attachments, and by \cref{thm:cerf,thm:framed} the relations among such composites are known and the
choices involved are contractible. The cleanest statement is obtained after extending the category
downwards, allowing manifolds with corners of all codimensions, to a symmetric monoidal
$(\infty,d)$-category $\Bord_d$.

\begin{theorem}[Cobordism hypothesis]\label{thm:CH}
\Sli[conjectured by Baez--Dolan \cite{BD95}; Lurie \cite{Lur09}; see also \cite{GP21}]
Let $\mathcal C$ be a symmetric monoidal $(\infty,d)$-category. Evaluation at the point is an equivalence
between symmetric monoidal functors $\Bord^{\mathrm{fr}}_d\to\mathcal C$, from the framed bordism category, and
the $\infty$-groupoid of fully dualizable objects of $\mathcal C$.
\end{theorem}

Lurie's paper gives a detailed outline of a proof; the case $d=2$ is worked out by generators and relations
in \cite{SP09}, and a proof in a geometric formulation has been given by Grady and Pavlov. For $d=1$ the
statement is that the oriented one-dimensional cobordism category is the free symmetric monoidal
$\infty$-category on a dualizable object. The value of the theory on the circle is then the trace of the
identity, $\chi(X)$ in the sense of \cref{ch:degree}; for $\mathcal C=\Sp$ and $X=\Sigma^\infty_+M$ it is the
Euler characteristic of $M$. The Euler characteristic of a closed manifold is the partition function, on
the circle, of the one-dimensional field theory it generates.

\section{Derived cobordism}

By \cref{thm:spivak} the cobordism ring of compact derived manifolds is $\mathfrak N_*$ again. The two
enlargements of this chapter are therefore complementary. Passing from the groups to the category
$\Cob_d$ adds information: the homotopy types of diffeomorphism groups. Passing from manifolds to derived
manifolds adds none at the level of cobordism classes, and that is its use: fibre products, fixed-point
sets and zero loci have classes without being put in general position first.

\section*{Notes}

Stong's \emph{Notes on Cobordism Theory} is the standard reference for the classical computations. The
identification of Thom spectra with colimits is from \cite{ABGHR14}. Cobordism categories as topological
categories go back to Segal's definition of conformal field theory and to Tillmann; \cite{GMTW09} determined
their homotopy type. The cobordism hypothesis was formulated in \cite{BD95}; besides \cite{Lur09}, see
Ayala and Francis for an approach through factorization homology.

\section*{Exercises}

\begin{exercise}\label{exer:chimod2}
Let $W$ be a compact odd-dimensional manifold with boundary. Show that $\chi(\partial W)=2\chi(W)$. Deduce
that the Euler characteristic modulo $2$ is a ring homomorphism $\mathfrak N_*\to\Z/2$, and that $\RP^{2m}$ is
not a boundary.
\end{exercise}

\begin{exercise}\label{exer:ranks}
Compute $\dim\mathfrak N_n$ for $n\le10$ from \cref{thm:thom}.
\end{exercise}

\begin{exercise}\label{exer:sw}
Verify the Stiefel--Whitney numbers in \cref{ex:sw}.
\end{exercise}

\begin{exercise}\label{exer:lowdim}
Show that $\mathfrak N_1=0$ and that the Klein bottle is a boundary. Using the classification of surfaces,
conclude that $\mathfrak N_2=\Z/2$, generated by $\RP^2$.
\end{exercise}

\begin{exercise}\label{exer:framed0}
Show directly that the Pontryagin--Thom collapse of a single point of $\R^k$ with the standard framing is a
map $S^k\to S^k$ of degree $1$, and that reversing one framing vector gives degree $-1$. Conclude
$\Omega^{\mathrm{fr}}_0\cong\Z$.
\end{exercise}
\chapter{Isotopy}\label{ch:isotopy}

Hirsch's eighth chapter is about paths in spaces of embeddings: an isotopy of a compact submanifold extends
to an isotopy of the ambient manifold; embeddings of discs are unique up to isotopy; manifolds glued by
isotopic maps are diffeomorphic. Each statement is about $\pi_0$ or about lifting paths. Each has a form
that concerns the whole homotopy type of a space of embeddings, and those forms are where the subject went
after 1976. This chapter states them and ends with the functor calculus that organizes spaces of
embeddings in high codimension. The companion volume uses this chapter more than any other.

\section{Isotopies and vector fields}

\begin{definition}\label{def:isotopy}
An \term{isotopy} of embeddings of $V$ in $M$ is a plot $I\to\Emb(V,M)$: a smooth map
$F\colon V\times I\to M$ with every $F_t$ an embedding. A \term{diffeotopy} of $M$ is an isotopy
$F\colon M\times I\to M$ of diffeomorphisms with $F_0=\id$. Embeddings are \term{isotopic} if joined by an
isotopy, and \term{ambient isotopic} if one is carried to the other by the end of a diffeotopy.
\end{definition}

So isotopy is concordance of points of the smooth set $\Emb(V,M)$ in the sense of \cref{def:shape}. (The
word concordance is used differently in the classical literature on embeddings, for embeddings of
$V\times I$ in $M\times I$ that need not preserve levels; we avoid that usage.)

\begin{theorem}\label{thm:flow}\Scl[Hirsch 8.1.1]
A time-dependent vector field $G$ on $M$ of bounded velocity, for instance one with compact support,
generates a unique diffeotopy $F$ with $\partial_tF_t(x)=G(F_t(x),t)$. Every diffeotopy with compact support
arises in this way from its velocity field.
\end{theorem}

So paths in $\Diff_c(M)$ starting at the identity are the same thing as paths in the vector space
$\mathfrak X_c(M)$ of compactly supported vector fields, which is the statement that $\Diff_c(M)$ is a Lie
group with that Lie algebra \cite{KM97}. The hypothesis on the velocity cannot be dropped: the field
$x^2\partial_x$ on $\R$ has solutions $x_0/(1-x_0t)$, which leave every compact set in finite time.

\section{Isotopy extension}

\begin{theorem}[Isotopy extension]\label{thm:isoext}\Scl[Thom, Palais, Cerf; Hirsch 8.1.3]
Let $V\subset M$ be a compact submanifold and $F\colon V\times I\to M$ an isotopy with image in
$M\setminus\partial M$. Then $F$ extends to a diffeotopy of $M$ with compact support.
\end{theorem}

The proof extends the velocity field of $F$, which is defined along the track of the isotopy in
$M\times I$, to a compactly supported field on $M\times I$ using a tubular neighbourhood of the track, and
integrates. The extension is a contractible choice. Carrying parameters through the argument gives the
stronger statement.

\begin{theorem}[Restriction is a fibration]\label{thm:palais}\Scl[Palais \cite{Pal60}; Cerf; Lima]
Let $V\subset M$ be a compact submanifold. The restriction map
$\Diff_c(M)\to\Emb(V,M)$ is a locally trivial fibre bundle with fibre the group $\Diff_c(M\text{ rel }V)$ of
diffeomorphisms fixing $V$. More generally, for compact $V\subset V'$ the restriction
$\Emb(V',M)\to\Emb(V,M)$ is a fibration.
\end{theorem}

\Cref{thm:isoext} is the path-lifting property of this map; \cref{thm:palais} says that all homotopies
lift. The consequence is a long exact sequence of homotopy groups. For $V$ a point it reads
\[\cdots\to\pi_1\Diff(M)\to\pi_1(M)\to\pi_0\Diff(M\text{ rel }x)\to\pi_0\Diff(M)\to\pi_0(M),\]
and the middle map sends a loop to the diffeomorphism that pushes the base point around it. For compact
$V$ the quotient of $\Emb(V,M)$ by $\Diff(V)$ is the space of submanifolds of $M$ diffeomorphic to $V$, and the
theorem says that the group $\Diff_c(M)$ acts on it with orbits that are unions of components.

Compactness is essential. Let $F_t\colon\R\to\R^3$ be a long knot whose knotted part sits at distance
$1/(1-t)$ along the line, for $t<1$, and let $F_1$ be the straight line. This is an isotopy in the sense of
\cref{def:isotopy}, smooth because on every compact set it is eventually constant, from a knotted line to an
unknotted one; it cannot be ambient (\cref{exer:noncompact}). In the language of \cref{ch:function}, it is a
plot of the mapping object that is not a path in the strong topology (\cref{prop:strongpath}).

\section{Embeddings of discs}

\begin{theorem}[Discs]\label{thm:discs}\Scl[Cerf, Palais; Hirsch 8.3.1 on $\pi_0$]
For an $n$-manifold $M$, the map $\Emb(D^n,M)\to\Fr(TM)$, $f\mapsto(f(0),df_0)$, to the frame bundle is a
homotopy equivalence. More generally $\Emb(D^k,M)$ is equivalent to the bundle of $k$-frames in $TM$, and
$\Emb(\coprod_jD^n,M)$ to the pullback of $\Fr(TM)^j$ to the configuration space $\Conf_j(M)$.
\end{theorem}

The proof is the Alexander-type shrinking of \cref{thm:tubcontract}: $f_t(x)=f(tx)$ rescaled deforms an
embedding to its linear part (\cref{exer:discRn}). On components this is Hirsch's theorem that two
embeddings of $D^k$ in a connected $M$ are isotopic, provided, when $k=n$ and $M$ is orientable, they induce
the same orientation: the bundle of $k$-frames is connected for $k<n$, and $\Fr(TM)$ has one or two
components according as $M$ is non-orientable or orientable. With \cref{thm:gluing} it makes the connected
sum of connected oriented manifolds well defined.

For $M=\R^n$ the theorem identifies the spaces of embeddings of several discs in one:
\[\Emb\bigl(\textstyle\coprod_j\R^n,\R^n\bigr)\ \simeq\ \Conf_j(\R^n)\times\Or(n)^j .\]
These spaces, with composition of embeddings, form the framed little $n$-discs operad; restricting to
embeddings that are rectilinear, or equivalently equipped with a trivialization of the derivative, gives the
operad $E_n$. Write $\cat{Disc}_n$ for the symmetric monoidal $\infty$-category of finite disjoint unions of
copies of $\R^n$ and their embeddings. Every $n$-manifold $M$ defines a presheaf
$E_M=\Emb(-,M)$ on it, and \cref{thm:discs} says that this presheaf is assembled from the configuration
spaces of $M$ and its tangent bundle.

\section{Diffeomorphisms of spheres}

\begin{theorem}\label{thm:spheres}\Scl[Hirsch 8.3.3; Smale; Hatcher; Cerf; Kervaire--Milnor]
The inclusion $\Or(n+1)\to\Diff(S^n)$ is a homotopy equivalence for $n=1,2,3$. For $n\ge5$ the group
$\pi_0\Diff^+(S^n)$ is isomorphic to the group $\Theta_{n+1}$ of exotic $(n+1)$-spheres, and is in general
not trivial.
\end{theorem}

Hirsch proves the statement on $\pi_0$ for $n=1$: every diffeomorphism of the circle is isotopic to the
identity or to a reflection. The case $n=2$ is Smale's, $n=3$ is Hatcher's, and for $n=4$ the inclusion is
known not to be an equivalence. The contrast with \cref{thm:discs} is instructive. Embeddings of a disc into
a manifold are governed by linear algebra in every dimension; diffeomorphisms of a disc fixing the boundary
are not, and the difference is the subject of pseudoisotopy theory.

\section{Immersions}

For immersions the passage from differential conditions to homotopy theory is complete.

\begin{theorem}[Smale--Hirsch]\label{thm:SH}\Scl[Smale \cite{Sma59}; Hirsch \cite{Hir59}]
Let $\dim V<\dim M$, or $\dim V=\dim M$ with $V$ having no closed component. Then the differential
\[\Imm(V,M)\longrightarrow\operatorname{Mono}(TV,TM)\]
to the space of fibrewise injective bundle maps is a weak homotopy equivalence.
\end{theorem}

In the diagram at the end of \cref{ch:function}, this says that for the open relation
$\mathcal R\subset J^1(V,M)$ of injective $1$-jets the inclusion of holonomic sections into all sections
induces an equivalence of shapes. Gromov's formulation explains why. The presheaf $U\mapsto\Imm(U,M)$ on
the open subsets of $V$ takes unions of open sets along handles to homotopy pullbacks, by a restriction
theorem like \cref{thm:palais} that holds for immersions without any compactness hypothesis on the
complement; so it is a homotopy sheaf, determined by its values on discs; and on a disc
$\Imm(D^m,M)\simeq\Emb(D^m,M)$ is the bundle of $m$-frames by \cref{thm:discs}. A homotopy sheaf with those
values is the sheaf of sections of a bundle, namely $\operatorname{Mono}(TV,TM)$.

\begin{corollary}\label{cor:WG}\Scl[Whitney--Graustein; Smale]
Regular homotopy classes of immersions $S^1\to\R^2$ are classified by the rotation number, an integer. All
immersions $S^2\to\R^3$ are regularly homotopic; in particular the sphere can be turned inside out.
\end{corollary}

\begin{proof}
Up to homotopy $\operatorname{Mono}(TS^1,T\R^2)$ is the free loop space of the circle of unit vectors, whose
components are indexed by degree. For the sphere the relevant group is $\pi_2$ of the space of $2$-frames in
$\R^3$, which is $\pi_2(\SO(3))=0$.
\end{proof}

The rotation numbers of the circle, the figure eight and the limaçon $r=1+2\cos\theta$ are $1$, $0$ and $2$;
the script computes them by accumulating the angle of the tangent.

\section{Embeddings, codimension, and calculus}

Embeddings do not satisfy an h-principle. Whether two embeddings are isotopic depends on the
codimension, and the two extremes are classical.

\begin{theorem}\label{thm:whitneyiso}\Scl[Whitney; Wu; Haefliger]
Let $V$ be a compact $m$-manifold. If $n\ge2m+2$, any two embeddings $V\to\R^n$ are isotopic. If $V$ is
connected, $m\ge2$ and $n=2m+1$, the same holds.
\end{theorem}

\begin{proof}[Proof of the first statement]
Join the embeddings by the straight-line homotopy and regard it as a family
$F\colon V\times I\to\R^n$. By the multijet transversality theorem one may perturb $F$, fixing the ends, so
that the set $\{(x,y,t):x\ne y,\ F_t(x)=F_t(y)\}$, of expected dimension $2m+1-n<0$, is empty, and so that the
set where $dF_t$ drops rank, of expected dimension $(m+1)-(n-m+1)<0$, is empty. Then every $F_t$ is an
embedding.
\end{proof}

This is the argument of \cref{cor:counts} with one more parameter. In codimension $2$ nothing of the kind
holds: $\pi_0\Emb(S^1,\R^3)$ is the set of knot types. Between the extremes there is a systematic theory.

\begin{theorem}[Manifold calculus]\label{thm:weiss}
\Sli[Weiss \cite{Wei99}; Goodwillie--Weiss \cite{GW99}; Goodwillie--Klein \cite{GK15}]
For a presheaf of spaces $F$ on the open subsets of an $m$-manifold $V$ that takes isotopy equivalences to
equivalences, let $T_kF(U)$ be the homotopy limit of $F(W)$ over the open $W\subset U$ diffeomorphic to a
disjoint union of at most $k$ discs. Then $T_kF$ is the universal approximation to $F$ by a presheaf that is
a homotopy sheaf for covers in which every set of at most $k$ points lies in some member. For
$F=\Emb(-,M)$ with $\dim M=n$: $T_1F=\Imm(-,M)$, and the map
$\Emb(V,M)\to T_k\Emb(V,M)$ is $(k(n-m-2)+1-m)$-connected. In particular, if $n-m\ge3$ then
$\Emb(V,M)\simeq\operatorname{holim}_kT_k\Emb(V,M)$.
\end{theorem}

So the Smale--Hirsch theorem is the linear approximation, and in codimension at least three the higher
approximations converge. The limit has a description that does not mention $V$'s open sets.

\begin{theorem}\label{thm:BdBW}\Sli[Boavida de Brito--Weiss \cite{BW13}]
$\operatorname{holim}_kT_k\Emb(V,M)$ is the space of derived maps $E_V\to E_M$ between the presheaves on
discs defined by $V$ and $M$.
\end{theorem}

In codimension at least three, then, the space of embeddings of $V$ in $M$ is determined by the
configuration spaces of the two manifolds, with their tangential data and the maps among them: by how
finitely many points sit in each. In codimension two the tower does not converge to the embedding space,
but its stages are still invariants of knots, related to those of finite type.

\section{Factorization homology}

Presheaves on discs classify embeddings; copresheaves on discs give invariants of manifolds. Let $A$ be an
algebra over the operad of \cref{thm:discs} in a symmetric monoidal $\infty$-category, that is, a symmetric
monoidal functor on $\cat{Disc}_n$. Its \term{factorization homology} over an $n$-manifold $M$ is the
colimit $\int_MA=\colim A(U)$ over the discs $U$ embedded in $M$.

\begin{theorem}\label{thm:AF}\Sli[Ayala--Francis \cite{AF15}; Lurie]
$M\mapsto\int_MA$ is symmetric monoidal and satisfies excision: if $M=M_-\cup_{M_0\times\R}M_+$ is a
decomposition along a collared hypersurface, then
$\int_MA\simeq\int_{M_-}A\otimes_{\int_{M_0\times\R}A}\int_{M_+}A$. These properties characterize it.
\end{theorem}

Excision is where \cref{thm:collar,thm:gluing} enter: a collar of $M_0$ is what makes $\int_{M_0\times\R}A$ an
algebra and the two sides modules over it. For $n=1$ and $M=S^1$ the invariant is Hochschild homology,
which for the algebra of functions on a manifold returns the differential forms of \cref{prop:three}.

\section*{Notes}

The fibration theorem is due independently to Palais and to Cerf, with a short proof by Lima. For the
homotopy types of diffeomorphism groups see Smale, Hatcher, and the survey literature on pseudoisotopy;
the failure in dimension four is due to Watanabe. Gromov's \emph{Partial Differential Relations} and the
book of Eliashberg and Mishachev treat the h-principle. Manifold calculus is surveyed by Munson; the
configuration-space description is \cite{BW13}. Factorization homology is developed in \cite{AF15} and in
Lurie's \emph{Higher Algebra}.

\section*{Exercises}

\begin{exercise}\label{exer:flowex}
Find the diffeotopy of $\R^2$ generated by $G(x,t)=(-x_2,x_1)$, and the velocity field of the diffeotopy
$F_t(x)=e^{t}x$. Show that $x^2\partial_x$ does not generate a diffeotopy of $\R$.
\end{exercise}

\begin{exercise}\label{exer:rotation}
Compute the rotation numbers of the unit circle, the figure eight $(\sin t,\sin t\cos t)$ and the limaçon
$r=1+2\cos\theta$. Which pairs are regularly homotopic?
\end{exercise}

\begin{exercise}\label{exer:discRn}
Show that $\Emb(D^n,\R^n)\to\R^n\times\GL_n(\R)$, $f\mapsto(f(0),df_0)$, is a homotopy equivalence.
\end{exercise}

\begin{exercise}\label{exer:noncompact}
Show that the isotopy of long knots described after \cref{thm:palais} is smooth, and that it is not induced
by a diffeotopy of $\R^3$ if the knot is non-trivial.
\end{exercise}

\begin{exercise}\label{exer:highcodim}
Show from \cref{thm:whitneyiso} that any two embeddings of a circle in $\R^4$ are isotopic, and explain where
the proof fails in $\R^3$.
\end{exercise}
\chapter{Surfaces}\label{ch:surfaces}

The source ends by classifying compact surfaces, using everything before it: Morse functions to cut a
surface into pieces, isotopy and gluing to recognise the pieces and put them back, the Euler characteristic
and orientability to tell the results apart. It is the one dimension above $1$ in which the programme of the
book can be carried to the end. The higher-categorical statements of the previous three chapters can also be
carried to the end in dimension two, and this chapter does so. The classification theorem turns out to be the
statement about components; the statements about higher homotopy are theorems of Earle and Eells and of
Madsen and Weiss.

\section{The classification}

Let $\Sigma_g$ be the closed orientable surface of genus $g$, the sphere with $g$ handles, and $N_h$ the
sphere with $h\ge1$ crosscaps.

\begin{theorem}[Classification of surfaces]\label{thm:classification}\Scl[Hirsch 9.3.5, 9.3.10, 9.3.11]
Two compact connected surfaces are diffeomorphic if and only if they have the same Euler characteristic,
the same number of boundary components, and are both orientable or both non-orientable. Every closed
connected surface is diffeomorphic to exactly one $\Sigma_g$, with $\chi=2-2g$, or $N_h$, with $\chi=2-h$.
Removing $b$ open discs lowers $\chi$ by $b$.
\end{theorem}

Hirsch's proof takes a Morse function, orders its critical points, and inducts on the number of saddles.
The pieces between consecutive critical levels are recognised by two lemmas: a connected surface carrying a
Morse function with critical points only of index $0$ and $2$ and boundary a level set is a disc or an
annulus (Hirsch 9.2), and a connected surface with a single critical point, a saddle, is a disc with two holes
if it is orientable and a Möbius band with one hole if it is not (Hirsch 9.3.1--9.3.3; \cref{exer:pants}).
The reassembly uses \cref{thm:gluing,thm:discs,thm:spheres}: gluing along circles depends only on the isotopy
class of the gluing map, and a diffeomorphism of the circle is isotopic to the identity or a reflection.

\section{The monoid of surfaces}

Connected sum is well defined on connected surfaces up to diffeomorphism, by \cref{thm:discs} and the fact
that every surface admits a diffeomorphism reversing the orientation of a small disc up to isotopy when it
is non-orientable, and that $\Sigma_g$ admits an orientation-reversing diffeomorphism. Write $T=\Sigma_1$ and
$P=\RP^2$.

\begin{corollary}\label{cor:monoid}\Scl[Hirsch 9.1.7]
Under connected sum, the diffeomorphism classes of closed connected surfaces form the commutative monoid
with unit $S^2$, generators $T$ and $P$, and the single relation $T\#P=P\#P\#P$.
\end{corollary}

A surface is often presented by a word: a polygon with $2n$ sides labelled by $n$ letters, each occurring
twice, with exponents giving the direction of identification. The quotient has one face and $n$ edges, so
$\chi=V-n+1$ with $V$ the number of vertex classes, and it is orientable exactly when every letter occurs
once with each exponent. The script computes $V$ by a union--find pass and identifies $aba^{-1}b^{-1}$ as
the torus, $abab$ as the projective plane, $abab^{-1}$ as the Klein bottle, and so on (\cref{exer:words}).

\section{The classification as a presentation}

Let $2\cat{Cob}$ be the symmetric monoidal category whose objects are finite disjoint unions of circles and
whose morphisms are diffeomorphism classes of compact oriented cobordisms: the homotopy category of
$\Cob^{\mathrm{or}}_2$ of \cref{def:cobcat}. By \cref{thm:handle} every cobordism is a composite of the four
elementary ones given by a single critical point: the disc as a cobordism $\emptyset\to S^1$ (index $0$) and
$S^1\to\emptyset$ (index $2$), and the pair of pants as a cobordism from two circles to one and from one to
two (index $1$).

\begin{theorem}\label{thm:2cob}\Scl[folklore; Abrams \cite{Abr96}; Kock \cite{Koc04}]
$2\cat{Cob}$ is generated as a symmetric monoidal category by
\[\eta\colon\emptyset\to S^1,\qquad\epsilon\colon S^1\to\emptyset,\qquad
\mu\colon S^1\sqcup S^1\to S^1,\qquad\delta\colon S^1\to S^1\sqcup S^1,\]
subject to the relations: $(\mu,\eta)$ is a commutative monoid, $(\delta,\epsilon)$ is a cocommutative
comonoid, and
$(\mu\sqcup\id)(\id\sqcup\delta)=\delta\mu=(\id\sqcup\mu)(\delta\sqcup\id)$.
\end{theorem}

\begin{proof}[Where the classification enters]
Generation is Morse theory. Using the relations, every connected composite of generators from $m$ circles to
$n$ circles can be brought to the normal form $\delta^{n-1}(\mu\delta)^g\mu^{m-1}$, with $g$ the number of
times a $\delta$ is followed by a $\mu$ that rejoins its two outputs. Two cobordisms with the same normal form
are equal in the presented category. Conversely, two connected oriented cobordisms with the same $m$, $n$
and $g$ are diffeomorphic by \cref{thm:classification}, since $g$ is the genus and $m+n$ the number of
boundary components. So the functor from the presented category to $2\cat{Cob}$ is faithful as well as full.
\end{proof}

The classification theorem is thus exactly the completeness of the relations. The same structure is visible
one level up: by \cref{thm:cerf}, two Morse functions on the same cobordism are joined by a path with
isolated births, deaths and crossings, and each such event is one of the relations of the theorem.

\section{Two-dimensional field theories}

A commutative \term{Frobenius algebra} over a field $k$ is a finite-dimensional commutative algebra $A$
with a linear form $\varepsilon\colon A\to k$ such that $(a,b)\mapsto\varepsilon(ab)$ is non-degenerate.

\begin{theorem}\label{thm:frob}\Scl[Dijkgraaf; Abrams \cite{Abr96}; \cite{Koc04}]
The category of symmetric monoidal functors $2\cat{Cob}\to\cat{Vect}_k$ is equivalent to the category of
commutative Frobenius algebras over $k$ and their isomorphisms, by $Z\mapsto Z(S^1)$. The comultiplication
is dual to the multiplication under the pairing, and for the closed surface of genus $g$,
\[Z(\Sigma_g)=\varepsilon(w^g),\qquad w=\mu\delta(1)=\sum_ie_ie_i^\#,\]
where $(e_i)$ is a basis and $(e_i^\#)$ the dual basis for the pairing.
\end{theorem}

\begin{example}\label{ex:frob}
(a) $A=k[x]/(x^2)$ with $\varepsilon(1)=0$, $\varepsilon(x)=1$: the cohomology of the sphere. The dual basis
of $(1,x)$ is $(x,1)$, so $w=2x$ and $Z(S^2)=0$, $Z(T^2)=2$, $Z(\Sigma_g)=0$ for $g\ge2$. The handle element is
the Euler class.

(b) $A=\C[G]$ for a finite abelian group $G$ of order $n$, with $\varepsilon(g)=\delta_{g,e}/n$. The dual
basis of $(g)$ is $(n\,g^{-1})$, so $w=n^2e$ and $Z(\Sigma_g)=\varepsilon(n^{2g}e)=n^{2g-1}$. This is
$\#\Hom(\pi_1\Sigma_g,G)/\#G$, the number of principal $G$-bundles on $\Sigma_g$ counted with weight one over
the order of the automorphism group.

In every case $Z(T^2)=\dim A$: the torus is the trace of the identity of $Z(S^1)$, as in \cref{thm:CH} for
$d=1$. The script checks the Frobenius relation and these values from the structure constants.
\end{example}

\section{Extended theories}

A field theory in the sense above assigns nothing to a point, and circles cannot be cut. Allowing
$1$-manifolds with boundary and surfaces with corners gives a symmetric monoidal bicategory $\Bord_2$.

\begin{theorem}\label{thm:SP}\Sli[Schommer-Pries \cite{SP09}]
The oriented bordism bicategory $\Bord^{\mathrm{or}}_2$ has a finite presentation by generators and relations,
obtained from the classification of the generic singularities of maps from surfaces to the plane and of
one-parameter families of them. Consequently, extended oriented two-dimensional field theories with values
in the bicategory of algebras, bimodules and bimodule maps are classified by separable symmetric Frobenius
algebras.
\end{theorem}

The proof is the method of \cref{thm:2cob} with $\R$ replaced by $\R^2$: a generic map to the plane has folds
and cusps where a Morse function has critical points, and the jet transversality theorem in one and two
parameters supplies the relations. For framed theories the theorem confirms \cref{thm:CH} in dimension two:
the fully dualizable algebras are the separable ones.

\section{Moduli}

By \cref{thm:ehresmann} the stack of bundles of closed connected surfaces has shape
\[\mathcal M_2\ \simeq\ \coprod_{g\ge0}B\Diff(\Sigma_g)\ \sqcup\ \coprod_{h\ge1}B\Diff(N_h),\]
and \cref{thm:classification} is the computation of $\pi_0\mathcal M_2$. The next question is the homotopy
type of each component.

\begin{theorem}\label{thm:EE}\Scl[Smale; Earle--Eells \cite{EE69}; Gramain]
The identity component of $\Diff(\Sigma_g)$ is contractible for $g\ge2$, is homotopy equivalent to the torus
for $g=1$, and to $\SO(3)$ for $g=0$. The identity component of $\Diff(N_h)$ is contractible for $h\ge3$, a
circle for the Klein bottle, and $\SO(3)$ for $\RP^2$.
\end{theorem}

So for $g\ge2$ the component $B\Diff^+(\Sigma_g)$ is the classifying space $B\Gamma_g$ of the mapping class
group $\Gamma_g=\pi_0\Diff^+(\Sigma_g)$: a surface bundle is determined by its monodromy. It is also the
shape of the moduli stack of Riemann surfaces of genus $g$, since the space of complex structures on a
surface is contractible (a choice in the sense of \cref{prin:choices}).

\begin{theorem}[Madsen--Weiss]\label{thm:MW}\Sli[\cite{MW07}; Harer; \cite{GMTW09}]
The cohomology of $B\Gamma_g$ is independent of $g$ in a range of degrees growing linearly with $g$, and the
stable cohomology is that of a component of $\Omega^\infty MTSO(2)$. Rationally,
\[H^*(B\Gamma_\infty;\mathbb Q)\cong\mathbb Q[\kappa_1,\kappa_2,\dots],\qquad|\kappa_i|=2i .\]
\end{theorem}

The second statement is \cref{thm:GMTW} for oriented surfaces together with the stability theorem; the
classes $\kappa_i$ are the characteristic classes of surface bundles defined by Mumford, Morita and Miller,
and the theorem was Mumford's conjecture.

\section{What the classification is the \texorpdfstring{$\pi_0$}{pi-0} of}

\Cref{tab:surfaces} collects the chapter. In each row the first entry is a statement of the source or a
direct consequence, and the second is the statement about a space or a category of which it is the shadow on
components.

\begin{table}[ht]\centering\small
\begin{tabular}{@{}p{0.43\textwidth}p{0.47\textwidth}@{}}
\toprule
classical statement & of which it is $\pi_0$ or the homotopy category\\
\midrule
closed surfaces are $\Sigma_g$, $N_h$ & $\mathcal M_2\simeq\coprod B\Diff$; \cref{thm:EE}\\
$\mathfrak N_2=\Z/2$, $\Omega_2=0$ & $B\Cob_2\simeq\Omega^{\infty-1}MTO(2)$; \cref{thm:GMTW}\\
generators and relations for $2\cat{Cob}$ & presentation of $\Bord_2$; \cref{thm:SP,thm:CH}\\
a diffeomorphism of $S^1$ is isotopic to an isometry & $\Diff(S^1)\simeq\Or(2)$\\
connected sum is well defined & $\Emb(D^2,M)\simeq\Fr(TM)$; \cref{thm:discs}\\
stable range for $H_*(B\Gamma_g)$ & $\Z\times B\Gamma_\infty^+\simeq\Omega^\infty MTSO(2)$; \cref{thm:MW}\\
\bottomrule
\end{tabular}
\caption{Surfaces: classical statements and the homotopy types behind them.}\label{tab:surfaces}
\end{table}

\section*{Notes}

The classification of surfaces is from the nineteenth century (Möbius, Jordan, Dyck); proofs by Morse
theory are in Hirsch and in Gramain's notes. The equivalence between two-dimensional field theories and
Frobenius algebras is folklore from the late 1980s; \cite{Koc04} is a careful account. The unoriented
analogue is due to Turaev and Turner. For mapping class groups see Farb and Margalit; for the Madsen--Weiss
theorem and its descendants, Galatius and Randal-Williams.

\section*{Exercises}

\begin{exercise}\label{exer:words}
Identify the closed surfaces given by the words $aba^{-1}b^{-1}$, $abab$, $abab^{-1}$, $aabb$,
$abca^{-1}b^{-1}c^{-1}$ and $abcabc$.
\end{exercise}

\begin{exercise}\label{exer:TP}
Deduce $T\#P\cong P\#P\#P$ from \cref{thm:classification}, using $\chi(M\#N)=\chi(M)+\chi(N)-2$.
\end{exercise}

\begin{exercise}\label{exer:frob}
Verify \cref{ex:frob}, and show that for a semisimple commutative Frobenius algebra $A=\bigoplus_ik\,e_i$ with
$\varepsilon(e_i)=\theta_i$ one has $Z(\Sigma_g)=\sum_i\theta_i^{1-g}$.
\end{exercise}

\begin{exercise}\label{exer:pants}
Let $W$ be a compact connected surface with a Morse function $f\colon W\to[a,b]$, $\partial W=f^{-1}\{a,b\}$,
having exactly one critical point, of index $1$. Show that $\chi(W)=-1$, and that $W$ is a disc with two
holes if it is orientable and a Möbius band with one hole if it is not.
\end{exercise}

\begin{exercise}\label{exer:torusdim}
Show from the relations of \cref{thm:2cob} that $Z(T^2)=\dim Z(S^1)$ for every two-dimensional field theory
with values in vector spaces.
\end{exercise}
\part{Assessment}
\chapter{What the Language Did}\label{ch:coda}

The two earlier volumes of this series ended with an assessment of how much of their higher-categorical
vocabulary was doing work. The same question is asked here, and the answer is different in both directions:
more of the vocabulary is essential, and less of the mathematics is touched by it.

\section{Where the proofs live}

\begin{principle}[Four inputs]\label{prin:inputs}\Sob
The theorems of the source cited in this book are derived there from four analytic facts and the topology of
function spaces: the inverse function theorem; the existence of partitions of unity; the Morse--Sard
theorem; and the existence, uniqueness and smooth dependence of solutions of ordinary differential
equations. The first gives \cref{thm:transpull} and tubular neighbourhoods. The second gives approximation,
collars, and every passage from local to global. The third gives transversality and so every genericity
statement. The fourth gives the regular interval theorem, isotopy extension, and the handle description of a
critical point.
\end{principle}

None of the four is proved, shortened, or explained by anything in this book. The categorical statements
begin where the four inputs end. \Cref{thm:CS} characterizes derived manifolds, but that the transverse
fibre products exist is the inverse function theorem, and that they are dense is Sard's theorem.
\Cref{thm:concordance} classifies concordance classes, but its proof uses partitions of unity in an
essential way. \Cref{thm:palais} says that restriction is a fibration, and the lifting is done by
integrating a vector field. What the language provides is the form of the conclusions, and the form matters
for what can be done next.

\section{The source as a book about components}

\begin{table}[ht]\centering\small
\begin{tabular}{@{}p{0.36\textwidth}p{0.52\textwidth}@{}}
\toprule
theorem of the source & the statement of which it is the shadow on $\pi_0$\\
\midrule
homotopic maps pull back isomorphic bundles; classification (4.2.4, 4.3.4) & $\Pi[M,B\GL_k]\simeq\Map(M,B\Or(k))$ (\cref{thm:concordance})\\
tubular neighbourhoods and collars are unique up to isotopy (4.5.3, 4.6) & the spaces of them are contractible (\cref{thm:tubcontract,thm:collar})\\
Hopf's theorem (5.1.10) & $S^n\to K(\Z,n)$ is $(n+1)$-connected; Pontryagin--Thom in $M$\\
invariance of intersection numbers (5.2.1) & derived fibre products have cobordism classes (\cref{thm:spivak})\\
$\chi$ by vector fields equals $\chi$ by homology (6.3.8) & a trace computed before and after a monoidal functor (\cref{thm:DP})\\
regular interval theorem (6.2.2) & families of compact manifolds are classified by $\coprod B\Diff$ (\cref{thm:ehresmann})\\
Thom's theorem (7.2.3) & $B\Cob_d\simeq\Omega^{\infty-1}MTO(d)$ (\cref{thm:GMTW})\\
isotopy extension (8.1.3) & $\Diff_c(M)\to\Emb(V,M)$ is a fibre bundle (\cref{thm:palais})\\
embeddings of discs are isotopic (8.3.1) & $\Emb(D^n,M)\simeq\Fr(TM)$ (\cref{thm:discs})\\
classification of surfaces (9.3.11) & $\mathcal M_2\simeq\coprod B\Diff$, with \cref{thm:EE,thm:MW}\\
\bottomrule
\end{tabular}
\caption{The main theorems of the source as statements about components.}\label{tab:pi0}
\end{table}

\Cref{tab:pi0} is the summary of the book. Read down the left column and one has Hirsch's table of contents;
read down the right and one has a good part of geometric topology since. In every row the left-hand
statement is what one can prove with one parameter of genericity (a path between two choices) and the
right-hand statement is what one gets by carrying all parameters. The source's method, described in the
Preface, already contains the idea: an invariant is independent of choices because two choices are joined by
a generic path. The higher statement is that the space of choices is contractible because spheres of choices
bound generic discs.

\section{What is higher here}

In the volume on game theory the $\infty$-categorical structure was largely decorative, and in the volume
on information theory it was concentrated in one place (the bar construction behind entropy). Here it is
needed in at least four places where no $1$-categorical substitute exists.

First, derived fibre products. The last row of \cref{tab:four} has no solution in any ordinary category of
ringed spaces, and \cref{thm:CS} characterizes the $\infty$-category that solves it.

Second, moduli. The stack $B\GL_k$ and the stack of manifold bundles are not sheaves of sets, and their shapes
$B\Or(k)$ and $\coprod B\Diff(W)$ are not discrete.

Third, composition of cobordisms, which is associative only up to the contractible choices of
\cref{ch:tubular}; the cobordism category is an $\infty$-category in an unavoidable way, and its homotopy type
is the content of \cref{thm:GMTW}.

Fourth, embedding spaces, whose description in \cref{thm:weiss,thm:BdBW} is by homotopy limits over
categories of discs.

In each case the higher statement was a theorem of someone else. This book contributes arrangement: the
observation that Hirsch's nine chapters are, in order, the $\pi_0$ of these theories, with the dictionary
made explicit and the status of each translation marked.

\section{What the language did not do}

It proved no theorem of differential topology. It did not make the analytic inputs of \cref{prin:inputs}
easier. It did not internalize genericity: the Baire property of the strong topology has no formulation in
terms of the sheaf $[M,N]$ that we know of (\cref{rem:nobaire}), and the density theorems are stated here
exactly as in the source.

It is blind, at the level of generality used here, to the phenomena that distinguish the smooth category
from the topological one. Exotic spheres appear in \cref{thm:spheres} as $\pi_0\Diff(S^n)$ and nowhere else;
nothing in the categorical formalism predicts them. The same formalism applied to topological or
piecewise-linear manifolds gives the same tables with different entries, and the differences among the
entries are the hard part of the subject.

And several of the statements quoted are much harder than the theorems they generalize. One should not
come away thinking that \cref{thm:GMTW} is a reformulation of Thom's theorem. It contains Thom's theorem as
its statement about components, and its proof uses new ideas.

\section{Directions}

\begin{problem}[Internal genericity]\label{prob:baire}\Spg
Find a property of a subobject $A\subset[M,N]$ of the mapping object, stated in terms of plots, that is
implied by residuality in the strong topology, implies the conclusion of the parametric transversality
theorem for every sufficiently rich plot, and is closed under countable intersections.
\end{problem}

\begin{problem}[Synthetic transversality]\label{prob:synthetic}\Spg
In a model of synthetic differential geometry, or in differential cohesive homotopy type theory, the
transverse fibre products are those preserved by passage to tangent complexes (\cref{thm:derivedtrans}).
Give a synthetic proof of \cref{lem:fibre} and isolate the axiom that plays the role of Sard's theorem.
\end{problem}

\begin{problem}[Families]\label{prob:families}\Spg
\Cref{prin:generic} concerns a single diagram. For a family of diagrams parametrized by a manifold $B$, the
limits form a derived manifold over $B$, and the points of $B$ where the fibre is not a manifold form a
closed set of positive codimension in good cases. Describe the stratification of $B$ by the type of the
singular fibre in terms of the tangent complex, in the generality of \cref{thm:CS}.
\end{problem}

The third problem is the one the companion volume takes up in a special case. A neural network architecture
is a smooth family of maps parametrized by its weights; the decision boundaries, level sets and fixed-point
sets of the network are fibres of fibre products over the parameter space; and the questions asked about
them in practice (when is the boundary a manifold, how does its topology change in training, what is the
set of parameters fitting the data) are questions about that stratification. The theorems of this book that
are used there are \cref{thm:transpull,thm:parametric,lem:fibre}, the openness results of
\cref{ch:function}, Morse theory in the form of \cref{cor:pushout,thm:cerf}, and the isotopy theory of
\cref{ch:isotopy}.
\appendix
\chapter{Categorical Toolkit}\label{app:toolkit}

This appendix fixes the meaning of the categorical terms used in the text. It is a list of definitions with
references and proves nothing.

\section*{Spaces and \texorpdfstring{$\infty$}{infinity}-categories}

A \emph{space} is an $\infty$-groupoid: a homotopy type, presented for instance by a Kan complex or a CW
complex. Spaces form an $\infty$-category $\Spc$. An \emph{$\infty$-category} has objects and, for each pair
of objects, a space of morphisms, with composition associative up to coherent homotopy; quasi-categories and
complete Segal spaces are equivalent models. An ordinary category is an $\infty$-category whose mapping
spaces are discrete. Limits and colimits in an $\infty$-category are homotopy limits and colimits in any
model; in particular a pushout in $\Spc$ is a homotopy pushout. The \emph{classifying space} $B\mathcal C$ of
an $\infty$-category is the space obtained by inverting all morphisms. A category in which composition is
defined after a choice from a contractible space, such as the cobordism category, presents an
$\infty$-category.

\section*{Sheaves on Euclidean spaces}

$\Cart$ is the category of the spaces $\R^n$ and smooth maps, with the coverage of
\cref{def:smset}. A sheaf of sets on it is a smooth set; a sheaf of spaces (a functor
$\Cart^{\op}\to\Spc$ satisfying descent for covers) is a smooth $\infty$-groupoid. The $\infty$-topos
$\mathbf H$ of these has a chain of adjoint functors
\[\Pi\ \dashv\ \operatorname{Disc}\ \dashv\ \Gamma\ \dashv\ \operatorname{coDisc}\colon\quad
\mathbf H\rightleftarrows\Spc ,\]
where $\Gamma$ is evaluation at the point, $\operatorname{Disc}$ is the constant sheaf, and $\Pi$ is the
shape \cite{Sch13}. For $X\in\mathbf H$ and a manifold $M$, a \emph{section of $X$ over $M$} is a map
$M\to X$ in $\mathbf H$; the mapping object $[M,X]$ has sections over $U$ the sections of $X$ over $U\times
M$. The delooping $BG$ of a sheaf of groups $G$ is the sheaf of groupoids with one object and automorphisms
$G$, sheafified; maps $M\to BG$ are principal $G$-bundles.

\section*{Dualizable objects and traces}

Let $(\mathcal C,\otimes,\mathbf 1)$ be a symmetric monoidal category or $\infty$-category. An object $X$ is
\emph{dualizable} if there are an object $X^\vee$ and maps $\eta\colon\mathbf 1\to X\otimes X^\vee$,
$\epsilon\colon X^\vee\otimes X\to\mathbf 1$ such that the composites
$X\to X\otimes X^\vee\otimes X\to X$ and $X^\vee\to X^\vee\otimes X\otimes X^\vee\to X^\vee$ are identities.
The \emph{trace} of $f\colon X\to X$ is $\epsilon\circ\text{swap}\circ(f\otimes1)\circ\eta\in
\operatorname{End}(\mathbf 1)$, and the \emph{Euler characteristic}, or dimension, of $X$ is
$\tr(\id_X)$ \cite{DP80,PS14}. In vector spaces the dualizable objects are the finite-dimensional ones; in
spectra they are the finite spectra. An object of a symmetric monoidal $(\infty,d)$-category is \emph{fully
dualizable} if it is dualizable and the evaluation and coevaluation maps have adjoints, which have adjoints,
and so on through dimension $d-1$ \cite{Lur09}.

\section*{Spectra and Thom spectra}

The $\infty$-category $\Sp$ of spectra is the stabilization of pointed spaces; it is symmetric monoidal under
the smash product, with unit the sphere spectrum $\mathbb S$, and $\pi_n\mathbb S$ is the $n$-th stable
homotopy group of spheres. There are adjoint functors $\Sigma^\infty_+\colon\Spc\to\Sp$ and
$\Omega^\infty\colon\Sp\to\Spc$, and $\Omega^{\infty-1}E$ means $\Omega^\infty\Sigma E$. The Thom spectrum of a
stable vector bundle $\xi$ over $X$ has $k$-th space the Thom space of the rank-$k$ part of $\xi$ over the
corresponding piece of $X$; equivalently it is the colimit of \cref{thm:ABGHR}. $MO$ is the Thom spectrum of
the universal stable bundle over $B\Or$, and $MTO(d)$ that of the virtual bundle $-\gamma_d$ over $B\Or(d)$.

\section*{\texorpdfstring{$C^\infty$}{C-infinity}-rings}

A $C^\infty$-ring is a product-preserving functor $\Cart\to\Set$ (\cref{def:cinf}); a simplicial
$C^\infty$-ring is a product-preserving functor $\Cart\to\Spc$ (\cref{def:scr}). The free $C^\infty$-ring on
$n$ generators is $C^\infty(\R^n)$. A $C^\infty$-ring is \emph{finitely presented} if it is
$C^\infty(\R^n)/(f_1,\dots,f_k)$; a simplicial one is finitely presented if it is a finite colimit of free
ones, and the opposite of the $\infty$-category of these is $\DMfd$ (\cref{thm:CS}). Hadamard's lemma, that a
smooth function vanishing on $\{x_1=0\}$ is divisible by $x_1$, is what makes quotients of $C^\infty$-rings
by coordinate functions behave as in algebra.

\section*{Notation}

\begin{center}\small
\begin{tabular}{@{}p{0.3\textwidth}p{0.6\textwidth}@{}}
\toprule
$\Mfd$, $\Cart$, $\cat{Open}$ & manifolds; Euclidean spaces; open subsets of Euclidean spaces\\
$\SmSet$, $\Dflg$, $\mathbf H$ & smooth sets; diffeological spaces; smooth $\infty$-groupoids\\
$\CinfRing$, $\DMfd$ & $C^\infty$-rings; derived manifolds\\
$[M,N]$, $\Pi$ & mapping object; shape\\
$\Emb$, $\Imm$, $\Diff$ & embeddings, immersions, diffeomorphisms (as smooth sets or their shapes)\\
$f\trans g$, $M\times^h_NP$ & transverse; derived fibre product\\
$\Th(\xi)$, $M^{-TM}$ & Thom space; Thom spectrum of the stable normal bundle\\
$\mathfrak N_n$, $\Omega_n$, $\Cob_d$, $\Bord_d$ & cobordism groups; cobordism categories\\
$\Sigma_g$, $N_h$ & closed surfaces, orientable of genus $g$ and non-orientable with $h$ crosscaps\\
\bottomrule
\end{tabular}
\end{center}
\chapter{Concordance with Hirsch}\label{app:concordance}

The table lists the sections of M.~W.~Hirsch, \emph{Differential Topology} (Graduate Texts in
Mathematics~33), with the place in this book where their content is treated and the form it takes there.
Section numbers are those of the source; ``2.1'' means Chapter~2, Section~1.

{\small
\begin{longtable}{@{}p{0.29\textwidth}p{0.14\textwidth}p{0.49\textwidth}@{}}
\toprule
Source & Here & Form\\
\midrule
\endhead
1.1--1.2 Differential structures; maps and tangent bundles & \cref{ch:manifolds}, \cref{ch:tangent} & the category $\Mfd$ and its four enlargements; $T$ as a Weil functor\\
1.3 Embeddings and immersions & \S\S1.6--1.7 & transverse fibre products; $\Mfd$ as an idempotent completion\\
1.4 Manifolds with boundary & \S1.8 & neat submanifolds; the half-line as a smooth set\\
2.1 The weak and strong topologies & \S\S3.1, 3.4 & open conditions as $D$-open subobjects of $[M,N]$\\
2.2 Approximations & \S3.1 & density; smoothing of structures (\cref{thm:smoothing})\\
2.3 Approximations on $\partial$-manifolds and pairs & \S1.8 & not treated separately\\
2.4 Jets and the Baire property & \S\S2.5, 3.5, 3.6 & jets compose; genericity is not internal to the topos\\
2.5 Analytic approximations & --- & not treated\\
3.1 The Morse--Sard theorem & \S4.1 & stated; the single analytic source of genericity\\
3.2 Transversality & \S\S4.2--4.6 & fibrewise lemma; derived manifolds; cobordism classes of derived fibre products\\
4.1--4.3 Vector bundles; classification & \cref{ch:bundles} & the stack $B\GL_k$; concordance classification; bundles as idempotents\\
4.4 Oriented vector bundles & \S5.5 & tangential structures\\
4.5 Tubular neighbourhoods & \S6.1 & contractible space of choices\\
4.6 Collars and tubular neighbourhoods of neat submanifolds & \S6.2 & gluing; why cobordism categories are $\infty$-categories\\
4.7 Analytic differential structures & --- & not treated\\
5.1 Degrees of maps & \S\S7.1--7.2 & Hopf's theorem as Pontryagin--Thom in $M$ and as a Postnikov statement\\
5.2 Intersection numbers and the Euler characteristic & \S\S7.3--7.6 & classes of derived intersections; traces; the diagonal\\
5.3 Historical remarks & --- & ---\\
6.1 Morse functions & \S8.1 & generic points of $[M,\R]$\\
6.2 Differential equations and regular level surfaces & \S8.2 & Ehresmann; families have no local moduli\\
6.3 Passing critical levels and attaching cells & \S\S8.3--8.5 & homotopy pushouts; the sublevel functor; barcodes\\
6.4 CW-complexes & \S8.3 & compact manifolds are finite spaces\\
7.1 Cobordism and transversality & \S9.1 & cobordism groups\\
7.2 The Thom homomorphism & \S\S9.2--9.5 & Thom spectra as colimits; the cobordism category\\
8.1 Extending isotopies & \S\S10.1--10.2 & restriction is a fibration\\
8.2 Gluing manifolds together & \S6.2 & contractible choices\\
8.3 Isotopies of disks & \S\S10.3--10.4 & $\Emb(D^n,M)\simeq\Fr(TM)$; the little discs operad\\
9.1 Models of surfaces & \S11.2 & the monoid of surfaces\\
9.2 Characterization of the disk & \S11.1 & used in the classification\\
9.3 The classification of compact surfaces & \S\S11.1--11.7 & presentation of the cobordism category; field theories; moduli\\
\bottomrule
\end{longtable}
}

Topics treated here that are not in the source: Weil algebras and tangent categories (\cref{ch:tangent});
the mapping object and its shape (\cref{ch:function}); derived manifolds (\cref{ch:transversality}); Atiyah
duality and traces (\cref{ch:tubular,ch:degree}); flow categories, Cerf theory and framed functions
(\cref{ch:morse}); the cobordism category and the cobordism hypothesis (\cref{ch:cobordism}); the
Smale--Hirsch theorem, manifold calculus and factorization homology (\cref{ch:isotopy}); field theories and
moduli of surfaces (\cref{ch:surfaces}).
\chapter{Computations}\label{app:computations}

The script \texttt{verify-differential-topology.py} that accompanies this book uses only the Python standard library and runs in a
few seconds. It performs 71 checks, each printing \texttt{PASS} or \texttt{FAIL}, and exits with status
zero if and only if all pass. The checks recompute the numerical examples in the text and test finite or
numerical instances of structural statements. They are evidence that the examples are stated correctly;
they do not prove the theorems.

{\small
\begin{longtable}{@{}p{0.13\textwidth}p{0.79\textwidth}@{}}
\toprule
Chapter & What is checked\\
\midrule
\endhead
\ref{ch:manifolds} & the differential of $A\mapsto A^{\mathsf T}A$ is onto at orthogonal matrices, $n=2,3,4$
(\cref{ex:On}); signed zero counts of $x^2-\varepsilon$; a smooth idempotent has rank equal to trace along
its fixed set (\cref{thm:karoubi}); the stereographic transition is an involution\\
\ref{ch:tangent} & the chain rule in dual numbers against finite differences, in one and two variables; the
expansion in $\R[\varepsilon,\eta]/(\varepsilon^2,\eta^2)$ and its symmetry (\cref{prop:flip}); the vertical
lift; associativity of composition of $2$-jets in one and several variables (\cref{prop:jets}); the reverse
chain rule and the transpose relation (\cref{prop:reverse})\\
\ref{ch:function} & \cref{ex:weakfail}: $g_n$ equals the identity on $[-n+1,n-1]$, has derivative $-1$
somewhere, and stays outside a strong neighbourhood; a degenerate critical point of $x-\beta(x-n)$
(\cref{exer:morseweak})\\
\ref{ch:transversality} & the dimension $\rho(m+n-\rho)$ of the tangent space to the rank-$\rho$ matrices in
four cases (\cref{prop:rank}); the linear algebra of \cref{lem:fibre} on random instances, including
non-transverse ones; the family $x^2-v$; the Koszul complex of $x^2$ in a polynomial model; the Euler
characteristic of the tangent complex; injectivity of random projections of a circle from $\R^5$ to $\R^3$
(\cref{cor:counts})\\
\ref{ch:bundles} & \cref{ex:taut}: the idempotent of the tautological line and its monodromy; the idempotent
of $TS^2$; the count of line bundles on the torus and Klein bottle (\cref{exer:lines})\\
\ref{ch:tubular} & the Jacobian $1+t$ of the normal exponential map of the circle; the focal radius $b^2/a$
of the ellipse (\cref{exer:focal}); convexity of inner products; the two linear homotopies of the shift
trick and the failure of the naive one (\cref{exer:shift})\\
\ref{ch:degree} & the degree of $z^k$ by preimages and by winding integral; the degree of the antipodal map;
intersection numbers on the torus (\cref{ex:torus}); fixed points and Lefschetz numbers of six toral
automorphisms (\cref{ex:lefschetz}); Poincaré--Hopf indices on $S^2$ and $S^3$; Euler characteristics and
mod $2$ Betti numbers of triangulations of $S^2$, $\RP^2$ and the torus; the Gauss linking integral
(\cref{ex:linking})\\
\ref{ch:morse} & the critical points and indices of \cref{ex:torusheight}; the Morse polynomial; the
birth--death family; the degree-zero barcode of random functions on a circle (\cref{prop:persist});
\cref{exer:rpn}; the deformation of \cref{exer:model}\\
\ref{ch:cobordism} & the ranks of $\mathfrak N_n$ for $n\le10$; the Stiefel--Whitney numbers of
\cref{ex:sw}; the arithmetic of \cref{exer:chimod2}; the degree of the collapse of a framed point
(\cref{exer:framed0})\\
\ref{ch:isotopy} & a time-dependent flow on $\R^2$ is injective with positive Jacobian; the rotation flow;
blow-up of $x^2\partial_x$; the rotation numbers of \cref{cor:WG}; the shrinking of \cref{exer:discRn}\\
\ref{ch:surfaces} & vertex counts, Euler characteristics and orientability for seven polygon words
(\cref{exer:words}); $T\#P$ and $P\#P\#P$; the Frobenius algebras of \cref{ex:frob}: handle elements,
$Z(\Sigma_g)$ for $g\le3$, the Frobenius relation, and $Z(T^2)=\dim A$\\
\bottomrule
\end{longtable}
}

Numerical tolerances are stated in the script. The random seed is fixed, so the output is reproducible.
\chapter{Solutions to the Exercises}\label{app:solutions}

Solutions to all exercises follow, by chapter. Where a solution relies on a computation, the computation is
also carried out by \texttt{verify-differential-topology.py}.
\section*{Chapter 1}

\sol{1.1} In $\Mfd$: a cone is $(S,u)$ with $u\colon S\to\R$ smooth and $u^2=0$, so $u=0$ and the cone is the
unique map $S\to\pt$. The same argument applies to every $S=\R^n$, so the sheaf-theoretic fibre product has
exactly one plot of each dimension and is the terminal smooth set. In $\CinfRing$ the pushout of
$C^\infty(\R)\xleftarrow{x\mapsto x^2}C^\infty(\R)\xrightarrow{\mathrm{ev}_0}\R$ is $C^\infty(\R)/(x^2)$, and
Hadamard's lemma $g=g(0)+g'(0)x+x^2h$ shows that $g\mapsto g(0)+g'(0)\varepsilon$ is an isomorphism onto
$\R[\varepsilon]/(\varepsilon^2)$. A $C^\infty$-ring map $\phi\colon C^\infty(M)\to\R[\varepsilon]$ has the
form $\phi(g)=\phi_0(g)+\phi_1(g)\varepsilon$. The component $\phi_0$ is an $\R$-algebra map to $\R$, hence
evaluation at a point $p$ (every character of $C^\infty(M)$ is a point evaluation, $M$ being second
countable). Multiplicativity of $\phi$ gives $\phi_1(gh)=g(p)\phi_1(h)+h(p)\phi_1(g)$, so $\phi_1$ is a
derivation at $p$, a tangent vector. Conversely a tangent vector $v$ at $p$ gives
$g\mapsto g(p)+v(g)\varepsilon$, which respects all smooth operations by the chain rule.

\sol{1.2} The proof in the text shows that each $x\in\Fix(e)$ has a neighbourhood $U$ on which $e$ has
constant rank $k=\tr de_x$. By the constant rank theorem there are charts near $x$ and $e(x)=x$ in which
$e$ is $(u,v)\mapsto(u,0)$ with $u\in\R^k$; shrinking $U$, the set $e(U)$ is a $k$-dimensional embedded
submanifold near $x$, and $\Fix(e)\cap U\subset e(U)$. Conversely $e(U)\cap U'\subset\Fix(e)$ for a smaller
neighbourhood $U'$, because $e(e(y))=e(y)$. So $\Fix(e)$ is an embedded submanifold near $x$; it is closed
as the equalizer of $e$ and $\id$ in a Hausdorff space. Smoothness of $e$ into $\Fix(e)$ follows since the
submanifold is embedded. For the example take $M=\R\sqcup\R^2$ and $e$ the map which is the identity on $\R$
and $(x,y)\mapsto(0,0)$ on $\R^2$; then $\Fix(e)=\R\sqcup\{0\}$. On a connected $M$ the set $\Fix(e)=e(M)$ is
connected, so the dimension is constant.

\sol{1.3} $f(A)=A^{\mathsf T}A$ is polynomial, hence smooth, with values in the vector space $\cat{Sym}_n$ of
dimension $\frac12n(n+1)$. Its differential at $A$ is $df_A(H)=A^{\mathsf T}H+H^{\mathsf T}A$. If
$A^{\mathsf T}A=I$ and $S$ is symmetric, then $H=\frac12AS$ gives
$\frac12A^{\mathsf T}AS+\frac12S^{\mathsf T}A^{\mathsf T}A=S$. So $I$ is a regular value and
$\dim\Or(n)=n^2-\frac12n(n+1)=\frac12n(n-1)$. The tangent space at $I$ is $\ker df_I=\{H:H+H^{\mathsf T}=0\}$.

\sol{1.4} A plot of $[M,N]$ of dimension $n$ is by definition a smooth map $F\colon\R^n\times M\to N$, and the
function it induces on points is $s\mapsto F(s,-)$, which determines $F$; so $[M,N]$ is concrete and its points
are $\Hom(\pt\times M,N)=C^\infty(M,N)$. For the adjunction, when $S=\R^n$ the bijection is the definition.
A general smooth set is a colimit of representables, $S=\colim_i\R^{n_i}$; the functor $-\times M$ preserves
colimits because $\SmSet$ is a topos, so
$\Hom(S\times M,N)=\lim_i\Hom(\R^{n_i}\times M,N)=\lim_i[M,N](\R^{n_i})=\Hom(S,[M,N])$.

\sol{1.5} Let $f(x)=g(y)=z$. The tangent space to $\Delta_N$ at $(z,z)$ is $\{(w,w)\}$. Transversality of
$f\times g$ to $\Delta_N$ at $(x,y)$ says that every $(a,b)\in T_zN\oplus T_zN$ is
$(df_xu+w,\,dg_yv+w)$ for some $u,v,w$. Subtracting, this holds if and only if every $a-b$ is of the form
$df_xu-dg_yv$, that is, if and only if $df_x(T_xM)+dg_y(T_yP)=T_zN$: given such a decomposition of $a-b$, set
$w=a-df_xu$.
\section*{Chapter 2}

\sol{2.1} For open subsets of Euclidean spaces,
$T(U\times V)=U\times V\times\R^m\times\R^n\cong TU\times TV$ by reordering coordinates, and the reordering
commutes with the maps $T(f\times g)$ and $Tf\times Tg$ since $D(f\times g)=Df\oplus Dg$. The isomorphism is
therefore natural and glues over atlases. For the second statement, $Z=f^{-1}(z)$ is a submanifold with
$T_xZ=\ker df_x$ by \cref{thm:transpull}. A vector $(x,v)\in TM$ satisfies $Tf(x,v)=0_z$ exactly when
$f(x)=z$ and $df_xv=0$, that is, when $(x,v)\in TZ$. One checks that $0_z$ is a regular value of $Tf$: in
coordinates $D(Tf)_{(x,v)}(\xi,\nu)=(Df_x\xi,\,D^2f_x(v,\xi)+Df_x\nu)$, which is onto when $Df_x$ is. So the
equality is one of submanifolds.

\sol{2.2} An $A$-point of $\R^n$ is $x+v\,t+\tfrac12a\,t^2$ with $t^3=0$, the $2$-jet at $0$ of the curve
$t\mapsto x+vt+\frac12at^2$; a map $f$ acts by Taylor expansion, which is the composition of $2$-jets, so
$T_AM$ is the bundle of $2$-jets of curves and its fibre has dimension $2n$. The map $x\mapsto
\varepsilon+\eta$ is a homomorphism since $(\varepsilon+\eta)^3=3\varepsilon^2\eta+3\varepsilon\eta^2=0$,
and it is injective: $1,\ \varepsilon+\eta,\ (\varepsilon+\eta)^2=2\varepsilon\eta$ are independent. On
points it sends $x+vt+\frac12at^2$ to $x+v\varepsilon+v\eta+a\varepsilon\eta$, that is, to $(x,v,v,a)$ in
the coordinates of \cref{tab:weil}. The fixed set of the flip is $\{v=w\}$, and $T(p)$ and $p_T$ are the two
projections $(x,v,w,a)\mapsto(x,v)$ and $(x,w)$, so the condition $T(p)=p_T$ is again $v=w$. The image is
exactly $\{(x,v,v,a)\}$, of fibre dimension $2n$, and the map is an embedding.

\sol{2.3} Write $f(x)=Ax+\frac12B(x,x)+\dots$ and $g(y)=Cy+\frac12E(y,y)+\dots$ with $B,E$ symmetric bilinear.
Then $g(f(x))=CAx+\frac12\bigl(C\,B(x,x)+E(Ax,Ax)\bigr)+O(|x|^3)$, so
\[(C,E)\circ(A,B)=\bigl(CA,\ C\circ B+E\circ(A\times A)\bigr).\]
For a third jet $(F,G)$,
$(F,G)\circ\bigl((C,E)\circ(A,B)\bigr)=\bigl(FCA,\ FCB+FE(A\times A)+G(CA\times CA)\bigr)$ and
$\bigl((F,G)\circ(C,E)\bigr)\circ(A,B)=\bigl(FCA,\ FCB+(FE+G(C\times C))(A\times A)\bigr)$, which agree.

\sol{2.4} A point of $TT\R^n=T_{W\otimes W}\R^n$ is $x+v\varepsilon+w\eta+a\varepsilon\eta$. A homomorphism
$\phi\colon A\to B$ acts on $A$-points by applying $\phi$ to the coefficients. Addition: a point of
$T_{W\times_\R W}$ is $x+v_1\varepsilon_1+v_2\varepsilon_2$, and $\varepsilon_i\mapsto\varepsilon$ sends it to
$x+(v_1+v_2)\varepsilon$. Vertical lift: $x+v\varepsilon\mapsto x+v\varepsilon\eta$, which is $(x,0,0,v)$.
Flip: exchanging $\varepsilon$ and $\eta$ sends $(x,v,w,a)$ to $(x,w,v,a)$. Each map is natural because it
is induced by a homomorphism (\cref{thm:weil}(iii)); directly, naturality of $\ell$ is the identity
$f(x+v\varepsilon\eta)=f(x)+Df_x(v)\varepsilon\eta$, since $(\varepsilon\eta)^2=0$.

\sol{2.5} Let $(f,f^\sharp)$, $(g,g^\sharp)$, $(h,h^\sharp)$ be composable. Both bracketings have forward part
$hgf$ and backward part
$(x,u')\mapsto f^\sharp\bigl(x,g^\sharp(f(x),h^\sharp(gf(x),u'))\bigr)$, as one sees by unfolding the
definition twice; so composition is associative. For $F=f_k\cdots f_1$ with scalar output, \cref{prop:reverse}
gives $R[F](x,1)=R[f_1](x_0,R[f_2](x_1,\dots R[f_k](x_{k-1},1)))$ with $x_0=x$ and $x_i=f_i(x_{i-1})$. The
forward pass computes and stores $x_0,\dots,x_{k-1}$; the backward pass applies the transposed Jacobians
$(Df_i)_{x_{i-1}}^{\mathsf T}$ to a single covector, for $i=k,\dots,1$. The result is the gradient of $F$ at
$x$.
\section*{Chapter 3}

\sol{3.1} The support of $\beta(\cdot-n)$ is $[n-1,n+1]$, so $g_n=\id$ on $[-n+1,n-1]$ and every weak
neighbourhood of $\id$, which constrains maps on a compact set only, contains $g_n$ for large $n$. Since
$\beta'$ attains its maximum $1$ at some $x^*$, $g_n'(n+x^*)=-1<0$ while $g_n'=1$ outside the support, so
$g_n$ is not monotone. For the strong topology, $|g_n(n)-n|=2\beta(0)>\beta(0)e^{-n^2}$, so no $g_n$ lies in the stated strong
neighbourhood of the identity. Not closed: $h_n(x)=x/n$ are diffeomorphisms converging weakly to the constant map $0$.

\sol{3.2} Fix $t_0$. We claim there are $\delta>0$ and a compact $K$ with $f_t=f_{t_0}$ off $K$ for
$|t-t_0|<\delta$. If not, for each $n$ there are $t_n$ with $|t_n-t_0|<1/n$ and $x_n\notin[-n,n]$ with
$f_{t_n}(x_n)\ne f_{t_0}(x_n)$; passing to a subsequence, the $x_n$ are distinct and tend to infinity.
Choose a continuous $\varepsilon\colon\R\to(0,\infty)$ with
$\varepsilon(x_n)<|f_{t_n}(x_n)-f_{t_0}(x_n)|$ for all $n$. The strong neighbourhood
$\{g:|g-f_{t_0}|<\varepsilon\}$ contains no $f_{t_n}$, contradicting continuity at $t_0$. Cover $[0,1]$ by
finitely many such intervals and take the union of the compact sets. Conversely, if $g=f$ off a compact
$K$, the straight-line path $f+t(g-f)$ is strongly continuous: it is constant off $K$, and on $K$ uniform
convergence suffices. So the path component of $f$ is as stated.

\sol{3.3} Let $F\colon\R^n\times M\to N$ be smooth and $s_0\in\R^n$. For a compact $K\subset M$ in a chart, $F$
and its first derivative in the $M$-direction are uniformly continuous on $\bar B(s_0,1)\times K$, so
$\sup_{x\in K}\bigl(|F_s(x)-F_{s_0}(x)|+\|D_xF_s(x)-D_xF_{s_0}(x)\|\bigr)\to0$ as $s\to s_0$; this is
convergence in $C^1_W$. For the example let $F_s(x)=(\gamma_s(x),0)$ with $\gamma_s(x)=x-2\beta(x-1/s)$ for
$s>0$ and $\gamma_s(x)=x$ for $s\le0$, with $\beta$ as in \cref{ex:weakfail}. This is smooth in $(s,x)$ because for $|x|<R$ and $0<s<1/(R+1)$ the
bump is not seen. $F_0$ is an embedding and $F_s$ is not injective for small $s>0$
(\cref{ex:weakfail}), so the set of parameters giving embeddings is not open at $s=0$.

\sol{3.4} Reflexivity: the constant plot $\R\to\pt\to X$. Symmetry: compose with $t\mapsto1-t$. Transitivity:
let $p,q\colon\R\to X$ be plots with $p(1)=q(0)$. Choose a smooth non-decreasing $\lambda\colon\R\to\R$ with
$\lambda(t)=0$ for $t\le\frac13$ and $\lambda(t)=1$ for $t\ge\frac23$. Then $p\lambda$ is a plot, constant
equal to $p(1)$ on $(\frac23,\infty)$, and $q\lambda(\cdot-1)$ is a plot, constant equal to $q(0)$ on
$(-\infty,\frac43)$. On the overlap $(\frac23,\frac43)$ of the open sets $(-\infty,\frac43)$ and
$(\frac23,\infty)$, which cover $\R$, the two plots restrict to the same constant plot. By the sheaf
condition (after pulling back along diffeomorphisms of the intervals with $\R$) they glue to a plot
$r\colon\R\to X$ with $r(0)=p(0)$ and $r(2)=q(1)$; reparametrize by $t\mapsto2t$.

\sol{3.5} Dense: given $f$ and a weak neighbourhood, which constrains $f$ on some $[-R,R]$, the functions
$f_a(x)=f(x)+ax$ are Morse for almost every $a$, because $f_a$ is Morse exactly when $-a$ is a regular value
of $f'$, and Sard's theorem applies to $f'$; and $f_a\to f$ on $[-R,R]$ in every $C^r$ norm as $a\to0$. Not
open: $f(x)=x$ is Morse. With $\beta$ as in \cref{ex:weakfail}, let $g_n(x)=x-\beta(x-n)$. At the point
$x^*$ where $\beta'$ attains its maximum $1$ we have $\beta''(x^*)=0$, so
$g_n'(n+x^*)=0$ and $g_n''(n+x^*)=0$: a degenerate critical point. And $g_n\to f$ weakly.
\section*{Chapter 4}

\sol{4.1} Let $F\colon\R^n\times M\to\R^n$, $F(v,x)=f(x)+v$. Its differential in the $v$-direction is the
identity, so $F$ is a submersion and in particular $F\trans A$. By \cref{thm:parametric}, $F_v=f+v$ is
transverse to $A$ for almost every $v$. If $M$ and $A$ are curves in $\R^3$ then
$\dim M+\dim A<3$, and a map of a $1$-manifold transverse to a $1$-dimensional submanifold of $\R^3$ cannot
meet it: at a common point the images of the two tangent spaces would have to span $\R^3$.

\sol{4.2} Let $X_0$ have rank $\rho$. After multiplying on both sides by invertible matrices, which are
diffeomorphisms of $\R^{m\times n}$ preserving rank, we may take $X_0=\bigl(\begin{smallmatrix}I_\rho&0\\0&0
\end{smallmatrix}\bigr)$. Write a nearby matrix in blocks
$X=\bigl(\begin{smallmatrix}P&Q\\R&S\end{smallmatrix}\bigr)$ with $P$ invertible. Row and column operations
give $\rk X=\rho+\rk(S-RP^{-1}Q)$. So near $X_0$ the set of rank-$\rho$ matrices is the zero set of
$\Phi(X)=S-RP^{-1}Q$, a map to $\R^{(m-\rho)\times(n-\rho)}$ whose differential in the $S$-direction is the
identity. It is a submersion, and the zero set is a submanifold of codimension $(m-\rho)(n-\rho)$.

\sol{4.3} Resolve $\R$ as a $C^\infty(\R)$-module by the Koszul complex
$C^\infty(\R)\xrightarrow{x}C^\infty(\R)$; it is exact on the left because $x$ is not a zero divisor, and the
cokernel is $\R$ by Hadamard's lemma. This resolution is also a cofibrant model as a simplicial
$C^\infty$-ring (it is free on one generator in degree $1$ over $C^\infty(\R)$). Tensoring with $\R$ along
evaluation at $0$ gives $\R\xrightarrow{0}\R$: the homotopy groups are $\R$ in degrees $0$ and $1$, as in
\cref{ex:derived}(b). For the second, resolve $\R$ in the same way and tensor with $C^\infty(\R)$ along
$x\mapsto x^2$: the complex is $C^\infty(\R)\xrightarrow{x^2}C^\infty(\R)$. If $x^2g=0$ then $g=0$ away from
$0$ and so everywhere; the map is injective, $\pi_1=0$, and
$\pi_0=C^\infty(\R)/(x^2)=\R[\varepsilon]/(\varepsilon^2)$, as in \cref{ex:derived}(a).

\sol{4.4} The critical points of $f_a$ are the solutions of $\nabla f(x)=-a$, and the Hessian of $f_a$ is
that of $f$, which is the Jacobian of the map $\nabla f\colon U\to\R^m$. So $f_a$ is Morse if and only if $-a$
is a regular value of $\nabla f$. By \cref{thm:sard}, applied to the smooth map $\nabla f$ between manifolds
of the same dimension, almost every $a$ qualifies.

\sol{4.5} Suppose $g\colon D^n\to D^n$ is continuous without fixed points. Then $|g(x)-x|\ge2\delta>0$ by
compactness. By \cref{thm:approx} there is a smooth $h\colon D^n\to\R^n$ with $|h-g|<\delta$; then
$h_1=h/(1+\delta)$ maps $D^n$ to itself, is smooth, and has $|h_1-g|<2\delta$, so $h_1$ has no fixed point.
Let $\rho(x)$ be the point where the ray from $h_1(x)$ through $x$ leaves $D^n$:
$\rho(x)=x+t(x)\,(x-h_1(x))$ with $t(x)\ge0$ the larger root of $|x+t(x-h_1(x))|^2=1$. The discriminant of this
quadratic is positive because $x\ne h_1(x)$ and $|x|\le1$, so $t$ is smooth, and $t=0$ when $|x|=1$. Thus
$\rho$ is a smooth retraction of $D^n$ onto $S^{n-1}$, contradicting \cref{cor:noretract}.
\section*{Chapter 5}

\sol{5.1} The fibre over the line $\ell_\theta$ spanned by $u(\theta)=(\cos\theta,\sin\theta)$ is $\ell_\theta$
itself. A nowhere-zero section would be $s(\theta)=\lambda(\theta)u(\theta)$ with $\lambda$ continuous and
non-zero on $[0,\pi]$, and since $u(\pi)=-u(0)$ while $s(\pi)=s(0)$ we need $\lambda(\pi)=-\lambda(0)$, so
$\lambda$ changes sign; contradiction. The total space is $\{(\theta,t)\}=[0,\pi]\times\R$ with $(0,t)$
glued to $(\pi,-t)$, the open Möbius band.

\sol{5.2} Since $e$ has constant rank, near any point one can choose $k$ columns of $e$ that are independent;
they are smooth sections of the image spanning it, so the image $\xi$ is a subbundle of $M\times\R^s$, with
complement the image of $1-e$. A section of $M\times\R^s$ is an element $v\in C^\infty(M)^s$; it lies in $\xi$
if and only if $ev=v$, so $\Gamma(\xi)=e\,C^\infty(M)^s$, and
$C^\infty(M)^s=e\,C^\infty(M)^s\oplus(1-e)\,C^\infty(M)^s$.

\sol{5.3} $TS^n=\{(x,v):|x|=1,\ \langle x,v\rangle=0\}$ and the normal bundle of $S^n\subset\R^{n+1}$ is
trivialized by the section $x\mapsto x$. So $TS^n\oplus\R\cong TS^n\oplus\nu=T\R^{n+1}|_{S^n}$, which is
trivial. The map $B\Or(n)\to B\Or(n+1)$ classifies $\xi\mapsto\xi\oplus\R$, so the composite
$S^n\to B\Or(n)\to B\Or(n+1)$ classifies a trivial bundle and is null-homotopic by \cref{thm:classify}; hence
so is the composite to $B\Or$.

\sol{5.4} If $\omega_1-\omega_0=dh$, let $\lambda\colon\R\to[0,1]$ be smooth with $\lambda(0)=0$,
$\lambda(1)=1$, and put $\Omega=\omega_0+d(\lambda(t)h)$ on $M\times\R$. It is closed and restricts to
$\omega_0$ and $\omega_1$. Conversely, let $\Omega=\alpha_t+g_t\,dt$ be closed on $M\times\R$, with
$\alpha_t$ a path of $1$-forms on $M$ and $g_t$ of functions. The $dt$-component of $d\Omega=0$ reads
$\partial_t\alpha_t=d_Mg_t$, so $\alpha_1-\alpha_0=d\int_0^1g_t\,dt$.

\sol{5.5} A line bundle has transition functions in $\GL_1(\R)$, homotopy equivalent to $\Z/2$, so
$B\Or(1)=\RP^\infty=K(\Z/2,1)$ and $[M,K(\Z/2,1)]=H^1(M;\Z/2)=\Hom(\pi_1M,\Z/2)$ for connected $M$. For the
torus $\pi_1=\Z^2$ and there are $4$ homomorphisms. For the Klein bottle
$\pi_1=\langle a,b\mid abab^{-1}\rangle$ has abelianization $\Z\oplus\Z/2$, and again
$\Hom(-,\Z/2)=(\Z/2)^2$ has $4$ elements. The determinant of the tangent bundle is trivial for the torus. For
the Klein bottle, presented as the square with $a$ the pair of sides glued by a reflection and $b$ the pair
glued by a translation, the orientation reverses along the loop that crosses the reflected sides, so
$\det T$ is the line bundle with $w_1$ non-zero on that loop and zero on the other.
\section*{Chapter 6}

\sol{6.1} If $g_0,g_1$ are positive definite symmetric bilinear forms then so is $(1-t)g_0+tg_1$, since
$(1-t)g_0(v,v)+tg_1(v,v)>0$ for $v\ne0$. Every $A\in\GL_k(\R)$ has a unique polar decomposition $A=QP$ with $Q$
orthogonal and $P=(A^{\mathsf T}A)^{1/2}$ positive definite symmetric, depending continuously on $A$. The
path $A_t=Q\bigl((1-t)P+tI\bigr)$ stays in $\GL_k$ by convexity, fixes $\Or(k)$, and ends in $\Or(k)$; so
$\Or(k)$ is a deformation retract. A homotopy equivalence of topological groups that is a homomorphism induces
an equivalence of classifying spaces.

\sol{6.2} Let $e_0,e_1\colon M\to\R^q\subset\R^\infty$ be embeddings. Let $S,S'\colon\R^\infty\to\R^\infty$ be the
linear injections placing a vector in the odd, respectively even, coordinates. The linear map
$(1-t)\id+tS$ is injective for $t\in[0,1]$: an element of its kernel would be an eigenvector of $S$ with
eigenvalue $-(1-t)/t\le0$, and $S$ has none (if $Sx=\lambda x$ with $\lambda\ne0,1$ then
$x_2=0$, $x_3=x_2/\lambda=0$, and so on, and $x_1=\lambda x_1$ forces $x_1=0$; and $S$ is injective). An
injective linear map composed with an embedding of a compact manifold is an embedding, so
$(1-t)e_1+tSe_1$ is an isotopy from $e_1$ to $Se_1$; similarly from $e_0$ to $S'e_0$. Finally
$(1-t)Se_1+tS'e_0$ is an embedding for every $t$: for $t<1$ the odd coordinates are $(1-t)e_1$, an injective
immersion, and for $t>0$ the even coordinates are $te_0$. For non-compact $M$ one uses closed embeddings and
the same formulas. The naive homotopy can fail: for $e_1=-e_0$ the path $(1-2t)e_0$ is constant at $t=\frac12$.

\sol{6.3} Parametrize by arc length, with $\gamma'=T$ and $n'=-\kappa T$ (Frenet, with $n$ the inward normal
and $\kappa>0$ for a convex curve traversed counterclockwise). Then $\partial_s=(1-t\kappa)T$ and
$\partial_t=n$, so the determinant is $1-t\kappa(s)$ up to the sign of the frame. The map is a local
diffeomorphism where $t\kappa(s)\ne1$, hence for $|t|<1/\kappa_{\max}$. For the ellipse
$(a\cos u,b\sin u)$ the curvature is $ab/(a^2\sin^2u+b^2\cos^2u)^{3/2}$, with maximum $a/b^2$ at the ends of
the major axis. So $r=b^2/a$. (Global injectivity may fail earlier in general; for the ellipse it does
not on the inward side before $t=b^2/a$, and it never fails on the outward side.)

\sol{6.4} The one-point compactification of a product of locally compact spaces is the smash product of the
compactifications, which gives the first identity; the second and third are the cases $\eta=\R^j$ over a
point, where $\Th=S^j$, and $\xi=0$, where $\Th=M_+$. For the last, $\RP^{n+1}\setminus\{[0:\dots:0:1]\}$ is
the total space of the tautological bundle over $\RP^n$ by $[x:t]\mapsto([x],\ t\,x/|x|^2)$, where
$t\,x/|x|^2$ is a vector in the line $[x]$ that is unchanged by $(x,t)\mapsto(\lambda x,\lambda t)$; this is
a diffeomorphism, and the remaining point is the point at infinity.

\sol{6.5} Extend $f$ radially: $F(rx)=rf(x)$ for $x\in S^{n-1}$, $0\le r\le1$. This is a homeomorphism of
$D^n$ (the Alexander trick), smooth away from the origin. The map $\Sigma_f=D^n\cup_fD^n\to D^n\cup_{\id}D^n=S^n$
that is the identity on the first disc and $F$ on the second is well defined and bijective, continuous, and
hence a homeomorphism. If $f$ extends to a diffeomorphism $G$ of $D^n$, the same construction with $G$ in
place of $F$ is smooth on each disc and compatible with the gluing; it is a diffeomorphism for the smooth
structures of \cref{thm:gluing} after adjusting $G$ near the boundary to respect collars, which is possible by
uniqueness of collars (\cref{thm:collar}).
\section*{Chapter 7}

\sol{7.1} For $k\ne0$ every point of $S^1$ is a regular value of $z\mapsto z^k$, with $|k|$ preimages, at each
of which the derivative $\theta\mapsto k\theta$ has the sign of $k$; so the degree is $k$. (For $k=0$ the map
is constant.) The antipodal map is the restriction of $-1$ on $\R^{n+1}$, a composite of $n+1$ reflections in
coordinate hyperplanes, each of which restricts to a diffeomorphism of $S^n$ of degree $-1$ (one preimage,
orientation reversed). So its degree is $(-1)^{n+1}$. If $v$ is a nowhere-zero vector field on $S^n$,
normalized to unit length, then $H(x,t)=\cos(\pi t)\,x+\sin(\pi t)\,v(x)$ is a homotopy on $S^n$ from the
identity to the antipodal map, so $1=(-1)^{n+1}$ and $n$ is odd.

\sol{7.2} A point $[x]$ is fixed if and only if $(A-1)x\in\Z^2$. The map $A-1$ is an injective endomorphism of
$\Z^2$ with image of index $|\det(A-1)|$, so the solutions $x\in(A-1)^{-1}\Z^2$ form a lattice containing
$\Z^2$ with index $|\det(A-1)|$; modulo $\Z^2$ there are that many. The differential of the torus map at any
point is $A$, so each fixed point is non-degenerate with index $\operatorname{sign}\det(1-A)$. The sum is
$\det(1-A)=1-\tr A+\det A$, which is $L(A)$ since $A$ acts on $H_1=\Z^2$ by $A$ and on $H_2=\Lambda^2H_1$ by
$\det A$.

\sol{7.3} Cyclicity: by the triangle identities, the dual map $g^\vee\colon Y^\vee\to X^\vee$ of
$g\colon X\to Y$ satisfies $(g\otimes1)\eta_X=(1\otimes g^\vee)\eta_Y$ and
$\epsilon_Y(1\otimes g)=\epsilon_X(g^\vee\otimes1)$. For $f\colon Y\to X$ one slides $g$ around the loop:
$\tr(gf)=\epsilon_Y\sigma(gf\otimes1)\eta_Y=\epsilon_Y\sigma(g\otimes1)(f\otimes1)\eta_Y$, and moving $g$
through $\epsilon$ as $g^\vee$ and then through $\eta$ back as $g$ on the other side yields
$\epsilon_X\sigma(fg\otimes1)\eta_X=\tr(fg)$. Multiplicativity: $X\otimes Y$ is dualizable with dual
$Y^\vee\otimes X^\vee$ and structure maps built from those of $X$ and $Y$; the trace diagram for $f\otimes g$
is the tensor product of the two trace diagrams after rearranging by symmetries, and
$\operatorname{End}(\mathbf 1)$ is commutative. For a graded vector space $V=\bigoplus V_i$ the coevaluation
is $1\mapsto\sum e_j\otimes e^j$ over a homogeneous basis, and the symmetry sends $e_j\otimes e^j$ to
$(-1)^{|e_j|\,|e^j|}e^j\otimes e_j=(-1)^{|e_j|}e^j\otimes e_j$, since $|e^j|=-|e_j|$. Evaluating gives
$\tr(f)=\sum_j(-1)^{|e_j|}\langle e^j,fe_j\rangle=\sum_i(-1)^i\tr(f|_{V_i})$.

\sol{7.4} On $S^n\subset\R^{n+1}$ let $v(x)=e_{n+1}-\langle x,e_{n+1}\rangle x$, the projection of the constant
field $e_{n+1}$. It vanishes at the poles $\pm e_{n+1}$. Near the south pole, in the chart
$y\mapsto(y,-\sqrt{1-|y|^2})$, $v$ is $y\mapsto y$ to first order, of index $+1$. Near the north pole it is
$y\mapsto-y$, of index $(-1)^n$. So $\chi(S^n)=1+(-1)^n$. Homologically $H_i(S^n)$ is $\mathbb Q$ for $i=0,n$
and zero otherwise, giving the same.

\sol{7.5} The disc $D=\{x^2+y^2\le1,\ z=0\}$ is bounded by $K$. The circle $L$ is
$\{(1+\cos t,0,\sin t)\}$; it meets the plane $z=0$ at $t=0$, the point $(2,0,0)$ outside $D$, and at
$t=\pi$, the point $(0,0,0)$ inside $D$. So $L$ meets $D$ once, transversally (its tangent there is
$(0,0,-1)$), and the linking number is $\pm1$, the sign depending on the orientations: with $D$ oriented by
the normal $e_3$ and $L$ by increasing $t$, it is $-1$.
\section*{Chapter 8}

\sol{8.1} $f_\theta=-(R+r\cos\varphi)\sin\theta$ and $f_\varphi=-r\sin\varphi\cos\theta$. Since
$R+r\cos\varphi>0$, $f_\theta=0$ forces $\sin\theta=0$; then $\cos\theta=\pm1$ and $f_\varphi=0$ forces
$\sin\varphi=0$. The second derivatives are $f_{\theta\theta}=-(R+r\cos\varphi)\cos\theta$,
$f_{\varphi\varphi}=-r\cos\varphi\cos\theta$, $f_{\theta\varphi}=r\sin\varphi\sin\theta=0$ at the critical
points. At $(0,0)$: $(-(R+r),-r)$, index $2$. At $(0,\pi)$: $(-(R-r),+r)$, index $1$. At $(\pi,\pi)$:
$(R-r,-r)$, index $1$. At $(\pi,0)$: $(R+r,r)$, index $0$. All are non-degenerate because $R>r>0$.

\sol{8.2} Write $M(t)=\sum\nu_kt^k$, $P(t)=\sum\beta_kt^k$, so $M-P=(1+t)Q$. Dividing by $1+t$ as a formal
power series, $Q(t)=(M(t)-P(t))\sum_j(-1)^jt^j$, whose coefficient of $t^k$ is
$\sum_{i\le k}(-1)^{k-i}(\nu_i-\beta_i)$. Non-negativity of the coefficients of $Q$ is exactly the stated
inequality. For a closed connected orientable surface of genus $g$, with $F=\mathbb Q$, $\beta=(1,2g,1)$, so
$\sum\nu_k\ge\sum\beta_k=2g+2$. For a non-orientable surface $N_h$ use $F=\Z/2$: $\beta=(1,h,1)$ and the
bound is $h+2$.

\sol{8.3} In the chart $x_i=1$ with coordinates $(x_j)_{j\ne i}$,
$f=\bigl(\lambda_i+\sum_{j\ne i}\lambda_jx_j^2\bigr)/\bigl(1+\sum_{j\ne i}x_j^2\bigr)
=\lambda_i+\sum_{j\ne i}(\lambda_j-\lambda_i)x_j^2+O(|x|^4)$. So the origin of each chart is a non-degenerate
critical point of index $\#\{j:\lambda_j<\lambda_i\}=i$. There are no others: a critical point of $f$ on
$\RP^n$ lifts to a critical point of the quadratic form on $S^n$, an eigenvector of
$\operatorname{diag}(\lambda_i)$, and the eigenvalues are distinct. So $\nu_k=1$ for $0\le k\le n$. With
$F=\Z/2$, $\beta_k=1$ for all $k$ and $Q=0$. With $F=\mathbb Q$, $\beta_0=1$, $\beta_n=1$ if $n$ is odd, and
the others vanish; for $n$ even $M-P=t+\dots+t^n=(1+t)(t+t^3+\dots+t^{n-1})$ and for $n$ odd
$M-P=t+\dots+t^{n-1}=(1+t)(t+t^3+\dots+t^{n-2})$.

\sol{8.4} Define $\mu(x,y)=\min\bigl(|y|,\sqrt{\max(|x|^2-\varepsilon,0)}\bigr)$, a continuous function, and
let $r_t(x,y)=(x,\ y_t)$ where $y_t=y$ if $y=0$ and otherwise
$y_t=\bigl((1-t)|y|+t\mu(x,y)\bigr)\,y/|y|$. Then $r_t$ is continuous, $r_0=\id$, and $|y_t|\le|y|$, so
$q(r_t(x,y))\le q(x,y)\le\varepsilon$ and the deformation stays in $\{q\le\varepsilon\}$. At $t=1$:
if $|x|^2\le\varepsilon$ then $\mu=0$ and $r_1(x,y)=(x,0)$, a point of the disc; if $|x|^2>\varepsilon$
then $|y_1|^2=\min(|y|^2,|x|^2-\varepsilon)$ and $q(r_1(x,y))\le-\varepsilon$. Points of the target are
fixed for all $t$: on the disc $y=0$, and on $\{q\le-\varepsilon\}$ we have $|y|^2\le|x|^2-\varepsilon$ so
$\mu=|y|$.

\sol{8.5} $f_t'=3x^2-t$ and $f_t''=6x$. For $t<0$ there are no critical points. For $t=0$ there is one, at
$0$, degenerate. For $t>0$ there are two: $x=-\sqrt{t/3}$ with $f''<0$, a local maximum (index $1$) with
value $\frac{2t}{3}\sqrt{t/3}$, and $x=+\sqrt{t/3}$ with $f''>0$, a local minimum (index $0$) with value
$-\frac{2t}{3}\sqrt{t/3}$. On $[-L,L]$ with $L$ large, $f_t$ is increasing near the ends. For $t<0$ the
sublevel sets are intervals $[-L,c']$, all contractible: the functor is constant after the minimum
$f_t(-L)$. For $t>0$ a second component is born at the level of the local minimum and merges with the first
at the level of the local maximum: the degree-zero barcode acquires one finite interval, of length
$\frac{4t}{3}\sqrt{t/3}$, which shrinks to nothing as $t\to0^+$.
\section*{Chapter 9}

\sol{9.1} Let $DW=W\cup_{\partial W}W$ be the double, a closed manifold of odd dimension, so $\chi(DW)=0$
(\cref{thm:morseineq}). By additivity, $\chi(DW)=2\chi(W)-\chi(\partial W)$. Hence $\chi(\partial W)=2\chi(W)$
is even. If $\dim W$ is even, $\partial W$ is closed of odd dimension and $\chi(\partial W)=0$. So $\chi$ mod
$2$ vanishes on boundaries; it is additive under disjoint union and multiplicative under products
(\cref{thm:euler}), hence a ring homomorphism on $\mathfrak N_*$. $\chi(\RP^{2m})=1$, as $\RP^{2m}$ has one
cell in each dimension $0,\dots,2m$.

\sol{9.2} The allowed parts are $2,4,5,6,8,9,10,\dots$ (excluding $1,3,7$). Partitions: $n=0$: the empty one;
$n=1$: none; $n=2$: $2$; $n=3$: none; $n=4$: $4$, $2{+}2$; $n=5$: $5$; $n=6$: $6$, $4{+}2$, $2{+}2{+}2$;
$n=7$: $5{+}2$; $n=8$: $8$, $6{+}2$, $4{+}4$, $4{+}2{+}2$, $2{+}2{+}2{+}2$; $n=9$: $9$, $5{+}4$,
$5{+}2{+}2$; $n=10$: $10$, $8{+}2$, $6{+}4$, $6{+}2{+}2$, $5{+}5$, $4{+}4{+}2$, $4{+}2{+}2{+}2$,
$2{+}2{+}2{+}2{+}2$. The counts are $1,0,1,0,2,1,3,1,5,3,8$.

\sol{9.3} $w(\RP^n)=(1+a)^{n+1}$ with $a^{n+1}=0$, coefficients modulo $2$. $n=2$: $(1+a)^3=1+a+a^2$, so
$w_1=a$, $w_2=a^2$, and $w_1^2=w_2=a^2$ evaluates to $1$. $n=3$: $(1+a)^4=1+a^4=1$. $n=4$:
$(1+a)^5=1+a+a^4$ since $\binom52,\binom53$ are even; so $w_1=a$, $w_2=w_3=0$, $w_4=a^4$, giving
$(w_1^4,w_1^2w_2,w_2^2,w_1w_3,w_4)=(1,0,0,0,1)$. For $\RP^2\times\RP^2$,
$w=(1+a+a^2)(1+b+b^2)$: $w_1=a+b$, $w_2=a^2+ab+b^2$, $w_3=a^2b+ab^2$, $w_4=a^2b^2$, and the numbers are the
coefficients of $a^2b^2$ modulo $2$: $w_1^4=(a+b)^4$ gives $6\equiv0$; $w_1^2w_2$ gives $1+2+1\equiv0$;
$w_2^2$ gives $2+1\equiv1$; $w_1w_3$ gives $2\equiv0$; $w_4$ gives $1$.

\sol{9.4} A closed $1$-manifold is a finite union of circles, each the boundary of a disc. The Klein bottle
is the boundary of the solid Klein bottle, the non-trivial $D^2$-bundle over $S^1$ (the mapping torus of a
reflection of $D^2$). By \cref{thm:classification} a closed connected surface is $\Sigma_g$ or $N_h$.
$\Sigma_g$ bounds a handlebody. $N_h$ is a connected sum of $h$ copies of $\RP^2$, and connected sum is
cobordant to disjoint union (attach a $1$-handle to the product cobordism), so $[N_h]=h[\RP^2]$; this is zero
for $h$ even since every element has order $2$, and equal to $[\RP^2]\ne0$ for $h$ odd by \cref{exer:chimod2}.

\sol{9.5} A tubular neighbourhood of the origin in $\R^k$ with the standard framing is the open unit ball
$B$ with the identification $B\cong\R^k$, $x\mapsto x/(1-|x|^2)$, which is the identity to first order at $0$.
The collapse $S^k=\R^k\cup\infty\to\R^k\cup\infty=S^k$ is this diffeomorphism on $B$ and constant at $\infty$
outside. The origin is a regular value with exactly one preimage, at which the differential is the
identity; so the degree is $1$. Reversing one framing vector composes the identification with a reflection,
and the degree becomes $-1$. A framed $0$-manifold in $\R^k$ is a finite set of points with signs, the
collapse has degree the sum of the signs, and by \cref{thm:hopf} the degree classifies maps $S^k\to S^k$; so
$\Omega^{\mathrm{fr}}_0\cong\pi_k(S^k)\cong\Z$.
\section*{Chapter 10}

\sol{10.1} The equation $\dot x=(-x_2,x_1)$ is linear with solution
$F_t(x)=(x_1\cos t-x_2\sin t,\ x_1\sin t+x_2\cos t)$, rotation by angle $t$. For $F_t(x)=e^tx$,
$\partial_tF_t(x)=e^tx=F_t(x)$, so $G(y,t)=y$. For $\dot x=x^2$, the solution through $x_0>0$ is
$x_0/(1-x_0t)$, which is unbounded as $t\to1/x_0$; so for every $t>0$ the time-$t$ map is undefined at
$x_0\ge1/t$, and there is no diffeotopy.

\sol{10.2} The rotation number is the degree of $t\mapsto\gamma'(t)/|\gamma'(t)|$. Circle: the tangent turns
once; $1$. Figure eight: $\gamma'=(\cos t,\cos2t)$; the curve has the symmetry $t\mapsto t+\pi$,
$\gamma\mapsto(-x,y)$, under which the contributions of the two lobes to the total turning cancel; $0$.
Limaçon: in complex notation $\gamma=(1+2\cos\theta)e^{i\theta}=e^{i\theta}+e^{2i\theta}+1$, so
$\gamma'=ie^{i\theta}(1+2e^{i\theta})$; the factor $e^{i\theta}$ has winding number $1$ and
$1+2e^{i\theta}$ winds once around $0$ since $|2e^{i\theta}|>1$; total $2$. By \cref{cor:WG} no two of the
three are regularly homotopic.

\sol{10.3} A homotopy inverse sends $(p,A)$ to the embedding $x\mapsto p+Ax$. One composite is the identity.
For the other, given an embedding $f$ define, for $0<t\le1$,
$f_t(x)=f(0)+\bigl(f(tx)-f(0)\bigr)/t$, and $f_0(x)=f(0)+df_0(x)$. Each $f_t$ is an embedding (a rescaling of
the restriction of $f$ to a smaller disc), $f_1=f$, and by Taylor's theorem
$f_t(x)=f(0)+df_0(x)+t\,r(x,t)$ with $r$ smooth, so the family is smooth down to $t=0$ and depends
continuously on $f$. It preserves $(f(0),df_0)$ throughout. So the space of embeddings deformation retracts
to the affine ones.

\sol{10.4} Parametrize so that $F_t(x)=(x,0,0)$ unless $|x-1/(1-t)|<1$, where a fixed knotted arc is
inserted, for $t<1$; put $F_1(x)=(x,0,0)$. On the open set $\{(x,t):x<R\}$ we have $F_t(x)=(x,0,0)$ as soon as
$1/(1-t)>R+1$, so $F$ is smooth near every point with $t=1$; it is smooth elsewhere by construction, and each
$F_t$ is an embedding. If a diffeotopy $\Phi_t$ of $\R^3$ satisfied $\Phi_1\circ F_0=F_1$, then $\Phi_1$ would
be a diffeomorphism of $\R^3$ carrying the knotted line to the straight line, hence a homeomorphism of the
complements. The complement of the straight line has fundamental group $\Z$, and the complement of a long
knot has the group of the corresponding closed knot, which is not abelian if the knot is non-trivial.

\sol{10.5} Here $m=1$ and $n=4=2m+2$, so \cref{thm:whitneyiso} applies. In $\R^3$ the dimension count for
double points of a one-parameter family gives $2m+1-n=0$: a generic path of maps $S^1\to\R^3$ has isolated
times at which the curve crosses itself transversally, and these crossings are exactly the events that change
the knot type. They cannot be removed by perturbation because a zero-dimensional transverse intersection is
stable.
\section*{Chapter 11}

\sol{11.1} For a word of length $2n$, $\chi=V-n+1$, and the surface is orientable if and only if each letter
occurs with both exponents. $aba^{-1}b^{-1}$: all four corners are identified, $V=1$, $\chi=0$,
orientable: the torus. $abab$: $V=2$, $\chi=1$, non-orientable: $\RP^2$. $abab^{-1}$: $V=1$, $\chi=0$,
non-orientable: the Klein bottle. $aabb$: $V=1$, $\chi=0$, non-orientable: the Klein bottle again, as
$P\#P$. $abca^{-1}b^{-1}c^{-1}$: $V=2$, $\chi=2-3+1=0$, orientable: the torus. $abcabc$: $V=3$,
$\chi=3-3+1=1$, non-orientable: $\RP^2$. (The vertex counts are those returned by the script.)

\sol{11.2} $\chi(M\#N)=\chi(M)+\chi(N)-2$ because a disc is removed from each (lowering each $\chi$ by $1$)
and the results are glued along a circle, of Euler characteristic $0$. So $\chi(T\#P)=0+1-2=-1$ and
$\chi(P\#P\#P)=1+1+1-4=-1$. Both are closed, connected and non-orientable (a connected sum is orientable only
if both summands are). By \cref{thm:classification} they are diffeomorphic, to $N_3$.

\sol{11.3} (a) The pairing matrix in the basis $(1,x)$ is
$\bigl(\begin{smallmatrix}0&1\\1&0\end{smallmatrix}\bigr)$, so $1^\#=x$ and $x^\#=1$, and
$w=1\cdot x+x\cdot1=2x$. Then $\varepsilon(w^0)=\varepsilon(1)=0$, $\varepsilon(w)=2$, and $w^2=4x^2=0$.
(b) $\varepsilon(g\cdot h)=\delta_{gh,e}/n$, so $g^\#=n\,g^{-1}$ and $w=\sum_gg\cdot ng^{-1}=n^2e$; hence
$\varepsilon(w^g)=n^{2g}\varepsilon(e)=n^{2g-1}$. A homomorphism $\pi_1\Sigma_g\to G$ to an abelian group is a
homomorphism from $H_1=\Z^{2g}$, and there are $n^{2g}$ of them; dividing by $n$ gives the same number.
Semisimple case: $e_ie_j=\delta_{ij}e_i$, so the pairing is diagonal with entries $\theta_i$, the dual basis
is $e_i^\#=e_i/\theta_i$, and $w=\sum_ie_i/\theta_i$. Then $w^g=\sum_ie_i/\theta_i^g$ and
$\varepsilon(w^g)=\sum_i\theta_i^{1-g}$.

\sol{11.4} By \cref{cor:pushout}, $W$ is obtained up to homotopy from $f^{-1}(a)$, a union of circles with
$\chi=0$, by attaching a $1$-cell; so $\chi(W)=-1$. Both $f^{-1}(a)$ and $f^{-1}(b)$ are non-empty: the minimum
and maximum of $f$ are not attained at the only critical point, a saddle, so they are attained on the
boundary. Hence $W$ has $b\ge2$ boundary components. If $W$ is orientable, $-1=2-2g-b$ gives
$(g,b)=(0,3)$, as $(1,1)$ is excluded: a sphere with three holes. If not, $-1=2-h-b$ with $h\ge1$, $b\ge2$
gives $(h,b)=(1,2)$: a projective plane with two holes, which is a Möbius band with one hole.

\sol{11.5} The torus is the composite
$\emptyset\xrightarrow{\eta}S^1\xrightarrow{\delta}S^1\sqcup S^1\xrightarrow{\mu}S^1\xrightarrow{\epsilon}
\emptyset$. Let $\beta=\epsilon\mu$, the pairing, and $\gamma=\delta\eta$, the copairing. The Frobenius
relation and the unit and counit laws give the zigzag identity
$(\beta\sqcup\id)(\id\sqcup\gamma)=\id$: indeed
$(\epsilon\sqcup\id)(\mu\sqcup\id)(\id\sqcup\delta)(\id\sqcup\eta)
=(\epsilon\sqcup\id)\,\delta\,\mu\,(\id\sqcup\eta)=\id$. So $Z(\gamma)$ and $Z(\beta)$ exhibit $A=Z(S^1)$ as its
own dual, with $Z(\gamma)$ the coevaluation and $Z(\beta)$ the evaluation, and by commutativity
$Z(T^2)=Z(\beta)Z(\gamma)=\tr(\id_A)=\dim A$.
\backmatter
\chapter{References}
\begingroup\small
\renewcommand{\chapter}[2]{}
\begin{thebibliography}{ABGHR14}\setlength{\itemsep}{1pt}
\newcommand{\bi}[1]{\bibitem[#1]{#1}}
\bi{ABGHR14} M.~Ando, A.~J.~Blumberg, D.~Gepner, M.~J.~Hopkins, C.~Rezk, An $\infty$-categorical approach to $R$-line bundles, $R$-module Thom spectra, and twisted $R$-homology, \emph{J.~Topol.} 7 (2014), 869--893.
\bi{Abr96} L.~Abrams, Two-dimensional topological quantum field theories and Frobenius algebras, \emph{J.~Knot Theory Ramifications} 5 (1996), 569--587.
\bi{AF15} D.~Ayala, J.~Francis, Factorization homology of topological manifolds, \emph{J.~Topol.} 8 (2015), 1045--1084.
\bi{Ati61} M.~F.~Atiyah, Thom complexes, \emph{Proc.\ London Math.\ Soc.} (3) 11 (1961), 291--310.
\bi{BBC21} K.~Bauer, M.~Burke, M.~Ching, Tangent $\infty$-categories and Goodwillie calculus, preprint, arXiv:2101.07819.
\bi{BD95} J.~C.~Baez, J.~Dolan, Higher-dimensional algebra and topological quantum field theory, \emph{J.~Math.\ Phys.} 36 (1995), 6073--6105.
\bi{BEBP24} D.~Berwick-Evans, P.~Boavida de Brito, D.~Pavlov, Classifying spaces of infinity-sheaves, \emph{Algebr.\ Geom.\ Topol.} 24 (2024), 4891--4937.
\bi{BH11} J.~C.~Baez, A.~E.~Hoffnung, Convenient categories of smooth spaces, \emph{Trans.\ Amer.\ Math.\ Soc.} 363 (2011), 5789--5825.
\bi{BW13} P.~Boavida de Brito, M.~Weiss, Manifold calculus and homotopy sheaves, \emph{Homology Homotopy Appl.} 15 (2013), 361--383.
\bi{CC14} J.~R.~B.~Cockett, G.~S.~H.~Cruttwell, Differential structure, tangent structure, and SDG, \emph{Appl.\ Categ.\ Structures} 22 (2014), 331--417.
\bi{CCG20} R.~Cockett, G.~Cruttwell, J.~Gallagher, J.-S.~P.~Lemay, B.~MacAdam, G.~Plotkin, D.~Pronk, Reverse derivative categories, in \emph{Computer Science Logic 2020}, LIPIcs 152, 18:1--18:16.
\bi{Cer70} J.~Cerf, La stratification naturelle des espaces de fonctions diff\'erentiables r\'eelles et le th\'eor\`eme de la pseudo-isotopie, \emph{Publ.\ Math.\ IH\'ES} 39 (1970), 5--173.
\bi{CJS95} R.~L.~Cohen, J.~D.~S.~Jones, G.~B.~Segal, Morse theory and classifying spaces, preprint, 1995.
\bi{CS19} D.~Carchedi, P.~Steffens, On the universal property of derived manifolds, preprint, arXiv:1905.06195.
\bi{CSW14} J.~D.~Christensen, G.~Sinnamon, E.~Wu, The $D$-topology for diffeological spaces, \emph{Pacific J.~Math.} 272 (2014), 87--110.
\bi{DP80} A.~Dold, D.~Puppe, Duality, trace, and transfer, in \emph{Proceedings of the International Conference on Geometric Topology (Warsaw, 1978)}, PWN, 1980, 81--102.
\bi{Dub79} E.~J.~Dubuc, Sur les mod\`eles de la g\'eom\'etrie diff\'erentielle synth\'etique, \emph{Cahiers Topologie G\'eom.\ Diff\'erentielle} 20 (1979), 231--279.
\bi{EE69} C.~J.~Earle, J.~Eells, A fibre bundle description of Teichm\"uller theory, \emph{J.~Differential Geometry} 3 (1969), 19--43.
\bi{EM12} Y.~M.~Eliashberg, N.~M.~Mishachev, The space of framed functions is contractible, in \emph{Essays in Mathematics and its Applications}, Springer, 2012, 81--109.
\bi{FST19} B.~Fong, D.~I.~Spivak, R.~Tuy\'eras, Backprop as functor: a compositional perspective on supervised learning, in \emph{LICS 2019}, IEEE, 2019.
\bi{GK15} T.~G.~Goodwillie, J.~R.~Klein, Multiple disjunction for spaces of smooth embeddings, \emph{J.~Topol.} 8 (2015), 651--674.
\bi{GMTW09} S.~Galatius, I.~Madsen, U.~Tillmann, M.~Weiss, The homotopy type of the cobordism category, \emph{Acta Math.} 202 (2009), 195--239.
\bi{God07} V.~Godin, Higher string topology operations, preprint, arXiv:0711.4859.
\bi{GP21} D.~Grady, D.~Pavlov, The geometric cobordism hypothesis, preprint, arXiv:2111.01095.
\bi{GW99} T.~G.~Goodwillie, M.~Weiss, Embeddings from the point of view of immersion theory, Part~II, \emph{Geom.\ Topol.} 3 (1999), 103--118.
\bi{Hir59} M.~W.~Hirsch, Immersions of manifolds, \emph{Trans.\ Amer.\ Math.\ Soc.} 93 (1959), 242--276.
\bi{Hir76} M.~W.~Hirsch, \emph{Differential Topology}, Graduate Texts in Mathematics 33, Springer, 1976.
\bi{Igu87} K.~Igusa, The space of framed functions, \emph{Trans.\ Amer.\ Math.\ Soc.} 301 (1987), 431--477.
\bi{IZ13} P.~Iglesias-Zemmour, \emph{Diffeology}, Mathematical Surveys and Monographs 185, American Mathematical Society, 2013.
\bi{Joy19} D.~Joyce, \emph{Algebraic Geometry over $C^\infty$-rings}, Mem.\ Amer.\ Math.\ Soc.\ 260 (2019), no.~1256.
\bi{KM97} A.~Kriegl, P.~W.~Michor, \emph{The Convenient Setting of Global Analysis}, Mathematical Surveys and Monographs 53, American Mathematical Society, 1997.
\bi{KMS93} I.~Kol\'a\v{r}, P.~W.~Michor, J.~Slov\'ak, \emph{Natural Operations in Differential Geometry}, Springer, 1993.
\bi{Koc04} J.~Kock, \emph{Frobenius Algebras and 2D Topological Quantum Field Theories}, London Math.\ Soc.\ Student Texts 59, Cambridge University Press, 2004.
\bi{Koc06} A.~Kock, \emph{Synthetic Differential Geometry}, 2nd ed., London Math.\ Soc.\ Lecture Note Series 333, Cambridge University Press, 2006.
\bi{Leu17} P.~Leung, Classifying tangent structures using Weil algebras, \emph{Theory Appl.\ Categ.} 32 (2017), 286--337.
\bi{Lur09} J.~Lurie, On the classification of topological field theories, in \emph{Current Developments in Mathematics 2008}, International Press, 2009, 129--280.
\bi{Mac17} A.~W.~Macpherson, The universal property of derived geometry, preprint, arXiv:1701.08359.
\bi{May01} J.~P.~May, The additivity of traces in triangulated categories, \emph{Adv.\ Math.} 163 (2001), 34--73.
\bi{MR91} I.~Moerdijk, G.~E.~Reyes, \emph{Models for Smooth Infinitesimal Analysis}, Springer, 1991.
\bi{MW07} I.~Madsen, M.~Weiss, The stable moduli space of Riemann surfaces: Mumford's conjecture, \emph{Ann.\ of Math.} (2) 165 (2007), 843--941.
\bi{Pal60} R.~S.~Palais, Local triviality of the restriction map for embeddings, \emph{Comment.\ Math.\ Helv.} 34 (1960), 305--312.
\bi{Pon55} L.~S.~Pontryagin, \emph{Smooth manifolds and their applications in homotopy theory}, Trudy Mat.\ Inst.\ Steklov 45 (1955); English transl.\ in Amer.\ Math.\ Soc.\ Transl.\ (2) 11 (1959).
\bi{PS14} K.~Ponto, M.~Shulman, Traces in symmetric monoidal categories, \emph{Expo.\ Math.} 32 (2014), 248--273.
\bi{Ros84} J.~Rosick\'y, Abstract tangent functors, \emph{Diagrammes} 12 (1984), JR1--JR11.
\bi{Sch13} U.~Schreiber, Differential cohomology in a cohesive infinity-topos, preprint, arXiv:1310.7930.
\bi{Sma59} S.~Smale, The classification of immersions of spheres in Euclidean spaces, \emph{Ann.\ of Math.} (2) 69 (1959), 327--344.
\bi{SP09} C.~J.~Schommer-Pries, \emph{The classification of two-dimensional extended topological field theories}, Ph.D.\ thesis, University of California, Berkeley, 2009; arXiv:1112.1000.
\bi{Spi10} D.~I.~Spivak, Derived smooth manifolds, \emph{Duke Math.\ J.} 153 (2010), 55--128.
\bi{Tho54} R.~Thom, Quelques propri\'et\'es globales des vari\'et\'es diff\'erentiables, \emph{Comment.\ Math.\ Helv.} 28 (1954), 17--86.
\bi{Wei99} M.~Weiss, Embeddings from the point of view of immersion theory, Part~I, \emph{Geom.\ Topol.} 3 (1999), 67--101.
\bi{ZC05} A.~Zomorodian, G.~Carlsson, Computing persistent homology, \emph{Discrete Comput.\ Geom.} 33 (2005), 249--274.
\end{thebibliography}
\endgroup
\end{document}

End of manuscript · Back to top ↑