Algebraic and Differential Topology in Higher Categories
Chains and framed manifolds as the linear algebra of the integers and of the sphere
A treatment, in categorical language, of the two monographs collected in L. S. Pontryagin, Selected Works, Volume 3: Algebraic and Differential Topology (ed. R. V. Gamkrelidze, Gordon and Breach, 1986): the Foundations of Algebraic Topology (complexes, homology, invariance, Sperner's lemma, the fixed point theorem of Lefschetz and Hopf) and Smooth Manifolds and Their Applications in Homotopy Theory (general position, framed manifolds, the degree, the Hopf invariant, and the groups $\pi_{n+1}(S^n)$ and $\pi_{n+2}(S^n)$). It reproduces none of the source's text or figures; a concordance maps its sections to this volume.
Format. Book, 80 pages, 13 chapters in four parts and four appendices, 35 figures and diagrams (31 numbered figures, about half of them computed), 48 exercises with solutions to all of them, 48 references, status tags on every numbered statement (classical, literature, reformulation, observation, programme), and a companion script with 63 computational checks.
The organizing statement
The first monograph is linear algebra over the integers and the second is linear algebra over the sphere spectrum $\mathbb S$. Write $\mathcal S$ for the $\infty$-category of homotopy types. Chains are the free $\mathbb Z$-module on a homotopy type, the suspension spectrum is the free $\mathbb S$-module, and the first is obtained from the second by extension of scalars along the Hurewicz map $\mathbb S\to H\mathbb Z$:
$$\begin{CD} \mathcal S @>{\Sigma^\infty_+}>> \mathrm{Mod}_{\mathbb S} \\ @V{\mathrm{id}}VV @VV{H\mathbb Z\otimes_{\mathbb S}-}V \\ \mathcal S @>>{C_*}> \mathrm{Mod}_{H\mathbb Z} \end{CD}$$
Both horizontal functors preserve colimits and are determined by their value on a point. The cycles of the lower theory are simplicial chains; those of the upper one are framed manifolds, by Pontryagin's theorem that $\pi_{n+k}(S^n)$ is the group of framed $k$-manifolds in $\mathbb R^{n+k}$ up to cobordism. The coefficients of the lower theory are $\mathbb Z$ in degree zero and nothing else. Those of the upper one are the stable stems $\mathbb Z,\ \mathbb Z/2,\ \mathbb Z/2,\ \mathbb Z/24,\dots$, and the map between them is an isomorphism in degree zero and zero above. That is why the first monograph is easy and the second is hard: the source computes the first two stems, Rokhlin reached the third by the same method, and there the method stopped.
Results: qualitative statement, then its quantitative form
| Topic | Qualitative | Quantitative (recomputed by the script) |
|---|---|---|
| Complexes | a polyhedron is a colimit of simplices; complexes embed in simplicial sets | model surfaces with $f$-vectors $(7,21,14)$ for the torus and $(6,15,10)$ for the projective plane; Euler characteristics $2,2,0,1,0,0$ |
| Homology | chains are a left adjoint. Integral homology determines the torsion coefficients, and mod-$p$ homology detects $p$-torsion by comparison with rational Betti numbers | Smith normal form: projective plane $\mathbb Z,\mathbb Z/2,0$; Klein bottle $\mathbb Z,\mathbb Z\oplus\mathbb Z/2,0$ |
| Invariance | subdivision is a functor with a homotopy inverse; the cone is a contraction | subdivided projective plane $(31,90,60)$ with the same homology; mesh $1.414, 0.745, 0.478, 0.283, 0.174$ |
| Sperner, Brouwer | a degree that cannot vanish | $17$ to $47$ complete cells, always odd, in $200$ labellings; displacement $0.059, 0.014, 0.0026$ on grids of side $8, 32, 128$; one winner in each of $300$ Hex boards |
| Homotopy | homotopies are $2$-cells; chains factor through homotopy types | prism identity $\partial P+P\partial=\mathrm{top}-\mathrm{bottom}$ on four complexes |
| Fixed points | the Lefschetz number is a trace of a dualizable object, in $\pi_0\mathbb S=\mathbb Z$ | Lefschetz numbers $0,1,3,4$ for $42$ automorphisms of the seven-vertex torus; periodic points $1,5,16,45,121,320,841,2205$ for the cat map |
| General position | transversality is generic | critical values of $x^3-3x$ near the critical set: length $/\delta^2=0.16855, 0.16685, 0.16669\to\frac16$; a trefoil projects with $3$ to $6$ double points |
| Degree and linking | a map to a sphere is a framed manifold | Gauss integrals $1, 0, 2$; Hopf invariants $1,4,9$ on one side and $1,2,3$ on the other |
| First stem | a framing relative to a reference gives a loop of rotations, and $\pi_1(\mathrm{SO}(n))=\mathbb Z/2$ for $n\ge3$ | turning numbers $0,1,2,3$ give classes $0,1,0,1$; a full turn lifts to a path from $1$ to $-1$ in the quaternions |
| Second stem | the invariant is the Arf invariant of a quadratic form, which is not linear | $(3,1)$, $(10,6)$, $(36,28)$ forms of Arf invariant $0$ and $1$ in genus $1,2,3$; $\mathrm{Sp}(4,\mathbb F_2)$ of order $720$ with orbits $10$ and $6$ |
| Stems | framed spheres give the image of $J$ | orders $24, 240, 504, 480, 264, 65520, 24, 16320$ from Bernoulli numbers |
What the language explains
1. The fixed point theorem. The fixed point index is a trace in spectra and the Lefschetz number a trace in chain complexes; they agree because a symmetric monoidal functor preserves traces and the two units have the same $\pi_0$. 2. An error and its correction. The volume's introductory survey recalls that the maps $S^{n+2}\to S^n$ were first classified wrongly and that the result was corrected in 1950. In the present terms the invariant of a framed surface is a quadratic form that fails to be linear, because $\mathbb S$ is not a product of Eilenberg-MacLane spectra: its first $k$-invariant is $Sq^2$. The earlier result is what one gets by assuming that it is. 3. Where the method stops. The source says that beyond the third stem the problem "so far defies solution by our method". The obstruction is surgery: to compute a stem from manifolds one must simplify an arbitrary framed manifold, and the obstructions to doing so are as hard as the groups sought. The Kervaire invariant is one such obstruction. Lin, Wang and Xu prove existence of a framed manifold of Kervaire invariant one in dimension 126, the final dimension left by earlier results. Preprint, first posted in 2024 and revised in 2025.
Two points of contact with the source
No numerical example of the source was found to disagree with a computation. Two conventions are worth knowing. The source's invariant $\delta$ of a framed circle is the class of a loop of rotations plus one unit per component. The unit is there because for a circle with the frame it inherits from a spanning disc the loop is the non-trivial one; accounts that take the bounding frame as the reference absorb the shift. And the source's "homology of framed manifolds" is what is now called cobordism, not the homology of its first monograph.
How much is "higher"
More than in the earlier volumes of the series, and the closing chapter says where. Chapters 1 to 4 and 7 are ordinary category theory: presheaves, colimits, adjunctions, pullbacks. Homotopies as $2$-cells enter in chapter 5. From chapter 8 on the central object is the sphere spectrum. Its infinite loop space is the free grouplike commutative monoid in spaces on a point. The source represents its homotopy classes by framed manifolds and computes the first two stems using geometric invariants.
What is and is not claimed
No new theorem. The long proofs, among them the general position arguments on which the second monograph rests and the classification of framed surfaces up to cobordism, are the source's and are indicated, not reproduced. The script checks the invariants that detect the first three stems on examples; it does not and could not check that they are complete. The orders of the stems beyond the second are quoted from the literature.
Files
algebraic-and-differential-topology-in-higher-categories.pdf— the book (80 pages).algebraic-and-differential-topology-in-higher-categories.tex— complete LaTeX source as a single file, figures included as TikZ (pdflatex, three passes).verify-topology.py— 63 checks, Python standard library only, about twenty seconds; exits 0 if and only if all pass. Homology is computed exactly, by Smith normal form.
The companion paper Algebraic and Differential Topology in Machine Learning and AI: Obstructions, Invariants, and Counts applies the first monograph, and the rotation groups of the second, to neural networks.
Status and authorship
Contribution. An exposition: known mathematics organized around one statement, with solved exercises, computed figures and reproducible checks.
Limitations. As above. In addition: the source was read from the text layer of a scanned copy, which is imperfect in formulas, and not every page was read with equal care; the bibliography was assembled from memory except for two entries (the 2024 preprint on the Kervaire invariant and a 2021 preprint on the cobordism hypothesis) whose details were checked online on 19 November 2026; historical remarks are from memory; and nothing has been independently refereed. Corrections are welcome.
AI 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, figures and typesetting.