Indexed text attachment
differential-topology-in-higher-categories.tex
Source for Differential Topology in Higher Categories: Mapping Objects, Derived Intersections, Spaces of Choices, and Cobordism
View paper · Download original file
Lines 3351–3450 of 3851. Source text is displayed literally.
3351 3352\sol{1.3} $f(A)=A^{\mathsf T}A$ is polynomial, hence smooth, with values in the vector space $\cat{Sym}_n$ of3353dimension $\frac12n(n+1)$. Its differential at $A$ is $df_A(H)=A^{\mathsf T}H+H^{\mathsf T}A$. If3354$A^{\mathsf T}A=I$ and $S$ is symmetric, then $H=\frac12AS$ gives3355$\frac12A^{\mathsf T}AS+\frac12S^{\mathsf T}A^{\mathsf T}A=S$. So $I$ is a regular value and3356$\dim\Or(n)=n^2-\frac12n(n+1)=\frac12n(n-1)$. The tangent space at $I$ is $\ker df_I=\{H:H+H^{\mathsf T}=0\}$.3357 3358\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 the3359function it induces on points is $s\mapsto F(s,-)$, which determines $F$; so $[M,N]$ is concrete and its points3360are $\Hom(\pt\times M,N)=C^\infty(M,N)$. For the adjunction, when $S=\R^n$ the bijection is the definition.3361A general smooth set is a colimit of representables, $S=\colim_i\R^{n_i}$; the functor $-\times M$ preserves3362colimits because $\SmSet$ is a topos, so3363$\Hom(S\times M,N)=\lim_i\Hom(\R^{n_i}\times M,N)=\lim_i[M,N](\R^{n_i})=\Hom(S,[M,N])$.3364 3365\sol{1.5} Let $f(x)=g(y)=z$. The tangent space to $\Delta_N$ at $(z,z)$ is $\{(w,w)\}$. Transversality of3366$f\times g$ to $\Delta_N$ at $(x,y)$ says that every $(a,b)\in T_zN\oplus T_zN$ is3367$(df_xu+w,\,dg_yv+w)$ for some $u,v,w$. Subtracting, this holds if and only if every $a-b$ is of the form3368$df_xu-dg_yv$, that is, if and only if $df_x(T_xM)+dg_y(T_yP)=T_zN$: given such a decomposition of $a-b$, set3369$w=a-df_xu$.3370%% ---- end sol/01.tex ----3371%% ---- begin sol/02.tex ----3372\section*{Chapter 2}3373 3374\sol{2.1} For open subsets of Euclidean spaces,3375$T(U\times V)=U\times V\times\R^m\times\R^n\cong TU\times TV$ by reordering coordinates, and the reordering3376commutes with the maps $T(f\times g)$ and $Tf\times Tg$ since $D(f\times g)=Df\oplus Dg$. The isomorphism is3377therefore natural and glues over atlases. For the second statement, $Z=f^{-1}(z)$ is a submanifold with3378$T_xZ=\ker df_x$ by \cref{thm:transpull}. A vector $(x,v)\in TM$ satisfies $Tf(x,v)=0_z$ exactly when3379$f(x)=z$ and $df_xv=0$, that is, when $(x,v)\in TZ$. One checks that $0_z$ is a regular value of $Tf$: in3380coordinates $D(Tf)_{(x,v)}(\xi,\nu)=(Df_x\xi,\,D^2f_x(v,\xi)+Df_x\nu)$, which is onto when $Df_x$ is. So the3381equality is one of submanifolds.3382 3383\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 curve3384$t\mapsto x+vt+\frac12at^2$; a map $f$ acts by Taylor expansion, which is the composition of $2$-jets, so3385$T_AM$ is the bundle of $2$-jets of curves and its fibre has dimension $2n$. The map $x\mapsto3386\varepsilon+\eta$ is a homomorphism since $(\varepsilon+\eta)^3=3\varepsilon^2\eta+3\varepsilon\eta^2=0$,3387and it is injective: $1,\ \varepsilon+\eta,\ (\varepsilon+\eta)^2=2\varepsilon\eta$ are independent. On3388points it sends $x+vt+\frac12at^2$ to $x+v\varepsilon+v\eta+a\varepsilon\eta$, that is, to $(x,v,v,a)$ in3389the coordinates of \cref{tab:weil}. The fixed set of the flip is $\{v=w\}$, and $T(p)$ and $p_T$ are the two3390projections $(x,v,w,a)\mapsto(x,v)$ and $(x,w)$, so the condition $T(p)=p_T$ is again $v=w$. The image is3391exactly $\{(x,v,v,a)\}$, of fibre dimension $2n$, and the map is an embedding.3392 3393\sol{2.3} Write $f(x)=Ax+\frac12B(x,x)+\dots$ and $g(y)=Cy+\frac12E(y,y)+\dots$ with $B,E$ symmetric bilinear.3394Then $g(f(x))=CAx+\frac12\bigl(C\,B(x,x)+E(Ax,Ax)\bigr)+O(|x|^3)$, so3395\[(C,E)\circ(A,B)=\bigl(CA,\ C\circ B+E\circ(A\times A)\bigr).\]3396For a third jet $(F,G)$,3397$(F,G)\circ\bigl((C,E)\circ(A,B)\bigr)=\bigl(FCA,\ FCB+FE(A\times A)+G(CA\times CA)\bigr)$ and3398$\bigl((F,G)\circ(C,E)\bigr)\circ(A,B)=\bigl(FCA,\ FCB+(FE+G(C\times C))(A\times A)\bigr)$, which agree.3399 3400\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 homomorphism3401$\phi\colon A\to B$ acts on $A$-points by applying $\phi$ to the coefficients. Addition: a point of3402$T_{W\times_\R W}$ is $x+v_1\varepsilon_1+v_2\varepsilon_2$, and $\varepsilon_i\mapsto\varepsilon$ sends it to3403$x+(v_1+v_2)\varepsilon$. Vertical lift: $x+v\varepsilon\mapsto x+v\varepsilon\eta$, which is $(x,0,0,v)$.3404Flip: exchanging $\varepsilon$ and $\eta$ sends $(x,v,w,a)$ to $(x,w,v,a)$. Each map is natural because it3405is induced by a homomorphism (\cref{thm:weil}(iii)); directly, naturality of $\ell$ is the identity3406$f(x+v\varepsilon\eta)=f(x)+Df_x(v)\varepsilon\eta$, since $(\varepsilon\eta)^2=0$.3407 3408\sol{2.5} Let $(f,f^\sharp)$, $(g,g^\sharp)$, $(h,h^\sharp)$ be composable. Both bracketings have forward part3409$hgf$ and backward part3410$(x,u')\mapsto f^\sharp\bigl(x,g^\sharp(f(x),h^\sharp(gf(x),u'))\bigr)$, as one sees by unfolding the3411definition twice; so composition is associative. For $F=f_k\cdots f_1$ with scalar output, \cref{prop:reverse}3412gives $R[F](x,1)=R[f_1](x_0,R[f_2](x_1,\dots R[f_k](x_{k-1},1)))$ with $x_0=x$ and $x_i=f_i(x_{i-1})$. The3413forward pass computes and stores $x_0,\dots,x_{k-1}$; the backward pass applies the transposed Jacobians3414$(Df_i)_{x_{i-1}}^{\mathsf T}$ to a single covector, for $i=k,\dots,1$. The result is the gradient of $F$ at3415$x$.3416%% ---- end sol/02.tex ----3417%% ---- begin sol/03.tex ----3418\section*{Chapter 3}3419 3420\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 weak3421neighbourhood of $\id$, which constrains maps on a compact set only, contains $g_n$ for large $n$. Since3422$\beta'$ attains its maximum $1$ at some $x^*$, $g_n'(n+x^*)=-1<0$ while $g_n'=1$ outside the support, so3423$g_n$ 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 strong3424neighbourhood of the identity. Not closed: $h_n(x)=x/n$ are diffeomorphisms converging weakly to the constant map $0$.3425 3426\sol{3.2} Fix $t_0$. We claim there are $\delta>0$ and a compact $K$ with $f_t=f_{t_0}$ off $K$ for3427$|t-t_0|<\delta$. If not, for each $n$ there are $t_n$ with $|t_n-t_0|<1/n$ and $x_n\notin[-n,n]$ with3428$f_{t_n}(x_n)\ne f_{t_0}(x_n)$; passing to a subsequence, the $x_n$ are distinct and tend to infinity.3429Choose a continuous $\varepsilon\colon\R\to(0,\infty)$ with3430$\varepsilon(x_n)<|f_{t_n}(x_n)-f_{t_0}(x_n)|$ for all $n$. The strong neighbourhood3431$\{g:|g-f_{t_0}|<\varepsilon\}$ contains no $f_{t_n}$, contradicting continuity at $t_0$. Cover $[0,1]$ by3432finitely many such intervals and take the union of the compact sets. Conversely, if $g=f$ off a compact3433$K$, the straight-line path $f+t(g-f)$ is strongly continuous: it is constant off $K$, and on $K$ uniform3434convergence suffices. So the path component of $f$ is as stated.3435 3436\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$3437and its first derivative in the $M$-direction are uniformly continuous on $\bar B(s_0,1)\times K$, so3438$\sup_{x\in K}\bigl(|F_s(x)-F_{s_0}(x)|+\|D_xF_s(x)-D_xF_{s_0}(x)\|\bigr)\to0$ as $s\to s_0$; this is3439convergence in $C^1_W$. For the example let $F_s(x)=(\gamma_s(x),0)$ with $\gamma_s(x)=x-2\beta(x-1/s)$ for3440$s>0$ 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 $0<s<1/(R+1)$ the3441bump is not seen. $F_0$ is an embedding and $F_s$ is not injective for small $s>0$3442(\cref{ex:weakfail}), so the set of parameters giving embeddings is not open at $s=0$.3443 3444\sol{3.4} Reflexivity: the constant plot $\R\to\pt\to X$. Symmetry: compose with $t\mapsto1-t$. Transitivity:3445let $p,q\colon\R\to X$ be plots with $p(1)=q(0)$. Choose a smooth non-decreasing $\lambda\colon\R\to\R$ with3446$\lambda(t)=0$ for $t\le\frac13$ and $\lambda(t)=1$ for $t\ge\frac23$. Then $p\lambda$ is a plot, constant3447equal to $p(1)$ on $(\frac23,\infty)$, and $q\lambda(\cdot-1)$ is a plot, constant equal to $q(0)$ on3448$(-\infty,\frac43)$. On the overlap $(\frac23,\frac43)$ of the open sets $(-\infty,\frac43)$ and3449$(\frac23,\infty)$, which cover $\R$, the two plots restrict to the same constant plot. By the sheaf3450condition (after pulling back along diffeomorphisms of the intervals with $\R$) they glue to a plot