Independent scholarship · Open accessThe monthly digest

Indexed text attachment

partial-differential-equations-in-higher-categories.tex

Source for Partial Differential Equations in Higher Categories

View paper · Download original file

Lines 2361–2460 of 3921. Source text is displayed literally.

2361\begin{exercise}\label{exer:bracket}2362Use \Cref{cor:mono} to bound the first Dirichlet eigenvalue of the unit disk between those of an inscribed and a2363circumscribed square, and compare with $z_{01}^2$.2364\end{exercise}2365 2366\begin{exercise}\label{exer:weylrect}2367Derive the two-term count for an $a\times b$ rectangle by counting lattice points in a quarter of an ellipse.2368\end{exercise}2369 2370\begin{exercise}\label{exer:corners}2371Verify that the constant term of \Cref{thm:heattrace} for a polygon gives $\tfrac14$ for a rectangle, and find it2372for an equilateral triangle.2373\end{exercise}2374%% ---- end ch/10-eigen.tex ----2375%% ---- begin ch/11-distributions.tex ----2376\chapter{Distributions and Transforms}\label{ch:distributions}2377 2378Several objects used so far were not functions: the initial state of the source function, the pole of a Green's2379function, a solution with a corner. The source legitimizes them late \src{Ch.~12}. In the language of2380\Cref{ch:equations} the step is a change of the second argument of the solution functor.2381 2382\section{Distributions}2383 2384A \term{test function} on an open set $U$ is a smooth function of compact support, and a \term{distribution} is a2385continuous linear functional on test functions \src{\S12.1}. Distributions restrict to smaller open sets and glue,2386so they form a sheaf $\DD'$, and differential operators act on them by duality: $(\pd f,\varphi)=-(f,\pd\varphi)$.2387So $\DD'$ is a sheaf of modules over the differential operators that contains the smooth functions, and2388$\Sol(P;\DD')$ is defined: these are the \term{weak solutions}.2389 2390\begin{example}2391The delta function is the limit of the source functions, $(S(\cdot,t),\varphi)=\varphi(0)+t\varphi''(0)+O(t^2)$ for2392$k=1$. The derivative of the step function is $\delta$, and the second derivative of $|x|$ is $2\delta$.2393\chk{T1, T2} The function $\tfrac12[\varphi(x+ct)+\varphi(x-ct)]$ is a weak solution of the wave equation for every2394locally integrable $\varphi$; this is \Cref{cor:par} stated in $\DD'$.2395\end{example}2396 2397\begin{theorem}[Kernels]\label{thm:kernel}2398\Scl[\cite{Sch57}] Continuous linear maps from test functions on $X$ to distributions on $Y$ correspond bijectively2399to distributions on $X\times Y$.2400\end{theorem}2401 2402\noindent This is the tensor--hom adjunction for the spaces in question; it holds because they are nuclear. It is2403the reason that every solution operator in this volume is given by a kernel: the source function, the Green's2404function, the Poisson kernel, the Kirchhoff kernel. One limit of the theory should be recorded now. There is no2405associative product on $\DD'$ extending the product of continuous functions \cite{Sch54}. Linear equations with2406smooth coefficients act on distributions; non-linear ones do not, and \Cref{ch:nonlinear} must find other notions2407of weak solution.2408 2409\section{Fundamental solutions}2410 2411A \term{fundamental solution} of $P$ is a distribution $E$ with $PE=\delta$; then $u=E*f$ solves $Pu=f$ for $f$ of2412compact support \src{\S12.2}.2413 2414\begin{theorem}\label{thm:fund}2415\Scl[\cite{Mal56,Ehr54}] Every non-zero differential operator with constant coefficients has a fundamental solution.2416\end{theorem}2417 2418\begin{center}\small2419\begin{tabular}{@{}lll@{}}2420\toprule2421operator & fundamental solution & singular support\\2422\midrule2423$\Lap$ in $\R^3$ & $-1/4\pi r$ & the origin\\2424$\pd_t-k\pd_x^2$ & $S(x,t)$ for $t>0$, else $0$ & the origin\\2425$\pd_t^2-c^2\pd_x^2$ & $1/2c$ for $|x|<ct$, else $0$ & the cone $|x|=ct$\\2426$\pd_t^2-c^2\Lap$ in $\R^3$ & $\delta(r-ct)/4\pi c\,r$ & the cone $r=ct$\\2427\bottomrule2428\end{tabular}2429\end{center}2430 2431\noindent The first line is checked against a test function. \chk{T3} The last column is the type of2432\Cref{thm:type} once more: an elliptic or parabolic fundamental solution is smooth away from the pole, so that all2433solutions are smooth where the right-hand side is; a hyperbolic one is singular along the characteristic cone,2434which is \Cref{thm:wf}. The last line is Huygens's principle: the support, and not only the singular support, is2435the cone.2436 2437\section{The Fourier transform}2438 2439\begin{theorem}\label{thm:fourier}2440\Scl[source, \S12.3] The Fourier transform $F(k)=\int f(x)e^{-ikx}dx$ is an isomorphism of the Schwartz space with2441itself and of the tempered distributions with themselves; it is unitary on $L^2$ up to the factor $2\pi$; and it2442exchanges $\pd_x$ with multiplication by $ik$ and multiplication by $x$ with $i\pd_k$.2443\end{theorem}2444 2445\noindent These properties are checked numerically. \chk{T4} Structurally: the line is its own Pontryagin dual, as2446the circle and the integers were dual in \Cref{thm:pontryagin}; and the transform is an automorphism of the Weyl2447algebra generated by $x$ and $\pd$ with $[\pd,x]=1$, since $[ik,i\pd_k]=1$. An equation with constant coefficients2448becomes an equation of order zero, $p(ik)\hat u=\hat f$, and solving it is division by a polynomial. The zero set2449of that polynomial is the characteristic variety of \Cref{ch:equations}, and the difficulty of the division there2450is the whole difficulty.2451 2452The duality is functorial for homomorphisms of groups: the dual of the quotient $\R\to\R/2l\Z$ is the inclusion of2453the lattice $\tfrac\pi l\Z\subset\R$. Applied to a function this is the Poisson summation formula, and applied to the2454heat kernel it is \Cref{thm:kernels} and \Cref{prop:rect}. The image method and Fourier's method are dual in the2455literal sense.2456 2457\section{Source functions}2458 2459With the transform the source functions are computed and not guessed \src{\S12.4}. For diffusion,2460$\hat S_t=-k^2\hat S$, $\hat S(k,0)=1$, so $\hat S=e^{-k^2t}$ and $S$ is the Gaussian. \chk{T5} For Laplace's equation