Indexed text attachment
partial-differential-equations-in-higher-categories.tex
Source for Partial Differential Equations in Higher Categories
View paper · Download original file
Lines 721–820 of 3921. Source text is displayed literally.
721heat equation give \Cref{ex:backward}; boundary values for the wave equation on a square give non-uniqueness722(\Cref{exer:dirwave}). The rest of the volume takes the three rows in turn, and \Cref{ch:coda} returns to the723table.724 725\section*{Notes}726 727The separation of field equations into topological balance laws and metric constitutive laws is Tonti's728\cite{Ton75}; in the language of differential forms it is the basis of compatible discretization729\cite{Bos98,AFW06}. Well-posedness is Hadamard's notion; his example is in \cite{Had23}. Fredholm theory for730elliptic boundary problems is treated in \cite{Eva10}.731 732\section*{Exercises}733 734\begin{exercise}\label{exer:dirwave}735Show that the Dirichlet problem for $u_{tt}=u_{xx}$ on the square $[0,\pi]^2$ in the $(x,t)$-plane does not have736unique solutions.737\end{exercise}738 739\begin{exercise}\label{exer:neumann}740Let $D$ be the incidence matrix of a connected graph. Show that the range of $D^{\T}D$ is the space of vectors of741mean zero.742\end{exercise}743 744\begin{exercise}\label{exer:backerr}745An initial temperature on $(0,\pi)$ is to be recovered from a measurement at time $0.1$ whose error is $10^{-10}$746in each mode. Which is the first mode in which the error of the reconstruction exceeds $1$?747\end{exercise}748 749\begin{exercise}\label{exer:robin-energy}750For $u_t=ku_{xx}$ on $(0,l)$ with the Robin conditions $u_x(0)=a_0u(0)$ and $u_x(l)=-a_lu(l)$, compute751$\tfrac{d}{dt}\tfrac12\int u^2$. For which signs of $a_0,a_l$ is it certain to be non-positive?752\end{exercise}753%% ---- end ch/02-archetypes.tex ----754\part{One Space Dimension}755%% ---- begin ch/03-waves.tex ----756\chapter{Waves}\label{ch:waves}757 758The wave equation on the line, $u_{tt}=c^2u_{xx}$, is the one archetype that can be solved in closed form by759looking at it. This chapter extracts four structures from the solution: a decomposition of the solution sheaf,760a causal structure, a symplectic form, and an action of reflection groups. Each reappears later in a setting761where no formula is available.762 763\section{Two transports glued along the constants}764 765\begin{theorem}[d'Alembert]\label{thm:dalembert}766\Scl[source, \S2.1] Let $q_\pm(x,t)=x\pm ct$. Every smooth solution on the plane is767$u=f\circ q_++g\circ q_-$, and $(f,g)$ is unique up to $(f+\kappa,g-\kappa)$:768\[ 0\longrightarrow\R\longrightarrow C^\infty(\R)\oplus C^\infty(\R)\xrightarrow{\ q_+^*+q_-^*\ }\Sol\longrightarrow0 \]769is exact. The solution with $u=\varphi$, $u_t=\psi$ at $t=0$ is770$u(x,t)=\tfrac12[\varphi(x+ct)+\varphi(x-ct)]+\tfrac1{2c}\int_{x-ct}^{x+ct}\psi$.771\end{theorem}772\begin{proof}773By \Cref{prop:factor} and \Cref{prop:char}: the two factors are transport operators with leaf spaces $q_\pm$, each774maps the solutions of the other onto themselves, and their common solutions are the constants. The formula is the775solution of two linear equations for $f'$ and $g'$. \chk{V1}776\end{proof}777 778\begin{corollary}[Parallelogram identity]\label{cor:par}779\Scl A function of the form $f\circ q_++g\circ q_-$ satisfies $u(A)+u(C)=u(B)+u(D)$ for the vertices $A,B,C,D$, in780order, of every parallelogram whose sides are characteristics. Conversely a continuous function with this property781is of that form.782\end{corollary}783 784\noindent The identity holds to rounding error for exact solutions. \chk{V2} It uses no derivative, so it defines785solutions that are merely continuous, or less: the first weak formulation of the volume. Restricted to a grid with786$c\,\Delta t=\Delta x$ it is an exact numerical scheme (\Cref{ch:numerics}).787 788\section{Causality}789 790For an interval $I$ of the $x$-axis let $D(I)$ be its \term{domain of dependence}: the points $(x,t)$ with791$[x-c|t|,x+c|t|]\subseteq I$, a diamond with waist $I$.792 793\begin{theorem}[Time slices]\label{thm:causal}794\Srf[source, \S2.2] Restriction to the waist is an isomorphism795\[ \Sol(D(I))\ \xrightarrow{\ \sim\ }\ C^\infty(I)\oplus C^\infty(I),\qquad u\mapsto(u,u_t)|_{t=0}, \]796natural for inclusions of intervals. Data outside $I$ do not affect the solution in $D(I)$, and data in $I$ affect797the solution only in the \term{domain of influence} of $I$.798\end{theorem}799 800\noindent Changing the data outside the interval of dependence of a point left the solution there unchanged to the801last bit; changing them inside did not. \chk{V3} Two remarks. The assignment $I\mapsto\Sol(D(I))$ is a functor on the802poset of intervals, and the theorem says it is determined by a much smaller amount of information than a803general sheaf on the plane would be; this is the property that field theory axiomatizes (\Cref{ch:fields}). And804the theorem has no counterpart for diffusion, where the solution at any point depends on all the data805(\Cref{ch:diffusions}).806 807\begin{center}808\begin{tikzpicture}[scale=0.78]809\begin{scope}810\fill[fwd!9] (-2.5,0)--(0,2.5)--(2.5,0)--(0,-2.5)--cycle;811\draw[thick,fwd] (-2.5,0)--(0,2.5)--(2.5,0)--(0,-2.5)--cycle;812\fill[fwd!24] (-0.6,0)--(0.4,1)--(1.4,0)--(0.4,-1)--cycle;813\draw[fwd] (-0.6,0)--(0.4,1)--(1.4,0)--(0.4,-1)--cycle;814\draw[->,ink2] (-3.1,0)--(3.3,0) node[right]{\small$x$};815\draw[->,ink2] (0,-2.8)--(0,3.0) node[above]{\small$t$};816\draw[very thick] (-2.5,0)--(2.5,0);817\node[below] at (-1.9,0) {\small$I$};818\node at (0.42,0.36) {\small$D(J)$};819\node at (-1.0,0.95) {\small$D(I)$};820\fill (1.25,1.0) circle (1.6pt) node[above right,inner sep=2pt]{\small$(x,t)$};