Indexed text attachment
partial-differential-equations-in-higher-categories.tex
Source for Partial Differential Equations in Higher Categories
View paper · Download original file
Lines 1721–1820 of 3921. Source text is displayed literally.
1721Let a \term{network} be a finite graph with positive conductances on its edges and a chosen set of1722\term{terminals}; a potential is harmonic if Kirchhoff's law holds at every non-terminal vertex. Eliminating the1723interior vertices from the Laplacian (a Schur complement, or Kron reduction) gives the matrix $\Lambda$ that sends1724terminal potentials to terminal currents: the discrete Dirichlet-to-Neumann map.1725 1726\begin{theorem}[Black-boxing is a functor]\label{thm:gluing}1727\Sli[\cite{BaFo18}] Networks with terminals are the morphisms of a category in which composition glues terminals.1728Sending a network to the graph of its Dirichlet-to-Neumann map is a functor to the category $\LagRel$ of1729symplectic vector spaces and Lagrangian relations. In particular the response of a glued network is determined1730by the responses of its parts.1731\end{theorem}1732 1733\noindent For ten pairs of random networks glued along two shared terminals, the response of the whole agreed with1734the composite of the responses of the parts to ten digits; conductances $2$ and $3$ in series give $6/5$. \chk{G6}1735The continuous statement is the same with \Cref{cor:dtn} in place of the matrix: if $D=D_1\cup_\Sigma D_2$, the1736Lagrangian subspace of $D$ is the composite of those of $D_1$ and $D_2$ over the interface $\Sigma$. Domain1737decomposition methods in numerical analysis are this functoriality used as an algorithm: solve on the pieces,1738then solve an equation on the interface.1739 1740\begin{center}1741\begin{tikzpicture}1742\tikzset{ntm/.style={circle,fill,inner sep=1.8pt},nin/.style={circle,draw,thick,fill=white,inner sep=1.6pt}}1743\begin{scope}1744\node[ntm] (a) at (0,0.6) {}; \node[nin] (b) at (1,1.3) {}; \node[nin] (c) at (1,-0.1) {};1745\node[ntm,accent] (s1) at (2.1,1.2) {}; \node[ntm,accent] (s2) at (2.1,0) {};1746\draw[thick] (a)--(b)--(c)--(a) (b)--(s1) (c)--(s2) (b)--(s2);1747\node at (1,-0.8) {\scriptsize $D_1$};1748\end{scope}1749\begin{scope}[xshift=3.3cm]1750\node[ntm,accent] (s1) at (0,1.2) {}; \node[ntm,accent] (s2) at (0,0) {};1751\node[nin] (d) at (1.1,0.6) {}; \node[ntm] (e) at (2.1,1.2) {}; \node[ntm] (f) at (2.1,0) {};1752\draw[thick] (s1)--(d)--(s2) (d)--(e)--(f)--(d);1753\node at (1,-0.8) {\scriptsize $D_2$};1754\end{scope}1755\draw[->,thick,ink2] (6.1,0.6)--(7.0,0.6);1756\begin{scope}[xshift=7.7cm]1757\node[ntm] (a) at (0,0.6) {}; \node[nin] (b) at (1,1.3) {}; \node[nin] (c) at (1,-0.1) {};1758\node[nin,accent] (s1) at (2.1,1.2) {}; \node[nin,accent] (s2) at (2.1,0) {};1759\node[nin] (d) at (3.2,0.6) {}; \node[ntm] (e) at (4.2,1.2) {}; \node[ntm] (f) at (4.2,0) {};1760\draw[thick] (a)--(b)--(c)--(a) (b)--(s1) (c)--(s2) (b)--(s2) (s1)--(d)--(s2) (d)--(e)--(f)--(d);1761\node at (2.1,-0.8) {\scriptsize $D_1\cup_\Sigma D_2$};1762\end{scope}1763\end{tikzpicture}1764\figcap{Gluing two networks along the shared terminals $\Sigma$ (in colour). Filled nodes are terminals, open ones are interior. After gluing, the nodes of $\Sigma$ are interior, and Kirchhoff's law there is the condition that the currents from the two sides cancel.}1765\end{center}1766 1767\[\begin{tikzcd}[column sep=2.2em,row sep=2.2em]1768 & & L_{D}\arrow[dl]\arrow[dr] & & \\1769 & L_{D_1}\arrow[dl]\arrow[dr] & & L_{D_2}\arrow[dl]\arrow[dr] & \\1770V_{\pd_1} & & V_\Sigma & & V_{\pd_2}1771\end{tikzcd}\]1772Each $V$ is a space of pairs (potential, current) on a piece of boundary, with the sign of the current reversed on1773one side of $\Sigma$ so that what leaves $D_1$ enters $D_2$. The top object is the image of the pullback: pairs of1774solutions on the two parts that agree on the interface in value and in flux, which is to say solutions on the1775whole.1776 1777\begin{remark}1778\Cref{thm:lagrangian}, \Cref{cor:dtn} and \Cref{thm:gluing} are one statement at three levels: boundary1779conditions, solutions and parts of a domain are all Lagrangian relations on spaces of boundary data, and a1780boundary-value problem is solved by composing them. An elliptic problem is a morphism from the empty boundary1781to the boundary data, and its solution is the composite with the boundary condition. This is as close as the1782elliptic theory comes to the time-evolution functors of the hyperbolic and parabolic theories.1783\end{remark}1784 1785\section*{Notes}1786 1787The source's Chapter 7 is followed for the identities, Dirichlet's principle and Green's functions. Black-boxing1788of passive linear networks as a functor to Lagrangian relations is due to Baez and Fong \cite{BaFo18}; Kron1789reduction is surveyed in \cite{DoBu13}. For Dirichlet-to-Neumann operators and domain decomposition see1790\cite{QuVa99}.1791 1792\section*{Exercises}1793 1794\begin{exercise}\label{exer:dtnsym}1795Deduce the symmetry of $\Lambda_D$ from Green's second identity.1796\end{exercise}1797 1798\begin{exercise}\label{exer:series}1799Compute the Kron reduction of two conductances $g_1,g_2$ in series onto the two end terminals.1800\end{exercise}1801 1802\begin{exercise}\label{exer:green1d}1803Show that $-d^2/dx^2$ on $(0,1)$ with Dirichlet conditions has the inverse kernel $x(1-y)$ for $x<y$, and expand1804it in eigenfunctions.1805\end{exercise}1806 1807\begin{exercise}\label{exer:kelvin}1808Show that for $|x|=a$ one has $|x-x_0^*|=(a/|x_0|)\,|x-x_0|$, and deduce that the Green's function of the ball1809vanishes on the sphere.1810\end{exercise}1811%% ---- end ch/07-green.tex ----1812%% ---- begin ch/08-space.tex ----1813\chapter{Waves and Quanta in Space}\label{ch:space}1814 1815In three dimensions the wave equation keeps its causal structure and gains a property it did not have on the1816line; the diffusion kernel becomes a product; and a third evolution equation, Schr\"odinger's, turns out to be1817both of the others at once \src{Ch.~9}.1818 1819\section{Cones}1820