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 1561–1660 of 3921. Source text is displayed literally.

1561$U$ if and only if its class in $H^1(U;\R)$ vanishes. On an annulus this class is the period1562$\oint\star\,\dd u$ around the hole. The period of $\log r$ is $2\pi$; the periods of $r^{\pm n}\cos n\theta$ and1563$r^{\pm n}\sin n\theta$ are zero.1564\end{proposition}1565 1566\noindent The source's solution of the Dirichlet problem in an annulus \src{\S6.4} has, besides the terms that1567appear for the disk, the functions $r^{-n}\cos n\theta$, $r^{-n}\sin n\theta$ and $\log r$. The first two families1568are the real parts of $z^{-n}$, holomorphic in the annulus. The last is not the real part of anything: it1569generates the cokernel of $\mathrm{Re}\colon\OO(U)\to\HH(U)$, which is $H^1(U;\R)=\R$. The three periods were1570computed numerically as $2\pi$, $0$, $0$. \chk{H5} This is the first cohomology class of the volume that is not1571zero, and its discrete counterpart is computed in \Cref{ch:numerics}.1572 1573\section{Corners}1574 1575\begin{proposition}\label{prop:wedge}1576\Scl[source, \S6.4] In a wedge of opening $\beta$ the harmonic functions vanishing on both edges are spanned near1577the vertex by $r^{n\pi/\beta}\sin(n\pi\theta/\beta)$. The gradient of the first behaves like $r^{\pi/\beta-1}$: it is1578bounded if $\beta\le\pi$ and unbounded if $\beta>\pi$.1579\end{proposition}1580 1581\noindent At a re-entrant corner of angle $3\pi/2$ the exponent is $\tfrac23$ and the gradient grows like $r^{-1/3}$.1582\chk{H6} Interior regularity of harmonic functions is unconditional; regularity at the boundary is a property of the1583boundary. The exponent $\pi/\beta$ is the reason that finite element methods lose accuracy on an L-shaped domain1584(\Cref{ch:numerics}), and it is again conformal naturality: $z\mapsto z^{\pi/\beta}$ opens the wedge to a1585half-plane.1586 1587\section*{Notes}1588 1589The source's Chapter 6 is followed throughout. For conformal invariance and the Kelvin transform see1590\cite{ABR01}; for Brownian motion and harmonic measure \cite{Kak44,MoPe10}; for corner singularities1591\cite{Gri85}. The Dirichlet problem in rectangles and cubes \src{\S6.2} is separation of variables, taken up in1592\Cref{ch:symmetry}.1593 1594\section*{Exercises}1595 1596\begin{exercise}\label{exer:mv}1597For a quadratic polynomial $u$ in the plane, show that the average over a circle of radius $r$ exceeds the value1598at the centre by $\tfrac{r^2}4\Lap u$.1599\end{exercise}1600 1601\begin{exercise}\label{exer:conf}1602Verify \Cref{thm:conformal}(b) for $f(z)=z^2$ and $u=x^2$.1603\end{exercise}1604 1605\begin{exercise}\label{exer:periods}1606Compute $\oint\star\,\dd u$ around the unit circle for $u=\log r$ and for $u=\mathrm{Re}(1/z)$.1607\end{exercise}1608 1609\begin{exercise}\label{exer:corner}1610Compare the behaviour of the gradient of the first singular function at the re-entrant corner of an L-shaped1611room and at one of its convex corners.1612\end{exercise}1613%% ---- end ch/06-harmonic.tex ----1614%% ---- begin ch/07-green.tex ----1615\chapter{Green's Identities and Green's Functions}\label{ch:green}1616 1617The source derives its remaining general facts about harmonic functions from two integral identities and then1618inverts the Laplacian by a kernel \src{Ch.~7}. Both steps have a structural reading, and the second has a1619compositional one: boundary-value problems can be glued.1620 1621\section{Stokes' theorem, twice}1622 1623\begin{proposition}[Green's identities]\label{prop:green}1624\Scl[source, \S7.1--7.2] For functions on a bounded domain $D$ with smooth boundary,1625\begin{align*}1626\int_Dv\,\Lap u+\int_D\nabla v\cdot\nabla u&=\int_{\pd D}v\,\pd_nu,\\1627\int_D(u\,\Lap v-v\,\Lap u)&=\int_{\pd D}(u\,\pd_nv-v\,\pd_nu).1628\end{align*}1629\end{proposition}1630 1631\noindent Both are Stokes' theorem for the form $v\star\dd u$. On a grid the first is the factorization of1632\Cref{prop:tonti}: the Dirichlet Laplacian is $D^{\T}D$. \chk{G1} The right-hand side of the second is a symplectic1633form on the space of pairs (boundary values, normal derivatives), the same form as in \Cref{thm:lagrangian}.1634 1635\begin{corollary}[The Dirichlet-to-Neumann map is symmetric]\label{cor:dtn}1636\Srf The pairs $(u|_{\pd D},\pd_nu|_{\pd D})$ for $u$ harmonic in $D$ form a Lagrangian subspace of the boundary data.1637It is the graph of the operator $\Lambda_D\colon h\mapsto\pd_nu$, where $u$ is the harmonic function with boundary1638values $h$; hence $\Lambda_D$ is symmetric, and by the first identity non-negative.1639\end{corollary}1640 1641\section{Dirichlet's principle}1642 1643\begin{theorem}\label{thm:dirichlet}1644\Scl[source, \S7.1] Among functions with given boundary values, the harmonic one has the least energy1645$E(w)=\tfrac12\int_D|\nabla w|^2$. More precisely, if $u$ is harmonic and $z$ vanishes on the boundary, then1646$E(u+z)=E(u)+E(z)$.1647\end{theorem}1648 1649\noindent The identity is the Pythagorean theorem for the energy inner product: harmonic functions are orthogonal to1650functions vanishing on the boundary. \chk{G2} So the solution operator of the Dirichlet problem is an orthogonal1651projection. The same holds in any subspace, which is the Rayleigh--Ritz or Galerkin method \src{\S7.1, Exercise16527}: the approximate solution in a trial space is the projection of the true one, and its error is orthogonal to1653the trial space. \chk{G7} Finite elements are this statement with a particular trial space (\Cref{ch:numerics}).1654 1655\section{The kernel of the inverse}1656 1657\begin{definition}[source, \S7.3]1658The \term{Green's function} of $D$ with pole $x_0$ is the function $G(x,x_0)$ that is harmonic in $D$ away from1659$x_0$, vanishes on $\pd D$, and differs from $-1/(4\pi|x-x_0|)$ by a function harmonic near $x_0$ (in three1660dimensions).