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 2181–2280 of 3921. Source text is displayed literally.

2181 2182For a general domain there are no formulas for the eigenvalues, and the source turns to what can be said without2183them: variational characterizations, completeness, and asymptotics \src{Ch.~11}. We add the question the2184asymptotics provoke, whether the spectrum determines the domain, because its answer is a statement about2185representations.2186 2187\section{Minimax, and what it makes functorial}2188 2189\begin{theorem}[Minimax]\label{thm:minimax}2190\Scl[source, \S11.1--11.2 and \S11.6] For the Dirichlet problem on a bounded domain $D$,2191\[ \lambda_n(D)=\min_{\substack{W\subset H^1_0(D)\\\dim W=n}}\ \max_{w\in W\setminus0}\ \frac{\int|\nabla w|^2}{\int w^2}. \]2192In particular the Rayleigh--Ritz values in any $n$-dimensional trial space are upper bounds for2193$\lambda_1,\dots,\lambda_n$.2194\end{theorem}2195 2196\noindent On $(0,1)$ the trial function $x(1-x)$ gives $10$, and adding $x^2(1-x)^2$ gives $9.86975$, against2197$\pi^2=9.86960$. \chk{L1} In finite dimensions the theorem is the interlacing of the eigenvalues of a matrix with2198those of its compressions. \chk{L2}2199 2200The formula depends on the domain only through the space of trial functions, and it is monotone in that space.2201 2202\begin{corollary}[Monotonicity]\label{cor:mono}2203\Scl[source, \S11.6, Theorems 4 and 5] If $D\subseteq D'$ then $\lambda_n(D)\ge\lambda_n(D')$ for every $n$:2204Dirichlet eigenvalues are a functor from domains and inclusions to decreasing sequences. For one domain the2205Neumann eigenvalues satisfy $\mu_n\le\lambda_n$. The Neumann eigenvalues are \emph{not} monotone in the domain.2206\end{corollary}2207\begin{proof}2208Extension by zero embeds $H^1_0(D)$ isometrically in $H^1_0(D')$, so the minimum is over a larger family. For the2209second statement $H^1_0(D)\subseteq H^1(D)$. For the third, a rectangle of sides $1.3$ and $0.1$ fits inside the unit2210square along its diagonal, and its first non-zero Neumann eigenvalue $\pi^2/1.69=5.84$ is smaller than that of the2211square, $\pi^2=9.87$.2212\end{proof}2213 2214\noindent An L-shaped region in a $6\times6$ grid has every Dirichlet eigenvalue above the corresponding one of the2215square. \chk{L3} The asymmetry has the same source as in \Cref{thm:images}: Dirichlet conditions are compatible2216with extension by zero and Neumann conditions with nothing of the kind. The spaces $H^1(D)$ are related by2217restriction, which does not preserve the Rayleigh quotient.2218 2219\section{Completeness}2220 2221\begin{theorem}[Spectral theorem for the Laplacian]\label{thm:spectral}2222\Scl[source, \S11.3--11.5] On a bounded domain, with Dirichlet, Neumann or Robin conditions, there is an orthonormal2223basis of $L^2(D)$ consisting of eigenfunctions, with eigenvalues $\lambda_n\to\infty$.2224\end{theorem}2225 2226\noindent The source proves this from the minimax principle. In general the spectral theorem says that a2227self-adjoint operator is unitarily equivalent to multiplication by a real function on a measure space, and its2228functional calculus $f\mapsto f(A)$ is a homomorphism of algebras. Every operator met so far is a function of one2229operator: the heat semigroup $e^{-tA}$, the wave group $\cos(t\sqrt A)$, the resolvent $(s+A)^{-1}$, the Green's2230operator $A^{-1}$, the Schr\"odinger group $e^{-itA}$. In finite dimensions this is the resolution of the identity2231by eigenvectors. \chk{L9}2232 2233\section{How many eigenvalues}2234 2235\begin{theorem}[Weyl]\label{thm:weyl}2236\Scl[source, \S11.6, Theorem 1] For the Dirichlet problem, $\lambda_n/n\to4\pi/A$ in the plane, with $A$ the area, and2237$\lambda_n^{3/2}/n\to6\pi^2/V$ in space, with $V$ the volume.2238\end{theorem}2239 2240\noindent The source's proof brackets the domain between unions of rectangles and uses \Cref{cor:mono}. For a2241$1\times\sqrt2$ rectangle the count $N(\lambda)$ of eigenvalues below $\lambda$ is $208$, $869$, $3531$ at2242$\lambda=2000$, $8000$, $32000$, and the two-term expression $A\lambda/4\pi-P\sqrt\lambda/4\pi$, with $P$ the2243perimeter, gives $207.9$, $865.9$, $3532.5$. \chk{L4}2244 2245\begin{center}2246\begin{tikzpicture}2247\begin{scope}[shift={(0,0)}]2248\draw[ink2] (0,4)--(0,0)--(8.4,0);2249\draw[ink2] (0,0)--(0,-0.07) node[below,inner sep=1.5pt]{\scriptsize $0$};2250\draw[ink2] (2.1,0)--(2.1,-0.07) node[below,inner sep=1.5pt]{\scriptsize $100$};2251\draw[ink2] (4.2,0)--(4.2,-0.07) node[below,inner sep=1.5pt]{\scriptsize $200$};2252\draw[ink2] (6.3,0)--(6.3,-0.07) node[below,inner sep=1.5pt]{\scriptsize $300$};2253\draw[ink2] (8.4,0)--(8.4,-0.07) node[below,inner sep=1.5pt]{\scriptsize $400$};2254\draw[ink2] (0,0)--(-0.07,0) node[left,inner sep=1.5pt]{\scriptsize $0$};2255\draw[ink2] (0,0.889)--(-0.07,0.889) node[left,inner sep=1.5pt]{\scriptsize $10$};2256\draw[ink2] (0,1.778)--(-0.07,1.778) node[left,inner sep=1.5pt]{\scriptsize $20$};2257\draw[ink2] (0,2.667)--(-0.07,2.667) node[left,inner sep=1.5pt]{\scriptsize $30$};2258\draw[ink2] (0,3.556)--(-0.07,3.556) node[left,inner sep=1.5pt]{\scriptsize $40$};2259\node[below] at (4.2,-0.38) {\scriptsize $\lambda$};2260\node[above,inner sep=1pt] at (0,4.05) {\scriptsize $N(\lambda)$};2261\draw[thick,fwd] plot coordinates {(0,0) (0.311,0) (0.311,0.089) (0.622,0.089) (0.622,0.178) (0.933,0.178) (0.933,0.267) (1.14,0.267) (1.14,0.356) (1.244,0.356) (1.244,0.444) (1.762,0.444) (1.762,0.533) (1.865,0.533) (1.865,0.622) (1.969,0.622) (1.969,0.711) (2.28,0.711) (2.28,0.8) (2.487,0.8) (2.487,0.889) (2.798,0.889) (2.798,0.978) (2.798,0.978) (2.798,1.067) (3.42,1.067) (3.42,1.156) (3.42,1.156) (3.42,1.244) (3.523,1.244) (3.523,1.333) (3.731,1.333) (3.731,1.422) (3.938,1.422) (3.938,1.511) (4.249,1.511) (4.249,1.6) (4.456,1.6) (4.456,1.689) (4.56,1.689) (4.56,1.778) (4.974,1.778) (4.974,1.867) (5.285,1.867) (5.285,1.956) (5.285,1.956) (5.285,2.044) (5.596,2.044) (5.596,2.133) (5.596,2.133) (5.596,2.222) (5.907,2.222) (5.907,2.311) (5.907,2.311) (5.907,2.4) (6.114,2.4) (6.114,2.489) (6.84,2.489) (6.84,2.578) (6.84,2.578) (6.84,2.667) (6.943,2.667) (6.943,2.756) (7.047,2.756) (7.047,2.844) (7.461,2.844) (7.461,2.933) (7.565,2.933) (7.565,3.022) (7.772,3.022) (7.772,3.111) (7.876,3.111) (7.876,3.2) (8.394,3.2) (8.394,3.289) (8.394,3.289) (8.394,3.378) (8.4,3.378)};2262\draw[ink2,densely dashed] plot coordinates {(0,0) (0.21,0.1) (0.42,0.2) (0.63,0.3) (0.84,0.4) (1.05,0.5) (1.26,0.6) (1.47,0.7) (1.68,0.8) (1.89,0.9) (2.1,1) (2.31,1.1) (2.52,1.2) (2.73,1.3) (2.94,1.4) (3.15,1.501) (3.36,1.601) (3.57,1.701) (3.78,1.801) (3.99,1.901) (4.2,2.001) (4.41,2.101) (4.62,2.201) (4.83,2.301) (5.04,2.401) (5.25,2.501) (5.46,2.601) (5.67,2.701) (5.88,2.801) (6.09,2.901) (6.3,3.001) (6.51,3.101) (6.72,3.201) (6.93,3.301) (7.14,3.401) (7.35,3.501) (7.56,3.601) (7.77,3.701) (7.98,3.801) (8.19,3.901)};2263\draw[thick,bwd] plot coordinates {(0,0) (0.105,0) (0.21,0) (0.315,0.018) (0.42,0.047) (0.525,0.079) (0.63,0.113) (0.735,0.148) (0.84,0.184) (0.945,0.221) (1.05,0.259) (1.155,0.297) (1.26,0.336) (1.365,0.375) (1.47,0.414) (1.575,0.454) (1.68,0.495) (1.785,0.535) (1.89,0.576) (1.995,0.617) (2.1,0.659) (2.205,0.7) (2.31,0.742) (2.415,0.784) (2.52,0.826) (2.625,0.869) (2.73,0.911) (2.835,0.954) (2.94,0.996) (3.045,1.039) (3.15,1.082) (3.255,1.125) (3.36,1.169) (3.465,1.212) (3.57,1.255) (3.675,1.299) (3.78,1.342) (3.885,1.386) (3.99,1.43) (4.095,1.474) (4.2,1.518) (4.305,1.562) (4.41,1.606) (4.515,1.65) (4.62,1.694) (4.725,1.738) (4.83,1.783) (4.935,1.827) (5.04,1.872) (5.145,1.916) (5.25,1.961) (5.355,2.005) (5.46,2.05) (5.565,2.095) (5.67,2.14) (5.775,2.185) (5.88,2.229) (5.985,2.274) (6.09,2.319) (6.195,2.364) (6.3,2.409) (6.405,2.455) (6.51,2.5) (6.615,2.545) (6.72,2.59) (6.825,2.635) (6.93,2.681) (7.035,2.726) (7.14,2.771) (7.245,2.817) (7.35,2.862) (7.455,2.908) (7.56,2.953) (7.665,2.999) (7.77,3.044) (7.875,3.09) (7.98,3.136) (8.085,3.181) (8.19,3.227) (8.295,3.273) (8.4,3.318)};2264\node[ink2,left] at (6.09,3.511) {\scriptsize $A\lambda/4\pi$};2265\node[bwd] at (6.3,1.156) {\scriptsize $A\lambda/4\pi-P\sqrt\lambda/4\pi$};2266\end{scope}2267\end{tikzpicture}2268\figcap{The counting function of the Dirichlet eigenvalues of the $1\times\sqrt2$ rectangle (steps, 38 eigenvalues below $400$), Weyl's leading term (dashed), and the two-term expression with the perimeter correction. The area is heard first and the perimeter second.}2269\end{center}2270 2271\section{The trace of the semigroup}2272 2273The function $Z(t)=\sum_ne^{-\lambda_nt}$ is the trace of the heat semigroup of \Cref{ch:diffusions}:2274$Z(t)=\int_DK(x,x,t)\,dx$.2275 2276\begin{theorem}[Hearing area, perimeter and holes]\label{thm:heattrace}2277\Sli[\cite{Kac66,McSi67}] For a plane domain with smooth boundary and $h$ holes,2278$Z(t)=\dfrac A{4\pi t}-\dfrac P{8\sqrt{\pi t}}+\dfrac{1-h}6+o(1)$ as $t\to0$. For a polygon the constant term is2279$\sum_i(\pi^2-\theta_i^2)/24\pi\theta_i$ over the interior angles $\theta_i$.2280\end{theorem}