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 1841–1940 of 3921. Source text is displayed literally.

1841 1842\noindent This is \term{Huygens's principle}: a disturbance of bounded extent is heard for a bounded time. The1843reason is visible for radial solutions: $(ru)_{tt}=c^2(ru)_{rr}$, so $ru$ solves the one-dimensional equation and1844$u=(F(r-ct)+G(r+ct))/r$. \chk{S1}1845 1846In two dimensions the source obtains the solution by \term{descent}: a function of $(x,y)$ is a function of1847$(x,y,z)$ that does not depend on $z$.1848 1849\begin{proposition}[Descent loses sharpness]\label{prop:descent}1850\Srf[source, \S9.2] Solutions in the plane are the solutions in space invariant under translation in $z$. The1851pushforward of the uniform measure on a sphere of radius $R$ under projection to the plane has the density1852$\bigl(2\pi R\sqrt{R^2-\rho^2}\bigr)^{-1}$ on the disk, so1853\[ u(x,t)=\frac1{2\pi c}\int_{|y-x|<ct}\frac{\psi(y)\,dy}{\sqrt{c^2t^2-|y-x|^2}}+\cdots \]1854depends on the data in the whole disk, with a tail of order $1/t$.1855\end{proposition}1856 1857\noindent For a source in the unit ball heard at distance $5$ with $c=1$, the three-dimensional signal is zero at1858$t=3$, non-zero at $t=5$ and exactly zero again at $t=6.5$; the two-dimensional one never stops, and $t\,u(t)$ is1859$1.29$, $1.002$, $1.0000$ times its limit at $t=8$, $80$, $800$. \chk{S2} Sharp propagation holds in odd dimensions1860from three upwards and fails in even ones; the source's remark that three is the best of all possible dimensions1861is the remark that it is the smallest of these. The structural point is that the property is not inherited by1862invariants under a non-compact group: the plane is the quotient of space by a line, and the shadow of a sphere1863is a disk.1864 1865\begin{center}1866\begin{tikzpicture}[scale=0.8]1867\begin{scope}1868\draw[->,ink2] (0,0)--(3.4,0) node[right]{\small$r$};1869\draw[->,ink2] (0,0)--(0,3.4) node[above]{\small$t$};1870\draw[very thick,fwd] (0,0)--(3.1,3.1);1871\draw[densely dashed,ink2] (2,0)--(2,3.2);1872\fill[bwd] (2,2) circle (2pt);1873\node at (1.6,-0.55) {\small three dimensions};1874\end{scope}1875\begin{scope}[xshift=6.4cm]1876\fill[fwd!18] (0,0)--(3.1,3.1)--(0,3.1)--cycle;1877\draw[->,ink2] (0,0)--(3.4,0) node[right]{\small$r$};1878\draw[->,ink2] (0,0)--(0,3.4) node[above]{\small$t$};1879\draw[very thick,fwd] (0,0)--(3.1,3.1);1880\draw[densely dashed,ink2] (2,0)--(2,3.2);1881\draw[very thick,bwd] (2,2)--(2,3.1);1882\node at (1.6,-0.55) {\small two dimensions};1883\end{scope}1884\end{tikzpicture}1885\figcap{The support of the response to an impulse at the origin. A listener at fixed distance (dashed) hears an1886instant in three dimensions and a tail that never ends in two.}1887\end{center}1888 1889\section{Singularities travel along rays}1890 1891\begin{theorem}[Propagation of singularities]\label{thm:wf}1892\Sli[\cite{Hor71,DuHo72}] For a solution $u$ of $Pu=0$ with $P$ of real principal type, the wave front set1893$\WF(u)\subseteq T^*M$ is contained in $\Char(P)$ and is a union of bicharacteristics, the integral curves of the1894Hamiltonian vector field of the principal symbol.1895\end{theorem}1896 1897\noindent For the wave equation the bicharacteristics project to light rays, and the theorem contains the source's1898statement that only characteristic surfaces can carry singularities \src{\S9.3, Theorem 3}. For an elliptic1899operator $\Char(P)$ is empty and solutions are smooth. For the heat equation the characteristic directions are1900$dt$ and singularities do not propagate at finite speed at all. The wave front set lives on the cotangent1901bundle; that singularities have a direction as well as a position is the starting point of microlocal analysis,1902where the solution sheaf is studied on $T^*M$ \cite{KaSc90}.1903 1904\section{Diffusion, and Schr\"odinger's equation}1905 1906In $n$ dimensions the diffusion kernel is the product of one-dimensional kernels \src{\S9.4}: the semigroup on1907$L^2(\R^n)=L^2(\R)^{\otimes n}$ is $T_t\otimes\cdots\otimes T_t$. Everything in \Cref{ch:diffusions} holds unchanged.1908 1909The free Schr\"odinger equation $iu_t=-\tfrac12\Lap u$ is formally the diffusion equation at imaginary time.1910 1911\begin{theorem}[One semigroup, two boundary values]\label{thm:wick}1912\Scl[\cite{ReSi75}] Let $H$ be a non-negative self-adjoint operator. Then $z\mapsto e^{-zH}$ is a holomorphic1913semigroup of contractions on $\mathrm{Re}\,z>0$, strongly continuous up to the imaginary axis. Its restriction to1914$z=t>0$ is the heat semigroup; its restriction to $z=it$ is the unitary group solving $iu_t=Hu$.1915\end{theorem}1916 1917\noindent The kernel $(2\pi it)^{-1/2}e^{ix^2/2t}$ solves the equation and equals the diffusion kernel evaluated at1918time $it$. A Gaussian packet keeps its norm exactly while its mean square width grows as $(1+t^2)/2$. \chk{S3}1919In the table of \Cref{ch:diffusions} Schr\"odinger's equation takes the first row from waves (a group, here1920unitary) and the generator from diffusions; it has infinite speed of propagation and no maximum principle, and it1921neither transports singularities sharply nor erases them, but disperses them. For the harmonic oscillator1922$-v''+x^2v=\lambda v$ the eigenvalues are $1,3,5,\dots$ \chk{S5}1923 1924\section{The hydrogen atom}1925 1926\begin{theorem}\label{thm:hydrogen}1927\Scl[source, \S9.5 and \S10.7] The equation $-\Lap v-\tfrac2rv=\lambda v$ in space has the square-integrable1928solutions $v=w(r)e^{-r/n}Y(\theta,\phi)$ with eigenvalues $\lambda=-1/n^2$, $n=1,2,\dots$; the eigenvalue $-1/n^2$ has1929multiplicity $n^2$.1930\end{theorem}1931 1932\noindent The radial series terminates exactly when the exponent is the reciprocal of an integer, and the1933polynomials for $n=1,2,3$ are $1$, $1-\tfrac12r$, $1-\tfrac23r+\tfrac2{27}r^2$, as in the source's table. \chk{S4} The1934multiplicity is $\sum_{l<n}(2l+1)$: rotational symmetry accounts for the factor $2l+1$ (\Cref{ch:symmetry}) and does1935not explain why different $l$ share an eigenvalue.1936 1937\begin{center}1938\begin{tikzpicture}1939\begin{scope}[shift={(0,0)}]1940\draw[ink2] (0,3.3)--(0,0)--(8.2,0);