% vim: tw=50 % 02/03/2022 10AM We can also look in cylindrical polar coordinates and seek solutions of the form \[ \psi = \psi(r) \] ($r^2 = x^2 + y^2$). Now we have \[ \nabla^2 \psi = \dfrac[2]{\psi}{r} + \frac{1}{r} \dfrac{\psi}{r} = \frac{1}{r} \dfrac{}{r} \left( r \dfrac{\psi}{r} \right) = 0 \] \[ \implies \psi = A \log r + B \] \begin{note*} In $\RR^n$, with $n \ge 3$, $\nabla^2 \psi = 0$ has the symmetric solution \[ \psi = \frac{A}{r^{n - 2}} + B \] \end{note*} \noindent We can get more solutions by differentiating, for example if $\psi = \frac{1}{r}$ is a solution in $\RR^3$, then so too \[ \tilde{\psi}(\bf{x}) = \bf{d} \cdot \nabla \left( \frac{1}{r} \right) = -\frac{\bf{d} \cdot \bf{x}}{r^3} \] with $\bf{d}$ a constant vector. (This is a potential for a \emph{dipole} in electromagnetism). \subsubsection*{Boundary Conditions} Often boundary conditions are important. For example: Solve \[ \nabla^2 \psi = \begin{cases} -\rho_0 & r \le R \\ 0 & r > R \end{cases} \] with $\rho_0$ constant. (for example gravitational potential for a planet with constant density). To get a unique solution, we require \begin{itemize} \item $\psi(r = 0)$ non-singular \item $\psi(r) \to 0)$ as $r \to \infty$ \item $\psi$ and $\psi'$ continuous. \end{itemize} Can check that if $\psi(r) = r^p$ then \[ \nabla^2 = p(p + 1)r^{p - 2} \] (in $\RR^3$). With spherical symmetry, we have \[ \psi(r) = \frac{A}{r} + B - \frac{1}{6} \rho_0 r^2 \qquad r \le R \] $\psi(r = 0)$ non-singular $\implies A = 0$. For $r > R$, \[ \psi(r) = \frac{c}{r} + D \qquad r > R \] $\psi(r \to \infty) \to 0 \implies D = 0$. Then we patch these together at $r = R$: \[ \psi(r = R) = B - \frac{1}{6} \rho_0 R^2 = \frac{C}{R} \] \[ \psi'(r = R) = -\frac{1}{3} \rho_0 R = -\frac{c}{R^2} \] \[ \implies \psi(r) = \begin{cases} \frac{1}{6} \rho_0 (3R^2 - r^2) & r \le R \\ \frac{1}{3} \rho_0 \frac{R^3}{r} & r > R \end{cases} \] \begin{center} \includegraphics[width=0.6\linewidth] {images/76b61b3e9a4111ec.png} \end{center} \subsubsection*{General Results} If we solve $\nabla^2 \psi = -\rho$ in some region $V \subset \RR^3$, then there are two common boundary conditions on $\partial V$. \begin{itemize} \item Dirichlet (D): Fix $\psi(\bf{x}) = f(\bf{x})$ on $\partial V$. \item Neumann (N): Fix $\bf{n} \cdot \nabla \psi(\bf{x}) = g(\bf{x})$ on $\partial V$, with $\bf{n}$ the outward pointing normal. \end{itemize} \begin{notation*} Sometimes this is written as \[ \dfrac{\psi}{\bf{n}} \equiv \bf{n} \cdot \nabla \psi \] or even \[ \dfrac{\psi}{n} \equiv \bf{n} \cdot \nabla \psi \] \end{notation*} \begin{claim*} There is a unique solution to the Poisson equation on $V$ with either D or N boundary conditions specified on each $\partial V$. \end{claim*} \begin{note*} Unique up to constant for N. \end{note*} \begin{proof} Let $\psi_1$ and $\psi_2$ satisfy Poisson and \[ \psi = \psi_1 - \psi_2 \] Then $\nabla^2 \phi = 0$ with $\psi = 0$ or $\bf{n} \cdot \nabla \psi = 0$ on each $\partial V$. \begin{align*} \int_v \nabla \cdot (\psi \nabla \psi) \dd V &= \int_V (\nabla \psi \cdot \nabla \psi + \psi \nabla^2 \psi) \dd V \\ &= \int_V |\nabla \psi|^2 \dd V \\ &= \int_{\partial V} \psi \nabla \psi \cdot \bf{\dd s} &&\text{by Divergence Theorem} \\ &= \int_{\partial V} \psi(\bf{n} \cdot \nabla \psi) \dd s \\ &= 0 &&\text{by one of the boundary conditions} \\ \implies |\nabla \psi|^2 &= 0 &&\text{in $V$} \\ \implies \nabla \psi &= 0 &&\text{in $V$} \\ \implies \psi &\text{is constant} \end{align*} If Dirichlet $\implies \psi = 0$ on $\partial V \implies \psi = 0$ everywhere. \end{proof} \begin{note*} Strictly for bounded $V$, but we can work harder and extend to, for example $\RR^3$. \end{note*} \begin{note*} \begin{itemize} \item If we can find, for example an isotropic solution, then this is \emph{the} solution. \item Sometimes there may be \emph{no} solution. \begin{example*} Solve $\nabla^2 \psi = \rho(x)$ with $\bf{n} \cdot \nabla \psi = g(\bf{x})$ on $\partial V$. Then \[ \int_V \nabla^2 \psi \dd V = \int_{\partial V} \nabla \psi \cdot \bf{\dd s} \] so a solution can exist only if \[ \int_V \rho \dd v = \int_{\partial V} g(\bf{x}) \dd s \] \end{example*} \item The proof uses Green's first identity: \begin{align*} \int_V \psi \nabla^2 \psi \dd V &= -\int \nabla \phi \cdot \nabla \psi \dd V + \int_S \phi \nabla \psi \cdot \bf{\dd s} \end{align*} (with $\phi = \psi$) This follows from the divergence theorem. Or by anti-symmetry \[ \int_V (\phi \nabla^2 \psi - \psi \nabla^2 \phi) \dd V = \int_S (\phi \nabla \psi - \psi \nabla \phi) \cdot \bf{\dd s} \] This is Green's second identity. \end{itemize} \end{note*}