1
$\begingroup$

Suppose we have a bounded domain $D$ with smooth boundary, with $G(x,y)$ being the Green's function for the Poisson equation on $D$, i.e. $G(x,\cdot)=0$ on $\partial D$ and $\Delta_y G(x,y)=\delta_x(y)$. We know that the solution of $\Delta u=f, u|_{\partial D}=0$ is given by $u(x)=\int_D{G(x,y)\,f(y)\mathrm dy}$. Its physical interpretation is exactly the Coulomb's law and the superposition principle.

Also we have a formula for the Dirichlet problem: $$u(x)=\int_{\partial D}{K(x,y)u(y)\mathrm dy}, \text{where }K(x,y)=\frac{\partial}{\partial n_y}G(x,y)$$

Realizing that the normal derivative of electric potential is the surface charge density on a conductor, I think there might be some physical interpretation of this using the Coulomb's law involving the surface charge. However I failed to find one because $\partial D$ cannot be a conductor if $u|_{\partial D}$ is nonconstant.

The Neumann condition is trickier. Consider a different Green's function $G_N(x,y)$ such that $\dfrac{\partial}{\partial n}G(x,\cdot)=c$ on $\partial D$ and $\Delta_y G(x,y)=\delta_x(y)$. The constant $c$ must be $1$ over the area of $\partial D$ by compatibility. (I read it from here.) By Green's representation formula, one can get $$u(x)=C-\int_{\partial D}{G_N(x,y)\frac{\partial u}{\partial n_y}\mathrm dy},$$where $C$ is a constant.

Again, is there any interpretation here? In this case a basic and not necessarily physical interpretation also helps because I know too few about Neumann condition. (I don't know the physical background of Neumann condition, nor can I find $G_N$ on simple domains like balls and half planes.)

$\endgroup$
0
$\begingroup$

There is an interpretation for the Dirichlet condition, i.e. $u(x)=\int_{\partial D}{K(x,y)u(y)\mathrm dy}, \text{where }K(x,y)=\frac{\partial}{\partial n_y}G(x,y)$, due to a physics friend of mine, H. Lam.

Given a charge distribution $Q$, we write the potential of these charges given by Coulomb's Law as $U(Q)$, a function on $\mathbb R^n$. Given a potential function $u$, we write the potential energy of the charge distribution $Q$ in $u$ as $uQ$. We have $U(Q_1)Q_2=U(Q_2)Q_1$. In other words $U(\cdot)\cdot$ is a symmetric bilinear functional on the space of charge distributions.

Now given $x\in D$. Consider the first scenario where $\mathbb R^n-D$ is a conductor and there is a unit point charge $Q$ at $x$. The induced charge $Q_i$ lies in $\partial D$ and we have $G(x,\cdot)=U(Q+Q_i)$ (here $G(x,\cdot)$ are defined to be $0$ outside $D$). By a property of conductors, $K(x,y)=\dfrac{\partial}{\partial n_y}G(x,y)$ is minus the surface charge density of $Q_i$.

Consider another scenario where the potential inside $D$ is given by $u$. Let $Q_u$ be a possible charge distribution that realizes this scenario. As $u$ is harmonic, $Q_u$ lies in $\mathbb R^n-D$.

Back to the formula. We have $u(x)=u(Q)=U(Q_u)Q$ and $RHS=\int_{\partial D}{K(x,y)u(y)\mathrm dy}=-U(Q_u)Q_i$. Then $LHS-RHS=U(Q_u)(Q+Q_i)=U(Q+Q_i)Q_u=G(x,\cdot)Q_u=0$ as $Q_u$ is complete outside $D$.

In spirit this is the same as the mathematical proof, as $U(Q_1)Q_2=\langle U(Q_1),\Delta U(Q_2) \rangle_{L^2(\mathbb R^n)}$, which can be proved to be symmetric by integration by parts.

As a side observation, this method can also give an interpretation of the symmetry of Green's function, or Principle of Reciprocity as is referred to in electrostatics.

Let $x,y\in D$, we want to prove $G(x,y)=G(y,x)$. Suppose the outside of $D$ is a conductor. In one scenario, there is a unit point charge $Q_x$ at $x$. The induced charge on $\partial D$ is denoted by $Q_x^i$. We have $U(Q_x+Q^i_x)=G(x,\cdot)$. In other scenario, we similarly define $Q_y$ and $Q^i_y$. Note that $U(Q_x+Q^i_x)Q^i_y=G(x,\cdot)Q^i_y=0$ as $Q^i_y$ lies on $\partial D$. Therefore

$$ G(x,y)=U(Q_x+Q^i_x)Q_y=U(Q_x+Q^i_x)(Q_y+Q^i_y)=U(Q_y+Q^i_y)(Q_x+Q^i_x)=G(y,x) $$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.