Laplace Equation, Harmonic Functions, Mean-Value Property, and Maximum Principle

Intuition Beginner

The Laplace equation says that a quantity is in equilibrium with its surroundings. Imagine a thin sheet of metal whose edges are held at fixed temperatures. After enough time has passed, the temperature inside the sheet stops changing. At each interior point the temperature is exactly the average of the temperatures at neighbouring points.

If the interior temperature were higher than the average of its neighbours, heat would flow out and the temperature would drop. If it were lower, heat would flow in and the temperature would rise. The steady state is reached when no such flow happens. The equation expressing this equilibrium is the Laplace equation, written in shorthand as Lu = 0 where L is the Laplacian operator that measures how much a function differs from the average of its neighbours.

The same equation governs a surprising range of physical settings. The electric potential in a region without electric charge satisfies it. The gravitational potential in a region without mass satisfies it. The shape of a soap film stretched across a wire loop satisfies it. The velocity potential of an ideal fluid in steady flow satisfies it. A function that satisfies the Laplace equation is called a harmonic function, and the field that studies harmonic functions is called potential theory.

Two properties make harmonic functions easy to recognise. First, the value at any interior point equals the average value on any sphere around that point that fits inside the region. This is the mean-value property. Second, a harmonic function cannot have a strict maximum or minimum inside the region; its largest and smallest values are always attained on the boundary. This is the maximum principle. Together, the two properties say that harmonic functions are as smooth and unsurprising as a function can be while still being non-constant.

Some examples. The function f(x, y) = 3x + 4y + 7 is harmonic in the plane: the Laplacian of a linear function is zero. The function f(x, y) = xx - yy is harmonic: the second derivative in x is 2 and the second derivative in y is -2, and they cancel. The function log r (where r is the distance from the origin) is harmonic everywhere in the plane except at the origin itself. In three dimensions, the function 1/r is harmonic everywhere except at the origin, and this is the function describing the gravitational potential of a point mass and the electric potential of a point charge.

The one-sentence takeaway: harmonic functions are the equilibrium shapes that arise when a quantity must average its neighbouring values at every interior point, and they obey a mean-value property and a maximum principle that make them the smoothest possible non-constant functions.

Visual Beginner

Picture a square region of metal with its four edges held at four different fixed temperatures. The temperature inside settles into a smooth surface whose shape is determined entirely by the boundary values. Hot spots near a hot edge fade smoothly into cold spots near a cold edge; there are no sudden bumps or hidden hot spots in the middle. The surface looks like a soap film: gently curved, with no kinks, attaining its highest and lowest temperatures only at the boundary.

The mean-value property is the picture in the middle panel: at the centre of any small sphere fitting inside the region, the function value equals the average of the function values on the sphere. The maximum principle is the picture in the right panel: the highest and lowest function values appear only on the boundary, never in the interior of the region.

Worked example Beginner

We verify directly that the function $u(x,y)=x^2-y^2$ is harmonic on the entire plane, and then use the mean-value property to compute its average value on the unit circle around the origin.

Step 1. Compute the second derivative of $u$ in the $x$ direction. Differentiating $u$ once with respect to $x$ gives $2x$. Differentiating again gives $2$.

Step 2. Compute the second derivative of $u$ in the $y$ direction. Differentiating $u$ once with respect to $y$ gives $-2y$. Differentiating again gives $-2$.

Step 3. Add the two second derivatives: $2+(-2)=0$. The Laplacian of $u$ is zero, so $u$ is harmonic.

Step 4. Apply the mean-value property at the origin. Since $u$ is harmonic and the origin is the centre of the unit circle, the average of $u$ on the unit circle equals $u(0,0)=0$.

Step 5. Sanity check by direct computation. On the unit circle, $x=\cos \theta$ and $y=\sin \theta$, so $u={\cos }^2\theta -{\sin }^2\theta =\cos (2\theta )$. The average of $\cos (2\theta )$ over one full revolution is $0$, matching the prediction.

What this tells us: the mean-value property is a real, computable consequence of harmonicity, and it lets us extract integral averages without computing them directly.

Check your understanding Beginner

Formal definition Intermediate+

Let $\Omega \subseteq {\mathbb{R}}^n$ be an open set and let $u:\Omega \to \mathbb{R}$ be a $C^2$ function. The Laplacian of $u$ is the second-order linear differential operator $$\Delta u=\sum _{i=1}^n\frac{{\partial }^{2}u}{\partial x_i^2}.$$ The function $u$ is called harmonic on $\Omega$ when $\Delta u=0$ pointwise on $\Omega$, subharmonic when $\Delta u\ge 0$, and superharmonic when $\Delta u\le 0$ ^{[Evans 2010 §2.2]}.

The Laplacian is invariant under rigid motions of ${\mathbb{R}}^n$: orthogonal transformations and translations preserve harmonicity. Up to multiplication by a non-zero scalar, the Laplacian is the unique second-order linear differential operator with constant coefficients enjoying this rigid-motion invariance.

Polar and spherical coordinates. In ${\mathbb{R}}^2$ with $(x,y)=(r\cos \theta ,r\sin \theta )$: $$\Delta =\frac{{\partial }^2}{\partial r^2}+\frac{1}{r}\frac{\partial }{\partial r}+\frac{1}{r^2}\frac{{\partial }^2}{\partial {\theta }^2}.$$ In ${\mathbb{R}}^n$ for $n\ge 3$ with spherical coordinates $(r,\omega )\in (0,\infty )\times S^{n-1}$: $$\Delta =\frac{{\partial }^2}{\partial r^2}+\frac{n-1}{r}\frac{\partial }{\partial r}+\frac{1}{r^2}{\Delta }_{S^{n-1}},$$ where ${\Delta }_{S^{n-1}}$ is the spherical Laplacian.

Separation of variables. In the plane, the ansatz $u(x,y)=X(x)Y(y)$ reduces the Laplace equation to $X^{\prime \prime }/X=-Y^{\prime \prime }/Y=-\lambda$ for some constant $\lambda$. For $\lambda >0$, the $x$-equation has solutions $\sin (\sqrt{\lambda }x)$ and $\cos (\sqrt{\lambda }x)$, and the $y$-equation has solutions $\sinh (\sqrt{\lambda }y)$ and $\cosh (\sqrt{\lambda }y)$. The roles reverse when $\lambda <0$, and the case $\lambda =0$ gives linear solutions. Sums of such product solutions, fitted to boundary conditions, produce the Fourier-series solution to the Dirichlet problem on rectangles.

Spherical harmonic separation. In ${\mathbb{R}}^3$, the ansatz $u(r,\omega )=R(r)Y(\omega )$ yields $R(r)=r^{\ell }$ or $R(r)=r^{-(\ell +1)}$ paired with $Y$ a spherical harmonic of degree $\ell$. The functions $r^{\ell }{Y}_{\ell }^m(\omega )$ and $r^{-(\ell +1)}{Y}_{\ell }^m(\omega )$ are the building blocks of multipole expansions in potential theory.

Counterexamples to common slips Intermediate+

• Harmonicity is not preserved under multiplication. If $u,v$ are harmonic, the product $uv$ generally is not: $\Delta (uv)=u\Delta v+v\Delta u+2\nabla u\cdot \nabla v=2\nabla u\cdot \nabla v$, which vanishes only when the gradients are orthogonal. So $x$ and $y$ are harmonic on ${\mathbb{R}}^2$, but $xy$ is harmonic (because $\nabla x\cdot \nabla y=0$), whereas $x^2$ is not harmonic.

• Harmonicity does not transfer to arbitrary coordinate changes. The Laplacian is invariant under rigid motions but not under general diffeomorphisms. Under a conformal change of coordinates in ${\mathbb{R}}^2$, the Laplacian rescales by the conformal factor: if $w=f(z)$ is holomorphic and $\tilde{u}(z)=u(f(z))$, then $\Delta \tilde{u}=|f^{\prime }(z){|}^2(\Delta u)(f(z))$, so harmonicity is preserved in $n=2$ under holomorphic changes (one face of the link to complex analysis). In $n\ge 3$ the only conformal maps are Möbius transformations, and harmonicity is preserved only up to a Kelvin-transform factor.

• Liouville's theorem requires boundedness, not just harmonicity. A harmonic function on all of ${\mathbb{R}}^n$ need not be constant: $u(x)=x_1$ is harmonic and non-constant. The statement that bounded harmonic functions on ${\mathbb{R}}^n$ are constant uses both harmonicity and boundedness.

• The fundamental solution differs by dimension. In $n=2$, the fundamental solution of $-\Delta$ has a logarithmic singularity; in $n\ge 3$ it has a power singularity. Confusing the two formulas leads to incorrect Newtonian-potential computations.

Key theorem with proof Intermediate+

Theorem (mean-value property for harmonic functions; Gauss 1840). Let $u$ be harmonic on the open set $\Omega \subseteq {\mathbb{R}}^n$ and let $\overline{B_r(x_0)}\subset \Omega$. Then $$u(x_0)=\frac{1}{|\partial B_r(x_0)|}{\int }_{\partial B_r(x_0)}u(y)dS(y)=\frac{1}{|B_r(x_0)|}{\int }_{B_r(x_0)}u(y)dy.$$ The value at the centre equals the average over any sphere (or ball) around it that fits inside $\Omega$.

Proof. Define $\phi (r)={\text{fint}}_{\partial B_r(x_0)}udS$, the spherical average over the sphere of radius $r$ centred at $x_0$. We will show $\phi$ is constant in $r$ and that ${\lim }_{r\to 0^{+}}\phi (r)=u(x_0)$.

Step 1 (limit at zero). Continuity of $u$ at $x_0$ implies ${\lim }_{r\to 0^{+}}{\text{fint}}_{\partial B_r(x_0)}udS=u(x_0)$, since the integrand approaches $u(x_0)$ uniformly on the shrinking sphere.

Step 2 (constant in $r$). Parametrise $\partial B_r(x_0)$ as $y=x_0+r\omega$ for $\omega \in \partial B_1(0)=S^{n-1}$, so $dS(y)=r^{n-1}dS(\omega )$ and $|\partial B_r(x_0)|=r^{n-1}|S^{n-1}|$. Therefore $$\phi (r)=\frac{1}{|S^{n-1}|}{\int }_{S^{n-1}}u(x_0+r\omega )dS(\omega ).$$ Differentiating under the integral sign (legitimate since $u$ is $C^2$ and the sphere is bounded): $${\phi }^{\prime }(r)=\frac{1}{|S^{n-1}|}{\int }_{S^{n-1}}\nabla u(x_0+r\omega )\cdot \omega dS(\omega ).$$ Change variables back to $\partial B_r(x_0)$: writing $\nu (y)=(y-x_0)/r$ for the outward unit normal, $${\phi }^{\prime }(r)=\frac{1}{|S^{n-1}|r^{n-1}}{\int }_{\partial B_r(x_0)}\nabla u(y)\cdot \nu (y)dS(y)=\frac{1}{|\partial B_r(x_0)|}{\int }_{\partial B_r(x_0)}\frac{\partial u}{\partial \nu }(y)dS(y).$$ By the divergence theorem [02.07.04, 03.04.05] applied to $\nabla u$ on $B_r(x_0)$, $${\int }_{\partial B_r(x_0)}\frac{\partial u}{\partial \nu }(y)dS(y)={\int }_{B_r(x_0)}\Delta u(y)dy=0,$$ because $u$ is harmonic. Therefore ${\phi }^{\prime }(r)=0$ for every $r>0$ with $\overline{B_r(x_0)}\subset \Omega$.

Step 3 (sphere mean equals centre value). Combining Steps 1 and 2, $\phi (r)={\lim }_{s\to 0^{+}}\phi (s)=u(x_0)$. This is the sphere version.

Step 4 (ball mean equals centre value). Integrate the sphere-mean identity in the radial variable: $${\int }_{B_r(x_0)}u(y)dy={\int }_0^r{\int }_{\partial B_s(x_0)}udSds={\int }_0^{r}u(x_0)|\partial B_s(x_0)|ds=u(x_0){\int }_0^r{s}^{n-1}|S^{n-1}|ds=u(x_0)|B_r(x_0)|.$$ Dividing by $|B_r(x_0)|$ gives the ball version.

$\square$

Bridge. The mean-value property builds toward 02.13.02 the Poisson equation, where the same identity acquires a non-zero right side encoding the contribution of sources, and appears again in 02.13.03 the heat equation as the parabolic mean-value identity over cylindrical heat balls. The central insight is that the divergence theorem on a ball converts the harmonic condition $\Delta u=0$ into a vanishing flux of $\nabla u$ across the sphere, and this is exactly the foundational reason every harmonic function is rigidly constrained: knowing values on one sphere forces the value at the centre. Putting these together with the simple-function approximation of integrals from 02.07.04, one identifies harmonic functions with the kernel of an averaging operator on a Hilbert space, the formulation behind Weyl's projection method ^{[Weyl 1940]}. The bridge is between the descriptive theory (a PDE governing pointwise behaviour) and the integral-geometric theory (a function pinned down by its boundary values via averaging), and the pattern generalises to subharmonic functions (where one inequality replaces the equation), to the Riesz decomposition theorem on Riesz potentials, and to Brownian-motion characterisations ^{[Kakutani 1944]}.

Exercises Intermediate+

Advanced results Master

The advanced theory of the Laplace equation organises around four pillars: the fundamental solution and the Newtonian potential, the Green-function and Poisson-kernel apparatus for solving the Dirichlet problem on bounded domains, the regularity theory (analyticity, Harnack, removable singularities), and the variational and probabilistic characterisations (Dirichlet principle, Weyl lemma, Brownian-motion mean).

Theorem 1 (fundamental solution; Laplace 1782, Poisson 1813). The function $$\Phi (x)=\{\begin{array}{ll}-\frac{1}{2\pi }\log |x|,& n=2,\\ \frac{1}{n(n-2){\omega }_n}|x{|}^{2-n},& n\ge 3,\end{array}$$ where ${\omega }_n=|B_1(0)|$ is the volume of the unit ball in ${\mathbb{R}}^n$, is the fundamental solution of the Laplace operator ^{[Laplace 1782]}. It satisfies $\Delta \Phi =0$ on ${\mathbb{R}}^n\setminus \{0\}$ and $-\Delta \Phi ={\delta }_0$ in the sense of distributions on ${\mathbb{R}}^n$. The normalisation constants are fixed by the requirement that the surface integral $-{\int }_{\partial B_r(0)}{\partial }_{\nu }\Phi dS$ equals $1$ for every $r>0$.

Theorem 2 (Newtonian potential; Poisson 1813). Let $f\in C_c^2({\mathbb{R}}^n)$ be a compactly supported $C^2$ function. The function $$u(x)={\int }_{{\mathbb{R}}^n}\Phi (x-y)f(y)dy$$ satisfies $-\Delta u=f$ on ${\mathbb{R}}^n$, with $u\in C^2({\mathbb{R}}^n)$ ^{[Poisson 1813]}. The Newtonian potential is the unique solution decaying at infinity (in $n\ge 3$); in $n=2$ uniqueness fails up to addition of a constant. The construction lifts directly to distributional $f$ via the convolution structure.

The construction is the convolution $u=\Phi \ast f$. Smoothness of $u$ comes from the smoothness of $f$ combined with the locally integrable singularity of $\Phi$. The identity $\Delta u=(\Delta \Phi )\ast f=-{\delta }_0\ast f=-f$ is the formal calculation that the distributional treatment makes rigorous (Schwartz 1950 Théorie des distributions).

Theorem 3 (Green function and Poisson kernel on the ball). Let $B_R=B_R(0)\subset {\mathbb{R}}^n$. The Green function for the Dirichlet problem on $B_R$ is $$G(x,y)=\Phi (x-y)-\Phi (\frac{|x|}{R}(y-x^{\ast }))$$ where $x^{\ast }=R^{2}x/|x{|}^2$ is the Kelvin inversion of $x$ through $\partial B_R$. The Green function is symmetric: $G(x,y)=G(y,x)$. The Poisson kernel is its inward normal derivative: $$P(x,y)=-\frac{\partial G}{\partial {\nu }_y}(x,y)=\frac{R^2-|x{|}^2}{R{\omega }_{n-1}|x-y{|}^n},x\in B_R,y\in \partial B_R,$$ where ${\omega }_{n-1}=|\partial B_1|$ is the surface area of the unit sphere ^{[Evans 2010 §2.2.4]}. The solution of the Dirichlet problem $\Delta u=0$ on $B_R$, $u{|}_{\partial B_R}=g$ for $g\in C(\partial B_R)$ is given by the Poisson integral: $$u(x)={\int }_{\partial B_R}P(x,y)g(y)dS(y).$$ The integral defines a function in $C^{\infty }(B_R)\cap C(\overline{B_R})$ taking the prescribed boundary values, and the formula recovers the centre-value mean-value identity when $x=0$: $P(0,y)=1/|\partial B_R|$, so $u(0)={\text{fint}}_{\partial B_R}gdS$.

Theorem 4 (smoothness and analyticity). Every harmonic function on $\Omega \subseteq {\mathbb{R}}^n$ is real-analytic on $\Omega$. In particular, harmonic functions are $C^{\infty }$, and the Cauchy estimates hold: for every multi-index $\alpha$ and every ball $B_r(x_0)\subset \Omega$ with $B_{2r}(x_0)\subset \Omega$, $$|D^{\alpha }u(x_0)|\le \frac{C_n^{|\alpha |}|\alpha |!}{r^{|\alpha |}}\underset{\overline{B_r(x_0)}}{\max }|u|.$$

The Cauchy estimates follow from the Poisson formula and differentiation under the integral sign; the analyticity follows from the Cauchy estimates and the Taylor remainder bound (Gilbarg-Trudinger §2.7).

Theorem 5 (Harnack inequality; Harnack 1887). For every connected open $\Omega \subseteq {\mathbb{R}}^n$ and every compact $K\subset \Omega$, there exists a constant $C=C(K,\Omega ,n)$ such that for every non-negative harmonic $u$ on $\Omega$, $$\underset{K}{\sup }u\le C\underset{K}{\inf }u.$$ On the ball case $K=\overline{B_{r/2}(x_0)}\subset B_r(x_0)\subset \Omega$, the constant takes the explicit form $C=3^n$ in ${\mathbb{R}}^n$ ^{[Evans 2010 §2.2.4]}.

The Harnack inequality is sharper than the maximum principle for non-negative harmonic functions: the maximum principle says $\inf u$ on $\overline{K}$ is attained at a point but gives no quantitative comparison with $\sup u$; Harnack quantifies the comparison, making harmonic functions effectively "equi-distributed" up to bounded constants on compactly contained subdomains.

Theorem 6 (strong maximum principle; Hopf 1927). Let $u\in C^2(\Omega )\cap C(\overline{\Omega })$ be harmonic on a connected open $\Omega \subseteq {\mathbb{R}}^n$. If $u$ attains its maximum at an interior point $x_0\in \Omega$, then $u$ is constant on $\Omega$. The same conclusion holds for subharmonic $u$ at an interior maximum.

Proof sketch. The set ${\Omega }^{\prime }=\{x\in \Omega :u(x)=M\}$ where $M=\max u$ is closed in $\Omega$ by continuity. Take $x_0\in {\Omega }^{\prime }$ and a ball $\overline{B_r(x_0)}\subset \Omega$. By the mean-value identity, $M=u(x_0)={\text{fint}}_{\partial B_r(x_0)}u\le M$, with equality forcing $u\equiv M$ on $\partial B_r(x_0)$. Extending radially, $u\equiv M$ on $\overline{B_r(x_0)}$, so ${\Omega }^{\prime }$ is also open. By connectedness of $\Omega$, ${\Omega }^{\prime }=\Omega$. $\square$

Theorem 7 (Hopf boundary point lemma; Hopf 1952). Let $u$ be harmonic on a bounded $C^2$ domain $\Omega$ and suppose $u$ attains its maximum at a boundary point $x_0\in \partial \Omega$. If $u$ is non-constant on $\Omega$, then the outward normal derivative at $x_0$ is strictly positive: $\partial u/\partial \nu (x_0)>0$. The boundary point lemma is the quantitative complement to the strong maximum principle and the main tool in Hopf-style comparison arguments for elliptic equations.

Theorem 8 (Liouville's theorem). Every bounded harmonic function on ${\mathbb{R}}^n$ is constant. More generally, a harmonic function $u$ on ${\mathbb{R}}^n$ with polynomial growth $|u(x)|\le C(1+|x{|}^k)$ is a harmonic polynomial of degree at most $k$.

Proof. For boundedness, Exercise 7 supplies the proof via the mean-value identity. For polynomial growth, differentiate the Poisson identity $k+1$ times: the Cauchy estimate gives $|D^{k+1}u(x_0)|\le C/R$ for every $R>0$ (the polynomial growth of $u$ is absorbed by $R^{n+k}$ in the denominator versus $R^{n+k+1}$), so $D^{k+1}u\equiv 0$, hence $u$ is a polynomial of degree at most $k$. $\square$

Theorem 9 (removable singularities; Painlevé 1888). Let $u$ be harmonic and bounded on $B_r(x_0)\setminus \{x_0\}$. Then $u$ extends uniquely to a harmonic function on $B_r(x_0)$. The same holds when $u(x)=o(|x-x_0{|}^{2-n})$ in $n\ge 3$ or $u(x)=o(\log |x-x_0|)$ in $n=2$ near $x_0$.

Proof sketch. Let $v$ be the harmonic extension of $u{|}_{\partial B_r(x_0)}$ to $\overline{B_r(x_0)}$ via the Poisson kernel. Then $u-v$ is harmonic on $B_r(x_0)\setminus \{x_0\}$, vanishes on $\partial B_r(x_0)$, and is bounded near $x_0$. Comparison with $\epsilon (\Phi (x-x_0)-\Phi (r{e}_1))$ for arbitrary $\epsilon >0$ via the maximum principle on $B_r(x_0)\setminus B_{\delta }(x_0)$ forces $|u-v|\le \epsilon (\Phi (x-x_0)-\Phi (r{e}_1))$; letting $\delta \to 0$ and then $\epsilon \to 0$ gives $u\equiv v$ on the punctured ball, and $v$ is the desired extension. $\square$

Theorem 10 (Schwarz reflection principle). Let ${\Omega }^{+}\subseteq {\mathbb{R}}_{+}^n=\{x_n>0\}$ be a region whose boundary $\partial {\Omega }^{+}$ contains a piece $\Sigma \subseteq \{x_n=0\}$ of the hyperplane. Let $u$ be harmonic on ${\Omega }^{+}$ and continuous on ${\Omega }^{+}\cup \Sigma$ with $u=0$ on $\Sigma$. Then the odd reflection $\tilde{u}(x_1,\dots ,x_{n-1},x_n)=-u(x_1,\dots ,x_{n-1},-x_n)$ defines a harmonic function on the open set ${\Omega }^{-}=\{(x_1,\dots ,x_{n-1},-x_n):(x_1,\dots ,x_n)\in {\Omega }^{+}\}$, and the combined function $u\cup \tilde{u}$ is harmonic on ${\Omega }^{+}\cup \Sigma \cup {\Omega }^{-}$. The result lifts every Dirichlet-trace-zero piece of a harmonic function to a harmonic function across the boundary.

Theorem 11 (Perron method for the Dirichlet problem; Perron 1923). Let $\Omega \subset {\mathbb{R}}^n$ be a bounded open set and $g\in C(\partial \Omega )$. The Perron solution of the Dirichlet problem is $$u(x)=\sup \{v(x):v\text{ subharmonic on }\Omega \text{ with }\underset{x\to y}{\mathrm{limsup}}v(x)\le g(y)\text{ for every }y\in \partial \Omega \}.$$ The function $u$ is harmonic on $\Omega$ ^{[Perron 1923]}. Whether $u$ continuously attains the boundary values $g$ at a given point $y_0\in \partial \Omega$ is a separate boundary-regularity question: it holds when $y_0$ is a regular boundary point, that is, when there exists a barrier function $w$ (a non-negative superharmonic function in $\Omega$ near $y_0$ with $w(y_0)=0$ and $w>0$ elsewhere on $\partial \Omega \cap B_{\delta }(y_0)$).

Theorem 12 (Wiener's criterion; Wiener 1924). A boundary point $y_0\in \partial \Omega$ is regular for the Dirichlet problem on $\Omega \subset {\mathbb{R}}^n$ if and only if the series $$\sum _{k=1}^{\infty }{2}^{(n-2)k}\cdot \mathrm{cap}\left(A_k(y_0)\cap {\Omega }^c\right)=\infty ,$$ where $A_k(y_0)=\{x:2^{-(k+1)}\le |x-y_0|\le 2^{-k}\}$ and $\mathrm{cap}$ denotes Newtonian capacity ^{[Wiener 1924]}. The Wiener criterion gives a complete characterisation of boundary-point regularity in terms of the capacity-thickness of the complement near the point.

The criterion identifies the boundary failures: an isolated cusp pointing inward has insufficient capacity-thickness and creates an irregular boundary point; a Lebesgue thorn (a cone pointing into ${\Omega }^c$) is regular by the Lebesgue thorn theorem 1912; the Lebesgue spine (a thorn pointing into $\Omega$) is the canonical example of an irregular boundary point.

Theorem 13 (Dirichlet principle and Weyl's lemma; Hilbert 1900, Weyl 1940). Let $\Omega \subset {\mathbb{R}}^n$ be a bounded $C^1$ domain and $g\in H^{1/2}(\partial \Omega )$. The Dirichlet energy is $$E(v)=\frac{1}{2}{\int }_{\Omega }|\nabla v{|}^{2}dx.$$ Among all $v\in H^1(\Omega )$ with trace $v{|}_{\partial \Omega }=g$, the energy $E$ attains a unique minimum, and the minimiser $u$ satisfies $\Delta u=0$ in the distributional sense on $\Omega$ ^{[Hilbert 1900]}.

Weyl's lemma ^{[Weyl 1940]}: every distributional solution of $\Delta u=0$ on $\Omega$ is in fact $C^{\infty }(\Omega )$ (in fact real-analytic). The Dirichlet principle and Weyl's lemma together resolved the rigorous gap left by Riemann and Hilbert: Riemann had used the Dirichlet principle without proving the existence of a minimiser (Weierstrass 1870 pointed out the gap with the counterexample of a coercive functional with no minimum on $C^1$); Hilbert's problem 20 in the 1900 Paris ICM address asked for a rigorous existence proof; Hilbert himself 1904 gave one via direct methods of the calculus of variations; Weyl 1940 closed the smoothness loop by showing the minimiser (initially only $H^1$) is actually smooth.

Theorem 14 (Brownian motion characterisation; Kakutani 1944). Let $\Omega \subset {\mathbb{R}}^n$ be a bounded open set and let $\{B_t\}$ be a Brownian motion in ${\mathbb{R}}^n$. For $x\in \Omega$, let ${\tau }_x=\inf \{t>0:B_t\notin \Omega \mid B_0=x\}$ be the first-exit time from $\Omega$ for a Brownian motion started at $x$. Let ${\nu }_x$ be the distribution of $B_{{\tau }_x}$ on $\partial \Omega$ (the harmonic measure on $\partial \Omega$ from $x$). Then the solution of the Dirichlet problem $\Delta u=0$ on $\Omega$ with continuous boundary values $g$ is $$u(x)={\int }_{\partial \Omega }g(y)d{\nu }_x(y)={\mathbb{E}}_x[g(B_{{\tau }_x})],$$ the expected value of $g$ at the Brownian first-exit point ^{[Kakutani 1944]}.

Theorem 15 (Hodge decomposition; Hodge 1941). Let $(M,g)$ be a compact oriented Riemannian manifold of dimension $n$ without boundary. The Hodge Laplacian on $k$-forms is ${\Delta }_H=d\delta +\delta d$. Every $L^2$ $k$-form $\omega$ has a unique orthogonal decomposition $\omega =d\alpha +\delta \beta +\gamma$ where $\alpha \in {\Omega }^{k-1}(M)$, $\beta \in {\Omega }^{k+1}(M)$, and $\gamma$ is harmonic (${\Delta }_H\gamma =0$). The space of harmonic $k$-forms is finite-dimensional and isomorphic to the de Rham cohomology $H_{dR}^k(M;\mathbb{R})$.

Synthesis. The theory of harmonic functions is the foundational reason that the Laplace operator is the canonical model for second-order elliptic PDEs. The central insight is that the divergence-theorem identity ${\int }_{\partial B}{\partial }_{\nu }udS={\int }_B\Delta udx$ converts the local PDE $\Delta u=0$ into the global integral-geometric mean-value identity, and this is exactly the structural bridge that makes harmonic functions rigid: a continuous function pinned to its sphere averages is forced to be real-analytic, to obey the maximum principle, and to be uniquely determined by its boundary values. Putting these together with the fundamental-solution apparatus, every harmonic function on a region with boundary is identified with a Poisson-integral representation in terms of its boundary trace, and every solution of $-\Delta u=f$ with a compactly supported source is identified with the Newtonian-potential convolution $\Phi \ast f$.

The pattern recurs through three escalations. First, harmonic functions on Euclidean space generalise to harmonic sections of vector bundles on Riemannian manifolds via the Hodge Laplacian 03.04.15; the Hodge decomposition theorem identifies the kernel with de Rham cohomology, the bridge between analysis and topology that Atiyah-Singer index theory later perfected. Second, the elliptic regularity of the Laplace operator generalises to general second-order divergence-form operators with measurable coefficients via De Giorgi 1957 and Nash 1958, opening modern non-linear analysis. Third, the Dirichlet principle generalises to general minimisation problems via the direct methods of the calculus of variations, and the Brownian-motion characterisation ^{[Kakutani 1944]} identifies harmonic measure with diffusion-process hitting measure, the bridge to stochastic potential theory and modern probabilistic methods.

The historical resolution of the Dirichlet principle by the joint work of Hilbert 1904, Lebesgue 1907, and Weyl 1940 is the foundational moment that converted nineteenth-century formal calculations into twentieth-century functional analysis: the descriptive existence-of-a-minimiser question is settled by direct-method weak-compactness arguments in Hilbert space (Hilbert), the regularity question is settled by Sobolev-space embedding and elliptic estimates (Weyl, then refined by Schauder and De Giorgi), and the boundary-attainment question is settled by Wiener's capacity-theoretic criterion 1924 and the Perron method 1923. The result is a complete theory in which every step from the boundary data to the smooth interior solution is rigorously controlled, and the same chain of ideas extends to general elliptic equations, parabolic equations, and degenerate problems via the same conceptual scaffolding.

Full proof set Master

Proposition 1 (uniqueness of the Dirichlet problem). Let $\Omega \subset {\mathbb{R}}^n$ be a bounded open set and $g\in C(\partial \Omega )$. If $u_1,u_2\in C^2(\Omega )\cap C(\overline{\Omega })$ both solve $\Delta u=0$ on $\Omega$ with $u{|}_{\partial \Omega }=g$, then $u_1=u_2$.

Proof. By linearity, $w=u_1-u_2$ is harmonic on $\Omega$ and continuous on $\overline{\Omega }$ with $w{|}_{\partial \Omega }=0$. By the weak maximum principle, ${\max }_{\overline{\Omega }}w={\max }_{\partial \Omega }w=0$. Applying the same to $-w$, ${\min }_{\overline{\Omega }}w=0$. Therefore $w\equiv 0$ on $\overline{\Omega }$.

$\square$

Proposition 2 (mean-value property characterises harmonicity; Koebe 1906). Let $u:\Omega \to \mathbb{R}$ be continuous on the open set $\Omega \subseteq {\mathbb{R}}^n$. If $u$ satisfies the sphere-mean identity $u(x_0)={\text{fint}}_{\partial B_r(x_0)}udS$ for every $x_0\in \Omega$ and every $r$ with $\overline{B_r(x_0)}\subset \Omega$, then $u\in C^{\infty }(\Omega )$ and $\Delta u=0$ on $\Omega$.

Proof. Choose a standard non-negative radial mollifier $\eta \in C_c^{\infty }(B_1(0))$ with $\int \eta =1$ and set ${\eta }_{\epsilon }(x)={\epsilon }^{-n}\eta (x/\epsilon )$. The convolution $u_{\epsilon }=u\ast {\eta }_{\epsilon }$ is $C^{\infty }$ on ${\Omega }_{\epsilon }=\{x:\mathrm{dist}(x,\partial \Omega )>\epsilon \}$.

For $x_0\in {\Omega }_{\epsilon }$: $$u_{\epsilon }(x_0)={\int }_{B_{\epsilon }(0)}u(x_0-y){\eta }_{\epsilon }(y)dy={\int }_0^{\epsilon }{\int }_{\partial B_s(0)}u(x_0-y)dS(y){\eta }_{\epsilon }(s)ds.$$

By hypothesis ${\int }_{\partial B_s(0)}u(x_0-y)dS(y)=u(x_0)|\partial B_s(0)|$ for every $s<\epsilon$, so: $$u_{\epsilon }(x_0)=u(x_0){\int }_0^{\epsilon }|\partial B_s(0)|{\eta }_{\epsilon }(s)ds=u(x_0){\int }_{B_{\epsilon }(0)}{\eta }_{\epsilon }=u(x_0).$$

So $u_{\epsilon }\equiv u$ on ${\Omega }_{\epsilon }$. Since $u_{\epsilon }$ is $C^{\infty }$, $u$ is $C^{\infty }$ on ${\Omega }_{\epsilon }$ for every $\epsilon >0$, hence $u\in C^{\infty }(\Omega )$.

Now $u$ is $C^2$. By differentiating the sphere-mean identity in $r$ as in the proof of Theorem (mean-value property), ${\int }_{B_r(x_0)}\Delta u=0$ for every ball. Letting $r\to 0$ and using continuity of $\Delta u$, $\Delta u(x_0)=0$.

$\square$

Proposition 3 (strong maximum principle). Let $\Omega \subset {\mathbb{R}}^n$ be open and connected, and let $u:\Omega \to \mathbb{R}$ be harmonic. If $u$ attains its maximum at an interior point, then $u$ is constant.

Proof. Let $M={\sup }_{\Omega }u$ and ${\Omega }_M=\{x\in \Omega :u(x)=M\}$. The set ${\Omega }_M$ is non-empty by hypothesis and closed in $\Omega$ by continuity of $u$. We show ${\Omega }_M$ is open.

Fix $x_0\in {\Omega }_M$ and choose $r>0$ with $\overline{B_r(x_0)}\subset \Omega$. By the ball mean-value property: $$M=u(x_0)=\frac{1}{|B_r(x_0)|}{\int }_{B_r(x_0)}u(y)dy.$$

Since $u\le M$ everywhere on $\overline{B_r(x_0)}$, the integrand $u(y)\le M$ pointwise. The integral can equal $M|B_r(x_0)|$ only when $u\equiv M$ almost everywhere on $B_r(x_0)$; continuity of $u$ promotes this to $u\equiv M$ everywhere on $B_r(x_0)$. So $B_r(x_0)\subset {\Omega }_M$, proving openness.

By connectedness of $\Omega$, the non-empty open-and-closed subset ${\Omega }_M$ equals $\Omega$. So $u\equiv M$ on $\Omega$.

$\square$

Proposition 4 (Poisson formula solves the Dirichlet problem on the ball). Let $g\in C(\partial B_R(0))$. The function $u:B_R(0)\to \mathbb{R}$ defined by $$u(x)={\int }_{\partial B_R(0)}P(x,y)g(y)dS(y),P(x,y)=\frac{R^2-|x{|}^2}{R{\omega }_{n-1}|x-y{|}^n},$$ satisfies $u\in C^{\infty }(B_R(0))\cap C(\overline{B_R(0)})$ with $\Delta u=0$ on $B_R(0)$ and $u{|}_{\partial B_R(0)}=g$.

Proof sketch. The smoothness is direct from differentiation under the integral sign, using that $|x-y|>0$ for $x\in B_R(0)$ and $y\in \partial B_R(0)$. The harmonicity ${\Delta }_{x}P(x,y)=0$ for $x\in B_R(0)$ follows by direct computation from $P(x,y)=-{\partial }_{{\nu }_y}G(x,y)$ with $G$ the Green function (Theorem 3); ${\Delta }_{x}G(x,y)=-{\delta }_y$ in distributions, so ${\Delta }_x{\partial }_{{\nu }_y}G(x,y)=0$ for $x\ne y$.

Boundary attainment uses that $P(x,y)dS(y)$ is a positive measure on $\partial B_R(0)$ with total mass $1$ (verified by $P(0,y)=1/|\partial B_R|$ and the maximum principle applied to $u\equiv 1$). As $x\to y_0\in \partial B_R(0)$, the measure $P(x,\cdot )dS$ concentrates at $y_0$: for $y$ bounded away from $y_0$, $P(x,y)\to 0$; for $y$ near $y_0$, the integrand is uniformly bounded times a function with total mass $1$. Continuity of $g$ then gives ${\lim }_{x\to y_0}u(x)=g(y_0)$. $\square$

Proposition 5 (Liouville's theorem and the polynomial-growth scale). Every bounded harmonic function on ${\mathbb{R}}^n$ is constant. More generally, a harmonic $u$ on ${\mathbb{R}}^n$ with $|u(x)|\le C(1+|x|{)}^k$ for some $k\ge 0$ and $C>0$ is a harmonic polynomial of degree at most $k$.

Proof. The bounded case is Exercise 7. For the polynomial-growth case, apply the Cauchy estimate to a derivative of order $k+1$: $$|D^{k+1}u(x_0)|\le \frac{C_n^{k+1}(k+1)!}{R^{k+1}}\underset{\overline{B_R(x_0)}}{\max }|u|\le \frac{C_n^{k+1}(k+1)!}{R^{k+1}}\cdot C(1+R+|x_0|{)}^k.$$ The right side has leading behaviour $\sim R^{-1}$ as $R\to \infty$ with $x_0$ fixed (the numerator is $\sim R^k$ from the $|u|$ bound, and the denominator is $R^{k+1}$). So $D^{k+1}u(x_0)=0$ for every $x_0$ and every multi-index of order $k+1$. By Taylor's theorem, $u$ is a polynomial of degree at most $k$. Such a polynomial is harmonic when its Laplacian vanishes, identifying $u$ with the space of harmonic polynomials of degree at most $k$.

$\square$

Connections Master

• Mean-value theorem on ${\mathbb{R}}^n$ 02.05.02. The classical multivariable mean-value theorem provides the differentiation framework on which the spherical-average derivative computation in Step 2 of the mean-value-property proof rests. The harmonic-function mean-value property is the integral-geometric strengthening: instead of an equality of a difference quotient to a derivative at a single intermediate point, it asserts that the value at the centre equals the integral average over an entire sphere, with the equation $\Delta u=0$ controlling the radial derivative of the average.

• Lebesgue integral and the monotone convergence theorem 02.07.04. Supplies the integration apparatus underpinning the Poisson kernel integral $u(x)={\int }_{\partial B_R}P(x,y)g(y)dS(y)$ and the Newtonian potential $u(x)=\int \Phi (x-y)f(y)dy$. The Lebesgue integral lets these integral representations be defined for general $L^p$ data $g$ and $f$; the monotone convergence theorem then justifies passing limits through the integral when constructing the Perron solution as the supremum of subharmonic envelopes.

• Poisson equation $-\Delta u=f$ 02.13.02. The downstream extension of the Laplace equation to allow non-vanishing sources. The Newtonian potential machinery developed here as Theorem 2 is the foundation of the Poisson-equation theory: every solution of $-\Delta u=f$ with compactly supported $f$ decomposes as the Newtonian potential of $f$ plus a harmonic function, and the boundary-value problem on a bounded domain decomposes via the Green-function representation $u(x)=-{\int }_{\partial \Omega }g(y){\partial }_{{\nu }_y}G(x,y)dS+{\int }_{\Omega }G(x,y)f(y)dy$.

• Heat equation $u_t=\Delta u$ 02.13.03. The parabolic relative of the Laplace equation. Steady states ($\partial u/\partial t=0$) of the heat equation are precisely the harmonic functions, identifying $t\to \infty$ limits of heat distributions with Dirichlet-problem solutions. The mean-value property has a parabolic analogue (the heat-ball mean-value identity), and the maximum principle has a parabolic version (the weak maximum principle holds on the parabolic boundary), giving a direct conceptual bridge from elliptic to parabolic theory.

• Laplace equation in electromagnetism 10.01.02. The physical setting that motivated the equation historically. Electrostatic potential in a charge-free region satisfies $\Delta \phi =0$, and the Dirichlet problem corresponds to specifying potential on conductor surfaces. The fundamental solution $\Phi$ identifies the potential of a point charge; the Poisson integral gives the potential inside a hollow conductor; the maximum principle is the formal statement of Earnshaw's theorem 1842 ^{[Earnshaw 1842]} on the impossibility of stable static charge configurations.

• Hodge Laplacian on a Riemannian manifold 03.04.15. Generalises the Euclidean Laplacian to forms on Riemannian manifolds via ${\Delta }_H=d\delta +\delta d$. The Hodge decomposition theorem identifies harmonic forms with de Rham cohomology, the bridge between analysis and topology that the Atiyah-Singer index theorem later refined. The mean-value identity has no direct analogue on curved manifolds, but the maximum principle and elliptic regularity carry over, making the Hodge Laplacian the canonical second-order elliptic operator on Riemannian geometry.

• Stokes theorem and the divergence theorem 03.04.05. Supplies the integration-by-parts identity used in the mean-value-property proof: ${\int }_{\partial B}{\partial }_{\nu }udS={\int }_B\Delta udx$. The integral form of the Laplace equation is exactly the divergence theorem applied to $\nabla u$, identifying harmonicity with vanishing total flux across every closed surface. The Green-function representation theorem is itself a special case of Stokes for the bilinear form $u\Delta v-v\Delta u=\nabla \cdot (u\nabla v-v\nabla u)$.

Historical & philosophical context Master

Laplace's 1782 Mémoires de l'Académie Royale des Sciences de Paris paper on the gravitational potential of spheroids ^{[Laplace 1782]} introduced the equation that now bears his name. Laplace's interest was celestial-mechanics: the gravitational potential outside an attracting body satisfies $\Delta \phi =0$, and Laplace used the spherical-harmonic expansion to solve the equation for an oblate spheroid (the shape of a rotating planet). The equation appeared earlier in Euler 1752 Principes généraux du mouvement des fluides in the velocity-potential form of incompressible irrotational fluid flow, but Laplace's 1782 paper crystallised it as a stand-alone object of mathematical study.

Poisson's 1813 Bulletin de la Société Philomathique de Paris note ^{[Poisson 1813]} extended Laplace's equation to allow non-zero right-hand sides: $-\Delta \phi =4\pi \rho$ inside a mass distribution of density $\rho$. The extension is the Poisson equation, the source-driven version that Theorem 2 above resolves via the Newtonian potential. Poisson also gave the Poisson formula on the disk (1820) and the half-plane (1823), the explicit representations that Theorem 3 generalises.

Gauss's 1840 Allgemeine Lehrsätze ^{[Gauss 1840]} proved the mean-value theorem for harmonic functions and used it as the key tool in the foundations of potential theory. Gauss's monograph was the first systematic treatment of what is now called potential theory; the modern names "harmonic function" and "potential theory" trace to mid-nineteenth-century commentaries on Gauss's work.

Earnshaw's 1842 Trans. Camb. Phil. Soc. paper ^{[Earnshaw 1842]} gave the maximum principle in the electrostatic context. Earnshaw's theorem states that no stable static configuration of point charges in vacuum can be maintained by Coulomb forces alone; the proof reduces to the maximum principle for the electrostatic potential, which forbids the existence of a strict local minimum at any charge-free point. Earnshaw's result was the first physical consequence of the maximum principle and remains a foundational fact in plasma physics and electromagnetic confinement.

Weierstrass's 1870 lecture pointed out the gap in Riemann's proof of the Dirichlet principle: Riemann had asserted that the Dirichlet energy $E(v)=\frac{1}{2}\int |\nabla v{|}^2$ attains a minimum among admissible $v$, but Weierstrass exhibited a coercive functional with no minimum on $C^1$, showing that existence-of-a-minimum is not automatic for variational problems. The gap was the central technical lacuna in Riemann's complex-analysis proofs and motivated Hilbert's 1900 ICM Problem 20 ^{[Hilbert 1900]} asking for a rigorous proof of the Dirichlet principle.

Hilbert resolved his own problem in 1904 by direct-method arguments and again in 1909 via Fredholm-type integral-equation methods. Lebesgue 1907 gave another proof using the new measure-theoretic integration. The full conceptual closure came with Weyl 1940 ^{[Weyl 1940]} who proved the regularity statement (Weyl's lemma): every distributional solution of $\Delta u=0$ is in fact smooth, so the minimiser of the Dirichlet energy is automatically a classical harmonic function. The arc from Riemann 1851 to Weyl 1940 spans ninety years and is the conceptual model for modern functional-analytic methods in PDE.

Perron's 1923 Math. Z. paper ^{[Perron 1923]} gave the supremum-of-subharmonic-envelope construction for the Dirichlet problem on general bounded domains. Perron's method bypasses the smoothness hypotheses on the boundary that earlier methods required and produces a harmonic function on the interior for arbitrary continuous boundary data; the boundary-attainment question is then deflected to the separate Wiener-criterion analysis of boundary regularity.

Wiener's 1924 J. Math. and Phys. paper ^{[Wiener 1924]} gave the complete characterisation of regular boundary points in terms of Newtonian capacity. The Wiener criterion is the definitive answer to the question "when does the Perron solution continuously attain its boundary values at a given boundary point?", and it remains the canonical statement for second-order elliptic problems. Wiener's framework introduced capacity-theoretic methods to PDE that later proved central in Frostman 1935, Choquet 1955, and modern potential theory.

Kakutani's 1944 Proc. Imp. Acad. Tokyo paper ^{[Kakutani 1944]} gave the Brownian-motion characterisation of harmonic functions: the solution of the Dirichlet problem is the expected value of the boundary data at the first-exit point of a Brownian motion. The result is the bridge between deterministic potential theory and probabilistic potential theory and underlies the modern stochastic-PDE methods of Doob, Hunt, and the Dynkin-Yushkevich school.

Hodge 1941 generalised the Euclidean Laplacian to differential forms on Riemannian manifolds and identified the harmonic forms with de Rham cohomology, the headline result of Hodge theory. The same year, Andreotti-Frankel and others extended these methods to Kähler geometry, foreshadowing the modern $\partial$-$\bar{\partial }$ Laplacian theory of complex manifolds. The analytic groundwork laid by Laplace, Gauss, and Hilbert reaches its fullest abstract expression in the Atiyah-Singer index theorem 1963 Ann. Math. 87, which encompasses the Hodge decomposition, the Gauss-Bonnet theorem, and the Riemann-Roch theorem as special cases of a single elliptic-operator index formula.

Bibliography Master

@article{Laplace1782,
  author  = {Laplace, Pierre-Simon},
  title   = {Th\'eorie des attractions des sph\'ero\"ides et de la figure des plan\`etes},
  journal = {M\'emoires de l'Acad\'emie Royale des Sciences de Paris},
  year    = {1782}
}

@article{Poisson1813,
  author  = {Poisson, Sim\'eon Denis},
  title   = {Remarques sur une \'equation qui se pr\'esente dans la th\'eorie des attractions des sph\'ero\"ides},
  journal = {Bulletin de la Soci\'et\'e Philomathique de Paris},
  volume  = {3},
  year    = {1813},
  pages   = {388--392}
}

@book{Gauss1840,
  author    = {Gauss, Carl Friedrich},
  title     = {Allgemeine Lehrs\"atze in Beziehung auf die im verkehrten Verh\"altnisse des Quadrats der Entfernung wirkenden Anziehungs- und Abstossungskr\"afte},
  publisher = {Weidmannsche Buchhandlung},
  address   = {Leipzig},
  year      = {1840}
}

@article{Earnshaw1842,
  author  = {Earnshaw, Samuel},
  title   = {On the nature of the molecular forces which regulate the constitution of the luminiferous ether},
  journal = {Transactions of the Cambridge Philosophical Society},
  volume  = {7},
  year    = {1842},
  pages   = {97--112}
}

@article{Hilbert1900,
  author  = {Hilbert, David},
  title   = {Mathematische Probleme --- Vortrag, gehalten auf dem internationalen Mathematiker-Kongress zu Paris 1900},
  journal = {Nachrichten von der K\"oniglichen Gesellschaft der Wissenschaften zu G\"ottingen},
  year    = {1900},
  pages   = {253--297}
}

@article{Perron1923,
  author  = {Perron, Oskar},
  title   = {Eine neue Behandlung der ersten Randwertaufgabe f\"ur $\Delta u = 0$},
  journal = {Mathematische Zeitschrift},
  volume  = {18},
  year    = {1923},
  pages   = {42--54}
}

@article{Wiener1924,
  author  = {Wiener, Norbert},
  title   = {The {D}irichlet problem},
  journal = {Journal of Mathematics and Physics},
  volume  = {3},
  year    = {1924},
  pages   = {127--146}
}

@article{Weyl1940,
  author  = {Weyl, Hermann},
  title   = {The method of orthogonal projection in potential theory},
  journal = {Duke Mathematical Journal},
  volume  = {7},
  year    = {1940},
  pages   = {411--444}
}

@article{Kakutani1944,
  author  = {Kakutani, Shizuo},
  title   = {Two-dimensional {B}rownian motion and harmonic functions},
  journal = {Proceedings of the Imperial Academy of Tokyo},
  volume  = {20},
  year    = {1944},
  pages   = {706--714}
}

@book{Evans2010,
  author    = {Evans, Lawrence C.},
  title     = {Partial Differential Equations},
  edition   = {2},
  publisher = {American Mathematical Society},
  series    = {Graduate Studies in Mathematics},
  volume    = {19},
  year      = {2010}
}

@book{GilbargTrudinger2001,
  author    = {Gilbarg, David and Trudinger, Neil S.},
  title     = {Elliptic Partial Differential Equations of Second Order},
  edition   = {2},
  publisher = {Springer},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {224},
  year      = {2001}
}

@book{HanLin2011,
  author    = {Han, Qing and Lin, Fang-Hua},
  title     = {Elliptic Partial Differential Equations},
  publisher = {American Mathematical Society / Courant Institute},
  series    = {Courant Lecture Notes in Mathematics},
  volume    = {1},
  year      = {2011}
}

@book{Strauss2008,
  author    = {Strauss, Walter A.},
  title     = {Partial Differential Equations: An Introduction},
  edition   = {2},
  publisher = {Wiley},
  year      = {2008}
}

@book{John1982,
  author    = {John, Fritz},
  title     = {Partial Differential Equations},
  edition   = {4},
  publisher = {Springer},
  year      = {1982}
}