---
title: 'Serrin''s Problem: Rigidity and Extensions'
url: https://www.emergentmind.com/topics/serrin-s-problem
type: topic
---

# Serrin's Problem: Rigidity and Extensions

Serrin’s problem is the overdetermined boundary value problem that asks when a domain can support a solution of a Poisson-type equation together with both constant Dirichlet and constant Neumann data on the boundary. In its classical Euclidean form, one seeks a bounded domain \(\Omega\subset\mathbb{R}^n\) and a function \(u\) such that
\[
\begin{cases}
\Delta u = -1 & \text{in } \Omega,\\
u = 0 & \text{on } \partial\Omega,\\
\partial_\nu u = -c & \text{on } \partial\Omega,
\end{cases}
\]
or equivalently \(u=0\) and \(|\nabla u|=c\) on \(\partial\Omega\), with \(c>0\). Serrin’s theorem states that, for a bounded \(C^2\) domain, the existence of such a solution forces \(\Omega\) to be a ball and \(u\) to be radial. This rigidity result became a prototype for overdetermined elliptic problems, linking PDE symmetry, geometric analysis, torsional rigidity, constant mean curvature phenomena, and shape optimization [1708.07392].

## 1. Classical formulation and core rigidity

The classical problem arises from the torsion equation for the Prandtl stress function. In the normalization used in several modern treatments, the torsion problem is
\[
\begin{cases}
-\Delta u = 1 & \text{in } \Omega,\\
u = 0 & \text{on } \partial\Omega,
\end{cases}
\]
and the overdetermined condition is \(\partial_\nu u = c\) on \(\partial\Omega\). The condition is overdetermined because the Dirichlet problem already determines \(u\), so prescribing a Neumann datum on the whole boundary is not generically compatible. Serrin’s theorem identifies the unique compatible geometry: \(\Omega\) must be a ball, and \(u\) is radial with respect to its center [1401.4385].

This rigidity is classically proved by the method of moving planes, together with the strong maximum principle, Hopf’s lemma, and Serrin’s corner lemma. An alternative proof due to Weinberger uses a \(P\)-function and integral identities. In the torsion normalization \(\Delta u=N\), one convenient reference radius is
\[
R=\frac{N|\Omega|}{|\Gamma|},\qquad \Gamma=\partial\Omega,
\]
and the quadratic comparison function
\[
q(x)=\frac12\bigl(|x-z|^2-a\bigr)
\]
leads to the harmonic remainder \(h=q-u\), which plays a central role in integral-identity approaches to symmetry and stability [1708.07392].

The problem is tightly connected to torsional rigidity. For the torsion function \(u\), the torsional rigidity is represented by \(\int_\Omega u\) or, depending on normalization, \(\int_\Omega |\nabla u|^2\). In shape optimization, the condition that \(\partial_\nu u\) be constant appears as the Euler–Lagrange condition for extremizers of torsional rigidity under a volume constraint [2411.00320].

## 2. Stability theory and quantitative symmetry

A major development after the qualitative theorem is quantitative stability: if the overdetermined condition is only approximately satisfied, then the domain is quantitatively close to a ball. A standard geometric measure of closeness uses inner and outer radii centered at a point \(z\),
\[
\rho_i=\min_{x\in\Gamma}|x-z|,\qquad \rho_e=\max_{x\in\Gamma}|x-z|,
\]
so that \(\rho_e-\rho_i\) measures the thickness of the smallest concentric annulus containing \(\partial\Omega\) [1903.04823].

For Serrin’s problem, a representative sharp-type estimate is
\[
\rho_e-\rho_i \le C\,\|u_\nu-R\|_{L^2(\Gamma)}^{\tau_N},
\]
with
\[
\tau_2=1,\qquad \tau_3 \text{ arbitrarily close to }1,\qquad \tau_N=\frac{2}{N-1}\ \text{for }N\ge 4,
\]
where \(u\) solves the torsion equation and \(R=N|\Omega|/|\Gamma|\) [1903.04823]. In the \(L^1\) setting, the exponent is halved:
\[
\rho_e-\rho_i \le C\,\|u_\nu-R\|_{L^1(\Gamma)}^{\tau_N/2}.
\]
These estimates are described as nearly optimal; in dimension \(2\), the exponent \(1\) is optimal, as shown by explicit ellipse computations [1903.04823].

A related integral-identity framework yields improved stability simultaneously for Serrin’s problem and Alexandrov’s Soap Bubble theorem. In that approach, the interior defect
\[
|\nabla^2 u|^2-\frac{(\Delta u)^2}{N}
\]
is the common analytic quantity governing both the deviation of \(u_\nu\) from a constant and the deviation of the mean curvature \(H\) from its reference value \(H_0\). This leads, for example, to
\[
\rho_e-\rho_i \le C\,\|H-H_0\|_{L^2(\Gamma)}^{\tau_N},
\]
with \(\tau_2=\tau_3=1\), \(\tau_4\) arbitrarily close to \(1\), and \(\tau_N=2/(N-2)\) for \(N\ge 5\) [1708.07392].

A distinct quantitative direction connects Serrin stability to dynamics. For the normalized overdetermined problem with \(|Du|=1\) on \(\partial\Omega\), the defect
\[
\big\||Du|^2-1\big\|_{L^2(\partial\Omega)}
\]
controls the symmetric-difference distance to the optimal ball linearly under geometric assumptions such as a \(C^2\) boundary, an interior ball condition, and an \(L_0\)-John condition [1707.06949]. That defect is exactly the energy dissipation rate for a quasi-static capillary drop model, which permits exponential convergence of regular solutions toward the spherical steady state [1707.06949].

The stability viewpoint also appears in a partial-data variant. If the equation is known only in \(\Omega\setminus\overline{\omega}\), for some \(\omega\Subset\Omega\), and the overdetermined condition is imposed on \(\partial\Omega\), then \(\Omega\) is quantitatively close to a ball when \(|\omega|\) or \(\mathcal{H}^{N-1}(\partial^*\omega)\) is small. The estimates control an \(L^2\) pseudo-distance to a sphere, Fraenkel-type asymmetry, and the annular thickness \(\rho_e-\rho_i\), thereby recovering the classical Serrin theorem when \(\omega=\varnothing\) [2005.04859].

## 3. Regularity thresholds and weak formulations

The classical theorem assumes a \(C^2\) boundary. A major recent result shows that this regularity is far from optimal. In a weak formulation, one extends \(u\) by zero outside \(\Omega\) and encodes the overdetermined condition distributionally as
\[
\Delta u = c\,\mathcal{H}^{n-1}\!\lfloor_{\partial^*\Omega} - 1_\Omega\,dx
\]
for a bounded indecomposable set of finite perimeter \(\Omega\). Under the density bound
\[
\mathcal{H}^{n-1}(B_r(x)\cap \partial^*\Omega)\le A r^{n-1}
\quad\text{for }\mathcal{H}^{n-1}\text{-a.e. }x\in\partial^*\Omega,\ \forall r\in(0,1),
\]
the only possible solution set is a ball, and \(u\) is the explicit quadratic torsion function [2407.02293]. This applies in particular to bounded Lipschitz domains and settles affirmatively the open question of whether Serrin’s theorem remains valid in that class [2407.02293].

The same framework also accommodates slit discontinuities. If \(\Sigma\subset\Omega\) is an \((n-1)\)-rectifiable interior slit and
\[
\Delta u = c\,\mathcal{H}^{n-1}\!\lfloor_{\partial^*\Omega}
      +2c\,\mathcal{H}^{n-1}\!\lfloor_{\Sigma}
      -1_\Omega\,dx,
\]
together with the symmetric blow-up condition
\[
\frac{u(x+rz)}{r}\to c\,|\nu_x\cdot z|
\quad\text{for }\mathcal{H}^{n-1}\text{-a.e. }x\in\Sigma,
\]
then the same spherical rigidity holds [2407.02293].

In the plane, weak formulations can be recast as harmonic quadrature identities. For a Jordan domain \(\Omega\subset\mathbb{C}\) with rectifiable boundary, the identity
\[
\int_\Omega h\,dA = c\int_{\partial\Omega} h\,d\mathcal{H}^1
\qquad \forall h\in C(\overline\Omega),\ \Delta h=0 \text{ in }\Omega
\]
is equivalent, on Jordan quasidisks, to the weak Serrin condition [2605.02034]. Within the class of rectifiable Jordan Smirnov domains, this identity forces \(\Omega\) to be a disk [2605.02034]. The Smirnov assumption is sharp: there exist rectifiable, non-Smirnov Jordan domains satisfying the same quadrature identity, hence supporting a weak Serrin formulation without being disks [2605.02034]. This exhibits a precise boundary-regularity threshold in the planar weak theory.

## 4. Fully nonlinear, two-phase, and non-elliptic variants

A broad generalization replaces the Laplacian by a fully nonlinear Hessian operator. In the planar real-analytic setting, one considers
\[
\begin{cases}
F(D^2u)=0 & \text{in }\Omega,\\
u=0,\quad |Du|=c & \text{on }\partial\Omega,
\end{cases}
\]
where \(F\) is rotationally invariant and not locally zero. If \(\Omega\subset\mathbb{R}^2\) is smooth, bounded, and simply connected, and \(u\in C^\omega(\overline\Omega)\) is nontrivial, then \(\Omega\) is a disk and \(u\) is radial, even though \(F\) is not assumed elliptic [1902.01744].

The proof mechanism is completely different from moving planes. Rotational invariance reduces the equation to a functional dependence between \(\Delta u\) and \(H(u)=\det D^2u\), yielding
\[
J[\Delta u,H(u)] = 0,
\]
and the analysis proceeds via the eigenline fields of the Hessian and a Poincaré–Hopf index argument on a simply connected planar domain [1902.01744]. The result is sharp in two separate senses: it fails if \(\Omega\) is not simply connected, and it fails if \(u\) is merely \(C^\infty\) rather than real analytic [1902.01744].

A different extension is the two-phase Serrin problem, motivated by shape optimization for composite torsional rigidity. Here one has a core \(D\Subset\Omega\) with piecewise-constant conductivity
\[
\sigma(x)=
\begin{cases}
\sigma_c & x\in D,\\
1 & x\in \Omega\setminus D,
\end{cases}
\qquad \sigma_c>0,\ \sigma_c\neq 1,
\]
and solves
\[
-\operatorname{div}(\sigma\nabla u)=1 \quad\text{in }\Omega,
\qquad
u=0,\ \partial_n u=c \quad\text{on }\partial\Omega,
\]
together with transmission conditions across \(\partial D\) [2411.00320]. In this setting, the outer shape \(\Omega\) is itself unknown.

Several qualitative features distinguish the two-phase problem from the one-phase case. Critical shapes of the two-phase torsional rigidity under a volume constraint are exactly those satisfying the overdetermined condition on \(\partial\Omega\), but such critical shapes are never local minimizers [2411.00320]. The same work proves that solutions have no tentacles, the outer boundary contains no flat parts, and if the outer boundary contains a spherical portion then the only possibility is concentric balls [2411.00320]. A strong parameter-rigidity result also holds: if a given configuration solves the two-phase problem for two distinct conductivity values \(\sigma_c=\alpha\neq\beta\), then the configuration must be concentric balls [2411.00320].

## 5. Curved ambient spaces, cones, and weighted analogues

On Riemannian manifolds, Serrin-type problems take the form
\[
\begin{cases}
\Delta_g u + nk\,u = -1 & \text{in }\Omega,\\
u>0 & \text{in }\Omega,\\
u=0,\quad |\nabla u|=c & \text{on }\partial\Omega,
\end{cases}
\]
with \(k\in\mathbb{R}\). Under \(\mathrm{Ric}\ge (n-1)k\,g\), a closed conformal vector field, and a curvature-weighted integral condition, one obtains rigidity: \(\Omega\) is a metric ball and \(u\) is radial [2305.19772]. In the Einstein case \(\mathrm{Ric}=(n-1)k\,g\), the curvature condition is automatic, so any positive solution of the overdetermined problem on such a manifold forces \(\Omega\) to be a metric ball [2305.19772].

The principal analytic tools in that setting are a new Pohozaev identity involving scalar curvature,
\[
\frac{n+2}{n}\int_\Omega \varphi u\,d\mu
=
c^2\int_\Omega \varphi\,d\mu
-\frac{n-2}{2n(n-1)}\int_\Omega u^2(\varphi R+X(R))\,d\mu
-2k\int_\Omega \varphi u^2\,d\mu,
\]
and a generalized Weinberger \(P\)-function,
\[
P(u)=|\nabla u|^2+\frac{2}{n}u+ku^2,
\]
which is subharmonic under the Ricci lower bound [2305.19772].

A related line of work treats convex cones in warped product manifolds. For a sector-like domain \(\Omega\subset\Sigma\) in a convex cone \(\Sigma\), the overdetermined problem becomes
\[
\begin{cases}
\Delta u + nk\,u = -1 & \text{in }\Omega,\\
u=0,\quad u_\nu=-c & \text{on }\Gamma,\\
u_\nu=0 & \text{on }\Gamma_1,
\end{cases}
\]
where \(\Gamma=\partial\Omega\cap\Sigma\) and \(\Gamma_1=\partial\Omega\setminus\overline{\Gamma}\) [2501.05551]. Under a Ricci lower bound and a compatibility condition involving the scalar curvature and the closed conformal field \(X=\rho\partial_t\), the only possible domains are intersections of geodesic balls with the cone [2501.05551]. In Einstein warped products the compatibility condition is automatic [2501.05551].

The same setting supports cone analogues of Alexandrov’s Soap Bubble theorem and of the Heintze–Karcher inequality, characterizing \(\Sigma\cap B_r(x_0)\) among sector-like domains. In the Euclidean cone case with drift Laplacian
\[
\Delta_f=\Delta+\langle \nabla\log f,\nabla\cdot\rangle
\]
for a homogeneous weight \(f\), the weighted overdetermined problem likewise forces the domain to be an intersection of a ball with the cone [2501.05551].

Warped products without cone structure were also treated by a direct Weinberger \(P\)-function method. For
\[
\Delta u + nk\,u = -1 \quad\text{in }\Omega,\qquad u=0,\ |\nabla u|=c \quad\text{on }\partial\Omega,
\]
in a warped product manifold with \(\mathrm{Ric}_M\ge (n-1)k\,g\), one defines
\[
P(u)=|\nabla u|^2+\frac{2}{n}u+ku^2.
\]
Under suitable assumptions on the warping function \(\sigma\), or under a compatibility condition
\[
\int_\Omega \left( k\sigma'(r)+\frac{(\sigma''\sigma^{n-1})'}{\sigma^{n-1}}\right)u^2\,dV \ge 0,
\]
the domain is a metric ball and \(u\) is radial [1912.09824]. In model manifolds, the argument becomes strong enough to show that the metric inside the relevant ball must actually be the space-form metric of curvature \(k\) [1912.09824].

## 6. Unbounded, periodic, and planar classification phenomena

Bounded domains are rigid in the classical theorem, but unbounded domains exhibit a richer geometry. One construction produces nontrivial unbounded periodic domains \(\Omega\subset\mathbb{R}^N\) of the form
\[
\Omega_\varphi
=
\{(z,t)\in\mathbb{R}^n\times\mathbb{R}^m:\ |z|<\varphi(t)\},
\]
where \(N=n+m\), \(\varphi\) is even in each \(t_j\), \(2\pi\)-periodic in each \(t_j\), and invariant under permutations of the \(t_j\). These domains bifurcate from straight cylinders \(\varphi\equiv\lambda_*\), and for small \(s\) one has
\[
\varphi_s(t)=\lambda_s+\psi_s(t),\qquad
\psi_s(t)=s\Bigl(\sum_{j=1}^m \cos t_j + u_s(t)\Bigr),
\]
with \(u_s\) orthogonal to the first harmonics [1603.05727]. On each such \(\Omega_{\varphi_s}\), the overdetermined problem
\[
-\Delta u_s=1 \quad\text{in }\Omega_{\varphi_s},
\qquad
u_s=0,\ \partial_\nu u_s=\lambda_s \quad\text{on }\partial\Omega_{\varphi_s}
\]
admits a solution [1603.05727]. These domains are also uniquely self-Cheeger relative to a period cell, with relative Cheeger constant \(1/\lambda_s\) [1603.05727].

A different unbounded phenomenon occurs for semilinear problems. For all \(N\ge 9\), there exist smooth entire epigraphs
\[
\Omega=\{x\in\mathbb{R}^N:\ x_N>F(x_1,\ldots,x_{N-1})\},
\]
which are not half-spaces and for which
\[
\Delta u + f(u)=0 \quad\text{in }\Omega,\qquad
u=0,\ \partial_\nu u=\text{constant}\quad\text{on }\partial\Omega
\]
has a positive bounded solution [1310.4528]. This gives a negative answer, in high dimension, to the Berestycki–Caffarelli–Nirenberg question for epigraphs [1310.4528]. The construction uses large dilations of nontrivial minimal or constant-mean-curvature hypersurfaces and gluing methods, thereby reinforcing the conceptual analogy between Serrin’s problem and Alexandrov’s theorem on constant mean curvature surfaces [1310.4528].

The sphere provides another setting where non-classical domains appear. On \(S^N\), the spherical overdetermined problem
\[
-\Delta_{S^N}u=1 \quad\text{in }\Omega,\qquad
u=0,\ \partial_\eta u=\text{const}\quad\text{on }\partial\Omega
\]
has, besides geodesic balls and straight tubular neighborhoods of the equator, families of nontrivial Serrin domains bifurcating from symmetric tubular neighborhoods [1612.03717]. These domains are parameterized by perturbations
\[
\varphi_s^j=\lambda_j(s)+v_s^j,\qquad
v_s^j=s(Y_j+w_s^j),
\]
where \(Y_j\) is an axially symmetric spherical harmonic of degree \(j\ge2\) and \(w_s^j\) is orthogonal to \(Y_j\) [1612.03717]. They are the first examples of Serrin domains in \(S^N\) not bounded by geodesic spheres [1612.03717].

In the planar doubly connected and periodic two-boundary setting, a recent classification identifies all smooth ring domains and periodic bands that solve Serrin’s classical problem with locally constant boundary data as algebro-geometric potentials of the mKdV hierarchy [2601.09649]. If \(g\) is the developing map of the domain, the quantity
\[
\eta(z)=\frac{g''(z)}{2g'(z)}
\]
satisfies a finite-gap relation
\[
Q_{\mathfrak m}[\eta]
=
a_0+\sum_{j=0}^{\mathfrak m-1} c_j Q_j[\eta],
\]
where \(Q_j\) are the mKdV hierarchy operators [2601.09649]. This organizes planar Serrin ring domains and periodic bands into finite-dimensional complexity levels indexed by the spectral genus \(\mathfrak m\). At the first nontrivial, elliptic level, the theory produces a global one-parameter family of periodic Serrin bands interpolating between a flat band and a chain of tangent disks, together with, for each \(n>1\), a two-dimensional moduli space \(\mathbf{T}_n\) of non-radial Serrin ring domains with dihedral symmetry of order \(2n\) [2601.09649].

Taken together, these developments show that Serrin’s problem is not a single rigidity statement but a broad research program. In the bounded Euclidean one-phase case it characterizes balls. In rough domains it remains rigid down to finite-perimeter settings with density control. In real-analytic fully nonlinear planar problems it survives beyond ellipticity. On manifolds and cones it interacts with curvature, closed conformal vector fields, and Reilly-type formulas. In two-phase media, periodic settings, spheres, and planar doubly connected domains, it reveals a substantial flexibility structured by spectral, geometric, and integrable mechanisms.

Source: https://www.emergentmind.com/topics/serrin-s-problem