--- title: Serrin Overdetermined Problem url: https://www.emergentmind.com/topics/serrin-overdetermined-problem type: topic --- # Serrin Overdetermined Problem The Serrin overdetermined problem is the rigidity question asking when an elliptic boundary value problem with both Dirichlet and constant Neumann data can be solved on a domain. In its classical form, if a bounded \(C^2\) domain \(\Omega\subset\mathbb{R}^n\) admits \(u\in C^2(\overline{\Omega})\) satisfying \[ \begin{cases} -\Delta u = 1 & \text{in } \Omega,\\ u = 0 & \text{on } \partial\Omega,\\ \partial_\nu u = c & \text{on } \partial\Omega, \end{cases} \] then \(\Omega\) is a ball and \(u\) is the quadratic radial profile. The sign of \(c\) depends on whether \(\nu\) is taken as the inward or outward unit normal, so equivalent formulations differ by orientation. The problem is “overdetermined” because the Dirichlet problem already determines the solution, and the extra Neumann condition constrains the geometry of the domain. Over time, the subject has expanded from this prototype to semilinear, quasilinear, nonlocal, Hessian, multi-phase, manifold, rough-boundary, unbounded, and shape-optimization settings [2402.09817] [2506.02423]. ## 1. Classical formulation and rigidity theorem Serrin’s theorem is the benchmark result of the subject. In one standard normalization, if \(\Omega\subset\mathbb{R}^n\) is connected, bounded, open, with \(C^2\) boundary, and there exists \(u\in C^2(\overline{\Omega})\) such that \[ \begin{cases} -\Delta u = 1 & \text{in } \Omega,\\ u = 0 & \text{on } \partial\Omega,\\ \dfrac{\partial u}{\partial \nu} = c & \text{on } \partial\Omega, \end{cases} \] then \(\Omega\) is a ball \(B_R(x_0)\) and \[ u(x)=\frac{R^2-|x-x_0|^2}{2n}. \] Equivalent versions appear with \(\Delta u=-1\) or with the outward normal, changing only signs and normalization [2402.09817] [2506.02423]. The theorem extends beyond the torsion equation. Serrin’s original framework also covers semilinear problems of the form \[ -\Delta u=f(u), \] under structural assumptions on \(f\) and assuming that \(u\) has constant sign; the conclusion remains that \(\Omega\) is a ball and \(u\) is radial with monotone radial profile [2402.09817] [1603.05727]. In this sense, the phrase “Serrin overdetermined problem” now designates a class of rigidity statements in which prescribing both \(u\) and a flux-type boundary datum forces spherical symmetry. The classical result also has a variational interpretation. In Euclidean space it characterizes stationary domains for torsional rigidity under volume-preserving deformations, and balls are precisely the stationary shapes in the one-phase setting [1406.0065]. This variational viewpoint became one of the main routes by which overdetermined boundary conditions entered shape optimization. ## 2. Proof paradigms, weak formulations, and quantitative stability The original proof uses the method of moving planes, inherited from Alexandrov’s reflection method for constant-mean-curvature hypersurfaces. In a fixed direction, one reflects caps of the domain across moving hyperplanes, compares the solution with its reflection, and uses the strong maximum principle together with Hopf’s boundary lemma and a refined corner-point lemma at the critical position. This remains the canonical qualitative proof strategy for many local Serrin-type problems [2402.09817] [2506.02423]. A distinct route is Weinberger’s \(P\)-function method, based on an auxiliary quantity and Pohožaev-type identities, which is especially effective in the Laplacian torsion case [2506.02423]. A major recent development concerns rough boundaries. The classical theorem was long tied to \(C^2\) domains, but it now extends to bounded indecomposable sets of finite perimeter satisfying a uniform upper density bound on the reduced boundary. In that formulation, the boundary condition is encoded distributionally as \[ \Delta u = c\,\mathcal{H}^{n-1}\mres\partial^*\Omega - \mathbf{1}_\Omega\,dx, \] with \(u=0\) almost everywhere outside \(\Omega\). This framework covers bounded Lipschitz domains and even allows slit discontinuities, while preserving the conclusion that \(\Omega\) must be a ball [2407.02293]. The argument replaces classical boundary regularity with geometric-measure-theoretic analysis of reduced boundaries, Green functions, and a rough-domain version of Weinberger’s maximum-principle strategy. Quantitative stability has developed in parallel with rigidity. In the fractional setting, for \[ (-\Delta)^s u = f(u)\quad \text{in }\Omega,\qquad u=0\quad \text{in }\mathbb{R}^n\setminus\Omega, \] with fractional normal derivative \(\partial_\nu^s u\), one has a stability estimate of the form \[ \rho(\Omega)\le C\,[\partial_\nu^s u]_{\partial\Omega}^{\frac{1}{s+2}}, \] where \(\rho(\Omega)\) measures the gap between inner and outer enclosing balls. For \(C^{2,\alpha}\) domains this exponent improves to \[ \frac{\alpha}{1+(s+1)\alpha}. \] The proof develops a quantitative nonlocal moving-planes apparatus, including an antisymmetric barrier, a quantitative nonlocal maximum principle, and nonlocal versions of Hopf’s lemma and Serrin’s corner lemma [2309.17119]. The same stability theme appears in perturbative local theory. For small perturbations of the constant Neumann datum around the ball, one can solve the perturbed overdetermined problem by parameterizing the boundary as a normal graph over the sphere and applying a modified implicit function theorem for Banach triplets that tolerates loss of derivatives away from the base point. This yields existence, local uniqueness up to translation, and linear stability estimates for the shape of the domain [2103.07072]. ## 3. Generalized operators and ambient geometries A central direction of generalization replaces the scalar Laplacian problem by more complex operators while retaining the overdetermined Dirichlet–Neumann structure. One extension treats a boundary-coupled elliptic system \[ \begin{cases} -\Delta u_i = f_i(u_i) & \text{in } \Omega,\\ u_i = 0 & \text{on } \partial\Omega,\qquad i=1,\dots,m,\\ F\!\left(\dfrac{\partial u_1}{\partial \nu},\dots,\dfrac{\partial u_m}{\partial \nu}\right)=c & \text{on } \partial\Omega, \end{cases} \] where each \(f_i=g_i+h_i\) with \(g_i\) locally Lipschitz and \(h_i\) nondecreasing, and the boundary nonlinearity \(F\) is nondecreasing in each variable and strictly increasing in at least one. Under these hypotheses, \(\Omega\) is a ball and each \(u_i\) is radially symmetric and strictly radially decreasing [2402.09817]. This is a genuine system analogue of Serrin’s theorem: the PDEs are decoupled in the interior, but the normal derivatives are coupled nonlinearly on the boundary. Another broad extension concerns quasilinear, potentially degenerate elliptic equations, \[ -\operatorname{div}\!\left(g(|\nabla u|)\frac{\nabla u}{|\nabla u|}\right)=f(u), \] with \(g\) continuous, strictly increasing, and \(g(0)=0\). In arbitrary bounded domains with no regularity assumed on \(\partial\Omega\), symmetry can still be recovered if \(u\) and \(|\nabla u|\) approach prescribed boundary values uniformly from the interior. The same continuous Steiner symmetrization method also gives a ring-domain analogue: in a connected domain \(\Omega=\Omega_0\setminus\overline{\Omega_1}\), the existence of a positive solution with appropriate two-sided overdetermined data forces both \(\Omega_0\) and \(\Omega_1\) to be balls, although not necessarily concentric [2506.02423]. This sharply contrasts with the bounded simply connected case, where the entire domain is forced to be a single ball. The theory also extends outside the uniformly elliptic regime. In dimension two, if \(u\in C^\omega(\overline{\Omega})\) solves a rotationally invariant Hessian equation \[ F(D^2u)=0 \] on a smooth bounded simply connected domain and satisfies \[ u=0,\qquad |Du|=c \quad \text{on }\partial\Omega, \] then \(\Omega\) is a disk and \(u\) is radial, even though \(F\) is not assumed elliptic [1902.01744]. The analyticity and simple connectivity are essential: the statement fails on non-simply-connected domains and fails for merely \(C^\infty\) solutions. The ambient geometry can also be changed. In convex cones, Serrin-type mixed problems with Dirichlet and constant Neumann data on the interior part of the boundary and homogeneous Neumann data on the conical part force the domain to be a spherical sector. This holds for a class of possibly degenerate operators of the form \[ L_f u=\operatorname{div}\!\left(f'(|\nabla u|)\frac{\nabla u}{|\nabla u|}\right), \] and, in space forms, for the operator \(\Delta+NK\), yielding geodesic spherical sectors [1806.08553]. On compact Riemannian manifolds, small Serrin-type domains exist as smooth perturbations of geodesic balls. Their centers are governed by scalar curvature: if \(p_0\) is a nondegenerate critical point of \(\mathrm{Scal}_g\), the boundaries of the resulting domains form a smooth foliation near \(p_0\), and conversely any concentrating family of such domains must converge to a critical point of \(\mathrm{Scal}_g\) [1406.0065]. ## 4. Coupled, two-phase, and multi-phase problems The two-phase Serrin problem replaces the Laplacian by a divergence-form operator with piecewise constant coefficient, \[ -\operatorname{div}(\sigma\nabla u)=1,\qquad \sigma(x)=1+(\sigma_c-1)\chi_D, \] in a pair of nested domains \(D\subset\Omega\), together with \[ u=0,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega. \] In this setting one distinguishes an outer problem, where \(D\) is fixed and \(\Omega\) is sought, and an inner problem, where \(\Omega\) is fixed and the inclusion \(D\) is sought [2107.05889] [1811.07156]. The concentric configuration \((D,\Omega)=(B_R,B_1)\) always provides a radial solution, but the one-phase rigidity picture no longer persists globally. Near a concentric two-phase configuration, local solvability can be proved by shape derivatives and an implicit function theorem. For fixed \(D=B_R\) and \(\Omega=B_1\), if the contrast \(\sigma_c\) avoids a finite resonance set \(\Sigma\subset(0,1]\), then every sufficiently small perturbation of the core induces a unique nearby perturbation of the outer boundary solving the two-phase overdetermined problem; numerically, this can be computed by minimizing a Kohn–Vogelius mismatch functional under a volume constraint [1811.07156]. This already shows that the outer boundary need not remain spherical. At resonant coefficient values, symmetry breaking occurs. Writing \(\sigma_c=s(m)+\lambda\) near special values \(s(m)\), the linearized operator loses invertibility in the spherical-harmonic mode of degree \(m\). Crandall–Rabinowitz bifurcation then yields nontrivial branches of solutions \[ g(\varepsilon)=\varepsilon Y_m+o(\varepsilon) \] for \(N\ge 3\), and \[ g(\varepsilon)=\varepsilon\cos(m\theta)+o(\varepsilon) \] for \(N=2\), where \(g\) parameterizes the outer boundary and \(Y_m\) is a degree-\(m\) spherical harmonic [2001.10212]. In this precise sense, the two-phase Serrin problem admits symmetry-breaking outer domains even with a spherical core. Quantitative stability survives near the one-phase regime. If a two-phase configuration solves the overdetermined problem and either \(|\sigma_c-1|\) is small or the inclusion volume \(|D|\) is small, then \(\Omega\) is quantitatively close to a ball, with explicit power-law control of the radius gap \(\rho_e-\rho_i\) by \(|\sigma_c-1|^{\tau_N}\) or \(|D|^{\tau_N/2}\), where \[ \tau_2=1,\qquad \tau_3=1-\theta,\qquad \tau_N=\frac{2}{N-1}\ \text{for }N\ge 4. \] These estimates imply nonexistence for the inner problem on a fixed non-ball when the contrast or inclusion size is too small [2107.05889]. The multi-phase problem exhibits a further structural change. For a layered conductivity \[ \sigma(x)=\sum_{k=1}^m \sigma_k\chi_{D_k}(x), \] one can ask whether constant normal derivatives of successive orders on the outer boundary force radiality. The answer depends sharply on the number of phases. For \(m=2\), a single overdetermined condition \(\partial_n u=\text{const}\) on the outer boundary does not force concentric balls, but imposing both \[ \partial_n u\equiv c_1,\qquad (\partial_n)^2u\equiv c_2 \quad\text{on }\partial\Omega_2 \] does, and this characterization is if and only if [2304.00791]. For \(m\ge 3\), the pattern collapses: there exist non-radial configurations for which \[ (\partial_n)^k u\equiv c_k \quad \text{on }\partial\Omega_m,\qquad \forall k\in\mathbb{N}, \] so countably infinitely many overdetermined conditions on the outer boundary still do not force concentricity [2304.00791]. This is one of the clearest demonstrations that Serrin-type rigidity is highly sensitive to how internal structure is encoded. ## 5. Unbounded, epigraph, periodic, and exterior configurations On bounded domains, the classical theorem forces balls. On unbounded domains, that classification fails. For the original torsion equation \[ -\Delta u=1,\qquad u=0,\qquad \partial_\nu u=\text{const}, \] there exist nontrivial unbounded domains that are periodic in some variables and radial in the others. These domains are of the form \[ \Omega_\phi=\{(z,t)\in\mathbb{R}^n\times\mathbb{R}^m: |z|<\phi(t)\}, \] with \(\phi\) even, \(2\pi\)-periodic, and permutation-invariant in the periodic variables, and they bifurcate from generalized cylinders or slabs. In each period cell they are uniquely self-Cheeger relative to that cell [1603.05727]. This gives a direct counterpoint to the bounded-domain ball theorem. Epigraphs form a different unbounded class. For \[ \Omega=\{(x',x_N)\in\mathbb{R}^{N-1}\times\mathbb{R}: x_N>g(x')\}, \] if \(\Omega\) is bounded from below and \(u\) is a classical solution of \[ \begin{cases} -\Delta u=f(u) & \text{in }\Omega,\\ u>0 & \text{in }\Omega,\\ u=0 & \text{on }\partial\Omega,\\ \dfrac{\partial u}{\partial\eta}=\mathfrak{c} & \text{on }\partial\Omega, \end{cases} \] then rigidity can return. For specified Allen–Cahn-type nonlinearities, one obtains that \(\Omega\) must be an affine half-space and \(u\) must be one-dimensional in all dimensions \(N\ge 2\); for locally Lipschitz \(f\), analogous results hold in dimensions \(N\le 3\); and when \(f(0)<0\), a new monotonicity theorem \(\partial_{x_N}u>0\) is proved in any dimension \(N\ge 2\) under \(C^{1,\gamma}\) hypotheses on the epigraph [2502.04812]. These results partially answer the Berestycki–Caffarelli–Nirenberg question in the epigraph setting. Exterior domains introduce yet another geometry. For Hessian equations in \(\mathbb{R}^d\setminus\overline{\Omega}\), \[ \sigma_k(\lambda(D^2u))=\binom{d}{k}, \] with \[ u=0,\qquad \frac{\partial u}{\partial\nu}=1 \quad \text{on }\partial\Omega, \] and prescribed quadratic asymptotics at infinity, two regimes appear. If the asymptotic quadratic form is isotropic, \(A=I\), there is a threshold \(c^*\) such that for \(c