Serrin-Type Overdetermined Systems

• Serrin-type systems are elliptic boundary value problems that impose both Dirichlet and Neumann conditions, creating overdetermined constraints that force domain rigidity.
• Techniques like the method of moving planes, P-function methods, and Pohozaev identities are employed to show that only canonical shapes such as balls or sectors admit solutions.
• Modern research extends the framework to anisotropic, rough, free-boundary, and multiphase problems, broadening its applications in geometric analysis and shape optimization.

A Serrin-type overdetermined system is an elliptic boundary value problem in which one prescribes, on the same boundary or on a distinguished part of it, both Dirichlet data and a Neumann-type datum. In its classical torsion form, the prototype is

$$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$

while semilinear, quasilinear, fully nonlinear, anisotropic, and geometric analogues replace the Laplacian, the source term, or the ambient space. The system is called overdetermined because a second-order elliptic equation ordinarily does not admit simultaneous global Dirichlet and Neumann prescriptions except in highly constrained geometries. The central theme of the subject is therefore rigidity: solvability often forces the domain to be a ball, a geodesic ball, a spherical sector, a Wulff shape, or another canonical symmetric object. At the same time, modern work shows that this rigidity can fail in unbounded, multiphase, weak, or high-dimensional settings, so the theory now sits at the intersection of elliptic PDE, geometric analysis, free-boundary theory, and shape optimization (Pino et al., 2013, Figalli et al., 2024).

1. Canonical formulation and the classical rigidity paradigm

The classical bounded-domain model is Serrin’s theorem: if a bounded $C^2$ domain admits a classical solution of the overdetermined Poisson problem, then the domain must be a Euclidean ball, and the solution is the corresponding radial quadratic profile. In the normalization used in the rough-domain formulation,

$$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$

the profile is

$u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$

up to translation (Figalli et al., 2024). This is the PDE analogue of Alexandrov’s theorem characterizing compact embedded constant-mean-curvature hypersurfaces as spheres, and that analogy remains structurally important in later developments (Pino et al., 2013).

A standard semilinear generalization replaces the torsion equation by

$$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$

with $f$ smooth and monostable: $$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$ This includes the Fisher–Kolmogorov and Allen–Cahn nonlinearities. Under these hypotheses there is a unique increasing one-dimensional profile $w$ solving

$w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$

and the half-space solution $u(x)=w(x_N)$ solves the overdetermined problem in $C^2$0 (Pino et al., 2013). This one-dimensional model becomes the local normal profile in several gluing constructions.

The modern literature uses “Serrin-type” in a broader sense. The overdetermined datum may be prescribed only on a portion of the boundary, may be replaced by $C^2$1, may involve a nonlinear functional $C^2$2, or may appear on an internal interface rather than the outer boundary (Celentano et al., 2024, Jia et al., 2023). The defining feature is not the precise PDE, but the coexistence of elliptic interior structure with boundary data that are too strong for generic domains.

2. Rigidity mechanisms and proof architectures

Two proof paradigms dominate the classical theory. The first is Serrin’s method of moving planes, which compares the solution with its reflections until a critical position is reached; internal tangency and corner-touching are then excluded by the strong maximum principle, Hopf’s lemma, and the corner lemma. This framework extends to semilinear systems

$C^2$3

provided $C^2$4 is nondecreasing in each variable and strictly increasing in at least one component; the conclusion is again that $C^2$5 is a ball and each $C^2$6 is radial (Celentano et al., 2024).

The second paradigm is Weinberger’s $C^2$7-function method. On model manifolds for the torsion equation $C^2$8, one introduces

$C^2$9

uses the Bochner formula and Cauchy–Schwarz to show $$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$0, and combines subharmonicity with boundary information to force constancy of $$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$1. In the Euclidean case this recovers Weinberger’s proof of Serrin’s theorem; on model manifolds an additional compatibility condition is required to obtain the same rigidity conclusion (Roncoroni, 2017). Variants of the same idea appear in rough domains, in Riemannian manifolds, in cones, in anisotropic problems, and in the Heisenberg group.

A third major mechanism is the Rellich–Pohozaev identity. For Hessian quotient equations and curvature quotient equations, the proofs proceed by pairing a $$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$2-function maximum principle with identities that force equality in Newton–Maclaurin inequalities. In the Euclidean Hessian quotient case, the model overdetermined problem is

$$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$3

and the combination of the Pohozaev identity and the $$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$4-function $$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$5 yields the unit ball and the quadratic profile (Gao et al., 2022).

More recent work shows that rigidity can also be proved without moving planes. For obstacle problems with radial obstacle $$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$6, the proof compares the solution in $$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$7 with radial obstacle-problem solutions in the largest and smallest concentric balls $$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$8, and uses superharmonicity to force $$\Delta u=-1 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c>0 \quad \text{on }\partial\Omega,$$9 (Nitti et al., 2023). For nonsmooth bounded domains and ring-shaped domains with potentially degenerate quasilinear operators, continuous Steiner symmetrization is used instead. There the key functional is

$u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$0

and local symmetry under continuous Steiner symmetrization ultimately forces the positivity set to be a ball, or both boundary components to be balls in the ring-shaped case (Cao et al., 3 Jun 2025).

3. Unbounded domains, high dimensions, and the failure of flatness

The sharpest departure from the classical bounded rigidity is the construction of nontrivial unbounded epigraphs. For every $u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$1 and every monostable $u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$2 as above, there exist smooth entire epigraphs

$u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$3

which are not half-spaces, yet support a positive bounded solution of

$u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$4

The model boundary is a small perturbation of a large dilation of the Bombieri–De Giorgi–Giusti minimal graph in $u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$5, and the solution is asymptotically of the form $u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$6, with $u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$7 the signed distance to $u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$8 (Pino et al., 2013). This gives a negative answer, for epigraphs in dimensions $u(x)=\frac{R^2-|x-x_0|^2}{2n},\qquad R=nc,$9, to the Berestycki–Caffarelli–Nirenberg question asking whether such domains must be a half-space, a cylinder, or the complement of a cylinder.

The same work pushes the Alexandrov–Serrin analogy into lower-dimensional noncompact geometry. If a hypersurface $$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$0 has suitable Jacobi nondegeneracy, a large dilation of $$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$1 can be perturbed to a nearby boundary carrying a Serrin solution. This produces domains near large dilations of compact nondegenerate CMC hypersurfaces, periodic cylindrically bounded domains in $$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$2 with boundary close to Delaunay surfaces, and domains near minimal surfaces of finite total curvature in $$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$3, including the catenoid and Costa-type surfaces (Pino et al., 2013).

Exterior-domain versions introduce a different kind of flexibility. For the $$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$4-Hessian equation

$$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$5

with interior-boundary conditions

$$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$6

and prescribed quadratic asymptotics at infinity, one obtains a free-boundary problem in which the bounded obstacle $$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$7 is determined by the far-field data. In the isotropic case $$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$8, the unique domain is a ball centered at $$\Delta u+f(u)=0 \quad \text{in }\Omega,\qquad u=0 \quad \text{on }\partial\Omega,\qquad \partial_\nu u=c \quad \text{on }\partial\Omega,$$9, but for anisotropic $f$0 the exterior domain need not be a ball (Wang et al., 2024). This suggests that overdetermined rigidity in exterior geometries is governed jointly by boundary data and asymptotic structure, rather than by boundary data alone.

4. Ambient geometry: manifolds, cones, and sub-Riemannian models

Once the Euclidean ambient space is replaced, the rigidity object changes with the geometry. On rotationally symmetric model manifolds

$f$1

the overdetermined torsion problem

$f$2

forces $f$3 to be an Euclidean ball centered at the pole, with

$f$4

provided $f$5 contains the pole, the Ricci curvature is nonnegative in the model sense, $f$6, and the solution satisfies the compatibility condition

$f$7

In Euclidean space this condition is automatic because $f$8; in curved settings it is a genuine restriction (Roncoroni, 2017).

In convex cones, the natural rigidity object is a spherical sector. For sector-like domains $f$9, the partially overdetermined problem

$$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$0

with

$$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$1

implies

$$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$2

so the domain is the intersection of the cone with a ball. The same conclusion holds in space forms for the equation $$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$3, yielding geodesic-ball sectors (Ciraolo et al., 2018). Related partially overdetermined results in convex cones and outside cones, especially in dimension $$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$4, identify ball sectors and exterior ball sectors by a $$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$5-function and integral identities (Lee et al., 2020).

In more general Riemannian manifolds, one can recover metric balls under structural assumptions on the ambient geometry. If $$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$6 admits a conformal vector field and

$$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$7

then a Pohozaev-type identity combined with Weinberger’s $$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$8-function

$$f(0)=f(1)=0,\qquad f(s)>0 \ \text{for } s\in(0,1),\qquad f'(1)<0.$$9

forces $w$0 to be a metric ball and $w$1 to be radial, provided $w$2 and a compatibility condition involving the scalar curvature and the conformal factor holds (Andrade et al., 2024).

In the Heisenberg group $w$3, Euclidean balls are replaced by gauge balls. The family

$w$4

with

$w$5

characterizes gauge balls within classes of cylindrically symmetric or toric symmetric domains; the model solution on $w$6 is

$w$7

(Martino et al., 2023).

Setting	Overdetermined problem	Rigidity object
Model manifolds	$w$8	Euclidean ball centered at the pole
Convex cones / space forms	Partial Dirichlet–Neumann overdetermination	Spherical sector or geodesic-ball sector
Riemannian manifolds with conformal field	$w$9	Metric ball
Heisenberg group	$w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$0	Gauge ball

5. Transmission media, free boundaries, and coupled variants

The two-phase Serrin problem replaces the Laplacian by a divergence-form operator with piecewise constant conductivity,

$w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$1

on a bounded domain $w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$2 containing an inclusion $w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$3. The overdetermined condition is

$w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$4

Near concentric balls $w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$5, $w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$6, the outer problem is locally solvable and unique for $w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$7, where $w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$8 is a finite resonance set defined explicitly in terms of $w''+f(w)=0,\qquad w(0)=0,\qquad w(+\infty)=1,$9, $u(x)=w(x_N)$0, and the spherical-harmonic degree $u(x)=w(x_N)$1. The proof uses shape derivatives and the implicit function theorem, and the numerical treatment minimizes a Kohn–Vogelius functional by a steepest descent algorithm (Cavallina et al., 2018).

At the exceptional conductivities $u(x)=w(x_N)$2, the linearized operator loses invertibility and symmetry breaking occurs. There are infinitely many bifurcating branches of nontrivial solutions $u(x)=w(x_N)$3, one for each critical mode $u(x)=w(x_N)$4, with

$u(x)=w(x_N)$5

Thus the two-phase problem admits nonradial outer boundaries even near a radial configuration (Cavallina et al., 2020).

Quantitative stability results show that this flexibility is limited near the one-phase regime. If $u(x)=w(x_N)$6 solves the one-phase torsion problem in $u(x)=w(x_N)$7, then

$u(x)=w(x_N)$8

with

$u(x)=w(x_N)$9

For the two-phase problem this yields

$C^2$00

so small contrast or small inclusion forces the outer boundary to be quantitatively close to a ball (Cavallina et al., 2021).

Other variants shift the overdetermination to incomplete or free-boundary settings. If the torsion equation is known only on $C^2$01,

$C^2$02

then a small unknown subdomain $C^2$03 still forces $C^2$04 to be close to the sphere $C^2$05, in boundary pseudodistance, Fraenkel-type asymmetry, and shell thickness $C^2$06 (Dipierro et al., 2020). For obstacle problems,

$C^2$07

radial obstacle data imply that $C^2$08 is a ball centered at the origin and $C^2$09 is radial; an approximate Neumann condition yields $C^2$10 (Nitti et al., 2023).

Partially overdetermined mixed problems in the half-space characterize capillary spherical caps. For

$C^2$11

one obtains

$C^2$12

and the contact angle satisfies $C^2$13 (Jia et al., 2023).

6. Weak formulations, rough domains, and sharp thresholds

A major recent development is the extension of Serrin rigidity to rough sets. For a bounded indecomposable set of finite perimeter $C^2$14 satisfying a uniform upper density bound on the reduced boundary, if

$C^2$15

and

$C^2$16

in the distributional sense, then $C^2$17 is a ball and $C^2$18 is exactly the Serrin profile. The result applies in particular to Lipschitz domains, answers the previously unresolved Lipschitz case affirmatively, and even extends to certain slit-domain formulations (Figalli et al., 2024).

An anisotropic Lipschitz-domain version replaces the Euclidean Laplacian by

$C^2$19

where $C^2$20 is the Wulff potential associated with a bounded ellipsoid $C^2$21. If

$C^2$22

and $C^2$23 satisfies the gradient nondegeneracy and $C^2$24-Dini-VMO assumptions near $C^2$25, then

$C^2$26

and $C^2$27 is homothetic to $C^2$28 (Dong et al., 5 Sep 2025).

In the plane, the weak formulation reveals a sharp regularity threshold. For a bounded Jordan domain with rectifiable boundary, the weak Serrin condition is equivalent to the harmonic quadrature identity

$C^2$29

If the domain is Smirnov, this identity forces $C^2$30 to be a disk. But there also exist rectifiable Jordan non-Smirnov domains satisfying the same identity, showing that the Smirnov hypothesis is sharp (Zhang, 3 May 2026). This resolves a common misconception: weak Serrin-type data do not by themselves force disks among all rectifiable Jordan domains; an additional factorization condition on the conformal map is essential.

The subject has also acquired a constructive perturbative branch. For the perturbed planar problem

$C^2$31

with

$C^2$32

analytic expansions

$C^2$33

produce infinitely many approximants whose PDE and boundary defects are super-polynomially small in $C^2$34, via an optimal truncation inspired by Nekhoroshev theory (Fortunati et al., 19 Mar 2026). This suggests a complementary viewpoint: beyond exact rigidity the overdetermined structure also supports highly accurate constructive near-solutions.

Taken together, these results show that Serrin-type overdetermined systems no longer denote a single theorem but a large rigidity–flexibility program. In smooth bounded Euclidean domains they still single out balls; in curved, anisotropic, and sub-Riemannian geometries they characterize the corresponding canonical shapes; in weak and rough settings they require refined measure-theoretic or conformal hypotheses; and in high-dimensional, multiphase, or asymptotic regimes they can support genuinely nonclassical geometries (Pino et al., 2013, Zhang, 3 May 2026).