Support-Function Approach

• Support-function approach is a method that represents convex objects via scalar dual encodings, simplifying geometric analysis and inference.
• It transforms complex moving hypersurface problems into fully nonlinear PDEs, allowing precise curvature control and convergence proofs.
• The method is applied in convex optimization, surface reconstruction, and econometric inference, ensuring strong differentiability and convexity properties.

The support-function approach denotes a family of formulations in which a convex body, a convex hypersurface, an oriented surface, or a closed convex identified set is encoded by its support function rather than by a moving embedding or an explicit boundary description. In geometric evolution, the support function of a convex hypersurface is $u(x)=\sup_{y\in M}\langle x,y\rangle$, attained at the unique point whose outer unit normal is $x$; for a compact convex set $K\subset \mathbb R^d$, it is $h_K(u)=\sup_{x\in K}\langle x,u\rangle$; for a closed convex identified set $\Theta(\phi)$, it is $$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$ (Ding et al., 2020, Soh et al., 2019, Liao et al., 2012). Across these settings, the method replaces the original object by a scalar function on the sphere or on a dual domain, and this replacement is used to derive fully nonlinear PDE, holomorphic representation formulas, finite-dimensional optimization schemes, and inferential procedures for set-valued parameters (Ivaki, 2015, Mendez et al., 2020, Antunes et al., 2018).

1. Foundational definition and dual encoding

For compact convex sets, the support function is a complete invariant: a compact convex set is determined by its support function, and distances between convex sets can be measured via support-function differences (Soh et al., 2019). The same dual encoding persists for general nonempty sets $A\subset X$ in finite-dimensional normed spaces, where the support function is

$\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$

with domain $$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$; moreover, $\sigma_A$ depends only on the closed convex hull of $x$0, and $x$1 (Zalinescu, 2013).

A central analytic advantage is that geometric quantities become explicit in support-function variables. For smooth strictly convex hypersurfaces in Euclidean space, the curvature radii matrix is

$x$2

and the Gauss curvature satisfies

$x$3

In the hypersurface setting of inverse Gauss map parametrization, the analogous formula is

$x$4

whose eigenvalues are the principal radii of curvature (Ivaki, 2015, Ding et al., 2020). This is the structural reason the support-function approach is effective: convexity becomes positivity of a matrix such as $x$5, while curvature becomes a function of its eigenvalues.

The same duality governs regularity questions. Since support functions are sublinear, they are convex and locally Lipschitz on the interior of their domain, and differentiability at $x$6 is equivalent to the subdifferential being a singleton. For support functions,

$x$7

so differentiability is equivalent to uniqueness of the maximizing exposed point (Zalinescu, 2013). This suggests a general principle: smoothness of the support function encodes geometric strictness of the underlying set.

2. Scalarization of convex hypersurface flows

In curvature-flow theory, the support-function approach converts a moving-hypersurface problem into a scalar fully nonlinear parabolic PDE on the sphere. In the expanding flow studied in “A class of curvature flows expanded by support function and curvature function” (Ding et al., 2020), the evolving hypersurface $x$8 satisfies

$x$9

where $K\subset \mathbb R^d$0 is the support function, $K\subset \mathbb R^d$1 is the outer unit normal, and $K\subset \mathbb R^d$2 is a smooth, symmetric, homogeneous of degree one, positive function of the principal curvature radii. After reparametrizing by the inverse Gauss map, the support function satisfies

$K\subset \mathbb R^d$3

in the unnormalized case, and

$K\subset \mathbb R^d$4

for the normalized flow (Ding et al., 2020).

The main theorem assumes

$K\subset \mathbb R^d$5

If the initial hypersurface is smooth, closed, uniformly convex, and encloses the origin, then the flow has a unique smooth uniformly convex solution for all time; after the appropriate normalization, the hypersurfaces converge exponentially in $K\subset \mathbb R^d$6 to a round sphere centered at the origin (Ding et al., 2020). The proof proceeds through $K\subset \mathbb R^d$7, gradient, and $K\subset \mathbb R^d$8 estimates for the support-function equation, followed by uniform parabolicity and higher-order regularity. The quantity

$K\subset \mathbb R^d$9

is studied via the maximum principle, and under the normalized flow

$h_K(u)=\sup_{x\in K}\langle x,u\rangle$0

is nonincreasing; the top eigenvalue of $h_K(u)=\sup_{x\in K}\langle x,u\rangle$1 satisfies a differential inequality of the form

$h_K(u)=\sup_{x\in K}\langle x,u\rangle$2

which yields curvature control (Ding et al., 2020).

A related formulation appears in “Deforming a hypersurface by Gauss curvature and support function” (Ivaki, 2015). There the normal flow is

$h_K(u)=\sup_{x\in K}\langle x,u\rangle$3

with $h_K(u)=\sup_{x\in K}\langle x,u\rangle$4, and the support function satisfies

$h_K(u)=\sup_{x\in K}\langle x,u\rangle$5

The support-function formulation yields a fixed-domain PDE on $h_K(u)=\sup_{x\in K}\langle x,u\rangle$6, makes strict convexity equivalent to ellipticity, and supports entropy-type functionals, monotonicity formulas, curvature estimates, and convergence to self-similar solutions such as origin-centered balls or ellipsoids, depending on the regime (Ivaki, 2015).

3. Representation theory for support-defined surface classes

In surface theory, the support-function approach is used not only to analyze evolution but also to reconstruct immersions from scalar data and holomorphic data. For an oriented surface $h_K(u)=\sup_{x\in K}\langle x,u\rangle$7 with Gauss map $h_K(u)=\sup_{x\in K}\langle x,u\rangle$8, the support function is

$h_K(u)=\sup_{x\in K}\langle x,u\rangle$9

and a standard reconstruction formula is

$\Theta(\phi)$0

where $\Theta(\phi)$1 is the support function written on the Gauss image (Mendez et al., 2020).

This framework underlies the class of SS-surfaces introduced in “Spherical type surfaces via support function” (Mendez et al., 2020). They are defined by the generalized Weingarten relation

$\Theta(\phi)$2

where $\Theta(\phi)$3. For surfaces with $\Theta(\phi)$4, this is equivalent to

$\Theta(\phi)$5

Writing $\Theta(\phi)$6 gives $\Theta(\phi)$7, so $\Theta(\phi)$8 is harmonic. The resulting Weierstrass-type representation depends on two holomorphic functions $\Theta(\phi)$9 and $$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$0, with $$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$1 prescribing the Gauss map and $$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$2 prescribing the support datum (Mendez et al., 2020). The same paper classifies rotational SS-surfaces and proves the rigidity statement that every compact connected SS-surface is a sphere (Mendez et al., 2020).

A more general harmonic-type construction appears in “Surfaces with quadratic support function of harmonic type” (Corro et al., 27 Jan 2026). HQSF-surfaces are defined by

$$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$3

which reduces to the earlier QSF relation when $$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$4. The support-function formalism leads to a Weierstrass-type representation with three holomorphic functions $$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$5, with

$$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$6

and

$$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$7

The same method yields a classification of rotational HQSF-surfaces into three cases according to the sign of

$$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$8

producing exponential-type, trigonometric, or degenerate linear-exponential profile curves (Corro et al., 27 Jan 2026).

4. Convex-body optimization, reconstruction, and exhauster reduction

In computational convex geometry, the support-function approach serves as a global parametrization of convex bodies. For a convex set $$S_\phi(\nu)=\sup_{\theta\in\Theta(\phi)}\theta^T\nu$$9, the support function

$A\subset X$0

determines the boundary through

$A\subset X$1

and truncating the Fourier expansion in two dimensions or the spherical-harmonic expansion in three dimensions turns shape optimization into a finite-dimensional constrained optimization problem (Antunes et al., 2018). Convexity becomes the inequality $A\subset X$2 in two dimensions, width constraints become pointwise conditions on $A\subset X$3, and constant width becomes an algebraic spectral condition: in two dimensions $A\subset X$4 and all even nonconstant Fourier modes vanish; in three dimensions the even nonconstant spherical harmonic coefficients vanish (Antunes et al., 2018). This framework is used for optimization of volume, perimeter, and Dirichlet Laplace eigenvalues under convexity, diameter, constant-width, and inclusion constraints, and it produces a numerical confirmation of Meissner’s conjecture for three-dimensional bodies of constant width (Antunes et al., 2018).

A different inverse problem is reconstruction from support-function evaluations. In “Fitting Tractable Convex Sets to Support Function Evaluations” (Soh et al., 2019), the data model is

$A\subset X$5

and the estimator is restricted to structured families of convex sets of the form $A\subset X$6, where $A\subset X$7 is a fixed tractable template such as the simplex $A\subset X$8 or the spectraplex $A\subset X$9. The key identity

$\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$0

reduces reconstruction to optimization over linear maps. Under the stated assumptions, the empirical minimizers are strongly consistent, asymptotic normality holds under stronger identifiability and differentiability assumptions, and exposed faces can be preserved under additional geometric conditions (Soh et al., 2019).

In nonsmooth optimization, support functions also provide a reduction mechanism for upper exhausters of positively homogeneous functions on $\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$1. For an upper exhauster $\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$2, each set $\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$3 is represented by the support-function curve

$\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$4

and the exhauster representation becomes the lower envelope

$\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$5

The reduction problem is then a geometric comparison of sinusoidal graphs in $\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$6-space: a set is removable if its curve never contributes essentially to the lower envelope, and inclusion-minimality means every curve contributes on a nontrivial angular interval (Tozkan, 2020).

5. Differentiability, convexity, and set-valued inference

The relation between convexity of a set and differentiability of its support function is classical for compact sets and substantially subtler for unbounded sets. For a compact convex set with nonempty interior in finite dimensions, the support function is differentiable excepting the origin if and only if the set is strictly convex (Zalinescu, 2013). More generally, if $\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$7 is unbounded, differentiability of $\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$8 on $\sigma_A(x^*) := \sup\{(x^*,u)\mid u\in A\},$9 is equivalent to a strict-convexity-type condition on

$$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$0

and under the cone-invariance hypothesis

$$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$1

with $$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$2 pointed, closed, convex, and with nonempty interior, differentiability on $$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$3 can force convexity of $$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$4 together with strong interior strict-convexity properties (Zalinescu, 2013).

The same support-function machinery has an economic interpretation. For a production function $$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$5, the upper level sets are

$$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$6

and the associated cost function is

$$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$7

Thus differentiability of the cost function is exactly differentiability of the support function of the level set at the opposite price vector (Zalinescu, 2013).

In econometrics, the support-function approach provides a natural Bayesian treatment of partially identified models. When the identified set $$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$8 is closed and convex, it is completely characterized by

$$\operatorname{dom}\sigma_A=\{x^*\in X^*: \sigma_A(x^*)<+\infty\}$$9

The posterior of $\sigma_A$0 depends only on the posterior of $\sigma_A$1, so inference on the identified set does not require specifying a prior on $\sigma_A$2 (Liao et al., 2012). The paper “Semi-parametric Bayesian Partially Identified Models based on Support Function” proves a Bernstein–von Mises theorem for the posterior of the support function and constructs the two-sided Bayesian credible set

$\sigma_A$3

which has asymptotically correct frequentist coverage probability for the identified set (Liao et al., 2012).

6. High-dimensional asymptotics and analytic extremals

The support-function approach also appears in asymptotic stochastic geometry. For the high-dimensional Poisson polytope

$\sigma_A$4

the support function in direction $\sigma_A$5 is

$\sigma_A$6

Its exact distribution is governed by spherical-cap volumes: $\sigma_A$7 As $\sigma_A$8, three regimes appear according to the scale of $\sigma_A$9: subcritical, critical, and supercritical. In each regime, the support function has explicit first-order asymptotics, and after suitable centering and scaling the fluctuations converge to a standard Gumbel random variable; analogous first-order statements hold for the infimum over a fixed $x$00-dimensional section of directions (Calka et al., 2024).

The phrase “support function” also appears in a distinct analytic sense in pluripotential theory. “An Optimal Support Function related to the strong openness conjecture” proves that the optimal support function for weighted $x$01 integrations on superlevel sets of a negative plurisubharmonic weight is

$x$02

More precisely, for the extremal quantity $x$03, the paper establishes

$x$04

and shows that the factor $x$05 is optimal; this quantitative estimate implies the strong openness property of multiplier ideal sheaves (Guan et al., 2021).

Taken together, these developments show that the support-function approach is not a single theorem but a recurrent method of scalarization and dualization. In geometric PDE it converts moving convex hypersurfaces into fully nonlinear equations on the sphere; in surface theory it yields holomorphic representation formulas; in convex optimization it turns difficult shape constraints into coefficient conditions; in inference it translates set estimation into support-function asymptotics; and in high-dimensional or analytic settings it isolates the extremal quantity that governs the problem (Ding et al., 2020, Antunes et al., 2018, Liao et al., 2012).