Inverse Curvature Flows Overview

• Inverse curvature flows are geometric evolution equations in which hypersurfaces move normally with speeds given by the reciprocal of curvature measures, underpinning shape analysis and rigidity results.
• Different formulations like the inverse mean curvature flow and inverse σₖ flows are developed under strict geometric and analytic conditions to ensure long-term existence and convergence to spherical configurations.
• These flows are applied across Euclidean, hyperbolic, warped product, and capillary settings, providing insights into pinching phenomena, asymptotic roundness, and the derivation of sharp geometric inequalities.

Searching arXiv for recent and foundational papers on inverse curvature flows to ground the article. Searching arXiv for "inverse curvature flow" and related variants. Inverse curvature flows are geometric evolution equations in which a hypersurface moves in its normal direction with speed given by the reciprocal, or an inverse power, of a curvature quantity. In Euclidean space a standard formulation is

$$(\partial_t F)^\perp=-\frac{1}{\rho(\kappa_1,\dots,\kappa_n)}\,\nu,$$

equivalently

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$

with the inverse mean curvature flow (IMCF) as the special case $f=\frac{1}{H}$, $H=\sum_{i=1}^n \kappa_i$ (Chin et al., 2018). Across Euclidean, hyperbolic, warped-product, capillary, and Lorentzian settings, the subject is organized by ellipticity, homogeneity, and concavity of the speed, and it is closely connected to pinching, asymptotical roundness, geometric inequalities, and general relativity (Scheuer, 2014).

1. Basic formulation and admissible speeds

The common structural feature is that the speed is inverse in curvature. In Euclidean space one often starts from a symmetric degree-$1$ curvature function $\rho$ on a cone $\Gamma\subset\mathbb{R}^n$ containing $(1,\dots,1)$, and writes the speed as $-1/\rho$. For well-posedness, the standard hypotheses are that $\rho\in C^2(\Gamma)$, is positive and symmetric, is homogeneous of degree $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$0, is strictly elliptic in the sense $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$1, and is concave in the sense that its Hessian is semi-negative definite on $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$2 (Chin et al., 2018). Equivalent formulations use a degree $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$3 speed $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$4 or, more generally, $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$5 with $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$6 and $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$7 1-homogeneous (Scheuer, 2014).

The model examples recur throughout the literature: $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$8, $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$9, $f=\frac{1}{H}$0, $f=\frac{1}{H}$1, normalized power means, and Hessian quotients all appear as admissible speeds under variants of the same monotonicity, homogeneity, and concavity assumptions (Chin et al., 2018, Wang et al., 2020, Lu, 2016). In Euclidean problems with $f=\frac{1}{H}$2, one typically allows $f=\frac{1}{H}$3 on a cone $f=\frac{1}{H}$4 with $f=\frac{1}{H}$5, whereas for $f=\frac{1}{H}$6 the regime $f=\frac{1}{H}$7 and strict convexity play a special role [(Scheuer, 2014); (Gerhardt, 2011)].

Admissibility is therefore geometric as well as analytic. A hypersurface is $f=\frac{1}{H}$8-admissible when its principal curvatures remain in the cone where the speed is defined; in hyperbolic and warped settings, star-shapedness or horo-convexity is frequently added to guarantee global graphical parametrizations and uniform parabolicity (Wang et al., 2020, Scheuer, 2017).

2. Geometric settings and equivalent formulations

In Euclidean space, star-shaped hypersurfaces are commonly written as radial graphs over $f=\frac{1}{H}$9, and the support function

$H=\sum_{i=1}^n \kappa_i$0

is central in both analysis and rigidity. Under star-shapedness, inverse curvature flows in large classes of degree-$H=\sum_{i=1}^n \kappa_i$1 speeds admit long-time graphical formulations, and rescaled flows approach spherical geometry [(Chin et al., 2018); (Scheuer, 2014)].

In hyperbolic space, the ambient geometry changes the natural notion of convexity. Horospheres have principal curvatures identically $H=\sum_{i=1}^n \kappa_i$2, so Wang–Wei–Zhou introduce shifted principal curvatures

$H=\sum_{i=1}^n \kappa_i$3

and call a hypersurface horo-convex precisely when $H=\sum_{i=1}^n \kappa_i$4 for all $H=\sum_{i=1}^n \kappa_i$5. They also use a horospherical support function $H=\sum_{i=1}^n \kappa_i$6 on $H=\sum_{i=1}^n \kappa_i$7, for which the shifted flow becomes a scalar parabolic equation $H=\sum_{i=1}^n \kappa_i$8 (Wang et al., 2020). This shifted formulation is tailored to hyperbolic geometry because geodesic spheres satisfy $H=\sum_{i=1}^n \kappa_i$9 while horospheres realize the baseline $1$0.

Warped-product and asymptotically hyperbolic settings generalize these graph descriptions. In Riemannian warped products with metric $1$1, graphical hypersurfaces carry a generalized support function $1$2, and the shape operator separates into a radial term $1$3 and Hessian terms of the graph function (Scheuer, 2017). In anti-de Sitter–Schwarzschild manifolds, the metric takes the form $1$4 with

$1$5

and inverse flows are studied non-parametrically as radial graphs over $1$6 (Chen et al., 2016). Capillary problems in the Euclidean unit ball add a boundary angle condition

$1$7

on $1$8, together with a tangential correction preserving the contact angle during the flow (Pan et al., 16 Jul 2025).

Low-dimensional analogues preserve the same inverse-curvature principle. For plane curves the basic equation is $1$9 (Kröner, 2014). In two-dimensional space forms the adapted speed

$\rho$0

is equivalent in the Euclidean case to standard inverse curvature flow after continuous rescaling (Kwong et al., 2021). For closed Legendre curves, the normal velocity is prescribed by $\rho$1, where $\rho$2 and $\rho$3 are the Legendre curvature data; this extends inverse curvature flow to fronts with cusp-type singularities (Kagaya et al., 6 Oct 2025).

3. Evolution equations, pinching, and asymptotic roundness

The analytic backbone is a family of evolution identities for geometric quantities under a normal flow $\rho$4: $\rho$5

$\rho$6

For inverse speeds these identities are combined with maximum principles, tensor maximum principles, and concavity inequalities to control the traceless second fundamental form, the curvature ratios, and the support or graph functions [(Chin et al., 2018); (Scheuer, 2014)].

In Euclidean space, the large-scale asymptotic picture is well developed. For flows

$\rho$7

with $\rho$8 positive, symmetric, monotone, concave, 1-homogeneous, and normalized by $\rho$9, Gerhardt showed that for $\Gamma\subset\mathbb{R}^n$0 star-shaped admissible hypersurfaces exist for all time and for $\Gamma\subset\mathbb{R}^n$1 strictly convex hypersurfaces expand to infinity in finite time; in both cases the properly rescaled flows converge to the unit sphere (Gerhardt, 2011). A sharper Euclidean result proves asymptotical roundness: there exists a point $\Gamma\subset\mathbb{R}^n$2 and a spherical solution $\Gamma\subset\mathbb{R}^n$3 centered at $\Gamma\subset\mathbb{R}^n$4 such that

$\Gamma\subset\mathbb{R}^n$5

and consequently $\Gamma\subset\mathbb{R}^n$6 at the same rate (Scheuer, 2014).

Pinching theory in space forms abstracts the same mechanism. Wei proved that for strictly convex hypersurfaces in $\Gamma\subset\mathbb{R}^n$7, with $\Gamma\subset\mathbb{R}^n$8, flowing by $\Gamma\subset\mathbb{R}^n$9 under inverse-concavity and boundary conditions on the dual speed, the ratio $(1,\dots,1)$0 stays uniformly controlled by its initial value. In hyperbolic space this yields smooth convergence, after normalization, to a geodesic sphere (Wei, 2017). Hyperbolic inverse flows by general 1-homogeneous concave speeds also admit long-time existence for star-shaped initial data, and the leaves become strongly convex exponentially fast and increasingly umbilic, with principal curvatures converging to $(1,\dots,1)$1 (Gerhardt, 2011).

Hyperbolic asymptotics are subtler than Euclidean ones. For the classical non-shifted inverse curvature flow in $(1,\dots,1)$2, principal curvatures may converge to $(1,\dots,1)$3 without asymptotic roundness; Hung–Wang constructed a counterexample in $(1,\dots,1)$4 showing that the limiting shape is not necessarily round (Wang et al., 2020). The shifted formulation corrects this: for horo-convex initial hypersurfaces and $(1,\dots,1)$5, shifted inverse curvature flows preserve horo-convexity, the shifted principal curvatures satisfy $(1,\dots,1)$6, and the hypersurfaces become arbitrarily close to geodesic spheres with

$(1,\dots,1)$7

under broad concavity or inverse-concavity assumptions (Wang et al., 2020).

Asymptotically hyperbolic and warped spaces exhibit analogous but ambient-dependent behavior. In anti-de Sitter–Schwarzschild manifolds, star-shaped solutions to $(1,\dots,1)$8 exist for all time and satisfy

$(1,\dots,1)$9

so the principal curvatures converge exponentially to $-1/\rho$0 (Chen et al., 2016). The inverse Hessian quotient flow in the same manifold yields the same qualitative conclusion for $-1/\rho$1 (Lu, 2016). In general Riemannian warped products with $-1/\rho$2 and $-1/\rho$3, compact graphical solutions to $-1/\rho$4, $-1/\rho$5, exist for all time and satisfy quantitative umbilicity estimates of the form

$-1/\rho$6

with sharper decay in bounded-$-1/\rho$7 regimes (Scheuer, 2017).

4. Self-similar, self-expanding, and self-conformal solutions

Self-similar solutions are the stationary profiles of inverse curvature dynamics after an ambient symmetry. For degree-$-1/\rho$8 speeds in Euclidean space, a homothetic self-similar solution satisfies

$-1/\rho$9

and for IMCF this is $\rho\in C^2(\Gamma)$0 (Chin et al., 2018). More generally, a self-conformal solution is generated by a conformal Killing field $\rho\in C^2(\Gamma)$1 with

$\rho\in C^2(\Gamma)$2

and in Euclidean space every conformal Killing field has the form

$\rho\in C^2(\Gamma)$3

with divergence an affine linear function (Chin et al., 2018).

A central rigidity theorem states that round spheres are the only compact self-expanders to a large class of inverse curvature flows by homogeneous symmetric functions. For the inverse $\rho\in C^2(\Gamma)$4-flow

$\rho\in C^2(\Gamma)$5

with $\rho\in C^2(\Gamma)$6 degree $\rho\in C^2(\Gamma)$7, symmetric, and parabolic, compact self-expanders satisfy $\rho\in C^2(\Gamma)$8, and the only compact solutions are round spheres (Chow et al., 2017). In the non-compact category, asymptotically cylindrical self-expanders must be rotationally symmetric, and under a uniform parabolicity condition there exist complete rotationally symmetric self-expanders asymptotic to two round cylinders with different radii (Chow et al., 2017).

The self-conformal theory strengthens the homothetic picture. For closed self-conformal IMCF, the only solutions are round spheres in dimension $\rho\in C^2(\Gamma)$9, and in dimension $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$00 the same conclusion holds when the generating conformal Killing field has constant divergence. For the broader class of inverse curvature speeds $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$01 with $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$02, the only closed, star-shaped self-conformal solutions are also round spheres (Chin et al., 2018). The proofs combine conformal invariance of the Willmore functional in low dimension, monotonicity of the Willmore energy along IMCF, Hsiung–Minkowski identities, and a conformally invariant tensor

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$03

which vanishes exactly on totally umbilic hypersurfaces (Chin et al., 2018).

5. Monotone quantities, constrained flows, and geometric inequalities

Inverse curvature flows are especially powerful when paired with monotone integral quantities. In Euclidean space, Minkowski formulas give

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$04

and more generally

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$05

For IMCF, the Willmore energy

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$06

is nonincreasing, with equality only when the hypersurface is umbilic, while for the flow by $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$07 the Guan–Li quantities $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$08 are monotone decreasing and constant exactly on round spheres (Chin et al., 2018).

A quantitative refinement appears in the stability theory of quermassintegrals. For the inverse $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$09-ratio flow

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$10

a rescaling $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$11, with $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$12, preserves $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$13 and makes $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$14 nonincreasing. Near the sphere, the decay rate of the $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$15-th quermassintegral dominates the decay of a natural asymmetry functional, leading to a stability inequality for nearly spherical sets (VanBlargan et al., 2022).

Warped-product analogues replace global rescaling by local lower-order corrections. In warped spaces one studies

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$16

or, in a broader notation,

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$17

Under suitable assumptions on the warping function and the initial graph, these locally constrained inverse curvature flows exist for all time and converge smoothly to a coordinate slice (Scheuer et al., 2017, Scheuer, 2020). The associated monotone quantities yield new Minkowski-type and weighted isoperimetric inequalities in anti-de Sitter–Schwarzschild and hyperbolic spaces (Scheuer et al., 2017, Scheuer, 2020).

Boundary geometry produces a further class of inequality-generating flows. For strictly convex capillary hypersurfaces in the unit ball with contact angle $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$18, inverse curvature flows with speed $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$19 and capillary boundary condition preserve the angle, exist up to a finite time $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$20, and converge smoothly to a flat ball $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$21. In the free-boundary case $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$22, IMCF then yields Alexandrov–Fenchel inequalities for weakly convex hypersurfaces with equality only for flat disks (Pan et al., 16 Jul 2025).

6. Low-dimensional, singular, and nonclassical variants

In one dimension, inverse curvature flow becomes especially explicit. For a strictly convex embedded plane curve,

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$23

the unnormalized length satisfies $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$24, and after the length-preserving normalization $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$25, the normalized equation is

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$26

An Andrews–Bryan chord–arc estimate yields

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$27

and the normalized flow converges smoothly to the unit circle (Kröner, 2014).

For convex curves in two-dimensional space forms, the adapted flow

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$28

preserves length, makes enclosed area nondecreasing, and converges exponentially fast to the unique geodesic circle centered at the reference point and having the same length as the initial curve (Kwong et al., 2021). This yields the isoperimetric inequality

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$29

for convex curves in $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$30, as well as weighted inequalities and a counterexample to the $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$31 case of a conjecture of Girão–Pinheiro (Kwong et al., 2021).

Inverse curvature dynamics also extend beyond regular embeddings. For $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$32-convex closed Legendre curves, the special inverse curvature flow is

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$33

After a normalization fixing $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$34, the curvature datum $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$35 satisfies the linear reaction–diffusion equation

$$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$36

so the entire asymptotic classification is controlled by Fourier modes. The zero number of $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$37, hence the number of singular cusps, is non-increasing, and normalized solutions converge to explicitly classified self-similar profiles (Kagaya et al., 6 Oct 2025).

Lorentzian and cosmological variants replace spheres by time slices. In ARW spacetimes, the rescaled leaves of the inverse curvature flow considered by Gerhardt converge to the graph of a constant function (Gerhardt, 2011). This suggests that, in addition to Euclidean and hyperbolic asymptotic roundness, inverse curvature flows can enforce asymptotic homogeneity relative to the ambient foliation.

Several open directions are explicit in the current literature. In the shifted hyperbolic setting, $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$38 is delicate because horo-convexity may be lost quickly, and the paper exhibits a counterexample for shifted IMCF with $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$39 (Wang et al., 2020). In capillary geometry, extending the free-boundary Alexandrov–Fenchel family to general contact angle $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$40 within the same IMCF framework remains open (Pan et al., 16 Jul 2025). In the Legendre setting, extending beyond $$\partial_t x=f(\kappa_1,\dots,\kappa_n)\,\nu,\qquad f(\lambda\kappa)=\lambda^{-1}f(\kappa),$$41-convexity and connecting with higher-dimensional Legendrian flows are identified as natural directions (Kagaya et al., 6 Oct 2025).