---
title: Locally Constrained Inverse Curvature Flow
url: https://www.emergentmind.com/topics/locally-constrained-inverse-curvature-flow
type: topic
---

# Locally Constrained Inverse Curvature Flow

Locally constrained inverse curvature flow is a class of geometric evolutions in which a hypersurface moves by an inverse-curvature speed modified by a pointwise lower-order term, typically built from the radial coordinate, a support function, or an ambient conformal Killing field. In the warped-product and space-form settings, representative speeds include
\[
\partial_t x=\left(\frac{n}{F}-\frac{u}{\lambda'}\right)\nu,\qquad
\partial_t x=\left(\frac{\vartheta'(u)}{F}-s\right)\nu,
\]
and, in de Sitter space,
\[
\partial_t x=\left(u-\frac{\lambda' E_1}{E_2}\right)\nu.
\]
The defining feature is that the correction term is local rather than a global average chosen to preserve enclosed volume or area. This local design is closely tied to Minkowski identities in warped products, and it is used to obtain long-time existence, smooth convergence to radial slices or geodesic spheres, and sharp geometric inequalities such as Minkowski, Alexandrov–Fenchel, Michael–Simon, and weighted isoperimetric inequalities [1708.06125] [2005.11236] [2507.11149].

## 1. Definition and characteristic structure

The phrase “locally constrained” refers to a pointwise correction imposed through local geometric quantities such as the support function and the ambient radial weight, rather than by subtracting a global average. In the warped-product formulation
\[
N=(a,b)\times S_0,\qquad \bar g=dr^2+\vartheta^2(r)\sigma,
\]
a central model is
\[
\partial_t x=\left(\frac{\vartheta'(u)}{F}-s\right)\nu,
\]
where \(s=\bar g(\vartheta \partial_r,\nu)\) is the support function and \(F\) is a symmetric, positive, strictly monotone, degree-one, concave curvature function on \(\Gamma_+\) with \(F|_{\partial\Gamma_+}=0\) and \(F(1,\dots,1)=1\) [2005.11236]. A closely related formulation in warped products over \(\mathbb S^n\) is
\[
\partial_t x=\left(\frac{n}{F}-\frac{u}{\lambda'}\right)\nu,
\]
with \(u=\bar g(\lambda\partial_r,\nu)\), again emphasizing that the lower-order correction is local and depends only on the same spacetime point [1708.06125].

In space forms, the inverse-curvature term is often expressed as a Hessian quotient. In hyperbolic space one studies
\[
\partial_t X=\left(\frac{\lambda'(r)}{F(\kappa)}-u\right)\nu,\qquad
F(\kappa)=\frac{E_m(\kappa)}{E_{m-1}(\kappa)},
\]
equivalently
\[
\partial_t X=\left(\lambda'(r)\frac{E_{m-1}(\kappa)}{E_m(\kappa)}-u\right)\nu,
\]
and also a shifted-curvature variant
\[
\partial_t X=\left((\lambda'(r)-u)\frac{E_{m-1}(\tilde\kappa)}{E_m(\tilde\kappa)}-u\right)\nu,\qquad \tilde\kappa_i=\kappa_i-1.
\]
These flows preserve a quermassintegral or a shifted quermassintegral and decrease the next one, with the compensating term \(u\) or \(\lambda'-u\) supplied pointwise by the hyperbolic Minkowski identities [2002.10643].

The same principle appears in low dimension. For convex curves in the simply connected space forms \(M^2(K)\), the speed
\[
\partial_t X=\left(\frac{\phi'(r)}{\kappa}-u\right)\nu
\]
is an inverse-curvature flow modified by the support function \(u=\langle \phi(r)\partial_r,\nu\rangle\). In the Euclidean plane, this is equivalent, up to rescaling, to the standard inverse curvature flow, while in the sphere and hyperbolic plane the factor \(\phi'(r)\) encodes the ambient curvature correction [2103.04338].

## 2. Geometric background and ambient identities

The ambient settings in which locally constrained inverse curvature flows are developed are predominantly warped products. In hyperbolic space,
\[
\mathbb H^{n+1}=[0,\infty)\times \mathbb S^n,\qquad \bar g=dr^2+\lambda(r)^2 g_{\mathbb S^n},\qquad \lambda(r)=\sinh r,\quad \lambda'(r)=\cosh r,
\]
and the conformal Killing field is \(V=\lambda\partial_r\), with support function \(u=\langle V,\nu\rangle\) [2002.10643]. In the sphere,
\[
\bar g=d\rho^2+\phi(\rho)^2dz^2,\qquad \phi(\rho)=\sin\rho,\qquad \phi'(\rho)=\cos\rho,
\]
and the corresponding support function is \(u=\langle V,\nu\rangle\) for \(V=\phi(\rho)\partial_\rho\) [2403.07281]. In de Sitter space,
\[
\mathbb S^n_1=(0,\infty)\times \mathbb S^n,\qquad \bar g=-dr^2+\lambda^2(r)\sigma,\qquad \lambda(r)=\cosh r,\quad \lambda'(r)=\sinh r,
\]
and a spacelike graph \(\Sigma=\{(r(\theta),\theta):\theta\in\mathbb S^n\}\) has induced metric
\[
g_{ij}=\lambda^2\sigma_{ij}-r_i r_j,
\]
future-directed timelike unit normal
\[
\nu=\frac1{\upsilon}\big(\partial_r+\lambda^{-2}Dr\big),\qquad \upsilon=\sqrt{1-\lambda^{-2}|Dr|^2},
\]
and support function
\[
u=-\bar g(\lambda\partial_r,\nu)=\lambda \upsilon^{-1}>1
\]
in one sign convention, while a related de Sitter formulation uses \(u=\bar g(V,\nu)\) and \(\Theta=u/\lambda'\) [2507.11149] [2512.19285].

The support function is not an auxiliary decoration. It is the local quantity that converts ambient conformal structure into integral identities. In hyperbolic space,
\[
\int_M \lambda' E_m(\kappa)\,d\mu=\int_M u\,E_{m+1}(\kappa)\,d\mu,
\]
and for shifted principal curvatures,
\[
\int_M (\lambda'-u)E_m(\tilde\kappa)\,d\mu=\int_M u\,E_{m+1}(\tilde\kappa)\,d\mu.
\]
In the sphere,
\[
\int_M p_{k+1}u\,d\mu=\int_M \phi' p_k\,d\mu.
\]
In de Sitter space,
\[
\int_\Sigma u E_k\, d\mu=\int_\Sigma \lambda' E_{k-1}\, d\mu.
\]
These Minkowski-type formulas are the mechanism by which one functional can be preserved while another becomes monotone [2002.10643] [2403.07281] [2507.11149].

The curvature data are usually expressed through normalized elementary symmetric functions. In de Sitter and hyperbolic formulations,
\[
E_k=\binom{n}{k}^{-1}\sigma_k,\qquad E_0=1,
\]
and the admissible cone is
\[
\Gamma_k^+=\{\kappa\in \mathbb R^n:\ E_l(\kappa)>0,\ l=1,\dots,k\}.
\]
In hyperbolic \(h\)-convexity, the shifted variables \(\tilde\kappa_i=\kappa_i-1\) are natural because \(h\)-convexity is precisely \(\kappa_i\ge 1\) [2002.10643] [2507.11149].

## 3. Analytic architecture

The standard curvature-flow architecture in this area consists of short-time existence, \(C^0\) bounds, gradient estimates, lower and upper curvature bounds, long-time existence, convergence, and then the geometric inequality. The local constraint enters at each stage, not only in the monotonicity formulas but also in the maximum-principle estimates [2507.11149].

Graph formulations are fundamental. In warped products one writes \(M_t=\{(r(t,y),y)\}\), or, after the change of variable \(\varphi(r)=\int_a^r \lambda(s)^{-1}\,ds\), a scalar PDE on \(\mathbb S^n\) for \(\varphi\) [1708.06125]. In de Sitter space the locally constrained inverse Hessian quotient flow
\[
\partial_t x=\left(u-\frac{\lambda' E_1}{E_2}\right)\nu
\]
becomes
\[
\partial_t r=\upsilon\left(u-\frac{\lambda'}{F}\right),\qquad F=\frac{E_2}{E_1},
\]
and short-time existence follows from fully nonlinear parabolic theory because \(F\) is increasing in \(\Gamma_2^+\) [2507.11149].

A characteristic \(C^0\) feature of local constraints is immediate radial control. In warped products one obtains
\[
\inf r_0\le r(t,\cdot)\le \sup r_0,
\]
and in the spherical \(p_2\)-flow
\[
\partial_t X=(\phi'-u p_2)\nu
\]
one similarly has
\[
\min_{\mathbb S^n}\rho(\cdot,0)\le \rho(z,t)\le \max_{\mathbb S^n}\rho(\cdot,0)
\]
by the maximum principle [1708.06125] [2403.07281].

Gradient control is typically encoded in the support function. In de Sitter space the scalar operator \(\mathscr L\) satisfies
\[
\mathscr L u = \lambda' u\left(1-\frac{1}{F^2}F^{pq}h_p^m h_{mq}\right) -\frac1F |\nabla \lambda'|^2,
\]
and since
\[
F^2\le F^{pq}h_p^m h_{mq},
\]
the coefficient of \(\lambda' u\) is nonpositive, yielding a uniform upper bound for \(u\). Because \(u=\lambda/\upsilon\), this preserves spacelikeness and controls the graphical gradient [2507.11149]. In two-dimensional space forms one gets two-sided support-function bounds
\[
C_1\ge u\ge C_2>0
\]
from radial and gradient estimates, and this keeps the curve strictly star-shaped [2103.04338].

Curvature control depends on the chosen inverse operator. In hyperbolic space, preservation of \(h\)-convexity for
\[
F=\frac{E_m}{E_{m-1}}
\]
and for the shifted quotient
\[
F(\tilde\kappa)=\frac{E_m(\tilde\kappa)}{E_{m-1}(\tilde\kappa)}
\]
is proved by Hamilton–Andrews tensor maximum principles applied to \(S_{ij}=h_{ij}-g_{ij}\), using concavity and inverse-concavity estimates [2002.10643]. In the sphere, the new algebraic estimate for \(F=p_2\),
\[
\frac{F^{ij}(h^2)_{ij}}{F^2}-1>\frac n2(F-1),
\]
substitutes for the missing quotient structure and closes the \(C^2\) estimate [2403.07281]. In de Sitter space, the auxiliary quantity
\[
\omega=\ln F+\ln u-\ln\lambda'
\]
yields, for \(F=E_2/E_1\), a uniform quotient bound
\[
\frac1C\le \frac{E_2}{E_1}\le C,
\]
which in the \(k=2\) case implies full curvature control [2507.11149].

## 4. Ambient settings and representative flows

The literature now contains several distinct ambient realizations of locally constrained inverse curvature flow, together with nearby but nonidentical frameworks.

| Ambient setting | Representative speed | Asymptotic limit |
|---|---|---|
| Warped products | \(\partial_t x=\left(\frac{\vartheta'(u)}{F}-s\right)\nu\) | radial slice |
| Hyperbolic space | \(\partial_t X=\left(\lambda'\frac{E_{m-1}}{E_m}-u\right)\nu\) | geodesic sphere |
| De Sitter space | \(\partial_t x=\left(u-\frac{\lambda' E_1}{E_2}\right)\nu\) or \(\partial_t x=\left(\frac{u}{\lambda'}-\frac{E_{k-1}}{E_k}\right)\nu\) | coordinate slice |
| Two-dimensional space forms | \(\partial_t X=\left(\frac{\phi'(r)}{\kappa}-u\right)\nu\) | centered geodesic circle |

In warped products beyond constant curvature, strict convex graphical hypersurfaces evolving by
\[
\partial_t x=\left(\frac{\vartheta'(u)}{F}-s\right)\nu
\]
exist for all time and converge smoothly to a radial slice under the stated structural assumptions on \(\vartheta\) and \(F\) [2005.11236]. In hyperbolic space, smooth closed \(h\)-convex hypersurfaces under
\[
\partial_t X=\left(\lambda'(r)\frac{E_{m-1}}{E_m}-u\right)\nu
\]
or its shifted analogue remain strictly \(h\)-convex for positive time and converge smoothly and exponentially to a geodesic sphere [2002.10643]. In de Sitter space, the inverse Hessian quotient flow with \(F=E_2/E_1\) exists for all \(t>0\) and converges to a coordinate slice, while a broader de Sitter flow
\[
\partial_t x=\left(\Theta-\frac1F\right)\nu,\qquad \Theta=\frac{u}{\lambda'},\quad F=\frac{E_k}{E_{k-1}},
\]
is proved to converge to a slice under the pinching condition \(\mathcal W\le \Theta \mathcal I\) [2507.11149] [2512.19285].

Not every locally constrained flow in the same literature is inverse. In the sphere, the main new analytic theorem of “Locally constrained flows and geometric inequalities in sphere” concerns
\[
\partial_t X=(\phi'-u p_2)\nu,
\]
which is locally constrained but not inverse; the same paper separately imports the genuinely inverse flow
\[
\partial_t X=\left(\frac{u}{F}-\phi'\right)\nu,\qquad F=\frac{p_k}{p_{k-1}},
\]
from Scheuer–Xia for its inequality applications [2403.07281]. By contrast, the cocompact Minkowski-space flow
\[
\partial_t X=-\frac1F\nu,\qquad F=s_k^{1/k},
\]
is explicitly described as unconstrained; its normalization arises only through the rescaling \(\tilde X=e^tX\), so it is a useful comparison point rather than an example of local constraint [2606.14082].

## 5. Monotone quantities and geometric inequalities

The main geometric reason for introducing local constraints is that the speed can be tuned to preserve one integral quantity while monotonically improving another. In warped products this produces new Minkowski-type inequalities. For
\[
\partial_t x=\left(\frac{n}{H}-\frac{u}{\lambda'}\right)\nu,
\]
the quantity
\[
\int_{M_t} H\lambda' \, d\mu_t - 2n\int_{\Omega_t}\frac{\lambda'''}{\lambda}\,dN
\]
is nonincreasing, while both
\[
\int_{\Omega_t}\lambda' \, dN
\]
and \(|M_t|\) are monotone. Passing to the slice limit yields new inequalities in anti-de-Sitter Schwarzschild manifolds and, in hyperbolic space, a weighted isoperimetric-type inequality [1708.06125].

In two dimensions, the warped-space flow
\[
\partial_t x=\left(\frac{\vartheta'(u)}{F}-s\right)\nu
\]
with the special choice
\[
F=\frac{H_2}{H_1}
\]
has two decisive monotone quantities. Along the flow, area is nondecreasing and the generalized \(W_2\)-functional
\[
\int_M H_1+\frac1n\int_{\hat M}\overline{\mathrm{Rc}(\partial_r,\partial_r)
\]
is nonincreasing. In dimension \(2\), this yields Minkowski-type inequalities in non-space-form warped products, and equality identifies radial slices [2005.11236].

Hyperbolic-space locally constrained inverse curvature flows realize the Alexandrov–Fenchel mechanism directly. For
\[
\eta=\lambda'\frac{E_{m-1}}{E_m}-u,
\]
one has
\[
\frac{d}{dt}W_m(\Omega_t)=0,\qquad \frac{d}{dt}W_{m+1}(\Omega_t)\le 0,
\]
and for the shifted-curvature flow one similarly preserves \(\widetilde W_m\) and decreases \(\widetilde W_{m+1}\). This produces sharp inequalities involving Gauss–Bonnet curvatures, weighted integrals of \(E_k\), and shifted mean curvatures, with equality only for geodesic spheres [2002.10643].

The de Sitter inverse Hessian quotient flow makes the same strategy Lorentzian. For
\[
\partial_t x=\left(u-\frac{\lambda' E_1}{E_2}\right)\nu,
\]
the first variation formula for the de Sitter quermassintegrals gives
\[
\frac{d}{dt}\mathcal A_1=0,\qquad \frac{d}{dt}\mathcal A_2\ge 0.
\]
Since the flow converges to a coordinate slice, one obtains
\[
\mathcal A_2(\Sigma)\le \varphi_2\circ \varphi_1^{-1}(\mathcal A_1(\Sigma)),
\]
with equality if and only if \(\Sigma\) is a coordinate slice [2507.11149]. A later de Sitter work proves, under the additional assumption of a Heintze–Karcher inequality for closed spacelike mean convex hypersurfaces, that
\[
\mathcal B_k(\Sigma)\le \varphi_k\circ \varphi_{-1}^{-1}(\mathcal B_{-1}(\Sigma)),
\]
again with slice rigidity [2512.19285].

The same local-constraint philosophy also yields sharp Michael–Simon type inequalities in hyperbolic space. For \(k\ge 2\), the imported flow
\[
\partial_t X=\left(\frac{E_{k-2}}{E_{k-1}}-\frac{u}{\lambda'}\right)\nu
\]
is combined with a weighted integral of \(E_{k-1}\) and weighted volume monotonicity to prove the sharp inequality of Theorem 1.17 for starshaped, strictly \(k\)-convex hypersurfaces [2205.12582]. In dimension one, the curve flow
\[
\partial_t X=\left(\frac{\phi'(r)}{\kappa}-u\right)\nu
\]
preserves length exactly,
\[
L'(t)=0,
\]
increases enclosed area, and yields the isoperimetric inequality together with weighted inequalities in all three simply connected two-dimensional space forms [2103.04338].

## 6. Scope, limitations, and adjacent frameworks

Several technical limitations are explicit in the current literature. For the de Sitter inverse Hessian quotient flow with \(F=E_2/E_1\), full curvature control is obtained only in the \(k=2\) case, and it is explicitly stated that analogous curvature estimates are not known for \(k\ge 3\) for that flow [2507.11149]. The broader de Sitter Alexandrov–Fenchel result based on
\[
\partial_t x=\left(\frac{u}{\lambda'}-\frac{E_{k-1}}{E_k}\right)\nu
\]
is conditional on an assumed Heintze–Karcher inequality in de Sitter space, which the paper does not prove [2512.19285].

There is also an important terminological boundary. Some papers develop locally constrained flows that are not inverse, such as the spherical flow
\[
\partial_t X=(\phi'-u p_2)\nu,
\]
while others study inverse curvature flows without any local constraint, such as the cocompact Minkowski-space flow
\[
\partial_t X=-\frac1F\nu.
\]
These belong to the same broader curvature-flow ecosystem but are analytically distinct [2403.07281] [2606.14082].

A further adjacent development is weak inverse mean curvature flow with an outer obstacle. There the local geometric restriction is not a support-function correction in the speed, but a variational obstacle condition enforcing tangential sticking to the boundary and encoded by the calibration trace condition
\[
[\nu\cdot \nu_\Omega]=1.
\]
This is not a locally constrained inverse curvature flow in the Brendle–Guan–Li or warped-product sense, but it shows that local constraints can also be formulated at the weak level through obstacle geometry rather than through a lower-order term in the normal velocity [2405.15181].

Taken together, these results suggest a coherent picture. In the established examples, locally constrained inverse curvature flow is most effective when three ingredients coincide: a warped-product ambient geometry with a conformal radial field, a support-function identity yielding a Minkowski formula, and an inverse curvature operator whose homogeneity and cone structure support maximum-principle estimates. Where these ingredients align, the flow can be tuned to preserve one quermassintegral-type quantity, improve another, and drive the hypersurface to a model slice or sphere; where one of them fails, the theory either becomes conditional or shifts to adjacent normalization frameworks rather than genuine local constraint.

Source: https://www.emergentmind.com/topics/locally-constrained-inverse-curvature-flow