---
title: Cutoff Level-Set Mean Curvature G-Equation
url: https://www.emergentmind.com/topics/cutoff-level-set-mean-curvature-g-equation
type: topic
---

# Cutoff Level-Set Mean Curvature G-Equation

Searching arXiv for recent and foundational papers on the cutoff level-set mean curvature G-equation and related homogenization/level-set curvature analysis.
The cutoff level-set mean curvature G-equation is a level-set model for flame front propagation in turbulent combustion in which the front is represented as the zero level set of a scalar function $G(x,t)$, with burnt region $\{G<0\}$ and unburnt region $\{G>0\}$. In its curvature-corrected form, the equation combines advection by an ambient incompressible flow, normal propagation at laminar flame speed, and a mean-curvature correction modulated by the Markstein number $d>0$. The cutoff is imposed through a positive-part operator to prevent a non-physical negative laminar flame speed, yielding the model
$$
G_t+\left(1-d\,\mathrm{div}\!\left(\frac{DG}{|DG|}\right)\right)_+|DG|+V(x)\cdot DG=0.
$$
For two-dimensional periodic incompressible flows, especially cellular flows, the principal analytical question is whether large-time front motion admits a direction-dependent effective burning velocity $\overline H(p)$, equivalently an effective Hamiltonian, so that solutions with planar initial data asymptotically propagate like planar fronts with uniform speed [2209.09228]. The equation also sits within the broader viscosity-solution and level-set framework for motion by mean curvature, where the geometric operator $|DG|\,\mathrm{div}(DG/|DG|)$ is the standard mean-curvature flow operator written in level-set form [1607.02069].

## 1. Model, geometric meaning, and cutoff mechanism

The inviscid G-equation is the Hamilton–Jacobi model
$$
G_t+V(x)\cdot DG+s_L|DG|=0,
$$
where the front normal speed is the laminar speed plus the normal component of the ambient flow. Incorporating Markstein curvature effect replaces the constant laminar speed by Markstein’s relation
$$
s_\ell=s_L^0(1-d\kappa)_+,
$$
with curvature
$$
\kappa=\mathrm{div}\!\left(\frac{DG}{|DG|}\right),
$$
and, after normalizing $s_L^0=1$, produces the cutoff mean-curvature G-equation [2209.09228].

The positive-part operator $(\cdot)_+$ is essential both physically and analytically. Physically, it ensures no “unburning,” because the laminar speed cannot become negative when curvature is large and positive. Analytically, it makes the operator more nonlinear and contributes to discontinuity at $DG=0$, which is one source of difficulty in homogenization and comparison arguments [2209.09228]. Without cutoff, the term $(1-d\kappa)|DG|$ can become negative when $\kappa$ is large and positive, which is regarded as unphysical in the combustion interpretation [2209.09228, 2303.16304].

The curvature contribution is the same geometric operator that appears in the level-set formulation of motion by mean curvature. If a level-set function $\phi$ represents a moving hypersurface, then
$$
\phi_t=|\nabla\phi|\,\mathrm{div}\!\left(\frac{\nabla\phi}{|\nabla\phi|}\right),
$$
and the operator can be written as
$$
|\nabla\phi|\,\mathrm{div}\!\left(\frac{\nabla\phi}{|\nabla\phi|}\right)
=\mathrm{tr}\!\big((I-n\otimes n)D^2\phi\big),\qquad n=\frac{\nabla\phi}{|\nabla\phi|},
$$
which makes its degenerate elliptic structure explicit [1607.02069]. In the G-equation setting, the curvature term tends to smooth the front and reduce reaction surface area, generally slowing propagation compared to the inviscid model [2209.09228].

## 2. Cellular-flow setting and effective burning velocity

A central setting is the two-dimensional incompressible periodic cellular flow with Hamiltonian
$$
H(x)=\sin x_1\sin x_2
$$
and amplitude $A>0$, giving the divergence-free velocity field
$$
V(x)=A(DH)^\perp
=A\big(-\cos x_2\sin x_1,\;\cos x_1\sin x_2\big).
$$
Its streamlines are the level curves of $H$, and the cell $Q=[0,\pi]\times[0,\pi]$ contains a single cell [2209.09228].

For any unit vector $p\in\mathbb R^2$, the main homogenization-type result states that there exists a positive number $\overline H_A(p)$ such that the solution of
$$
G_t+\left(1-d\,\mathrm{div}\!\left(\frac{DG}{|DG|}\right)\right)_+|DG|+V(x)\cdot DG=0,\qquad G(x,0)=p\cdot x,
$$
satisfies
$$
|G(x,t)-p\cdot x+\overline H_A(p)t|\le C
\qquad\text{in }\mathbb R^2\times[0,\infty),
$$
for a constant $C$ depending only on the Markstein number $d$ and the cellular flow amplitude $A$ [2209.09228]. In the rescaled thin-reaction-zone regime, with $V(x)\mapsto V(x/\epsilon)$ and $d\mapsto d\epsilon$, the corresponding bound becomes
$$
|G_\epsilon(x,t)-p\cdot x+\overline H_A(p)t|\le C\epsilon
\qquad\text{for all }(x,t)\in\mathbb R^2\times[0,\infty).
$$
The quantity $\overline H_A(p)$ is the effective burning velocity, also described as the turbulent flame speed or ergodic constant [2209.09228].

The effective Hamiltonian admits a stationary cell-problem characterization. One seeks a periodic corrector $w$ solving
$$
\left(1-d\,\mathrm{div}\!\left(\frac{p+Dw}{|p+Dw|}\right)\right)_+|p+Dw|
+V(x)\cdot(p+Dw)=\overline H(p),
$$
in the viscosity sense [2209.09228]. The authors construct correctors by ergodic penalization: for $\lambda>0$, solve
$$
\lambda v_\lambda(x)+
\left(1-d\,\mathrm{div}\!\left(\frac{p+Dv_\lambda}{|p+Dv_\lambda|}\right)\right)_+|p+Dv_\lambda|
+V(x)\cdot(p+Dv_\lambda)=0,
$$
and show
$$
\lambda v_\lambda\to-\overline H(p)
$$
uniformly as $\lambda\to0$ [2209.09228].

The resulting effective Hamiltonian is strictly positive for $p\neq0$, positively homogeneous of degree one, and continuous in $p$:
$$
\overline H(\lambda p)=\lambda\,\overline H(p),\qquad
\overline H(p)>0\;(p\neq0),\qquad
\overline H\in C(\mathbb R^2).
$$
In the inviscid case $d=0$, the cellular-flow benchmark due to Xin–Yu gives the sharp law
$$
\frac{A\pi(|p_1|+|p_2|)}{2\log A+C_2}\le \overline H_A(p)\le
\frac{A\pi(|p_1|+|p_2|)}{2\log A+C_1},
$$
while with curvature the same $O(A/\log A)$ scaling is conjectured [2209.09228].

## 3. Viscosity solutions, homogenization difficulties, and correctors

The cutoff level-set mean curvature G-equation is treated in the viscosity-solution framework. This is consistent with the general theory for level-set mean-curvature equations, where the operator is degenerate parabolic, undefined at $\nabla\phi=0$ in classical terms, and nevertheless admits continuous viscosity solutions with uniqueness via comparison principles [1607.02069]. In the cutoff G-equation, the relevant difficulties are sharper because the Hamiltonian is non-coercive in the gradient and the mean-curvature operator is discontinuous at $DG=0$ [2209.09228].

The non-coercivity comes from the linear advection term $V\cdot DG$, while degeneracy arises because the mean-curvature term vanishes at $DG=0$ and is discontinuous as an operator there [2209.09228]. Consequently, the standard perturbed test function method is not directly applicable. The paper therefore establishes existence of ergodic constants and sub- and supersolution correctors through PDE comparison combined with game-theoretic dynamics, rather than through the standard convex/coercive Hamilton–Jacobi homogenization route [2209.09228].

A key step is a negative linear drift estimate of the form
$$
G(x,t)-p\cdot x\le -\gamma t+C,
$$
which is then inserted into the Alvarez–Bardi penalization scheme for the stationary $\lambda$-equation. For small $\lambda$, one obtains
$$
\max_x \lambda v_\lambda(x)\le -\gamma/2,
$$
so that $v_\lambda$ becomes a viscosity supersolution of a stationary equation with strictly positive right-hand side [2209.09228]. A minimum value principle can then be applied to control minima on suitable regions. Combined with reachability and flow invariance, this yields uniform oscillation bounds on $u_\lambda=p\cdot x+v_\lambda$, which in turn imply uniform convergence of $\lambda v_\lambda$ to $-\overline H(p)$ [2209.09228].

This analytical structure places the cutoff curvature G-equation within the modern PDE theory of front propagation, while also showing why it is materially harder than the inviscid G-equation. For incompressible periodic flows, homogenization of the inviscid G-equation is well established, but the cutoff curvature model requires additional structure, especially in exploiting the underlying flow geometry [2209.09228].

## 4. Deterministic game representation and streamline geometry

The proof in two-dimensional cellular flow relies on a deterministic game characterization in the sense of Kohn–Serfaty. For time step $\tau>0$ and $N$ steps, the game trajectory $\{x_n\}$ evolves by
$$
x_{n+1}=x_n+\tau\sqrt{2d}\,b_n\eta_n+\tau^2\eta_n^\perp-\tau^2V(x_n),
$$
where $\eta_n\in\overline B_1(0)$ is chosen first by Player I, $b_n\in\{-1,1\}$ is then chosen by Player II, and $\eta^\perp=(-\eta_2,\eta_1)$ [2209.09228]. The value function satisfies the dynamic programming principle
$$
u_\tau(x,k\tau^2)=\inf_{|\eta|\le1}\max_{b=\pm1}
u_\tau\!\left(x+\tau\sqrt{2d}\,b\eta+\tau^2\eta^\perp-\tau^2V(x),(k-1)\tau^2\right),
$$
and as $\tau\to0$ with $N\tau^2\to t$, one has local uniform convergence to the viscosity solution of the cutoff curvature G-equation [2209.09228].

Allowing the choice $\eta=0$ encodes the positive-part cutoff. This is the mechanism by which Player I can “not burn” rather than “unburn,” and it is an essential link between the game and the cutoff operator [2209.09228]. The continuum consistency calculation is expressed through the semicontinuous envelopes $\underline F,\overline F$ of the curvature Hamiltonian [2209.09228].

The geometry of cellular streamlines is then used to design effective reachability strategies. For $\mu\in(0,1)$, the sets
$$
Q_\mu=\{x\in Q:\,H(x)>\mu\}
$$
are invariant under the reverse flow $\dot\xi=-V(\xi)$ [2209.09228]. The level sets $\{H=\mu\}$ are convex inside cells, boundary strips near cell edges can be driven into interior invariant sets, and transitions between adjacent cells can be performed in uniformly bounded time [2209.09228]. Specifically, the paper proves that from any point $P_1\in Q_\mu$, the curve $\{H=\mu\}$ is reachable in time $\le 3\sqrt2$, and any point on $\{H=\mu\}$ can be reached by flowing along the curve in time $O(1+|\log\mu|)$ [2209.09228]. It also proves uniform-time passage from boundary strips $\Gamma_\mu$ into $Q_{2\mu_0}$, and uniform-time transitions between neighboring cells [2209.09228].

These reachability results are combined with a minimum value principle for stationary problems to produce the oscillation bounds needed for the ergodic limit [2209.09228]. An auxiliary construction in the appendix uses an expanding or shrinking ellipse whose boundary speed is strictly less than the PDE’s curvature-plus-advection speed in a strip near a vertical segment with $V\cdot(1,0)=0$, ensuring the front crosses that segment within a fixed time [2209.09228].

## 5. Comparisons with related curvature and level-set theories

The cutoff curvature G-equation is closely related to, but distinct from, the pure level-set mean-curvature flow studied in geometric PDE. In the latter, mean-convex arrival-time solutions satisfy the elliptic equation
$$
-1=|\nabla u|\,\mathrm{div}\!\left(\frac{\nabla u}{|\nabla u|}\right),
$$
and there are strong regularity results: in the mean-convex case, the arrival time is twice differentiable everywhere, smooth away from the critical set, and solves the PDE classically everywhere [1607.02069]. Those results are specific to the curvature-only setting and rely on geometric properties of mean-curvature flow [1607.02069].

For the flame-front equation, the curvature term appears with advection and with cutoff, so the sharp regularity theory of mean-curvature flow does not directly transfer. A plausible implication is that the geometric insight from arrival-time regularity is structurally relevant, but the cutoff G-equation requires separate analysis because advection and the positive-part operator fundamentally alter the stationary and large-time problems [1607.02069, 2209.09228].

There is also a numerical comparison point. In level-set computation, curvature is typically evaluated from
$$
\kappa=\nabla\cdot\left(\frac{\nabla\phi}{\|\nabla\phi\|}\right),
\qquad
n=\frac{\nabla\phi}{\|\nabla\phi\|},
$$
but standard central differences fail near “kinks,” namely regions equidistant to two or more interfaces where $\nabla\phi$ is discontinuous [1409.6555]. The paper on curvature calculations shows that such kinks produce spurious oscillations, large curvature spikes of order $O(1/\Delta x)$, and incorrect curvature sign, and develops improved discretization strategies including directional differences for normals, Macklin–Lowengrub’s method, Lervåg’s method, and Salac–Lu’s method [1409.6555]. This suggests that numerical treatment of the cutoff curvature G-equation is inseparable from robust curvature evaluation when multiple fronts or equal-distance ridges are present.

A later three-dimensional study proposes data-driven mean-curvature correction for level-set methods, using feedforward networks that ingest transformed level-set, gradient, and curvature data on a $3\times3\times3$ stencil, together with Gaussian-curvature-based stencil classification [2208.09047]. That work is formulated for mean-curvature computation rather than homogenization theory, but it demonstrates that under-resolved curvature estimation can be improved substantially relative to baseline finite differences [2208.09047]. A plausible implication is that any computational study of the cutoff G-equation in three dimensions must account for the numerical fragility of curvature computation independently of the analytical nonexistence issues.

## 6. Extensions, bifurcation phenomena, and large-time dynamics

The two-dimensional existence theory extends beyond the basic cellular flow. The same approach is described as adaptable to broader two-dimensional periodic incompressible flows
$$
V=(-H_{x_2},H_{x_1}),
$$
including non-convex cells and “cat’s-eye” dynamics, under reasonable geometric conditions such as nondegenerate critical points [2209.09228]. The extension again depends on geometric reachability across convex or concave boundaries and between islands and unbounded regions [2209.09228].

In dimensions three and higher, however, the existence of effective burning velocity can fail. For shear flows of the form
$$
V(x)=(0,\dots,0,A f(x')),
$$
the paper on bifurcation proves that in dimension $n+1\ge3$ there exists a finite threshold $A_0(P)$ such that the set of intensities admitting effective burning velocity is precisely
$$
S_H(P)=[0,A_0(P)].
$$
Moreover, there is a saturation threshold $A_1(P)\in(0,A_0]$ such that
$$
\overline H_+(P,d,A)=A\,F(P)\qquad\text{for all }A\in[A_1,A_0],
$$
while no effective burning velocity exists for $A>A_0(P)$ [2303.16304]. In physical three-dimensional shear flow, corresponding to $n=2$, the result sharpens to
$$
A_0(P)=A_1(P),
$$
so the saturation threshold and the breakdown threshold coincide [2303.16304].

The cutoff and non-cutoff effective velocities are related by
$$
\overline H_+(P,d,A)=\max\{\overline H(P,d,A),\,A F(P)\},
$$
whenever the cutoff effective constant exists [2303.16304]. The mechanism of nonhomogenization is attributed to cutoff degeneracy: the local laminar speed
$$
s_\ell=(1-d\kappa)_+
$$
can vanish at isolated points where curvature is large, creating localized stagnation sets in the cell problem and obstructing periodic stationary solutions [2303.16304]. In three-dimensional shear, the characterization of the bifurcation point is linked to regularity theory for two-dimensional minimal-surface type equations, specifically through Simon’s Harnack inequality and a removability analysis for isolated singularities [2303.16304].

A different extension introduces a non-negative source term and studies large-time behavior for
$$
u_t+\left(-a^{ij}(Du)(D^2u)_{ij}+|Du|\right)_+ + \vec W(x)\cdot Du=f(x)
$$
on the torus, where
$$
a^{ij}(p)=\delta_{ij}-\frac{p_i p_j}{|p|^2}\qquad (p\neq0).
$$
Under the assumptions
$$
A:=\{x\in T^n:\,f(x)=0\}\subseteq\{x\in T^n:\,\vec W(x)=\vec0\},
\qquad A\neq\emptyset,
$$
the unique viscosity solution converges uniformly as $t\to\infty$ to a viscosity solution of the stationary problem [2509.16341]. The set $A$ serves simultaneously as a monotonicity set and a uniqueness set, and the proof uses half-relaxed limits plus a comparison principle on $A$ [2509.16341]. In the radially symmetric case with $\vec W\equiv0$, the reduced Hamiltonian
$$
H(r,p)=\left(-\frac{n-1}{r}p+|p|\right)_+-F(r)
$$
is convex in $p$, allowing an optimal-control representation formula for the solution and its large-time limit [2509.16341]. This source-term model is not the same homogenization problem as the cellular-flow setting, but it shows that monotonicity induced by a non-negative source can restore asymptotic convergence in situations where standard Hamilton–Jacobi large-time theory is unavailable [2509.16341].

## 7. Parameter dependence, misconceptions, and current directions

Several qualitative dependencies are established or conjectured. The effective burning velocity is strictly positive for nonzero propagation direction, positively homogeneous, and continuous in direction $p$ [2209.09228]. In shear flows, it decreases strictly with the Markstein number $d$, consistent with the physical smoothing effect of curvature [2209.09228]. If the curvature term is replaced by Laplacian diffusion, then turbulent flame speeds are dramatically slowed down: in two-dimensional cellular flows, $\overline H_A(p)$ remains uniformly bounded in $A$ [2209.09228]. Numerical evidence also indicates that the cutoff is active for moderate intensities, for example $A\in(0,45]$ and $d=0.1$, affecting $\overline H$ relative to the non-cutoff model [2209.09228].

One common misconception is that the cutoff is merely a technical regularization. In the cited work it is instead a physical constraint that prevents negative laminar flame speed and thereby excludes “unburning” [2209.09228, 2303.16304]. Another misconception is that curvature correction is interchangeable with viscous diffusion. The available results distinguish these effects sharply: curvature slows fronts while preserving geometric dependence on level sets, whereas replacing curvature by Laplacian diffusion changes the asymptotic speed law qualitatively, including boundedness in $A$ for cellular flows [2209.09228].

A further misconception is that homogenization should persist in higher dimensions whenever it holds in two dimensions. The shear-flow bifurcation result shows otherwise: in dimensions three and higher, effective burning velocity may cease to exist once advection exceeds a threshold, even though the non-cutoff problem homogenizes for all intensities [2303.16304]. This identifies the cutoff itself, rather than curvature alone, as a mechanism that can stall effective propagation in certain geometries [2303.16304].

Current directions follow directly from the cited results. In two-dimensional cellular flow, the precise asymptotic scaling of $\overline H_A(p)$ with curvature remains open; the paper conjectures the same $O(A/\log A)$ growth as in the inviscid case and asks whether curvature significantly reduces the prefactor [2209.09228]. In broader two-dimensional flows, the geometry of anisotropy and directional dependence is described as future work [2209.09228]. In three dimensions, an open problem is whether more mixing-dominated flows, such as swirled or chaotic incompressible periodic flows, may still admit effective burning velocities for all intensities despite the shear-flow bifurcation [2303.16304]. Together, these questions define the present research frontier of the cutoff level-set mean curvature G-equation.

Source: https://www.emergentmind.com/topics/cutoff-level-set-mean-curvature-g-equation