Modulus of Continuity Methods

• The method of modulus of continuity is a quantitative regularity tool that bounds function oscillations using auxiliary functions to control and estimate behavior in diverse PDE and stochastic frameworks.
• It employs proof templates such as two-point comparison, oscillation decay, and multiscale induction, yielding results from Hölder to logarithmic continuity.
• The framework is versatile, with applications in elliptic, parabolic, transport, and complex-geometric problems, revealing the optimal gauges or limitations of naïve continuity transfer.

Searching arXiv for recent and foundational papers on modulus of continuity methods across PDE, dynamics, probability, and analysis. The method of modulus of continuity is a quantitative regularity technique in which oscillation is encoded by a function such as $\omega$, $\varphi$, $\mu$, or $\theta$, and the main analytic task is to prove that this function is preserved, improved, or explicitly estimated under an equation, operator, stochastic limit, or dynamical iteration. Depending on context, the modulus may be prescribed, optimal, minimal, or constructed a posteriori. Representative formulations include the optimal two-point modulus

$$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$

the periodic increment modulus

$$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$

and general subadditive moduli $\theta$ for random fields (Li, 2015). Across the literature, the method appears as a one-dimensional comparison principle, a De Giorgi-type oscillation reduction, a stability estimate, a transport-distortion argument, or a multiscale iteration, with outputs ranging from Hölder continuity to logarithmic and iterated-logarithmic moduli (Xu, 2022).

1. Definitions and functional roles of a modulus

A modulus of continuity is not defined identically in every subfield, but the common content is a quantitative bound on increments. For viscosity solutions, a function $f:[0,\infty)\to\mathbb R_+$ is called a modulus of continuity if

$|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$

and the optimal modulus is defined by a two-point supremum (Li, 2015). In harmonic analysis of sparse Fourier series, the modulus is

$$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$

which is used to compare frequency growth and coefficient decay (Formica et al., 2021). For martingale random fields, the admissible class consists of continuous, increasing, subadditive functions $\varphi$0 with $\varphi$1, and regularity is encoded by boundedness of

$\varphi$2

This makes the modulus a seminormed continuity scale rather than only a qualitative property (Miftakhov, 2020).

Several papers distinguish between a merely admissible modulus and a canonical one. The minimal modulus of a continuous function $\varphi$3 on an interval is

$\varphi$4

and satisfies $\varphi$5 for every modulus $\varphi$6 of $\varphi$7 (Breneis, 2020). In KAM theory, the space $\varphi$8 replaces $\varphi$9 by requiring top derivatives to be controlled by a general modulus $\mu$0, with structural assumptions such as semi separability and weak homogeneity (Tong et al., 2023). In fractional Orlicz-Sobolev theory, the target is a space $\mu$1 with seminorm

$\mu$2

and the problem is to identify the optimal $\mu$3 produced by the Orlicz growth (Alberico et al., 2024).

These formulations show that the method is not tied to a single regularity scale. Power laws, logarithmic moduli, iterated logarithms, Dini-type gauges, and moduli defined through inverse Young functions all occur naturally. This suggests that the method is best understood as a scale-sensitive framework for oscillation control rather than as a synonym for Hölder estimates.

2. Recurring proof templates

Across the cited works, a small number of proof patterns recur.

Template	Core device	Representative papers
Two-point comparison	Auxiliary function in $\mu$4 and a 1D comparison equation	(Li, 2015, Andrews et al., 25 Jul 2025)
Oscillation decay	De Giorgi iteration, intrinsic cylinders, or section iteration	(Baroni et al., 2014, Liao, 2024, Cheng et al., 2020)
Stability-to-modulus	Control $\mu$5 by an $\mu$6 or Orlicz perturbation, then regularize	(Liu, 26 Oct 2025)
Transport distortion	Follow the modulus along flows with log-Lipschitz or Osgood control	(Jeon, 2023, Clop et al., 2017)
Multiscale induction	Avalanche Principle, dyadic chaining, or branching estimates	(Duarte et al., 2014, Hammond, 2017, Decrouez et al., 2012)

For local quasilinear and geometric flows, the standard device is the doubled-variable function

$\mu$7

combined with the Crandall–Ishii–Lions parabolic maximum principle for semicontinuous functions. The resulting jets imply that the modulus of continuity is a viscosity subsolution of a one-dimensional equation such as

$\mu$8

or, for level set mean curvature flow,

$\mu$9

(Li, 2015). In the non-local heat setting, second-derivative tests are replaced by reflection coupling under a radial kernel, and the odd extension $\theta$0 of the modulus must solve a one-dimensional non-local heat equation with reduced kernel $\theta$1 (Andrews et al., 25 Jul 2025).

For degenerate, singular, or nonlocal PDE, the method is typically an oscillation-reduction scheme. The $\theta$2-degenerate two-phase Stefan problem uses intrinsic cylinders, a two-alternative argument, Caccioppoli inequalities, a logarithmic forward-in-time lemma, and weak Harnack estimates to close an explicit logarithmic modulus (Baroni et al., 2014). Nonlocal parabolic equations with measurable kernels use intrinsic cylinders, De Giorgi lemmas, expansion of positivity, and a time-dependent truncation level $\theta$3 that absorbs the far-field contribution measured by the tail (Liao, 2024). Interior Monge–Ampère estimates are organized by sections and affine normalization, with a sequence of normalized sections playing the role of scales (Cheng et al., 2020).

A third template begins from a stability estimate and converts it into continuity. On compact Hermitian manifolds, a smoothing operator $\theta$4 built from the holomorphic part of the exponential map is combined with a stability estimate of the form

$\theta$5

and the bound on $\theta$6 is then upgraded to pointwise log-continuity (Liu, 26 Oct 2025).

3. Elliptic, parabolic, and complex-geometric PDE implementations

In planar singular elliptic theory, the method yields an explicit logarithmic modulus for weak solutions of

$\theta$7

under the assumptions $\theta$8 bounded Lipschitz, $\theta$9 for some $$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$0, and $$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$1. The proof combines a dyadic decomposition of $$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$2, logarithmic energy estimates, boundedness of all logarithmic moments of

$$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$3

and a splitting $$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$4 into homogeneous and forced parts. The resulting estimate is

$$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$5

with arbitrary $$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$6, in sharp contrast with the degenerate planar result of Onninen–Zhong, where the logarithmic exponent must be suitably small (Xu, 2022).

For the $$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$7-degenerate two-phase Stefan problem, the quantitative interior modulus is

$$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$8

derived by a refined intrinsic oscillation-decay argument on cylinders

$$\omega(s,t)=\sup\left\{\frac{u(y,t)-u(x,t)}{2}\,\middle|\, \frac{|y-x|}{2}=s\right\},$$9

The authors state that this discards one logarithm iteration and obtains an explicit value for the exponent $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$0, even in the classical case $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$1 (Baroni et al., 2014).

For nonlocal parabolic equations of fractional $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$2-Laplacian type with measurable kernels, the method quantifies a general modulus of continuity for locally bounded local weak solutions under a minimal $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$3-in-time tail condition. The oscillation scale is updated by

$$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$4

and the final modulus contains both a geometric decay term and an explicit tail contribution. Under a stronger $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$5-in-time tail assumption, the same framework yields local Hölder continuity (Liao, 2024).

Complex-geometric equations furnish several distinct modulus-of-continuity mechanisms. For the Dirichlet problem for complex Hessian equations on a strongly $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$6-pseudoconvex domain, the solution satisfies the sharp estimate

$$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$7

obtained from barriers, a linearized characterization via operators $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$8, and the comparison principle (Charabati, 2014). For the Monge–Ampère Dirichlet problem

$$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$9

Cheng–O’Neill estimate $\theta$0 and $\theta$1 in terms of the modulus

$\theta$2

using sections, affine invariance, and iterative normalization. In the Hölder regime $\theta$3, the corollaries give nonlinear bounds

$\theta$4

confirming the nonlinear dependence found by Figalli, Jhaveri and Mooney (Cheng et al., 2020). On compact Hermitian manifolds, a pure PDE stability approach adapted from Guo–Phong–Tong–Wang and Guo–Guo yields

$\theta$5

for solutions of the global complex Monge–Ampère equation; the Kähler endpoint $\theta$6 is not claimed in the Hermitian case (Liu, 26 Oct 2025).

4. Transport equations, Euler dynamics, and propagation versus breakdown

For transport-type problems, the method tracks how a modulus deforms under the flow map. In the continuity equation

$\theta$7

well-posedness is proved under a growth condition

$\theta$8

with

$\theta$9

and an Osgood-type modulus-of-continuity condition

$f:[0,\infty)\to\mathbb R_+$0

The proof does not use divergence bounds; instead it uses optimal transport with a cost adapted to $f:[0,\infty)\to\mathbb R_+$1, splitting the solution into positive and negative parts and proving uniqueness by a transport-cost differential inequality (Clop et al., 2017).

For the 2D Euler equation, two complementary results delineate what can and cannot be propagated. In the loglog-vortex problem, the perturbation is measured in

$f:[0,\infty)\to\mathbb R_+$2

and the decomposition

$f:[0,\infty)\to\mathbb R_+$3

leads to a forced transport equation for $f:[0,\infty)\to\mathbb R_+$4. The forcing term matches the same modulus when $f:[0,\infty)\to\mathbb R_+$5, and the perturbation remains in $f:[0,\infty)\to\mathbb R_+$6 for all time; by contrast, for every $f:[0,\infty)\to\mathbb R_+$7 there exists a smooth compactly supported perturbation whose $f:[0,\infty)\to\mathbb R_+$8 norm blows up instantly (Jeon, 2023). In a different Euler setting, logarithmic moduli

$f:[0,\infty)\to\mathbb R_+$9

and also

$|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$0

are propagated by the Yudovich log-Lipschitz flow, but the paper also constructs a family of moduli that are not propagated. The negative theorem shows that the answer to the general question “Given a modulus of continuity for the 2D Euler equations, can we always propagate it?” is no (Khalil, 2024).

These results eliminate a common oversimplification. The method does not assert that “continuity is preserved” in a uniform sense; rather, it identifies which gauges are compatible with the distortion law of the flow. In Euler, compatibility is governed by log-Lipschitz transport; in the continuity equation, by an Osgood modulus.

5. Harmonic analysis, random fields, KPZ, KAM, and cocycles

Outside PDE regularity theory, the method frequently becomes a quantitative bridge between local oscillation and another analytic structure.

For superlacunar trigonometric series,

$|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$1

the paper establishes bilateral interrelations between Fourier coefficients and modulus of continuity. One direction is

$|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$2

and the reverse direction is encoded by

$|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$3

In the lacunar example $|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$4, the modulus is of logarithmic order $|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$5, while in the superlacunar example $|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$6 it becomes of iterated-logarithmic order $|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$7. The same framework is then applied to Gaussian periodic stationary processes, where the covariance may be very rough while the process itself remains continuous (Formica et al., 2021).

In Brownian last passage percolation, polymer weight profiles in scaled KPZ coordinates have a modulus of continuity of order

$|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$8

for a broad class of initial data. The proof combines a two-point estimate for narrow wedge profiles, Brownian Gibbs regularity, and a chaining argument. Weak limits of the profiles inherit the same modulus almost surely (Hammond, 2017). For Canonical Embedded Branching Processes, the almost sure modulus is

$|u(y)-u(x)| \le 2 f\!\left(\frac{|y-x|}{2}\right),$9

obtained by adapting the Barlow–Perkins strategy to a generally non-Markovian setting via crossing-tree estimates (Decrouez et al., 2012). For martingale sequences of random fields, uniform control of

$$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$0

and $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$1 allows construction of a modification of the pointwise limit that preserves the same modulus $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$2; the same mechanism extends to derivatives in the $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$3 setting (Miftakhov, 2020).

In abstract dynamics, the continuity modulus itself becomes the target of the theorem. For Lyapunov exponents of linear cocycles, large deviation type estimates, a deterministic Avalanche Principle, and a multiscale iteration yield a local modulus

$$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$4

so the sharpness of the continuity modulus is explicitly dictated by the decay of the exceptional-set function $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$5 (Duarte et al., 2014). In finite-smooth KAM theory, Jackson approximation is extended from Hölder scales to $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$6 spaces, the analytic approximation error becomes $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$7, and the final regularity of the invariant torus is described by moduli $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$8 derived from the convergence rate of the KAM scheme (Tong et al., 2023). In the uncentered Hardy–Littlewood maximal operator, the method yields exact or sharp integral formulas for the modulus of continuity of $$w[f](\delta)=\sup_{|h|<\delta}\sup_t |f(t+h)-f(t)|,$$9 in terms of that of $\varphi$00, including

$\varphi$01

on proper subintervals of $\varphi$02, and

$\varphi$03

on $\varphi$04 (Aldaz et al., 2010).

6. Sharpness, thresholds, and failures of naïve transfer

A recurring feature of the method is that it often identifies a borderline scale rather than merely proving continuity. Several papers make this explicit. In the planar singular elliptic equation with coefficient matrix $\varphi$05, the reciprocal logarithmic factor may have an arbitrarily large power, whereas the degenerate planar result of Onninen–Zhong requires a suitably small exponent (Xu, 2022). In the two-phase Stefan problem, the modulus

$\varphi$06

is conjectured to be optimal (Baroni et al., 2014). In the Hermitian complex Monge–Ampère equation, the theorem is stated for strict sub-endpoints $\varphi$07, while the Kähler case reaches the endpoint (Liu, 26 Oct 2025).

Equally important are the negative results. For bounded-variation functions, the variation function always satisfies

$\varphi$08

but the converse direction fails drastically: given any modulus $\varphi$09 weaker than Lipschitz and any bounded modulus $\varphi$10, there exists $\varphi$11 with $\varphi$12 and $\varphi$13. In particular, the open problem asking whether the variation function of an $\varphi$14-Hölder continuous function must be $\varphi$15-Hölder continuous is resolved negatively (Breneis, 2020). For 2D Euler, not every modulus is propagated (Khalil, 2024). For non-local heat equations on bounded domains, a one-dimensional regional theorem holds on an interval, but a counterexample with thin rectangles suggests that a non-local analogue of the Payne–Weinberger inequality in higher dimensions would depend on more than diameter alone (Andrews et al., 25 Jul 2025).

These examples clarify a central point. The method of modulus of continuity is not a universal monotonicity principle saying that any chosen modulus should survive an evolution or pass through an operator. Rather, it is a mechanism for identifying the correct continuity gauge produced by the geometry of the equation, the operator, or the stochastic structure. When the gauge is compatible with that structure, one gets preservation or sharp quantitative control; when it is not, the same method can reveal failure just as decisively as success.