Dafermos Regularization in Nonlinear PDEs

• Dafermos regularization is a method that employs self-similar viscous perturbations to ensure the uniqueness of weak solutions in nonlinear PDEs.
• It distinguishes admissible solutions by using viscous wave fan profiles and entropy rate criteria, crucial for capturing the correct Riemann solution.
• The approach underpins numerical schemes by selectively dissipating entropy, bridging theoretical analysis with practical computation in conservation laws.

In the cited literature, Dafermos regularization denotes a family of admissibility and approximation mechanisms associated with weak solutions of nonlinear PDEs. In the classical conservation-law setting, it is the self-similar viscous perturbation $U_t+F(U)_x=\varepsilon\,t\,U_{xx}$, designed so that Riemann solutions remain compatible with the similarity variable $\xi=x/t$. In a distinct but related line of work, especially in numerical analysis, it denotes Dafermos’ entropy rate criterion, according to which the selected weak solution should dissipate total entropy at least as fast as every competing weak solution. These two usages are connected by a common selection principle—viscous or variational—which is used to distinguish physically relevant weak solutions from other admissible distributions (Sourdis, 2023, Klein, 2022, Vorotnikov, 9 Jan 2025).

1. Self-similar viscous regularization of conservation laws

For the scalar conservation law

$U_t + f(U)_x = 0,$

with Riemann initial data

$$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$

Dafermos regularization replaces the equation by

$$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$

This is explicitly distinguished from the standard viscous regularization $U_t+f(U)_x=\varepsilon U_{xx}$: the factor $t$ makes the perturbation compatible with self-similar scaling. Under the ansatz $U(x,t)=u(\xi)$, $\xi=x/t$, one obtains the profile ODE

$$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$

with endstates

$\xi=x/t$0

Solutions of this two-point problem are the viscous wave fan profiles, that is, smooth self-similar internal layers representing the viscously spread-out analogue of the Riemann wave fan (Sourdis, 2023).

The same self-similar viscous structure also appears in system settings. In the chromatography model studied after a change of variables,

$\xi=x/t$1

the regularization is written in vector form as

$\xi=x/t$2

which again reduces, in $\xi=x/t$3, to

$\xi=x/t$4

For a Keyfitz–Kranzer-type $\xi=x/t$5 system, the cited analysis states that the Dafermos-style vanishing-viscosity/self-similar viewpoint is used implicitly: add a small viscosity, rescale into similarity variables, reduce to an ODE, and study the singular limit $\xi=x/t$6 (Tsikkou, 2015, Culver et al., 8 Aug 2025).

2. Viscous wave fan profiles and the scalar uniqueness theory

A central recent result is the uniqueness theory for scalar viscous wave fan profiles. For

$\xi=x/t$7

Sourdis proves that if $\xi=x/t$8 is Lipschitz on the interval between $\xi=x/t$9 and $U_t + f(U)_x = 0,$0 (or $U_t + f(U)_x = 0,$1 in the trivial equal-data case), then for every $U_t + f(U)_x = 0,$2 there exists a unique classical solution. The paper stresses that the argument is not restricted to the small self-similar viscosity regime; it applies for all $U_t + f(U)_x = 0,$3. A direct corollary is that there is at most one Riemann solution satisfying the viscous wave fan profile criterion (Sourdis, 2023).

The proof is based on qualitative comparison arguments adapted from elliptic PDE theory. For $U_t + f(U)_x = 0,$4, the profiles are strictly increasing, and the argument uses the sliding method with translations

$U_t + f(U)_x = 0,$5

For $U_t + f(U)_x = 0,$6, where profiles are decreasing, the paper instead uses Serrin’s sweeping principle with the modified family

$U_t + f(U)_x = 0,$7

where $U_t + f(U)_x = 0,$8 is a Lipschitz constant for $U_t + f(U)_x = 0,$9 on the relevant interval. In both cases, the shift is arranged so that the translated family becomes a strict supersolution, and a maximum-principle comparison rules out first contact except in the trivial case. The significance of the result lies precisely in its minimal hypotheses: monotonicity, Lipschitz control of $$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$0, and maximum-principle arguments suffice, without asymptotic perturbation theory and without a small-$$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$1 assumption (Sourdis, 2023).

The same paper also formulates a profile criterion for self-similar Riemann solutions. A self-similar Riemann solution $$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$2 satisfies the viscous wave fan profile criterion if and only if there exists a solution $$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$3 of the profile ODE such that

$$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$4

in the appropriate local sense stated there. This ties uniqueness of the ODE profile directly to uniqueness of the selected Riemann solution (Sourdis, 2023).

3. Burgers equation and the role of unbounded profiles

For Burgers’ flux

$$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$5

the viscous profile equation becomes

$$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$6

This case is used to display the subtlety of the uniqueness problem. The paper proves the existence of a unique increasing solution of

$$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$7

with $$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$8 satisfying the growth bound

$$U(x,0)= \begin{cases} U_L, & x<0,\ U_R, & x>0, \end{cases}$$9

for some constants $$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$0, and states that the decay rate on the $$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$1 side can be sharpened to $$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$2. This solution is unbounded, growing like $$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$3 as $$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$4, and is not itself a Riemann profile; its role is auxiliary but structurally important (Sourdis, 2023).

The Burgers profile ODE admits the first integral

$$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$5

which is used to sharpen the asymptotic analysis. The cited work then combines the uniqueness theorem with the auxiliary unbounded solution to obtain a precise profile-level description of Dafermos regularization for Burgers rarefaction waves. When $$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$6,

$$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$7

and the profile obeys the symmetry relation

$$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$8

The remainder is stated to be uniformly bounded in $$U_t + f(U)_x = \varepsilon\, t\, U_{xx},\qquad \varepsilon>0.$$9 as $U_t+f(U)_x=\varepsilon U_{xx}$0. This yields not merely convergence to the inviscid rarefaction wave, but a sharp description of the self-similar viscous profile and its exponentially small correction (Sourdis, 2023).

A common misconception is that Dafermos regularization always produces bounded viscous layers. The Burgers analysis shows that unbounded profile solutions can be essential even when the final Riemann profile is bounded. This suggests that auxiliary singular or unbounded orbits are part of the internal geometric structure of the regularized problem rather than accidental artifacts (Sourdis, 2023).

4. Singular shocks, delta shocks, and geometric singular perturbation

For systems, Dafermos regularization is closely tied to the emergence of nonclassical self-similar solutions. In the chromatography model, the paper states that in the relevant region no classical bounded Riemann solution exists, so one must seek unbounded viscous profiles as $U_t+f(U)_x=\varepsilon U_{xx}$1. The central theorem establishes existence of a singular shock connecting $U_t+f(U)_x=\varepsilon U_{xx}$2 and $U_t+f(U)_x=\varepsilon U_{xx}$3, realized as a self-similar viscous profile solving the Dafermos ODE and becoming unbounded as $U_t+f(U)_x=\varepsilon U_{xx}$4. The proof uses Geometric Singular Perturbation Theory, a blow-up transformation, normally hyperbolic invariant manifolds, and the Exchange Lemma and Corner Lemma to connect outer states through an inner singular layer (Tsikkou, 2015).

In that setting, the inner layer is narrower than a standard viscous shock profile: $U_t+f(U)_x=\varepsilon U_{xx}$5 and the limiting object satisfies generalized Rankine–Hugoniot relations with a positive deficit

$U_t+f(U)_x=\varepsilon U_{xx}$6

The cited interpretation is that the viscous profile concentrates mass into a spike whose amplitude diverges as $U_t+f(U)_x=\varepsilon U_{xx}$7, thereby producing a singular shock rather than an ordinary Lax shock (Tsikkou, 2015).

The Keyfitz–Kranzer-type $U_t+f(U)_x=\varepsilon U_{xx}$8 analysis develops an analogous selection picture in phase space. The paper classifies self-similar Riemann solutions into elementary pieces labeled $U_t+f(U)_x=\varepsilon U_{xx}$9, $t$0, $t$1, $t$2, Vacuum, and $t$3, and organizes the phase portrait by curves $t$4, $t$5, and $t$6. Its stated conclusion is that the Dafermos regularization mechanism selects which composite pattern survives in the vanishing-viscosity limit: shock plus contact, rarefaction plus contact, vacuum-mediated composites, or $t$7 when the right state lies below the limiting curve $t$8 (Culver et al., 8 Aug 2025).

The phase-curve relation is written there as

$t$9

with the singularity at $U(x,t)=u(\xi)$0 driving a blow-up-type analysis of the internal layer. The paper explicitly interprets this geometry as that of a geometric singular perturbation analysis: an outer flow on inviscid characteristic manifolds, an inner layer near the singular set, and a limiting composite wave obtained by matching. In this usage, Dafermos regularization does not merely smooth the PDE; it provides the selection mechanism by which classical waves, vacuum states, and overcompressive delta shocks arise as limiting self-similar solutions (Culver et al., 8 Aug 2025).

5. Entropy rate criterion as numerical regularization

A second major usage of the term appears in numerical analysis. Here Dafermos regularization means implementing Dafermos’ entropy rate criterion rather than inserting the viscous term $U(x,t)=u(\xi)$1. For a conservation law with convex entropy $U(x,t)=u(\xi)$2, total entropy is written as

$U(x,t)=u(\xi)$3

and the criterion is stated as

$U(x,t)=u(\xi)$4

for all other weak solutions $U(x,t)=u(\xi)$5. The numerical objective is then to construct schemes whose discrete evolution dissipates entropy at least as fast as the admissible lower bound predicted by this criterion (Klein, 2022, Klein, 2022).

In finite volume form, one approach is the convex combination

$U(x,t)=u(\xi)$6

where $U(x,t)=u(\xi)$7 is the Godunov flux and $U(x,t)=u(\xi)$8 is Tadmor’s entropy conservative flux. The parameter $U(x,t)=u(\xi)$9 acts as a local entropy-dissipation indicator: the cited construction aims to suppress dissipation in smooth regions and activate it near shocks, thereby approximating the entropy-rate selection rule rather than merely enforcing a discrete entropy inequality (Klein, 2022).

For discontinuous Galerkin methods, the stabilization is formulated as a correction to a baseline DG update. In the 2022 paper, the correction is posed as a constrained minimization of a discrete cell entropy $\xi=x/t$0, yielding semi-discrete DDG and fully discrete DRKDG variants. In the semi-discrete version, the corrected derivative is obtained from

$\xi=x/t$1

under conservation and error-budget constraints; in the fully discrete version, the new state is the minimizer of $\xi=x/t$2 over the corresponding admissible set. The paper characterizes this as a constrained entropy-minimizing correction that preserves conservation and limits the correction by an error estimate (Klein, 2022).

The later systems paper rewrites the corrected DG evolution as

$\xi=x/t$3

with $\xi=x/t$4 produced by a conservative entropy-dissipative filter generator $\xi=x/t$5, and $\xi=x/t$6 chosen so that the discrete entropy production does not exceed the bound predicted by a local entropy dissipation estimator. A key bound for Riemann-type data is the HLL-based predictor

$\xi=x/t$7

with

$\xi=x/t$8

The same paper states that the classical Lax–Friedrichs flux is the maximally entropy-dissipative consistent conservative three-point flux for systems of conservation laws, and uses this as a benchmark for local entropy-rate prediction (Klein, 2023).

On unstructured triangular grids, the stabilization is extended to multidimensional DG with a corrected semidiscrete system

$\xi=x/t$9

and the paper emphasizes that the resulting schemes are free of tunable viscosity parameters. Entropy dissipation is estimated face-by-face, including boundary contributions for coupling and reflecting boundaries, and the correction strength is determined from entropy accounting rather than from a hand-tuned artificial viscosity coefficient (Klein, 2024).

A related finite-volume line combines the entropy-rate criterion with optimal recovery. There, Dafermos regularization is implemented as a flux-selection regularization over an admissible state or flux set, for example

$$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$0

The admissible set is derived from reconstruction uncertainty, encoded through a radius of information, and the resulting scheme is described as a recovery-based alternative to ENO and WENO. The accompanying viscosity redistribution is introduced because the raw entropy-based viscosity can become unbounded when entropy-variable jumps are small (Klein et al., 2023).

6. Abstract generalizations and conceptual scope

The entropy-rate interpretation has also been extended well beyond hyperbolic conservation laws. In the abstract framework

$$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$1

posed on $$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$2, the 2025 paper identifies an entropy-like quantity

$$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$3

generated by a strictly convex $$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$4-function, leading naturally to an anisotropic Orlicz space. Using a Brenier-style dual matrix-valued variational formulation with time-adaptive weights $$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$5, the paper proves solvability of the dual problem and large-time consistency of the duality scheme (Vorotnikov, 9 Jan 2025).

In that setting, the author formulates Dafermos’ principle as follows: no subsolution can dissipate total entropy earlier or faster than the strong solution on the interval where the strong solution exists. If $$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$6 is the entropy of the strong solution and $$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$7 the entropy of a subsolution, then for any $$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$8 it cannot happen simultaneously that

$$\varepsilon u''(\xi)=\bigl(f'(u(\xi))-\xi\bigr)u'(\xi),\qquad \xi\in\mathbb{R},$$9

and

$\xi=x/t$00

The paper states that this result is new even for “isotropic” problems such as the incompressible Euler system (Vorotnikov, 9 Jan 2025).

This broader usage clarifies an important terminological point. Dafermos regularization is not uniform across the literature. In PDE analysis of Riemann problems it usually means the self-similar viscous perturbation $\xi=x/t$01. In computational and variational work it often means a selection principle based on maximal entropy dissipation, implemented through optimization, flux design, filtering, or duality. The common content is not a single formula but the insistence that admissibility should be encoded through a quantitatively strongest allowable entropy decay (Sourdis, 2023, Klein, 2022, Vorotnikov, 9 Jan 2025).

7. Mathematical significance and recurring themes

Across these uses, several themes recur. First, Dafermos regularization is fundamentally a selection mechanism for nonunique weak solutions. In the scalar Riemann problem, uniqueness of viscous wave fan profiles implies uniqueness of the associated Riemann solution satisfying the profile criterion. In system problems with concentration, the same mechanism selects singular shocks, vacuum-mediated composites, or delta shocks when classical self-similar solutions are insufficient (Sourdis, 2023, Tsikkou, 2015, Culver et al., 8 Aug 2025).

Second, the approach is closely tied to self-similarity and internal-layer analysis. The factor $\xi=x/t$02 in $\xi=x/t$03 is not a minor modification of standard viscosity; it is the structural device that reduces the regularized PDE to an ODE in $\xi=x/t$04. This is what makes the framework especially natural for Riemann problems and for geometric singular perturbation analyses of shock layers and concentration profiles (Sourdis, 2023, Tsikkou, 2015).

Third, the entropy-rate reinterpretation shows that the same admissibility philosophy can be turned into a computational design principle. Rather than adding arbitrary dissipation, the cited DG and finite-volume schemes estimate how much entropy dissipation is admissible and add only the correction needed to realize that budget. This suggests a conceptual bridge between vanishing-viscosity selection, entropy inequalities, and high-order shock capturing, although the underlying mechanisms differ substantially from paper to paper (Klein, 2022, Klein, 2023, Klein, 2024).

Finally, the recent dispersive and duality-based work indicates that Dafermos’ principle is not confined to classical conservation-law viscosity. A plausible implication is that “Dafermos regularization” names a broader admissibility paradigm: identify a convex entropy functional, compare competing weak or subsolutions by the speed of entropy decay, and recover the physically relevant dynamics either by self-similar viscous approximation or by a variational selection principle (Vorotnikov, 9 Jan 2025).