---
title: a-Contraction with Shifts in Shock Stability
url: https://www.emergentmind.com/topics/a-contraction-with-shifts
type: topic
---

# a-Contraction with Shifts in Shock Stability

In the analysis of shock stability for systems of conservation laws, \(a\)-contraction with shifts denotes a weighted relative-entropy method in which an entropic Rankine–Hugoniot shock \((u_L,u_R,\sigma_{LR})\) is compared against a solution \(u(t,x)\) through a pseudo-distance that is both asymmetric and translation-modulated. For a weight \(a>0\) and a shift \(h(t)\), the central functional is
\[
E(t) = \int_{-\infty}^{h(t)} a\, \eta(u(t,x)\mid u_L)\,dx + \int_{h(t)}^{\infty} \eta(u(t,x)\mid u_R)\,dx,
\]
where \(\eta(u\mid v)\) is the relative entropy associated with a strictly convex entropy–entropy flux pair \((\eta,q)\). The goal is to choose \(h(t)\) so that \(E(t)\) is nonincreasing, thereby proving orbital \(L^2\) stability up to translation; recent work develops necessary and sufficient conditions for this mechanism in the small-shock regime for general systems, especially for interior characteristic families [2504.03920].

## 1. Hyperbolic–entropy formulation

The underlying PDE is the one-dimensional system of conservation laws
\[
u_t + (f(u))_x = 0,\quad t>0,\ x\in \mathbb{R},
\]
with \(u=(u_1,\dots,u_n):[0,\infty)\times\mathbb{R}\to \mathcal{V}\), \(\mathcal{V}\subset \mathbb{R}^n\) open and connected, and \(f\in C^4(\mathcal{V};\mathbb{R}^n)\). The flux Jacobian is \(A(u)=f'(u)\). The system is assumed to admit a convex entropy–entropy flux pair \((\eta,q)\) satisfying
\[
\eta'(u) f'(u) = q'(u),
\]
together with the entropy inequality
\[
\eta(u)_t + q(u)_x \le 0
\]
in the sense of distributions, and strict convexity \(\nabla^2 \eta(u)>0\) for all \(u\in \mathcal{V}\) [2504.03920].

For any fixed \(v\in \mathcal{V}\), the relative entropy and relative entropy flux are
\[
\eta(u\mid v) := \eta(u) - \eta(v) - \nabla\eta(v)\cdot (u-v),
\]
\[
q(u;v) := q(u) - q(v) - \nabla\eta(v)\cdot (f(u)-f(v)).
\]
The weighted pseudo-distance with shift,
\[
E(t) = \int_{-\infty}^{h(t)} a\, \eta(u(t,x)\mid u_L)\,dx + \int_{h(t)}^{\infty} \eta(u(t,x)\mid u_R)\,dx,
\]
is equivalent to a weighted \(L^2\) norm on compact subsets: for any compact \(V\subset\subset \mathcal{V}\) there exist \(0<c\le C\) such that
\[
c|u-v|^2 \le \eta(u\mid v)\le C|u-v|^2,\quad (u,v)\in \mathcal{V}\times V.
\]
The shift \(h(t)\) is chosen to follow a generalized characteristic in continuous regions or the shock path at discontinuities. Along a smooth path \(h(t)\), one has for almost all \(t\),
\[
f(u(t,h(t)+)) - f(u(t,h(t)-)) = \dot h(t)\big(u(t,h(t)+) - u(t,h(t)-)\big),
\]
and
\[
q(u(t,h(t)+)) - q(u(t,h(t)-)) \le \dot h(t)\big(\eta(u(t,h(t)+)) - \eta(u(t,h(t)-))\big).
\]
When no jump is present and \(h(t)\) is chosen as a generalized characteristic of the \(i\)-th family, \(\dot h(t)=\lambda_i(u(t,h(t)))\) [2504.03920].

A standard abbreviation is
\[
\tilde\eta(u) := a\,\eta(u\mid u_L) - \eta(u\mid u_R),\quad
\tilde q(u) := a\, q(u;u_L) - q(u;u_R),\quad
\Pi:=\{u:\tilde\eta(u)<0\}.
\]
With strong traces, the basic differential inequality is
\[
\frac{d}{dt} E(t) \le \dot h(t)\big(a\,\eta(u\mid u_L)-\eta(u\mid u_R)\big) - \big(a\, q(u;u_L) - q(u;u_R)\big).
\]
This is the analytic core of the method: the weight \(a\) compensates for the asymmetry of the two states, while the shift removes the neutral translational mode.

## 2. Dissipation mechanism and \(a\)-relative entropy stability

The method separates dissipation into a continuous contribution and a Rankine–Hugoniot contribution. When \(\dot h(t)=\lambda_i(u(t,h(t)))\), the continuous dissipation is
\[
D_{\mathrm{cont}}(u) := -\tilde q(u) + \lambda_i(u)\,\tilde\eta(u).
\]
When \(h(t)\) crosses a discontinuity and \(\dot h(t)=\sigma_\pm\), the discontinuous dissipation is
\[
D_{\mathrm{RH}}(u_\pm,\sigma_\pm) := \big(q(u_+;u_R) - \sigma_\pm \eta(u_+ \mid u_R)\big) - a\big(q(u_-;u_L) - \sigma_\pm \eta(u_- \mid u_L)\big).
\]
An entropic shock \((u_L,u_R,\sigma_{LR})\) is called \(a\)-relative entropy stable if the following hold:

\[
(\mathcal{H}1)\quad \text{For any }u\in\partial\Pi,\ D_{\mathrm{cont}}(u)\le 0.
\]

\[
(\mathcal{H}2)\quad \text{For any entropic shock }(u_-,u_+,\sigma_\pm)\text{ with }\tilde\eta(u_-)<0<\tilde\eta(u_+),\ D_{\mathrm{RH}}(u_\pm,\sigma_\pm)\le 0.
\]

A theorem attributed in the paper to Kang–Vasseur states that if, for any \(u\in \mathcal{S}_{\mathrm{weak}}\), there exists a Lipschitz \(h(t)\) such that \(E(t)\) is nonincreasing, then the shock is \(a\)-relative entropy stable, hence satisfies \((\mathcal{H}1)\)–\((\mathcal{H}2)\) [2504.03920].

In the small-shock local-attractor regime, the target estimate becomes quantitative:
\[
\frac{d}{dt} E(t) \le -C\, |u_-(t)-u_L|^2.
\]
This formulation is stronger than mere nonincrease of \(E(t)\): it identifies a coercive mechanism tied to the left trace \(u_-(t)\), and it is precisely this strengthened dissipation that recent work characterizes for general systems.

## 3. Necessary and sufficient conditions for small shocks

For small shocks of a genuinely nonlinear \(i\)-th family, the decisive criterion is a matrix inequality on the complement of the \(i\)-th eigendirection. Let \(u_R=S^i_{u_L}(s)\), with Rankine–Hugoniot speed \(\sigma(s)\), and let \(r_i(u_L)\) and \(\lambda_i(u_L)\) denote the right eigenvector and characteristic speed of \(f'(u_L)\). A sufficient condition is the existence of \(C\in\mathbb{R}\) such that
\[
-C\,\nabla^2 \eta(u_L)\big(f'(u_L)-\lambda_i(u_L)I\big) + r_i(u_L)^\top \nabla^2 \eta(u_L)\,f''(u_L)
\]
is negative definite on \(\mathrm{span}\{r_k(u_L)\}_{k\ne i}\). Under this hypothesis, there exists \(s_0>0\) such that for all \(0<s<s_0\) one can choose
\[
a(s)=1+Cs
\]
and \(\epsilon>0\) so that for perturbations \(u\in \mathcal{S}^{\epsilon}_{\mathrm{weak}(u_L,S^i_{u_L}(s))}\),
\[
\frac{d}{dt} E(t) \le -C\,|u_-(t)-u_L|^2.
\]
Hence the shock is a local attractor [2504.03920].

The corresponding necessary condition has the same geometric content. If for all small \(s\in(0,s_0)\) the shock \((u_L,S^i_{u_L}(s),\sigma(s))\) is a local attractor with weight \(a(s)\), then
\[
-a'(0)\,\nabla^2 \eta(u_L)\big(f'(u_L) - \lambda_i(u_L)I\big) + r_i(u_L)^\top \nabla^2 \eta(u_L)\, f''(u_L)
\]
is negative semidefinite on \(\mathrm{span}\{r_k(u_L)\}_{k\ne i}\) [2504.03920].

This criterion is obtained through the analysis of a “maximal shock” \(u^+(u)\) and of the associated functional \(D_{\max}(u)\), together with the asymptotic Hessian
\[
\frac{1}{s}\,\nabla^2 D_{\max}(u_L)\to -a'(0)\,\nabla^2\eta(u_L)\big(f'(u_L)-\lambda_i(u_L)I\big) + r_i(u_L)^\top \nabla^2 \eta(u_L)\, f''(u_L)
\]
on the transverse eigenspace. A plausible implication is that \(a\)-contraction is governed by a genuinely transverse second-order compatibility between entropy geometry, shock family, and nonlinear flux curvature.

## 4. Extremal families, interior families, and examples

The theory sharply distinguishes extremal and interior characteristic families. For extremal families \(i=1\) or \(i=n\), the operator \(f'(u_L)-\lambda_i(u_L)I\) is sign-definite on \(\mathrm{span}\{r_k\}_{k\ne i}\), so the sufficient condition is always satisfiable. This yields \(a\)-contraction with appropriate weights even for large perturbations in \(\mathcal{S}_{\mathrm{weak}}\). For interior families \(1<i<n\), the necessary matrix condition can fail; when it is not negative semidefinite, the paper states that for all small \(s>0\) and any \(\delta>0\), there exist nearby states \((u_-,u_+)\) with \(u_+=S^i_{u_-}(t)\) and \(|u_- - u_L|+|u_+ - u_R|<\delta\) but
\[
D_{\mathrm{RH}}(u_\pm,\sigma_\pm)>0,
\]
so no local-attractor \(a\)-contraction holds [2504.03920].

The article also places the single-entropy, constant-weight theory in context. In “rich systems,” Serre–Vasseur proved contraction with shifts for any Liu–Majda stable shock by constructing two distinct entropies \(\eta_\pm\) depending on the shock, including for intermediate families. They also treated the Euler contact discontinuities using a single entropy in a special contact structure case. By contrast, the recent small-shock theory works in general systems with a single entropy and constant weights \(a\), and it shows explicitly that some interior shocks cannot be local attractors under this framework [2504.03920].

A model \(3\times 3\) example illustrates positive results. For the system with state \(U=(u,v,w)\), entropy \(\eta(U)=\frac12|U|^2\), and intermediate family \(i=2\), the critical matrix becomes
\[
\begin{pmatrix}
-2C-2 & 0 & 2\alpha\\
0 & 2 & 0\\
2\alpha & 0 & 2C-6
\end{pmatrix},
\]
which is negative definite on \(\mathrm{span}\{r_1,r_3\}\) for \(|\alpha|<4\) and \(|\alpha|-1<C<3-|\alpha|\). Hence small \(i=2\) shocks are local attractors with \(a(s)=1+Cs\). By contrast, for 2D isentropic MHD and interior family \(i=2\), the analogous matrix fails to be negative semidefinite on \(V=\mathrm{span}\{r_1,r_3,r_4\}\) for many states, including large specific volume \(v\), yielding cases where small interior shocks are not local attractors for any constant weight \(a(s)\) [2504.03920].

A common misconception is that the weight in \(a\)-contraction is always a spatial profile attached to the shock. In the small-shock hyperbolic theory just described, the weight is constant in space and depends only on the shock amplitude \(s\); the paper states explicitly that there is no ODE for \(a(x)\) along the shock profile.

## 5. Viscous and viscous–dispersive shock applications

For the one-dimensional barotropic Navier–Stokes system in Lagrangian coordinates,
\[
\begin{cases}
v_t - u_x = 0,\\
u_t + p(v)_x = \left(\mu(v)\,\dfrac{u_x}{v}\right)_x,
\end{cases}
\]
the method is adapted to viscous shock profiles \((\widetilde v,\widetilde u)\) by introducing a shock-aligned weight
\[
a(t,x):=1+\frac{u_- - \widetilde{u}(x-\sigma t)}{\sqrt{\delta}},
\]
together with a shifted weight \(a^X(t,x)=a(x-\sigma t-X(t))\) and the weighted relative entropy
\[
E_a(t):=\int_{\mathbb{R} a^X(t,x)\,\eta\big(U(t,x)\mid \widetilde{U}^X(t,x)\big)\,dx.
\]
A central point of the 2024 analysis is that it does not employ the effective velocity variable \(h\) even for higher order estimates. The resulting contraction estimate has the form
\[
\int_{\mathbb{R} a\,\eta(U\mid \widetilde{U}^X)(t)\,dx + \int_0^t\Big(\delta\,|\dot{X}|^2 + \mathcal{G}_1 + G^S + \mathcal{D}\Big)\,ds
\le C\int_{\mathbb{R} a\,\eta(U\mid \widetilde{U}^X)(0)\,dx,
\]
and yields global existence, convergence in sup norm toward the shifted viscous shock, and \(\dot X(t)\to 0\) [2410.12875].

For the Navier–Stokes–Korteweg system, the 2026 extension reintroduces an effective velocity
\[
h(t,x) = u(t,x) -\tau_1\,\mu(v)\,\frac{v_x}{v},
\qquad \tau_1 = \frac{1+\sqrt{1-4c}}{2},
\]
which transforms the system into two degenerate parabolic equations. The paper then applies the method of \(a\)-contraction with shifts to prove the contraction property of any large solution perturbed from a viscous-dispersive shock wave. The contraction property does not depend on the strengths of viscosity and capillarity. This uniformity yields global existence for large perturbations and zero viscosity–capillarity limits, on which Riemann shocks are unique and stable up to shifts [2602.13788].

Taken together, these results show that \(a\)-contraction with shifts is not merely a hyperbolic relative-entropy criterion; it is a transportable stability architecture for viscous and viscous–dispersive regularizations, with the shift absorbing translation and the \(a\)-weight encoding the asymmetry of the shock.

## 6. Related meanings of contraction with shifts

The phrase “contraction with shifts” also appears in other branches of analysis, but with different technical content. In nonlinear dynamical systems, contraction “with shifts” is also called transverse contraction, horizontal contraction, or contraction modulo time/phase reparametrization. There, contraction is required only in directions transverse to the flow \(f\), while tangential directions along \(f\) are neutral. Two trajectories are considered equivalent if there exists a reparameterization \(t\mapsto \theta(t)\) such that
\[
\|x_2(\theta(t)) - x_1(t)\|\to 0\quad \text{as } t\to\infty,
\]
and uniform transverse contraction on a compact forward-invariant set without equilibria implies existence of a unique periodic orbit together with exponential orbital convergence [2203.01367].

In symbolic dynamics, the analogous structure is a contraction homotopy
\[
h: I \times X \times X \to X,\qquad I=\{0,1\}^G,
\]
satisfying
\[
h(\bar 0,x,y)=x,\qquad h(\bar 1,x,y)=y.
\]
Here the “time parameter” is itself a binary full shift, and homotopies must be block maps commuting with the group action. The notion is used to characterize retracts of full shifts, dense periodic points under finite periodic asymptotic dimension, and several map-extension properties [2401.16774].

A further, unrelated fixed-point usage employs “shifting distance functions” \((\Phi,\Psi)\) satisfying
\[
\Phi\big(d(Tx, Ty)\big) \le \Psi\big(d(x, y)\big),
\]
which generalizes the Banach contraction principle on complete metric spaces [1310.0995].

These usages are mathematically distinct. What they share is the replacement of literal pointwise contraction by contraction after a symmetry reduction, a shift, or a transformed comparison. In the conservation-law setting, that symmetry is spatial translation of a shock profile, and the resulting \(a\)-contraction with shifts has become a precise tool for proving orbital \(L^2\) stability and for identifying when shocks are, or are not, local attractors.

Source: https://www.emergentmind.com/topics/a-contraction-with-shifts