---
title: Parabolic-Hyperbolic Skeleton Equation
url: https://www.emergentmind.com/topics/parabolic-hyperbolic-skeleton-equation
type: topic
---

# Parabolic-Hyperbolic Skeleton Equation

The parabolic-hyperbolic skeleton equation denotes a class of canonical or limiting equations in which parabolic and hyperbolic dynamics are coupled either across a type-change interface, through a nonlocal gluing law, or via a singular-perturbation or large-deviation reduction. In the mixed-type boundary-value literature, the canonical model is the piecewise equation
$$
0=\begin{cases}
u_{xx}-u_y, & y>0,\\
u_{yy}-u_{xx}, & y<0,
\end{cases}
$$
posed on a domain partitioned by the interface $\Gamma=\{y=0\}$; in the non-equilibrium large-deviations literature, the skeleton equation is
$$
\partial_t \rho=\Delta\Phi(\rho)-\nabla\cdot\big(\Phi^{1/2}(\rho)\,g\big);
$$
and in the weakly dissipative Kirchhoff setting, the parabolic limit
$$
b(t)u'(t)+m(|A^{1/2}u(t)|^2)Au(t)=0
$$
acts as the skeleton for a hyperbolic equation with small inertia [2302.04583], [1910.11860], [1203.6581]. This suggests that the expression is contextual rather than universal.

## 1. Terminological scope and model classes

Within the supplied literature, the term “skeleton equation” is used explicitly in the large-deviation analysis of conservative interacting particle systems, where it names a degenerate parabolic-hyperbolic PDE with irregular drift on $\mathbb T^d\times(0,T)$ [1910.11860]. By contrast, the 2023 boundary-value study does not use the term and instead treats a canonical mixed parabolic–hyperbolic PDE in a bounded planar domain split by $y=0$ [2302.04583]. A fractional analogue replaces the parabolic part by a Caputo fractional diffusion equation and couples it to a hyperbolic equation by an integral gluing condition on the interface segment $AB=\{(x,0):0<x<1\}$ [1305.4992]. In a different but related usage, the weakly dissipative Kirchhoff literature identifies the parabolic limit of a hyperbolic equation as the “skeleton” or leading-order dynamics in the singular limit $\varepsilon\to0$ [1203.6581].

These usages share a structural theme rather than a single governing formula. Each formulation isolates a reduced dynamics that mediates between diffusion-dominated and wave- or transport-dominated behavior. In the planar mixed-type problems, the reduction is geometric and interfacial: the PDE changes type across a prescribed curve. In the large-deviation setting, the reduction is variational and control-theoretic: the skeleton equation is the controlled hydrodynamic constraint entering the rate function. In the Kirchhoff setting, the reduction is asymptotic: the skeleton is the parabolic equation obtained by formally suppressing the inertial term.

## 2. Canonical mixed-type equation and geometric classification

The canonical mixed-type model is
$$
0=\begin{cases}
u_{xx}-u_y, & y>0,\\
u_{yy}-u_{xx}, & y<0.
\end{cases}
$$
For $y>0$, the second-order part has $A=1$, $B=0$, $C=0$, hence $D=AC-B^2=0$, so the equation is parabolic. For $y<0$, the second-order part is $(-1)u_{xx}+u_{yy}$, with $A=-1$, $B=0$, $C=1$, hence $D=-1<0$, so the equation is hyperbolic [2302.04583].

The domain $D$ is a simply connected bounded subset of the $(x,y)$-plane. In the upper part $D_1=\{y>0\}$ it is bounded by the vertical segments on $x=0$ and $x=1$ and by the line $y=1$. In the lower part $D_2=\{y<0\}$ it is bounded by the characteristics
$$
AC:\ x+y=0,\qquad BC:\ x-y=1,
$$
issuing from $A(0,0)$ and $B(1,0)$. The type-change curve is the horizontal line
$$
\Gamma:\ y=0.
$$
The change of type is therefore imposed by switching the governing operator across $\Gamma$, not by a single operator whose discriminant vanishes on the interface [2302.04583].

The corresponding boundary data are also split by type. In the parabolic region,
$$
u(0,y)=\phi_0(y),\qquad u(1,y)=\phi_1(y),\qquad y>0.
$$
In the hyperbolic region, the data are prescribed as a linear combination of characteristic traces:
$$
a\,u(x/2,-x/2)+b\,u((x+1)/2,(x-1)/2)=\psi(x),\qquad 0\le x\le1,
$$
with real constants $a,b$ satisfying $a^2+b^2>0$ [2302.04583]. This choice is characteristic of mixed-type formulations in which the upper and lower subdomains require different boundary mechanisms.

A fractional Tricomi-type variant is posed on
$$
\Omega=\Omega^+\cup\Omega^-\cup AB,
$$
with
$$
\Omega^+=\{(x,y):0<x<1,\ 0<y<1\},\qquad
\Omega^-=\{(x,y):-y<x<y+1,\ -1/2<y<0\},
$$
and equations
$$
u_{xx}(x,y)-{}^C D_{0y}^{\alpha}u(x,y)=0,\qquad (x,y)\in\Omega^+,\ 0<\alpha<1,
$$
$$
u_{xx}(x,y)-u_{yy}(x,y)=0,\qquad (x,y)\in\Omega^-.
$$
Here
$$
{}^C D_{0y}^{\alpha} u(x,y)=\frac{1}{\Gamma(1-\alpha)}\int_{0}^{y}\frac{\partial u}{\partial y}(x,\tau)(y-\tau)^{-\alpha}\,d\tau,
$$
so the upper region is a time-fractional diffusion-type regime and the lower region remains hyperbolic [1305.4992].

## 3. Interface traces, transmission, and gluing laws

For the canonical mixed problem, regularity imposes continuity across $\Gamma$:
$$
u|_{\Gamma^-}=u|_{\Gamma^+}=\tau(x),
$$
where
$$
\tau(x):=\lim_{y\to+0}u(x,y).
$$
The parabolic equation above the interface implies the relation
$$
\nu(x):=\lim_{y\to+0}u_y(x,y)=\tau''(x).
$$
The hyperbolic region is then treated as a Cauchy problem with data
$$
u(x,0)=\tau(x),\qquad u_y(x,0)=\nu(x),
$$
so the interface is the locus where the two subproblems are glued through continuity and the PDE-implied identity $\nu=\tau''$ [2302.04583].

The boundary data must satisfy a compatibility condition obtained by evaluating the characteristic relation at the endpoints:
$$
a^2\lim_{y\to+0}\phi_0(y)-b^2\lim_{y\to+0}\phi_1(y)=a\psi(0)-b\psi(1).
$$
In the theorem of the model problem, the data regularity is
$$
\phi_0,\phi_1\in C[0,1],\qquad \psi\in C^1[0,1]\cap C^2(0,1),
$$
together with $a\ne b$ [2302.04583].

The fractional problem replaces local gluing by a nonlocal integral condition on $AB$:
$$
\lim_{y\to+0} y^{1-\alpha}u_y(x,y)
=
\gamma_1 u_y(x,-0)+\gamma_2\int_{0}^{x}u_y(t,-0)\,Q(x,t)\,dt,
\qquad 0<x<1.
$$
With the trace notation
$$
\tau_1(x)=u(x,+0),\quad \tau_2(x)=u(x,-0),\quad \nu_2(x)=u_y(x,-0),\quad \nu_1(x)=\lim_{y\to+0}y^{1-\alpha}u_y(x,y),
$$
the interface system becomes
$$
\tau_1'(x)-\Gamma(\alpha)\nu_1(x)=0,
$$
$$
\nu_1(x)=\gamma_1\nu_2(x)+\gamma_2\int_0^x \nu_2(t)Q(x,t)\,dt.
$$
The hyperbolic side yields
$$
\tau_2'(x)-2\psi'(x/2)=\nu_2(x).
$$
This gluing condition generalizes the standard continuity/flux match and reduces to a local classical gluing when $Q\equiv0$ [1305.4992].

A recurring misconception is that transmission for mixed-type problems must be encoded by an independent jump or flux law. In the canonical constant-coefficient model, no separate jump or flux conditions are imposed beyond continuity across $\Gamma$ and the PDE-implied relation $\nu=\tau''$ [2302.04583]. In the fractional model, by contrast, the interface law is explicitly nonlocal and weighted by $y^{1-\alpha}$ on the parabolic side [1305.4992].

## 4. Interfacial reduction, explicit representations, and well-posedness

For the canonical model, the hyperbolic Cauchy problem in $y<0$ is solved by the d’Alembert-type formula
$$
u(x,y)=\frac{\tau(x+y)+\tau(x-y)}{2}
-\frac12\int_{x+y}^{x-y}\nu(\xi)\,d\xi.
$$
Substituting this into the characteristic boundary condition and differentiating gives
$$
\nu(x)=\frac{a+b}{a-b}\tau'(x)-\frac{2}{a-b}\psi'(x).
$$
Combining this with $\nu=\tau''$ produces the linear ODE
$$
\tau''(x)-\frac{a+b}{a-b}\tau'(x)=-\frac{2}{a-b}\psi'(x),
$$
subject to
$$
\tau(0)=\phi_0(0),\qquad \tau(1)=\phi_1(0).
$$
Multiplication by the integrating factor
$$
p(x)=\exp\!\left(-\frac{a+b}{a-b}x\right)
$$
reduces the problem to a first-order equation for $p(x)\tau'(x)$, and this yields a unique $\tau\in C^2[0,1]$ under the stated hypotheses [2302.04583].

Once $\tau$ is determined, the solution is reconstructed explicitly in each subdomain. In $D_2$, the d’Alembert formula applies. In $D_1$, the parabolic component is represented by
$$
u(x,y)
=
\int_0^y G_\xi(x,y;0,\tau)\phi_0(\tau)\,d\tau
-
\int_0^y G_\xi(x,y;1,\tau)\phi_1(\tau)\,d\tau
+
\int_0^1 G(x,y;\xi,0)\tau(\xi)\,d\xi,
$$
where the Dirichlet heat kernel on $(0,1)$ is
$$
G(x,y;\xi,\eta)=2\sum_{n=1}^{\infty}\sin(\pi n x)\sin(\pi n \xi)e^{-\pi^2n^2(y-\eta)}.
$$
These formulas define a function
$$
u\in C(\overline D)\cap C^2(D)
$$
satisfying the PDE, all boundary data, and the transmission conditions [2302.04583].

Uniqueness in the homogeneous case is based on the energy-type identity
$$
\int_0^1 \tau(\xi)\nu(\xi)\,d\xi
=
\frac{a+b}{2(a-b)}\big[\tau^2(1)-\tau^2(0)\big]
=0.
$$
With $\tau(0)=\tau(1)=0$, this forces $\tau\equiv0$, hence $\nu\equiv0$, then $u\equiv0$ in both subdomains. The resulting theorem states that if
$$
\phi_0,\phi_1\in C[0,1],\qquad \psi\in C^1[0,1]\cap C^2(0,1),\qquad a\ne b,
$$
and the compatibility condition holds, then the problem has a unique regular solution in $C(\overline D)\cap C^2(D)$. No Lopatinskii–Shapiro or smallness assumptions are invoked; existence and uniqueness are proved by explicit interfacial reduction to an ODE for $\tau$ combined with explicit solution operators in each subdomain [2302.04583].

The fractional Tricomi problem follows a parallel but more nonlocal route. The interface equations reduce the problem to a second-order ODE for $\tau_1$ and then to a Fredholm equation of the second kind,
$$
\tau_1(x)-\int_0^1 \tau_1(\xi)K(x,\xi)\,d\xi=F_2(x),
$$
whose unique solution is written via the resolvent kernel $R$:
$$
\tau_1(x)=F_2(x)-\int_0^1 F_2(\xi)R(x,\xi)\,d\xi.
$$
The uniqueness theorem assumes
$$
\gamma_2\ge0,\qquad \frac{\partial Q(x,t)}{\partial t}=-Q_1(x)Q_2(t),\qquad Q(x,x)>0\ \text{on}\ [0,1],
$$
while the existence theorem requires
$$
\phi_1,\phi_2,\psi\in C[0,1]\cap C^1(0,1),\qquad Q\in C([0,1]\times[0,1]).
$$
The solution in $\Omega^+$ is represented by a Green function for the fractional diffusion equation involving a Wright-type kernel
$$
e_{1,\beta}(z)=\sum_{n=0}^{\infty}\frac{z^n}{n!\Gamma(\beta-\beta n)},
\qquad \beta=\alpha/2,
$$
and the hyperbolic region is reconstructed by the Cauchy formula [1305.4992].

## 5. Energy-critical skeleton PDE in non-equilibrium large deviations

In the large-deviation theory of conservative interacting particle systems, the skeleton equation is the controlled hydrodynamic PDE
$$
\partial_t \rho=\Delta\Phi(\rho)-\nabla\cdot\big(\Phi^{1/2}(\rho)\,g\big)
\quad\text{in }\mathbb T^d\times(0,T),\qquad
\rho(\cdot,0)=\rho_0,
$$
with nonnegative increasing $\Phi\in C([0,\infty))\cap C_{\mathrm{loc}}^1((0,\infty))$ and irregular conservative drift
$$
g\in L^2([0,T];L^2(\mathbb T^d;\mathbb R^d)).
$$
The entropy functional is
$$
\Psi_\Phi(\xi):=\int_0^\xi \log(\Phi(\xi'))\,d\xi',
$$
and the initial data belong to
$$
\mathrm{Ent}_\Phi(\mathbb T^d)
=
\Big\{\rho_0\in L^1(\mathbb T^d):\rho_0\ge0\ \text{a.e.},\ \int_{\mathbb T^d}\Psi_\Phi(\rho_0)\,dx<\infty\Big\}.
$$
For porous medium nonlinearities $\Phi(\rho)=\rho^m$, the equation is energy-critical for $g\in L_t^2L_x^2$ and $\rho_0\in L_x^1$; it is supercritical for $L^r$ initial data with $r>1$ [1910.11860].

The analytic framework is built on renormalized kinetic solutions. The kinetic function is
$$
\chi(x,\xi,t):=\overline\chi(\rho(x,t),\xi),
$$
and the parabolic defect measure is
$$
p=\delta_0(\xi-\rho)\,\frac{4\,\Phi(\xi)}{\Phi'(\xi)}\,|\nabla\Phi^{1/2}(\rho)|^2.
$$
The kinetic equation takes the conservative form
$$
\partial_t \chi
=
\Phi'(\xi)\Delta_x\chi
-\partial_\xi\big(\Phi^{1/2}(\xi)\,g(x,t)\cdot\nabla_x\chi\big)
+\nabla_x\cdot\big(\Phi^{1/2}(\xi)\,g(x,t)\,\partial_\xi\chi\big)
+\partial_\xi p.
$$
A renormalized kinetic solution is a nonnegative $\rho\in L^\infty([0,T];L^1(\mathbb T^d))$ such that
$$
\Phi^{1/2}(\rho)\in L^2([0,T];H^1(\mathbb T^d))
$$
and the renormalized kinetic identity holds against compactly supported test functions. Classical weak solutions are defined by the same optimal regularity together with the integrated identity obtained from
$$
\partial_t\rho
=
2\nabla\cdot\big(\Phi^{1/2}(\rho)\nabla\Phi^{1/2}(\rho)\big)
-
\nabla\cdot\big(\Phi^{1/2}(\rho)\,g\big).
$$
Under the structural hypotheses denoted as_unique and as_equiv, renormalized kinetic solutions coincide with classical weak solutions [1910.11860].

The key a priori estimate is the entropy-energy inequality
$$
\sup_{t\in[0,T]}\int_{\mathbb T^d}\Psi_\Phi(\rho)\,dx
+
\int_0^T\int_{\mathbb T^d}\big|\nabla\Phi^{1/2}(\rho)\big|^2\,dx\,dt
\le
\int_{\mathbb T^d}\Psi_\Phi(\rho_0)\,dx
+
c\int_0^T\int_{\mathbb T^d}|g|^2\,dx\,dt.
$$
Under as_unique, solutions satisfy the $L^1$ contraction
$$
\|\rho^1-\rho^2\|_{L^\infty([0,T];L^1(\mathbb T^d))}
\le
\|\rho_0^1-\rho_0^2\|_{L^1(\mathbb T^d)}.
$$
Under as_unique and as_compact, existence holds for every $\rho_0\in \mathrm{Ent}_\Phi(\mathbb T^d)$ and $g\in L_t^2L_x^2$, and every renormalized kinetic solution admits an $L^1(\mathbb T^d)$-continuous representative in time. Stability is strong in $L^1([0,T];L^1(\mathbb T^d))$ under weak convergence of controls and initial data, provided uniform entropy bounds hold [1910.11860].

This PDE enters the rate function
$$
I(\rho)
=
\frac12
\inf\Big\{
\|g\|_{L^2([0,T];L^2(\mathbb T^d;\mathbb R^d))}^2:
\partial_t\rho=\Delta\Phi(\rho)-\nabla\cdot(\Phi^{1/2}(\rho)\,g)
\Big\},
$$
and the well-posedness theory supports several applications: identification of the lower semicontinuous envelope of restricted rate functions for zero range processes, zero-noise large deviations for conservative singular SPDE, and $\Gamma$-convergence of finite-mode rate functions. The paper states that the identification of the lower semicontinuous envelope solves a longstanding open problem in the large deviations for zero range processes [1910.11860].

## 6. Parabolic regime as skeleton limit for weakly dissipative hyperbolic equations

A distinct notion of parabolic-hyperbolic skeleton equation appears in the coercive Kirchhoff problem
$$
\varepsilon u_\varepsilon''(t)+b(t)u_\varepsilon'(t)+m(|A^{1/2}u_\varepsilon(t)|^2)Au_\varepsilon(t)=0,
\qquad
b(t)=(1+t)^{-p},
$$
posed in a real Hilbert space $H$ with a self-adjoint nonnegative coercive operator $A$ satisfying
$$
\langle Au,u\rangle\ge \nu |u|^2,\qquad \nu>0,
$$
and a nondegenerate locally Lipschitz Kirchhoff nonlinearity $m$ with
$$
\inf_{\sigma\ge0}m(\sigma)=:\mu>0.
$$
Formally setting $\varepsilon=0$ yields the parabolic limit
$$
b(t)u'(t)+m(|A^{1/2}u(t)|^2)Au(t)=0,
$$
which serves as the skeleton or leading-order dynamics [1203.6581].

The dissipation threshold is determined by
$$
\int_0^\infty b(s)\,ds=\infty \iff p\le1.
$$
For $p\le1$ this is the parabolic regime; for $p>1$ it is the hyperbolic regime. In the coercive case, the parabolic problem has a unique global solution and satisfies
$$
|u(t)|^2+|A^{1/2}u(t)|^2+|Au(t)|^2+(1+t)^{2p}|u'(t)|^2
\le
C\exp\!\left\{-\frac{2\mu\nu}{1+p}\big[(1+t)^{1+p}-1\big]\right\}.
$$
For the hyperbolic equation with $p\in[0,1]$ and small $\varepsilon$, the functional
$$
\Gamma_\varepsilon(t)
=
|u_\varepsilon(t)|^2+|A^{1/2}u_\varepsilon(t)|^2+|Au_\varepsilon(t)|^2+|u_\varepsilon'(t)|^2+\varepsilon|A^{1/2}u_\varepsilon'(t)|^2
$$
obeys the optimal decay bounds
$$
\Gamma_\varepsilon(t)\le C e^{-\beta t}\qquad (p=0,\ \beta<2\mu\nu),
$$
$$
\Gamma_\varepsilon(t)\le C\exp\{-\beta(1+t)^{1-p}\}\qquad (0<p<1,\ \beta>0),
$$
$$
\Gamma_\varepsilon(t)\le C(1+t)^{-\beta}\qquad (p=1,\ \beta>0).
$$
The paper emphasizes a “quite surprising fact”: in the coercive case the analogy between parabolic equations and dissipative hyperbolic equations is weaker than in the noncoercive case [1203.6581].

The singular perturbation is sharpened by a boundary-layer corrector $\theta_\varepsilon$, defined by
$$
\varepsilon \theta_\varepsilon''(t)+b(t)\theta_\varepsilon'(t)=0,\qquad
\theta_\varepsilon(0)=0,\qquad
\theta_\varepsilon'(0)=u_1+m(|A^{1/2}u_0|^2)Au_0.
$$
Writing
$$
u_\varepsilon(t)=u(t)+\theta_\varepsilon(t)+r_\varepsilon(t)=u(t)+\rho_\varepsilon(t),
$$
the remainder satisfies decay-error estimates of order $\varepsilon^2$ with the same decay profiles as the hyperbolic problem. For example, if $(u_0,u_1)\in D(A^{3/2})\times D(A^{1/2})$,
$$
\Gamma_{r,\varepsilon}(t)\le C\varepsilon^2 e^{-\beta t}\quad (p=0),
$$
$$
\Gamma_{r,\varepsilon}(t)\le C\varepsilon^2\exp\{-\beta(1+t)^{1-p}\}\quad (0<p<1),
$$
$$
\Gamma_{r,\varepsilon}(t)\le C\varepsilon^2(1+t)^{-\beta}\quad (p=1).
$$
The optimality theorem shows that no faster decay can hold for nontrivial hyperbolic solutions, so the parabolic limit is the skeleton only after the corrector and the sharp decay mismatch are taken into account [1203.6581].

Taken together, these formulations show that “parabolic-hyperbolic skeleton equation” is best understood as a structural designation. It may refer to a piecewise mixed-type PDE with explicit interfacial reconstruction, a fractional diffusion–wave system with nonlocal gluing, an energy-critical controlled hydrodynamic equation underpinning non-equilibrium large deviations, or the parabolic limit of a weakly dissipative hyperbolic evolution. The common mathematical content is the extraction of a reduced equation that organizes well-posedness, interface coupling, decay, or variational rate-function structure across a parabolic-hyperbolic divide.

Source: https://www.emergentmind.com/topics/parabolic-hyperbolic-skeleton-equation