---
title: 'Burgers–Huxley Equation with Memory: Models & Analysis'
url: https://www.emergentmind.com/topics/generalized-burgers-huxley-equation-with-memory
type: topic
---

# Burgers–Huxley Equation with Memory: Models & Analysis

Searching arXiv for recent papers on the generalized Burgers–Huxley equation with memory and related numerical/stabilization analyses.
The generalized Burgers–Huxley equation with memory is a nonlinear advection–diffusion–reaction model in which the diffusive part is augmented by a hereditary convolution term, typically acting on the Laplacian through either a weakly singular kernel or an exponentially decaying fading-memory kernel. In the recent arXiv literature, the model appears in finite element, discontinuous Galerkin, hp-time-stepping, a posteriori estimation, and feedback-stabilization settings, with homogeneous Dirichlet, mixed Dirichlet–Neumann, and controlled boundary conditions treated in different formulations [2309.01636], [2403.08269], [2503.21887], [2508.02654].

## 1. Governing equations and memory mechanisms

A standard weakly singular-kernel formulation on a bounded domain \(\Omega\subset\mathbb{R}^d\), \(d\in\{2,3\}\), is
\[
\begin{aligned}
\partial_t u(x,t) + \alpha\,u(x,t)^{\delta}\sum_{i=1}^d \partial_{x_i}u(x,t) - \nu\,\Delta u(x,t)
-\eta \int_0^t K(t-\tau)\,\Delta u(x,\tau)\,d\tau \\
= \beta\,u(x,t)\,\big(1-u(x,t)^{\delta}\big)\,\big(u(x,t)^{\delta}-\gamma\big) + f(x,t),
\qquad (x,t)\in\Omega\times(0,T),
\end{aligned}
\]
with homogeneous Dirichlet boundary data and initial condition \(u(x,0)=u_0(x)\). Here \(\alpha,\nu,\beta\ge 0\), \(\eta\ge 0\), \(\delta\ge 1\), and \(\gamma\in(0,1)\). The memory term is a Volterra convolution in time applied to \(\Delta u\), so the hereditary effect is through diffusion rather than through a discrete delay state \(u(x,t-\tau)\) [2403.08269], [2309.01636].

The associated stationary generalized Burgers–Huxley equation removes both \(\partial_t u\) and the memory contribution:
\[
\alpha\,u^\delta \sum_{i=1}^d \partial_{x_i}u - \nu\,\Delta u
= \beta\,u(1-u^\delta)(u^\delta-\gamma) + f
\quad \text{in }\Omega,\qquad u=0 \text{ on }\partial\Omega.
\]

For weak formulations, the nonlinear structures are often written through
\[
b(u,v,w)=\int_\Omega u^\delta \Big(\sum_{i=1}^d \partial_{x_i}v\Big)\,w\,dx,
\qquad
c(u)=u\,(1-u^\delta)\,(u^\delta-\gamma).
\]
In this notation, the memory term becomes a time convolution on gradients after integration by parts:
\[
\eta\int_0^t K(t-\tau)\,(\nabla u(\tau),\nabla v)\,d\tau.
\]

Two kernel classes dominate the literature. For weakly singular memory, one prototypical choice is
\[
K(t)=\frac{1}{\Gamma(\gamma)}\,t^{\gamma-1},\qquad 0<\gamma<1,
\]
or, in computational studies, \(K(t)=t^{-1/2}\). For stabilization problems, the kernel is typically exponential:
\[
K(t)=e^{-\lambda t}
\quad\text{or}\quad
K(t)=e^{-\delta t},
\]
which is positive, strictly decaying, integrable, completely monotone, and admits the explicit Laplace transform \(\widehat K(\xi)=1/(\delta+\xi)\) in the boundary-control analysis [2403.08269], [2503.21887], [2508.02654].

## 2. Variational structure, positivity, and regularity

The basic analytical framework uses the Gelfand triplet \(H_0^1(\Omega)\hookrightarrow L^2(\Omega)\hookrightarrow H^{-1}(\Omega)\), augmented by \(L^{2(\delta+1)}(\Omega)\) control to handle the Huxley nonlinearity. For the time-dependent weakly singular-kernel problem, one seeks
\[
u\in L^\infty(0,T;L^2(\Omega))\cap L^2(0,T;H_0^1(\Omega))
\cap L^{2(\delta+1)}(0,T;L^{2(\delta+1)}(\Omega)),
\]
with \(\partial_t u\) in a dual space, and tests against \(v\in H_0^1(\Omega)\cap L^{2(\delta+1)}(\Omega)\) [2403.08269].

A structural assumption common to the weakly singular analyses is that \(K\in L^1(0,T)\) is of positive type:
\[
\int_0^T\!\!\int_0^t K(t-\tau)\,w(\tau)\,w(t)\,d\tau\,dt\ge0
\qquad \forall\,w\in L^2(0,T).
\]
This positivity is repeatedly used in energy identities and stability estimates. In the a posteriori DG analysis, positivity of type and integrability near \(0\) are sufficient to control the memory contribution, and no fractional Grönwall inequality is required [2403.08269]. In the conforming and nonconforming finite element analyses, the same positivity enters the continuous and discrete energy inequalities [2309.01636], [2310.07788].

Existence, uniqueness, and regularity are established under standard data assumptions. For weak solutions, one result assumes \(u_0\in L^2(\Omega)\), \(f\in L^2(0,T;H^{-1}(\Omega))\), and \(K\in L^1(0,T)\) of positive type; the resulting solution belongs to
\[
L^{\infty}(0,T;L^{2}(\Omega))\cap L^{2}(0,T;H_0^1(\Omega))
\cap L^{2(\delta+1)}(0,T;L^{2(\delta+1)}(\Omega)),
\]
with
\[
\partial_tu\in L^{\frac{2(\delta+1)}{2\delta+1}}
\big(0,T; H^{-1}(\Omega)+ L^{\frac{2(\delta+1)}{2\delta+1}}(\Omega)\big),
\]
and \(u\in C([0,T];L^2(\Omega))\) [2309.01636]. Uniqueness is shown either under \(u_0\in L^{d\delta}(\Omega)\) or under the parameter condition
\[
\beta\,\nu>(2^{\delta}\alpha)^2.
\]

For stronger regularity, if \(\Omega\) is convex or has \(C^2\)-boundary, \(u_0\in H_0^1(\Omega)\cap L^{2(\delta+1)}(\Omega)\), and \(f\in L^2(0,T;L^2(\Omega))\), then
\[
u\in C([0,T];H_0^1(\Omega))\cap L^2(0,T;H^2(\Omega))
\cap L^{2(\delta+1)}(0,T;L^{6(\delta+1)}(\Omega)),
\qquad
\partial_tu\in L^2(0,T;L^2(\Omega)).
\]
Under \(u_0\in H^2(\Omega)\cap H_0^1(\Omega)\) and \(f\in H^1(0,T;L^2(\Omega))\), one further obtains
\[
\partial_t u\in L^\infty(0,T;L^2(\Omega))\cap L^2(0,T;H_0^1(\Omega)),
\]
and with \(f\in L^2(0,T;H^1(\Omega))\),
\[
u\in L^\infty(0,T;H^2(\Omega)).
\]
The stabilization literature with exponential kernels uses related weak and strong solution spaces, but under mixed boundary conditions and, in the boundary-control case, nonhomogeneous Dirichlet data on \(\Gamma_1\) and homogeneous Neumann data on \(\Gamma_2\) [2309.01636], [2508.02654].

## 3. Discretization frameworks

The numerical literature on the GBHE with memory covers conforming FEM, nonconforming Crouzeix–Raviart FEM, SIPG-type DGFEM in space, backward Euler and Crank–Nicolson time discretizations, and hp-DG time-stepping [2309.01636], [2310.07788], [2403.08269], [2409.00818].

| Formulation | Space discretization | Time treatment |
|---|---|---|
| Conforming approximation | \(V_h\subset H_0^1(\Omega)\), typically \(\mathbb P_1\) | semi-discrete, backward Euler, or hp-DG |
| Nonconforming approximation | Crouzeix–Raviart space \(V_h\) | backward Euler with positive memory quadrature |
| DG approximation | SIPG / DGFEM space \(V_h^{DG}\) or \(V_{DG}\) | semi-discrete, backward Euler, Crank–Nicolson, or hp-DG |

For conforming FEM, the semi-discrete scheme seeks \(u_h(t)\in V_h\) such that
\[
\langle\partial_tu_h(t),\chi\rangle+\nu (\nabla u_h(t),\nabla \chi)
+\alpha\,b(u_h(t),u_h(t),\chi)
+\eta\big((K*\nabla u_h)(t),\nabla \chi\big)
=\beta(c(u_h(t)),\chi)+\langle f(t),\chi\rangle
\]
for all \(\chi\in V_h\) [2309.01636].

For nonconforming CR methods, the central device is a skew-symmetrized convection form
\[
b_{CR}(u;u,w):=\frac{1}{\delta+2}\Bigl(u^{\delta}\sum_{i=1}^d \frac{\partial u}{\partial x_i},w\Bigr)_{\mathcal{T}_h}
-\frac{1}{\delta+2}\Bigl(u^{\delta}\sum_{i=1}^d \frac{\partial w}{\partial x_i},u\Bigr)_{\mathcal{T}_h},
\]
which satisfies \(b_{CR}(u;u,u)=0\). This identity is used to obtain stability without parameter restrictions [2310.07788].

For DG methods in space, the diffusion term is typically discretized by the symmetric interior penalty form
\[
\begin{aligned}
a_{DG}(u,v) &= \sum_{K}(\nabla u,\nabla v)_K
- \sum_{E}\int_E \{\!\{\nabla u\}\!\}\cdot [\![v]\!]\,ds
- \sum_{E}\int_E \{\!\{\nabla v\}\!\}\cdot [\![u]\!]\,ds \\
&\quad+ \sum_{E}\int_E \gamma_h\,[\![u]\!]\cdot [\![v]\!]\,ds,
\end{aligned}
\]
with DG norm
\[
\|v\|_{DG}^2
=
\sum_{K}\|\nabla v\|_{L^2(K)}^2
+\sum_{E}\frac{\gamma}{h_E}\|[v]\|_{L^2(E)}^2
\]
or, in the adaptive a posteriori analysis,
\[
\|\,|\,|v\|\,|\,|^2
=
\sum_{K}\|\nabla v\|_{0,K}^2
+
\sum_{E}\gamma_h\|[\![v]\!]\|_{0,E}^2.
\]
The DG advection form is built with upwind fluxes and satisfies \(b_{DG}(w;v,v)=0\) and \(b_{DG}(w;u,v)=-b_{DG}(w;v,u)\) in the analyses that exploit energy cancellation [2403.08269], [2310.07788].

Time discretization of memory is usually handled by product integration. In backward Euler form,
\[
J(\psi)(t_k)=\int_0^{t_k} K(t_k-s)\psi(s)\,ds
\approx
\sum_{j=1}^{k}\omega_{kj}\,\Delta t\,\psi^j,
\]
with
\[
\omega_{kj}
=
\frac{1}{(\Delta t)^2}\int_{t_{k-1}}^{t_k}\int_{t_{j-1}}^{\min(t,t_j)}K(t-s)\,ds\,dt.
\]
The fully discrete DG a posteriori paper additionally treats Crank–Nicolson with midpoint quadrature and mesh-transfer operators \(I^k\), and states that no CFL restriction is imposed in the analysis; stability follows from positivity of the kernel and elliptic coercivity of the DG form [2403.08269].

The hp-DG time-stepping formulation introduces a time partition \(J_n=(t_{n-1},t_n]\) and piecewise polynomial spaces
\[
S(\mathcal{M},p)
=
\{ v:[0,T]\to H_0^1(\Omega): v|_{J_n}\in \mathbb{P}_{p_n}\},
\]
with jumps \([v]^n=v^n-v_n\). In the case \(p=(0,\dots,0)\), the method reduces to a backward-Euler-type update [2409.00818].

## 4. Error analysis: a priori, a posteriori, and optimality

The weakly singular-kernel literature contains both a priori and a posteriori error analyses. In conforming FEM, a semi-discrete error estimate under minimal regularity reads
\[
\|u_h-u\|_{L^{\infty}(0,T;L^2(\Omega))}^2+\|u_h-u\|_{L^2(0,T;H_0^1(\Omega))}^2
\le C\Big\{\|u_0^h-u_0\|_{L^2}^2+h^2\int_0^T\|\partial_t u\|_{H_0^1}^2\,dt
+h^2\int_0^T(\|u\|_{H_0^1}^2+\|u\|_{H^2}^2)\,dt\Big\},
\]
and the fully discrete backward-Euler scheme satisfies
\[
\|u- u_{kh}\|_{L^{\infty}(0,T;L^2(\Omega))}^2 + \|u - u_{kh}\|_{L^2(0,T;H_0^1(\Omega))}^2
\le \eta^2(\Delta t)^2\sup_{k,j} \overline{K}_{kj}^2\big(\|f\|_{H^1(0,T;L^2(\Omega))}^2 + \|u_0\|_{H^2}^2\big)
+ C\big((\Delta t)^2 + h^2\big)\big(\|f\|_{H^1(0,T;L^2(\Omega))}^2 + \|u_0\|_{H^2}^2\big).
\]
If \(|K(t)|\lesssim t^{-\alpha}\) with \(0<\alpha<1\), then \(\sup_{k,j}\overline K_{kj}^2=O((\Delta t)^{-2\alpha})\), and the temporal error becomes \(O((\Delta t)^{2(1-\alpha)})\) [2309.01636].

For nonconforming CR and DG spatial discretizations, semi-discrete error estimates of the form
\[
\|u_h-u\|_{L^{\infty}(0,T;L^2)}^2+|\!|\!|u_h-u|\!|\!|^2
\le C\bigl(\|u_0^h-u_0\|_{L^2}^2+h^2\,\Theta(u)\bigr)
\]
are derived, together with fully discrete bounds containing
\[
C(f,u_0)\Bigl(h^2+(\Delta t)^2+\eta^2(\Delta t)^2\,\sup_{k,j}\omega_{kj}^2\Bigr)
\]
for CR and the analogous DG estimate with the DG energy norm [2310.07788].

The hp-DG time-stepping analysis establishes energy-norm optimality with explicit dependence on local time-step sizes \(k_n\), polynomial degrees \(p_n\), spatial mesh sizes \(h_K\), and spatial degrees \(r_K\). For analytic solutions, the local projection estimates yield exponential convergence in \(p_n\); for algebraic regularity, the estimates involve the factor
\[
I_{p,q}=\frac{\Gamma(p+1-q)}{\Gamma(p+1+q)}.
\]
In the fully discrete conforming and DG settings, the resulting bounds are optimal in the energy norm with respect to \(k,h,p\), while the \(L^2\)-in-time error is suboptimal by one power of \(p\); for DG in space, the \(r\)-dependence is also suboptimal relative to the conforming case [2409.00818].

The a posteriori analysis for DG formulations introduces residual-based estimators for the stationary problem, the semi-discrete problem with memory, and fully discrete backward Euler and Crank–Nicolson schemes. For the stationary GBHE, local residuals are
\[
R_K =
\big(f_h + \nu \Delta u_h - \alpha\,u_h^\delta \sum_{i=1}^d \partial_{x_i}u_h + \beta\,u_h(1-u_h^\delta)(u_h^\delta-\gamma)\big)\Big|_K,
\]
with flux-jump residuals \(R_e\) and jump terms \(\zeta_{J_K}\). The global estimator \(\zeta\) satisfies
\[
\|\,|\,|u-u_h\|\,|\,| \le C\,\big(\zeta+\mathcal{F}\big),
\qquad
\zeta \le C\,\big(\|\,|\,|u-u_h\|\,|\,|+\mathcal{F}\big).
\]
For the semi-discrete memory problem,
\[
\|u-u_h\|_{L^\infty(0,T;L^2(\Omega))}^2 + \int_0^T \|\,|\,|u(t)-u_h(t)\|\,|\,|^2\,dt \le C\,\Theta,
\]
where \(\Theta\) contains cell residuals, face residuals, jump terms, and explicit memory-jump contributions. For fully discrete backward Euler and Crank–Nicolson schemes, the reliability bounds combine time indicators \(\Xi_k\), spatial indicators \(\Upsilon_k\), and kernel-quadrature oscillation terms \(\mathcal K_k\) [2403.08269].

The same paper proves optimal \(L^2\)-error estimates for the stationary and semi-discrete evolutionary problems. In particular, for DG degree \(k\ge1\),
\[
\|u-u_h\|_{L^2(\Omega)} \le C\,h^{k+1}\,\|u\|_{H^{k+1}(\Omega)}
\]
for the stationary problem, and for the semi-discrete time-dependent problem
\[
\|e\|_{L^\infty(0,T;L^2(\Omega))}^2 + \int_0^T \|\,|\,|e(t)\|\,|\,|^2\,dt
\le C\Big(\|e(0)\|_0^2 + h^{2(p+1)}\big(\|u\|_{L^\infty H^{p+1}}^2 + \|u_t\|_{L^\infty H^{p+1}}^2\big)\Big).
\]
Fully discrete \(L^2\)-rates in \(\Delta t\) are not proved there, but numerical results report first-order temporal convergence for backward Euler and second-order temporal convergence for Crank–Nicolson [2403.08269].

## 5. Feedback stabilization and control-theoretic formulations

A separate line of work studies GBHEs with exponential memory kernels from the viewpoint of stabilizability around zero or non-constant steady states. One formulation on a bounded \(C^2\) domain with homogeneous Dirichlet boundary conditions and distributed interior control is
\[
\left\{
\begin{aligned}
& y_t - \eta \Delta y + \alpha\, y^\delta \sum_{i=1}^d \frac{\partial y}{\partial x_i}
- \kappa \int_0^t e^{-\lambda (t-s)} \Delta y(\cdot,s)\,ds \\
&\qquad = \beta\, y(1 - y^\delta)(y^\delta - \gamma) + f + u\,\chi_{\mathcal O}, \\
& y = 0 \quad \text{on } \Gamma \times (0,\infty), \\
& y(\cdot,0) = y_0 \quad \text{in } \Omega .
\end{aligned}
\right.
\]
Introducing
\[
z(\cdot,t)=\int_0^t e^{-\lambda (t-s)} y(\cdot,s)\,ds,
\qquad z_t+\lambda z-y=0,
\]
converts the memory equation into a coupled system amenable to semigroup and Riccati analysis [2503.21887].

For the linearized principal system around a steady state, the state \(X=(w,v)\) satisfies
\[
X'(t)=\mathcal A X(t)+\mathcal B u(t),
\]
where \(\mathcal A\) generates an analytic semigroup on \(L^2(\Omega)\times L^2(\Omega)\). The spectral analysis produces explicit eigenvalues
\[
\mu_k^\pm = \frac{-(\beta\gamma + \lambda + \eta \Lambda_k) \pm \sqrt{(\beta\gamma + \lambda + \eta \Lambda_k)^2 - 4(\beta\gamma\lambda + (\eta\lambda + \kappa)\Lambda_k)}}{2},
\]
with \(-\Delta\phi_k=\Lambda_k\phi_k\), and all eigenvalues satisfy \(\Re(\mu_k^\pm)<0\). After a spectral shift \(\mathcal A_\nu=\mathcal A+\nu I\), a feedback operator is obtained from the algebraic Riccati equation
\[
\mathcal A_\nu^* P + P \mathcal A_\nu - P \mathcal B \mathcal B^* P + I = 0,
\qquad P=P^*\ge 0,
\]
leading to the closed-loop generator \(\mathcal A_\nu-\mathcal B\mathcal B^*P\), which is exponentially stable. The nonlinear stabilization results are then obtained by Banach fixed point arguments in the solution space
\[
D = L^2(0,\infty;H^2(\Omega)) \cap L^\infty(0,\infty;H^1_0(\Omega))
\cap L^{2(\delta+1)}(0,\infty;L^{6(\delta+1)}(\Omega))
\cap H^1(0,\infty;L^2(\Omega)),
\]
with smallness assumptions on \(y_0\) and, for non-constant steady states, on \(\|y_\infty\|_{H^2}\) [2503.21887].

The boundary-control literature treats mixed boundary conditions
\[
y|_{\Gamma_1}=u,\qquad \partial_n y|_{\Gamma_2}=0,
\]
and a fading memory term
\[
\int_0^t e^{-\delta(t-s)}\,\Delta y(s)\,ds.
\]
After shifting by a decay factor \(e^{\omega t}\) and performing an elliptic lifting \(D\tilde u\), the feedback is constructed from the eigenfunctions \(\{\varphi_j\}\) of the Laplacian with Dirichlet conditions on \(\Gamma_1\) and Neumann conditions on \(\Gamma_2\). Under the linear independence hypothesis
\[
\left\{\frac{\partial\varphi_i}{\partial n}\Big|_{\Gamma_1}\right\}_{i=1}^{N_\omega}
\text{ is linearly independent in }L^2(\Gamma_1),
\]
the finite-dimensional Dirichlet boundary controller is
\[
u|_{\Gamma_1}(t)
=
\lambda_1 \sum_{j=1}^{N_\omega} \mu_j\,\langle w(t),\varphi_j\rangle_{L^2(\Omega)}\,\Phi_j,
\qquad
\mu_j := \frac{k+\eta\lambda_j}{k+\eta(\lambda_j-\lambda_1)}.
\]
The linear closed-loop system satisfies
\[
\|w(t)\|_{L^2(\Omega)} \le C\,e^{-(\omega+\alpha)t}\|w_0\|_{L^2(\Omega)},
\]
and the full nonlinear system is stabilized by the same controller through a Banach fixed point theorem on
\[
\mathcal D
=
L^\infty(0,\infty;H^1)\cap L^2(0,\infty;H^2)\cap H^1(0,\infty;L^2)
\cap L^{2(\kappa+1)}(0,\infty;L^{6(\kappa+1)}).
\]
If the steady state is zero, the additional constraint \(\omega<\delta\) is not needed [2508.02654].

These stabilization results concern exponential kernels rather than weakly singular kernels. A plausible implication is that the explicit auxiliary-variable reformulation \(z_t+\lambda z-y=0\) and the resolvent-based spectral analysis are especially well aligned with fading-memory kernels of exponential type.

## 6. Numerical behavior, adaptive refinement, and current limitations

The numerical studies cover manufactured solutions, singular solutions, adaptive refinement, finite-dimensional feedback simulation, and qualitative pattern-formation problems [2403.08269], [2503.21887], [2309.01636], [2310.07788], [2409.00818].

For weakly singular kernels, conforming, nonconforming, and DG discretizations on \(\Omega=(0,1)^d\) with \(K(t)=t^{-1/2}\) repeatedly show first-order spatial convergence for piecewise linear spatial approximation in the relevant energy norm. In the conforming approximation paper, smooth-kernel tests with \(K(t)=e^{-\delta t}\) also exhibit first-order spatial convergence in 2D and 3D, both with memory \((\eta=1)\) and without memory \((\eta=0)\) [2309.01636]. The hp-DG time-stepping paper reports, for \(p=0,1,2\) in time, the expected increase in order, including \(L^2\) errors of order approximately \(O(h^3)\) and \(O(h^4)\) and \(H^1\) errors of order approximately \(O(h^2)\) and \(O(h^3)\) in the quadratic-space tests, as well as analogous optimal rates in 3D and in Caputo-type fractional experiments [2409.00818].

The adaptive a posteriori paper applies the standard loop
\[
\text{Solve} \to \text{Estimate} \to \text{Mark} \to \text{Refine}
\]
with a max-strategy marking criterion
\[
\zeta_K \ge \gamma_1\,\max_{L\in \mathcal T_h}\zeta_L,
\qquad 0<\gamma_1<1.
\]
On L-shaped domains, adaptive refinement concentrates near steep localized gradients or re-entrant corners. For the singular solution \(u=(x^2+y^2)^{1/4}\), uniform meshes are suboptimal, but adaptive refinement recovers optimal rates. A time-varying singularity test without memory \((\eta=0)\) shows that a backward-Euler-plus-DG adaptive algorithm tracks a moving peak by refining around the moving singularity at each time step [2403.08269].

The stabilization papers include finite element simulations on \(\Omega=(0,1)^2\) with \(P_1\) FEM semi-discretization. In the Riccati-based interior-control setting, the discrete algebraic Riccati equation is solved and the closed-loop eigenvalues are moved into the negative half-plane; the zero steady state and the non-constant steady state \(y_\infty(x)=\sin(\pi x_1)\sin(\pi x_2)\) both exhibit energy decay under feedback, whereas the uncontrolled shifted dynamics do not [2503.21887].

Several application-oriented computations treat coupled or phenomenological systems. One conforming FEM study considers a FitzHugh–Nagumo-type system with memory and reports that spiral-wave dynamics respond to the nonlinearity exponent \(\delta\) while the weakly singular memory modulates pattern formation [2309.01636]. The nonconforming/DG paper also examines a fractional-time-derivative extension and 2D spiral-wave behavior on \(\Omega=(0,300)^2\), reporting that small \(\eta\) preserves spiral structure while larger \(\eta\) affects or reverses it [2310.07788]. The hp-DG paper includes a prey–predator application in which increasing memory prolongs dynamics and affects attractors [2409.00818].

Several limitations are explicit. The weakly singular-kernel analyses generally assume \(K\in L^1\) and positive type; stronger singularities or nonpositive kernels are not covered in the conforming, nonconforming, and a posteriori works [2309.01636], [2310.07788], [2403.08269]. Most nonlinear estimates are restricted to \(d\le 3\), with upper bounds on the exponent in three dimensions such as \(\delta\in\{1,2\}\) or \(\kappa\in\{1,2\}\) in the stabilization setting [2503.21887], [2508.02654]. The fully discrete analysis in the adaptive DG paper is a posteriori rather than a priori in \(\Delta t\), even though the observed temporal orders match first-order backward Euler and second-order Crank–Nicolson [2403.08269]. In the same work, the discrete memory sum requires a growing history at each time step, so the cost and storage are \(O(N^2)\) without compression; fast convolution or sum-of-exponentials acceleration is noted as beyond the scope [2403.08269].

A recurrent misconception is that the term “delayed GBHE” denotes a discrete delay equation. In the finite element papers, it does not: no term of the form \(u(x,t-\tau)\) appears, and the entire delay mechanism is hereditary memory represented by a time convolution with \(K\) acting on \(\Delta u\) or \(\nabla u\) [2309.01636], [2310.07788].

Source: https://www.emergentmind.com/topics/generalized-burgers-huxley-equation-with-memory