Time Fractional Fisher-KPP Equation Overview

• Time Fractional Fisher-KPP equations are reaction-diffusion models that replace the classical time derivative with nonlocal operators to incorporate memory and subdiffusive behavior.
• Multiple formulations (Caputo, diffusion-memory, and fractal-time) are used to study effects on front propagation, well-posedness, and numerical stability.
• Numerical methods and asymptotic analyses reveal that the placement of the fractional operator critically influences front dynamics, speeds, and stability.

Time-fractional Fisher-KPP equations are Fisher-KPP reaction-diffusion models in which the classical first-order temporal evolution is replaced or modified by a nonlocal operator in time. In the current literature, this includes at least three distinct constructions: the Caputo formulation $\partial_t^\alpha u = u_{xx}+f(u)$, formulations in which a Riemann–Liouville fractional derivative acts on the diffusion term, and more specialized fractal-time models based on $F^\alpha$-calculus. Across these variants, the common theme is the incorporation of memory or sub-diffusive transport into the Fisher-KPP mechanism of diffusion, proliferation, and saturation, but the analytical framework, admissible notion of front propagation, and numerical treatment depend strongly on where the fractional operator is placed (Ishii, 2023, Fritz et al., 7 Nov 2025, Gortsas, 22 Aug 2025).

1. Formulations and operator placement

A standard time-fractional Fisher-KPP equation is the one-dimensional Caputo model

$$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$

where

$\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$

and $f$ is monostable and of KPP type: $$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$ A canonical example is $f(u)=u(1-u)$. In this formulation, the memory enters through the time derivative itself and is interpreted as sub-diffusion (Ishii, 2023).

A broader two-dimensional diffusion-reaction form treated numerically is a time-fractional equation of the form “fractional or fractal-fractional time derivative $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$,” with

$$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$

This is a generalized Fisher-KPP reaction term, logistic-like in the sense that it combines linear growth and nonlinear saturation. The same work considers both the Caputo derivative and a fractal-fractional derivative in the Riemann-Liouville sense,

$${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$

$F^\alpha$0

together with the relation

$F^\alpha$1

These definitions make explicit that “time-fractional Fisher-KPP” is not a single equation but a family of memory-driven models (Gortsas, 22 Aug 2025).

A separate line of work argues that the fractional operator should act on the transport mechanism rather than on the whole evolution law. In that formulation, the paper studies a time-fractional Fisher-KPP equation in which the Riemann–Liouville fractional derivative acts on the diffusion term, rather than replacing the whole time derivative by a Caputo derivative as is common in the literature. The authors emphasize that the resulting model and the conventional Caputo model are not equivalent in behavior, even though both reduce to the classical Fisher-KPP equation when $F^\alpha$2 (Fritz et al., 7 Nov 2025).

Formulation	Temporal operator	Characteristic emphasis
Caputo Fisher-KPP	Caputo derivative on the full evolution	Sub-diffusion, memory, asymptotic traveling waves
Diffusion-memory Fisher-KPP	Riemann–Liouville derivative acts on diffusion term	Physically consistent anomalous transport
Fractal-time nonlocal Fisher-KPP	$F^\alpha$3-derivative on a fractal time set	Fractally interrupted time evolution and weak-diffusion asymptotics

2. Propagation in the Caputo model

For the classical Fisher-KPP equation, one studies traveling waves of the form $F^\alpha$4. In the time-fractional Caputo equation this reduction fails: substituting $F^\alpha$5 gives

$F^\alpha$6

and the right-hand side still depends on $F^\alpha$7 through the upper limit of integration. Hence the resulting equation is not autonomous in $F^\alpha$8, and the standard definition of a traveling wave does not apply (Ishii, 2023).

To address this, the notion of an asymptotic traveling wave solution is introduced through the formal large-time limit

$F^\alpha$9

with front conditions

$$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$0

This is not an exact traveling-wave reduction of the PDE; it is a candidate asymptotic object for long-time front dynamics. Under the additional hypothesis

$$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$1

the paper defines a critical speed $$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$2 and proves that for any $$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$3, there exists an asymptotic traveling wave solution with an increasing profile $$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$4 satisfying $$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$5 and $$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$6. If $$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$7, the leading edge is governed by the smallest positive root $$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$8 of

$$\partial_t^\alpha u = u_{xx}+f(u), \qquad 0<\alpha<1,$$9

and the solution can be normalized so that

$\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$0

The same analysis shows that $\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$1 converges to the classical Fisher-KPP threshold,

$\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$2

The construction is based on a monotone iteration method for the nonlocal operator

$\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$3

together with a maximum principle. The paper builds explicit upper and lower solutions, iterates them through a Green-function representation, and obtains a monotone front. It also proves that if $\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$4 with $\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$5 from the theorem, then $\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$6 is a subsolution of the original PDE. Numerical simulations for

$\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$7

show that a sharp front forms and propagates, that the front moves roughly like $\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$8, and that the numerical propagation speed $\partial_t^\alpha u(t,x) := \frac{1}{\Gamma(1-\alpha)} \int_0^t \frac{\partial u}{\partial t}(s,x)}{(t-s)^\alpha}\,ds,$9 is close to $f$0, especially when $f$1 is near $f$2. The authors explicitly note that they have not proved that $f$3 is the exact minimal wave speed for the nonlinear problem (Ishii, 2023).

3. Memory acting on diffusion: well-posedness and stability

A distinct time-fractional Fisher-KPP framework places the memory operator on the diffusion term rather than on the full evolution law. This model is motivated by tumor growth in heterogeneous tissue and by subdiffusive transport due to trapping, crowding, viscoelastic hindrance, and heterogeneous microstructure. The stated modeling claim is that, in derivations based on continuous-time random walks and subordination, the memory naturally acts on the diffusion operator; the paper therefore contrasts its formulation with the more common Caputo-in-time model and regards the latter as mathematically convenient but physically less consistent (Fritz et al., 7 Nov 2025).

On a bounded Lipschitz domain $f$4, $f$5, the paper proves local well-posedness of weak solutions. For any

$f$6

there exists

$f$7

such that the problem has a weak solution

$f$8

satisfying the corresponding weak formulation. The proof uses a Galerkin approximation built from Laplace eigenfunctions, and the key a priori estimate is closed by a Henry–Gronwall–Bihari inequality rather than a standard Gronwall argument. This is necessitated by an estimate of the form

$f$9

which contains a cubic-type nonlinearity at the energy level.

For sufficiently small initial data, the same paper proves global well-posedness. There exists $$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$0 such that if

$$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$1

then the equation has a unique global weak solution

$$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$2

with $$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$3 weakly. The extension argument restarts the equation at a putative maximal time and introduces a history force

$$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$4

which is shown to lie in the required fractional space, allowing the local theory to be re-applied.

For homogeneous Dirichlet boundary conditions, the paper derives asymptotic stability through the bound

$$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$5

so that

$$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$6

For Neumann boundary conditions, decay to zero is not guaranteed because the mean mode may persist. The same paper observes that the conventional Caputo model admits a weak comparison principle more easily and is analytically simpler, but emphasizes that the two models are not behaviorally equivalent (Fritz et al., 7 Nov 2025).

4. Numerical methods and computational structure

The numerical analysis of time-fractional Fisher-KPP equations is shaped by two coupled difficulties: nonlinear reaction and temporal nonlocality. One approach develops a Local Domain Boundary Element Method (LD-BEM) for nonlinear time-fractional Fisher-KPP problems in two dimensions. The method starts from the Laplace fundamental solution

$$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$7

and works with a time-discrete integral representation in which transient and nonlinear contributions appear as volume integrals. To avoid the dense global matrices of conventional BEM, the domain is partitioned into conforming non-overlapping subregions, the boundary integral equation is applied locally on each subdomain, and the assembled algebraic system is

$$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$8

with $$f\in C^1,\qquad f(0)=f(1)=0,\qquad f(u)>0\ (0<u<1),\qquad f(u)\le f'(0)\,u \ (0<u<1).$$9 sparse and non-symmetric (Gortsas, 22 Aug 2025).

In that framework, the Caputo derivative is discretized by

$f(u)=u(1-u)$0

and the fractal-fractional derivative by

$f(u)=u(1-u)$1

The nonlinear term is handled by lagging linearization,

$f(u)=u(1-u)$2

and interior unknowns are eliminated through a local inversion of a small matrix $f(u)=u(1-u)$3, with the inverse computed iteratively using the Henderson–Searle formula. The reported numerical campaign covers six 2D problems, compares LD-BEM with the meshless Fragile Points Method (FPM), finds good agreement with analytical solutions when available, and reports that in one rectangular-domain test the line errors $f(u)=u(1-u)$4 and $f(u)=u(1-u)$5 are on the order of $f(u)=u(1-u)$6 to $f(u)=u(1-u)$7 (Gortsas, 22 Aug 2025).

A second numerical framework is attached to the diffusion-memory model. There the time grid is graded,

$f(u)=u(1-u)$8

and the quadrature weights are

$f(u)=u(1-u)$9

The scheme uses conforming $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$0 finite elements in space and nonuniform convolution quadrature in time. The reported simulations compare the physically consistent model with the conventional Caputo formulation and find that the former shows smoother, more monotone mass evolution, slower, subdiffusive front motion, and more compact fronts, whereas the Caputo model shows overshoots and faster apparent spread. Decreasing $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$1 strengthens memory and subdiffusion: interfaces become thicker, front motion is delayed, interior saturation decreases, and gradients flatten (Fritz et al., 7 Nov 2025).

5. Fractal-time and nonlocal competitive generalizations

A more specialized generalization replaces the ordinary time derivative by the $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$2-derivative in the sense of Parvate–Gangal calculus and combines it with weak diffusion and nonlocal quadratic competition. The resulting one-dimensional equation has the form

$=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$3

where $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$4, $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$5 is a fractal time set, $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$6 is its characteristic function, and $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$7 is the weak diffusion parameter. The loss term is explicitly nonlocal: $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$8 and reduces to a local quadratic loss when $=\lambda\nabla^2\phi-N(\mathbf m,\phi)+f$9 is localized (Shapovalov et al., 19 Mar 2025).

The analytical machinery is semiclassical and based on the Maslov method. The solution is decomposed into quasiparticle components

$$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$0

where each $$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$1 lies in a class of trajectory-concentrated functions localized near a moving trajectory $$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$2. This leads to a system of moment equations, the fractal Einstein–Ehrenfest system, for the masses $$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$3, trajectories $$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$4, and second central moments $$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$5. At second order, the simplest moment law is

$$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$6

The nonlinear PDE is then linked to an associated fractal linear equation and solved asymptotically by a hierarchy

$$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$7

The worked example takes $$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$8, Gaussian initial data, a Gaussian nonlocal kernel, and constant growth $$N(\mathbf m,\phi(\mathbf r,t))=\phi(\mathbf r,t)\big(c-d\,\phi^b(\mathbf r,t)\big), \qquad \mathbf m=[b,c,d]^T\in\mathbb R^3.$$9. In that setting,

$${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$0

The reported qualitative conclusion is that, as $${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$1 decreases, the evolution accelerates: zeroth moments reach quasi-stationarity faster, trajectories separate faster, and two peaks can merge into a more Gaussian-like profile sooner. The case $${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$2 reproduces the standard first-order time-derivative dynamics studied previously. This is a distinct branch of the time-fractional Fisher-KPP literature because the time nonlocality is encoded by the geometry of a fractal set rather than by Caputo or Riemann–Liouville memory kernels (Shapovalov et al., 19 Mar 2025).

6. Distinction from spatially fractional Fisher-KPP equations

A persistent source of confusion is the phrase “fractional Fisher-KPP equation.” Much of the earlier literature uses “fractional” to mean fractional diffusion in space, typically through the fractional Laplacian, not a time-fractional derivative. Representative models include

$${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$3

in periodic media, and

$${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$4

with nonlinear fractional diffusion. These are not time-fractional equations; time remains classical first order (Cabre et al., 2012, Stan et al., 2013).

The qualitative front dynamics are correspondingly different. In the spatially fractional setting, several works prove exponentially fast invasion. For compactly supported or fast decaying initial data, front positions scale like

$${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$5

with exponents such as

$${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$6

depending on the model. One paper proves that there are no nonconstant planar traveling wave solutions of the form $${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$7 for the spatially fractional equation; another proves that, in periodic media, the spreading radius is asymptotically spherical and independent of direction; another proves gradient decay estimates

$${}^{C}D_t^\alpha \phi(\mathbf r,t) = \frac{1}{\Gamma(1-\alpha)} \int_0^t (t-\tau)^{-\alpha}\,\frac{\partial \phi(\mathbf r,\tau)}{\partial \tau}\,d\tau,$$8

leading to quantitative circularization of level sets (Cabre et al., 2012, Cabre et al., 2012, Roquejoffre et al., 2015).

This contrasts sharply with the time-fractional setting. One of the spatial-fractional papers explicitly remarks that time-fractional diffusion usually leads to slower-than-classical propagation because of memory effects, whereas spatial fractional diffusion produces accelerated fronts due to long jumps. The time-fractional Caputo model therefore requires asymptotic traveling waves rather than exact traveling-wave reduction, and the diffusion-memory formulation leads to slower, subdiffusive front motion in numerical experiments. A plausible implication is that the placement of the fractional operator—time versus space, and within time, on the whole evolution law versus only on diffusion—is not a technical detail but a primary determinant of the front-selection mechanism (Roquejoffre et al., 2015, Ishii, 2023, Fritz et al., 7 Nov 2025).