**Overview**

This page details a model governing behavior of fronts in heterogeneous environments, as given in paper *Propagation of fronts in the Fisher-Kolmogorov equation with spatially varying diffusion* by Curtis and Bortz^{[1]}. Paper provides an analytic framework that describes propagation of the front up to the minimum of the diffusion coefficient and shows that standard traveling coordinate frames do not properly describe front propagation.

**Introduction**

Wave travelling in positive direction

Reaction-diffusion systems are systems involving components locally transformed into each other by chemical reactions and transported in space by diffusion. They arise, quite naturally, in chemistry and chemical engineering but also serve as a reference for the study of a wide range of phenomena encountered beyond the strict realm of chemical science such as environmental and life sciences

^{[2]}. Mathematically, reaction–diffusion systems take the form of semi-linear parabolic partial differential equations. They can be represented in the general form

where each component of the vector represents the concentration of one substance, is a diagonal matrix of diffusion coefficients, and accounts for all local reactions. The solutions of reaction–diffusion equations display a wide range of behaviours, including the formation of travelling waves and wave-like phenomena as well as other self-organized patterns like stripes, hexagons or more intricate structure like dissipative solitions. ^{[3]} The choice yields Fisher's equation . Fisher proposed this equation to describe the spatial spread of an advantageous allele and explored its travelling wave, or propagating front, solutions. ^{[4]}

**History**

Front propagation in reaction-diffusion equations (RDEs) is an important topic in the physical and biological sciences.
Interest in the behavior of fronts in heterogeneous environments has increased over the last several years. In particular, traveling waves, or propagating fronts, through certain classes of time and spatially varying environments have been well studied. Nonlinear, density-dependent diffusion, where the diffusion coefficient depends on u, has also been thoroughly examined.
In the biological literature, spatially heterogeneous diffusion coefficients are discussed repeatedly, but continuously varying models are rarely investigated rigorously. A morphogenesis phenomenon in which has been given serious consideration is in a general RDE model for the slug stage of Dictyostelium discoideum. The resulting diffusion profile across a gap junction is a hyperbolic cosine, i.e. . Moreover, recent work has considered the impact of letting (for D constant and η small) in the context of the avascular growth phase of cancer as well as on Turing bifurcations of standing wave solutions to RDEs.

**Mathematical model**

This paper considers the case of a heterogeneous Fisher-Kolmogorov (FK), or Kolmogorov-Petrovskii-Piskunov (KPP), equation where heterogeneities are represented by spatially varying diffusion, i.e.

And .

The form of discussed in this page is . This choice offers a good canonical model for a diffusion-mediated barrier. For example, one could use this mathematical structure to model a geographic barrier such as a mountain. Further, this choice also allows us to study the effect of strongly varying diffusion.

Initial condition taken is the step initial condition

Where denotes the location of the front at time t.

**Analysis from the paper**

**The Soft Front Approximation**

When, we assume that the solution has a *soft *front, i.e., a condition defined via the asymptotic relationship

This asymptotic condition is denoted as the Soft Front Approximation (SFA). Using the SFA, we now must solve the semilinear hyperbolic equation,

Then using method of characteristics, we get the solution

which follows the characteristic curve . In the case that , we see that information propagates towards the origin as desired for a traveling front. However, if , then all information again propagates to the origin, and thus the soft front model cannot describe a front propagating past the origin. Hence, we have a mechanism that explains front propagation that follows a decreasing diffusion coefficient; i.e., a(x) is strictly decreasing when . As can be seen from (1), the front does not propagate as a traveling wave; i.e., there is not a constant wave speed such that . However, it is not hard to see that the front travels along curves of the form

In this log-transformed spatial coordinate we can see the front propagates with speed . We refer to (2) as a traveling wave coordinate (TWC), and we see that the TWC and the SFA allow us to compute a generalized notion of wave speed. They argue that the SFA holds in a region around and this region must be matched, via intermediate layers, to the far field of , which should have a much steeper decay profile. To study this regime, they linearize the FK equation around ; i.e., let and then collect all terms in , so we get

Assuming that is given by the WKB ansatz we get the leading order problem

Using the method of characteristics and noting that above equation is a Hamilton-Jacobi equation, with Hamiltonian , we get that characteristic associated with the choice of diffusion coefficient are given by

Thus we see that for , the characteristics are to leading order given by**Failed to parse (lexing error): x_0∼xe^{\mp 2\tilde H t }**
.

In the FK equation, the minimum of the diffusion coefficient represents a turning point; i.e., the sign of a changes through the minimum. Through the turning point, or when , we get the leading order behavior

The WKB analysis shows how the SFA eventually breaks down and as the front approaches the origin, something like a shock, or steepened front, must form.

**Softening Sharp Initial Contditions**

We assume throughout this section that . The asymptotic condition describing the sharp front is that

We also choose a parameter so that . We choose Neumann boundary conditions to allow for analytical tractability. Introduce the fast time and using the ansatz,

we get the equations

Leading order equation can be solved using separation of variables, which leads to expansion for

where and solve the Sturm-Liouville problem

with initial conditions
Sturm-Liouville theory ensures the functions
form a complete, orthonormal set with respect to norm.
Since we assume that , it easily follows that

Moving to the second term , we see, using Duhamel’s principle, that we can write as

where

Since , one can see that a possibility for a secularity, which means the asymptotic series becomes invalid on time scales, arises from computing

Thus enforcing below condition removes secularity

which has the solution

can also be calculated:

Thus, on timescales , or , one has

From this, since for one has by orthonormality

it follows that the average of to leading order is given by , or

This result shows that the average should increase exponentially fast on time scales of O(1) or less, which amounts to
showing that the transition region of the solution becomes smoother and “softer” rapidly. Having done this, the solution then enters the SFA regime.

**Front Beyond The Turning Point: Reduction To A Stationary Equation**

Since the SFA only leads to a model that describes propagation of a front up to the turning point, we must now find some other means of trying to describe propagation of the front past the turning point. To do this, by taking , so that , we suppose that , which means the FK equation becomes

By choosing

we get the stationary FK equation

with coordinate

The variable is another instance of a TWC, and so again we see that strongly varying diffusion requires a generalization of
the definition of a traveling front.

**Numerical results**

Figures 1 and 2 depict the development of the solution in the (t, x) plane as a surface and as a contour plot, respectively. Figure 3 shows the front before, at, and after the turning point. In Figure 4, we compare the numerical propagation of the front and our asymptotic the two curves represent the function at which , where denotes the numerical approximation to . In the case of the asymptotic curve, we take , and use the same step initial condition as used in the numerics in (4). The agreement between the SFA and numerics is convincing, and thus we have confirmation that the SFA is valid up to the turning point. As can be seen the SFA breaks down at the turning point. From Figures 1, 2, and 3, we see that the front steepens substantially.

In Figure 5, we plot the time duration that the simulated front spends near as a function of , where we define near as . Note the value 0.4 is the width of the spatial mesh step in the numerics, and thus the smallest scale on which phenomena can be distinguished. Figure 5 shows that a least squares fit of the times to a curve growing like is accurate.

On the other side of the turning point, as seen in Figure 2, the solution propagates in a manner consistent with the log-transformed TWC, i.e., information is propagating along curves of the form

**Conclusion**

After diffusion smooths the initial conditions such that the SFA is valid, the choice of a strongly varying diffusion coefficient with a global minimum implies the following

- The definition of a traveling front must be generalized via the Traveling Wave Coordinate
- The minimum, or turning point, of causes the formation of shocklike behavior and leads to a trapping of the front on asymptotically long time scales.
- The behavior of the front on either side of the turning point is fundamentally different, and on either side, the TWC is necessary to describe dynamics.
- After the front is past the turning point, the TWC allows us to transform into a stationary FK equation, thus simplifying all subsequent analysis

**Extensions**

- Quadratic spatial diffusion could be used to model barriers such as mountains, and results will give estimates of a time to barrier transit. For example, the invasion of the Midwestern United States by gypsy moths has been extensively studied
^{[5]}. It is also well known that the moths’ spatial diffusion rate is tightly correlated with the local habitat. By estimating the population growth rates and fitting a quadratic curve to the local diffusion rates from capture-mark-recapture experiments, the method developed in this paper could be used to estimate the waiting time until the moths reach a certain habitat.
- Nontrivial spatially varying FK equation, studied here, could be reduced to a much simpler stationary problem. This will allow for a greater degree of control and flexibility in curve-fitting routines as it will only require a snapshot at a single point in time. With the gypsy moth invasion example, one could envision estimating the wave speed from a single snapshot.

- It would be of particular interest to develop asymptotic matching techniques that would allow for connecting the various regimes of the front, i.e., the sharp to soft front transition, and then the trapping layer around the minimum of . Likewise, it would also be of interest to develop higher order expansions for the speedcin the SFA regime.

**References**

- ↑ C. W. Curtis and D. M. Bortz, Propagation of fronts in the Fisher-Kolmogorov equation with quadratically varying diffusion. Phys. Rev. E, 86(6), 066108, Dec. 2012.
- ↑ http://www.scholarpedia.org/article/Reaction-diffusion_systems
- ↑ http://en.wikipedia.org/wiki/Reaction%E2%80%93diffusion_system
- ↑ Fisher, R. A., The genetical theory of natural selection. Oxford University Press, 1930. Oxford University Press, USA, New Ed edition, 2000
- ↑ P. C. Tobin, S. L. Whitmire, D. M. Johnson, O. N. Bjørnstad, and A. M. Liebhold,Ecol. Lett.10, 36 (2007)