Modeling  III  Linear  Stability  Analysis  

Nondimensional  Equa9ons  From previous lecture, we have a system of nondimensional PDEs:

(22.1)

(22.2)

(22.3)

where here the “*” sign has been dropped for convenience.

Parameters  The parameter values:

Initial conditions:

Boundary conditions: zero-flux on all boundaries.

Homogeneous  Steady  States  

leaving

A homogeneous steady state of a PDE model is a solution that is constant in space in time. For equations (22.1), (22.2), and (22.3) we take

(22.4)

(22.5)

(22.6)

Homogeneous  Steady  States  

Hence, there are four possible steady state points:

Possibilities of steady state points:

(22.7)

Inhomogeneous  Perturba9ons  

•  It is of interest to determine whether or not the steady states are stable.

•  We analyze stability properties by considering the effect of small perturbations.

•  To do this, we must look at spatially non-uniform (also called inhomogeneous) perturbations and explore whether they are amplified or attenuated.

•  If an amplification occurs, then a situation close to the spatially uniform steady state will destabilize, leading to some new state in which spatial variations predominate.

Inhomogeneous  Perturba9ons  We take the distributions of the variables

where are small.

Using the facts that are constants and uniform, the temporal and spatial derivatives give

Inhomogeneous  Perturba9ons  The second-order spatial derivatives give:

For taxis terms: (22.8)

Inhomogeneous  Perturba9ons  The terms

are quadratic in the perturbations or their derivatives and consequently are of smaller magnitude than other terms, thus they can be omitted, leaving

and, similarly

(22.9)

(22.10)

Inhomogeneous  Perturba9ons  And for the reaction terms:

Inhomogeneous  Perturba9ons  Combining all together we rewrite the approximate linearized equations of (22.1) – (22.3) as

which are linear in the quantities

(22.11)

(22.12)

(22.13)

Finding  Eigenvalues  We find eigenvalues by setting the equations (22.11), (22.12), and (22.13) to have no spatial variations, or

Then let

(22.14)

(22.15)

(22.16)

Finding  Eigenvalues  Differentiating with respect to gives us a Jacobian matrix of the reaction terms:

Eigenvalues are obtained by taking (22.17)

(22.18)

Stability  of  the  Steady  States  Steady states in (22.7) are linearly stable if Re λ < 0 since in this case the perturbations go to zero as time goes to infinity.

Using the parameter values given and substituting into (22.18), we obtain the stability of the steady state:

gives 2 λ > 0 and 1 λ < 0 : unstable

gives 1 λ > 0 and 2 λ < 0 : unstable

gives 3 λ < 0 : stable

gives 1 λ > 0 and 2 λ < 0 : unstable

Dispersion  Rela9on  Now we consider the full equations (22.11) – (22.13) and differentiate with respect to the second-order spatial derivatives

to get the transport Jacobian

(22.19)

Dispersion  Rela9on  The linearized system of equations (22.11) – (22.13) can now be represented in a compact form

to be solved in a domain with zero-flux boundary conditions

where

(22.20)

(22.21)

Dispersion  Rela9on  To solve the system of equations in (22.20) subject to the boundary conditions, we first define W(r) to be the time-independent solution of the spatial eigenvalue problem, defined by

where k is the eigenvalue. For example, if the domain is 1D, say , then

where n is an integer. This satisfies zero-flux boundary conditions at x = 0 and x = L.

(22.22)

Dispersion  Rela9on  The eigenvalue in this case is

So

We shall refer to k in this context as the wavenumber.

is a measure of the wavelike pattern: the eigenvalue k is called the wavenumber and 1/k is proportional to the wavelength ω:

Dispersion  Rela9on  With finite domains there is a discrete set of possible wavenumbers since n is an integer.

We now look for solutions of (22.20) in the form

Substituting (22.23) into (22.20) with (22.21) and canceling we get, for each k,

(22.23)

Dispersion  Rela9on  We require nontrivial solutions for Wk so the now the λ are determined by the roots of

Evaluating the determinant with JT and JR we get the eigenvalues λ(k) as functions of the wavenumber k.

(22.24)

Dispersion  Rela9on  Dispersion relation from steady state (0,0,0)

•  Max real part (blue line) of eigenvalues is 0.25

•  It means that the perturbation grows with time.

•  Imaginary part (red line) is zero.

•  With the max real part, there are a range of k where the eigenvalues are positive.

k

λ(k)

Dispersion  Rela9on  Dispersion relation from steady state (0,1,0)

•  Max real part (blue line) of eigenvalues is 0.25.

•  The perturbation grows with time.

•  Imaginary part (red line) is zero.

•  There are a range of k (between k=0 and k⋍30) where the eigenvalues are positive.

k

λ(k)

Dispersion  Rela9on  Dispersion relation from steady state

•  Max real part (blue line) of eigenvalues is -0.0477.

•  The perturbation is damped away.

•  Imaginary part (red line) is zero.

k

λ(k)

Dispersion  Rela9on  Dispersion relation from steady state

•  Max real part (blue line) of eigenvalues is 3.725.

•  Max imaginary part (red line) is 2.1374.

•  The perturbations grows with time.

•  Imaginary part creates oscillating solutions.

k

λ(k)

Simula9on  Results  

References  (1) Mathematical Biology II: Spatial Models and Biomedical

Applications, J.D. Murray, Springer, Third Edition.