Skip to main content

Dispersion Relation

Module 4: Space and Pattern Formation

NoteReading
  • Compulsory: Britton (2003), Chapter 7, Section 7.3 (pp. 211 to 214): the dispersion relation, potential Turing instability and the mode-\(n\) bifurcation.
  • Recommended: Turing (1952), Section 9 (pp. 54 to 56): the instability \(I(U)\) as a function of the wavelength and the “chemical wave-length” of the fastest-growing wave.

The Turing conditions tell you whether some mode grows. The dispersion relation tells you which modes grow and how fast (Murray 2003, sec. 2.5 and 2.6).

Growth Rate Versus Wavenumber

For each wavenumber \(k\), the growth rates are the roots of

\[ \sigma^2 - \operatorname{tr}A(k^2)\,\sigma + h(k^2) = 0, \qquad A(k^2) = \gamma J - k^2\operatorname{diag}(1, d), \tag{1}\]

with \(h(k^2) = \det A(k^2) = d\,k^4 - \gamma (d f_u + g_v) k^2 + \gamma^2 \det J\), as on the Turing Conditions page. The function \(k \mapsto \operatorname{Re}\sigma(k^2)\), taking the larger root, is the dispersion relation. Britton writes it as \(\rho(\lambda) = \max_i \operatorname{Re}\sigma_i(\lambda)\) with \(\lambda = k^2\), and recalls that the name comes from waves in fluids, where waves of different wavelength travel at different speeds (Britton 2003, sec. 7.3). Turing derived the same function for the continuum as his Equation (9.1), \((p - a + \mu' U)(p - d + \nu' U) = bc\) with \(U = k^2\), and called the real part of its largest root \(I(U)\) (Turing 1952, sec. 9).

The Unstable Band

\(\sigma > 0\) exactly when \(h(k^2) < 0\). The roots of \(h\) give the band

\[ k_1^2 < k^2 < k_2^2, \qquad k_{1,2}^2 = \frac{\gamma}{2d}\left[(d f_u + g_v) \mp \sqrt{(d f_u + g_v)^2 - 4 d \det J}\right]. \tag{2}\]

The band has width proportional to \(\gamma\) in \(k^2\). At \(d = d_c\) it shrinks to the single value \(k_c^2 = \gamma (d_c f_u + g_v)/(2 d_c)\) (Murray 2003, sec. 2.3).

The Fastest-Growing Mode

The fastest-growing wavenumber \(k_{\max}\) maximizes \(\operatorname{Re}\sigma(k^2)\). Near onset it is close to \(k_m\), the minimizer of \(h\), but the two are not equal in general, because \(\sigma\) also depends on the trace. You find \(k_{\max}\) numerically. The corresponding wavelength \(2\pi/k_{\max}\) predicts the spacing of the first visible pattern. Nonlinear effects can change the spacing later (Murray 2003, sec. 2.6). Turing called \(2\pi/k_{\max}\) the chemical wavelength. On a ring it is generally not a divisor of the circumference, and the number of lobes that form is one of the two integers nearest to circumference divided by chemical wavelength, usually the nearest (Turing 1952, sec. 8 and 11). For the numbers below, \(L k_{\max}/\pi = 5.46\), and the fastest Neumann mode is \(n = 5\).

On a finite domain only discrete wavenumbers exist. With Neumann conditions on \((0, L)\), \(k_n = n\pi/L\). The mode that grows fastest is the admissible \(k_n\) with the largest \(\sigma\). If no \(k_n\) falls into the band, the uniform state stays stable even when the Turing conditions hold.

Figure 1: Dispersion relation of Gierer-Meinhardt with \(a = 0.4\), \(b = 1\), \(\gamma = 1\). Dots mark the Neumann modes \(k_n = n\pi/L\) for \(L = 40\).

For \(d = 20\) every mode decays. At \(d = d_c \approx 26.24\) the curve touches zero. For \(d = 30\) a band of modes grows.

import numpy as np

from amlab.turing import dispersion_relation, gierer_meinhardt_jacobian, unstable_band

jac = gierer_meinhardt_jacobian(0.40, 1.00)
d = 30
k1_sq, k2_sq = unstable_band(*jac, d)
k = np.linspace(np.sqrt(k1_sq), np.sqrt(k2_sq), 2001)
k_max = k[np.argmax(dispersion_relation(k**2, *jac, d))]

print(f"unstable band: {np.sqrt(k1_sq):.3f} < k < {np.sqrt(k2_sq):.3f}")
print(f"fastest-growing k = {k_max:.3f}, wavelength = {2 * np.pi / k_max:.1f}")
print(f"closest Neumann mode on L = 40: n = {k_max * 40 / np.pi:.2f}")
unstable band: 0.349 < k < 0.523
fastest-growing k = 0.429, wavelength = 14.6
closest Neumann mode on L = 40: n = 5.46

The fastest wavenumber \(k_{\max} \approx 0.43\) corresponds to a wavelength of about \(14.6\), which lies between the Neumann modes \(n = 5\) and \(n = 6\) on \(L = 40\). Both modes grow, at almost the same rate, and mode \(5\) is slightly faster. The laboratory checks which one the simulation selects.

Explore the instability

The figure below draws the dispersion relation for the current parameters and, next to it, solves the 1D model with explicit Euler from the uniform state plus 1% noise, with Neumann boundaries, exactly as in the lab. Red dots are the Neumann modes \(k_n = n\pi/L\) that grow.

  1. With \(a = 0.4\) and \(d = 30\), let the pattern grow and count its half-wavelengths on \((0, 40)\). Which Neumann mode does the simulation select? Compare with the mode marked \(n\) on the left.
  2. Lower \(d\) to 20 and press Restart from noise. Why does the noise die out? Then try \(d = 27\), just above \(d_c \approx 26.24\): why does the pattern take so much longer to appear?
  3. Keep \(d = 30\) and set \(L = 5\). Which Neumann modes fall inside the shaded band? Does a pattern form, even though \(d > d_c\)? Now try \(L = 80\) and count again.
  4. Change the kinetics: set \(a = 0.2\), then \(a = 0.6\). How do \(d_c\), the width of the band and the speed of growth change?

Build this in Python. In the lab you write the code behind this figure: Unstable Modes in 1D for the dispersion relation and its Neumann modes, and Gierer-Meinhardt Solver in 1D for the explicit Euler simulation that grows the pattern.

Domain Size as the Bifurcation Parameter

Britton treats the size of the domain as the parameter that is varied. A mode \(n\) is unstable only for an interval of \(\gamma\) (equivalently, of \(L\)), and the uniform state can be stable for domains of intermediate size. The closed forms of these intervals, a figure of the mode curves \(k_n = n\pi/L\) crossing the band, and a comparison with simulations are on Domain Size and Mode Selection.

NoteExercise

Keep \(d = 30\) and \(\gamma = 1\) with Neumann conditions. Mode \(n\) is unstable when \(k_1 < n\pi/L < k_2\). For which domain lengths \(L < 20\) is no mode \(n \ge 1\) unstable?

The code above gives \(k_1 \approx 0.349\) and \(k_2 \approx 0.523\). Mode \(n\) is unstable for \(n\pi/k_2 < L < n\pi/k_1\):

  • \(n = 1\): \(6.01 < L < 8.99\),
  • \(n = 2\): \(12.02 < L < 17.99\),
  • \(n = 3\): \(18.03 < L < 26.98\).

No mode is unstable for \(L \le 6.01\), for \(8.99 \le L \le 12.02\), and in the narrow gap \(17.99 \le L \le 18.03\). Beyond that the intervals overlap. Small domains and some intermediate sizes stay uniform even though the Turing conditions hold (Murray 2003, sec. 2.5).

References

Britton, Nicholas F. 2003. Essential Mathematical Biology. Springer Undergraduate Mathematics Series. Springer. https://doi.org/10.1007/978-1-4471-0049-2.
Murray, James D. 2003. Mathematical Biology II: Spatial Models and Biomedical Applications. 3rd ed. Vol. 18. Interdisciplinary Applied Mathematics. Springer. https://doi.org/10.1007/b98869.
Turing, Alan M. 1952. “The Chemical Basis of Morphogenesis.” Philosophical Transactions of the Royal Society B 237 (641): 37–72. https://doi.org/10.1098/rstb.1952.0012.