A full derivation is given in Ref. [44]:f u + g v < 0, (2.19)f u g v − f v g u > 0, (2.20)df u + g v > 0, (2.21)(df u + g v ) 2 − 4d(f u g v − f v g u ) > 0. (2.22)Eqs. (2.19) - 2.20) give the Hopf bifurcation while Eqs. (2.21) - 2.22) give the <strong>Turing</strong>bifurcation. For the LE model the linearized reaction terms are given byf u = ∂f∂u∣ = 3α2 − 5u0 ,v 01 + α 2 , f v = ∂fα∂v ∣ = −4u0 ,v 01 + α2, (2.23)g u = ∂g∂u∣ = σ b 2α2u0 ,v 01 + α 2, g v = ∂gα∂v∣ = −σ bu0 ,v 01 + α 2 (2.24)<strong>and</strong> d = σ c. Inserting Eqs. (2.23 - 2.24) into Eqs. (2.19 - 2.22) yields the conditionsσ bα > 3α 2 − 5, (2.25)σ bα> 0,1 + α2 (2.26)σ bα < (3α 2 − 5)d, (2.27)9d 2 α 4 − 30d 2 α 2 − 26dα 3 σb + 25d 2 − 10dσbα + σ 2 b 2 α 2 > 0. (2.28)Assuming that σ,d,α > 0 <strong>and</strong> considering Eq. (2.27), Eq. (2.28) yields for the <strong>Turing</strong>bifurcationb T = c (13a 2 + 125 − 4 √ √ )10a a5a2 + 25(2.29)Additionally Eq. (2.25) yields for the Hopf bifurcationb H = 1 ( 3σ 5 a − 25 ), (2.30)a<strong>and</strong> the region <strong>of</strong> <strong>Turing</strong> instability is given byb H < b < b T , (2.31)i.e. the region <strong>of</strong> the parameter space, where the homogeneous steady state is stablewithout diffusion but unstable, when diffusion is taken into account. The curvesb H <strong>and</strong> b T are shown in section 4.1 for parameters that resemble the experimentalconditions. Additionally in regions where the homogeneous steady state is unstablewith <strong>and</strong> without diffusion the convergence to heterogeneous steady states is <strong>of</strong>tenfavorable. Only in regions, where b H > b > b T <strong>and</strong> when b T ≈ b H , oscillatory <strong>and</strong>spatio-temporal solutions can occur.24
Linear stability analysis distinguishes convergent from divergent modes, howeveronly the modes with a real part <strong>of</strong> the exponent, which is exactly zero will give alinearly stable st<strong>and</strong>ing wave pattern. However the full nonlinear reaction termsexp<strong>and</strong> the number <strong>of</strong> stable modes <strong>and</strong> bound the divergent modes, so that complexheterogeneous patterns can form in the system from r<strong>and</strong>omly disturbed initialconditions.2.3 Other reaction-diffusion modelsApart form the LE model which is based on the chemical mechanism <strong>of</strong> the CIMAreaction, phenomenological models with simple reaction kinetics for pattern formationin reaction-diffusion systems have been suggested. A well-established model,proposed in Ref. [16] is the so-called Brusselator, which is motivated by simple butunphysical chemical reaction equations with two diffusing <strong>and</strong> reacting intermediatesu <strong>and</strong> v. The corresponding reaction-diffusion system is given by∂u∂t = D u∇ 2 u + a − (b + 1)u + u 2 v , (2.32)∂v∂t = D v∇ 2 v + bu − u 2 v , (2.33)with constants D u , D v , a <strong>and</strong> b. A detailed analysis <strong>of</strong> pattern formation in theBrusselator model can be found in Ref. [48].A generic approach to a reaction-diffusion system leads to the model proposed <strong>and</strong>analyzed in Ref. [6], referred to as the generic Brusselator, where general nonlinearreaction terms are exp<strong>and</strong>ed around a steady state up to third order <strong>and</strong> the coefficientsare conveniently chosen to keep the model simple, but generate a diversity <strong>of</strong>patterns. The specific form <strong>of</strong> the system is:∂u∂t = Dδ∇2 u + α u(1 − r 1 v 2 ) + v (1 − r 2 u), (2.34)∂v∂t = δ∇2 v + v (β + αr 1 uv) + u(γ + r 2 v), (2.35)with constants D, δ, α, β, r 1 <strong>and</strong> r 2 . The generic Brusselator model has alsobeen used to study the effect <strong>of</strong> noise on <strong>Turing</strong> pattern formation [31]. Otherreaction-diffusion systems with pattern formation are the Schnakenberg [57], Gierer-Meinhardt [15], Gray-Scott [49], Thomas [60] <strong>and</strong> Bazykin model 4 [36], which allhave similar phenomenological reaction terms.While the kinetics <strong>of</strong> these models are greatly simplified we show in chapter 5 that theconcentration pr<strong>of</strong>iles <strong>of</strong> similar states are morphologically similar. This indicatesthat the difference between experimental <strong>and</strong> simulated patterns, shown in chapter5 is not due to the specific form <strong>of</strong> the reaction kinetics, but appears to be a moregeneric property <strong>of</strong> deterministic reaction-diffusion <strong>Turing</strong> models.4 The Bazykin model is a modified Lotka-Volterra predator-prey system [64].25