Speed Selection in Coupled Fisher Waves Martin R. Evans SUPA, School of Physics and Astronomy, University of Edinburgh, U.K. September 13, 2013 Collaborators: Juan Venegas-Ortiz, Rosalind Allen
Plan: Speed Selection in Coupled Fisher Waves Plan I Recap of Fisher equation, travelling waves + speed selection principle II Coupled Fisher waves + new speed selection mechanism References: J. Venegas-Ortiz, R. J. Allen and M.R. Evans (2013)
I Fisher-KPP Equation (1d) Population density u(x, t) obeys u = D 2 u + αu(k u) x 2 Fisher (1937) - The wave of advance of advantageous gene Kolmogorov, Petrovsky, Piskunov (1937) Study of the diffusion equation with growth of the quantity of matter and its application to a biological problem. Skellam (1951) Random dispersals in theoretical populations. Basically a nonlinear diffusion equation - nonlinearity is growth with saturation Linear growth rate αk, Carrying capacity K
I Fisher-KPP Equation (1d) Population density u(x, t) obeys u = D 2 u + αu(k u) x 2 Fisher (1937) - The wave of advance of advantageous gene Kolmogorov, Petrovsky, Piskunov (1937) Study of the diffusion equation with growth of the quantity of matter and its application to a biological problem. Skellam (1951) Random dispersals in theoretical populations. Basically a nonlinear diffusion equation - nonlinearity is growth with saturation Linear growth rate αk, Carrying capacity K Appears ubiquitously in: disordered systems such as Directed Polymers in random media; nonequilibrium systems such as directed percolation; computer science search problems etc
Travelling Wave Solutions Population density Front Tip Space Nonlinear wave with finite width travelling at speed v. Stable f.p. solution u = K invades unstable f.p. solution u = 0
Travelling Wave Solutions Population density Front Space Tip Nonlinear wave with finite width travelling at speed v. Stable f.p. solution u = K invades unstable f.p. solution u = 0 Speed selection principle The selected speed of the front is determined by tip dynamics and initial conditions
Physical Examples Ordering Dynamics e.g. Time-dependent Landau-Ginzburg equation m = D 2 m m f (m) f (m) = am2 + bm 4 Contact Process 1 0 rate 1 01 11 rate λ 10 11 rate λ Mean field for density ρ i ρ i ρ = ρ i + λ(1 ρ i )ρ i+1 + λρ i 1 (1 ρ i ) λ 2 ρ + (2λ 1)ρ 2λρ2 x 2 Generally stable phase (fixed point) invades unstable phase (fixed point)
Speed Selection Principle (physics approach) I Simple Linear Analysis: linearise at tip u 1 u = D 2 u x 2 + αuk
Speed Selection Principle (physics approach) I Simple Linear Analysis: linearise at tip u 1 u = D 2 u x 2 + αuk Try wave solution u(x, t) = f (z) where z = x vt vf = Df + αkf Exponential solutions f = e λz where vλ = Dλ 2 + αk. Solve for λ λ = v ± (v 2 4αDK ) 1/2 2D Real if v > v = 2 αdk
Speed Selection Principle (physics approach) I Simple Linear Analysis: linearise at tip u 1 u = D 2 u x 2 + αuk Try wave solution u(x, t) = f (z) where z = x vt vf = Df + αkf Exponential solutions f = e λz where vλ = Dλ 2 + αk. Solve for λ λ = v ± (v 2 4αDK ) 1/2 2D Real if v > v = 2 αdk Speed selection principle The marginal speed v is selected for sufficiently steep initial conditions.
Speed Selection Principle II: More involved Linear Analysis u = D 2 u x 2 + αuk Solve for initial condition u(x, 0) = e λx [ for x 0: ( u(x, t) = (1/2)e λ(x v(λ)t) 1 + erf (x 2Dλt)/(2 )] Dt) where v(λ) = Dλ + αk λ.
Speed Selection Principle II: More involved Linear Analysis u = D 2 u x 2 + αuk Solve for initial condition u(x, 0) = e λx [ for x 0: ( u(x, t) = (1/2)e λ(x v(λ)t) 1 + erf (x 2Dλt)/(2 )] Dt) where Far tip x 2Dλt Use erf(z) 1 for z 1 Thus the tip speed is v(λ) v(λ) = Dλ + αk λ. u(x, t) exp [ λ(x v(λ)t]
Speed Selection Principle II: More involved Linear Analysis u = D 2 u x 2 + αuk Solve for initial condition u(x, 0) = e λx [ for x 0: ( u(x, t) = (1/2)e λ(x v(λ)t) 1 + erf (x 2Dλt)/(2 )] Dt) where Far tip x 2Dλt Use erf(z) 1 for z 1 Thus the tip speed is v(λ) Front x 2Dλt v(λ) = Dλ + αk λ. u(x, t) exp [ λ(x v(λ)t] Use erf( z) 1 + e z2 z π for z 1 u exp [ λ (x v t) (x v t) 2 /(4Dt) ] where v = 2(DαK ) 1/2.
Speed Selection Principle III: Crossover point between far tip and front regimes x = 2Dλt. If λ > λ = v /2D (steep i.c.), then the crossover point advances faster than the tip, and front has speed v. If λ < λ, the front catches up with crossover point which falls behind tip, the speed of the front will be controlled by the tip v(λ).
Speed Selection Principle III: Crossover point between far tip and front regimes x = 2Dλt. If λ > λ = v /2D (steep i.c.), then the crossover point advances faster than the tip, and front has speed v. If λ < λ, the front catches up with crossover point which falls behind tip, the speed of the front will be controlled by the tip v(λ). Speed selection principle If the initial profile decays faster than u(x, 0) e λ x, where λ = v /2D, then wave travels at marginal speed v = v = 2 αdk If the initial profile decays less steeply, say as e λx where λ < λ, the wave advances at faster speed v(λ) = Dλ + αk /λ.
Propagation of coupled Fisher-KPP waves Motivation Horizontal gene transfer within bacterial population More generally acquisation of trait two sub populations N A and N B : those with trait and those with without
Propagation of coupled Fisher-KPP waves Motivation Horizontal gene transfer within bacterial population More generally acquisation of trait two sub populations N A and N B : those with trait and those with without (1) N A = D 2 N A x 2 + αn A(K N T ) βn A + γn A N B, (2) N B = D 2 N B x 2 + αn B(K N T ) + βn A γn A N B N T = N A + N B is the total population density New parameters γ rate of horizontal transmission of trait β loss rate of trait
Propagation of coupled Fisher-KPP waves Motivation Horizontal gene transfer within bacterial population More generally acquisation of trait two sub populations N A and N B : those with trait and those with without (1) N A = D 2 N A x 2 + αn A(K N T ) βn A + γn A N B, (2) N B = D 2 N B x 2 + αn B(K N T ) + βn A γn A N B N T = N A + N B is the total population density New parameters γ rate of horizontal transmission of trait β loss rate of trait Summing (1,2) gives a standard Fisher-KPP equation: (3) N T = D 2 N T x 2 + αn T(K N T ), with speed v T = 2 αkd
Propagation of coupled Fisher-KPP waves N A N T Numerics reveal = D 2 N A x 2 + αn A(K N T ) βn A + γn A N B, = D 2 N T x 2 + αn T(K N T ) Population density v c v tot Space i.e. an invading population wave which does not carry the trait in turn being invaded by a trait wave with different speed
Fixed point Structure N A N T = D 2 N A x 2 + αn A(K N T ) βn A + γn A N B, = D 2 N T x 2 + αn T(K N T ) Choose intermediate transmission rate α > γ > β/k. There are three homogeneous density, steady-state solutions (N T, N A) with dynamical stabilities: The fixed point (N T, N A) = (0, 0) is unstable. (N T, N A) = (K, 0) is a saddle point. (N T, N A) = (K, K β/γ) is stable
Fixed point Structure N A N T = D 2 N A x 2 + αn A(K N T ) βn A + γn A N B, = D 2 N T x 2 + αn T(K N T ) Choose intermediate transmission rate α > γ > β/k. There are three homogeneous density, steady-state solutions (N T, N A) with dynamical stabilities: The fixed point (N T, N A) = (0, 0) is unstable. (N T, N A) = (K, 0) is a saddle point. (N T, N A) = (K, K β/γ) is stable The coupled travelling waves we observe correspond to a homogeneous spatial domain of (0, 0) being invaded by the solution (K, 0) being invaded by the solution (K, K β/γ) (i.e. the trait wave).
Effect of different initial conditions Spread of trait in established Population Initially N T = K throughout the domain (and remains so) 4 x/10000 v s 0 t/100000 1
Effect of different initial conditions Spread of trait in established Population Initially N T = K throughout the domain (and remains so) 4 x/10000 v s 0 t/100000 Spread of trait in expanding population Domain initially empty N T = 0. 4 1 x/10000 v c v s t/100000 0 1
Summary of different speeds I v T = 2 DαK speed of total population wave v s speed of trait wave invading saturated population (N T = K ) v c speed of trait wave invading expandng population
Summary of different speeds I v T = 2 DαK speed of total population wave v s speed of trait wave invading saturated population (N T = K ) v c speed of trait wave invading expandng population also we have v tip speed of far tip of trait wave (N A, N T 1)
Summary of different speeds I v T = 2 DαK speed of total population wave v s speed of trait wave invading saturated population (N T = K ) v c speed of trait wave invading expandng population also we have v tip speed of far tip of trait wave (N A, N T 1) v s : Set N T = K yields F-KPP equation N A = D 2 N A x 2 + γn A ( K β ) γ N A. Thus usual speed selection predicts v s = 2 D(γK β)
Summary of different speeds I v T = 2 DαK speed of total population wave v s speed of trait wave invading saturated population (N T = K ) v c speed of trait wave invading expandng population also we have v tip speed of far tip of trait wave (N A, N T 1) v s : Set N T = K yields F-KPP equation N A = D 2 N A x 2 + γn A ( K β ) γ N A. Thus usual speed selection predicts v s = 2 D(γK β) v tip : Set N T, N A 1 and linearise N A = D 2 N A x 2 + N A (αk β). Thus usual speed selection predicts v tip = 2 D(αK β)
Summary of different speeds II v T = 2 DαK speed of total population wave v s = 2 D(γK β) speed of trait wave invading saturated population (N T = K ) v tip = 2 D(αK β) speed of far tip of trait wave (N A, N T 1)
Summary of different speeds II v T = 2 DαK speed of total population wave v s = 2 D(γK β) speed of trait wave invading saturated population (N T = K ) v tip = 2 D(αK β) speed of far tip of trait wave (N A, N T 1) We show that Speed of trait wave invading expandng population v tip (v T v tip ) v c = v T v T (v T v tip ) 2 + vtip 2 v s 2 N.B. v T > v tip > v c > v s
Determination of v c : the kink in the trait wave Population density 1 0.8 0.6 0.4 0.2 0 (a) N A v c N T v tot 2000 3000 4000 5000 6000 x Population density x 10-52 3 2 1 (c) N A N T v tot z L z R 5100 5105 5110 5115 5120 5125 5130 x Kink at x = v T t Log[Population Density] -401-467 -534-601 -668 (b) N A N T v tip 5500 5550 5600 5650 5700 5750 x Far tip at x >> v T t
Speed selection mechanism for the trait wave Ahead of kink Linearise N A, N B 1 to get N A / = D( 2 N A / x 2 ) + N A (αk β) Then v tip = 2 D(αK β) is selected and defining z R = x v T t we get N A (z R, t) exp[ v tip (z R + (v T v tip )t)/(2d)
Speed selection mechanism for the trait wave Ahead of kink Linearise N A, N B 1 to get N A / = D( 2 N A / x 2 ) + N A (αk β) Then v tip = 2 D(αK β) is selected and defining z R = x v T t we get N A (z R, t) exp[ v tip (z R + (v T v tip )t)/(2d) Behind kink Write N A (z L, t) exp[ at + bz L ], where z L = v T t x and a and b are unknown constants such that v c = v T a/b
Speed selection mechanism for the trait wave Ahead of kink Linearise N A, N B 1 to get N A / = D( 2 N A / x 2 ) + N A (αk β) Then v tip = 2 D(αK β) is selected and defining z R = x v T t we get N A (z R, t) exp[ v tip (z R + (v T v tip )t)/(2d) Behind kink Write N A (z L, t) exp[ at + bz L ], where z L = v T t x and a and b are unknown constants such that v c = v T a/b By demanding that our two asymptotic expressions must match at the kink: i.e. N A (z R = 0, t) = N A (z L = 0, t), we can determine a = v tip (v T v tip )/(2D).
Speed selection mechanism for the trait wave II To find the remaining constant b, we linearize in the region behind the kink, where the total population is large (N T K ), but the amplitude of the trait wave is still small (N A 1). This gives N A / = D( 2 N A / x 2 ) + N A (γk β) But now we need a new selection mechanism for solution Substituting in N A (z L, t) exp[ at + bz L ] with a now determined by matching gives a quadratic for b with root ( ) b = (1/(2D)) v T (v T v tip ) 2 + vtip 2 v s 2. Using v c = v T a/b we finally obtain the following expression for the speed of the trait wave: v tip (v T v tip ) v c = v T v T (v T v tip ) 2 + vtip 2 v s 2
Analytical prediction and simulation results for the speed v c of the trait wave 2 6 wave speed 1.5 1 0.5 wave speed 5 4 3 2 1 0 0.2 0.4 0.6 0.8 1 γ 2 4 6 8 10 K The black lines show the analytical result while the red circles show simulation results for the trait wave speed For comparison the blue lines show the speed of the total population wave v T = 2 αdk while the purple lines show the speed of the trait wave as it invades an established population, v s = 2 D(γK β).
Recap: Spread of trait in expanding population Consider an expanding population starting with the spatial domain initially empty.
Recap: Spread of trait in expanding population Consider an expanding population starting with the spatial domain initially empty. 4 x/10000 v c v s t/100000 0 1 Invasion of Expanding Population Clearly the trait wave advances at speed v c that is greater than the speed v s at which it invades an established population but still less than the speed v T of the total population wave. At very long times, when the total population wave is very far ahead of the trait wave, the speed of the trait wave reverts to v s
Initial-condition dependent speeding-up transition v s Population density 1 0.8 0.6 0.4 0.2 d x(t) 30000 20000 10000 v s v c (a) 0 200 400 600 800 x (b) 20000 40000 60000 80000 t Initially system partially occupied by non-trait population.
Summary Summary Fisher-KPP equation exhibits nonlinear travelling waves We study coupled Fisher waves and find new selection mechanism
Summary Summary Fisher-KPP equation exhibits nonlinear travelling waves We study coupled Fisher waves and find new selection mechanism Outlook Many other generalisations to coupled Fisher waves remain to be studied Important to consider effect of noise