Towards metastability in a model of population dynamics with competition
Loren Coquille (HCM Bonn) Joint work with A. Bovier
Essen — June 26, 2014
Outline
1 The model
2 Law of large numbers (large population)
3 Large population and rare mutations Deterministic limit
Probabilistic limit
4 Towards metastability
The model
Outline
1 The model
2 Law of large numbers (large population)
3 Large population and rare mutations Deterministic limit
Probabilistic limit
4 Towards metastability
The model
The model
Trait space ={0,1, . . . ,L}
Number of individuals of trait i =Xi(t)
Generator of the processX = (X0(t), . . . ,XL(t))∈NL+1 : Lf(X) =
L
X
i=0
(f(X+ei)−f(X))·bi(1−ε)Xi clonal birth
+
L
X
i=0
(f(X−ei)−f(X))·(di +
L
X
j=1
cijXj)Xi natural death and competition
+
L
X
i=0
X
j∼i
(f(X+ej)−f(X))·Xiεbi/2 mutation where ei = (0, . . . ,1, . . . ,0).
Law of large numbers (large population)
Outline
1 The model
2 Law of large numbers (large population)
3 Large population and rare mutations Deterministic limit
Probabilistic limit
4 Towards metastability
Law of large numbers (large population)
Law of large numbers
[Fournier, Méléard, 2004]The rescaled process XK = K1(X0(t), . . . ,XL(t))∈(K1N)L+1 with generator
Lf(X) =
L
X
i=0
(f(X +ei
K)−f(X))·bi(1−ε)KXi +
L
X
i=0
(f(X −ei
K)−f(X))·(di +
L
X
j=1
cij
KKXj)KXi +
L
X
i=0
X
j∼i
(f(X+ ej
K)−f(X))·KXiεbi/2, bounded parameters and convergence of the initial condition,
Law of large numbers (large population)
Law of large numbers
[Fournier, Méléard, 2004]converges in law, asK → ∞, towards the solution of the non-linear system of differential equations :
dxiε dt =
(1−2ε)bi −di−
L
X
j=0
cijxjε
xiε+ε
X
j∼i
xjεbj 2
, i =0, . . . ,L
Canonical vocabulary:
Monomorphic equilibrium : x¯i = (bi −di)/cii Invasion fitness : fij =bi−di−cijx¯j
Law of large numbers (large population)
Law of large numbers
[Fournier, Méléard, 2004]converges in law, asK → ∞, towards the solution of the non-linear system of differential equations :
dxiε dt =
(1−2ε)bi −di−
L
X
j=0
cijxjε
xiε+ε
X
j∼i
xjεbj 2
, i =0, . . . ,L
Canonical vocabulary:
Monomorphic equilibrium : x¯i = (bi −di)/cii Invasion fitness : fij =bi−di−cijx¯j
Large population and rare mutations
Outline
1 The model
2 Law of large numbers (large population)
3 Large population and rare mutations Deterministic limit
Probabilistic limit
4 Towards metastability
Large population and rare mutations Deterministic limit
Deterministic limit K → ∞ followed by ε → 0
Theorem
Start with initial condition : xε(0) = ¯x0e0.
If the invasion fitnesses fi0 andfiL satisfy : æ æ
æ æ
æ
à à
à à
ì ì ì ì ìà
then the sequence of rescaled deterministic processes
x0ε(t log(1/ε)), . . . ,xLε(t log(1/ε))
t>0
converges, as ε→0, towards the process x(t) =
¯x0e0 pour06t6L/fL0
¯
xLeL pour t>L/fL0
Large population and rare mutations Deterministic limit
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
1 2 3 4 5 6 7
<——————>
∼ fL
L0 log(1/ε)
Large population and rare mutations Deterministic limit
Idea of the proof : look at the logarithmic scale
log(O(1))
log(O(ε))
log(O(ε2))
log(O(ε3))
log(O(ε4))
0.2 0.4 0.6 0.8 1.0 1.2 1.4
-80 -60 -40 -20
<——> <—–>
T1 T2
Large population and rare mutations Deterministic limit
During time T
1The process is close to the solution of the linear system
dy dt =
0 0 0 0 0 0
εb0
2 f10 0 0 0 0
0 εb21 f20 0 0 0
0 0 . .. ... 0 0
0 0 0 εbL−22 fL−1,0 0
0 0 0 0 εbL−12 fL0
y
with initial condition y(0) = (¯x0,0, . . . ,0).
Indeed, d
dt(xiε−yi) =fi0(xiε−yi) +O(εxi+1ε ) +Error is negative until time t such that xL−1 feels the presence of xL:
−O(εL−1) +O(εL+1efL0t)>0 ⇒ T1= 2
fL0 log(1/ε)(1+o(1))
Large population and rare mutations Deterministic limit
During time T
1The process is close to the solution of the linear system
dy dt =
0 0 0 0 0 0
εb0
2 f10 0 0 0 0
0 εb21 f20 0 0 0
0 0 . .. ... 0 0
0 0 0 εbL−22 fL−1,0 0
0 0 0 0 εbL−12 fL0
y
with initial condition y(0) = (¯x0,0, . . . ,0). Indeed, d
dt(xiε−yi) =fi0(xiε−yi) +O(εxi+1ε ) +Error is negative until time t such that xL−1 feels the presence of xL:
−O(εL−1) +O(εL+1efL0t)>0 ⇒ T1= 2
fL0 log(1/ε)(1+o(1))
Large population and rare mutations Deterministic limit
During time T
2The process is close to the solution of the linear system
dy dt =
0 0 0 0
εb0
2 f10 0 0
0 . .. ... 0
0 0 εbL−22 fL−2,0
0
0 fL−1,0 εb2L
0 fL0
y
with initial condition y(0) = (¯x0,O(ε), . . . ,O(εL−2)).
Indeed, dtd(xiε−yi) is negative until time t such that xL−2 feels the presence of xL−1:
T2= 2
fL0 log(1/ε)(1+o(1))
Large population and rare mutations Deterministic limit
During time T
2The process is close to the solution of the linear system
dy dt =
0 0 0 0
εb0
2 f10 0 0
0 . .. ... 0
0 0 εbL−22 fL−2,0
0
0 fL−1,0 εb2L
0 fL0
y
with initial condition y(0) = (¯x0,O(ε), . . . ,O(εL−2)).
Indeed, dtd(xiε−yi) is negative until time t such that xL−2 feels the presence of xL−1:
T2= 2
fL0 log(1/ε)(1+o(1))
Large population and rare mutations Deterministic limit
Step by step towards the swap
If L>5, we continue to compare the system with :
dy dt =
0 0 0 0 0
εb0
2 f10 0 0 0
0 εb21 f20 0 0
0 0 . .. . .. 0
0 0 0 εbk−22 fk−1,0
0
0
fk,0 εbk+12 0 0
0 . .. . .. 0
0 0 fL−1,0 εb2L
0 0 0 fL0
y
If Lis even, the time to reach the swap is L
fL0log(1/ε)(1+o(1)).
Large population and rare mutations Deterministic limit
The swap : Lotka-Volterra system with 2 traits
The system is now close to the solution of:
dy0/dt = (b0−d0−c00y0−c0LyL)y0 dyL/dt = (bL−dL−c0Ly0−cLLyL)yL dyi/dt =0, ∀0<i <L
with initial condition
y0(0) = ¯x0 yL(0) =η >0
yi(0) =O(εmin{L−i,i}), ∀0<i <L
unique stable equilibrium(0, . . . ,0,x¯L)
time to enter an η−neighborhood of this equilibrium isO(1).
Large population and rare mutations Deterministic limit
The swap : Lotka-Volterra system with 2 traits
The system is now close to the solution of:
dy0/dt = (b0−d0−c00y0−c0LyL)y0 dyL/dt = (bL−dL−c0Ly0−cLLyL)yL dyi/dt =0, ∀0<i <L
with initial condition
y0(0) = ¯x0 yL(0) =η >0
yi(0) =O(εmin{L−i,i}), ∀0<i <L unique stable equilibrium(0, . . . ,0,x¯L)
time to enter an η−neighborhood of this equilibrium isO(1).
Large population and rare mutations Deterministic limit
Step by step towards equilibrium
We continue to compare the system with the linear two blocs system:
dy dt =
f0L 0 0 0 0
εb0
2 f1L 0 0 0
0 εb21 f2L 0 0
0 0 . .. . .. 0
0 0 0 εbL−k−22 fL−k−1,L
0
0
fL−k,L
εbL−k+1
2 0 0
0 . .. . .. 0
0 0 fL−1,L εbL
2
0 0 0 0
y
The system reaches equilibrium after a time L
f0L log(1/ε)(1+o(1)).
Large population and rare mutations Probabilistic limit
Probabilistic limit (K , ε) → (∞, 0) with ε
LK 1
Theorem
Assume same hypothesis on the fitness landscape. Consider the process (XtK)t>0 with initial condition X0K = NK0Ke0 such that NK0K law→ ¯x0 >0as
(K, ε)→(∞,0) with εLK 1 Then we have
K→∞lim XtKlog(1/ε)(d)=
¯x0e0 for06t 6L/fL0
¯xLeL for t >L/fL0 under the total variation norm.
Towards metastability
Outline
1 The model
2 Law of large numbers (large population)
3 Large population and rare mutations Deterministic limit
Probabilistic limit
4 Towards metastability
Towards metastability
Towards metastability
Natural questions :
study the limit(K, ε)→(∞,0)on a non-trivial scale get a random time of swap
exponentially distributed ?
what are the strategies used by the system to swap at a given time ?
There is no literature about metastability in models with competition.
!!! non-reversibility !!!
Towards metastability
Thank you !