From phase to micro-phase separation in flockingmodels
A. Solon, H. Chaté, J. Tailleur
Laboratoire MSCCNRS - Université Paris Diderot
Advances in Nonequilibrium Statistical Mechanics
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 1 / 16
Energy consumption at the microscopic scale Self-propulsion
Aligning interactions
Collective motion (with long range-order?)
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 2 / 16
The Vicsek model [Vicsek et al. PRL 75, 1226 (1995)]
N self-propelled particles off-lattice
Local alignment rule
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 3 / 16
The Vicsek model [Vicsek et al. PRL 75, 1226 (1995)]
N self-propelled particles off-lattice
Local alignment rule
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 3 / 16
The Vicsek model [Vicsek et al. PRL 75, 1226 (1995)]
N self-propelled particles off-lattice
Local alignment rule
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 3 / 16
The Vicsek model [Vicsek et al. PRL 75, 1226 (1995)]
N self-propelled particles off-lattice
Local alignment rule
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 3 / 16
The Vicsek model [Vicsek et al. PRL 75, 1226 (1995)]
N self-propelled particles off-lattice
Local alignment rule
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 3 / 16
Flocking transition [Grégoire, Chaté, PRL (2004)]
Disordered Inhomogeneous Fluctuatingflocking state
noise or density
Non-equilibrium transition to long-range order in d = 2
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 4 / 16
A long-standing debate
Simulations are simple but strong finite size effects2nd-order (1995) vs 1st-order (2004) [Gregoire and Chate, PRL 2004]
hard to study numerically
Analytical descriptions: Boltzmann (Bertin et al.),phenomenological equations (Toner&Tu, Marchetti et al.)
hard to solve analytically
Use a much simpler model: active Ising spinson latticediscrete symmetry
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 5 / 16
A long-standing debate
Simulations are simple but strong finite size effects2nd-order (1995) vs 1st-order (2004) [Gregoire and Chate, PRL 2004]
hard to study numerically
Analytical descriptions: Boltzmann (Bertin et al.),phenomenological equations (Toner&Tu, Marchetti et al.)
hard to solve analytically
Use a much simpler model: active Ising spinson latticediscrete symmetry
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 5 / 16
A long-standing debate
Simulations are simple but strong finite size effects2nd-order (1995) vs 1st-order (2004) [Gregoire and Chate, PRL 2004]
hard to study numerically
Analytical descriptions: Boltzmann (Bertin et al.),phenomenological equations (Toner&Tu, Marchetti et al.)
hard to solve analytically
Use a much simpler model: active Ising spinson latticediscrete symmetry
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 5 / 16
Active Ising model
1 2 3 4 5 . . . L
D(1−ε) D(1+ε) D(1+ε) D(1−ε)
Biased diffusionSpin-flip
i
W−iW+
i
Density ρi = n+i + n−i Magnetisation mi = n+i − n−i
Local alignment W±i = exp(±βmiρi
)
Fully connected Ising models on each site
Self-propulsion Diffusion biased by the spins for ε 6= 0
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 6 / 16
Active Ising model
1 2 3 4 5 . . . L
D(1−ε) D(1+ε) D(1+ε) D(1−ε)
Biased diffusion
Spin-flip
i
W−iW+
i
Density ρi = n+i + n−i Magnetisation mi = n+i − n−i
Local alignment W±i = exp(±βmiρi
)
Fully connected Ising models on each site
Self-propulsion Diffusion biased by the spins for ε 6= 0
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 6 / 16
Active Ising model
1 2 3 4 5 . . . L
D(1−ε) D(1+ε) D(1+ε) D(1−ε)
Biased diffusionSpin-flip
i
W−iW+
i
Density ρi = n+i + n−i Magnetisation mi = n+i − n−i
Local alignment W±i = exp(±βmiρi
)
Fully connected Ising models on each site
Self-propulsion Diffusion biased by the spins for ε 6= 0
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 6 / 16
Phase diagram in 2d
m
Quench ordered •Quench disordered •
0 10 20
0.4
0.6
0.8
1.0T
ρ0
GG+L
Lρ`
ρh
0 100 200 300−0.5
0
0.5
1
1.5
2
2.5
0 100 200 300
0
2
4
6
8
0 100 200 300
0
2
4
6
8
ρ(x)
m(x)
Liquid/gas
Gas Liquid
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 7 / 16
Mean-field and beyond
Mean-field equations 〈f(n±i )〉 ' f(〈n±i 〉)
ρ = D∆ρ−v∂xm v ∝ ε
m = D∆m−v∂xρ+2m(β − 1)− αm3
ρ2
ρ = ρ0 m = 0 always linearly unstable for T < Tc = 1
No clustersContinuous transition
MF only valid at ρ =∞ Refined-Mean-Field-Model (RMFM)
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 8 / 16
Mean-field and beyond
Mean-field equations 〈f(n±i )〉 ' f(〈n±i 〉)
ρ = D∆ρ−v∂xm v ∝ ε
m = D∆m−v∂xρ+2m(β − 1)− αm3
ρ2
ρ = ρ0 m = 0 always linearly unstable for T < Tc = 1
No clustersContinuous transition
MF only valid at ρ =∞ Refined-Mean-Field-Model (RMFM)
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 8 / 16
Mean-field and beyond
Mean-field equations 〈f(n±i )〉 ' f(〈n±i 〉)
ρ = D∆ρ−v∂xm v ∝ ε
m = D∆m−v∂xρ+2m(β − 1)− αm3
ρ2
ρ = ρ0 m = 0 always linearly unstable for T < Tc = 1
No clustersContinuous transition
MF only valid at ρ =∞ Refined-Mean-Field-Model (RMFM)
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 8 / 16
Mean-field and beyond
Mean-field equations 〈f(n±i )〉 ' f(〈n±i 〉)
ρ = D∆ρ−v∂xm v ∝ ε
m = D∆m−v∂xρ+2m(β − 1)− αm3
ρ2
ρ = ρ0 m = 0 always linearly unstable for T < Tc = 1
No clustersContinuous transitionMF only valid at ρ =∞
Refined-Mean-Field-Model (RMFM)
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 8 / 16
Mean-field and beyond
Mean-field equations 〈f(n±i )〉 ' f(〈n±i 〉)
ρ = D∆ρ−v∂xm v ∝ ε
m = D∆m−v∂xρ+2m(β − 1)− αm3
ρ2
Finite density: fluctuations βc = 1 + r/ρ
Continuous transition
MF only valid at ρ =∞
Refined-Mean-Field-Model (RMFM)
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 8 / 16
Mean-field and beyond
Mean-field equations 〈f(n±i )〉 ' f(〈n±i 〉)
ρ = D∆ρ−v∂xm v ∝ ε
m = D∆m−v∂xρ+2m(β−1− r
ρ)− αm
3
ρ2
Finite density: fluctuations βc = 1 + r/ρ
Continuous transition
MF only valid at ρ =∞
Refined-Mean-Field-Model (RMFM)
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 8 / 16
Mean-field and beyond
Mean-field equations 〈f(n±i )〉 ' f(〈n±i 〉)
ρ = D∆ρ−v∂xm v ∝ ε
m = D∆m−v∂xρ+2m(β−1− r
ρ)− αm
3
ρ2
Finite density: fluctuations βc = 1 + r/ρ
Continuous transition
MF only valid at ρ =∞ Refined-Mean-Field-Model (RMFM)
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 8 / 16
Simulations of the RMFM
0 5 10 15
0.6
0.8
1.0T
ρ0
Gas
Liquid
ρ1
ρ2
ρ`
ρh
0 50 100
0
1
2
0 50 100
0
2
4
0 50 100
0
2
4
6m(x)
ρ(x)
Spinodals
Coexistence
G L+G L
Same phenomenology as microscopic model
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 9 / 16
Hysteresis loops
1 2 3 4 5 60
0.2
0.4
0.6
0.8
1 φ
ρ0interface effect
nucleation
spinodaldecomposition
updown
0 200 400 600 800
123456
ρ
x
ρ0 =1.2
ρ0 =2
ρ0 =3
ρ0 =4
1.5 2.0 2.5 3.0 3.5 4.00
0.2
0.4
0.6
0.8
1 φ
ρ0
updown
0 200 400 600 800
2
3
4 ρ
x
ρ0 =1.7
ρ0 =2
ρ0 =2.4
ρ0 =2.8
Micro 2d
RMFM
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 10 / 16
Active spins – summary
New flocking model with discrete symmetry using active spins
Flocking trans. Liquid-gas transition in canonical ensemble
Symmetry of the liquid phase ρc =∞
T
ρ0
G L+G L
Tc, ρcEquilibirum
Liquid-gas
Tc, ρc =∞
Active
Liquid-gas
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 11 / 16
Back to the Vicsek model (with H. Chaté)
Phase diagram: liquid-gas picture seems ok
But phase separation micro-phase separationρ1 ≤ ρ2
AIM
VM
Back to the Vicsek model (with H. Chaté)
Phase diagram: liquid-gas picture seems ok
But phase separation micro-phase separationρ1 ≤ ρ2
AIM
VM
Spinodals Quenches shows different regimes • •
Finite-size scaling of order parameter
Hysteresis:
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 13 / 16
Spinodals Quenches shows different regimes • •Finite-size scaling of order parameter
Hysteresis:
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 13 / 16
Spinodals Quenches shows different regimes • •Finite-size scaling of order parameter
Hysteresis:
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 13 / 16
Hydrodynamics: phase vs micro-phase separationBoth type of propagative solutions exist for generichydrodynamic descriptions [Caussin et al., PRL 2014]
Scalar order parameter (Ising)
∂tρ = −v∂xm
∂tm+ ξm∂xm = D∇2m− λ∂xρ+
[(ρ− ρc)−
m2
P 20 ρ
]m
Vectorial order parameter (Vicsek)
∂tρ = −∇.~m
∂t ~m+ ξ(~m∇).~m = D∇2 ~m− λ∇ρ+
[(ρ− ρc)−
|~m|2
P 20 ρ
]~m
0.8
1.2
1.6
ρ
PDE Ising PDE Vicsek
Hydrodynamics: phase vs micro-phase separationBoth type of propagative solutions exist for generichydrodynamic descriptions [Caussin et al., PRL 2014]Scalar order parameter (Ising)
∂tρ = −v∂xm
∂tm+ ξm∂xm = D∇2m− λ∂xρ+
[(ρ− ρc)−
m2
P 20 ρ
]m
Vectorial order parameter (Vicsek)
∂tρ = −∇.~m
∂t ~m+ ξ(~m∇).~m = D∇2 ~m− λ∇ρ+
[(ρ− ρc)−
|~m|2
P 20 ρ
]~m
0.8
1.2
1.6
ρ
PDE Ising PDE Vicsek
Hydrodynamics: phase vs micro-phase separationBoth type of propagative solutions exist for generichydrodynamic descriptions [Caussin et al., PRL 2014]Scalar order parameter (Ising)
∂tρ = −v∂xm
∂tm+ ξm∂xm = D∇2m− λ∂xρ+
[(ρ− ρc)−
m2
P 20 ρ
]m
Vectorial order parameter (Vicsek)
∂tρ = −∇.~m
∂t ~m+ ξ(~m∇).~m = D∇2 ~m− λ∇ρ+
[(ρ− ρc)−
|~m|2
P 20 ρ
]~m
0.8
1.2
1.6
ρ
PDE Ising PDE Vicsek
Fluctuations play a crucial role
PDEs + noises do a good job: • •
∂tm = [...] + η ∂t ~m = [...] + ~η
t=400 t=106
scalarm
vectorial~m
0
1
2
3ρ
103 104 105 106 107
101
102
103
104 ∆n
nn0.5
n0.8
sSDEvSDEAIMVM
The nature of the phase-separated states stems from theinterplay between fluctuations and symmetry of the orderparameter
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 15 / 16
Fluctuations play a crucial role
PDEs + noises do a good job: • •
∂tm = [...] + η ∂t ~m = [...] + ~η
t=400 t=106
scalarm
vectorial~m
0
1
2
3ρ
103 104 105 106 107
101
102
103
104 ∆n
nn0.5
n0.8
sSDEvSDEAIMVM
The nature of the phase-separated states stems from theinterplay between fluctuations and symmetry of the orderparameter
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 15 / 16
Conclusion
Flocking trans. Liquid-gas transition in canonical ensemble
Symmetry of the liquid phase ρc =∞
Different universality classes:
Ising phase separationVicsek micro-phase separation
Active Ising Model [A. Solon, J.T., PRL 111 078101, (2013)]
Study of Hydrodynamic equations [JB. Caussin, A. Solon, A. Peshkov, H.
Chaté , T. Dauxois, J.T., V. Vitelli, D. Bartolo et al., PRL 112 148102, (2014)]
AIM (follow-up) and Vicsek: hopefully next week on the arxiv !
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 16 / 16
Phase-separated profiles
Propagating shocks between ρ`,m` = 0 and ρh,mh 6= 0
Stationary solutions in comoving frame of velocity c
Dρ′′ + cρ′ − vm′ = 0 (1)
Dm′′ + cm′ − vρ′ + 2m(β − 1− r
ρ)− αm
3
ρ2= 0 (2)
Solvable at large densities ρ1 = rβ−1 � r
1: Solve (1) to get ρ = ρ` + vc
∑∞k=0
(− D
c ∇)km
2: Expand (2) around ρ1, inject ρ(m) and truncate
D(1 + v2
c2)m′′ + [c− v2
c −2Dvrc2ρ21
m]m′−2r(ρ1−ρ`)ρ21
m+ 2rvcρ21m2− αm3
ρ21= 0
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 17 / 16
Phase-separated profiles
Propagating shocks between ρ`,m` = 0 and ρh,mh 6= 0
Stationary solutions in comoving frame of velocity c
Dρ′′ + cρ′ − vm′ = 0 (1)
Dm′′ + cm′ − vρ′ + 2m(β − 1− r
ρ)− αm
3
ρ2= 0 (2)
Solvable at large densities ρ1 = rβ−1 � r
1: Solve (1) to get ρ = ρ` + vc
∑∞k=0
(− D
c ∇)km
2: Expand (2) around ρ1, inject ρ(m) and truncate
D(1 + v2
c2)m′′ + [c− v2
c −2Dvrc2ρ21
m]m′−2r(ρ1−ρ`)ρ21
m+ 2rvcρ21m2− αm3
ρ21= 0
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 17 / 16
Phase-separated profiles
Propagating shocks between ρ`,m` = 0 and ρh,mh 6= 0
Stationary solutions in comoving frame of velocity c
Dρ′′ + cρ′ − vm′ = 0 (1)
Dm′′ + cm′ − vρ′ + 2m(β − 1− r
ρ)− αm
3
ρ2= 0 (2)
Solvable at large densities ρ1 = rβ−1 � r
1: Solve (1) to get ρ = ρ` + vc
∑∞k=0
(− D
c ∇)km
2: Expand (2) around ρ1, inject ρ(m) and truncate
D(1 + v2
c2)m′′ + [c− v2
c −2Dvrc2ρ21
m]m′−2r(ρ1−ρ`)ρ21
m+ 2rvcρ21m2− αm3
ρ21= 0
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 17 / 16
Symmetric front solutions β ' 1
m±(x) =mh
2(tanh(q±(x− ct)) + 1)
c= v mh =4r
3αq± = ± β − 1
3√αD
' ±0.0518
0 100 200 300 4000.2
0.0
0.2
0.4
0.6
0.8
1.0
x
ρ−ρlρh−ρl
q=0.051 q=-0.051
T=0.83
microfit tanh
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 18 / 16
Symmetric front solutions β ' 1
m±(x) =mh
2(tanh(q±(x− ct)) + 1)
c= v mh =4r
3αq± = ± β − 1
3√αD' ±0.0518
0 100 200 300 4000.2
0.0
0.2
0.4
0.6
0.8
1.0
x
ρ−ρlρh−ρl
q=0.051 q=-0.051
T=0.83
microfit tanh
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 18 / 16
Asymmetric front solutions β > 1
m±(x) =mh
2(tanh(q±(x− ct)) + 1)
c = v+2Dr2
3vαρ21q± = ± r
3ρ1√αD− r2
6αvρ21mh =
4r
3α− 8Dr3
9v2α2ρ21
0 100 200 300 4000.2
0.0
0.2
0.4
0.6
0.8
1.0
x
ρ−ρlρh−ρl
q=0.078 q=-0.14
T=0.5
microfit tanh
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 19 / 16
The flock fly faster than the birds
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.00.95
1.00
1.05
1.10
1.15
1.20
1.25 cv
T
micromean-field
0 200 4001
3
5
7
0 200 4007
9
11
13
15
c = v+ 2Dr2
3vαρ21
v microscopic velocities
2Dr2
3vαρ21FKPP-like contribution
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 20 / 16
The flock fly faster than the birds
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.00.95
1.00
1.05
1.10
1.15
1.20
1.25 cv
T
micromean-field
0 200 4001
3
5
7
0 200 4007
9
11
13
15
c = v+ 2Dr2
3vαρ21
v microscopic velocities
2Dr2
3vαρ21FKPP-like contribution
J. Tailleur (CNRS-Univ Paris Diderot) GGI-28/05/2014 20 / 16