Towards Global Existence and Optimal Equilibriation Rates ... · Towards Global Existence and...
Transcript of Towards Global Existence and Optimal Equilibriation Rates ... · Towards Global Existence and...
Towards Global Existence andOptimal Equilibriation Rates for
Reaction-Diffusion Models
Klemens Fellner
Institute of Mathematics and Scientific Computing,
University of Graz
with J. A. Canizo, L. Desvillettes, E. Latos, S. Rosenberger, Bao Q. Tang
Banff 02.07.2014 – p. 1/25
OverviewTowards bright sun and white snow for RD models
Single Reversible Reaction: Quadratic RD system
Entropy-Dissipation structure
Global existence (classical vs. weak)
Convergence to equilibrium (optimal?)
Networks of Reversible Reactions
Motivation
Entropy-Dissipation structure?
Quasi-Steady-State-Approximation
Volume-Surface RD models
Banff 02.07.2014 – p. 2/25
Reaction-Diffusion SystemsPrototypical quadratic system: A1 +A2 ↔ A3 +A4
One reversible reaction of four species Ai
A1 +A2 ↔ A3 +A4
mass action law kinetics: reaction rates ∼ a1a2 − a3a4
bounded, smooth domain Ω ⊂ RN , |Ω| = 1
homogeneous Neumann boundary conditions
Banff 02.07.2014 – p. 3/25
Reaction-Diffusion SystemsPrototypical quadratic system: A1 +A2 ↔ A3 +A4
Concentrations ai(t, x) of Ai, different diffusivities di > 0
∂ta1 − d1xa1 = −a1 a2 + a3 a4
∂ta2 − d2xa2 = −a1 a2 + a3 a4
∂ta3 − d3xa3 = +a1 a2 − a3 a4
∂ta4 − d4xa4 = +a1 a2 − a3 a4
quadratic non-linearities, Bootstrap?
no comparison principle, no invariant regions,Turing instability?
NO, there is an entropy functional!
Banff 02.07.2014 – p. 3/25
Entropy and Entropy DissipationPrototypical quadratic model: A1 +A2 ↔ A3 +A4
Kinetic (free energy) entropy
H(ai) =
∫
Ω
4∑
i=1
(ai ln(ai)− ai) dx
Entropy dissipation ddt
H = −D ≤ 0
D(ai) = 44
∑
i=1
∫
Ω
di|∇√
ai|2 dx +
∫
Ω
(a1 a2 − a3 a4) lna1 a2
a3 a4
dx ≥ 0
Programme
Global (classical, weak, renormalised) solutions
Large-time behaviour: explicit exponential convergence
Banff 02.07.2014 – p. 4/25
Entropy and Entropy DissipationPrototypical quadratic model: A1 +A2 ↔ A3 +A4
Equilibrium state ai∞i=1..4 is the unique vector of positiveconstants balancing the reaction rate
a1∞ a3∞ = a2∞ a4∞,
and satisfying the three (linear indep.) mass-conservationlaws (homogeneous Neumann boundary conditions)
a1∞ + a2∞ =1
|Ω|
∫
Ω
(a10 + a20) dx,
a1∞ + a4∞ =1
|Ω|
∫
Ω
(a10 + a40) dx,
a2∞ + a3∞ =1
|Ω|
∫
Ω
(a20 + a30) dx.
Banff 02.07.2014 – p. 5/25
The Entropy MethodQuantitative large-time behaviour
E(f) non-increasing convex entropy functional
D(f) entropy dissipation, f∞ entropy minimising equilibrium
d
dtE(f) =
d
dt(E(f)− E(f∞)) = −D(f) ≤ 0
provided conservation laws: D(f) = 0⇐⇒ f = f∞
D ≥ Φ(E(f)− E(f∞)), Φ(0) = 0, Φ ≥ 0
⇒ explicit convergence in entropy, exponential if Φ′(0) > 0
⇒ convergence in L1 : ‖f − f∞‖21 ≤ C(E(f)− E(f∞))
Cziszár-Kullback-Pinsker inequalities of convex entropies
Banff 02.07.2014 – p. 6/25
The Entropy MethodMy Personal Entropy Method Dictionary
Understanding Entropy-Dissipation Structure ⇐⇒Entropy Entropy-Dissipation (EDD) Estimate
D ≥ Φ(E(f)− E(f∞)), Φ(0) = 0, Φ ≥ 0
Really Understanding ED Structure ⇐⇒ OptimalRate/Constant in EED Estimate
Gradient flow structure?
Banff 02.07.2014 – p. 7/25
Entropy Entropy-Dissipation EstimateD ≥ C (E − E∞)
Theorem:a For any functions ai, i = 1, 2, 3, 4 measurable,non-negative, satisfying the conservation laws holds
D(ai) ≥ C(Mij)(E(ai|ai,∞)).
Proof: Additivity E(ai|ai,∞) = E(ai|ai) + E(ai|ai,∞)
E(ai|ai) =4
∑
i=1
∫
Ω
ai ln
(
ai
ai
)
dx ≤ L(Ω)4
∑
i=1
∫
Ω
|∇x
√ai|2 dx ,
+ Long long calculations + Conservation laws!
⇒ obtain Functional Inequality Not sharp!a [L. Desvillettes, K.F.]
Banff 02.07.2014 – p. 8/25
Entropy Entropy-Dissipation EstimateD ≥ C (E − E∞)
Entopry Entropy-Dissipation Estimate
D(ai) ≥ C(Mij)(E(ai|ai,∞))
+ Gronwall argument + Cziszár-Kullback inequality⇒ explicit exponential convergence to equilibrium in L1.
As long a solutions exists?
Banff 02.07.2014 – p. 9/25
Entropy A-priori EstimatesEntropy decay
The entropy decays
H(T ) = H(0)−∫ T
0
D(s) ds
with the entropy dissipation
D(ai) = 44
∑
i=1
∫
Ω
di|∇√
ai|2 dx +
∫
Ω
(a1 a2 − a3 a4) lna1 a2
a3 a4
dx
Banff 02.07.2014 – p. 10/25
Entropy A-priori EstimatesEntropy decay
H(T ) :
ai ∈ L∞([0,+∞);L log L(Ω)) ∀i = 1, .., 4∫ T
0
D(s) : √
ai ∈ L2([0,+∞);H1(Ω)) ∀i = 1, .., 4 : di > 0
in 1D: ‖ai‖3−εL3−ε([0,T ]×[0,1]) ≤ C(1 + T ) + parabolic bootstrap
⇒ ‖ai‖L∞([0,T ]×[0,1]) ≤ C(
1 + T212
)
⇒ global classical solutions
in 2D: a2i ≤ ai e
sai‖√ai(t)‖2H1(Ω) +
ai‖√
ai(t)‖2H1(Ω)
sln(a2
i ) for ln(ai) > 1
Trudinger ineq. ⇒ global L2 bound: ‖ai‖2L2([0,T ]×Ω) ≤ C(1 + T )
⇒ global weak (super-)solutions [M. Pierre 2003]a
a[Desvillettes F. Pierre Vovelle]
Banff 02.07.2014 – p. 10/25
Entropy A-priori EstimatesEntropy decay
H(T ) :
ai ∈ L∞([0,+∞);L log L(Ω)) ∀i = 1, .., 4
∫ T
0
D(s) :
√ai ∈ L2([0,+∞);H1(Ω)) ∀i = 1, .., 4 : di > 0
∫ T
0
∫
Ω(a1 a2 − a3 a4) ln
(
a1 a2
a3 a4
)
dxdt ≤ C
in 3+D: ‖ai‖1+2/N
L1+2/N ([0,T ]×Ω)≤ C(1 + T )
⇒ renormalised solutions (in the sense of [DiPerna, Lions]) a
a[Desvillettes F. Pierre Vovelle]
Banff 02.07.2014 – p. 10/25
Duality Argument for Entropy DensityEntropy density equation: A1 +A2 ↔ A3 +A4
Denote zi = ai ln(ai)− ai ⇒ H(ai) =∫
Ω
∑4i=1 zi dx
∂t
(∑4
i=1 zi
)
−∆x
(∑4
i=1 di zi
)
≤ 0, t ∈ [0, T ], x ∈ Ω,
n · ∇xzi = 0, t ∈ [0, T ], x ∈ ∂Ω,
Banff 02.07.2014 – p. 11/25
Duality Argument for Entropy DensityEntropy density equation: A1 +A2 ↔ A3 +A4
Denote zi = ai ln(ai)− ai ⇒ H(ai) =∫
Ω
∑4i=1 zi dx
∂t
(∑4
i=1 zi
)
−∆x
(∑4
i=1 di zi
)
≤ 0, t ∈ [0, T ], x ∈ Ω,
n · ∇xzi = 0, t ∈ [0, T ], x ∈ ∂Ω,
rewrite with z :=∑4
i=1 zi and M(t, x) :=P4
i=1 di zi
zas
∂tz −∆x[M z] ≤ 0, t ∈ [0, T ], x ∈ Ω,
n · ∇x[M z] = 0, t ∈ [0, T ], x ∈ ∂Ω,
coefficient M bounded by mindi ≤M(t, x) ≤ maxdi!
Banff 02.07.2014 – p. 11/25
Duality Argument for Entropy DensityEntropy density equation: A1 +A2 ↔ A3 +A4
Simplest case: 0 < d0 ≤ mindi ≤M(t, x) ≤ maxdi <∞
∂t
(
∑ki=1 zi
)
−∆x
(
∑ki=1 di zi
)
≤ 0, t ∈ [0, T ], x ∈ Ω,
n · ∇xzi = 0, t ∈ [0, T ], x ∈ ∂Ω,
uniform L2(log L)2 bound + quadratic non-linearities
⇒ uniform integrability of non-linearities
⇒ convergence in L1(QT ) of approximating sequence
⇒ global L2-weak solutions in all space dimensions!
Banff 02.07.2014 – p. 11/25
Duality Argument for Entropy DensityGeneral systems ∂tai −∇x ·(di∇xai) = fi(a)
Quadratic Lotka-Volterra systems in RN : For z ∈ (0,∞)q
∂tai = di∆xai + ai
∑qj=1 pij(aj − zj), i = 1 . . . q
∇xai ·n = 0 on ∂Ω, ai(0, ·) = ai,0(·) ∈ L2(Ω),
Then, there exists a global weak solution in L2(Ω) in RN .
Banff 02.07.2014 – p. 11/25
Improved Duality ArgumentAn improved duality argument
Let Ω ⊂ RN be a bounded domain with ∂Ω ∈ C2+α. Let T > 0,
∂tu−∆x(M(t, x)u) = 0 on ΩT ,
u(0, x) = u0(x) ∈ Lp(Ω) for x ∈ Ω,
∇xu · ν(x) = 0 on [0, T ]× ∂Ω,
with 0 < a ≤M(t, x) ≤ b < +∞ for (t, x) ∈ ΩT .
Then, any weak solution u satisfies (1/p + 1/p′ = 1)
‖u‖Lp(ΩT ) ≤ (1 + bDa,b,p′) T 1/p ‖u0‖Lp(Ω), p ∈ (2,+∞),
where Da,b,p′ :=C a+b
2 ,p′
1−C a+b2 ,p′
b−a2
as long as C a+b2
,p′b−a2
< 1. a
a[Canizo Desvillettes F., CPDE]Banff 02.07.2014 – p. 12/25
SummaryResults A1 +A2 ↔ A3 +A4
1D: global classical solutions based on entropy structure≈ alternative to theory of [Amann],... (more general)explicit exponential decay (rates) in all Sobolev norms.
2D: global classical solutions [Goudon, Vasseur] [CañizoDesvillettes F.]explicit exponential decay (rates) in L2.
allD: global weak L2-solutionsexplicit exponential decay (rates) in Lp, 1 ≤ p < 2.Blow-up example [M. Pierre D. Schmidt]
Banff 02.07.2014 – p. 13/25
Towards optimal constants/rates?Nonlinear mass action law system αA ↔ βB
∂ta− daa =
∂tb− dbb =
−α α
β −β
·
l aα
k bβ
Linearisation around equilibrium u = a− a∞ and v = b− b∞
⇒ Rescaled linearised system
∂tu− dau =
∂tv − dbv =
−α2 αβ
αβ −β2
·
u
v
Fourier expansion u =∑∞
k=0 uk(t)ϕk, v =∑∞
k=0 vk(t)ϕk
∆ϕk = λkϕk in Ω
n · ∇ϕk = 0 on ∂Ωk = 0, 1, . . .
Banff 02.07.2014 – p. 14/25
Towards optimal constants/rates?Nonlinear mass action law system αA ↔ βB
For any eigenmode k ∈ N, we have
∂t
(
uk
vk
)
=
daλk − α2 αβ
αβ dbλk − β2
·(
uk
vk
)
, k = 0, 1, . . .
a pair of eigenvalues µi, i = 1, 2 and eigenvectors ei, i = 1, 2:
µ1(0) = 0, µ2(0) = −(α2 + β2) < 0,
µ1(k) = da+db
2λk − α2+β2
2+
√
(λk(db−da)+α2−β2)2
4+ α2β2 < 0,
µ2(k) = da+db
2λk − α2+β2
2−
√
(λk(db−da)+α2−β2)2
4+ α2β2 < 0,
We have µ2(k) < µ1(k) and µ1(k + 1) < µ1(k) for all k ∈ N.Banff 02.07.2014 – p. 14/25
Towards optimal constants/rates?Nonlinear mass action law system αA ↔ βB
Two dominant negative eigenvalues
µ2(0) = −(α2 + β2) < 0,
µ1(1) =da + db
2λk−
α2 + β2
2+
√
(λk(db − da) + α2 − β2)2
4+ α2β2,
Special case: d1 = d2 = d ⇒ µ1(1) = dλk.
Optimal rate of convergence depends on
|µ1(1)| > |µ2(0)| ⇔ π2 >α2
d2+
β2
d1
Geometry + stoichiometric coefficients + diffusion rates
Banff 02.07.2014 – p. 14/25
Towards optimal constants/rates?Nonlinear mass action law system αA ↔ βB
Optimal constant in Entropy Entropy-Dissipation Estimatea
I = minβu+αv=0,u,v∈C∞
da
∫
Ω|∇u|2 + db
∫
Ω|∇v|2dx +
∫
Ω(αu− βv)2dx
∫
Ωu2 + v2
= minc0=0
∑∞k=0 |µ1(k)|(ck)2 + |µ2(k)|(dk)2
∑∞k=0(c
k)2 + (dk)2
≥ min|µ1(1)|, |µ2(0)|
Minimising functions are ∼ ϕ0 (“large diffusion”) and ∼ ϕ1
(“small diffusion”)!?
Conjecture: nonlinear constants/minimisers are the same!
a[G. Pissante, E. Latos, K. F.]
Banff 02.07.2014 – p. 14/25
OverviewTowards bright sun and white snow for RD models
Single Reversible Reaction: Quadratic RD system
Entropy-Dissipation Structure
Global existence (classical vs. weak)
Convergence to equilibrium (optimal?)
Networks of Reversible Reactions
Motivation
Entropy-Dissipation Structure? (beyond detailedbalance)
Quasi-Steady-State-Approximation
Volume-Surface RD models
Banff 02.07.2014 – p. 15/25
MotivationProtein-localisation before asymmetric stem-cell division
Asymmetric stem-cell division:
Cell-diversity by localisation of cell-fate determinants into oneside of the cell cortex and into one of two daughter cells.a
aGFP-Pon in SOP precursor cells in living Drosophila larvae [Meyer, Emery,
Berdnik, Wirtz-Peitz, Knoblich, Current Biology, 2005]
Banff 02.07.2014 – p. 16/25
MotivationProtein-localisation before asymmetric stem-cell division
Mathematical model:
“high” concentrations, insignificant stochastic effects
system of (reversible) reaction-diffusion equations
volume(cytoplasm)-surface(membran) dynamics
Banff 02.07.2014 – p. 16/25
Mixed Volume-Surface Reaction-Diffusion SystemsModel Assumptions and Quantities
Key protein Lgl in cell cortex and cytoplasm.Kinase aPKC phosphorylates Lgl on a part Γ2 of cell cortex.
L(t, x): cytoplasmic concentration of Lgll(t, x): cortical concentration of Lglp(t, x): cortical phosphorylated LglP (t, x): cytoplasmic phosphorylated Lgl
L(t, x)α←−−−β
P (t, x)
λ
y
γ ξ|Γ2
x
l(t, x)σ|Γ2−−−→ p(t, x)
Banff 02.07.2014 – p. 17/25
Mixed Volume-Surface Reaction-Diffusion SystemsA prototypical model I
Volume equations with diffusion coefficients dL, dP > 0
(V)
Lt − dL∆L = αP − βL, x ∈ Ω, t > 0,
Pt − dP ∆P = −αP + βL, x ∈ Ω, t > 0,
L(0, x) = L0(x), P (0, x) = P0(x), x ∈ Ω
Boundary conditions on ∂Ω = Γ =Γ1 ∪ Γ2 and Γ1 ∩ Γ2 = ∅
(BC)
dL∂L∂ν
= γl − λL, x ∈ Γ, t > 0,
dP∂P∂ν
= 0, x ∈ Γ1, t > 0,
dP∂P∂ν
= ξp, x ∈ Γ2, t > 0,
Reaction rates α, β, γ, λ, σ, ξ are positive constantsBanff 02.07.2014 – p. 17/25
Mixed Volume-Surface Reaction-Diffusion SystemsA prototypical model II
Boundary dynamics
(BD)
lt − dl∆Γl = λL− γl − σχΓ2l, x ∈ Γ, t > 0
pt − dp∆Γ2p = σl − ξp, x ∈ Γ2, t > 0,
dp∂p
∂νΓ2= 0, x ∈ ∂Γ2,
l(0, x) = l0(x), x ∈ Γ,
p(0, x) = p0(x), x ∈ Γ2,
∆ is the usual Laplacian in the domain Ω
∆Γ and ∆Γ2 are Laplace-Beltrami operator on Γ and Γ2
χΓ2 is the characteristic function of Γ2
Banff 02.07.2014 – p. 17/25
Mixed Volume-Surface Reaction-Diffusion SystemsLocal well-posedness
Conservation law:
d
dt
[
∫
Ω
(L(t, x) + P (t, x)) +
∫
Γ
l(t, x) +
∫
Γ2
p(t, x)]
= 0
For a T > 0 there exists a unique weak solution L,P, l, p on(0, T ), which remains nonnegative if the intital data are so.
Proof: [Bao Quoc Tang, S. Rosenberger, K. F.]
fix-point argument between (V),(BC) and (BD).
local weak solutions of (V),(BC) and (BD)
non-negativity of solutions of (V),(BC) and (BD)
uniqueness
Banff 02.07.2014 – p. 17/25
Mixed Volume-Surface Reaction-Diffusion SystemsGlobal existence
The unique solution (L,P, l, p) exists globally.
Proof: H(t) = 12
(
‖L(t)‖2Ω + ‖P (t)‖2Ω + ‖l(t)‖2Γ + ‖p(t)‖2Γ2
)
dH
dt≤ ηH
where η = max
dL + α+β2
, dP + α+β2
, CP (λ+γ)2
4dL+ σ
2, CP ξ2
4dP+ σ
2
Convergence to equilibrium for all initial data andparameter?
Problem: Decaying entropy functional?
Banff 02.07.2014 – p. 17/25
Simple 1D modelEquilibrium: Asymmetric Protein Localisation
1D Cell x ∈ [0, 1], aPKC is located at x = 0.
⇒ Explicit solution (β = 0): l(∞,1)l(∞,0)
= 1 + a(1 + bh(g))
with a = σγ, b = λ
dL, g2 = α
dP, h(x) = (ex−1)2
x(e2x−1),
Remark: ξ does not enter stationary state!Banff 02.07.2014 – p. 18/25
Systems of Reaction-Diffusion EquationsConvergence of Network of Reversible Linear Reactions
Deviation around the stationary state
X := L− L∞, Y := P − P∞, x := l − l∞, y := p− p∞
Network of four linear reversible reactions: general structure
Xα←−−−β
Y
λ
y
γ ξ
x
µ
xσ−−−→ν
y
Conservation law∫
[X + Y + x + y] = 0, ∀t ≥ 0.
Banff 02.07.2014 – p. 19/25
Systems of Reaction-Diffusion EquationsDetailed balance vs. complex kinetics
Then, the detailed balance condition αλσξβµνγ
= 1 is sufficient(and necessary?) for a convex entropy of the form:
d
dt
∫
[Φ1(X) + Φ2(Y ) + Φ3(x) + Φ4(y)]≤ 0.
Incompatibility of diffusion and reaction eigen-structure.
RD-systems allow potentially for oscillations ([BZ]).
From numerics: Decaying entropy after transient phase?
Banff 02.07.2014 – p. 19/25
Systems of Reaction-Diffusion EquationsNetwork of Reversible Linear Reactions
Closed system of reactions between n substances:
Ui Uj i, j = 1, 2, . . . , naij
aji
Concentrations U = [u1, u2, . . . , un]T , D = diag(d1, d2, . . . , dn)
Ut = DU + RU,
∂U∂ν
= 0, U(0) = [u1,0, u2,0, . . . , un,0]T ≥ 0.
aii = −∑nj=1,j 6=i aij, R =
a11 a12 . . . a1n
a21 a22 . . . a2n
. . . . . . . . . . . .
an1 an2 . . . ann
Banff 02.07.2014 – p. 20/25
Systems of Reaction-Diffusion EquationsNetwork of Reversible Linear Reactions
Convergence to the unique equilibrium?
∑nj=1 ajiui,∞ = 0, i = 1, 2, . . . , n,
|Ω|∑ni=1 ui,∞ = M > 0
Exists a quadratic entropy E[U ](t) = 12
∑ni=1 αi
∫
Ω|ui(t, x)|2dx,
for a multiplier B = diag(α1, α2, . . . , αn) with αi > 0 to get
D[U ](t) = −dE
dt=
n∑
i=1
αidi‖∇ui‖2 − UT BRU ≥ 0 ?
Banff 02.07.2014 – p. 20/25
Systems of Reaction-Diffusion EquationsNetwork of Reversible Linear Reactions
Algebra Theorem: Assume that R = (aij)ni,j=1 such that
(i) aij ≥ 0, i, j = 1, 2, . . . , n,
(ii)∑n
i=1 aji = 0 for all i = 1, 2, . . . , n,
(iii)∑n
j=1 a2ji ×
∑nk=1 a2
ik 6= 0 for all i = 1, 2, . . . , n.
Then, the matrix BR, where B = diag(1/ρ11, 1/ρ22, . . . , 1/ρnn),is negative semi-definite, ρij are some co-factors.
BR has n eigenvalues 0 = λ1 > λ2 ≥ λ3 ≥ . . . λn.
Interpretation: B transforms to relative entropy!
Banff 02.07.2014 – p. 20/25
Systems of Reaction-Diffusion EquationsNetwork of Reversible Linear Reactions
Friedmann-Neumann-Rannacher modela
u1 u3 u4 u5
u6u7u2u0
rimp2 rdelay rdelay
rdelay
rdelayrdelay
ractpJAK
rexp
rimp
Yes, if assume that reactions occur only within one domain
aE. Friedmann, R. Neumann, R. Rannacher, Well-posedness of a linear spatio-
temporal model of the JAK2/STAT5 signaling pathway, Comm. Math. Anal. 15
(2013) 76-102.
Banff 02.07.2014 – p. 20/25
Quasi-steady-state approximationQSSA as ξ → +∞
Fast expulsion limit ξ → +∞:
L P
ℓ p
αβ
γλ
σ
ξ
ξ→+∞−−−−→
L P
l
α
β
σγλ
Convergence towards reduced QSSAa
Proof: ideas from duality method and entropy dissipation.
a[T.Q.Bao, K.F., S. Rosenberger]
Banff 02.07.2014 – p. 21/25
Nonlinear Boundary ModelToy model system
Volume-concentrations u(x, t), Surface-concentration v(x, t)
Nonlinear Robin-type boundary condition and matchingreversible reaction source term on Γ = ∂Ω
(NBV)
ut − δu∆u = 0, x ∈ Ω, t ≥ 0,
δu∂u∂ν
= −α(kuuα − kvv
β), x ∈ Γ, t ≥ 0,
vt − δv∆Γv = β(kuuα − kvv
β), x ∈ Γ, t ≥ 0,
u(0, x) = u0(x) ≥ 0, x ∈ Ω,
v(0, x) = v0(x) ≥ 0, x ∈ Γ
Stoichiometric coefficients α, β ∈ [1,+∞)
Reaction rates ku(t, x), kv(t, x)
Banff 02.07.2014 – p. 22/25
Nonlinear Boundary ModelModel properties
Mass conservation law
M = β
∫
Ω
u(t, x) dx + α
∫
Γ
v(t, x) dS, ∀t ≥ 0.
The unique equilibrium (u∞, v∞) balances the reaction,
uα∞ = vβ
∞,
and satisfies the mass conservation
β|Ω|u∞ + α|Γ|v∞ = M.
Uniqueness from monotonicity of uα∞ =
(
Mα|Γ| −
β|Ωα|Γ| |u∞)
)β
Banff 02.07.2014 – p. 22/25
Exponential Convergence to EquilibriumEntropy and entropy dissipation
Entropy functional:
E(u, v) =
∫
Ω
u(log u− 1)dx +
∫
Γ
v(log v − 1)dS
Entropy dissipation:
D(u, v) = − d
dtE(u, v)
= δu
∫
Ω
|∇u|2u
dx + δv
∫
Γ
|∇Γv|2v
dS
+
∫
Γ
(vβ − uα) logvβ
uαdS ≥ 0
Two cases: non-degenerate δv > 0 and degenerate δv = 0.Banff 02.07.2014 – p. 23/25
Exponential Convergence to EquilibriumExplicit exponential convergence to equilibrium
Theorem:
Assume Ω ⊂ Rn with smooth boundary Γ = ∂Ω.
Assume initial data (u0, v0) ∈ L∞(Ω)× L∞(Γ).
Then, the global bounded solution (u, v) satisfies the followingexponential convergence to equilibrium
‖u(t)− u∞‖2L1(Ω) + ‖v(t)− v∞‖2L1(Γ) ≤C1e
−C0t (E(u0, v0)− E(u∞, v∞)) ,
where the constants C0 > 0, C1 > 0 depending on α, β, Ω andthe initial mass M = β
∫
Ωu0dx + α
∫
Γv0dS.
Banff 02.07.2014 – p. 23/25
Exponential Convergence to EquilibriumEntropy entropy dissipation estimate
For all functions u : Ω→ R+ and v : Ω→ R+ satisfying massconservation
β
∫
Ω
u dx + α
∫
Γ
v dS = M,
there exists C0 > 0 such that
D(u, v) ≥ C0 (E(u, v)− E(u∞, v∞))
Proof for two cases:non-degenerate boundary diffusion δv > 0
degenerate boundary diffusion δv = 0: need L∞-bounds
Banff 02.07.2014 – p. 23/25
Conclusions and Open ProblemsTowards bright sun and white snow for RD models
Good existence theory in 2D for quadratic nonlinearities
Higher dimensions?
Algebraic Structure of Networks
How to combine entropy and duality method?
THANK YOU VERY MUCH!!
Banff 02.07.2014 – p. 24/25
ReferencesJ. A. Cañizo, L. Desvillettes, K. F., Improved duality estimates and
applications to reaction-diffusion equations, to appear in CPDE
L. Desvillettes, K. F., M. Pierre, J. Vovelle, About Global Existence
for Quadratic Systems of Reaction-Diffusion, J. Advanced
Nonlinear Studies, 7 no. 3 (2007) pp. 491–511.
L. Desvillettes, K. F., Entropy Methods for Reaction-Diffusion
Equations: Slowly Growing A-priori Bounds, Revista
Matemática Iberoamericana, 24 no. 2 (2008) pp. 407–431.
T. Goudon, A. Vasseur. Regularity analysis for systems of
reaction- diffusion equations, Ann. Sci. Ec. Norm. Super., (4)
43 no. 1 (2010) pp. 117–141.
M. Pierre, Weak solutions and supersolutions in L1 for
reaction-diffusion systems, J. Evol. Equ., 3, no. 1 (2003)
153–168.Banff 02.07.2014 – p. 25/25