Optimality in critical and subcritical inequalities:
flows, linearization and entropy methods
Jean Dolbeault
http://www.ceremade.dauphine.fr/∼dolbeaul Ceremade, Universit´e Paris-Dauphine
December 6, 2016
Tsinghua Sanya International Mathematics Forum Nonlinear Partial Differential Equations and Mathematical Physics
Workshop, December 5–9, 2016
Outline
BSymmetry breaking and linearization
The critical Caffarelli-Kohn-Nirenberg inequality
A family of sub-critical Caffarelli-Kohn-Nirenberg inequalities Linearization and spectrum
BWithout weights: Gagliardo-Nirenberg inequalities and fast diffusion flows
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
BWith weights: Caffarelli-Kohn-Nirenberg inequalities and weighted nonlinear flows
Towards a parabolic proof
Large time asymptotics and spectral gaps A discussion of optimality cases
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
Collaborations
Collaboration with...
M.J. Esteban and M. Loss (symmetry, critical case) M.J. Esteban, M. Loss and M. Muratori (symmetry, subcritical case) M. Bonforte, M. Muratori and B. Nazaret (linearization and large time asymptotics for the evolution problem) M. del Pino, G. Toscani (nonlinear flows and entropy methods) A. Blanchet, G. Grillo, J.L. V´azquez (large time asymptotics and linearization for the evolution equations) ...and also
S. Filippas, A. Tertikas, G. Tarantello, M. Kowalczyk ...
Background references (partial)
Rigidity methods, uniqueness in nonlinear elliptic PDE’s:
[B. Gidas, J. Spruck, 1981],[M.-F. Bidaut-V´eron, L. V´eron, 1991]
Probabilistic methods (Markov processes), semi-group theory and carr´e du champmethods (Γ2theory): [D. Bakry, M. Emery, 1984], [Bakry, Ledoux, 1996], [Demange, 2008],[JD, Esteban, Loss, 2014 & 2015] →D. Bakry, I. Gentil, and M. Ledoux.
Analysis and geometry of Markov diffusion operators (2014) Entropy methods in PDEs
BEntropy-entropy production inequalities: Arnold, Carrillo, Desvillettes, JD, J¨ungel, Lederman, Markowich, Toscani, Unterreiter, Villani...,[del Pino, JD, 2001],[Blanchet, Bonforte, JD, Grillo, V´azquez]→A. J¨ungel, Entropy Methods for Diffusive Partial Differential Equations (2016)
BMass transportation: [Otto]→C. Villani, Optimal transport.
Old and new (2009)
BR´enyi entropy powers (information theory)[Savar´e, Toscani, 2014], [Dolbeault, Toscani]
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
Critical Caffarelli-Kohn-Nirenberg inequality Subcritical Caffarelli-Kohn-Nirenberg inequalities Linearization and spectrum
Symmetry and symmetry breaking results
BThe critical Caffarelli-Kohn-Nirenberg inequality
BA family of sub-critical Caffarelli-Kohn-Nirenberg inequalities BLinearization and spectrum
Critical Caffarelli-Kohn-Nirenberg inequality
LetDa,b:=n
v ∈Lp Rd,|x|−bdx
: |x|−a|∇v| ∈L2 Rd,dx o Z
Rd
|v|p
|x|b p dx 2/p
≤ Ca,b Z
Rd
|∇v|2
|x|2a dx ∀v ∈ Da,b holds under conditions ona andb
p= 2d
d−2 + 2 (b−a) (critical case) BAn optimal function among radial functions:
v?(x) =
1 +|x|(p−2) (ac−a)−p−22
and C?a,b= k |x|−bv?k2p
k |x|−a∇v?k22
Question: Ca,b= C?a,b (symmetry) orCa,b>C?a,b (symmetry breaking) ?
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
Critical Caffarelli-Kohn-Nirenberg inequality Subcritical Caffarelli-Kohn-Nirenberg inequalities Linearization and spectrum
Critical CKN: range of the parameters
Figure: d = 3 Z
Rd
|v|p
|x|b p dx 2/p
≤ Ca,b Z
Rd
|∇v|2
|x|2a dx
a b
0 1
−1
b=a
b=a+ 1
a=d−22 p
a≤b≤a+ 1ifd≥3
a<b≤a+ 1ifd= 2,a+ 1/2<b≤a+ 1ifd= 1 anda<ac := (d−2)/2
p= 2d
d−2 + 2 (b−a)
[Glaser, Martin, Grosse, Thirring (1976)]
[Caffarelli, Kohn, Nirenberg (1984)]
[F. Catrina, Z.-Q. Wang (2001)]
Linear instability of radial minimizers:
the Felli-Schneider curve
The Felli & Schneider curve bFS(a) := d(ac−a)
2p
(ac−a)2+d−1 +a−ac
a b
0
[Smets],[Smets, Willem],[Catrina, Wang],[Felli, Schneider]
The functional C?a,b
Z
Rd
|∇v|2
|x|2a dx− Z
Rd
|v|p
|x|b p dx 2/p
is linearly instable atv =v?
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
Critical Caffarelli-Kohn-Nirenberg inequality Subcritical Caffarelli-Kohn-Nirenberg inequalities Linearization and spectrum
Symmetry versus symmetry breaking:
the sharp result in the critical case
[JD, Esteban, Loss (Inventiones 2016)]
a b
0
Theorem
Let d≥2and p<2∗. If either a∈[0,ac)and b>0, or a<0 and b≥bFS(a), then the optimal functions for the critical
Caffarelli-Kohn-Nirenberg inequalities are radially symmetric
The Emden-Fowler transformation and the cylinder
BWith an Emden-Fowler transformation, critical the
Caffarelli-Kohn-Nirenberg inequality on the Euclidean space are equivalent to Gagliardo-Nirenberg inequalities on a cylinder
v(r, ω) =ra−acϕ(s, ω) with r =|x|, s=−logr and ω= x r With this transformation, the Caffarelli-Kohn-Nirenberg inequalities can be rewritten as thesubcriticalinterpolation inequality
k∂sϕk2L2(C)+k∇ωϕk2L2(C)+ Λkϕk2L2(C)≥µ(Λ)kϕk2Lp(C) ∀ϕ∈H1(C) whereΛ := (ac−a)2,C=R×Sd−1 and the optimal constantµ(Λ)is
µ(Λ) = 1
Ca,b with a=ac±√
Λ and b= d p ±√
Λ
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
Critical Caffarelli-Kohn-Nirenberg inequality Subcritical Caffarelli-Kohn-Nirenberg inequalities Linearization and spectrum
Linearization around symmetric critical points
Up to a normalization and a scaling
ϕ?(s, ω) = (coshs)−p−21
is a critical point of
H1(C)3ϕ7→ k∂sϕk2L2(C)+k∇ωϕk2L2(C)+ Λkϕk2L2(C)
under a constraint onkϕk2Lp(C)
ϕ? is notoptimal for (CKN) if the P¨oschl-Teller operator
−∂s2−∆ω+ Λ−ϕp−2? =−∂s2−∆ω+ Λ− 1 (coshs)2 hasa negative eigenvalue
Subcritical Caffarelli-Kohn-Nirenberg inequalities
Norms: kwkLq,γ(Rd) := R
Rd|w|q|x|−γ dx1/q
,kwkLq(Rd) :=kwkLq,0(Rd)
(some)Caffarelli-Kohn-Nirenberg interpolation inequalities(1984) kwkL2p,γ(Rd)≤Cβ,γ,pk∇wkϑL2,β(Rd)kwk1−ϑLp+1,γ(Rd) (CKN) HereCβ,γ,p denotes the optimal constant, the parameters satisfy
d≥2, γ−2< β < d−2d γ , γ∈(−∞,d), p∈(1,p?] withp?:= d−β−2d−γ and the exponentϑis determined by the scaling invariance, i.e.,
ϑ= (d−γ) (p−1)
p d+β+2−2γ−p(d−β−2)
Is the equality case achieved by the Barenblatt / Aubin-Talenti type function
w?(x) = 1 +|x|2+β−γ−1/(p−1)
∀x ∈Rd ?
Do we know (symmetry) that the equality case is achieved among radial functions?
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
Critical Caffarelli-Kohn-Nirenberg inequality Subcritical Caffarelli-Kohn-Nirenberg inequalities Linearization and spectrum
Range of the parameters
Herepis given
Symmetry and symmetry breaking
[JD, Esteban, Loss, Muratori, 2016]
Let us defineβFS(γ) :=d−2−p
(d−γ)2−4 (d−1) Theorem
Symmetry breaking holds in (CKN) if
γ <0 and βFS(γ)< β < d−2
d γ
In the rangeβFS(γ)< β <d−2d γ
w?(x) = 1 +|x|2+β−γ−1/(p−1) is notoptimal
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
Critical Caffarelli-Kohn-Nirenberg inequality Subcritical Caffarelli-Kohn-Nirenberg inequalities Linearization and spectrum
The green area is the region of symmetry, while the red area is the region of symmetry breaking. The threshold is determined by the hyperbola
(d−γ)2−(β−d+ 2)2−4 (d−1) = 0
A useful change of variables
With
α= 1 + β−γ
2 and n= 2 d−γ β+ 2−γ, (CKN) can be rewritten for a functionv(|x|α−1x) =w(x)as
kvkL2p,d−n(Rd)≤Kα,n,pkDαvkϑL2,d−n(Rd)kvk1−ϑLp+1,d−n(Rd)
with the notationss=|x|,Dαv = α∂v∂s,1s ∇ωv
. Parameters are in the range
d≥2, α >0, n>d and p∈(1,p?], p?:= n n−2 By our change of variables,w? is changed into
v?(x) := 1 +|x|2−p−11
∀x∈Rd
The symmetry breaking condition (Felli-Schneider) now reads α < αFS with αFS:=
rd−1 n−1
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
Critical Caffarelli-Kohn-Nirenberg inequality Subcritical Caffarelli-Kohn-Nirenberg inequalities Linearization and spectrum
The second variation
J[v] :=ϑlog kDαvkL2,d−n(Rd)
+ (1−ϑ) log kvkLp+1,d−n(Rd)
+ log Kα,n,p−log kvkL2p,d−n(Rd)
Let us definedµδ:=µδ(x)dx, where µδ(x) := (1 +|x|2)−δ. Sincev? is a critical point ofJ, a Taylor expansion at orderε2shows that
kDαv?k2L2,d−n(Rd)J
v?+ε µδ/2f
=12ε2ϑQ[f]+o(ε2) withδ=p−12p and
Q[f] = Z
Rd
|Dαf|2|x|n−ddµδ− 4pα2 p−1
Z
Rd
|f|2|x|n−ddµδ+1
We assume thatR
Rdf|x|n−ddµδ+1= 0(mass conservation)
Symmetry breaking: the proof
Proposition (Hardy-Poincar´e inequality)
Let d≥2,α∈(0,+∞), n>d andδ≥n. If f has0 average, then Z
Rd
|Dαf|2|x|n−ddµδ ≥Λ Z
Rd
|f|2|x|n−ddµδ+1
with optimal constantΛ = min{2α2(2δ−n),2α2δ η} whereη is the unique positive solution toη(η+n−2) = (d−1)/α2. The corresponding eigenfunction is not radially symmetric ifα2> (d−1)δ2
n(2δ−n) (δ−1) Q ≥0iff 4pp−1α2 ≤Λand symmetry breaking occurs in (CKN) if
2α2δ η < 4pα2
p−1 ⇐⇒ η <1
⇐⇒ d−1
α2 =η(η+n−2)<n−1 ⇐⇒ α > αFS
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
Inequalities without weights and fast diffusion equations
BR´enyi entropy powers
BSelf-similar variables and relative entropies BThe role of the spectral gap
R´ enyi entropy powers and fast diffusion
BR´enyi entropy powers, the entropy approach without rescaling:
[Savar´e, Toscani]: scalings, nonlinearity and a concavity property inspired by information theory
BFaster rates of convergence: [Carrillo, Toscani],[JD, Toscani]
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
The fast diffusion equation in original variables
Consider the nonlinear diffusion equation inRd,d ≥1
∂v
∂t = ∆vm
with initial datumv(x,t = 0) =v0(x)≥0such thatR
Rdv0dx= 1 and R
Rd|x|2v0dx<+∞. The large time behavior of the solutions is governed by the source-type Barenblatt solutions
U?(t,x) := 1
κt1/µdB? x κt1/µ
where
µ:= 2 +d(m−1), κ:=2µm m−1
1/µ
andB? is the Barenblatt profile
B?(x) :=
C?− |x|21/(m−1)
+ ifm>1 C?+|x|21/(m−1)
ifm<1
The R´ enyi entropy power F
Theentropy is defined by E :=
Z
Rd
vmdx and theFisher informationby
I :=
Z
Rd
v|∇p|2dx with p = m m−1vm−1 Ifv solves the fast diffusion equation, then
E0 = (1−m) I To computeI0, we will use the fact that
∂p
∂t = (m−1) p ∆p +|∇p|2 F := Eσ with σ= µ
d(1−m) = 1+ 2 1−m
1
d +m−1
= 2 d
1 1−m−1 has a linear growth asymptotically ast →+∞
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
The concavity property
Theorem
[Toscani-Savar´e]Assume that m≥1−1d if d>1 and m>0 if d= 1.
Then F(t)is increasing,(1−m) F00(t)≤0 and
t→+∞lim 1
t F(t) = (1−m)σ lim
t→+∞Eσ−1I = (1−m)σEσ−1? I? [Dolbeault-Toscani]The inequality
Eσ−1I≥Eσ−1? I?
is equivalent to the Gagliardo-Nirenberg inequality k∇wkθL2(Rd)kwk1−θLq+1(Rd)≥CGNkwkL2q(Rd)
if1−1d ≤m<1. Hint: vm−1/2= kwkw
L2q(Rd),q=2m−11
The proof
Lemma
If v solves ∂v∂t = ∆vm with 1d ≤m<1, then I0= d
dt Z
Rd
v|∇p|2dx =−2 Z
Rd
vm
kD2pk2+ (m−1) (∆p)2 dx Explicit arithmetic geometric inequality
kD2pk2− 1
d (∆p)2=
D2p− 1 d∆pId
2
There are no boundary terms in the integrations by parts
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
Remainder terms
F00=−σ(1−m) R[v]. Thepressure variableisP = 1−mm vm−1
R[v] := (σ−1) (1−m) Eσ−1 Z
Rd
vm ∆P−
R
Rdv|∇P|2dx R
Rdvmdx
2
dx + 2 Eσ−1
Z
Rd
vmD2P−d1∆PId2dx Let
G[v] := F[v] σ(1−m) =
Z
Rd
vmdx σ−1Z
Rd
v|∇P|2dx The Gagliardo-Nirenberg inequality is equivalent toG[v0]≥G[v?] Proposition
G[v0]≥G[v?] + Z ∞
0
R[v(t,·)]dt
Self-similar variables and relative entropies
The large time behavior of the solution of ∂v∂t = ∆vm is governed by the source-typeBarenblatt solutions
v?(t,x) := 1 κd(µt)d/µB?
x κ(µt)1/µ
where µ:= 2 +d(m−1) whereB?is the Barenblatt profile (with appropriate mass)
B?(x) := 1 +|x|21/(m−1) A time-dependent rescaling: self-similar variables
v(t,x) = 1 κdRd u
τ, x κR
where dR
dt =R1−µ, τ(t) := 12 log R(t)
R0
Then the functionusolvesa Fokker-Planck type equation
∂u
∂τ +∇ ·h
u ∇um−1− 2x i
= 0
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
Free energy and Fisher information
The functionusolves a Fokker-Planck type equation
∂u
∂τ +∇ ·h
u ∇um−1− 2x i
= 0 [Ralston, Newman, 1984]Lyapunov functional:
Generalized entropyor Free energy E[u] :=
Z
Rd
−um
m +|x|2u
dx− E0
Entropy production is measured by theGeneralized Fisher information
d
dtE[u] =−I[u], I[u] :=
Z
Rd
u∇um−1+ 2x2 dx
Without weights: relative entropy, entropy production
Stationary solution: chooseC such thatku∞kL1=kukL1=M>0 u∞(x) := C+ |x|2−1/(1−m)
+
Relative entropy: FixE0 so thatE[u∞] = 0
Entropy – entropy production inequality[del Pino, J.D.]
Theorem
d≥3, m∈[d−1d ,+∞), m>12, m6= 1 I[u]≥4E[u]
Corollary
[del Pino, J.D.]A solution u with initial data u0∈L1+(Rd)such that
|x|2u0∈L1(Rd), u0m∈L1(Rd)satisfies
E[u(t,·)]≤ E[u0]e−4t
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
A computation on a large ball, with boundary terms
∂u
∂τ +∇ ·h
u ∇um−1− 2x i
= 0 τ >0, x ∈BR
whereBR is a centered ball in Rd with radiusR>0, and assume that usatisfies zero-flux boundary conditions
∇um−1−2x
· x
|x| = 0 τ >0, x ∈∂BR. Withz(τ,x) :=∇Q(τ,x) :=∇um−1− 2x, therelative Fisher informationis such that
d dτ
Z
BR
u|z|2dx+ 4 Z
BR
u|z|2dx + 21−mm
Z
BR
um
D2Q2− (1−m) (∆Q)2 dx
= Z
∂BR
um ω· ∇|z|2
dσ≤0 (by Grisvard’s lemma)
Another improvement of the GN inequalities
Let us define therelative entropy E[u] :=−1
m Z
Rd
um− Bm? − mBm−1? (u− B?) dx therelative Fisher information
I[u] :=
Z
Rd
u|z|2dx= Z
Rd
u ∇um−1−2x2dx
and R[u] := 21−m m
Z
Rd
umD2Q2− (1−m) (∆Q)2 dx
Proposition
If1−1/d≤m<1and d ≥2, then I[u0]− 4E[u0]≥
Z ∞
0 R[u(τ,·)]dτ
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
Entropy – entropy production, Gagliardo-Nirenberg ineq.
4E[u]≤ I[u]
Rewrite it withp=2m−11 ,u=w2p,um=wp+1 as 1
2 2m
2m−1 2Z
Rd
|∇w|2dx+ 1
1−m −d Z
Rd
|w|1+pdx−K ≥0 for some γ,K =K0 R
Rdu dx =R
Rdw2pdxγ w =w∞=v∞1/2p is optimal
Theorem
[Del Pino, J.D.]With1<p≤ d−2d (fast diffusion case) and d ≥3 kwkL2p(Rd) ≤ Cp,dGNk∇wkθL2(Rd)kwk1−θLp+1(Rd)
Cp,dGN=y(p−1)2
2πd
θ2
2y−d 2y
2p1 Γ(y)
Γ(y−d2)
θd
,θ= p(d+2−(d−2)p)d(p−1) , y =p−1p+1
Sharp asymptotic rates of convergence
Assumptions on the initial datumv0
(H1)VD0 ≤v0≤VD1 for someD0>D1>0
(H2)if d≥3 andm≤m∗,(v0−VD)is integrable for a suitable D∈[D1,D0]
Theorem
[Blanchet, Bonforte, J.D., Grillo, V´azquez]Under Assumptions (H1)-(H2), if m<1 and m6=m∗:= d−4d−2, the entropy decays according to
E[v(t,·)]≤C e−2 (1−m) Λα,dt ∀t ≥0
whereΛα,d >0is the best constant in the Hardy–Poincar´e inequality Λα,d
Z
Rd
|f|2dµα−1≤ Z
Rd
|∇f|2dµα ∀f ∈H1(dµα) withα:= 1/(m−1)<0, dµα:=hαdx , hα(x) := (1 +|x|2)α
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
Plots (d = 5)
λ01=−4α−2d
λ10=−2α λ11=−6α−2 (d+ 2) λ02=−8α−4 (d+ 2)
λ20=−4α λ30
λ21λ12 λ03
λcont α,d:=14(d+ 2α−2)2
α=−d α=−(d+ 2)
α=−d+22 α=−d−22 α=−d+62
α 0 Essential spectrum ofLα,d
α=−√ d−1−d2 α=−√
d−1−d+42 α=−√2d−d+22
(d= 5) Spectrum ofLα,d
mc=d−2d
m1=d−1d m2=d+1d+2
! m1=d+2d
! m2=d+4d+6
m Spectrum of (1−m)L1/(m−1),d
(d= 5)
Essential spectrum of (1
1
−m)L1/(m−1),d
2 4 6
Improved asymptotic rates
[Bonforte, J.D., Grillo, V´azquez]Assume thatm∈(m1,1),d≥3.
Under Assumption (H1), ifv is a solution of the fast diffusion equation with initial datumv0such thatR
Rdx v0dx= 0, then the asymptotic convergence holds with an improved rate corresponding to the improved spectral gap.
m1=d+2d
m1=d−1d m2=d+4d+6
m2=d+1d+2
4
2
m 1 mc=d−2
d (d= 5)
γ(m)
0
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
R´enyi entropy powers
Self-similar variables and relative entropies The role of the spectral gap
Spectral gaps and best constants
mc=d−2d 0
m1=d−1d
m2=d+1d+2
! m2:=d+4d+6
m 1
2 4
Case 1 Case 2 Case 3
γ(m)
(d= 5)
! m1:=d+2d
Comments
The spectral gap corresponding to the red curves relies on a refined notion of relative entropy with respect tobest matching Barenblatt profiles[J.D., Toscani]
A result by[Denzler, Koch, McCann]Higher order time asymptotics of fast diffusion in Euclidean space: a dynamical systems approach
The constantC in
E[v(t,·)]≤Ce−2γ(m)t ∀t≥0
can be made explicit, under additional restrictions on the initial data[Bonforte, J.D., Grillo, V´azquez]
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
Weighted nonlinear flows:
Caffarelli-Kohn-Nirenberg inequalities
BEntropy and Caffarelli-Kohn-Nirenberg inequalities BLarge time asymptotics and spectral gaps
BOptimality cases
CKN and entropy – entropy production inequalities
When symmetry holds, (CKN) can be written as anentropy – entropy productioninequality
1−m
m (2 +β−γ)2E[v]≤ I[v]
and equality is achieved byBβ,γ. Here thefree energyand the relative Fisher informationare defined by
E[v] := 1 m−1
Z
Rd
vm−Bmβ,γ−mBm−1β,γ (v−Bβ,γ) dx
|x|γ I[v] :=
Z
Rd
v∇vm−1− ∇Bm−1β,γ
2 dx
|x|β. Ifv solves theFokker-Planck type equation
vt+|x|γ∇ ·h
|x|−βv∇ vm−1− |x|2+β−γi
= 0 (WFDE-FP)
then d
dtE[v(t,·)] =− m
1−mI[v(t,·)]
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
Proposition
Let m=p+12p and consider a solution to (WFDE-FP) with nonnegative initial datum u0∈L1,γ(Rd)such that ku0mkL1,γ(Rd) and
R
Rdu0|x|2+β−2γdx are finite. Then
E[v(t,·)]≤ E[u0]e−(2+β−γ)2t ∀t≥0 if one of the following two conditions is satisfied:
(i) either u0 is a.e. radially symmetric (ii) or symmetry holds in (CKN)
A useful change of variables (reminder)
With
α= 1 + β−γ
2 and n= 2 d−γ β+ 2−γ, (CKN) can be rewritten for a functionv(|x|α−1x) =w(x)as
kvkL2p,d−n(Rd)≤Kα,n,pkDαvkϑL2,d−n(Rd)kvk1−ϑLp+1,d−n(Rd)
with the notationss=|x|,Dαv = α∂v∂s,1s ∇ωv
. Parameters are in the range
d≥2, α >0, n>d and p∈(1,p?], p?:= n n−2 By our change of variables,w? is changed into
v?(x) := 1 +|x|2−1/(p−1)
∀x∈Rd
The symmetry breaking condition (Felli-Schneider) now reads α > αFS with αFS:=
rd−1 n−1
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
Towards a parabolic proof
For anyα≥1, let DαW = α ∂rW,r−1∇ωW
so that Dα =∇+ (α−1) x
|x|2(x· ∇) =∇+ (α−1)ω ∂r
and define the diffusion operatorLα by Lα=−D∗αDα =α2
∂2r +n−1 r ∂r
+∆ω
r2 where∆ω denotes the Laplace-Beltrami operator onSd−1
∂g
∂t = Lαgm is changed into
∂u
∂τ = D∗α(u z), z := Dαq, q :=um−1− Bαm−1, Bα(x) :=
1 +|x|2
α2 m−11
by the change of variables
g(t,x) = 1 κnRnu
τ, x κR
where
dR
dt =R1−µ, R(0) =R0
τ(t) = 12 logR(t)
R0
If the weight does not introduce any singularity atx= 0...
m 1−m
d dτ
Z
BR
u|z|2dµn
= Z
∂BR
um ω·Dα|z|2
|x|n−ddσ (≤0by Grisvard’s lemma)
−21−mm m−1 +1n Z
BR
um|Lαq|2dµn
− Z
BR
um
α4m1
q00−qr0 −α2(n−1)∆ωq r2
2
+2rα22
∇ωq0−∇rωq
2 dµn
−(n−2) α2FS−α2Z
BR
|∇ωq|2 r4 dµn
A formal computation that still needs to be justified (singularity atx= 0?)
Other potential application: the computation of Bakry, Gentil and Ledoux (chapter 6) for non-integer dimensions; weights on manifolds
[...]
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
Fast diffusion equations with weights: large time asymptotics
Relative uniform convergence Asymptotic rates of convergence From asymptotic to global estimates
Herev solves the Fokker-Planck type equation vt+|x|γ∇ ·h
|x|−βv∇ vm−1− |x|2+β−γi
= 0 (WFDE-FP)
Joint work with M. Bonforte, M. Muratori and B. Nazaret
Relative uniform convergence
ζ:= 1− 1−(1−m)2−mq
1−2−m1−mθ θ:= (1−m) (2+β−γ)
(1−m) (2+β)+2+β−γ is in the range0< θ < 1−m2−m <1 Theorem
For “good” initial data, there exist positive constantsKand t0such that, for all q∈2−m
1−m,∞
, the function w=v/Bsatisfies
kw(t)−1kLq,γ(Rd) ≤ Ke−2(1−m)22−m Λζ(t−t0) ∀t ≥t0
in the caseγ∈(0,d), and
kw(t)−1kLq,γ(Rd)≤ Ke−2(1−m)22−m Λ (t−t0) ∀t ≥t0 in the caseγ≤0
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
0
Λ0,1
Λ1,0
Λess
Essential spectrum
δ4 δ
δ1 δ2 δ5
Λ0,1
Λ1,0 Λess Essential spectrum
δ4 δ5:=n 2−η
The spectrum ofLas a function ofδ=1−m1 , withn= 5. The essential spectrum corresponds to the grey area, and its bottom is determined by the parabolaδ7→Λess(δ). The two eigenvaluesΛ0,1and Λ1,0are given by the plain, half-lines, away from the essential
spectrum. The spectral gap determines the asymptotic rate of convergence to the Barenblatt functions
Main steps of the proof:
Existence of weak solutions, L1,γ contraction, Comparison Principle, conservation of relative mass
Self-similar variables and the Ornstein-Uhlenbeck equation in relative variables: the ratiow(t,x) :=v(t,x)/B(x)solves
|x|−γwt =−B1 ∇ ·
|x|−βBw∇ (wm−1−1)Bm−1 in R+×Rd
w(0,·) =w0:=v0/B in Rd
Regularity: [Chiarenza, Serapioni], Harnack inequalities; relative uniform convergence (without rates) and asymptotic rates
(linearization)
The relative free energy and the relative Fisher information:
linearized free energy and linearized Fisher information A Duhamel formula and a bootstrap
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
Asymptotic rates of convergence
Corollary
Assume that m∈(0,1), with m6=m∗:= n−4n−2. Under the relative mass condition, for any “good solution” v there exists a positive constantC such that
E[v(t)]≤ Ce−2 (1−m) Λt ∀t ≥0.
With Csisz´ar-Kullback-Pinsker inequalities, these estimates provide a rate of convergence in L1,γ(Rd)
Improved estimates can be obtained using “best matching techniques”
From asymptotic to global estimates
When symmetry holds (CKN) can be written as anentropy – entropy productioninequality
(2 +β−γ)2E[v]≤ m 1−mI[v]
so that
E[v(t)]≤ E[v(0)]e−2 (1−m) Λ?t ∀t ≥0 with Λ?:= (2 +β−γ)2 2 (1−m)
Let us consider again theentropy – entropy production inequality K(M)E[v]≤ I[v] ∀v ∈L1,γ(Rd) such that kvkL1,γ(Rd)=M, whereK(M)is the best constant: withΛ(M) := m2(1−m)−2K(M)
E[v(t)]≤ E[v(0)]e−2 (1−m) Λ(M)t
∀t≥0
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
Symmetry breaking and global entropy – entropy production inequalities
Proposition
•In the symmetry breaking range of (CKN), for any M >0, we have 0<K(M)≤m2 (1−m)2Λ0,1
•If symmetry holds in (CKN) then
K(M)≥ 1−mm (2 +β−γ)2 Corollary
Assume that m∈[m1,1)
(i) For any M>0, if Λ(M) = Λ? thenβ=βFS(γ)
(ii) Ifβ > βFS(γ)thenΛ0,1<Λ? andΛ(M)∈(0,Λ0,1]for any M>0 (iii) For any M>0, if β < βFS(γ)and if symmetry holds in (CKN), then
Λ(M)>Λ?
Linearization and optimality
Joint work with M.J. Esteban and M. Loss
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
Linearization and scalar products
Withuεsuch that
uε=B? 1 +εfB?1−m
and Z
Rd
uεdx=M? at first order inε→0 we obtain thatf solves
∂f
∂t =Lf where Lf := (1−m)Bm−2? |x|γD∗α |x|−βB?Dαf Using the scalar products
hf1,f2i= Z
Rd
f1f2B2−m? |x|−γ dx and hhf1,f2ii= Z
Rd
Dαf1·Dαf2B?|x|−βdx we compute
1 2
d
dthf,fi=hf,Lfi= Z
Rd
f (Lf)B?2−m|x|−γ dx=− Z
Rd
|Dαf|2B?|x|−β dx=− hhf,fii for anyf smooth enough:
1 2
d
dthhf,fii= Z
Rd
Dαf ·Dα(Lf)B?|x|−βdx=− hhf,Lfii
Linearization of the flow, eigenvalues and spectral gap
Now let us consider an eigenfunction associated with the smallest positive eigenvalueλ1ofL
− Lf1=λ1f1
so thatf1realizes the equality case in theHardy-Poincar´e inequality hhg,gii=− hg,Lgi ≥λ1kg−g¯k2, g¯ :=hg,1i/h1,1i
− hhg,Lgii ≥λ1hhg,gii Proof: expansion of the square :
− hh(g−g¯),L(g −g¯)ii=hL(g−g¯),L(g −g¯)i=kL(g −g¯)k2 Key observation:
λ1≥4 ⇐⇒ α≤αFS :=
rd−1 n−1
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
Symmetry breaking in CKN inequalities
Symmetry holds in (CKN) ifJ[w]≥ J[w?]with J[w] :=ϑlog kDαwkL2,δ(Rd)
+(1−ϑ) log kwkLp+1,δ(Rd)
−log kwkL2p,δ(Rd)
withδ:=d−nand
J[w?+εg] =ε2Q[g]+o(ε2) where
2
ϑkDαw?k2L2,d−n(Rd)Q[g]
=kDαgk2L2,d−n(Rd)+p(2+β−γ)(p−1)2
d−γ−p(d−2−β) Z
Rd
|g|21+|x||x|n−d2 dx
−p(2p−1)(2+β−γ)(p−1)22 Z
Rd
|g|2(1+|x|x|n−d|2)2 dx is a nonnegative quadratic form if and only ifα≤αFS
Symmetry breaking holds ifα > αFS
Information – production of information inequality
LetK[u]be such that d
dτI[u(τ,·)] =− K[u(τ,·)] =−(sum of squares) Ifα≤αFS, thenλ1≥4and
u7→ K[u]
I[u] −4 is a nonnegative functional
Withuε=B? 1 +εfB1−m?
, we observe that
4≤ C2:= inf
u
K[u]
I[u] ≤ lim
ε→0inf
f
K[uε] I[uε] = inf
f
hhf,Lfii
hhf,fii = hhf1,Lf1ii hhf1,f1ii =λ1
ifλ1= 4, that is, if α=αFS, theninfK/I= 4is achieved in the asymptotic regime asu→ B?and determined by the spectral gap ofL
ifλ1>4, that is, if α < αFS, thenK/I >4
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
Symmetry in Caffarelli-Kohn-Nirenberg inequalities
Ifα≤αFS, the fact thatK/I ≥4has an important consequence.
Indeed we know that d
dτ (I[u(τ,·)]− 4E[u(τ,·)])≤0 so that
I[u]−4E[u]≥ I[B?]−4E[B?] = 0
This inequality is equivalent toJ[w]≥ J[w?], which establishes that optimality in (CKN) is achieved among symmetric functions. In other words, the linearized problem shows that forα≤αFS, the function
τ7→ I[u(τ,·)]−4E[u(τ,·)]
is monotone decreasing
This explains why the method based on nonlinear flows provides theoptimal range for symmetry
Entropy – production of entropy inequality
Using dτd (I[u(τ,·)]− C2E[u(τ,·)])≤0, we know that I[u]− C2E[u]≥ I[B?]− C2E[B?] = 0 As a consequence, we have that
C1:= inf
u
I[u]
E[u] ≥C2= inf
u
K[u]
I[u]
Withuε=B? 1 +εfB1−m?
, we observe that
C1≤ lim
ε→0inf
f
I[uε] E[uε] = inf
f
hf,Lfi
hf,fi = hf1,Lf1i hf1,fi1
=λ1= lim
ε→0inf
f
K[uε] I[uε] Iflimε→0inff K[uε]
I[uε] =C2, thenC1=C2=λ1
This happens ifα=αFS and in particular in the case without weights (Gagliardo-Nirenberg inequalities)
J. Dolbeault Flows, linearization, entropy methods
Symmetry breaking and linearization Entropy methods without weights Weighted nonlinear flows and CKN inequalities
A parabolic proof ?
Large time asymptotics and spectral gaps Linearization and optimality
These slides can be found at
http://www.ceremade.dauphine.fr/∼dolbeaul/Conferences/
BLectures The papers can be found at
http://www.ceremade.dauphine.fr/∼dolbeaul/Preprints/list/
BPreprints / papers
For final versions, useDolbeaultas login andJeanas password