Measure concentration, functional inequalities, and curvature of metric measure spaces
M. Ledoux, Institut de Math´ematiques de Toulouse, France
Luminy, March 2008
circleof ideas
between analysis, geometry and probability theory
• concentration of measure phenomenon
• geometric, measure/information theoretic, functional inequalities
• curvature of metric measure spaces
theory of optimal transportation
circleof ideas
between analysis, geometry and probability theory
• concentration of measure phenomenon
• geometric, measure/information theoretic, functional inequalities
• curvature of metric measure spaces
theory of optimal transportation
circleof ideas
between analysis, geometry and probability theory
• concentration of measure phenomenon
• geometric, measure/information theoretic, functional inequalities
• curvature of metric measure spaces
theory of optimal transportation
circleof ideas
between analysis, geometry and probability theory
• concentration of measure phenomenon
• geometric, measure/information theoretic, functional inequalities
• curvature of metric measure spaces
theory of optimal transportation
circleof ideas
between analysis, geometry and probability theory
• concentration of measure phenomenon
• geometric, measure/information theoretic, functional inequalities
• curvature of metric measure spaces
theory of optimal transportation
concentration of measure phenomenon
V. Milman 1970
Dvoretzky’s theorem onspherical sections of convex bodies inhigh dimension
for allε >0, there exists δ(ε)>0such that for every convex body K in Rn, there is a subspace F of Rnwith dim(F)≥δ(ε) logn and anellipsoidE in F such that
Milman’s idea : find F random using aconcentration property ofspherical measures in high dimension
concentration of measure phenomenon
V. Milman 1970
Dvoretzky’s theorem onspherical sections of convex bodies inhigh dimension
for allε >0, there exists δ(ε)>0such that for every convex body K in Rn, there is a subspace F of Rnwith dim(F)≥δ(ε) logn and anellipsoidE in F such that
Milman’s idea : find F random using aconcentration property ofspherical measures in high dimension
concentration of measure phenomenon
V. Milman 1970
Dvoretzky’s theorem onspherical sections of convex bodies inhigh dimension
for allε >0, there exists δ(ε)>0such that for every convex body K in Rn, there is a subspace F of Rnwith dim(F)≥δ(ε) logn and anellipsoidE in F such that
Milman’s idea : find F random using aconcentration property ofspherical measures in high dimension
concentration of measure phenomenon
V. Milman 1970
Dvoretzky’s theorem onspherical sections of convex bodies inhigh dimension
for allε >0, there exists δ(ε)>0such that for every convex body K in Rn,
there is a subspace F of Rnwith dim(F)≥δ(ε) logn and anellipsoidE in F such that
Milman’s idea : find F random using aconcentration property ofspherical measures in high dimension
concentration of measure phenomenon
V. Milman 1970
Dvoretzky’s theorem onspherical sections of convex bodies inhigh dimension
for allε >0, there exists δ(ε)>0such that for every convex body K in Rn, there is a subspace F of Rnwith dim(F)≥δ(ε) logn and anellipsoidE in F
such that
Milman’s idea : find F random using aconcentration property ofspherical measures in high dimension
concentration of measure phenomenon
V. Milman 1970
Dvoretzky’s theorem onspherical sections of convex bodies inhigh dimension
for allε >0, there exists δ(ε)>0such that for every convex body K in Rn, there is a subspace F of Rnwith dim(F)≥δ(ε) logn and anellipsoidE in F such that
K ∩F
Milman’s idea : find F random using aconcentration property ofspherical measures in high dimension
concentration of measure phenomenon
V. Milman 1970
Dvoretzky’s theorem onspherical sections of convex bodies inhigh dimension
for allε >0, there exists δ(ε)>0such that for every convex body K in Rn, there is a subspace F of Rnwith dim(F)≥δ(ε) logn and anellipsoidE in F such that
(1−ε)E ⊂K ∩F ⊂(1 +ε)E
Milman’s idea : find F random using aconcentration property ofspherical measures in high dimension
concentration of measure phenomenon
V. Milman 1970
Dvoretzky’s theorem onspherical sections of convex bodies inhigh dimension
for allε >0, there exists δ(ε)>0such that for every convex body K in Rn, there is a subspace F of Rnwith dim(F)≥δ(ε) logn and anellipsoidE in F such that
(1−ε)E ⊂K ∩F ⊂(1 +ε)E
Milman’s idea : find F random
using aconcentration property ofspherical measures in high dimension
concentration of measure phenomenon
V. Milman 1970
Dvoretzky’s theorem onspherical sections of convex bodies inhigh dimension
for allε >0, there exists δ(ε)>0such that for every convex body K in Rn, there is a subspace F of Rnwith dim(F)≥δ(ε) logn and anellipsoidE in F such that
(1−ε)E ⊂K ∩F ⊂(1 +ε)E
Milman’s idea : find F random using aconcentration property ofspherical measures in high dimension
spherical concentration
isoperimetric inequalityon Sn⊂Rn+1
geodesic ballsare the extremal sets
if vol(A) ≥ vol(B), B geodesic ball
then vol(Ar) ≥ vol(Br), r >0
normalized volume µ(·) = vol(·) vol(Sn)
spherical concentration
isoperimetric inequalityon Sn⊂Rn+1
geodesic ballsare the extremal sets
if vol(A) ≥ vol(B), B geodesic ball
then vol(Ar) ≥ vol(Br), r >0
normalized volume µ(·) = vol(·) vol(Sn)
spherical concentration
isoperimetric inequalityon Sn⊂Rn+1
geodesic ballsare the extremal sets
if vol(A) ≥ vol(B), B geodesic ball
then vol(Ar) ≥ vol(Br), r >0
normalized volume µ(·) = vol(·) vol(Sn)
spherical concentration
isoperimetric inequalityon Sn⊂Rn+1
geodesic ballsare the extremal sets
if vol(A) ≥ vol(B), B geodesic ball
then vol(Ar) ≥ vol(Br), r >0
normalized volume µ(·) = vol(·) vol(Sn)
spherical concentration
isoperimetric inequalityon Sn⊂Rn+1
geodesic ballsare the extremal sets
if vol(A) ≥ vol(B), B geodesic ball
then vol(Ar) ≥ vol(Br), r >0
normalized volume µ(·) = vol(·) vol(Sn)
spherical concentration
isoperimetric inequalityon Sn⊂Rn+1
geodesic ballsare the extremal sets
if vol(A) ≥ vol(B), B geodesic ball
then vol(Ar) ≥ vol(Br), r >0
normalized volume µ(·) = vol(·) vol(Sn)
if µ(A) ≥ 1 2
(B half-sphere)
then, for all r >0,
µ(Ar) ≥ µ(Br) ≥ 1−e−(n−1)r2/2
r∼ 1
√n (n→ ∞)
µ(Ar) ≈ 1
if µ(A) ≥ 1
2 (B half-sphere)
then, for all r >0,
µ(Ar) ≥ µ(Br) ≥ 1−e−(n−1)r2/2
r∼ 1
√n (n→ ∞)
µ(Ar) ≈ 1
if µ(A) ≥ 1
2 (B half-sphere)
then, for all r >0,
µ(Ar) ≥ µ(Br)
≥ 1−e−(n−1)r2/2
r∼ 1
√n (n→ ∞)
µ(Ar) ≈ 1
if µ(A) ≥ 1
2 (B half-sphere)
then, for all r >0,
µ(Ar) ≥ µ(Br) ≥ 1−e−(n−1)r2/2
r∼ 1
√n (n→ ∞)
µ(Ar) ≈ 1
if µ(A) ≥ 1
2 (B half-sphere)
then, for all r >0,
µ(Ar) ≥ µ(Br) ≥ 1−e−(n−1)r2/2
r∼ 1
√n (n→ ∞)
µ(Ar) ≈ 1
if µ(A) ≥ 1
2 (B half-sphere)
then, for all r >0,
µ(Ar) ≥ µ(Br) ≥ 1−e−(n−1)r2/2
r∼ 1
√n (n→ ∞)
µ(Ar) ≈ 1
application byV. Milman
m (normalized) Haar measure on On+1
m {T ∈ On+1;Tx ∈A}
= µ(A), x ∈Rn+1,A⊂Sn k · k gauge of K
showthat for dim(E)≥δ(ε) logn
mn
T ∈ On+1; |z|
1 +ε ≤ kTzk ≤ |z|
1−ε, z ∈Eo
> 0
F =ImT|E, E=T(BE)
application byV. Milman
m (normalized) Haar measure on On+1
m {T ∈ On+1;Tx ∈A}
= µ(A), x ∈Rn+1,A⊂Sn k · k gauge of K
showthat for dim(E)≥δ(ε) logn
mn
T ∈ On+1; |z|
1 +ε ≤ kTzk ≤ |z|
1−ε, z ∈Eo
≈ 1
F =ImT|E, E=T(BE)
framework for measure concentration
metric measure space(X,d, µ)
(X,d) metric space
µ Borel measureonX, µ(X) = 1
concentration function
α(r) =α(X,d,µ)(r) = sup
1−µ(Ar);µ(A)≥1/2 , r >0
Ar =
x ∈X;d(x,A)<r
µ uniform on Sn⊂Rn+1 : α(r) ≤ e−(n−1)r2/2, r >0
framework for measure concentration
metric measure space(X,d, µ) (X,d) metric space
µ Borel measureonX, µ(X) = 1
concentration function
α(r) =α(X,d,µ)(r) = sup
1−µ(Ar);µ(A)≥1/2 , r >0
Ar =
x ∈X;d(x,A)<r
µ uniform on Sn⊂Rn+1 : α(r) ≤ e−(n−1)r2/2, r >0
framework for measure concentration
metric measure space(X,d, µ) (X,d) metric space
µ Borel measureonX, µ(X) = 1
concentration function
α(r) =α(X,d,µ)(r) = sup
1−µ(Ar);µ(A)≥1/2 , r >0
Ar =
x ∈X;d(x,A)<r
µ uniform on Sn⊂Rn+1 : α(r) ≤ e−(n−1)r2/2, r >0
framework for measure concentration
metric measure space(X,d, µ) (X,d) metric space
µ Borel measureonX, µ(X) = 1
concentration function
α(r) =α(X,d,µ)(r) = sup
1−µ(Ar);µ(A)≥1/2 , r >0
Ar =
x ∈X;d(x,A)<r
µ uniform on Sn⊂Rn+1 : α(r) ≤ e−(n−1)r2/2, r >0
framework for measure concentration
metric measure space(X,d, µ) (X,d) metric space
µ Borel measureonX, µ(X) = 1
concentration function
α(r) =α(X,d,µ)(r) = sup
1−µ(Ar);µ(A)≥1/2 , r >0
Ar =
x ∈X;d(x,A)<r
µ uniform on Sn⊂Rn+1 : α(r) ≤ e−(n−1)r2/2, r >0
framework for measure concentration
metric measure space(X,d, µ) (X,d) metric space
µ Borel measureonX, µ(X) = 1
concentration function
α(r) =α(X,d,µ)(r) = sup
1−µ(Ar);µ(A)≥1/2 , r >0
Ar =
x ∈X;d(x,A)<r µ uniform on Sn⊂Rn+1 :
α(r) ≤ e−(n−1)r2/2, r >0
framework for measure concentration
metric measure space(X,d, µ) (X,d) metric space
µ Borel measureonX, µ(X) = 1
concentration function
α(r) =α(X,d,µ)(r) = sup
1−µ(Ar);µ(A)≥1/2 , r >0
Ar =
x ∈X;d(x,A)<r
µ uniform on Sn⊂Rn+1 : α(r) ≤ e−(n−1)r2/2, r >0
equivalent formulation on functions
F :X →R 1- Lipschitz m=mF median of F for µ
µ(F ≤m), µ(F ≥m) ≥ 1 2
A={F ≤m} =⇒ Ar ⊂ {F <m+r}
µ
|F −m|<r ≥ 1−2α(r), r >0
equivalent formulation on functions
F :X →R 1- Lipschitz
m=mF median of F for µ µ(F ≤m), µ(F ≥m) ≥ 1
2
A={F ≤m} =⇒ Ar ⊂ {F <m+r}
µ
|F −m|<r ≥ 1−2α(r), r >0
equivalent formulation on functions
F :X →R 1- Lipschitz m=mF median of F for µ
µ(F ≤m), µ(F ≥m) ≥ 1 2
A={F ≤m} =⇒ Ar ⊂ {F <m+r}
µ
|F −m|<r ≥ 1−2α(r), r >0
equivalent formulation on functions
F :X →R 1- Lipschitz m=mF median of F for µ
µ(F ≤m), µ(F ≥m) ≥ 1 2
A={F ≤m} =⇒ Ar ⊂ {F <m+r}
µ
|F −m|<r ≥ 1−2α(r), r >0
equivalent formulation on functions
F :X →R 1- Lipschitz m=mF median of F for µ
µ(F ≤m), µ(F ≥m) ≥ 1 2
A={F ≤m} =⇒ Ar ⊂ {F <m+r}
µ
|F −m|<r ≥ 1−2α(r), r >0
dual description : observable diameter(M. Gromov)
κ >0 (κ= 10−10)
PartDiamµ(X,d)
= inf
D ≥0,∃A⊂X,Diam(A)≤D, µ(A)≥1−κ
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
F :µ→µF
dual description : observable diameter(M. Gromov)
κ >0 (κ= 10−10)
PartDiamµ(X,d)
= inf
D ≥0,∃A⊂X,Diam(A)≤D, µ(A)≥1−κ
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
F :µ→µF
dual description : observable diameter(M. Gromov)
κ >0 (κ= 10−10)
PartDiamµ(X,d)
= inf
D ≥0,∃A⊂X,Diam(A)≤D, µ(A)≥1−κ
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
F :µ→µF
dual description : observable diameter(M. Gromov)
κ >0 (κ= 10−10)
PartDiamµ(X,d)
= inf
D ≥0,∃A⊂X,Diam(A)≤D, µ(A)≥1−κ
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
F :µ→µF
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
µ stateon the configuration space (X,d)
F :X →R observable
yielding thetomographic image µF on R
µF ]m−r,m+r[
=µ
|F −m|<r ≥ 1−2α(r)
ObsDiamµ(X,d) ∼ α−1(κ)
ObsDiam(Sn) = O 1
√n
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
µ stateon the configuration space (X,d)
F :X →R observable
yielding thetomographic image µF on R
µF ]m−r,m+r[
=µ
|F −m|<r ≥ 1−2α(r)
ObsDiamµ(X,d) ∼ α−1(κ)
ObsDiam(Sn) = O 1
√n
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
µ stateon the configuration space (X,d)
F :X →R observable
yielding thetomographic image µF on R
µF ]m−r,m+r[
=µ
|F −m|<r ≥ 1−2α(r)
ObsDiamµ(X,d) ∼ α−1(κ)
ObsDiam(Sn) = O 1
√n
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
µ stateon the configuration space (X,d)
F :X →R observable
yielding thetomographic image µF on R
µF ]m−r,m+r[
=µ
|F −m|<r ≥ 1−2α(r)
ObsDiamµ(X,d) ∼ α−1(κ)
ObsDiam(Sn) = O 1
√n
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
µ stateon the configuration space (X,d)
F :X →R observable
yielding thetomographic image µF on R
µF ]m−r,m+r[
=µ
|F −m|<r ≥ 1−2α(r)
ObsDiamµ(X,d) ∼ α−1(κ)
ObsDiam(Sn) = O 1
√n
ObsDiamµ(X,d) = sup
F 1−Lip
PartDiamµF(R)
µ stateon the configuration space (X,d)
F :X →R observable
yielding thetomographic image µF on R
µF ]m−r,m+r[
=µ
|F −m|<r ≥ 1−2α(r)
ObsDiamµ(X,d) ∼ α−1(κ)
ObsDiam(Sn) = O 1
√n
measure concentration property
less restrictivethanisoperimetry, widely shared
variety of examples and tools
• spectral methods
• combinatorial tools
• product measures(M. Talagrand 1995)
• geometric, measure/information theoretic, functional inequalities
measure concentration property less restrictivethanisoperimetry, widely shared
variety of examples and tools
• spectral methods
• combinatorial tools
• product measures(M. Talagrand 1995)
• geometric, measure/information theoretic, functional inequalities
measure concentration property less restrictivethanisoperimetry, widely shared
variety of examples and tools
• spectral methods
• combinatorial tools
• product measures(M. Talagrand 1995)
• geometric, measure/information theoretic, functional inequalities
measure concentration property less restrictivethanisoperimetry, widely shared
variety of examples and tools
• spectral methods
• combinatorial tools
• product measures(M. Talagrand 1995)
• geometric, measure/information theoretic, functional inequalities
measure concentration property less restrictivethanisoperimetry, widely shared
variety of examples and tools
• spectral methods
• combinatorial tools
• product measures(M. Talagrand 1995)
• geometric, measure/information theoretic, functional inequalities
measure concentration property less restrictivethanisoperimetry, widely shared
variety of examples and tools
• spectral methods
• combinatorial tools
• product measures
(M. Talagrand 1995)
• geometric, measure/information theoretic, functional inequalities
measure concentration property less restrictivethanisoperimetry, widely shared
variety of examples and tools
• spectral methods
• combinatorial tools
• product measures(M. Talagrand 1995)
• geometric, measure/information theoretic, functional inequalities
measure concentration property less restrictivethanisoperimetry, widely shared
variety of examples and tools
• spectral methods
• combinatorial tools
• product measures(M. Talagrand 1995)
• geometric, measure/information theoretic, functional inequalities
measure concentration property less restrictive than isoperimetry, widely shared
variety of examples and tools
• spectral methods
• combinatorial tools
• product measures(M. Talagrand 1995)
• geometric, measure/information theoretic, functional inequalities
dimension free inequalities
model : Gaussian measure dγ(x) =e−|x|2/2 dx
(2π)k/2 on (Rk,| · |) concentration property
R ∼√ n
µ uniformon Sn√n → γ Gaussian Gauss space : curvature 1 dimension ∞
dimension free inequalities
model : Gaussian measure dγ(x) =e−|x|2/2 dx
(2π)k/2 on (Rk,| · |)
concentration property
R ∼√ n
µ uniformon Sn√n → γ Gaussian Gauss space : curvature 1 dimension ∞
dimension free inequalities
model : Gaussian measure dγ(x) =e−|x|2/2 dx
(2π)k/2 on (Rk,| · |) concentration property
R ∼√ n
µ uniformon Sn√n → γ Gaussian Gauss space : curvature 1 dimension ∞
dimension free inequalities
model : Gaussian measure dγ(x) =e−|x|2/2 dx
(2π)k/2 on (Rk,| · |) concentration property
Sn: α(r) ≤ e−(n−1)r2/2
R ∼√ n
µ uniformon Sn√n → γ Gaussian Gauss space : curvature 1 dimension ∞
dimension free inequalities
model : Gaussian measure dγ(x) =e−|x|2/2 dx
(2π)k/2 on (Rk,| · |) concentration property
SnR : α(r) ≤ e−(n−1)r2/2R2
R ∼√ n
µ uniformon Sn√n → γ Gaussian Gauss space : curvature 1 dimension ∞
dimension free inequalities
model : Gaussian measure dγ(x) =e−|x|2/2 dx
(2π)k/2 on (Rk,| · |) concentration property
SnR : α(r) ≤ e−(n−1)r2/2R2 R ∼√
n
µ uniformon Sn√n → γ Gaussian Gauss space : curvature 1 dimension ∞
dimension free inequalities
model : Gaussian measure dγ(x) =e−|x|2/2 dx
(2π)k/2 on (Rk,| · |) concentration property
SnR : α(r) ≤ e−(n−1)r2/2R2 R ∼√
n
µ uniformon Sn√n
→ γ Gaussian Gauss space : curvature 1 dimension ∞
dimension free inequalities
model : Gaussian measure dγ(x) =e−|x|2/2 dx
(2π)k/2 on (Rk,| · |) concentration property
SnR : α(r) ≤ e−(n−1)r2/2R2 R ∼√
n
µ uniformon Sn√n → γ Gaussian
Gauss space : curvature 1 dimension ∞
dimension free inequalities
model : Gaussian measure dγ(x) =e−|x|2/2 dx
(2π)k/2 on (Rk,| · |) concentration property
SnR : α(r) ≤ e−(n−1)r2/2R2 R ∼√
n
µ uniformon Sn√n → γ Gaussian Gauss space : curvature 1 dimension ∞
Gaussian concentration α(r) ≤ e−r2/2
if A⊂Rk, γ(A) ≥ 1 2 then γ(Ar) ≥ 1−e−r2/2, r >0
ObsDiamγ(Rk) = O(1)
independentof the dimension infinite dimensional analysis
Gaussian concentration α(r) ≤ e−r2/2
if A⊂Rk, γ(A) ≥ 1 2
then γ(Ar) ≥ 1−e−r2/2, r >0
ObsDiamγ(Rk) = O(1)
independentof the dimension infinite dimensional analysis
Gaussian concentration α(r) ≤ e−r2/2
if A⊂Rk, γ(A) ≥ 1 2 then γ(Ar) ≥ 1−e−r2/2, r >0
ObsDiamγ(Rk) = O(1)
independentof the dimension infinite dimensional analysis
Gaussian concentration α(r) ≤ e−r2/2
if A⊂Rk, γ(A) ≥ 1 2 then γ(Ar) ≥ 1−e−r2/2, r >0
ObsDiamγ(Rk) = O(1)
independentof the dimension infinite dimensional analysis
Gaussian concentration α(r) ≤ e−r2/2
if A⊂Rk, γ(A) ≥ 1 2 then γ(Ar) ≥ 1−e−r2/2, r >0
ObsDiamγ(Rk) = O(1)
independentof the dimension
infinite dimensional analysis
Gaussian concentration α(r) ≤ e−r2/2
if A⊂Rk, γ(A) ≥ 1 2 then γ(Ar) ≥ 1−e−r2/2, r >0
ObsDiamγ(Rk) = O(1)
independentof the dimension infinite dimensional analysis
triple description of Gaussian concentration α(r) ≤ e−r2/2
• geometric
• measure/information theoretic
• functional
notion of curvaturein metric measure spaces
(PDEand calculus of variations viewpoint)
triple description of Gaussian concentration α(r) ≤ e−r2/2
• geometric
• measure/information theoretic
• functional
notion of curvaturein metric measure spaces
(PDEand calculus of variations viewpoint)
triple description of Gaussian concentration α(r) ≤ e−r2/2
• geometric
• measure/information theoretic
• functional
notion of curvaturein metric measure spaces
(PDEand calculus of variations viewpoint)
triple description of Gaussian concentration α(r) ≤ e−r2/2
• geometric
• measure/information theoretic
• functional
notion of curvaturein metric measure spaces
(PDEand calculus of variations viewpoint)
triple description of Gaussian concentration α(r) ≤ e−r2/2
• geometric
• measure/information theoretic
• functional
notion of curvaturein metric measure spaces
(PDEand calculus of variations viewpoint)
triple description of Gaussian concentration α(r) ≤ e−r2/2
• geometric
• measure/information theoretic
• functional
notion of curvaturein metric measure spaces
(PDEand calculus of variations viewpoint)
geometric description : Brunn-Minkowski inequality
Pr´ekopa-Leindler theorem
θ∈[0,1], u,v,w :Rn→R+
if w θx+ (1−θ)y
≥ u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dx ≥ Z
u dx θ Z
v dx 1−θ
u =χA,v=χB,multiplicative form ofBrunn-Minkowski voln θA+ (1−θ)B
≥ voln(A)θvoln(B)1−θ
geometric description : Brunn-Minkowski inequality Pr´ekopa-Leindler theorem
θ∈[0,1], u,v,w :Rn→R+
if w θx+ (1−θ)y
≥ u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dx ≥ Z
u dx θ Z
v dx 1−θ
u =χA,v=χB,multiplicative form ofBrunn-Minkowski voln θA+ (1−θ)B
≥ voln(A)θvoln(B)1−θ
geometric description : Brunn-Minkowski inequality Pr´ekopa-Leindler theorem
θ∈[0,1], u,v,w :Rn→R+
if w θx+ (1−θ)y
≥ u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dx ≥ Z
u dx θ Z
v dx 1−θ
u =χA,v=χB,multiplicative form ofBrunn-Minkowski voln θA+ (1−θ)B
≥ voln(A)θvoln(B)1−θ
geometric description : Brunn-Minkowski inequality Pr´ekopa-Leindler theorem
θ∈[0,1], u,v,w :Rn→R+
if w θx+ (1−θ)y
≥ u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dx ≥ Z
u dx θ Z
v dx 1−θ
u =χA,v=χB,multiplicative form ofBrunn-Minkowski voln θA+ (1−θ)B
≥ voln(A)θvoln(B)1−θ
geometric description : Brunn-Minkowski inequality Pr´ekopa-Leindler theorem
θ∈[0,1], u,v,w :Rn→R+
if w θx+ (1−θ)y
≥ u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dx ≥ Z
u dx θ Z
v dx 1−θ
u =χA,v=χB,multiplicative form ofBrunn-Minkowski voln θA+ (1−θ)B
≥ voln(A)θvoln(B)1−θ
geometric description : Brunn-Minkowski inequality Pr´ekopa-Leindler theorem
θ∈[0,1], u,v,w :Rn→R+
if w θx+ (1−θ)y
≥ u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dx ≥ Z
u dx θ Z
v dx 1−θ
u =χA,v=χB,
multiplicative form ofBrunn-Minkowski voln θA+ (1−θ)B
≥ voln(A)θvoln(B)1−θ
geometric description : Brunn-Minkowski inequality Pr´ekopa-Leindler theorem
θ∈[0,1], u,v,w :Rn→R+
if w θx+ (1−θ)y
≥ u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dx ≥ Z
u dx θ Z
v dx 1−θ
u =χA,v=χB,multiplicative form ofBrunn-Minkowski voln θA+ (1−θ)B
≥ voln(A)θvoln(B)1−θ
dx → dγ(x) =e−|x|2/2 dx (2π)n/2
then Z
w dγ ≥ Z
u dγ θ Z
v dγ 1−θ
choose θ= 1/2, w ≡1, v =χA, u =ed(·,A)2/4 Z
ed(·,A)2/4dγ ≤ 1 γ(A) γ(Ar) ≥ 1−2e−r2/4, γ(A) ≥ 1
2 extension : dµ=e−Vdx, V00≥c >0
dx → dγ(x) =e−|x|2/2 dx (2π)n/2 if w θx+ (1−θ)y
≥ u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dγ ≥ Z
u dγ θ Z
v dγ 1−θ
choose θ= 1/2, w ≡1, v =χA, u =ed(·,A)2/4 Z
ed(·,A)2/4dγ ≤ 1 γ(A) γ(Ar) ≥ 1−2e−r2/4, γ(A) ≥ 1
2 extension : dµ=e−Vdx, V00≥c >0
dx → dγ(x) =e−|x|2/2 dx (2π)n/2 if w θx+ (1−θ)y
≥ e−θ(1−θ)|x−y|2/2u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dγ ≥ Z
u dγ θ Z
v dγ 1−θ
choose θ= 1/2, w ≡1, v =χA, u =ed(·,A)2/4 Z
ed(·,A)2/4dγ ≤ 1 γ(A) γ(Ar) ≥ 1−2e−r2/4, γ(A) ≥ 1
2 extension : dµ=e−Vdx, V00≥c >0
dx → dγ(x) =e−|x|2/2 dx (2π)n/2 if w θx+ (1−θ)y
≥ e−θ(1−θ)|x−y|2/2u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dγ ≥ Z
u dγ θ Z
v dγ 1−θ
choose θ= 1/2, w ≡1, v =χA, u =ed(·,A)2/4 Z
ed(·,A)2/4dγ ≤ 1 γ(A) γ(Ar) ≥ 1−2e−r2/4, γ(A) ≥ 1
2 extension : dµ=e−Vdx, V00≥c >0
dx → dγ(x) =e−|x|2/2 dx (2π)n/2 if w θx+ (1−θ)y
≥ e−θ(1−θ)|x−y|2/2u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dγ ≥ Z
u dγ θ Z
v dγ 1−θ
choose θ= 1/2, w ≡1, v =χA,
u =ed(·,A)2/4 Z
ed(·,A)2/4dγ ≤ 1 γ(A) γ(Ar) ≥ 1−2e−r2/4, γ(A) ≥ 1
2 extension : dµ=e−Vdx, V00≥c >0
dx → dγ(x) =e−|x|2/2 dx (2π)n/2 if w θx+ (1−θ)y
≥ e−θ(1−θ)|x−y|2/2u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dγ ≥ Z
u dγ θ Z
v dγ 1−θ
choose θ= 1/2, w ≡1, v =χA, u =ed(·,A)2/4
Z
ed(·,A)2/4dγ ≤ 1 γ(A) γ(Ar) ≥ 1−2e−r2/4, γ(A) ≥ 1
2 extension : dµ=e−Vdx, V00≥c >0
dx → dγ(x) =e−|x|2/2 dx (2π)n/2 if w θx+ (1−θ)y
≥ e−θ(1−θ)|x−y|2/2u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dγ ≥ Z
u dγ θ Z
v dγ 1−θ
choose θ= 1/2, w ≡1, v =χA, u =ed(·,A)2/4 Z
ed(·,A)2/4dγ ≤ 1 γ(A)
γ(Ar) ≥ 1−2e−r2/4, γ(A) ≥ 1 2 extension : dµ=e−Vdx, V00≥c >0
dx → dγ(x) =e−|x|2/2 dx (2π)n/2 if w θx+ (1−θ)y
≥ e−θ(1−θ)|x−y|2/2u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dγ ≥ Z
u dγ θ Z
v dγ 1−θ
choose θ= 1/2, w ≡1, v =χA, u =ed(·,A)2/4 Z
ed(·,A)2/4dγ ≤ 1 γ(A) γ(Ar) ≥ 1−2e−r2/4, γ(A) ≥ 1
2
extension : dµ=e−Vdx, V00≥c >0
dx → dγ(x) =e−|x|2/2 dx (2π)n/2 if w θx+ (1−θ)y
≥ e−θ(1−θ)|x−y|2/2u(x)θv(y)1−θ, x,y ∈Rn
then Z
w dγ ≥ Z
u dγ θ Z
v dγ 1−θ
choose θ= 1/2, w ≡1, v =χA, u =ed(·,A)2/4 Z
ed(·,A)2/4dγ ≤ 1 γ(A) γ(Ar) ≥ 1−2e−r2/4, γ(A) ≥ 1
2 extension : dµ=e−Vdx, V00≥c >0
measure/information theoretic description : transportation cost inequality
ν probability measure on Rn, ν << γ W2(ν, γ)2 ≤ H(ν|γ)
W2(ν, γ)2 = inf
ν←π→γ
Z Z 1
2|x−y|2dπ(x,y) Wasserstein distance
H(ν|γ) = Z
log dν dγ dν relative entropy M. Talagrand 1996
measure/information theoretic description : transportation cost inequality
ν probability measure on Rn, ν << γ W2(ν, γ)2 ≤ H(ν|γ)
W2(ν, γ)2 = inf
ν←π→γ
Z Z 1
2|x−y|2dπ(x,y) Wasserstein distance
H(ν|γ) = Z
log dν dγ dν relative entropy M. Talagrand 1996
measure/information theoretic description : transportation cost inequality
ν probability measure on Rn, ν << γ W2(ν, γ)2 ≤ H(ν|γ)
W2(ν, γ)2 = inf
ν←π→γ
Z Z 1
2|x−y|2dπ(x,y) Wasserstein distance
H(ν|γ) = Z
log dν dγ dν relative entropy M. Talagrand 1996
measure/information theoretic description : transportation cost inequality
ν probability measure on Rn, ν << γ W2(ν, γ)2 ≤ H(ν|γ)
W2(ν, γ)2 = inf
ν←π→γ
Z Z 1
2|x−y|2dπ(x,y) Wasserstein distance
H(ν|γ) = Z
log dν dγ dν relative entropy
M. Talagrand 1996
measure/information theoretic description : transportation cost inequality
ν probability measure on Rn, ν << γ W2(ν, γ)2 ≤ H(ν|γ)
W2(ν, γ)2 = inf
ν←π→γ
Z Z 1
2|x−y|2dπ(x,y) Wasserstein distance
H(ν|γ) = Z
log dν dγ dν relative entropy M. Talagrand 1996
concentrationvia the transportation cost inequality (K. Marton)
A,B⊂Rn, d(A,B)≥r>0 γA = γ(· |A), γB = γ(· |B)
W2(γA, γB) ≤ log 1
γ(A) 1/2
+ log 1
γ(B) 1/2
√r
2 ≤ W2(γA, γB) ≤ log 1
γ(A) 1/2
+
log 1 γ(B)
1/2
γ(A)≥1/2, B =complementof Ar
γ(Ar) ≥ 1−e−r2/4, r ≥r0
functional description : logarithmic Sobolev inequality
f ≥0 smooth, Z
f dγ = 1 Z
f logf dγ ≤ 1 2
Z |∇f|2 f dγ
Rf logf dγ = H(fdγ|dγ) entropy Z |∇f|2
f dγ Fisher information f →f2:
Z
f2logf2dγ ≤ 2 Z
|∇f|2dγ
A. Stam 1959, L. Gross 1975
functional description : logarithmic Sobolev inequality
f ≥0 smooth, Z
f dγ = 1
Z
f logf dγ ≤ 1 2
Z |∇f|2 f dγ
Rf logf dγ = H(fdγ|dγ) entropy Z |∇f|2
f dγ Fisher information f →f2:
Z
f2logf2dγ ≤ 2 Z
|∇f|2dγ
A. Stam 1959, L. Gross 1975
functional description : logarithmic Sobolev inequality
f ≥0 smooth, Z
f dγ = 1 Z
f logf dγ ≤ 1 2
Z |∇f|2 f dγ
Rf logf dγ = H(fdγ|dγ) entropy Z |∇f|2
f dγ Fisher information f →f2:
Z
f2logf2dγ ≤ 2 Z
|∇f|2dγ
A. Stam 1959, L. Gross 1975
functional description : logarithmic Sobolev inequality
f ≥0 smooth, Z
f dγ = 1 Z
f logf dγ ≤ 1 2
Z |∇f|2 f dγ
Rf logf dγ = H(fdγ|dγ) entropy
Z |∇f|2
f dγ Fisher information f →f2:
Z
f2logf2dγ ≤ 2 Z
|∇f|2dγ
A. Stam 1959, L. Gross 1975
functional description : logarithmic Sobolev inequality
f ≥0 smooth, Z
f dγ = 1 Z
f logf dγ ≤ 1 2
Z |∇f|2 f dγ
Rf logf dγ = H(fdγ|dγ) entropy Z |∇f|2
f dγ Fisher information
f →f2: Z
f2logf2dγ ≤ 2 Z
|∇f|2dγ
A. Stam 1959, L. Gross 1975
functional description : logarithmic Sobolev inequality
f ≥0 smooth, Z
f dγ = 1 Z
f logf dγ ≤ 1 2
Z |∇f|2 f dγ
Rf logf dγ = H(fdγ|dγ) entropy Z |∇f|2
f dγ Fisher information f →f2:
Z
f2logf2dγ ≤ 2 Z
|∇f|2dγ
A. Stam 1959, L. Gross 1975
functional description : logarithmic Sobolev inequality
f ≥0 smooth, Z
f dγ = 1 Z
f logf dγ ≤ 1 2
Z |∇f|2 f dγ
Rf logf dγ = H(fdγ|dγ) entropy Z |∇f|2
f dγ Fisher information f →f2:
Z
f2logf2dγ ≤ 2 Z
|∇f|2dγ
A. Stam 1959, L. Gross 1975
concentrationvia the logarithmic Sobolev inequality(I. Herbst)
F :Rn→R 1-Lipschitz, R
F dγ = 0
f =eλF/R
eλFdγ, λ∈R differential inequalityon R
eλFdγ Z
eλFdγ ≤ eλ2/2, λ∈R γ {F <r}
≥ 1−e−r2/2, r >0
Brunn-Minkowski inequality
logarithmic Sobolev inequality
transportation cost inequality
hierarchy
Brunn-Minkowski inequality
⇓
logarithmic Sobolev inequality
⇓
transportation cost inequality
⇓
dimension free measure concentration
tools toestablish theseinequalities parametrisation methods
hierarchy
Brunn-Minkowski inequality
⇓
logarithmic Sobolev inequality
⇓
transportation cost inequality
⇓
dimension free measure concentration
tools toestablish theseinequalities parametrisation methods
hierarchy
Brunn-Minkowski inequality
⇓
logarithmic Sobolev inequality
⇓
transportation cost inequality
⇓
dimension free measure concentration
tools toestablish theseinequalities
parametrisation methods
hierarchy
Brunn-Minkowski inequality
⇓
logarithmic Sobolev inequality
⇓
transportation cost inequality
⇓
dimension free measure concentration
tools toestablish theseinequalities parametrisation methods
heat kernel parametrisation
logarithmic Sobolev inequality R
f logf dγ ≤ 12R |∇f|2 f dγ generator Lf = ∆f −x· ∇f, Pt=etL semigroup
γ invariant measure
d dt
Z
Ptf logPtf dγ = −1 2
Z |∇Ptf|2 Ptf dγ commutation |∇Ptf| ≤ e−tPt(|∇f|)
equivalent to a curvature condition
heat kernel parametrisation
logarithmic Sobolev inequality R
f logf dγ ≤ 12R |∇f|2 f dγ
generator Lf = ∆f −x· ∇f, Pt=etL semigroup γ invariant measure
d dt
Z
Ptf logPtf dγ = −1 2
Z |∇Ptf|2 Ptf dγ commutation |∇Ptf| ≤ e−tPt(|∇f|)
equivalent to a curvature condition
heat kernel parametrisation
logarithmic Sobolev inequality R
f logf dγ ≤ 12R |∇f|2 f dγ generator Lf = ∆f −x· ∇f,
Pt=etL semigroup γ invariant measure
d dt
Z
Ptf logPtf dγ = −1 2
Z |∇Ptf|2 Ptf dγ commutation |∇Ptf| ≤ e−tPt(|∇f|)
equivalent to a curvature condition
heat kernel parametrisation
logarithmic Sobolev inequality R
f logf dγ ≤ 12R |∇f|2 f dγ generator Lf = ∆f −x· ∇f, Pt=etL semigroup
γ invariant measure
d dt
Z
Ptf logPtf dγ = −1 2
Z |∇Ptf|2 Ptf dγ commutation |∇Ptf| ≤ e−tPt(|∇f|)
equivalent to a curvature condition
heat kernel parametrisation
logarithmic Sobolev inequality R
f logf dγ ≤ 12R |∇f|2 f dγ generator Lf = ∆f −x· ∇f, Pt=etL semigroup
γ invariant measure
d dt
Z
Ptf logPtf dγ = −1 2
Z |∇Ptf|2 Ptf dγ commutation |∇Ptf| ≤ e−tPt(|∇f|)
equivalent to a curvature condition
heat kernel parametrisation
logarithmic Sobolev inequality R
f logf dγ ≤ 12R |∇f|2 f dγ generator Lf = ∆f −x· ∇f, Pt=etL semigroup
γ invariant measure
d dt
Z
Ptf logPtf dγ = −1 2
Z |∇Ptf|2 Ptf dγ
commutation |∇Ptf| ≤ e−tPt(|∇f|) equivalent to a curvature condition
heat kernel parametrisation
logarithmic Sobolev inequality R
f logf dγ ≤ 12R |∇f|2 f dγ generator Lf = ∆f −x· ∇f, Pt=etL semigroup
γ invariant measure
d dt
Z
Ptf logPtf dγ = −1 2
Z |∇Ptf|2 Ptf dγ commutation |∇Ptf| ≤ e−tPt(|∇f|)
equivalent to a curvature condition
heat kernel parametrisation
logarithmic Sobolev inequality R
f logf dγ ≤ 12R |∇f|2 f dγ generator Lf = ∆f −x· ∇f, Pt=etL semigroup
γ invariant measure
d dt
Z
Ptf logPtf dγ = −1 2
Z |∇Ptf|2 Ptf dγ commutation |∇Ptf| ≤ e−tPt(|∇f|)
equivalent to a curvature condition
extensions
dµ=e−Vdx on Rn, V00≥c >0 Riemannian manifolds Ric≥c >0
Bochner’s formula 1
2∆ |∇f|2
− ∇f · ∇(∆f) = kHess(f)k2+Ric(∇f,∇f) ≥ c|∇f|2
equivalent to |∇Ptf| ≤ e−ctPt(|∇f|) (Gauss space : curvature 1 )
manifolds withweights dµ=e−Vdx, Ric+Hess(V)≥c >0
extensions
dµ=e−Vdx on Rn, V00≥c >0
Riemannian manifolds Ric≥c >0 Bochner’s formula
1
2∆ |∇f|2
− ∇f · ∇(∆f) = kHess(f)k2+Ric(∇f,∇f) ≥ c|∇f|2
equivalent to |∇Ptf| ≤ e−ctPt(|∇f|) (Gauss space : curvature 1 )
manifolds withweights dµ=e−Vdx, Ric+Hess(V)≥c >0
extensions
dµ=e−Vdx on Rn, V00≥c >0 Riemannian manifolds Ric≥c >0
Bochner’s formula 1
2∆ |∇f|2
− ∇f · ∇(∆f) = kHess(f)k2+Ric(∇f,∇f) ≥ c|∇f|2
equivalent to |∇Ptf| ≤ e−ctPt(|∇f|) (Gauss space : curvature 1 )
manifolds withweights dµ=e−Vdx, Ric+Hess(V)≥c >0
extensions
dµ=e−Vdx on Rn, V00≥c >0 Riemannian manifolds Ric≥c >0
Bochner’s formula 1
2∆ |∇f|2
− ∇f · ∇(∆f) = kHess(f)k2+Ric(∇f,∇f)
≥ c|∇f|2
equivalent to |∇Ptf| ≤ e−ctPt(|∇f|) (Gauss space : curvature 1 )
manifolds withweights dµ=e−Vdx, Ric+Hess(V)≥c >0