20.04.2014 Views

Uniform Log-Sobolev inequality for Boltzmann measures with ...

Uniform Log-Sobolev inequality for Boltzmann measures with ...

Uniform Log-Sobolev inequality for Boltzmann measures with ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>Uni<strong>for</strong>m</strong> <strong>Log</strong>-<strong>Sobolev</strong> <strong>inequality</strong> <strong>for</strong><br />

<strong>Boltzmann</strong> <strong>measures</strong> <strong>with</strong> exterior<br />

magnetic field over spheres ∗<br />

Zhengliang Zhang, † Bin Qian, ‡ Yutao Ma §<br />

April 2009<br />

Abstract<br />

In this paper, We obtain the uni<strong>for</strong>m <strong>Log</strong>-<strong>Sobolev</strong> <strong>inequality</strong> <strong>for</strong> the <strong>Boltzmann</strong><br />

<strong>measures</strong> by reducing multi-dimensional <strong>measures</strong> to one-dimensional<br />

<strong>measures</strong>, and then applying the characterization on the constant of <strong>Log</strong>-<br />

<strong>Sobolev</strong> <strong>inequality</strong> <strong>for</strong> a probability measure on the real line .<br />

Keywords: <strong>Boltzmann</strong> measure, <strong>Log</strong>-<strong>Sobolev</strong> <strong>inequality</strong>, T 2 -transportation <strong>inequality</strong>,<br />

Poincaré <strong>inequality</strong>.<br />

AMS Classification Subjects 2000: 60E15 39B62 26Dxx<br />

1 Introduction<br />

Let S n−1 be the unit sphere in R n (n ≥ 3), and µ the normalized Lebesgue measure<br />

on S n−1 , i.e. µ = σ n−1 /s n−1 , where σ n−1 and s n−1 are the uni<strong>for</strong>m surface measure<br />

and total volume respectively on S n−1 . For any h > 0 and e 1 = (1, 0, · · · , 0) ∈ S n−1 ,<br />

let µ h be the probability on S n−1 given by<br />

dµ h (x) = eh<br />

c n (h) dµ(x), x ∈ Sn−1 ,<br />

where c n (h) is the normalizing constant. This is the so-called <strong>Boltzmann</strong> measure<br />

<strong>with</strong> exterior magnetic field [9].<br />

∗ Research supported by National Science Funds of China (10626041, 10871153) and China<br />

Scholarship council 2007U13020.<br />

† Department of Mathematics and Statistics, Wuhan University, 430072-Hubei, China. E-mail<br />

address: zzl731205@yahoo.com.cn; zhlzhang.math@whu.edu.cn<br />

‡ Department of Mathematics and Statistics, Wuhan University, 430072-Hubei, China. E-mail<br />

address: binqiancn@yahoo.com.cn<br />

§ Department of Mathematics, Beijing Normal University, 100875-Beijing, China. E-mail address:<br />

mayt@bnu.edu.cn<br />

1


We say that a probability measure µ on a manifold X satisfies a logarithmic<br />

<strong>Sobolev</strong> <strong>inequality</strong> if there exists a constant C > 0 such that <strong>for</strong> every smooth<br />

function f : X → R, one has<br />

∫<br />

Ent µ (f 2 ) ≤ C |∇f| 2 dµ,<br />

where the entropy is defined by<br />

∫<br />

(∫<br />

Ent µ (f 2 ) = f 2 log f 2 dµ −<br />

) (∫<br />

f 2 dµ log<br />

)<br />

f 2 dµ .<br />

To be short we write µ ∈ LSI(C) <strong>for</strong> this <strong>inequality</strong>. Its property was well introduced<br />

by Gross [11]. Though here we just study the <strong>Log</strong>-<strong>Sobolev</strong> <strong>inequality</strong>, it’s<br />

well-known that Poincaré, transportation and <strong>Log</strong>-<strong>Sobolev</strong> inequalities are essential<br />

tools in the study of concentration of <strong>measures</strong> (see e.g. [12], [10]), and there exist<br />

some relations among the three types of inequalities: log-<strong>Sobolev</strong> <strong>inequality</strong> is<br />

strictly stronger than Talagrand’s transportation cost <strong>inequality</strong> ([12], [6] and [2]),<br />

which is stronger than Poincaré <strong>inequality</strong> (see [4], section 4.1).<br />

The main method we applied is to reduce the proof of the <strong>measures</strong> µ x on spheres<br />

to that of a probability <strong>measures</strong> ν x on [0, π], and then give an estimate on the<br />

characterization of a probability measure on R which satisfies a logarithmic <strong>Sobolev</strong><br />

<strong>inequality</strong>.<br />

2 Main result<br />

In this paper, we get the uni<strong>for</strong>m <strong>Log</strong>-<strong>Sobolev</strong> <strong>inequality</strong> in h > 0 <strong>for</strong> the <strong>Boltzmann</strong><br />

<strong>measures</strong>. The main results are shown as follows:<br />

Theorem 2.1. For any h > 0, the probability <strong>measures</strong> µ h satisfy the uni<strong>for</strong>m <strong>Log</strong>-<br />

<strong>Sobolev</strong> <strong>inequality</strong>, that is, <strong>for</strong> any smooth function f : S n−1 → R, the following<br />

<strong>inequality</strong> holds:<br />

Ent µh (f 2 ) ≤ C LS<br />

∫S n−1 |∇ S n−1f| 2 dµ,<br />

where C LS is independent of h > 0, but depends on the dimension n.<br />

We know that <strong>Log</strong>-<strong>Sobolev</strong> <strong>inequality</strong> is stronger than T 2 -transportation <strong>inequality</strong>,<br />

and T 2 -transportation <strong>inequality</strong> stronger than Poincaré <strong>inequality</strong>, so we have<br />

the following corollaries.<br />

Corollary 2.2. For any h > 0, the probability <strong>measures</strong> µ h ∈ P I(C P ), C P independent<br />

of h, that is, <strong>for</strong> any smooth function f : S n−1 → R, if the following <strong>inequality</strong><br />

holds:<br />

where C P is independent of h > 0.<br />

Var µh (f) ≤ C P<br />

∫S n−1 |∇ S n−1f| 2 dµ,<br />

2


Corollary 2.3. For any h > 0, the probability <strong>measures</strong> µ h ∈ T 2 (C), C independent<br />

of h.<br />

Remark: In fact, in [7], F.Q. Gao and L.M. Wu have already given the uni<strong>for</strong>m<br />

Poincaré <strong>inequality</strong> <strong>for</strong> the <strong>Boltzmann</strong> <strong>measures</strong>. Moreover, they find a more precise<br />

uni<strong>for</strong>m spectral gap. By their result on Poincaré <strong>inequality</strong> and an additional<br />

condition given by Miao and the first author in [15], we can also obtain Corollary<br />

2.3 quickly <strong>with</strong>out Theorem 2.1 .<br />

3 Idea of Proof<br />

In this paper, following an idea in [2], the method consists in reducing the proof<br />

to some one dimensional <strong>measures</strong> <strong>for</strong> which precise estimates of the log-<strong>Sobolev</strong><br />

constant are known see [3].<br />

Notice that the case h = 0, i.e. µ h = µ, is the simplest one, and <strong>for</strong> any h > 0<br />

fixed, it’s well-known that <strong>Log</strong>-<strong>Sobolev</strong> <strong>inequality</strong> holds <strong>for</strong> the <strong>measures</strong> µ h , so we<br />

just need to consider the cases that h is large enough. Through the whole paper,<br />

we assume that h is large enough.<br />

Let ν h be the image of µ h by the map x → d(e 1 , x). It is a probability on [0, π]<br />

<strong>with</strong> density<br />

ρ n,h := dν h<br />

dθ = 1<br />

C n (h) eh cos θ (sin θ) n−2 ,<br />

where C n (h) is the normalizing constant, whose order of h is e h h − n−1<br />

2 as h goes to<br />

infinity (see Lemma 5.1).<br />

In [2], F. Barthe and the first author give the following lemma, which makes it<br />

possible <strong>for</strong> us to reduce to the discussion on one-dimensional <strong>measures</strong>.<br />

Lemma 3.1. Let λ be a probability measure on S n−1 <strong>with</strong><br />

dλ(x) = φ(d(x, e 1 ))dσ n−1 (x), x ∈ S n−1 ,<br />

where φ is measurable. Let ν be the image of λ by the map x → d(x, e 1 ). Assume that<br />

ν satisfies a <strong>Log</strong>-<strong>Sobolev</strong> <strong>inequality</strong> <strong>with</strong> constant C LS , then λ satisfies a <strong>Log</strong>-<strong>Sobolev</strong><br />

<strong>inequality</strong> <strong>with</strong> constant max(C LS , c n−2 ), where c n−2 is the <strong>Log</strong>-<strong>Sobolev</strong> constant of<br />

the uni<strong>for</strong>m probability σ n−2 /s n−2 on S n−2 , that is, <strong>for</strong> every smooth function f :<br />

S n−1 → R,<br />

∫<br />

Ent λ (f 2 ) ≤ max(C LS , c n−2 ) |∇ S n−1f| 2 dλ.<br />

S n−1<br />

As far as the measure µ h is concerned, by the Lemma 3.1 we just need to discuss<br />

the measure ν h . The generator associated <strong>with</strong> ν h is<br />

L h f(θ) = d2<br />

dθ 2 f + (−h sin θ + (n − 2) cot θ) d dθ f<br />

It’s well-known that Bakry-Emery criteria gives a curvature condition to get the<br />

<strong>Log</strong>-<strong>Sobolev</strong> <strong>inequality</strong>. However, the curvature condition requires a non-negative<br />

lower bound.<br />

3


Theorem 3.2. (Bakry and Emery [1]) Let dν = e −Ψ dx be a probability measure<br />

on a manifold M, such that Ψ ∈ C 2 (M) and D 2 Ψ + Ric ≥ ρI n , ρ > 0. Then<br />

dν ∈ LSI(ρ).<br />

For the measure dν h = e h cos θ (sin θ) n−2 dθ/C n (h) on [0, π], we have<br />

Ψ = −(h cos θ + (n − 2) log sin θ) + log C n (h),<br />

D 2 Ψ = h cos θ + (n − 2) csc 2 θ.<br />

Since θ = π − ε the second derivative D 2 Ψ behaves like −h + (n − 2)ε −2 , the<br />

curvature (or the second derivative if you prefer) is not bounded below by some<br />

positive constant <strong>for</strong> all h, and then Bakry-Emery criterion does not apply.<br />

4 Proofs of main result<br />

Here we’ll make use of the refined characterization from Barthe and Roberto [3]. Of<br />

course, we may also use that one from Bobkov, and Götze [5], after all, we don’t<br />

want to give a sharp estimate on the logarithmic <strong>Sobolev</strong> constants. Now we state<br />

the refined characterization as follows.<br />

Theorem 4.1. ([3]) Let µ, ν be Borel <strong>measures</strong> on R <strong>with</strong> µ(R) = 1 and dν(x) =<br />

n(x)dx. Let m be a median of µ. Let C be the optimal constant such that <strong>for</strong> every<br />

smooth function f : R → R, one has<br />

∫<br />

Ent µ (f 2 ) ≤ C f ′2 dν.<br />

Then max(b − , b + ) ≤ C ≤ 4 max(B − , B + ), where<br />

(<br />

) ∫<br />

1<br />

x<br />

1<br />

b + = sup µ([x, ∞)) log 1 +<br />

x>m<br />

2µ([x, ∞)) m n ,<br />

(<br />

)<br />

e 2 ∫ x<br />

1<br />

B + = sup µ([x, ∞)) log 1 +<br />

x>m<br />

µ([x, ∞)) m n ,<br />

(<br />

) ∫<br />

1<br />

m<br />

1<br />

b − = sup µ((−∞, x]) log 1 +<br />

x


where m h is the median of ν h , and it tends to 0 as h → +∞ (see Lemma 5.2 in<br />

appendix part). Moreover, we also give the speed rate of m h : there exist positive<br />

constants M 0 , M, s.t. M 0 ≤ h(1 − cos m h )) ≤ M (see Lemma 5.3).<br />

By a change of variables θ = arccos x, <strong>for</strong> arbitrary α, β ∈ (0, π), we have<br />

∫ β<br />

α<br />

ρ n,h (θ)dθ =<br />

∫ β<br />

α<br />

∫ β<br />

α<br />

e h cos θ (sin θ) n−2<br />

C n (h)<br />

1<br />

ρ n,h (θ) dθ =<br />

∫ cos α<br />

cos β<br />

∫ cos α<br />

dθ =<br />

cos β<br />

C n (h)<br />

e hx (1 − x 2 ) n−1<br />

2<br />

e hx (1 − x 2 ) n−3<br />

2<br />

C n (h)<br />

Since we just want to prove that C LS is bounded, c = 1/2 or c = e 2 has no direct<br />

influence on our proof. For simplicity, we’ll just estimate S − (h, 1) and S + (h, 1),<br />

denoted by S − (h), and S + (h) respectively <strong>for</strong> short. In order to avoid some trouble,<br />

we’ll denote all the constants independent of h by the uni<strong>for</strong>m constant C.<br />

(I) First we study the supper bound on S − (h). Let r = h(1 − t),<br />

S − (h) =<br />

sup<br />

α∈(0,m h )<br />

= sup<br />

t∈(cos m h ,1)<br />

(∫ α<br />

)<br />

ρ n,h (θ)dθ log<br />

0<br />

∫ 1<br />

·<br />

t<br />

(<br />

1 +<br />

e hx (1 − x 2 ) n−3<br />

2 dx log<br />

(1 +<br />

( ∫ t<br />

cos(m h )<br />

dx<br />

e hx (1 − x 2 ) n−1<br />

2<br />

1<br />

∫ α<br />

ρ 0 n,h(θ)dθ<br />

Since 0 < cos m h < t < 1 (when h is large enough), we have<br />

∫ 1<br />

t<br />

∫ 1<br />

e hx (1 − x) n−3<br />

2 dx ≤<br />

∫ 1<br />

t<br />

t<br />

)<br />

e hx (1 − x 2 ) n−3<br />

n−3<br />

2 dx ≤ 2 2<br />

e hx (1 − x) n−3<br />

2 dx =<br />

e h<br />

h n−1<br />

2<br />

.<br />

∫ h(1−t)<br />

0<br />

dx.<br />

α<br />

dx,<br />

) (∫ mh<br />

)<br />

1<br />

ρ n,h (θ) dθ<br />

C n (h)<br />

∫ 1<br />

t<br />

ehx (1 − x 2 ) n−3<br />

2 dx<br />

∫ 1<br />

t<br />

e −y y n−3<br />

2 dy,<br />

and (noting that 0 < M 0 ≤ h(1 − cos m h ) ≤ M (see Lemma 5.3))<br />

)<br />

e hx (1 − x) n−3<br />

2 dx,<br />

∫ t<br />

cos(m h )<br />

dx<br />

e hx (1 − x 2 ) n−1<br />

2<br />

≤ h ∫ n−3<br />

2 h(1−cos mh )<br />

e h h(1−t)<br />

e y dy<br />

y n−1<br />

2<br />

≤ h n−3<br />

2<br />

e h<br />

∫ M<br />

h(1−t)<br />

e y dy<br />

,<br />

y n−1<br />

2<br />

Thus <strong>with</strong> C n (h)h n−1<br />

2 e −h ≤ C (see Lemma 5.1)), we get<br />

S − (h) ≤<br />

≤<br />

sup<br />

r∈(0,M)<br />

sup<br />

r∈(0,M)<br />

2 n−3<br />

2<br />

h<br />

2 n−3<br />

2<br />

h<br />

∫ r<br />

0<br />

r n−1<br />

2<br />

n−1<br />

2<br />

2 n+1<br />

2 e M<br />

≤<br />

(n − 1)(n − 3) · 1<br />

h<br />

e −y y n−3<br />

2 dy log<br />

(1 +<br />

(<br />

log<br />

1 + n − 1<br />

2<br />

(<br />

sup r log<br />

r∈(0,M)<br />

C<br />

∫ r<br />

0 e−y y n−3<br />

Cr − n−1<br />

2<br />

) e<br />

M<br />

2 dy<br />

n−3<br />

r − n−3<br />

2<br />

2<br />

1 + n − 1 Cr − n−1<br />

2<br />

2<br />

) (∫ M<br />

)<br />

r<br />

)<br />

e y dy<br />

y n−1<br />

2<br />

5


Clearly, S − (h) is uni<strong>for</strong>mly bounded in h > 0.<br />

(II) Second we study the supper bound on S + (h).<br />

(∫ π<br />

) (<br />

) (∫<br />

1<br />

α<br />

)<br />

1<br />

S + (h) = sup ρ n,h (θ)dθ log 1 + ∫ π<br />

α∈(m h ,π) α<br />

ρ α n,h(θ)dθ m h<br />

ρ n,h (θ) dθ<br />

∫ )<br />

t<br />

= sup e hx (1 − x 2 ) n−3<br />

C n (h)<br />

2 dx log<br />

(1 + ∫ t<br />

t∈(−1,cos m h ) −1<br />

−1 ehx (1 − x 2 ) n−3<br />

2 dx<br />

( ∫ )<br />

cos mh<br />

dx<br />

·<br />

t<br />

e hx (1 − x 2 ) n−1<br />

2<br />

(1) When −1 < t < 0,<br />

First we work on the first term of S + (h) in the right hand side,<br />

∫ t<br />

−1<br />

∫ t<br />

e hx (1 − x 2 ) n−3<br />

n−3<br />

2 dx ≤ 2 2<br />

then on the third term,<br />

−1<br />

e hx (1 + x) n−3 2 n−3<br />

2<br />

2 dx ≤<br />

h n−1<br />

0<br />

2 e h ∫ h(1+t)<br />

e y y n−3<br />

2 dy,<br />

∫ cos mh<br />

t<br />

dx<br />

e hx (1 − x 2 ) n−1<br />

2<br />

≤<br />

∫ 0<br />

t<br />

dx<br />

e hx (1 + x) n−1<br />

2<br />

+<br />

∫ cos mh<br />

0<br />

dx<br />

e hx (1 − x) n−1<br />

2<br />

,<br />

notice that<br />

∫ cos mh<br />

0<br />

dx<br />

e hx (1 − x) n−1<br />

2<br />

∫ cos mh<br />

≤ 2 n−1<br />

2 +<br />

1/2<br />

≤ 2 n−1<br />

2 +<br />

dx<br />

e hx (1 − x) n−1<br />

2<br />

e −h/2 h n−1<br />

2<br />

(h(1 − cos m h )) n−1<br />

2<br />

≤ C (independent of h),<br />

and<br />

∫ 0<br />

t<br />

dx<br />

e hx (1 + x) n−1<br />

2<br />

∫ h<br />

= h n−3<br />

2 e<br />

h<br />

h(1+t)<br />

dy<br />

e y y n−1<br />

2<br />

∫ +∞<br />

≤ h n−3<br />

2 e<br />

h<br />

h(1+t)<br />

dy<br />

e y y n−1<br />

2<br />

.<br />

Let q = h(1 + t), by the monotonicity of x log(1 + C ) in x > 0,<br />

x<br />

S + (h) ≤ sup C2 n−3<br />

2<br />

q∈(0,h)<br />

+ sup<br />

q∈(0,h)<br />

2 n−3<br />

2<br />

h<br />

h n−1<br />

∫ q<br />

0<br />

∫<br />

1 q<br />

2 e h<br />

0<br />

e y y n−3<br />

2 dy log<br />

(<br />

=: sup I 1 (h) + sup I 2 (h).<br />

q∈(0,h)<br />

q∈(0,h)<br />

i). When lim h→+∞ q = 0,<br />

e y y n−3<br />

2 dy log<br />

(<br />

6<br />

1 + h n−1<br />

2 e h C n (h)<br />

1 + h n−1<br />

2 e h C n (h)<br />

∫<br />

2 n−3 q<br />

2<br />

0 ey y n−3<br />

∫<br />

2 n−3 q<br />

2<br />

0 ey y n−3<br />

2 dy<br />

2 dy<br />

) ∫ +∞<br />

q<br />

)<br />

dy<br />

e y y n−1<br />

2


I 2 (h) ≤ 2 n−3<br />

2<br />

h<br />

= 2 n−3<br />

2<br />

n−1<br />

2<br />

∫ q<br />

0<br />

n−3<br />

2<br />

y n−3<br />

2 dy<br />

∫ +∞<br />

q n−1<br />

2<br />

h<br />

1<br />

q n−3<br />

2<br />

≤C q (<br />

)<br />

h log 1 + Ce 2h q − n−1<br />

2 → 0.<br />

I 1 (h) → 0,<br />

)<br />

dy<br />

log<br />

(1 + h n−1<br />

2 e h C n (h)<br />

∫<br />

q y n−1<br />

2 2 n−3 q<br />

2 y n−3<br />

2 dy<br />

0<br />

(<br />

n−1<br />

2<br />

log 1 +<br />

h )<br />

n−1<br />

2 e h C n (h)<br />

, (C<br />

2 n−3<br />

2 q n−1<br />

n (h) ∼ eh<br />

2<br />

ii). When lim h→+∞ q = +∞ : note that 0 < q < h,<br />

(<br />

)<br />

I 1 (h) ≤ C2 n−3 q n−3<br />

2 e q<br />

2<br />

h n−1<br />

2 e log 1 + h n−1<br />

2 e h C n (h)<br />

h 2 n−3<br />

2 q n−3<br />

2 e q<br />

≤ C log(1 + Ce2h )<br />

h<br />

I 2 (h) ≤ 2 n−3<br />

2 q n−3<br />

2 e q<br />

log(1 + Ce 2h )e −q 2 n−3<br />

q− 2 = C log(1 + Ce2h )<br />

h<br />

n − 3 h<br />

h n−1<br />

2<br />

Clearly I 1 (h), I 2 (h) are bounded in h in the case −1 < t < 0, so is S + (h).<br />

(2) When 0 ≤ t < cos m h :<br />

Let p = h(1 − t) ≥ h(1 − cos m h ) ≥ M 0 > 0,<br />

∫ t<br />

−1<br />

∫ cos mh<br />

t<br />

∫ t<br />

e hx (1 − x 2 ) n−3<br />

n−3<br />

2 dx ≤ 2 2<br />

dx<br />

e hx (1 − x 2 ) n−1<br />

2<br />

≤ h n−3<br />

2<br />

e h<br />

−1<br />

e hx (1 − x) n−3 2 n−3<br />

2 e h<br />

2 dx ≤<br />

∫ h(1−t)<br />

h(1−cos m h )<br />

By the monotonicity of x log(1 + C ) in x > 0,<br />

x<br />

S + (h) ≤<br />

≤<br />

sup<br />

p∈(M 0 ,2h)<br />

sup<br />

p∈(M 0 ,2h)<br />

·<br />

2 n−3<br />

2<br />

h<br />

(∫ p<br />

·<br />

2 n−1<br />

2<br />

p<br />

( ∫ p/2<br />

M 0<br />

∫ 2h<br />

p<br />

e −y y n−3<br />

2 dy log<br />

(<br />

e y y − n−1<br />

h(1−cos m h )<br />

∫ +∞<br />

p<br />

2 dy<br />

)<br />

e −y y n−3<br />

2 dy log<br />

(1 +<br />

e y y − n−1<br />

2 dy +<br />

∫ p<br />

p/2<br />

h n−1<br />

2<br />

e y y − n−1 h n−3<br />

2<br />

2 dy ≤<br />

e h<br />

∫ 2h<br />

p<br />

∫ p<br />

1 + e−h h n−1<br />

2 C n (h)<br />

2 n−3<br />

2<br />

∫ +∞<br />

∫ 2h<br />

p<br />

e −y y n−3<br />

2 dy<br />

C<br />

)<br />

e −y y n−3<br />

2 dy,<br />

M 0<br />

e y y − n−1<br />

2 dy.<br />

)<br />

)<br />

e<br />

p<br />

−y y n−3<br />

2 dy<br />

)<br />

e y y − n−1<br />

2 dy , (C n (h) ∼ eh<br />

h n−1<br />

2<br />

).<br />

7


Notice that<br />

∫ +∞<br />

p<br />

e −y y n−3<br />

n−1<br />

2 dy ≤ p 2<br />

∫ +∞<br />

1<br />

= e −p p n−1<br />

2<br />

≤ Ce −p p n−1<br />

2<br />

≤ Ce −p p n−3<br />

2 .<br />

e −pz z n−3<br />

2 dz<br />

∫ +∞<br />

0<br />

∫ +∞<br />

Then by the monotonicity of x log(1 + C ), we have<br />

x<br />

)<br />

S + (h) ≤<br />

≤<br />

sup<br />

p∈(M 0 ,2h)<br />

sup<br />

p∈(M 0 ,+∞)<br />

2 n−1<br />

2 Cp n−3<br />

2<br />

pe p<br />

C<br />

(<br />

e − p 2 p<br />

n−3<br />

(<br />

log<br />

0<br />

e −pz (z + 1) n−3<br />

2 dz<br />

e −pz e M 0z/2 dz<br />

(<br />

1 + Ce p p − n−3 2<br />

2 ·<br />

n − 3<br />

(<br />

) log<br />

2<br />

0 + 2 n−3<br />

2<br />

p<br />

2 M<br />

− n−3<br />

e p 2<br />

M n−3<br />

2<br />

0<br />

1 + Ce p p n−3<br />

2<br />

)<br />

+ ep 2 n−3<br />

2<br />

p n−3<br />

2<br />

From the quality above, we see that S + (h) is still bounded as p → +∞. There<strong>for</strong>e,<br />

we get that S + (h) is bounded in h in the case 0 < t < cos m h .<br />

Summarizing the cases (1) and (2), we obtain S + (h) is uni<strong>for</strong>mly bounded in<br />

h > 0.<br />

By the boundness of S − (h) and S + (h) and Theorem 4.1, we finish our proof.<br />

.<br />

)<br />

5 Appendix<br />

Lemma 5.1. There exist positive constants C 1 (n) and C 2 (n) such that <strong>for</strong> say h ≥ 1,<br />

C 1 (n)e h h − n−1<br />

2 ≤ C n (h) ≤ C 2 (n)e h h − n−1<br />

2 .<br />

Lemma 5.2. The median m h of the measure ν h tends to 0 as h → +∞.<br />

Proof. By the definition of median, we have<br />

∫ π<br />

e h cos θ (sin θ) n−2 dθ = C n(h)<br />

m h<br />

2<br />

However<br />

∫ π<br />

m h<br />

e h cos θ (sin θ) n−2 dθ =<br />

∫ cos mh<br />

−1<br />

≤ 2 n−3<br />

2 + 2 n−3<br />

2<br />

= 2 n−3<br />

2 + 2 n−3<br />

2<br />

e hx (1 − x 2 ) n−3<br />

2 dx<br />

∫ cos mh<br />

0<br />

e h<br />

h n−1<br />

2<br />

∫ h<br />

∼ e h h − n−1<br />

2 .<br />

e hx (1 − x) n−3<br />

2 dx<br />

h(1−cos m h )<br />

e −y y n−3<br />

2 dy<br />

∼ e h cos mh · {(1 − cos m h ) n−3<br />

2 h<br />

−1 ∨ h − n−1<br />

2 } (as h large).<br />

8


By the comparison of the order of h, cos m h must tend to 1 as h → +∞. Thus<br />

we reach the conclusion required.<br />

Lemma 5.3. {h(1 − cos m h ); h > 0} is uni<strong>for</strong>mly bounded in h. Moreover, there<br />

exist positive constants M 0 , M, s.t. 0 < M 0 ≤ h(1 − cos m h ) ≤ M when h is large<br />

enough.<br />

Proof. Since<br />

and by lim h→+∞ cos m h = 1<br />

∫ mh<br />

0<br />

e h cos θ (sin θ) n−2 dθ = C n(h)<br />

,<br />

2<br />

We have<br />

∫ mh<br />

0<br />

∫ 1<br />

e h cos θ (sin θ) n−2 dθ = e hx (1 − x 2 ) n−3<br />

2 dx<br />

cos m h<br />

≃2 n−3<br />

2<br />

∫ 1<br />

=2 n−3<br />

2 e h h − n−1<br />

2<br />

cos m h<br />

e hx (1 − x) n−3<br />

2 dx<br />

∫ h(1−cos mh )<br />

0<br />

e −y y n−3<br />

2 dy,<br />

∫ h(1−cos mh )<br />

0<br />

e −y y n−3<br />

2 dy ≃ 2<br />

1−n<br />

2 e −h h n−1<br />

2 Cn (h)<br />

{<br />

≤ 2 1−n<br />

2 e −h h n−1<br />

2 2 n−3<br />

2 + 2 n−3<br />

2<br />

≤ 1 2 h n−1<br />

2 e −h + 1 2<br />

→ 1 2<br />

∫ +∞<br />

0<br />

∫ +∞<br />

0<br />

e h<br />

h n−1<br />

2<br />

e −y y n−3<br />

2 dy<br />

∫ +∞<br />

e −y y n−3<br />

2 dy (as h → +∞).<br />

0<br />

e −y y n−3<br />

2 dy<br />

}<br />

Clearly, by the <strong>inequality</strong> above we know that h(1−cos m h ) doesn’t tend to infinity<br />

as h → +∞, that is, {h(1 − cos m h ); h > 0} uni<strong>for</strong>mly bounded in h. Moreover,<br />

from lim h→+∞ 2 1−n<br />

2 e −h h n−1<br />

2 C n (h) > 0, we have<br />

lim h(1 − cos m h) > 0.<br />

h→+∞<br />

Example : When n = 3, dν h (θ) = eh cos θ sin θ<br />

dθ, where C 3 (h) is the normalized<br />

C 3 (h)<br />

constant, given as follows:<br />

C 3 (h) =<br />

∫ π<br />

0<br />

e h cos θ sin θdθ = eh − e −h<br />

.<br />

h<br />

9


m h is the median of ν h , by the definition of median, we have<br />

Then<br />

Moreover,<br />

∫ mh<br />

0<br />

e h cos θ (sin θ) n−2 dθ = C 3(h)<br />

,<br />

2<br />

( )<br />

1<br />

m h = arccos<br />

h log eh + e −h<br />

.<br />

2<br />

lim h(1 − cos m h) = log 2.<br />

h→+∞<br />

References<br />

[1] D. Bakry and M. Emery, Diffusions hypercontractivies, In Sém. Proba. XIX,<br />

LNM, 1123, Springer (1985), 177-206.<br />

[2] F. Barthe and Z.L. Zhang, Poincare, transportation and logarithmic <strong>Sobolev</strong><br />

inequalities <strong>for</strong> harmonic <strong>measures</strong> on spheres, to submit.<br />

[3] F. Barthe and C. Roberto, <strong>Sobolev</strong> inequalities <strong>for</strong> probabilty <strong>measures</strong> on the<br />

real line, Studia Mathematica, 159 (3) (2003), 481-497.<br />

[4] S. G. Bobkov, I. Gentil and M. Ledoux, Hypercontractivity of Hamilton-Jacobi<br />

equations, J. Math. Pures Appl. 80 (2001) 669-696.<br />

[5] S. G. Bobkov and F. Götze, Exponential integrability and transportation cost<br />

related to logarithmic <strong>Sobolev</strong> inequalities, J. Funct. Anal. 163 (1999), 1-28.<br />

[6] P. Cattiaux and A. Guillin, On quadratic transpotation cost inequalities, J.<br />

Math. Pures Appl. 86 (2006) 342-361.<br />

[7] F.Q. Gao and L.M. Wu, Transportatoin-In<strong>for</strong>mation inequalities <strong>for</strong> Gibbs<br />

<strong>measures</strong>. (in preparation)<br />

[8] H. Gjellout, A. Guillin and L. Wu, Transportation cost-in<strong>for</strong>mation inequalities<br />

and applications to random dynamical systems and diffusons, Ann. Proba., Vol<br />

32, No. 3B (2004), 2702-2732.<br />

[9] J. Glimm and A. Jaffe, Quantum Physics, A Functional Integral Point of View,<br />

Springer-Verlag, New York Heidelberg Berlin.<br />

[10] A. Guionnet and B. Zegarlinski, Lectures on logarithmic <strong>Sobolev</strong> inequalities,<br />

in: Séminaire de Probabilité XXXVI, Lecture Notes in Math. 1801, Springer,<br />

Berlin, 2003, 1-134.<br />

[11] L. Gross, <strong>Log</strong>arithmic <strong>Sobolev</strong> inequalities, Amer. J. Math. 97 (1975), 1061-<br />

1083.<br />

10


[12] M. Ledoux, The Concentration of Measures Phenomenon, Math. Surveys Monographs<br />

89, Amer. Math. Soc., Providence, RI, 2001.<br />

[13] K. Marton, Bounding ¯d-distance by in<strong>for</strong>mational divergence: a method to<br />

prove measure concentration, Ann. Prob. 24 (1996) 857-866.<br />

[14] F. Otto and C. Villani, Generalization of an <strong>inequality</strong> by Talagrand and links<br />

<strong>with</strong> the logarithmic <strong>Sobolev</strong> <strong>inequality</strong>, J. Funct. Anal. 173 (2000), no. 2, 361–<br />

400.<br />

[15] Z.L. Zhang and Y. Miao, An equivalent condition between Poincare <strong>inequality</strong><br />

and T 2 -transportation cost <strong>inequality</strong>, Acta Appl. Math.DOI 10.1007/s10440-<br />

008-9379-z.<br />

11

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!