14.02.2013 Views

Gruber P. Convex and Discrete Geometry

Gruber P. Convex and Discrete Geometry

Gruber P. Convex and Discrete Geometry

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.

164 <strong>Convex</strong> Bodies<br />

An application of the inequality of the arithmetic <strong>and</strong> the geometric mean to the right<br />

h<strong>and</strong>-side expression shows that the latter is bounded below by V (S) 1−λ<br />

d V (T ) λ d .<br />

Now raise both sides to the dth power to get<br />

V � (1 − λ)S + λT � ≥ V (S) 1−λ V (T ) λ .<br />

(ii)⇒(i) If V (S) = 0orV (T ) = 0, the ordinary Brunn–Minkowski inequality<br />

clearly holds. Assume now that V (S), V (T )>0. Define:<br />

(4) U =<br />

1<br />

V (S) 1 1<br />

S, V =<br />

d V (T ) 1 λV (T )<br />

T, µ=<br />

d<br />

1 d<br />

(1 − λ)V (S) 1 d + λV (T ) 1 .<br />

d<br />

Then<br />

(1 − λ)S + λT<br />

(5) (1 − µ)U + µV =<br />

(1 − λ)V (S) 1 d + λV (T ) 1 .<br />

d<br />

The multiplicative version of the Brunn–Minkowski inequality applied to U, V,µ<br />

shows that<br />

V � (1 − µ)U + µV � ≥ V (U) 1−µ V (V ) µ ≥ 1<br />

by (4). Now use (5) to obtain<br />

V � (1 − λ)S + λT � 1 d ≥ (1 − λ)V (S) 1 d + λV (T ) 1 d . ⊓⊔<br />

8.6 General Isoperimetric Inequalities <strong>and</strong> Concentration of Measure<br />

There are natural extensions of the isoperimetric inequality to the sphere S d <strong>and</strong> the<br />

hyperbolic space H d . In many cases, these extensions are based on symmetrization<br />

arguments or on Brunn–Minkowski type inequalities, see Lévy [653] <strong>and</strong>, in particular,<br />

Schmidt [891] <strong>and</strong> Dinghas [269]. More recently, these isoperimetric inequalities<br />

were extended even further in the context of metric probability spaces, a study initiated<br />

by Milman. Some of these results are rather surprising <strong>and</strong> are well described<br />

by “concentration of measure”.<br />

This section gives a description of the general isoperimetric problem in metric<br />

probability spaces. Then the cases of the sphere <strong>and</strong> of the Gaussian measure on E d<br />

are discussed. Our exposition follows Ball [53].<br />

For more material on general isoperimetric inequalities <strong>and</strong> on concentration<br />

of measure we refer to Burago <strong>and</strong> Zalgaller [178], Ball [53], Ledoux [634],<br />

Schechtman [885] <strong>and</strong> the references cited below.<br />

The Euclidean Case<br />

Let C be a compact set in Ed . The Brunn–Minkowski theorem for compact sets then<br />

gives the following: Let ϱ ≥ 0 such that V (ϱB d ) = V (C). Then:<br />

(1) V (C + εB d ) ≥ � V (C) 1 d + V (εB d ) 1 �d � 1<br />

d<br />

d<br />

= V (ϱB ) d + V (εB d ) 1 �d d<br />

= V (ϱB d + εB d ) for ε ≥ 0.

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

Saved successfully!

Ooh no, something went wrong!