JOHN S DECOMPOSITION OF THE IDENTITY IN THE NON-CONVEX CASE. Jesus Bastero* and Miguel Romance**

Similar documents
Boundaries and the James theorem

An Extremal Property of the Regular Simplex

Minimal surface area position of a convex body is not always an M-position

Chapter 2. Periodic points of toral. automorphisms. 2.1 General introduction

Chapter IV Integration Theory

Lecture Notes for Analysis Class

Sequences and Series of Functions

Convergence of random variables. (telegram style notes) P.J.C. Spreij

b i u x i U a i j u x i u x j

Theorem 3. A subset S of a topological space X is compact if and only if every open cover of S by open sets in X has a finite subcover.

(A sequence also can be thought of as the list of function values attained for a function f :ℵ X, where f (n) = x n for n 1.) x 1 x N +k x N +4 x 3

f n (x) f m (x) < ɛ/3 for all x A. By continuity of f n and f m we can find δ > 0 such that d(x, x 0 ) < δ implies that

Math Solutions to homework 6

Definition 4.2. (a) A sequence {x n } in a Banach space X is a basis for X if. unique scalars a n (x) such that x = n. a n (x) x n. (4.

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014.

Homework Set #3 - Solutions

Seunghee Ye Ma 8: Week 5 Oct 28

Chapter 7 Isoperimetric problem

BETWEEN QUASICONVEX AND CONVEX SET-VALUED MAPPINGS. 1. Introduction. Throughout the paper we denote by X a linear space and by Y a topological linear

Several properties of new ellipsoids

6 Integers Modulo n. integer k can be written as k = qn + r, with q,r, 0 r b. So any integer.

1. By using truth tables prove that, for all statements P and Q, the statement

OFF-DIAGONAL MULTILINEAR INTERPOLATION BETWEEN ADJOINT OPERATORS

Metric Space Properties

A remark on p-summing norms of operators

APPLICATION OF YOUNG S INEQUALITY TO VOLUMES OF CONVEX SETS

Measure and Measurable Functions

Analytic Continuation

A NOTE ON INVARIANT SETS OF ITERATED FUNCTION SYSTEMS

1 Introduction. 1.1 Notation and Terminology

n p (Ω). This means that the

Stochastic Matrices in a Finite Field

Chapter 6 Infinite Series

A REMARK ON A PROBLEM OF KLEE

McGill University Math 354: Honors Analysis 3 Fall 2012 Solutions to selected problems

If a subset E of R contains no open interval, is it of zero measure? For instance, is the set of irrationals in [0, 1] is of measure zero?

Brief Review of Functions of Several Variables

Solutions to home assignments (sketches)

TENSOR PRODUCTS AND PARTIAL TRACES

Supplemental Material: Proofs

Infinite Sequences and Series

2 Banach spaces and Hilbert spaces

Singular Continuous Measures by Michael Pejic 5/14/10

5.1. The Rayleigh s quotient. Definition 49. Let A = A be a self-adjoint matrix. quotient is the function. R(x) = x,ax, for x = 0.

Chapter 3 Inner Product Spaces. Hilbert Spaces

(VII.A) Review of Orthogonality

4 The Sperner property.

THE ASYMPTOTIC COMPLEXITY OF MATRIX REDUCTION OVER FINITE FIELDS

Lecture 10: Bounded Linear Operators and Orthogonality in Hilbert Spaces

Homework 2. Show that if h is a bounded sesquilinear form on the Hilbert spaces X and Y, then h has the representation

On equivalent strictly G-convex renormings of Banach spaces

INEQUALITIES BJORN POONEN

ON MEAN ERGODIC CONVERGENCE IN THE CALKIN ALGEBRAS

The log-behavior of n p(n) and n p(n)/n

2.1. The Algebraic and Order Properties of R Definition. A binary operation on a set F is a function B : F F! F.

A Proof of Birkhoff s Ergodic Theorem

Properties of Fuzzy Length on Fuzzy Set

Square-Congruence Modulo n

Introduction to Functional Analysis

Advanced Analysis. Min Yan Department of Mathematics Hong Kong University of Science and Technology

PRELIM PROBLEM SOLUTIONS

Resolution Proofs of Generalized Pigeonhole Principles

The Wasserstein distances

AN EXTENSION OF SIMONS INEQUALITY AND APPLICATIONS. Robert DEVILLE and Catherine FINET

FUNDAMENTALS OF REAL ANALYSIS by. V.1. Product measures

A Hadamard-type lower bound for symmetric diagonally dominant positive matrices

MATH10212 Linear Algebra B Proof Problems

Ma 4121: Introduction to Lebesgue Integration Solutions to Homework Assignment 5

Lecture 2. The Lovász Local Lemma

Recitation 4: Lagrange Multipliers and Integration

6.3 Testing Series With Positive Terms

Introductory Analysis I Fall 2014 Homework #7 Solutions

Math 220A Fall 2007 Homework #2. Will Garner A

ALGEBRAIC GEOMETRY COURSE NOTES, LECTURE 5: SINGULARITIES.

A constructive analysis of convex-valued demand correspondence for weakly uniformly rotund and monotonic preference

Integrable Functions. { f n } is called a determining sequence for f. If f is integrable with respect to, then f d does exist as a finite real number

On Random Line Segments in the Unit Square

The Multiplication Operator in Sobolev Spaces with Respect to Measures

1 Duality revisited. AM 221: Advanced Optimization Spring 2016

Algebra of Least Squares

Lecture 19: Convergence

The random version of Dvoretzky s theorem in l n

Fall 2013 MTH431/531 Real analysis Section Notes

Real Numbers R ) - LUB(B) may or may not belong to B. (Ex; B= { y: y = 1 x, - Note that A B LUB( A) LUB( B)

1 Approximating Integrals using Taylor Polynomials

ON THE LEHMER CONSTANT OF FINITE CYCLIC GROUPS

sin(n) + 2 cos(2n) n 3/2 3 sin(n) 2cos(2n) n 3/2 a n =

PRACTICE FINAL/STUDY GUIDE SOLUTIONS

Chapter 4. Fourier Series

Mathematical Foundations -1- Sets and Sequences. Sets and Sequences

Fundamental Theorem of Algebra. Yvonne Lai March 2010

Large holes in quasi-random graphs

7.1 Convergence of sequences of random variables

A NOTE ON BOUNDARY BLOW-UP PROBLEM OF u = u p

Riesz-Fischer Sequences and Lower Frame Bounds

Beurling Integers: Part 2

Introduction to Optimization Techniques

Optimally Sparse SVMs

Machine Learning Theory Tübingen University, WS 2016/2017 Lecture 12

Math 451: Euclidean and Non-Euclidean Geometry MWF 3pm, Gasson 204 Homework 3 Solutions

Transcription:

JOHN S DECOMPOSITION OF THE IDENTITY IN THE NON-CONVEX CASE Jesus Bastero* ad Miguel Romace** Abstract We prove a extesio of the classical Joh s Theorem, that characterices the ellipsoid of maximal volume positio iside a covex body by the existece of some kid of decompositio of the idetity, obtaiig some results for maximal volume positio of a compact ad coected set iside a covex set with oempty iterior. By usig those results we give some estimates for the outer volume ratio of bodies ot ecesarily covex. 1. Itroductio ad Notatio Throughout this paper, we cosider R with the caoical basis (e 1,..., e ) ad its usual Euclidea structure,. Let B 2 = {x R ; x = x, x 1/2 1} be the euclidea ball o R. If K R, the it K, K c ad K will deote the iterior, the complemetary ad the border of K, respectively; cov(k) will be the covex hull of K, K 0 will deote the polar of K with respect to the origi, i.e. K 0 = {y R ; x, y 1, x K} ad vol (K) represets the Lebesgue measure o R of K. Followig [TJ], if K 1 K 2 R, we call a pair (x, y) R R a cotact pair for (K 1, K 2 ) if it satisfies: i) x K 1 K 2, ii) y K 0 2 ad iii) x, y = 1. As it is usual y x deotes the liear trasformatio o R defied by y x(z) = z, y x ad I will be the idetity map o R. Joh s ellipsoid theorem is a classical tool i the theory of covex bodies; it says how far a covex body is from beig a ellipsoid. Joh showed that each covex body cotais a uique ellipsoid of maximal volume ad characterized it. The decompositio of the idetity associated to this characterizatio gives a effective method to itroduce a appropiated euclidea structure i fiite dimesioal ormed spaces, whe we cosider cetrally symmetric covex bodies. We ca state Joh s theorem i the followig way: Theorem ([J], [Ba2], [Ba3]). Let K be a covex body i R ad suppose that the euclidea ball B2 is cotaied i K, the the followig assertios are equivalet: (i) B2 is the ellipsoid of maximal volume cotaied i K, (ii) there exist λ 1,..., λ m > 0 ad u 1,..., u m K B2, with m ( + 3)/2 such m m that I = u i u i ad λ i u i = 0. (iii) B 2 is the uique ellipsoid of maximal volume cotaied i K. * Partially supported by Grat D.G.E.S.I.C. (Spai) ** Supported by F.P.I. Grat D.G.E.S.I.C. (Spai) 1

Oe ca cosider this situatio for ay geeral couple of covex bodies (K 1, K 2 ) or, eve more, for ay couple of compact sets i R istead of (B 2, K). Suppose that K 1 is a compact set i R with vol (K 1 ) > 0 ad K 2 is aother compact set i R, with it K 2. A compactess argumet shows that there exists a affie positio of K 1, amely K 1, such that K 1 K 2 ad vol ( K 1 ) = max{vol (a + T (K 1 )); a + T (K 1 ) K 2, a R, T GL()}. This positio K 1 is called maximal volume positio of K 1 iside K 2. Very recetly, Giaopoulos, Perisiaki ad Tsolomitis have cosidered the covex situatio ad proved the followig Theorem [G-P-T]. Let K 1 K 2 be two smooth eough covex bodies i R such that K 1 is i maximal volume positio iside K 2. The for every poit z i the iterior of K 1, there exist λ 1,..., λ N > 0, with N 2 + + 1 ad cotact pairs for (K 1 z, K 2 z), (x i z, y i ), (1 i N) such that: (i) λ i y i = 0 ad (ii) I = λ i y i x i. Furthermore, if we assume the extra assumptio for K 1 to be a polytope ad K 2 to have C (2 boudary with strictly positive curvature, the a ceter z ca be chose i K 1 \ {vertices of K 1 } for which we have (i), (ii) ad also (iii) 1 λ i x i = z. The special case of cosiderig K 1 ad K 2 cetrally symmetric covex bodies was first observed by Milma (see [TJ], Theorem 14.5). The aim of this paper is to exted this result to the o-covex case. We obtai a geeral result which is valid for K 1 a compact, coected set i R with vol (K 1 ) > 0, K 1 K 2, where K 2 is a compact i R such that it cov(k 2 ) = ad K 1 is i maximal volume positio iside cov(k 2 ) (o extra assumptios o the boudary of the bodies are used). The method we develop to prove our result is differet from that i [G-P-T]. We follow the ideas give i [Ba3], with suitable modificatios ad the mai result we achieve is the followig Theorem 1.1. Let K 1 R be a coected, compact set with vol (K 1 ) > 0 ad K 2 R be a compact set such that K 1 K 2. If K 1 is i maximal volume positio iside cov(k 2 ), for every z it cov(k 2 ) there exist N N, N 2 +, (x i, y i ) cotact pairs for (K 1 z, K 2 z) ad λ i > 0 for all i = 1,..., N such that: λ k y k x k = 1 I λ k y k = 0. It is well kow that if there exists a decompositio of the idetity i the sese of theorem 1.1 we ca ot expect that K 1 were the uique maximal volume positio iside 2

K 2 eve for covex bodies, as it ca be show by cosiderig simpleces or octahedra iscribed i the cube. Furthermore, a equivalece as it appears i Joh s Theorem is ot true i geeral. We study this problem ad as a cosequece we obtai Theorem 1.2. Let K 1 R be a coected, compact set with vol (K 1 ) > 0 ad K 2 R be a compact set such that K 1 K 2. Let z be a fixed poit i it cov (K 2 ). The the followig asumptios are equivalets: (i) vol (K 1 ) = max{vol (a + S(K 1 )); a R, a + S(K 1 ) cov(k 2 )}, where S rus over all symmetric positive defiite matrices. (ii) There exist N N, N 2 +3 2, (x i, y i ) cotact pairs for (K 1 z, K 2 z) ad λ i > 0 for all i = 1,..., N such that: λ k (y k x k + x k y k ) = 1 I (1) λ k y k = 0. (iii) K 1 is the uique positio of K 1 verifyig (i). I sectio 3 we exted the upper estimates of the volume ratio proved i [G-P-T] by defiig the outer volume ratio of a compact K 1 with respect to a covex body K 2, by cosiderig a appropriate idex. We follow the methods that appears i [G-P-T] by usig Brascamp-Lieb ad reverse Brascamp-Lieb iequalities as the mai tools. 2. Proofs of Mai Theorems Throughout this sectio K 1 will be a coected, compact set i R with vol (K 1 ) > 0 ad K 2 will be a compact i R such that K 1 K 2. Therorem 1.1 gives us a sufficiet coditio for the existece of some kid of Joh s decompositio of the idetity ad it follows the spirit of the work of K.M. Ball (see for istace [Ba2] or [Ba3]). Proof of Theorem 1.1: First of all, otice that it cov (K 2 ), sice K 1 cov (K 2 ) ad vol (K 1 ) > 0. Without loss of geerality we ca assume that z = 0. Furthermore, sice K 1 K 2 ad cov(k 2 ) 0 = K2 0 a cotact pair for (K 1, cov(k 2 )) is also a cotact pair for (K 1, K 2 ), so, we may suppose cov(k 2 ) = K 2. Let A = {(y x, y) L(R, R ) R ; (x, y) is a cotact pair for (K 1, K 2 )}. By usig the maximality of the volume of K 1, the covexity of K 2 ad sice 0 it K 2 it is easy to prove that A is a o empty subset of L(R, R ) R. We will show that ( 1 I, 0) cov(a), where cov(a) is the covex hull of A i L(R, R ) R. Suppose that ( 1 I, 0) / cov(a). The by usig a separatio theorem, there exist H L(R, R ) ad b R such that: 1 I, H tr + 0, b > y x, H tr + b, y for all (x, y) cotact pair ad where, tr deotes the trace duality o L(R, R ), i.e. T, S tr = tr ST. 3

Thus for every cotact pair (x, y) 1 tr H > Hx, y + b, y. (2) There is o loss of geerality to assume tr H = 0. Ideed, we ca choose H = H I L(R, R ), which is a liear operator with trace zero that verifies: tr H Hx, y + b, y = Hx, y tr H x, y + b, y < 0 = tr H for all (x, y) cotact pair. By usig the liear map defied by the matrix H ad b R we are goig to costruct a family of affie maps S δ s, with 0 < δ < δ 1, such that det S δ 1 ad S δ (K 1 ) it K 2, which cotradicts the maximality of the volume of K 1. We will divide the proof of that fact ito 3 steps. By cotiuity, there exists a positive umber δ 0 > 0 such that I δh is ivertible for all 0 < δ < δ 0. For each 0 < δ < δ 0 we take S δ : R R defied by S δ (z) = (I δh) 1 (z) + δ(i δh) 1 (b). Step 1: There exists 0 < δ 1 δ 0 such that S δ (K 1 ) K 2 =, for all 0 < δ < δ 1. Cosider M = { x K 2 ; y K 0 2 such that x, y = 1 ad Hx, y + b, y 0 }. It is easy to check that M is a compact subset of K 2 ad also M K 1 =. If there exists a x M K 1 K 2 K 1, there would exist y K 0 2 such that x, y = 1 (so (x, y) is a cotact pair) ad Hx, y + b, y 0 = tr H which would cotradict (2). Therefore M K c 1; by compactess of M ad by cotiuity, there exists 0 < δ 1 ( δ 0 ) such that (I δh)(m) δb K c 1, for all 0 < δ < δ 1, ad so S δ (K 1 ) M =. Now let x K 2. We will prove that x / S δ (K 1 ). We oly have to cosider the case x / M. The Hx, y + b, y < 0 for all y K 0 2 such that x, y = 1. Sice 0 it K 2 ad K 2 is a covex body there exists y 0 K 0 2 such that x, y 0 = 1, so we have that x δhx δb, y 0 = 1 δ ( Hx, y 0 + b, y 0 ) > 1 for all δ > 0 ad i particular for all 0 < δ < δ 1. equivaletly x / S δ (K 1 ). Hece x δhx δb / K 1, or Step 2: For every 0 < δ < δ 1 there exists λ δ > 1 such that λ δ S δ (K 1 ) it K 2. Note that S δ (K 1 ) is coected ad S δ (K 1 ) K 2 =, therefore either S δ (K 1 ) it K 2 or S δ (K 1 ) K c 2. Fix x K 1 it K 2, it exists sice vol (K 1 ) = 0, ad take C x = {S δ (x) ; 0 δ < δ 1 }. It is easy to check that C x is coected, C x K 2 =, C x it K 2 ad C x it K 2. Therefore S δ (K 1 ) it K 2 ad by coectedess of K 1 we coclude that 4

S δ (K 1 ) it K 2, for all 0 < δ < δ 1. Now, by a compactess argumet ad the fact that S δ (K 1 ) it K 2 we coclude that for every 0 < δ < δ 1 there exists λ δ > 1 such that λ δ S δ (K 1 ) it K 2. Step 3: vol (λ δ S δ (K 1 )) > vol (K 1 ), for all 0 < δ < δ 1. Ideed, vol (λ δ S δ (K 1 )) = λ δ vol (K 1) det(i δh). Now, by usig the iequality betwee arithmetic mea ad geometric mea (deoted briefly AM-GM iequality) we obtai that det(i δh) 1 1 tr (I δh) = 1, ad so vol (λ δ S δ (K 1 )) λδ vol (K 1 ) > vol (K 1 ). Therefore, if ( 1 I, 0) / cova, the K 1 is ot i maximal volume positio iside K 2. Note that the fact that N 2 + is deduced from the classical Caratheodory s theorem sice {(y x 1 I, y) ; (x, y) is a cotact pair} is cotaied i a ( 2 + 1) dimesioal vector space. Remarks. 1) For every K R with vol (K) > 0 it is easy to check that K is i maximal volume positio iside cov (K). 2) We ote that K 1 K 2 ad K 1 is i maximal volume positio iside cov (K 2 ) implies that K 1 is i maximal volume positio iside K 2. The coverse is ot true, as the followig example shows. Cosider K 1 = {x R ; x = max 1 i x i 1} K 2 = K 1 2 K 1. It is trivial to see that K 1 is i maximal volume positio iside K 2, K 1 is ot i maximal volume positio iside cov(k 2 ) ad there is o decompositio of the idetity as before, sice there are o cotact pairs for (K 1, K 2 ). Corolary 2.1. Let K 1 K 2 R be as i the theorem 1.1. If cov (K 1 ) is a polytope, cov (K 2 ) has C (2) boudary with strictly positive curvature ad K 1 is maximal volume positio iside cov (K 2 ), the there exist z cov (K 1 ), N N, N 2 +, (x k, y k ) cotact pairs of (K 1 z, K 2 z) ad λ k > 0 for all k = 1,..., N such that λ k y k x k = 1 I λ k y k = λ k x k = 0. Proof: It is easy to prove that the fact that K 1 is i maximal volume positio iside cov (K 2 ) implies that cov (K 1 ) is i maximal volume positio iside cov (K 2 ) ad K 1 (cov (K 2 )) = (cov (K 1 )) (cov (K 2 )). Therefore we ca assume that 5

K 1 is a polytope ad K 2 is a covex which has a C (2) boudary with strictly positive curvature. But otice that this case was studied by A. Giaopoulos, I. Perissiaki ad A. Tsolomitis (see [G-P-T]) cocludig the result eeded. Now we ca ask if the existece of some kid of decompositio of the idetity i R would imply that K 1 were the uique maximal volume positio iside K 2, as it happes i the classical Joh s Theorem. It is well kow that we ca t expect such a thig, simply by cosiderig simplices or octahedra iscribed i the cube. Theorem 1.2 shows, loosely speakig, that ot oly the existece of a modified Joh s decompositio of the idetity for a pair (K 1, K 2 ) implies that K 1 is the uique pseudo maximal volume positio iside K 2, but also that this pseudo maximality implies the existece of a modified decompositio of the idetity too. Proof of Theorem 1.2: As before, we ca show that it cov (K 2 ). We ca also assume z = 0 ad K 2 covex. (i) (ii) Let B = {( 1 2 (y x + x y), y) L(R, R ) R ; (x, y) is a cotact pair}. By usig the maximality of the volume of K 1, the covexity of K 2 ad sice 0 it K 2 it is easy to prove that B is a o empty subset of L(R, R ) R. As i the proof of theorem 1.1, we will show that ( 1 I, 0) cov (B). Suppose, o the cotrary, that ( 1 I, 0) / cov (B). The by usig a separatio theorem, there exist H L(R, R ) ad θ R such that: 1 I, H tr + 0, θ > 1 2 ( y x, H tr + x y, H tr ) + θ, y for all (x, y) cotact pair. Therefore 1 tr H > 1 ( Hx, y + x, Hy ) + θ, y. 2 There is o loss of geerality to assume that: 1) H is a symmetric matrix because i other case we could take H = 1 2 (H + H ) which is a symmetric matrix that verifies that Hx, y + x, Hy = Hx, y + x, Hy ad therefore for every cotact pair (x, y) 1 tr H > Hx, y + θ, y. 2) tr H = 0 sice i other case we ca choose H = H tr H I L(R, R ) which is a liear operator with trace zero that verifies: Hx, y + θ, y = Hx, y tr H x, y + θ, y < 0 = tr H for all (x, y) cotact pair. Therefore there exist a symmetric matrix H L(R, R ) with tr H = 0 ad θ R such that: 0 > Hx, y + θ, y 6

for all (x, y) cotact pair. By usig the liear map defied by the matrix H ad θ R we are goig to costruct a family of affie maps T δ ( ) = S δ ( ) + b δ, with S δ symmetric positive defiite matrix for all 0 < δ < δ 1, such that det S δ 1 ad T δ (K 1 ) it K 2, which cotradicts the maximality of K 1. By cotiuity, there exists a positive umber δ 0 > 0 such that I δh is ivertible ad symmetric positive defiite for all 0 < δ < δ 0. For each 0 < δ < δ 0 we take T δ : R R defied by T δ (z) = (I δh) 1 (z) + δ(i δh) 1 (θ). By the same methods as i the proof of Theorem 1.1, we ca show that: 1) There exists 0 < δ 1 δ 0 such that T δ (K 1 ) K 2 =, for all 0 < δ < δ 1. 2) For every 0 < δ < δ 1 there exists λ δ > 1 such that λ δ T δ (K 1 ) it K 2. 3) vol (λ δ T δ (K 1 )) > vol (K 1 ), for all 0 < δ < δ 1 which cotradicts the maximality of K 1. (ii) (iii) Let T ( ) = S( ) + a be such that T (K 1 ) K 2, a R ad S is a symmetric positive defiite matrix. It is well kow that we ca fid a orthogoal matrix U O() ad a diagoal matrix D with diagoal elemets α 1,..., α > 0 such that S = U D U ad therefore ( ) vol (T (K 1 )) = det (U D U) vol (K 1 ) = α k vol (K 1 ). (3) Hece we have to estimate α k. O the oe had, we obtai that U D Ux, y = α j U e j, x U e j, y j=1 for all x, y R, by straightforward computatio. O the other had, if (x, y) is a cotact pair the T x, y 1 ad therefore 1 = = = 1 λ k λ k λ k T x k, y k = j=1 λ k U D Ux k, y k α j U e j, x k U e j, y k = α j U e j, U e j = 1 j=1 α j. j=1 ( j=1 α j N λ k U e j, x k U e j, y k Now by usig the AM-GM iequality, we coclude that 1 1 αj ( α j ) 1, which implies that i (3) we obtai vol (T (K 1 )) vol (K 1 ). I additio to this, ote that if T is such that vol (T (K 1 )) = vol (K 1 ), the, by the equality case i the AM-GM iequality we would have that α 1 =... = α = 1, so T = I + a. Therefore we would obtai that 1 T x, y = x + a, y = 1 + a, y 7 )

for all (x, y) cotact pair ad, i particular, a, y k 0 for all (x k, y k ) cotact pair that appears i the decompositio of the idetity. But we also would have that: λ k a, y k = a, λ k y k = 0 which would imply that, a, y k = 0 for all (x k, y k ) ad the we would coclude that Hece T =I. 1 N a, a = λ k a, y k a, x k = 0. Corolary 2.2. Let K 1 K 2 be as i theorem 1.1. Fix z it cov (K 2 ). The the followig assumptios are equivalets: (i) vol (K 1 ) = max{vol (a + S(K 1 )); a R, a + S(K 1 ) cov(k 2 )}, where S rus over all symmetric positive defiite matrices. (ii) For every S GL() symmetric matrix ad every θ R there exists a cotact pair (x, y) for (K 1 z, K 2 z) such that tr S Sx, y + θ, y. Proof: As before, we ca show that it cov (K 2 ). We ca also assume z = 0 ad K 2 covex. (i) (ii) By Theorem 1.2 there exist (x i, y i ) cotact pairs for (K 1 z, K 2 z) ad λ i > 0 for all i = 1,..., N such that: λ k (y k x k + x k y k ) = 1 I ad λ k y k = 0. Suppose that there would exist S GL() symmetric matrix ad θ R such that for every (x, y) cotact pair Therefore tr S > Sx, y + θ, y. tr S = tr S = < N + θ, λ i y i = (S, θ), ( 1 I, λ k y k ) = λ i (S, θ), ( 1 2 (y i x i + x i y i ), y i ) = λ i tr S = tr S λ i ( Sx i, y i + θ, y k ) < 8

which leads us to a cotradictio. (ii) (i) By usig the hypothesis, for every H GL(), θ R, there exists a cotact pair such that 1 tr H > 1 ( Hx, y + x, Hy ) + θ, y 2 which make that ( 1, 0) cov ({( 1 2 (y x + x y), y) L(R, R ) R ; where (x, y) is a cotact pair }) Remarks. 1) If K 1 is i maximal volume positio iside cov (K 2 ), the K 1 is uique if we oly cosider affie trasformatios give by symmetric, positive defiite matrices. Ideed, this is due to the fact that 1 I = N λ k y k x k implies that 1 I = N λ k x k y k. 2) If we suppose either K 1 =B2 or cov(k 2 )=B2 i the last theorem, we obtai a stroger coclusio, sice the existece of cotact pairs (x k, y k ) ad λ k > 0 such that λ k 2 (y k x k + x k y k ) = 1 I ad λ k y k = 0 is equivalet to the fact that vol (K 1 ) = max{vol (a + T (K 1 )) such that a + T (K 1 ) cov (K 2 ), a R ad T GL()} ad this maximum is oly attaied at K 1, up to orthogoal trasformatio (i.e. if vol (T (K 1 )) = vol (K 1 ), the T is a orthogoal trasformatio). This is the classical Joh s result. Let s see it briefly. Suppose that there exists a decompositio of the idetity (i the sese of (1)). If we take cov(k 2 ) = B 2 ad T is a affie trasformatio such that T (K 1 ) B 2, the there exist orthogoal matrices U, V, a diagoal matrix D with diagoal elemets α 1,..., α > 0 ad a R such that T ( ) = V DU( ) + a. Now if we choose T ( ) = U DU( ) + (V U) (a) the it is easy to check that this map verifies: (a) U DU is a symmetric positive defiite matrix. (b) T (K 1 ) (V U) (B 2 ) = B 2 (sice T ( ) = (V U) T ( )). (c) vol ( T (K 1 )) = vol (T (K 1 )). Therefore by usig (ii) (iii) i theorem 1.2 ad sice T satisfies (a) ad (b) we coclude that vol (T (K 1 )) = vol ( T (K 1 )) vol (K 1 ) ad the equality is oly attaied if T = I, ad so T is a orthogoal trasformatio. Note that a similar reasoig ca be applied to the case K 1 = B 2. 9

3. Some estimates for the outer volume ratio of compact sets We ca exted the otio of volume ratio to a pair (K 1, K 2 ) R R, where K 2 is a covex body ad K 1 is a compact set with vol (K 1 ) > 0, simply by Defiitio 3.1. Let K 1 R be compact set with vol (K 1 ) > 0 ad K 2 R be a covex body. We defie outer volume ratio as { vol (K 2 ) 1 vr(k 2 ; K 1 ) = if ; T affie trasformatio with T (K vol (T (K 1 )) 1 1 ) K 2 }. It is quite easy to show that we caot expect ay upper estimate without asumig extra asumptios. We are goig to itroduce a idex for compact sets with positive volume i order to get geeral bouds, depedig oly o the dimesio ad o the idex, for the outer volume ratio with respect to a covex body. We recall that a set K R is p-covex, (0 < p 1) if λx + µy K, for every x, y K ad for every λ, µ 0 such that λ p + µ p = 1. The p-covex hull of a set K, which we deote by p cov (K), is defied as the itersectio of all p-covex sets that cotai K. It is easy to see that 0 p cov (K). Defiitio 3.2. Let K R a compact set. We defie p(k) as p(k) = { sup {p (0, 1] ; a R with p cov{(extk) a} K a} if it exists 0 otherwise where extk deotes the set of extreme poits of K. Remarks: 1) If p (0, 1] verifies that there exist a a R such that p cov{(extk) a} K a the a K, sice 0 is iside the clausure of p cov{(extk) a}, which is embedded i K a ad so a K. 2) p(k) is a affie ivariat of K, i.e. if T = a + S is a affie trasformatio o R with a R ad S GL() the p(t (K)) = p(k). 3) The supremum i the last defiitio ca be replaced by maximum, simply by usig compactess ad cotiuity argumets. 4)If K is a p-covex body with 0 < p 1 the p(k) p, but if 0 < p < 1 the there are compact sets K with p(k) p which are ot p-covex. Notice that p(k) = 1 if ad oly if K is covex, simply by usig Krei-Milma s theorem. Now we are goig to state ad prove some upper estimates for the volume ratio of a pair (K 1, K 2 ) where K 1 is a compact set with vol (K 1 ) > 0 ad p(k 1 ) > 0, ad K 2 is a covex body. We ca assume that K 1 is i maximal volume positio iside K 2, sice i other case, there would exist a affie trasformatio T such that T (K 1 ) would be i maximal volume positio iside K 2 ad therefore K 1 would work with the pair (T (K 1 ), K 2 ). Hece if p(k 1 ) = p, vr(k 2 ; K 1 ) = vol (K 2) 1 vol (K 1 ) 1 vol (K 2 ) 1 vol (p cov {(extk 1 ) a}) 1 10

for some a K 1. Therefore vr(k 2 ; K 1 ) vol (K 2 a) 1 vol (cov {(extk 1 ) a}) 1 vol (cov {(extk 1 ) a}) 1. vol (p cov {(extk 1 ) a}) 1 It ca be show that cov {(extk 1 ) a} = cov (K 1 a) ad sice cov (K 1 a) is i maximal volume positio iside K 2 a we get vol (cov {(extk 1 ) a}) 1 vr(k 2 ; K 1 ) vr(k 2 ; cov (K 1 )). vol (p cov {(extk 1 ) a}) 1 It is easy to check that cov {(extk 1 ) a} 1 p 1 (p cov {(extk) a}). Ideed, sice a K 1 the cov {(extk 1 ) a} = cov (K 1 a) = cov { x K1 [0, x a]} ad we ca use a stroger versio of Caratheodory s theorem appearig i [E] that asserts that for every x cov {(extk 1 ) a} there exist x i (extk) a ad α i 0, i = 1,..., such that x = α ix i ad α i = 1. Therefore ( α p i ) 1 p 1 p 1 which implies that x 1 p 1 p cov {(extk 1 ) a}. O the other had a result of Giaopoulos, Perissiaki ad Tsolomitis (see [G-P-T]) shows that vr(k 2 ; cov {(extk 1 ) a}) ad thus we sumarize all these thigs i the followig result Propositio 3.3. Let K 1, K 2 R be such that K 1 is a compact set with vol (K 1 ) > 0, p(k 1 ) = p > 0 ad K 2 a covex body. The vr(k 2 ; K 1 ) 1 p. Next we are goig to prove that if K 1 or K 2 has some kid of symmetry properties the this geeral estimate ca be slightly improved by usig decompositios of the idetity i the sese of theorem 1.1, followig the spirit of K.M. Ball (see [Ba1]) ad A. Giaopoulos, I. Perisiaki, A. Tsolomitis ([G-P-T]). We start with a result which ca be foud i [G-P-T] ad whose proof ivolves Cauchy-Biet formula. Lemma 3.4. Let λ 1,..., λ N > 0. Let x 1,..., x N ad y 1,..., y N be vectors i R satisfyig x k, y k = 1, for all k = 1,..., N ad λ k y k x k = I. The D x D y 1, where D x ad D y are defied by { det( } N D x = if λ kα k x k x k ) ; α k > 0, k = 1,..., N N αλ k k 11 α i, (4)

{ det( } N D y = if λ kα k y k y k ) ; α k > 0, k = 1,..., N. (5) N αλ k k Propositio 3.5. Let K 1, K 2 R be such that K 1 is a symmetric compact set with vol (K 1 ) > 0, p(k 1 ) = p > 0 ad K 2 is a symmetric covex body. The vr(k 2 ; K 1 )! 1 1 p 1. Proof: First of all it is easy to check that we ca assume that K 1 ad K 2 are cetrally symmetric ad so it is ext K 1. By usig the same argumets tha before we coclude that vr(k 2 ; K 1 ) vr(k 2 ; cov (K 1 )) 1/p 1. Next we are goig to give a upper estimate for vr(k 2 ; L), where K 2 ad L = cov (K 1 ) are cetrally symmetric covex bodies ad L is i maximal volume positio iside K 2. By usig theorem 1.1, we ca fid cotact pairs (x i, y i ) ad λ i > 0, for all i = 1,..., N, N 2 +, such that λ k y k x k = I ad λ k y k = 0. If we take X = cov {±x 1,..., ±x N } L ad Y = {y R ; y, y k 1 k = 1,..., N} K 2 Y, we obtai that vr(k 2 ; L) = vol (K 2) 1 vol (L) 1 vol (Y ) 1. vol (X) 1 Therefore if we fid some upper estimate for vol (Y ) ad lower estimate for vol (X) we will obtai some upper estimates for vr(k 2 ; K 1 ). Claim 1: vol (Y ) 2 Dy Cosider g j : R R, j = 1,..., N, defied by g j (t) = χ [ 1,1] (t). By usig the Brascamp-Liev iequality (see [Bar]) we obtai that R N (g k ( x, y k )) λ k dx 1 Dy N ( R ) λk g k (t) dt = 1 ( 1 Dy 1 ) λ k dt where D y was defied i (5). O the other had, we coclude that Therefore vol (Y ) R N (g k ( x, y k )) λ k dx = χ Y (x) dx = vol (Y ). R 2 Dy 12

Claim 2: vol (X) 2 D x! We defie for every x R { N N(x) = if α k ; x = } α k x k which is a itegrable fuctio that verifies { N } e N(x) dx = sup e αp k ; αk 0, x = α k x k dx R R { N } = sup f k (θ k ) λ k ; x = λ k θ k x k dx R where f k : R R is defied by f k (t) = e t. Now, if we use the reverse of the Brascamp-Liev iequality (see [Bar]) we ca assert that { N sup f k (θ k ) λ k ; x = R } λ k θ k x k dx D x N = D x N = D x 2 ( 2 λ k R ) λk f k (t) dt where D x was defied i (4). O the other had, we ca compute directly the itegral of e N(x) by e N(x) dx = R R + N(x) e t dt dx = + 0 e t {N(x) t} dxdt. It is easy to check that {x R ; N(x) t} = tx, for all t > 0, ad hece R e N(x) dx = + 0 (6) e t t vol (X) dt =! vol (X). (7) So, combiig (6) ad (7) we coclude the desired lower estimate for vol (X) ad by usig Claim 1, Claim 2 ad lemma 3.4 we obtai that ad hece, the result holds. vr(k 2 ; L)! 1 By usig similar argumets we ca prove the followig result 13

Propositio 3.6. Let K 1, K 2 R are such that K 1 is a compact set with vol (K 1 ) > 0, p(k 1 ) = p > 0 ad K 2 is a covex body, the: (1) If K 1 is symmetric, vr(k 2 ; K 1 ) vr(k 2 ; K 1 ) e 2 (!) 1 1 p 1. (2) If K 2 is symmetric,vr(k 2 ; K 1 ) vr(k 2 ; K 1 ) 2 (!) 1 1 p 1. Proof: (1) Take g j (t) = e t χ (,1] (t) istead of g j (t) i the proof of propositio 3.5. (2) Take f j (t) = e t χ [0,+ ) (t) istead of f j (t) i the proof of propositio 3.5 ad substitute N(x) by Ñ(x) = { N } if α k ; α k 0, x = α k x k if it exists + otherwise. Refereces [Ba1] K.M. Ball, Volume Ratios ad a reverse isoperimetric iequality, J. Lodo Math. Soc. (2) 44 (1991), pp. 351-359. [Ba2] K.M. Ball, Ellipsoids of Maximal Volume i covex bodies, Geometria Dedicata 41 (1992), pp. 241-250. [Ba3] K.M. Ball, A Elemetary Itroductio to Moder Covex Geometry i Flavours of Geometry. Edited by S. Levy. Cambridge Uiversity Press. (1997), pp. 1-55. [Bar] F. Barthe, Iégalités de Brascamp-Liev et covexité, C.R. Acad. Sci. Paris 324 (1997), pp. 885-888. [E] H.G. Egglesto, Covexity, Cambridge Tracts i Math., vol. Uiv. Press, Lodo ad New York (1969). 47, Cambridge [G-P-T] A. Giaopoulos, I. Perissiaki ad A. Tsolomitis, Joh s Theorem for a arbitrary pair of covex bodies, preprit. [J] F. Joh, Extremum problems with iequalities as subsidiary coditios, Courat Aiversary Volume, New York (1948), pp. 187-204 [TJ] N. Tomczak-Jaegerma, Baach-Mazur distaces ad fiite dimesioal operator ideals, Pitma Moographs 38 (1989), Pitma, Lodo. Jesus Bastero: departameto de matemáticas, uiversidad de zaragoza, 50009-zaragoza, spai. e-mail:bastero@posta.uizar.es Miguel Romace: departameto de matemáticas, uiversidad de zaragoza, 50009-zaragoza, spai. e-mail:mromace@posta.uizar.es 14