THE STEINER FORMULA FOR EROSIONS. then one shows in integral geometry that the volume of A ρk (ρ 0) is a polynomial of degree n:

Similar documents
arxiv: v2 [math.ds] 9 Jun 2013

Maths 212: Homework Solutions

CHAPTER 7. Connectedness

Notes for Functional Analysis

Locally convex spaces, the hyperplane separation theorem, and the Krein-Milman theorem

GAUSSIAN MEASURE OF SECTIONS OF DILATES AND TRANSLATIONS OF CONVEX BODIES. 2π) n

Math 421, Homework #6 Solutions. (1) Let E R n Show that = (E c ) o, i.e. the complement of the closure is the interior of the complement.

A. Iosevich and I. Laba January 9, Introduction

REAL ANALYSIS LECTURE NOTES: 1.4 OUTER MEASURE

NAME: Mathematics 205A, Fall 2008, Final Examination. Answer Key

REAL ANALYSIS I HOMEWORK 4

Appendix B Convex analysis

Course 212: Academic Year Section 1: Metric Spaces

Topology. Xiaolong Han. Department of Mathematics, California State University, Northridge, CA 91330, USA address:

Convexity in R N Supplemental Notes 1

Notes on Distributions

Star bodies with completely symmetric sections

3 (Due ). Let A X consist of points (x, y) such that either x or y is a rational number. Is A measurable? What is its Lebesgue measure?

U e = E (U\E) e E e + U\E e. (1.6)

Geometry and topology of continuous best and near best approximations

Problem Set 2: Solutions Math 201A: Fall 2016

A LOCALIZATION PROPERTY AT THE BOUNDARY FOR MONGE-AMPERE EQUATION

The small ball property in Banach spaces (quantitative results)

Geometric and isoperimetric properties of sets of positive reach in E d

FUNCTIONAL ANALYSIS LECTURE NOTES: COMPACT SETS AND FINITE-DIMENSIONAL SPACES. 1. Compact Sets

Austin Mohr Math 730 Homework. f(x) = y for some x λ Λ

1 Topology Definition of a topology Basis (Base) of a topology The subspace topology & the product topology on X Y 3

g 2 (x) (1/3)M 1 = (1/3)(2/3)M.

Mathematics for Economists

Proofs for Large Sample Properties of Generalized Method of Moments Estimators

A Brunn Minkowski theory for coconvex sets of finite volume

MATH 54 - TOPOLOGY SUMMER 2015 FINAL EXAMINATION. Problem 1

The Geometric Approach for Computing the Joint Spectral Radius

LECTURE 15: COMPLETENESS AND CONVEXITY

Measuring Ellipsoids 1

POLARS AND DUAL CONES

EQUIVALENCE OF TOPOLOGIES AND BOREL FIELDS FOR COUNTABLY-HILBERT SPACES

MATH 722, COMPLEX ANALYSIS, SPRING 2009 PART 5

Chapter 1. Measure Spaces. 1.1 Algebras and σ algebras of sets Notation and preliminaries

Problem set 1, Real Analysis I, Spring, 2015.

Topology, Math 581, Fall 2017 last updated: November 24, Topology 1, Math 581, Fall 2017: Notes and homework Krzysztof Chris Ciesielski

Real Analysis Notes. Thomas Goller

1 Lecture 4: Set topology on metric spaces, 8/17/2012

Contents: 1. Minimization. 2. The theorem of Lions-Stampacchia for variational inequalities. 3. Γ -Convergence. 4. Duality mapping.

THE CONLEY ATTRACTORS OF AN ITERATED FUNCTION SYSTEM

Convexity in R n. The following lemma will be needed in a while. Lemma 1 Let x E, u R n. If τ I(x, u), τ 0, define. f(x + τu) f(x). τ.

Analysis III Theorems, Propositions & Lemmas... Oh My!

Convergence of a Generalized Midpoint Iteration

Lecture Notes in Advanced Calculus 1 (80315) Raz Kupferman Institute of Mathematics The Hebrew University

Reflections of planar convex bodies

Set, functions and Euclidean space. Seungjin Han

SHARP BOUNDARY TRACE INEQUALITIES. 1. Introduction

Recall that if X is a compact metric space, C(X), the space of continuous (real-valued) functions on X, is a Banach space with the norm

Banach Spaces V: A Closer Look at the w- and the w -Topologies

Convex Geometry. Carsten Schütt

Def. A topological space X is disconnected if it admits a non-trivial splitting: (We ll abbreviate disjoint union of two subsets A and B meaning A B =

1 Directional Derivatives and Differentiability

Spheres with maximum inner diameter

Deviation Measures and Normals of Convex Bodies

Your first day at work MATH 806 (Fall 2015)

10 Typical compact sets

SIMPLICIAL COMPLEXES WITH LATTICE STRUCTURES GEORGE M. BERGMAN

Real Analysis Math 131AH Rudin, Chapter #1. Dominique Abdi

General Convexity, General Concavity, Fixed Points, Geometry, and Min-Max Points

Economics 204 Fall 2012 Problem Set 3 Suggested Solutions

Convex Sets. Prof. Dan A. Simovici UMB

Part V. 17 Introduction: What are measures and why measurable sets. Lebesgue Integration Theory

NAKAJIMA S PROBLEM: CONVEX BODIES OF CONSTANT WIDTH AND CONSTANT BRIGHTNESS

(c) For each α R \ {0}, the mapping x αx is a homeomorphism of X.

Richard F. Bass Krzysztof Burdzy University of Washington

Introduction to Dynamical Systems

Some Background Material

3 COUNTABILITY AND CONNECTEDNESS AXIOMS

Math 341: Convex Geometry. Xi Chen

Definition 2.1. A metric (or distance function) defined on a non-empty set X is a function d: X X R that satisfies: For all x, y, and z in X :

PICARD S THEOREM STEFAN FRIEDL

PROBLEMS. (b) (Polarization Identity) Show that in any inner product space

SCALE INVARIANT FOURIER RESTRICTION TO A HYPERBOLIC SURFACE

Discrete Geometry. Problem 1. Austin Mohr. April 26, 2012

Centre for Mathematics and Its Applications The Australian National University Canberra, ACT 0200 Australia. 1. Introduction

In English, this means that if we travel on a straight line between any two points in C, then we never leave C.

Math General Topology Fall 2012 Homework 8 Solutions

Best approximations in normed vector spaces

Functional Analysis I

Some Properties of Convex Hulls of Integer Points Contained in General Convex Sets

ECARES Université Libre de Bruxelles MATH CAMP Basic Topology

(x 1, y 1 ) = (x 2, y 2 ) if and only if x 1 = x 2 and y 1 = y 2.

Topics in Stochastic Geometry. Lecture 4 The Boolean model

Introduction and Preliminaries

arxiv: v1 [math.fa] 14 Jul 2018

Chapter 2 Convex Analysis

Quasi-conformal maps and Beltrami equation

The Minimal Element Theorem

From now on, we will represent a metric space with (X, d). Here are some examples: i=1 (x i y i ) p ) 1 p, p 1.

Analysis III. Exam 1

LECTURE 10: THE ATIYAH-GUILLEMIN-STERNBERG CONVEXITY THEOREM

On John type ellipsoids

MATH 426, TOPOLOGY. p 1.

Introduction to Geometric Measure Theory

Behaviour of Lipschitz functions on negligible sets. Non-differentiability in R. Outline

Transcription:

THE STEINER FORMULA FOR EROSIONS. G. MATHERON Abstract. If A and K are compact convex sets in R n, a Steiner-type formula is valid for the erosion of A by K if and only if A is open with respect to K. The if part is proved in the general case. The only if part is conjectured, and proved only in the case n = 2. Keywords: erosion, Steiner formula. If A and K are compact and convex in R n, and if denotes Minowsi sum B C = {b + c, b B, c C}, then one shows in integral geometry that the volume of A ρk (ρ 0) is a polynomial of degree n: V (A ρk) = n W (A, K)ρ. (1) The coefficients of this polynomial are the values in (A, K) of the mixed volume functionals of Minowsi. In particular, W 0 (A, K) = V (A) and W n (A, K) = V (K). More generally, one also has n p p W p (A ρk, K) = W +p (A, K)ρ, p = 0, 1,..., n. (2) These mixed functionals (in the two variables A and K) are read as usual functionals of a single variable (Quermassintegrale) by the intermediary of passage to means of rotation: if Ω denotes the group of rotations of R n and ω(dω) is the unique invariant probability measure on Ω, one has, in effect, the classical relation W (ωa, K) ω(dω) = 1 W (A)W n (K) (3) b n Ω where b n denotes the volume of the unit ball in R n. In particular, the mean volume of the rotation of A ρk is given by V (ωa ρk) ω(dω) = 1 n W (A)W n (K)ρ. (4) b n Ω In contrast, one does not generally have such simple results for Minowsi subtraction. Rather than Minowsi subtraction, it is interesting to consider the erosion of A by ρk, in Translation by Erin P. J. Pearse, May 20, 2005. Full reference for the original article appears at the end. 1

other words, the difference A ρǩ where Ǩ = K is the symmetric of K in relation to the origin. (In a general manner, B Č = {x C. x B}, where C x denotes the translate of C by x.) My objective, in what follows, is to examine conditions under which the formulas (1), (2), and (4) remain valid for erosions: A must be open by K, i.e., A = A K, with A K = (A Ǩ) K. This first result can be improved. Suppose that all the functionals W p, p = 1, 2,..., n, satisfy the relation in question. I mae a conjecture stronger than the converse: it suffices only to suppose that the volume of the erosion satisfies the formula for which A is open by K. Similarly, if one passes to means of rotation, one deduces from the first result a general demonstration of a conjecture of [Miles] (demonstrated by this author in the case n = 2) according to which the Steiner formula is applied to the mean of rotation of the erosion for which A and all its rotational transforms are open by K. In my turn, I put forth a new conjecture for which condition is not just sufficient, but also necessary. I demonstrate these two new conjectures in paragraph 3, but only for 2 dimensions. In the last paragraph, I give (always in R 2 ) the necessary and sufficient condition for the volume of the erosion to be a polynomial of second degree. 1. The formula (2) for erosions. We first give the principal result. Theorem 1. Let A and K be compact and convex in R n. One has: n p W p (A ρǩ, K) = p ( 1) ρ W +p (A, K) (2 ) for p = 0, 1,..., n and all ρ (0, 1) if and only if A is open by K. Proof. ( ) Suppose that A is open by K, so A = A 1 K with A 1 = A Ǩ. So for 0 r 1, one also has A rǩ = A 1 (1 r)k and, according to (2): n p W p (A rǩ, K) = p (1 r) W +p (A 1, K) = Q p (r), is a polynomial in r. By the same formula, for r 0 one also finds n p p W p (A rk, K) = r W +p (A, K) = P p (r). But one has A rk = A 1 (1 + r)k and thus again: n p p W p (A rk, K) = (1 + r) W +p (A 1, K) = Q p ( r). Consequently, Q p ( r) = P p (r) for r 0: as it is a question of polynomials, this also gives rise to Q p (r) = P p ( r) for r (0, 1), i.e., W p (A rǩ, K) = P p( r). 2

( ) Conversely, suppose: W p (A Ǩ, K) = P p( 1) for p = 0, 1,..., n. Applying the formula (1) to the opening A K = (A Ǩ) K gives n n V (A K ) = W (A Ǩ, K) = P ( 1). An elementary calculation thus gives ( n ) P ( 1) = V (A). Lie A K A, the equality V (A K ) = V (A) brings forth the equality A K = A: A is open by K. Corollary (1). If A is open by K, for all ρ (0, 1) one has: n V (A ρǩ) = ( 1) W (A, K)ρ. (1 ) This corollary is trivial, but its converse is not. Below, we state a form of this conjecture for R n, and we prove only the case n = 2. Conjecture 1. (Strong converse). If (1 ) is true, then A is open by K. We will see just below that the volume of the erosion is differentiable in ρ with derivative: d dρ V (A ρǩ) = nw 1(A ρǩ, K), ρ (0, 1) (5) (provided that A Ǩ ). It follows that (2 ) hence (1 ) is verified for p = 1. For two dimensions, this suffices (since W 2 (A ρǩ, K) = V (K) trivially verifies (2 )), and Conjecture 2 is a simple corollary of the theorem. But it is no longer the same when n 3. In this regard to the means of rotation we obtain the result of [Miles] named in the conjecture (and proved by him in the case n = 2). Corollary (2). If A and all its rotations are open by K, then for all ρ (0, 1), one has: V (ωa ρǩ) ω(dω) = 1 n ( 1) W (A)W n (K)ρ. (4 ) b n Ω In effect, it suffices to apply (1 ) to ωa and to use the relation (3). Remar. It is easy to see that our condition (ωa open by K for all rotations ω Ω) is equivalent to that given by [Miles] (the sup of the radii of the curvature of K must be bounded by the inf of the radii of the curvature of A) or rather: there exists a ball Br of the ray r 0 such that A is open by Br and Br is open by K. We prove further the converse of the corollary in the case n = 2. For n > 2, we give a conjecture. 3

Conjecture 2. If (4 ) is true, then A and all its rotations are open by K. 2. Differentiability and convexity of V (A ρǩ) In the following, A and K always designate compact convex sets for which the erosion A Ǩ is not empty. As the many functionals studied here are invariant under translation, we may assume 0 K A, which lets us use the support functions h K and h A of these convex bodies. Our first objective is to establish the differentiability of V (A ρǩ), and the relation (5) used to prove the conjecture in the case n = 2. We first note an inequality. From A A ρk = (A ρǩ) ρk and the relation (1), we have V (A) V (A ρǩ) + nρw 1(A ρǩ, K) +... (taing into account the continuity of W 1 ) and by the following, for ρ tending to 0: V (A) V (A ρǩ) lim inf nw 1 (A, K). (6) ρ We use the following lemma (which we will prove in a moment). Lemma 1 (Convexity Lemma). For all ρ (0, 1), one has V (A ρk) V (A) V (A) V (A ρǩ). (7) According to this lemma, one obtains for ρ tending to 0: lim sup V (A) V (A ρǩ) ρ lim V (A ρk) V (A) nw 1 (A, K). ρ Taing (6) into account, this indicates that V (A ρǩ) admits a right derivative at ρ = 0 equal to nw 1 (A, K). Upon replacing A by A ρǩ, on finds immediately that for 0 ρ < 1 d + V (A ρǩ) dρ = nw 1 (A ρǩ, K). For 0 < ρ 1, analogous reasons show that the left derivative exists and again taes the value nw 1 (A ρǩ, K). This establishes differentiability and the relation (5). Moreover, this derivative is increasing (because W 1 (C, K) is increasing function of the convex body C) so that the function ρ V (A ρǩ) is convex on (0, ). If we put { V (A ρǩ), ρ 0, Φ(ρ) = V (A ρǩ), ρ < 0, this function Φ admits the same left and right derivative nw 1 (A ρǩ, K) at ρ = 0. As it is convex on the positive half-line, and also, after (1), on the negative half-line, this function is convex on the entire interval (, ). 4

Proof of the convexity lemma. We initially consider the case where A is a polyhedron (with nonempty interior). Let F i, i = 1,..., N be the faces of A, and S i the (n 1)-volume of F i, and let u i be the exterior unit normal vector of F i. If h K is the support function of K, one has: N V (A ρk) V (A) + ρ S i h K (u i ) and nw 1 (A, K) = i=1 N S i h K (u i ). We show that V (A ρǩ) has right derivative nw 1(A, K) at ρ = 0. Let x A be a point of A. One has x A Ǩ iff x F i ρǩ for one or more faces F i. But it is easy to see that the volume of A (F i ρǩ) has an upper bound of the form i=1 V ((F i ρǩ) A) ρ(s ih K (u i ) + ε i ) with ε i 0 if ρ 0. (h K is the support function of K). Therefore: N V (A) V (A ρǩ) V [ N (F i ρǩ) A] ρ V (S i h K (u i ) + ε i ) and lim sup i=1 V (A) V (A ρǩ) ρ i=1 N S i h K (u i ) = nw 1 (A, K). Taing into account (6), this implies the existence of a right derivative at ρ = 0. Just as this derivative is nw 1 (A, K), the function { V (A ρǩ), ρ 0, Φ(ρ) = V (A ρk), ρ < 0, similarly admits a derivative i=1 Φ (0) = nw 1 (A, K) at ρ = 0. For ρ > 0, A ρǩ is again a polyhedron (as an intersection of translates of A), and one can apply the same reasoning replacing A by A ρǩ. Whence one sees that Φ has, for all ρ (, 1), the derivative { Φ nw 1 (A ρǩ, K), ρ 0, (ρ) = nw 1 (A ρk, K), ρ < 0. This derivative is increasing so that Φ is convex. In particular, A verifies the relation (7). Now if A is not a polyhedron, consider a decreasing sequence of convex polyhedrons {A n } such that A = A n : one has A ρk = (A n ρk), A ρǩ = (A n ρǩ), 5

so that the relation (7), which holds for each A n, passes to the limit. The surface measure. When A is a convex polyhedron, one has for all compact convex K: nw 1 (A, K) = h K (u)g A (du), (8) where G A is the measure on the unit sphere defined by G A = N i=1 S iδ ui. Because the support functions h K of the compact convex sets K form a dense subspace of the continuous functions on the unit sphere (under uniform convergence), this relation (8) passes to the limit: for all compact convex sets A, there is a measure G A on the unit sphere such that (8) is valid and the function A G A is continuous. (If A n A for the Hausdorff metric, then G An converges vaguely to G A.) Moreover, if A is a polyhedron, the (n 1)-volumes S i (ρ) of the faces of the erosion A ρǩ are bounded above by the (n 1)-volumes S i of the corresponding faces of A. Then one also has G A ρ Ǩ G A. This relation similarly passes to the limit: for all compact convex A, the function ρ G A ρ Ǩ is decreasing. In particular, we see the following result. Lemma 2. In 2 dimensions, A is open by A Ǩ whenever A Ǩ. Proof. In effect, in R 2, a measure G on the unit circle is a perimeter measure of a compact convex (defined up to a translation) iff e iθ G(dθ) = 0 (condition of the closure of the contour). If A Ǩ, one has G A Ǩ G A by the above. Thus, the measure G = G A G A Ǩ is positive, and as it satisfies the conditions of closure, G is the perimeter measure of a compact convex set K 1. But G A = G A Ǩ + G K1 is equivalent to A + (A Ǩ) K 1. Thus A is open by A Ǩ. Remar. Lemma 2 is not true for n 3. Counterexample: in R 3, let A be a cone with a circular base, and let K be a segment b parallel to the base of A. Then A Ǩ is the intersection A A b, and one easily sees that A is not open by A A b. In R n, we have A 1 = A Ǩ, K 1 = A Ǎ1 = A (Ǎ K). It is easy to see that K 1 is the closure of K by the complement A c of A (i.e., the intersection of the translates of A which contain K): K 1 = K Ac = (K Ǎc ) A c = {A b. A b K}. So A Ǩ1 = A 1 (because A Ǩ1 is the closure of A 1 by A c, but A 1 is already closed by A c insofar as the intersection of the translates of A) and one has A K1 = A A1 = A 1 K 1 A. But the inclusion is strict in general, and it s only in the case n = 2 that the equality A = A 1 K 1 is guaranteed. 6

In R n, if A is open by K, let A = A 1 K and one necessarily has K 1 = K. In effect, A 1 K = A A 1 K 1 A 1 K follows from the equalities A 1 K 1 = A 1 K and K 1 = K. For n = 2, the converse is true: in R 2, A is open by K iff K 1 = K. In effect, according to lemma 2, in R 2 A is open by A 1 = A Ǩ. One has therefore A = A 1 C for the compact convex C which satisfies C = A Ǎ1 = K 1. Thus A = A 1 K 1. Consequently, if K 1 = K, one has A = A 1 K and A is open by K. 3. Demonstration of the Conjectures 1 and 2 for n = 2 In the case n = 2, one has an important inequality. Theorem 2. If A and K are compact and convex in R 2 and A Ǩ, one has: V (A ρǩ) V (A) 2ρW 1(A, K) + ρ 2 V (K) (9) for all ρ (0, 1), with equality iff A is open by ρk. Proof. In effect, by (5) one immediately has d dρ V (A ρǩ) = 2W 1(A ρǩ, K) and A A ρk = (A ρǩ) ρk follows immediately from (2): W 1 (A, K) W 1 (A ρǩ, K) + ρv (K). Consequently: d dρ V (A ρǩ) 2ρV (K) 2W 1(A, K) and by integrating from 0 to ρ 0 1: V (A ρ 0 Ǩ) V (A) 2ρ 0 W 1 (A, K) + ρ 2 0 V (K). If the equality is attained, one has d dρ V (A ρǩ) = 2ρV (K) 2W 1(A, K) for 0 ρ ρ 0, i.e., W 1 (A ρǩ) = W 1(A, K) ρv(k). Upon transferring these results into V (A ρ0 K) = V (A ρ 0 Ǩ) + 2ρ 0 W 1 (A ρ 0 Ǩ, K) + ρ 2 0 V (K) one thus finds V (A ρ0 K) = V (A), and consequently A ρ0 K = A is open by ρ 0 K. Corollary. The conjectures 1 and 2 are true in R 2. 7

Proof. For conjecture 1, this is immediate. Suppose we have V (ωa ρǩ) ω(dω) = V (A) (ρ/2π)s(a)s(k) + ρ2 V (K) Ω where S(A) = 2W 1 (A) and S(K) = 2W 1 (K) are the perimeters of A and K. inequality (9), this indicates that one has: V (ωa ρǩ) = V (A) 2ρW 1(A, K) + ρ 2 V (K), By the thus ωa is open by K, for ω-almost every rotation ω Ω. As the collection of compact convex sets open by K is closed (under the Hausdorff metric), this relation occurs, as a matter of fact, for all ω Ω, and conjecture 2 follows. At this stage, we put forth a third conjecture (from which the first and second conjectures would be immediate consequences). Conjecture 3. In R n, one has for all ρ (0, 1): n V (A ρǩ) ( 1) W (A, K)ρ with equality iff A is open by ρk. 4. Conditions for which V (A ρǩ) will be a polynomial When A isn t open by K, the function ρ V (A ρǩ) may be very complicated (for example, if one erodes a ball by a cube), but may, in certain cases, be a polynomial of degree n (by example, if one erodes a cube by a ball). In the case of 2 dimensions, one can obtain a precise result in this regard (which does not generalize to higher dimensions). In effect, for n = 2 we can put By Lemma 2, one has One can also write A ρ = A ρǩ and K ρ = A Ǎρ, ρ (0, 1). A = A ρ K ρ, and ρk K ρ ρk 1. A 1 = A Ǩ = (A ρǩ) (1 ρ)ǩ = A ρ (1 ρ)ǩ. By the same Lemma 2, it follows that A ρ is open by A 1, so let: with X ρ = A ρ Ǎ1. Whence A ρ = A 1 X ρ (10) A = A ρ K ρ = A 1 K ρ X ρ. As one also has A = A 1 K 1, it further follows that K 1 = K ρ X ρ. (10 ) 8

Therefore, K 1 is open by K ρ = A Ǎρ for all ρ (0, 1). As one also has ρk ρk 1, it follows that A ρǩ = A ρ A ρǩ1, thus V (A ρǩ) V (A ρǩ1) = V (A) 2ρW 1 (A, K 1 ) + ρ 2 V (K 1 ) (11) with equality when ρ = 0 and ρ = 1 (because A Ǩ = A Ǩ1). For ρ other than 0 or 1, the inequality (11) becomes an inequality iff A ρǩ = A ρǩ1, that is to say, if: A 1 X ρ = (A 1 K 1 ) ρǩ1 = A 1 (1 ρ)k 1, so finally also if X ρ = (1 ρ)k 1. But X ρ = (1 ρ)k 1,, by (10 ) is equivalent to K ρ = ρk 1. Whence a first result. Lemma 3. In R 2, the inequality (11) is true, and the equality is true for ρ (0, 1) iff K ρ = ρk 1. More generally, we have the following theorem. Theorem 3. In R 2, V (A ρǩ) is a polynomial of second degree in ρ iff K ρ = ρk 1 for every ρ (0, 1), and in this case one has { V (A ρǩ) = V (A) 2ρW (A, K 1 ) + ρ 2 V (K 1 ) W 1 (A, K 1 ) = W 1 (A, K). Proof. If K ρ = ρk 1, these relation follow from Lemma 3 and formula (5). Conversely, we suppose that V (A ρ ) is a second degree polynomial, V (A ρ ) = V (A) 2aρ + Hρ 2. Upon application of (5), we find a = W 1 (A, K) and W 1 (A ρ, K) = W 1 (A, K) ρh. But A = A ρ K ρ and W 1 (A, K) = W 1 (A ρ, K) + W 1 (K ρ, K). Consequently, In particular, for ρ = 1 it becomes So one can write for ρ = 1 ρh = W 1 (K ρ, K), ρ (0, 1). H = W 1 (K 1, K). V (A 1 ) = V (A) 2W 1 (A, K) + W 1 (K 1, K) = V (A) 2W 1 (A 1, K) W 1 (K 1, K) (seeing that A = A 1 K 1 ), since (by K K 1 ) we have V (A 1 ) V (A) 2W 1 (A 1, K 1 ) V (K 1 ) = V (A) 2W 1 (A, K 1 ) + V (K 1 ) = V (A 1 ). 9

It follows that the preceding inequalities are, in fact, equalities: W 1 (K 1, K) = V (K 1 ) W 1 (A 1, K) = W 1 (A 1, K 1 ) W 1 (A, K) = W 1 (A, K 1 ). Thus one has a = W 1 (A, K 1 ) and H = V (K 1 ). By Lemma 3, this follows from A ρ = A ρǩ1, and K ρ = ρk 1 for all ρ (0, 1). Remar. Theorem 3 is not true for n 3. For example, tae the unit ball in R 3 for A, and let K be a straight segment of unit length. Then for ρ (0, 1), one finds: V (A ρǩ) = π (1 32 6 ρ + 12 ) ρ3. This is a polynomial of third degree. But the relation K ρ = ρk 1 is not satisfied. References [Math] Matheron, G., La Formule de Steiner pour les érosions, J. Appl. Prob. 15, pp. 126 135 (1978) [Miles] Miles, R. E., On the elimination of edge efects in planar sampling., in Stochastic Geometry. E. F. Harding and D. G. Kendall, eds., Wiley, London (1974) 10