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.

11 Approximation of <strong>Convex</strong> Bodies <strong>and</strong> Its Applications 205<br />

� det(A+ B) � 1 d ≥ � det A � 1 d + � det B � 1 d for A, B ∈ P, where equality holds<br />

precisely in case where A, B are proportional.<br />

A proof of this inequality can be obtained by diagonalizing A <strong>and</strong> B simultaneously<br />

<strong>and</strong> then using the following inequality which is a consequence of the arithmetic–<br />

geometric mean inequality,<br />

� (x1 + y1) ···(xd + yd) � 1 d ≥ (x1 ···xd) 1 d + (y1 ···yd) 1 d for xi, yi ≥ 0,<br />

where equality holds if <strong>and</strong> only if x = (x1,...,xd), y = (y1,...,yd) are<br />

linearly dependent.<br />

See also Sect. 29.4.<br />

Proof of Theorem 11.1. A simple compactness argument yields the existence of at<br />

least one ellipsoid in C of maximum volume.<br />

To see uniqueness, assume that, on the contrary, there are two distinct maximum<br />

volume ellipsoids in C. Without loss of generality we may assume that their<br />

volumes are equal to that of B d . By (1) these ellipsoids can be represented in the<br />

form AB d , BB d with suitable d × d-matrices A, B ∈ P, where A �= B <strong>and</strong><br />

det A = det B = 1. Then 1 2 (A + B)Bd ⊆ C by the convexity of C. Since<br />

det � 1 2 (A + B) � > 1 by (2), the ellipsoid 1 2 (A + B)Bd has greater volume than<br />

the maximum volume ellipsoids AB d , BB d , a contradiction. ⊓⊔<br />

Proof of Theorem 11.2. (i)⇒(ii) By (1), the family of all ellipsoids in C is represented<br />

by the following set of symmetric positive definite matrices in P.<br />

E = � A ∈ P : Au · v = A · u ⊗ v ≤ hC(v) for all u,v ∈ bd B d� .<br />

Clearly, E is the intersection of the family of the closed halfspaces<br />

(3) � A ∈ E 1 2 d(d+1) : A · u ⊗ v ≤ hC(v) � : u,v ∈ bd B d ,<br />

O<br />

K<br />

H<br />

P<br />

D<br />

u2 ⊗ u2<br />

E I<br />

I<br />

u1 ⊗ u1<br />

Fig. 11.1. Proof of John’s theorem

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

Saved successfully!

Ooh no, something went wrong!