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.

54 <strong>Convex</strong> Bodies<br />

in one of the two closed halfspaces determined by H. In this case, we denote the<br />

halfspace containing C by H − , the other one by H + . H − is called a support halfspace<br />

of C at x. In general, we represent H in the form H ={z : u · z = u · x}, where<br />

u is a normal (unit) vector of H pointing into H + . Then H − ={z : u · z ≤ u · x}<br />

<strong>and</strong> H + ={z : u · z ≥ u · x}. u is an exterior normal (unit) vector of H, ofC or of<br />

bd C at x. Note that H may not be unique. The intersection C ∩ H is the support set<br />

of C determined by H or, with exterior normal (unit) vector u.<br />

In Theorems 1.2 <strong>and</strong> 2.3, it was shown that a convex function has affine support<br />

at each point in the interior of its domain of definition. The following theorem is the<br />

corresponding result for convex sets.<br />

Theorem 4.1. Let C ⊆ E d be a closed convex set. For each x ∈ bd C, there is a<br />

support hyperplane HC(x) of C at x, not necessarily unique. If C is compact, then<br />

for each vector u ∈ E d \{o}, there is a unique support hyperplane HC(u) of C with<br />

exterior normal vector u.<br />

Let S d−1 denote the Euclidean unit sphere in E d .<br />

Proof. First, the following will be shown.<br />

(1) Let y ∈ E d \C. Then the hyperplane H through pC(y) ∈ bd C, orthogonal<br />

to y − pC(y), supports C at pC(y).<br />

It is sufficient to show that H separates y <strong>and</strong> C. If this does not hold, there is a point<br />

z ∈ C which is not separated from y by H. Then the line segment [pC(y), z] ⊆C<br />

contains a point of C which is closer to y than pC(y). This contradicts the definition<br />

of pC(y) <strong>and</strong> thus concludes the proof of (1).<br />

Next we claim the following.<br />

(2) Let Hn ={z : un · z = xn · un} be support hyperplanes of C at the points<br />

xn ∈ bd C, n = 1, 2,... Assume that un → u ∈ E d \{o} <strong>and</strong> xn → x (∈<br />

bd C) as n →∞. Then H ={z : u · z = u · x} is a support hyperplane of<br />

C at x.<br />

Clearly, x ∈ H. It is sufficient to show that C ⊆ H − .Letz ∈ C. Then un ·z ≤ un · xn<br />

for n = 1, 2,... Letting n →∞, we see that u · z ≤ u · x, or z ∈ H − , concluding<br />

the proof of (2).<br />

For the proof of the first assertion in the theorem, choose points yn ∈ E d \C, n =<br />

1, 2,...,such that yn → x. By Lemma 4.1, xn = pC(yn)(∈ bd C) → x = pC(x).<br />

Proposition (1) shows that, for n = 1, 2,..., there is a support hyperplane of C at<br />

xn,sayHn ={z : un · z = un · xn}, where un ∈ S d−1 . By considering a subsequence<br />

<strong>and</strong> re-numbering, if necessary, we may suppose that un → u ∈ S d−1 ,say.An<br />

application of (2) then implies that H ={z : u · z = u · x} is a support hyperplane of<br />

C at x.<br />

To see the second assertion, note that the compactness of C implies that u · x =<br />

sup{u · z : z ∈ C} for a suitable x ∈ C. Clearly, x ∈ bd C <strong>and</strong> H ={z : u · z = u · x}<br />

is a support hyperplane of C (at x) with exterior normal vector u. ⊓⊔

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

Saved successfully!

Ooh no, something went wrong!