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.

Characterization of Support Functions<br />

The next result reveals the simple nature of support functions.<br />

4 Support <strong>and</strong> Separation 57<br />

Theorem 4.3. Let h : E d → R. Then the following statements are equivalent:<br />

(i) h is the support function of a (unique) convex body C, i.e. h = hC.<br />

(ii) h has the following properties:<br />

h(λu) = λh(u) for u ∈ E d ,λ≥ 0<br />

h(u + v) ≤ h(u) + h(v) for u,v ∈ E d<br />

Statement (ii) means that h is positively homogeneous of degree 1 <strong>and</strong> subadditive<br />

<strong>and</strong> thus, in particular, convex.<br />

Proof. (i)⇒(ii) Let h = hC, where C is a suitable convex body. Then<br />

hC(λu) = sup{λu · x : x ∈ C} =λ sup{u · x : x ∈ C} =λhC(u),<br />

hC(u + v) = sup{(u + v) · x : x ∈ C}<br />

≤ sup{u · x : x ∈ C}+sup{v · x : x ∈ C} =hC(u) + hC(v)<br />

for u,v ∈ E d ,λ≥ 0.<br />

(ii)⇒(i) Define<br />

(5) C = � x : v · x ≤ h(v) for all v ∈ E d� = �<br />

{x : v · x ≤ h(v)}.<br />

v∈E d<br />

Being an intersection of closed halfspaces, C is closed <strong>and</strong> convex. Taking v =<br />

±b1,...,±bd, where {b1,...,bd} is the st<strong>and</strong>ard basis of E d , shows that C is<br />

bounded. If C �= ∅, the definition (5), of C, implies that hC ≤ h (note (4)). Thus, to<br />

finish the proof, it is sufficient to show that<br />

(6) C �= ∅<strong>and</strong> h ≤ hC.<br />

Let u ∈ E d \{o}. By (ii), the epigraph epi h, ofh, is a closed convex cone in<br />

E d+1 = E d × R with apex at the origin (o, 0), directed upwards <strong>and</strong> with nonempty<br />

interior. By Theorem 4.1, there is a support hyperplane H of epi h at the point<br />

� u, h(u) � ∈ bd epi h, where � u, h(u) � �= (o, 0). Since epi h is a convex cone with<br />

non-empty interior <strong>and</strong> apex (o, 0), it follows that H supports epi h also at (o, 0).<br />

(Fig. 4.2)<br />

The exterior normal vectors of H at (o, 0) point below E d . Thus we may choose<br />

such a vector of the form (x, −1). Hence H = {(v, s) : v · x − s = 0}. Then<br />

H − ={(v, s) : v · x ≤ s} ⊇epi h <strong>and</strong> thus, in particular, v · x ≤ h(v) for each point<br />

(v, h(v)) ∈ bd epi h <strong>and</strong> therefore v · x ≤ h(v) for all v ∈ E d . Hence x ∈ C, bythe<br />

definition of C in (5), <strong>and</strong> thus C �= ∅. Since H is a support hyperplane of epi h at<br />

(u, h(u)), we have u · x = h(u). For x ∈ C, the definition of the support function<br />

hC implies that hC(u) ≥ u · x. Thus hC(u) ≥ h(u). The proof of (6) is complete,<br />

concluding the proof of the implication (ii)⇒(i). ⊓⊔<br />

Warning. We warn the reader: the above proof cannot be trivialized, since, a priori,<br />

it is not clear that the boundary hyperplanes of the halfspaces appearing in (5) all<br />

touch C.

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

Saved successfully!

Ooh no, something went wrong!