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.

24 <strong>Convex</strong> Functions<br />

First-Order Differentiability <strong>and</strong> Partial Derivatives<br />

A function f : C → R is differentiable at a point x ∈ C in the sense of Stolz or<br />

Fréchet if there is a (necessarily unique) vector u ∈ E d , such that<br />

f (y) = f (x) + u · (y − x) + o(�y − x�) as y → x, y ∈ C.<br />

A weaker notion of differentiability is differentiability in the sense of Gâteaux. Since<br />

both notions of differentiability coincide for convex functions in finite dimensions,<br />

only the former will be considered.<br />

Theorem 2.5. Let f : C → R be convex <strong>and</strong> x ∈ int C. Then the following statements<br />

are equivalent:<br />

(i) f is differentiable at x.<br />

(ii) The partial derivatives fxi (x), i = 1,...,d, exist.<br />

Proof. (i)⇒(ii) This is well known from calculus <strong>and</strong> easy to prove.<br />

(ii)⇒(i) Let ui = fxi (x), i = 1,...,d, let u = (u1,...,ud) <strong>and</strong> let {b1,...,bd}<br />

be the st<strong>and</strong>ard basis of E d . Then<br />

f (x + tbi) = f (x) + uit + o(|t|) as t → 0fori = 1,...,d.<br />

Combined with Jensen’s inequality, this shows that<br />

(7) f (y) = f � x + (y − x) �<br />

= f<br />

�<br />

1 � � 1 � �<br />

x + d(y1 − x1)b1 +···+ x + d(yd − xd)bd<br />

d<br />

d<br />

�<br />

≤ 1<br />

d f � � 1<br />

x + d(y1 − x1)b1 +···+<br />

d f � �<br />

x + d(yd − xd)bd<br />

= f (x) + u1(y1 − x1) +···+ud(yd − xd)<br />

+ o(|y1 − xy|) +···+o(|yd − xd|)<br />

= f (x) + u · (y − x) + o(�y − x�) as y → x, y ∈ C.<br />

A similar argument yields the following:<br />

(8) f (2x − y) ≤ f (x) + u · (x − y) + o(�y − x�) as y → x, 2x − y ∈ C.<br />

The final inequality we are seeking follows from the convexity of f :<br />

�<br />

1 1<br />

�<br />

(9) f (x) = f y + (2x − y) ≤<br />

2 2 1 1<br />

f (y) + f (2x − y) for y, 2x − y ∈ C.<br />

2 2<br />

The inequalities (9) <strong>and</strong> (8) now imply that<br />

f (x) + u · (y − x) ≤ 1 1<br />

f (y) + f (2x − y) + u · (y − x)<br />

2 2<br />

≤ 1 1<br />

f (y) +<br />

2 2<br />

1<br />

f (x) + u · (y − x) + o(�y − x�) as y → x, y ∈ C,<br />

2

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

Saved successfully!

Ooh no, something went wrong!