Logarithmically convex function explained

In mathematics, a function f is logarithmically convex or superconvex[1] if

{log}\circf

, the composition of the logarithm with f, is itself a convex function.

Definition

Let be a convex subset of a real vector space, and let be a function taking non-negative values. Then is:

{log}\circf

is convex, and

{log}\circf

is strictly convex.Here we interpret

log0

as

-infty

.

Explicitly, is logarithmically convex if and only if, for all and all, the two following equivalent conditions hold:

\begin{align} logf(tx1+(1-t)x2)&\letlogf(x1)+(1-t)logf(x2),\\ f(tx1+(1-t)x2)&\le

1-t
f(x
2)

. \end{align}

Similarly, is strictly logarithmically convex if and only if, in the above two expressions, strict inequality holds for all .

The above definition permits to be zero, but if is logarithmically convex and vanishes anywhere in, then it vanishes everywhere in the interior of .

Equivalent conditions

If is a differentiable function defined on an interval, then is logarithmically convex if and only if the following condition holds for all and in :

logf(x)\gelogf(y)+

f'(y)
f(y)

(x-y).

This is equivalent to the condition that, whenever and are in and,
\left(f(x)
f(y)
1
x-y
\right)

\ge\exp\left(

f'(y)
f(y)

\right).

Moreover, is strictly logarithmically convex if and only if these inequalities are always strict.

If is twice differentiable, then it is logarithmically convex if and only if, for all in,

f''(x)f(x)\gef'(x)2.

If the inequality is always strict, then is strictly logarithmically convex. However, the converse is false: It is possible that is strictly logarithmically convex and that, for some, we have

f''(x)f(x)=f'(x)2

. For example, if

f(x)=\exp(x4)

, then is strictly logarithmically convex, but

f''(0)f(0)=0=f'(0)2

.

Furthermore,

f\colonI\to(0,infty)

is logarithmically convex if and only if

e\alphaf(x)

is convex for all

\alpha\inR

.[2] [3]

Sufficient conditions

If

f1,\ldots,fn

are logarithmically convex, and if

w1,\ldots,wn

are non-negative real numbers, then
w1
f
1

wn
f
n
is logarithmically convex.

If

\{fi\}i

is any family of logarithmically convex functions, then

g=\supifi

is logarithmically convex.

If

f\colonX\toI\subseteqR

is convex and

g\colonI\toR\ge

is logarithmically convex and non-decreasing, then

g\circf

is logarithmically convex.

Properties

A logarithmically convex function f is a convex function since it is the composite of the increasing convex function

\exp

and the function

log\circf

, which is by definition convex. However, being logarithmically convex is a strictly stronger property than being convex. For example, the squaring function

f(x)=x2

is convex, but its logarithm

logf(x)=2log|x|

is not. Therefore the squaring function is not logarithmically convex.

Examples

f(x)=\exp(|x|p)

is logarithmically convex when

p\ge1

and strictly logarithmically convex when

p>1

.

f(x)=

1
xp
is strictly logarithmically convex on

(0,infty)

for all

p>0.

See also

References

Notes and References

  1. Kingman, J.F.C. 1961. A convexity property of positive matrices. Quart. J. Math. Oxford (2) 12,283-284.
  2. .
  3. .