next up previous
Next: Inequalities for convex functions Up: Inequalities Previous: The power mean inequality


Convex functions

A function $ f(x)$ is convex if for any real numbers $ a<b$, each point $ (c,d)$ on the line segment joining $ (a,f(a))$ and $ (b,f(b))$ lies above or at the point $ (c,f(c))$ on the graph of $ f$ with the same $ x$-coordinate.

Algebraically, this condition says that

$\displaystyle f((1-t)a+tb) \le (1-t)f(a) + t f(b).$ (1)

whenever $ a<b$ and for all $ t \in [0,1]$. (The left hand side represents the height of the graph of the function above the $ x$-value $ x=(1-t)a+tb$ which is a fraction $ t$ of the way from $ a$ to $ b$, and the right hand side represents the height of the line segment above the same $ x$-value.)

Those who know what a convex set in geometry is can interpret the condition as saying that the set $ S=\{(x,y):y \ge f(x)\}$ of points above the graph of $ f$ is a convex set. Loosely speaking, this will hold if the graph of $ f$ curves in the shape of a smile instead of a frown. For example, the function $ f(x)=x^2$ is convex, as is $ f(x)=x^n$ for any positive even integer.

One can also speak of a function $ f(x)$ being convex on an interval $ I$. This means that the condition (1) above holds at least when $ a,b \in I$ (and $ a<b$ and $ t \in [0,1]$). For example, one can show that $ f(x)=x^3$ is convex on $ [0,\infty)$, and that $ f(x)=\sin x$ is convex on $ (-\pi,0)$.

Finally one says that a function $ f(x)$ on an interval $ I$ is strictly convex, if

$\displaystyle f((1-t)a+tb) < (1-t)f(a) + t f(b)$

whenever $ a,b \in I$ and $ a<b$ and $ t \in (0,1)$. In other words, the line segment connecting two points on the graph of $ f$ should lie entirely above the graph of $ f$, except where it touches at its endpoints.

For convenience, here is a brief list of some convex functions. In these, $ k$ represents a positive integer, $ r,s$ represent real constants, and $ x$ is the variable. In fact, all of these are strictly convex on the interval given, except for $ x^r$ and $ -x^r$ when $ r$ is 0 or $ 1$.

$\displaystyle x^{2k},$ on all of $ {\mathbb{R}}$    
$\displaystyle x^r,$ on $ [0,\infty)$, if $ r \ge 1$    
$\displaystyle -x^r,$ on $ [0,\infty)$, if $ r \in [0,1]$    
$\displaystyle x^r,$ on $ (0,\infty)$, if $ r \le 0$    
$\displaystyle -\log x,$ on $ (0,\infty)$    
$\displaystyle -\sin x,$ on $ [0,\pi]$    
$\displaystyle -\cos x,$ on $ [-\pi/2,\pi/2]$    
$\displaystyle \tan x,$ on $ [0,\pi/2)$    
$\displaystyle e^x,$ on all of $ {\mathbb{R}}$    
$\displaystyle r/(s+x)$ on $ (-s,\infty)$, if $ r>0$    

A sum of convex functions is convex. Adding a constant or linear function to a function does not affect convexity.


Remarks (for those who know about continuity and derivatives):
If one wants to prove rigorously that a function is convex, instead of just guessing it from the graph, it is often easier to use one of the criteria below instead of the definition of convexity.

  1. Let $ f(x)$ be a continuous function on an interval $ I$. Then $ f(x)$ is convex if and only if $ (f(a)+f(b))/2 \ge f((a+b)/2)$ holds for all $ a,b \in I$. Also, $ f(x)$ is strictly convex if and only if $ (f(a)+f(b))/2 > f((a+b)/2)$ whenever $ a,b \in I$ and $ a<b$.

  2. Let $ f(x)$ be a differentiable function on an interval $ I$. Then $ f(x)$ is convex if and only if $ f'(x)$ is increasing on $ I$. Also, $ f(x)$ is strictly convex if and only if $ f'(x)$ is strictly increasing on the interior of $ I$.

  3. Let $ f(x)$ be a twice differentiable function on an interval $ I$. Then $ f(x)$ is convex if and only if $ f''(x) \ge 0$ for all $ x \in I$. Also, $ f(x)$ is strictly convex if and only if $ f''(x) > 0$ for all $ x$ in the interior of $ I$.


next up previous
Next: Inequalities for convex functions Up: Inequalities Previous: The power mean inequality
Zvezdelina Stankova-Frenkel 2001-11-18