Jensen Inequality

Published: 2020-03-07

Jensen Inequality is an inequality in mathematics that relates to the concave/convex function. A function is concave if the line segment between any two points on it lies below the graph of the function. Mathematically we can write:

function f(x)f(x) is concave if, if any a,b,αa,b,\alpha satisfies:

f(αa+(1α)b)αf(a)+(1α)f(b)(1) f(\alpha a+(1-\alpha)b) \geq \alpha f(a) + (1-\alpha) f(b) \tag{1}

where 0α10 \leq \alpha \leq 1, and a,bXa,b \in X. Here, XX is a concave set.

Whereas, in a convex function, the line segment between any two points is lies above the graph. So, we only need to change the \geq sign in equation (1)(1) to \leq.

For example log(x)\log(x) is a concave function, and log(x)-\log(x) is a convex function. The definition above can be visualized as follows:

That’s the case for 2 points. Now, consider if we have 3 points instead. If α1+α2+α3=1\alpha_1+\alpha_2+\alpha_3 = 1 and αk0\alpha_k \geq 0 and f(x)f(x) is concave:

f(α1a1+α2a2+α3a3)α1f(a1)+α2f(a2)+α3f(a3) f(\alpha_1 a_1+\alpha_2 a_2+\alpha_3 a_3) \geq \alpha_1 f(a_1)+\alpha_2 f(a_2)+\alpha_3 f(a_3)

Thus, using the definition of expectation, we can generalize above equation to (concave and convex respectively):

f(EP(x)x)EP(x)f(x)f(EP(x)x)EP(x)f(x)f(\mathbb{E}_{P(x)}x) \geq \mathbb{E}_{P(x)} f(x) \\ f(\mathbb{E}_{P(x)}x) \leq \mathbb{E}_{P(x)} f(x)

Thats it, Jensen Inequality is the generalized form of the statement that the secant line of a convex function lies above the graph of the function.

References: