Homepage Solution manuals Terence Tao Probability Theory Exercise 1.40 (Complex Jensen inequality)

Exercise 1.40 (Complex Jensen inequality)

Let f : C R be a convex function, and let X be a complex random variable with X and f(X) both absolutely integrable. Show that

f(EX) Ef(X).

Answers

We make use of the supporting hyperplane theorem:

Theorem 1. (Supporting hyperplane, complex version) Let f : be a convex function. Then, for each x0 there exists an affine function l(x) = ax + b such that

f(x) l(x)

holds for all x n and l supports f at the point x0:

f(x0) = l(x0).

We apply the theorem to our assertion by setting x0 = EX; i.e, we find an affine lEX(x) = ax + b such that lEX(EX) = f(EX). In particular,

f(X) aX + b.

Taking expectations and using linearity of expectation, we conclude

Ef(X) E(aX + b) = aEX + b = lEX(EX) = f(EX).

We now provide the proof for the supporting hyperplane theorem that we have used in this theorem.

Proof. (Supporting hyperplane) Let Δ > 0 be arbitrary. Notice that the point f(x0) is in the middle of f(x0 + Δ) and f(x0 Δ); thus,

f(x0) = f (1 2(x0 + Δ) + 1 2(x0 Δ)) 1 2f(x0 + Δ) + 1 2f(x0 Δ).

From this we immediate

1 2f(x0) 1 2f(x0 Δ) 1 2f(x0 + Δ) 1 2(x0)

and thus,

f(x0) f(x0 Δ) Δ f(x0 + Δ) f(x0) Δ .

We can take limits while preserving inequality

sup Δ>0f(x0) f(x0 Δ) Δ inf Δ>0f(x0 + Δ) f(x0) Δ .

Let a be an arbitrary number between the two limits. Define

l : ,xa(x x0) + f(x0).

We then have

l(x0) = f(x0)

and

  • If x x0 then with Δ := x x0 we have f(x) f(x0) = f(x + Δ) f(x0) aΔ + f(x0) by construction of a. Therefore, f(x) l(x).
  • If x x0 we similarly obtain f(x) l(x).

Thus, l satisfies all of the desired properties. □

User profile picture
2021-09-04 00:00
Comments