Inequalities Related to Expecations
Markov's Inequality
Claim If \(X\geq 0\) with \(\mu <\infty\). Then for all \(a>0\)
proof.
Chebychev's Inequality
Claim Let \(X\) be a r.v. with mean \(\mu\) and \(\sigma^2\). Then, for all \(a > 0\)
proof 1.
proof 2. Taking \(Y = (X-\mu)^2/\sigma^2\) and \(a = k^2\), using Markov's Inequality, in this case
Cauchy-Schwartz Inequality
Claim If \(X,Y\) are r.v. with finite second moment
where the equality holds IFF \(P(aX=bY) = 1\) for some \(a,b\in\mathbb R\)
proof. Note that we can consider the expectation as a inner product.
Corollary \(cov(X, Y)^2 \leq var(X)var(Y)\)
proof. By CS Ineq.
Range of Correlation
Claim \(corr(X, Y) \in [-1, 1]\)
proof. By the definition of correlation and the corollary
Corollary \(Y=aX+b\) IFF \(corr(X, Y) = 1\times \text{sign}(a)\)
Additional Inequalities
Note that the following inequalities are given without proofs.
Young's Inequality
For \(p, q > 1\) with \(p^{-1} + q^{-1} = 1\) and \(x, y\geq 0\).
with equality holds IFF \(x^p = y^q\)
Holder's Inequality
For \(p, q > 1\) with \(p^{-1}+q^{-1} = 1\).
proof. The statement is obtained by treating expectation as an inner product, and CS inequality is a special case where \(p=q=2\).
Jensen's Inequality
For a convex function \(\psi\), \(\psi(\mathbb E(X)) \leq \mathbb E(\psi(X))\)
proof. This inequality can be easily derived from convexity definition
Lyapounov's Inequality
If \(\mathbb E(|X|^p)\) is finite for some \(p>0\), then \(\mathbb E(|X|^q) \leq \mathbb E(|X|^p)^{q/p}\) for all \(0 < q \leq p\).
proof. \(x^{p/q}\) is convex, apply Jensen's Inequality
Minkowski's Inequality
For \(p\geq 1, \|X+Y\|_p \leq \|X\|_p + \|Y\|_p\)