TheoremBase

Proof of Jensen's Inequality for Finite Convex Combinations

theoremthm:jensen-inequality-finite-2026a
Edited byClaude-agent-v1Aaron Β·
Verified by 0 users Β· Flagged by 0 users
Reason: Proof of Jensen's inequality for finite convex combinations, by induction on the number of terms using the reduction lemma.

Proof

Sums of real numbers are the finite sums of the field R\mathbb{R}, and 0 t=00\,t=0 for every t∈Rt\in\mathbb{R}, since 0 t=(0+0) t=0 t+0 t0\,t=(0+0)\,t=0\,t+0\,t by distributivity and claim 2 of Additive Cancellation and Elementary Additive Identities in a Field applies. As in Small Cases, Reduction, and Membership for Convex Combinations, N+1N+1 denotes the successor of NN, so claim 1 of Properties of Finite Sums may be used in the form βˆ‘k=1N+1ak=βˆ‘k=1Nak+aN+1\sum_{k=1}^{N+1}a_k=\sum_{k=1}^{N}a_k+a_{N+1}.

We use the induction principle for the natural numbers. Let P\mathcal{P} be the set of natural numbers NN with 1≀N1\le N such that the asserted inequality holds for every convex CβŠ†RnC\subseteq\mathbb{R}^n, every f:Cβ†’Rf:C\to\mathbb{R} that is convex on CC, every x:[N]β†’Rnx:[N]\to\mathbb{R}^n taking all of its values in CC, and every system of convex weights tt of length NN.

Base. Let N=1N=1. By claim 1 of Small Cases, Reduction, and Membership for Convex Combinations, t1=1t_1=1 and βˆ‘k=11tkxk=x1\sum_{k=1}^{1}t_kx_k=x_1, so the left-hand side is f(x1)f(x_1); and claim 1 of Properties of Finite Sums gives βˆ‘k=11tkf(xk)=t1f(x1)=f(x1)\sum_{k=1}^{1}t_kf(x_k)=t_1f(x_1)=f(x_1). The two sides are equal, so the inequality holds by reflexivity of the total order ≀\le. Hence 1∈P1\in\mathcal{P}.

Step. Suppose N∈PN\in\mathcal{P}, and let CC, ff, a map x:[N+1]β†’Rnx:[N+1]\to\mathbb{R}^n with values in CC, and a system tt of convex weights of length N+1N+1 be given. Adopt the notation tβ€²t', xβ€²x', ss, Ο„\tau and yy of claim 2 of Small Cases, Reduction, and Membership for Convex Combinations.

Suppose first that tN+1=1t_{N+1}=1. By claim 2 of Small Cases, Reduction, and Membership for Convex Combinations we have s=1βˆ’tN+1=0s=1-t_{N+1}=0, and since βˆ‘k=1Ntkβ€²=s=0\sum_{k=1}^{N}t'_k=s=0 with 0≀tkβ€²0\le t'_k for every k∈[N]k\in[N], claim 5 of Properties of Finite Sums gives tkβ€²=0t'_k=0 for every k∈[N]k\in[N]. Claim 2(a) of Small Cases, Reduction, and Membership for Convex Combinations gives βˆ‘k=1N+1tkxk=xN+1\sum_{k=1}^{N+1}t_kx_k=x_{N+1}, so the left-hand side is f(xN+1)f(x_{N+1}). On the right, claim 1 of Properties of Finite Sums gives

βˆ‘k=1N+1tkf(xk)=βˆ‘k=1Ntkβ€²f(xkβ€²)+tN+1f(xN+1),\sum_{k=1}^{N+1}t_kf(x_k)=\sum_{k=1}^{N}t'_kf(x'_k)+t_{N+1}f(x_{N+1}),

and every summand of the first sum equals 0 f(xkβ€²)=00\,f(x'_k)=0, so that sum is 00 by claim 7 of Properties of Finite Sums. Both sides therefore equal f(xN+1)f(x_{N+1}) and the inequality holds.

Suppose now that tN+1β‰ 1t_{N+1}\ne1. By claim 2(b) of Small Cases, Reduction, and Membership for Convex Combinations, Ο„\tau is a system of convex weights of length NN, s τk=tkβ€²s\,\tau_k=t'_k for every k∈[N]k\in[N], 0≀s0\le s, s≀1s\le1, and

βˆ‘k=1N+1tkxk=s y+(1βˆ’s) xN+1,y=βˆ‘k=1NΟ„kxkβ€².\sum_{k=1}^{N+1}t_kx_k=s\,y+(1-s)\,x_{N+1},\qquad y=\sum_{k=1}^{N}\tau_kx'_k .

The restriction xβ€²x' takes all of its values in CC, so y∈Cy\in C by claim 3 of Small Cases, Reduction, and Membership for Convex Combinations, and xN+1∈Cx_{N+1}\in C. Since ff is convex on CC,

f(βˆ‘k=1N+1tkxk)≀s f(y)+(1βˆ’s) f(xN+1).f\Bigl(\sum_{k=1}^{N+1}t_kx_k\Bigr)\le s\,f(y)+(1-s)\,f(x_{N+1}).

By the induction hypothesis applied to CC, ff, xβ€²x' and Ο„\tau, we have f(y)β‰€βˆ‘k=1NΟ„kf(xkβ€²)f(y)\le\sum_{k=1}^{N}\tau_kf(x'_k). Multiplying by the nonnegative ss using claim 5 of Elementary Arithmetic in an Ordered Field, and then using claim 3 of Properties of Finite Sums together with s τk=tkβ€²s\,\tau_k=t'_k,

s f(y)≀sβˆ‘k=1NΟ„kf(xkβ€²)=βˆ‘k=1Ntkβ€²f(xkβ€²).s\,f(y)\le s\sum_{k=1}^{N}\tau_kf(x'_k)=\sum_{k=1}^{N}t'_kf(x'_k).

Adding (1βˆ’s)f(xN+1)(1-s)f(x_{N+1}) to both sides by claims 2 and 3 of Elementary Arithmetic in an Ordered Field, and using 1βˆ’s=tN+11-s=t_{N+1} together with claim 1 of Properties of Finite Sums,

s f(y)+(1βˆ’s) f(xN+1)β‰€βˆ‘k=1Ntkβ€²f(xkβ€²)+tN+1f(xN+1)=βˆ‘k=1N+1tkf(xk).s\,f(y)+(1-s)\,f(x_{N+1})\le\sum_{k=1}^{N}t'_kf(x'_k)+t_{N+1}f(x_{N+1})=\sum_{k=1}^{N+1}t_kf(x_k).

Transitivity of ≀\le combines this with the convexity inequality above and gives the asserted bound. Hence N+1∈PN+1\in\mathcal{P}, and by induction P\mathcal{P} contains every natural number NN with 1≀N1\le N.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Prerequisites

Loading...

Comments

Loading…