TheoremBase

The conjugate exponent is found by solving a linear equation. Young's inequality is the convexity of the exponential function, obtained from the bound exp(u) >= 1 + u, evaluated at the logarithms of apa^p and bqb^q.

Proof

Conjugate exponents. Since p>1p>1, we have p−1>0p-1>0, so q=p/(p−1)q=p/(p-1) is defined and

1q=p−1p=1−1p.\frac{1}{q}=\frac{p-1}{p}=1-\frac{1}{p} .

Conversely, if a real number q′q' satisfies 1/p+1/q′=11/p+1/q'=1, then 1/q′=(p−1)/p1/q'=(p-1)/p, and so q′=p/(p−1)=qq'=p/(p-1)=q. Finally p>p−1>0p>p-1>0 gives q>1q>1.

A convexity inequality for the exponential function. Let x,y,θ∈Rx,y,\theta\in\mathbb{R} with 0<θ<10<\theta<1, and put m=θx+(1−θ)ym=\theta x+(1-\theta)y. Since exp⁡(m)>0\exp(m)>0 and exp⁡(u)≥1+u\exp(u)\ge1+u for every real uu (Basic Properties of the Exponential Function (iii) and Basic Properties of the Exponential Function (v)), Basic Properties of the Exponential Function (ii) gives

exp⁡(x)=exp⁡(m)exp⁡(x−m)≥exp⁡(m)(1+x−m),exp⁡(y)=exp⁡(m)exp⁡(y−m)≥exp⁡(m)(1+y−m).\exp(x)=\exp(m)\exp(x-m)\ge\exp(m)(1+x-m),\qquad \exp(y)=\exp(m)\exp(y-m)\ge\exp(m)(1+y-m).

Multiplying the first inequality by θ>0\theta>0 and the second by 1−θ>01-\theta>0 and adding, and using θ(x−m)+(1−θ)(y−m)=m−m=0\theta(x-m)+(1-\theta)(y-m)=m-m=0, we obtain

θexp⁡(x)+(1−θ)exp⁡(y)≥exp⁡(θx+(1−θ)y).\theta\exp(x)+(1-\theta)\exp(y)\ge\exp\bigl(\theta x+(1-\theta)y\bigr).

Young's inequality. If a=0a=0 or b=0b=0, then ab=0ab=0, while ap≥0a^{p}\ge0 and bq≥0b^{q}\ge0 by Properties of Real Powers (ii); so the inequality holds. Let now a,b>0a,b>0, and put

x=log⁡ap=plog⁡a,y=log⁡bq=qlog⁡b,θ=1p,x=\log a^{p}=p\log a,\qquad y=\log b^{q}=q\log b,\qquad \theta=\frac{1}{p},

by Properties of Real Powers (iii) (ap,bq>0a^{p},b^{q}>0 by Properties of Real Powers (ii)). Then 0<θ<10<\theta<1 and 1−θ=1/q1-\theta=1/q, so θx+(1−θ)y=log⁡a+log⁡b=log⁡(ab)\theta x+(1-\theta)y=\log a+\log b=\log(ab) by Properties of the Natural Logarithm: Values at 1 and e, Products, Quotients, Monotonicity, Continuity, Derivative and the Bound log t at most t - 1 (ii). By The Natural Logarithm (ii), exp⁡(x)=ap\exp(x)=a^{p}, exp⁡(y)=bq\exp(y)=b^{q} and exp⁡(log⁡(ab))=ab\exp(\log(ab))=ab, and the convexity inequality becomes

app+bqq≥ab.\frac{a^{p}}{p}+\frac{b^{q}}{q}\ge ab .

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…