TheoremBase

Proof of The Function slogss\log s: Continuity, Young's Inequality and Lower Bounds, with the Elementary Bounds for the Exponential and the Logarithm

lemmalem:entropy-function-real-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 9,657 chars · 18 deps · depth 16 Reason: Proof of lem:entropy-function-real-2026a (E2 Stage 0).

The bound exp(u) >= 1+u for negative u comes from the mean value theorem on [u,0]; the two-sided logarithm bound follows by substituting u = log t and passing to 1/t. Continuity at 0 uses |s log s| <= 2 sqrt(s) for 0<s<1, measurability follows by composing f with a continuous extension of phi to the real line, and Young's inequality and the lower bounds follow from the exponential bound and the logarithm bound.

Proof

Each result cited below is universally quantified over the data in its own statement. Throughout, we use the identities log(exp(u))=u\log(\exp(u))=u for uRu\in\mathbb{R} and exp(logt)=t\exp(\log t)=t for positive tt, and the rule log(st)=logs+logt\log(st)=\log s+\log t for positive s,ts,t, all recorded in The Natural Logarithm. By claim 1 of Basic Properties of the Exponential Function, exp(0)=1\exp(0)=1, hence log1=log(exp(0))=0\log 1=\log(\exp(0))=0.

Claim 1. Let uRu\in\mathbb{R}. If 0u0\le u, then 1+uexp(u)1+u\le\exp(u) by claim 4 of Basic Properties of the Exponential Function. Suppose now u<0u<0, and let g:[u,0]Rg:[u,0]\to\mathbb{R} be the restriction of exp\exp to the closed interval [u,0][u,0]. The set R\mathbb{R} is an interval, and each xRx\in\mathbb{R} is an interior point of it, since x1<x<x+1x-1<x<x+1. By claim 3 of Basic Properties of the Exponential Function, exp\exp is differentiable at every xRx\in\mathbb{R} with derivative exp(x)\exp(x); hence, by Differentiability at an Interior Point Implies Continuity There, exp\exp is continuous at every xx relative to R\mathbb{R}. Consequently gg is continuous on [u,0][u,0], because the condition defining continuity relative to [u,0][u,0] only concerns points yy of [u,0]R[u,0]\subseteq\mathbb{R}. Likewise, for every cc with u<c<0u<c<0 the point cc is an interior point of [u,0][u,0], and gg is differentiable at cc with g(c)=exp(c)g'(c)=\exp(c): a δ\delta that serves for exp\exp at cc in the definition of the derivative also serves for gg, because the condition for gg only concerns those hh with c+h[u,0]c+h\in[u,0]. By Mean Value Theorem on a Closed Real Interval there is cc with u<c<0u<c<0 and

exp(c)=exp(0)exp(u)0u,that is,1exp(u)=exp(c)(u).\exp(c)=\frac{\exp(0)-\exp(u)}{0-u},\qquad\text{that is,}\qquad 1-\exp(u)=\exp(c)\,(-u).

Since exp\exp is strictly increasing by claim 4 of Basic Properties of the Exponential Function and c<0c<0, we have exp(c)<exp(0)=1\exp(c)<\exp(0)=1. As 0u0\le-u, multiplying exp(c)1\exp(c)\le1 by u-u (claim 5 of Elementary Arithmetic in an Ordered Field) gives 1exp(u)=exp(c)(u)u1-\exp(u)=\exp(c)(-u)\le-u, that is, 1+uexp(u)1+u\le\exp(u).

Claim 2. Let tt be positive. Claim 1 with u=logtu=\log t gives 1+logtexp(logt)=t1+\log t\le\exp(\log t)=t, so logtt1\log t\le t-1. By claim 7 of Elementary Order Arithmetic in an Ordered Field, t1t^{-1} exists and is positive, and 0=log1=log(tt1)=logt+log(t1)0=\log 1=\log(t\,t^{-1})=\log t+\log(t^{-1}), so log(t1)=logt\log(t^{-1})=-\log t. The upper bound just proved, applied to t1t^{-1}, gives logtt11-\log t\le t^{-1}-1, that is, 1t1logt1-t^{-1}\le\log t.

Claim 3. Step 1 (continuity of the logarithm). Let tt be positive and let 0<ε0<\varepsilon. Put a=exp(logtε)a=\exp(\log t-\varepsilon) and b=exp(logt+ε)b=\exp(\log t+\varepsilon); since exp\exp is strictly increasing (claim 4 of Basic Properties of the Exponential Function) and exp(logt)=t\exp(\log t)=t, we have 0<a<t<b0<a<t<b, where 0<a0<a is claim 2 there. By claim 9 of Elementary Order Arithmetic in an Ordered Field let δ\delta be the smaller of tat-a and btb-t; then 0<δ0<\delta. Let yy be positive with yt<δ|y-t|<\delta. By claim 9 of Properties of the Absolute Value in an Ordered Field, tδ<y<t+δt-\delta<y<t+\delta, and as atδa\le t-\delta and t+δbt+\delta\le b, claim 2 of Elementary Order Arithmetic in an Ordered Field gives a<y<ba<y<b. Since log\log is strictly increasing on (0,)(0,\infty) by claim 2 of Real Powers Through the Exponential, and Elementary Asymptotic Tools: Monotonicity, Null Sequences of Negative Powers, Exponential Domination, Integer Rounding, and Square-Root and Exponential Inequalities, and loga=logtε\log a=\log t-\varepsilon, logb=logt+ε\log b=\log t+\varepsilon, we get logtε<logy<logt+ε\log t-\varepsilon<\log y<\log t+\varepsilon, so logylogt<ε|\log y-\log t|<\varepsilon by claim 9 of Properties of the Absolute Value in an Ordered Field. Thus log\log is continuous at tt relative to (0,)(0,\infty), as a map from the subset (0,)(0,\infty) of the real line (R,dR)(\mathbb{R},d_{\mathbb{R}}) of The Absolute Value Metric on the Real Line into (R,dR)(\mathbb{R},d_{\mathbb{R}}).

Step 2 (continuity at a positive point). Let ss be positive and put A=(0,)A=(0,\infty). The map yyy\mapsto y on AA is continuous at ss relative to AA (given 0<ε0<\varepsilon, take δ=ε\delta=\varepsilon), and so is log\log by Step 1; hence their product yylogyy\mapsto y\log y, which agrees with ϕ\phi on AA, is continuous at ss relative to AA by claim 3 of Continuity of Sums and Products of Real-Valued Functions on a Metric Space. Let 0<ε0<\varepsilon, let δ1\delta_{1} be positive with ϕ(y)ϕ(s)<ε|\phi(y)-\phi(s)|<\varepsilon for every yAy\in A with ys<δ1|y-s|<\delta_{1}, and let δ\delta be the smaller of δ1\delta_{1} and ss (claim 9 of Elementary Order Arithmetic in an Ordered Field). If y[0,)y\in[0,\infty) and ys<δ|y-s|<\delta, then sysy=ys<ss-y\le|s-y|=|y-s|<s, by claim 3 of Properties of the Absolute Value in an Ordered Field applied to sys-y, claim 2 there (as sy=(ys)s-y=-(y-s)), and claim 2 of Elementary Order Arithmetic in an Ordered Field (as ys<δs|y-s|<\delta\le s); so 0<y0<y by claim 1 of Elementary Order Arithmetic in an Ordered Field, and yAy\in A, and ys<δ1|y-s|<\delta_{1}; hence ϕ(y)ϕ(s)<ε|\phi(y)-\phi(s)|<\varepsilon. So ϕ\phi is continuous at ss relative to [0,)[0,\infty).

Step 3 (continuity at 00). First let 0<y<10<y<1; we show ϕ(y)2r|\phi(y)|\le2r, where rr is the nonnegative real number with r2=yr^{2}=y given by Existence and Uniqueness of the Nonnegative Square Root. Since r2=y0r^{2}=y\ne0, r0r\ne0, so rr is positive; and r<1r<1 by claim 1 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field, since r2=y<1=12r^{2}=y<1=1^{2}. Now logy=log(rr)=2logr\log y=\log(r\,r)=2\log r, and 1r1logr1-r^{-1}\le\log r by Claim 2, so multiplying by 22, which is positive by claim 8 of Elementary Order Arithmetic in an Ordered Field, gives 22r1logy2-2r^{-1}\le\log y (claim 5 of Elementary Arithmetic in an Ordered Field); multiplying by y=r2y=r^{2}, which is nonnegative (claim 5 of Elementary Arithmetic in an Ordered Field), gives 2r22rylogy2r^{2}-2r\le y\log y, and 2r22r2r2r^{2}-2r\ge-2r because 0<2r20<2r^{2} (claim 5 of Elementary Order Arithmetic in an Ordered Field, as 0<20<2 and 0<r20<r^{2}). On the other hand logyy1<0\log y\le y-1<0 by Claim 2, so multiplying logy0\log y\le0 by yy (claim 5 of Elementary Arithmetic in an Ordered Field) gives ylogy02ry\log y\le0\le 2r, where 0<2r0<2r by claim 5 of Elementary Order Arithmetic in an Ordered Field. Hence 2rϕ(y)2r-2r\le\phi(y)\le2r, and ϕ(y)2r|\phi(y)|\le2r by claim 6 of Properties of the Absolute Value in an Ordered Field.

Now let 0<ε0<\varepsilon, and by claim 9 of Elementary Order Arithmetic in an Ordered Field let δ\delta be the smaller of 11 and ε2/4\varepsilon^{2}/4; then 0<δ0<\delta. Let y[0,)y\in[0,\infty) with y0<δ|y-0|<\delta. If y=0y=0, then ϕ(y)ϕ(0)=0<ε|\phi(y)-\phi(0)|=0<\varepsilon. Otherwise 0<y<10<y<1, and with rr as above, r2=y<ε2/4=(ε/2)2r^{2}=y<\varepsilon^{2}/4=(\varepsilon/2)^{2}, so r<ε/2r<\varepsilon/2 by claim 1 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field; hence ϕ(y)ϕ(0)=ϕ(y)2r<ε|\phi(y)-\phi(0)|=|\phi(y)|\le2r<\varepsilon. So ϕ\phi is continuous at 00 relative to [0,)[0,\infty), and with Step 2, ϕ\phi is continuous on [0,)[0,\infty).

Step 4 (measurability). For xRx\in\mathbb{R} let x+=xx^{+}=x if 0x0\le x and x+=0x^{+}=0 otherwise, so that x+[0,)x^{+}\in[0,\infty). For all x,yRx,y\in\mathbb{R} we have y+x+yx|y^{+}-x^{+}|\le|y-x|: if x,yx,y are both nonnegative the two sides are equal; if both are negative the left side is 00; if 0y0\le y and x<0x<0 the left side is yyx=yxy\le y-x=|y-x|; and the remaining case is symmetric. Let ψ:RR\psi:\mathbb{R}\to\mathbb{R} be given by ψ(x)=ϕ(x+)\psi(x)=\phi(x^{+}). Then ψ\psi is continuous on R\mathbb{R}: given xRx\in\mathbb{R} and 0<ε0<\varepsilon, let δ\delta be positive with ϕ(z)ϕ(x+)<ε|\phi(z)-\phi(x^{+})|<\varepsilon for all z[0,)z\in[0,\infty) with zx+<δ|z-x^{+}|<\delta (Steps 2 and 3); if yx<δ|y-x|<\delta then y+x+<δ|y^{+}-x^{+}|<\delta by claim 2 of Elementary Order Arithmetic in an Ordered Field, so ψ(y)ψ(x)<ε|\psi(y)-\psi(x)|<\varepsilon.

By claim 2 of Borel Measurability and Bounded Integration on a Metric Space, the Borel σ\sigma-algebra of the metric space (R,dR)(\mathbb{R},d_{\mathbb{R}}) is B(R)\mathcal{B}(\mathbb{R}); so by claim 3 of that lemma, applied with both metric spaces equal to (R,dR)(\mathbb{R},d_{\mathbb{R}}), ψ\psi is measurable with respect to B(R)\mathcal{B}(\mathbb{R}) and B(R)\mathcal{B}(\mathbb{R}). Let (X,F,μ)(X,\mathcal{F},\mu) be a measure space and let f:XRf:X\to\mathbb{R} be measurable with 0f(x)0\le f(x) for every xXx\in X. By Measure Spaces and the Lebesgue Integral: Standing Notation §measurable, ff is measurable with respect to F\mathcal{F} and B(R)\mathcal{B}(\mathbb{R}). By claim 4 of Borel Measurability and Bounded Integration on a Metric Space, applied with the metric space (R,dR)(\mathbb{R},d_{\mathbb{R}}), the measurable space (X,F)(X,\mathcal{F}) in the role of (Ω,F)(\Omega,\mathcal{F}), the measurable space (R,B(R))(\mathbb{R},\mathcal{B}(\mathbb{R})) in the role of (Z,G)(Z,\mathcal{G}), and the maps ff and ψ\psi, the composite ψf\psi\circ f is measurable with respect to F\mathcal{F} and B(R)\mathcal{B}(\mathbb{R}), that is, measurable in the sense of Measure Spaces and the Lebesgue Integral: Standing Notation §measurable. Since f(x)+=f(x)f(x)^{+}=f(x) for every xXx\in X, ψf=ϕf\psi\circ f=\phi\circ f, which is therefore measurable.

Claim 4. Let aRa\in\mathbb{R} and s[0,)s\in[0,\infty). If s=0s=0, then as=0<exp(a1)=ϕ(0)+exp(a1)a\,s=0<\exp(a-1)=\phi(0)+\exp(a-1) by claim 2 of Basic Properties of the Exponential Function. Let ss be positive. By claim 1 of Basic Properties of the Exponential Function,

exp(a1)=exp(logs+(a1logs))=exp(logs)exp(a1logs)=sexp(a1logs).\exp(a-1)=\exp\bigl(\log s+(a-1-\log s)\bigr)=\exp(\log s)\,\exp(a-1-\log s)=s\,\exp(a-1-\log s).

By Claim 1, alogs=1+(a1logs)exp(a1logs)a-\log s=1+(a-1-\log s)\le\exp(a-1-\log s), and multiplying by ss (claim 5 of Elementary Arithmetic in an Ordered Field) gives asslogsexp(a1)a\,s-s\log s\le\exp(a-1), that is, asϕ(s)+exp(a1)a\,s\le\phi(s)+\exp(a-1).

Claim 5. Let s[0,)s\in[0,\infty). If s=0s=0, then s1=1<0=ϕ(s)s-1=-1<0=\phi(s). If ss is positive, then 1s1logs1-s^{-1}\le\log s by Claim 2, and multiplying by ss (claim 5 of Elementary Arithmetic in an Ordered Field) gives s1slogs=ϕ(s)s-1\le s\log s=\phi(s). Finally, Claim 4 with a=0a=0 gives 0ϕ(s)+exp(1)0\le\phi(s)+\exp(-1), that is, exp(1)ϕ(s)-\exp(-1)\le\phi(s).

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Comments

Loading…