TheoremBase

On a compact subinterval, the set where the difference quotients are unbounded has outer measure at most (total increment)/M for every M, and for rationals u<v the set where the quotients dip below u and rise above v infinitely often is shown to have outer measure s satisfying vs <= us, by two applications of the Vitali covering theorem; hence s=0 and the quotients converge off a null set.

Proof

Notation. For real p<qp<q write J=[p,q]J=[p,q] for the closed interval with these endpoints, ∣J∣=q−p|J|=q-p and, when J⊆IJ\subseteq I, ΔJ=g(q)−g(p)\Delta J=g(q)-g(p); since gg is nondecreasing, ΔJ≥0\Delta J\ge0. By claim 4 of Lebesgue Measure on the Real Line: Existence, Lengths of Intervals and Uniqueness we have λ(J)=∣J∣\lambda(J)=|J|, and JJ is the closed ball with centre (p+q)/2(p+q)/2 and radius (q−p)/2(q-p)/2, so families of such intervals are families of closed balls in the sense of The Vitali Covering Theorem in Rn\mathbb{R}^n, an interval of length ℓ\ell having radius ℓ/2\ell/2.

Step S: increments add up inside an interval. Let J⊆IJ\subseteq I be a closed interval and let K1,…,KM⊆JK_{1},\dots,K_{M}\subseteq J be pairwise disjoint closed intervals. Then ∑j=1MΔKj≤ΔJ\sum_{j=1}^{M}\Delta K_{j}\le\Delta J. Indeed, relabelling the finitely many intervals we may assume their left endpoints increase, writing Kj=[pj,qj]K_{j}=[p_{j},q_{j}] with p1≤⋯≤pMp_{1}\le\dots\le p_{M}; disjointness then forces qj<pj+1q_{j}<p_{j+1} for j<Mj<M, and gg nondecreasing gives g(qj)≤g(pj+1)g(q_{j})\le g(p_{j+1}). Writing J=[p,q]J=[p,q] we have p≤p1p\le p_{1} and qM≤qq_{M}\le q, so

∑j=1M(g(qj)−g(pj))≤∑j=1M(g(qj)−g(pj))+∑j=1M−1(g(pj+1)−g(qj))=g(qM)−g(p1)≤g(q)−g(p),\sum_{j=1}^{M}\bigl(g(q_{j})-g(p_{j})\bigr)\le\sum_{j=1}^{M}\bigl(g(q_{j})-g(p_{j})\bigr)+\sum_{j=1}^{M-1}\bigl(g(p_{j+1})-g(q_{j})\bigr)=g(q_{M})-g(p_{1})\le g(q)-g(p),

all the discarded terms being nonnegative.

Step 0: reduction to a compact subinterval. For every x∈Ix\in I there are rationals a<ba<b with x∈(a,b)x\in(a,b) and [a,b]⊆I[a,b]\subseteq I: since II is open there is a real ρ>0\rho>0 with {t:∣t−x∣<ρ}⊆I\{t:|t-x|<\rho\}\subseteq I, and The Rational Numbers are Dense in the Real Numbers supplies rationals aa and bb with x−ρ<a<x<b<x+ρx-\rho<a<x<b<x+\rho. The set of pairs of rationals is countable by The Integers and the Rational Numbers are Countable and Products and Powers of Countable Sets, so II is the union of a countable family of open intervals (a,b)(a,b) with a,ba,b rational and [a,b]⊆I[a,b]\subseteq I. Since a countable union of null sets is null by the null-set claim, it suffices to prove that (a,b)∖D(a,b)\setminus D is null for each such pair.

Fix such a<ba<b, put J0=[a,b]⊆IJ_{0}=[a,b]\subseteq I and T=ΔJ0=g(b)−g(a)≥0T=\Delta J_{0}=g(b)-g(a)\ge0. All intervals below are contained in (a,b)(a,b), and every subset of (a,b)(a,b) has outer measure at most b−a<∞b-a<\infty by monotonicity and agreement on Borel sets.

Step 1: the points of unbounded difference quotients. For M∈NM\in\mathbb{N} let GMG_{M} be the set of x∈(a,b)x\in(a,b) such that for every real η>0\eta>0 there is a closed interval JJ with x∈J⊆(a,b)x\in J\subseteq(a,b), 0<∣J∣<η0<|J|<\eta and ΔJ>M∣J∣\Delta J>M|J|. We claim λ∗(GM)≤T/M\lambda^{\ast}(G_{M})\le T/M.

The pairs carrying these intervals finely cover GMG_{M} in the sense of the fine covering condition: given x∈GMx\in G_{M} and η>0\eta>0, apply the definition with 2η2\eta in place of η\eta to obtain such a JJ containing xx of radius ∣J∣/2<η|J|/2<\eta. Let ε>0\varepsilon>0. By the Vitali covering theorem there are finitely many pairwise disjoint intervals J1,…,JNJ_{1},\dots,J_{N} of this family with λ∗(GM∖⋃iJi)≤ε\lambda^{\ast}(G_{M}\setminus\bigcup_{i}J_{i})\le\varepsilon. Using subadditivity and monotonicity of λ∗\lambda^{\ast}, then M∣Ji∣<ΔJiM|J_{i}|<\Delta J_{i}, and finally Step S with J=J0J=J_{0},

λ∗(GM)≤ε+∑i=1N∣Ji∣≤ε+1M∑i=1NΔJi≤ε+TM.\lambda^{\ast}(G_{M})\le\varepsilon+\sum_{i=1}^{N}|J_{i}|\le\varepsilon+\frac{1}{M}\sum_{i=1}^{N}\Delta J_{i}\le\varepsilon+\frac{T}{M}.

This holds for every real ε>0\varepsilon>0; were λ∗(GM)>T/M\lambda^{\ast}(G_{M})>T/M, taking ε=(λ∗(GM)−T/M)/2\varepsilon=\bigl(\lambda^{\ast}(G_{M})-T/M\bigr)/2 would contradict it. Hence λ∗(GM)≤T/M\lambda^{\ast}(G_{M})\le T/M, as claimed. Consequently P=⋂M∈NGMP=\bigcap_{M\in\mathbb{N}}G_{M} satisfies λ∗(P)≤T/M\lambda^{\ast}(P)\le T/M for every M∈NM\in\mathbb{N}, by monotonicity; if λ∗(P)\lambda^{\ast}(P) were positive then The Archimedean Property of the Real Numbers would supply M∈NM\in\mathbb{N} with T/M<λ∗(P)T/M<\lambda^{\ast}(P), a contradiction. So λ∗(P)=0\lambda^{\ast}(P)=0 and PP is null by the null-set claim.

Step 2: the points of oscillating difference quotients. Let u,vu,v be rationals with 0≤u<v0\le u<v and let Eu,vE_{u,v} be the set of x∈(a,b)x\in(a,b) such that for every real η>0\eta>0 there are closed intervals J,KJ,K with x∈J⊆(a,b)x\in J\subseteq(a,b), x∈K⊆(a,b)x\in K\subseteq(a,b), 0<∣J∣<η0<|J|<\eta, 0<∣K∣<η0<|K|<\eta,

ΔJ<u ∣J∣andΔK>v ∣K∣.\Delta J<u\,|J|\qquad\text{and}\qquad \Delta K>v\,|K| .

We claim λ∗(Eu,v)=0\lambda^{\ast}(E_{u,v})=0. Put s=λ∗(Eu,v)s=\lambda^{\ast}(E_{u,v}), a real number with 0≤s≤b−a0\le s\le b-a, and suppose 0<s0<s. Let ε\varepsilon be a real number with 0<ε<s/20<\varepsilon<s/2.

By outer regularity of the outer measure there is an open O′O' with Eu,v⊆O′E_{u,v}\subseteq O' and λ(O′)≤s+ε\lambda(O')\le s+\varepsilon; put O=O′∩(a,b)O=O'\cap(a,b), again open, still containing Eu,vE_{u,v}, with λ(O)≤s+ε\lambda(O)\le s+\varepsilon.

First application. Let F1\mathcal{F}_{1} consist of the pairs carrying the closed intervals JJ with J⊆OJ\subseteq O and ΔJ<u∣J∣\Delta J<u|J|. This family finely covers Eu,vE_{u,v}: for x∈Eu,vx\in E_{u,v} and η>0\eta>0, choose a real ρ>0\rho>0 with {t:∣t−x∣<ρ}⊆O\{t:|t-x|<\rho\}\subseteq O and apply the definition of Eu,vE_{u,v} with min⁡{2η,ρ}\min\{2\eta,\rho\} in place of η\eta; the resulting JJ contains xx, has ∣J∣<ρ|J|<\rho and therefore J⊆OJ\subseteq O, and has radius <η<\eta. By the Vitali covering theorem there are pairwise disjoint J1,…,JN∈F1J_{1},\dots,J_{N}\in\mathcal{F}_{1} with λ∗(Eu,v∖⋃iJi)≤ε\lambda^{\ast}(E_{u,v}\setminus\bigcup_{i}J_{i})\le\varepsilon. Being pairwise disjoint Borel subsets of OO, they satisfy ∑i∣Ji∣≤λ(O)≤s+ε\sum_{i}|J_{i}|\le\lambda(O)\le s+\varepsilon, whence

∑i=1NΔJi<u∑i=1N∣Ji∣≤u(s+ε).(1)\sum_{i=1}^{N}\Delta J_{i}<u\sum_{i=1}^{N}|J_{i}|\le u(s+\varepsilon). \tag{1}

Second application. Let A=Eu,v∩⋃i=1N(pi,qi)A=E_{u,v}\cap\bigcup_{i=1}^{N}(p_{i},q_{i}), where Ji=[pi,qi]J_{i}=[p_{i},q_{i}]. The finite set FF of the endpoints pi,qip_{i},q_{i} is null, being a finite union of singletons, each of measure 00 by claim 4 of Lebesgue Measure on the Real Line: Existence, Lengths of Intervals and Uniqueness applied to an interval with equal endpoints, and a countable union of null sets being null by the null-set claim; moreover Eu,v⊆(Eu,v∖⋃iJi)∪A∪FE_{u,v}\subseteq(E_{u,v}\setminus\bigcup_{i}J_{i})\cup A\cup F, so subadditivity gives s≤ε+λ∗(A)s\le\varepsilon+\lambda^{\ast}(A), that is λ∗(A)≥s−ε\lambda^{\ast}(A)\ge s-\varepsilon.

Let F2\mathcal{F}_{2} consist of the pairs carrying the closed intervals KK with K⊆(pi,qi)K\subseteq(p_{i},q_{i}) for some ii and ΔK>v∣K∣\Delta K>v|K|. This family finely covers AA: for x∈Ax\in A pick ii with x∈(pi,qi)x\in(p_{i},q_{i}) and a real ρ>0\rho>0 with {t:∣t−x∣<ρ}⊆(pi,qi)\{t:|t-x|<\rho\}\subseteq(p_{i},q_{i}), then apply the definition of Eu,vE_{u,v} with min⁡{2η,ρ}\min\{2\eta,\rho\} in place of η\eta. By the Vitali covering theorem there are pairwise disjoint K1,…,KM∈F2K_{1},\dots,K_{M}\in\mathcal{F}_{2} with λ∗(A∖⋃jKj)≤ε\lambda^{\ast}(A\setminus\bigcup_{j}K_{j})\le\varepsilon, so subadditivity gives

∑j=1M∣Kj∣≥λ∗(A)−ε≥s−2ε>0.\sum_{j=1}^{M}|K_{j}|\ge\lambda^{\ast}(A)-\varepsilon\ge s-2\varepsilon>0 .

Each KjK_{j} lies in exactly one JiJ_{i}, the JiJ_{i} being pairwise disjoint; grouping the KjK_{j} accordingly and applying Step S inside each JiJ_{i} gives ∑jΔKj≤∑iΔJi\sum_{j}\Delta K_{j}\le\sum_{i}\Delta J_{i}. Combining this with ΔKj>v∣Kj∣\Delta K_{j}>v|K_{j}| and with (1)(1),

v (s−2ε)≤v∑j=1M∣Kj∣<∑j=1MΔKj≤∑i=1NΔJi<u (s+ε).v\,(s-2\varepsilon)\le v\sum_{j=1}^{M}|K_{j}|<\sum_{j=1}^{M}\Delta K_{j}\le\sum_{i=1}^{N}\Delta J_{i}<u\,(s+\varepsilon).

This holds for every real ε\varepsilon with 0<ε<s/20<\varepsilon<s/2. If vs>usvs>us then, vv and uu being fixed, the Archimedean property supplies such an ε\varepsilon with (2v+u)ε<vs−us(2v+u)\varepsilon<vs-us, contradicting the display; hence vs≤usvs\le us, so (v−u)s≤0(v-u)s\le0. As u<vu<v and s>0s>0 this is impossible, and therefore s=0s=0.

Step 3: assembling. Let x∈(a,b)x\in(a,b) belong to none of the sets ⋂MGM\bigcap_{M}G_{M} and Eu,vE_{u,v} (over rationals 0≤u<v0\le u<v). We show x∈Dx\in D.

For h≠0h\ne0 with x+h∈(a,b)x+h\in(a,b) let JhJ^{h} be the closed interval with endpoints xx and x+hx+h and put q(h)=(g(x+h)−g(x))/hq(h)=\bigl(g(x+h)-g(x)\bigr)/h. Then ∣Jh∣=∣h∣|J^{h}|=|h| and ΔJh=∣g(x+h)−g(x)∣\Delta J^{h}=|g(x+h)-g(x)|, because gg is nondecreasing; as g(x+h)−g(x)g(x+h)-g(x) and hh have the same sign, q(h)=ΔJh/∣Jh∣≥0q(h)=\Delta J^{h}/|J^{h}|\ge0.

Since x∉⋂MGMx\notin\bigcap_{M}G_{M} there is M∈NM\in\mathbb{N} with x∉GMx\notin G_{M}, so there is a real η0>0\eta_{0}>0 such that every closed interval JJ with x∈J⊆(a,b)x\in J\subseteq(a,b) and 0<∣J∣<η00<|J|<\eta_{0} has ΔJ≤M∣J∣\Delta J\le M|J|. Applying this to JhJ^{h} gives 0≤q(h)≤M0\le q(h)\le M whenever 0<∣h∣<η00<|h|<\eta_{0} and x+h∈(a,b)x+h\in(a,b). For real η\eta with 0<η≤η00<\eta\le\eta_{0} let

Q(η)={q(h):0<∣h∣<η, x+h∈(a,b)},Q(\eta)=\{q(h):0<|h|<\eta,\ x+h\in(a,b)\},

a nonempty subset of [0,M][0,M]; let m(η)m(\eta) be its greatest lower bound and mˉ(η)\bar{m}(\eta) its least upper bound, which exist by the least upper bound property and by Existence of the Infimum of a Nonempty Subset of R\mathbb{R} Bounded Below. As η\eta decreases Q(η)Q(\eta) shrinks, so mm does not decrease and mˉ\bar m does not increase; both stay in [0,M][0,M]. Let α\alpha be the least upper bound of {m(η):0<η≤η0}\{m(\eta):0<\eta\le\eta_{0}\} and β\beta the greatest lower bound of {mˉ(η):0<η≤η0}\{\bar m(\eta):0<\eta\le\eta_{0}\}; then 0≤α≤β≤M0\le\alpha\le\beta\le M, since m(η)≤mˉ(η)m(\eta)\le\bar m(\eta) for every η\eta and m(η)≤mˉ(η′)m(\eta)\le\bar m(\eta') for all η,η′\eta,\eta' by comparing at min⁡{η,η′}\min\{\eta,\eta'\}.

Suppose α<β\alpha<\beta. By The Rational Numbers are Dense in the Real Numbers choose rationals u,vu,v with α<u<v<β\alpha<u<v<\beta; then 0≤u0\le u. For every η\eta with 0<η≤η00<\eta\le\eta_{0} we have m(η)≤α<um(\eta)\le\alpha<u, so uu is not a lower bound of Q(η)Q(\eta) and there is hh with 0<∣h∣<η0<|h|<\eta and q(h)<uq(h)<u, that is ΔJh<u∣Jh∣\Delta J^{h}<u|J^{h}|; and mˉ(η)≥β>v\bar m(\eta)\ge\beta>v, so there is h′h' with 0<∣h′∣<η0<|h'|<\eta and ΔJh′>v∣Jh′∣\Delta J^{h'}>v|J^{h'}|. Hence x∈Eu,vx\in E_{u,v}, contrary to the choice of xx. Therefore α=β\alpha=\beta; call this common value cc.

Finally, let ε>0\varepsilon>0. There is η1\eta_{1} with m(η1)>c−εm(\eta_{1})>c-\varepsilon and η2\eta_{2} with mˉ(η2)<c+ε\bar m(\eta_{2})<c+\varepsilon, since cc is the least upper bound of the m(η)m(\eta) and the greatest lower bound of the mˉ(η)\bar m(\eta). Put δ=min⁡{η1,η2,x−a,b−x}\delta=\min\{\eta_{1},\eta_{2},x-a,b-x\}, a positive real because a<x<ba<x<b. If 0<∣h∣<δ0<|h|<\delta and x+h∈Ix+h\in I, then ∣h∣<min⁡{x−a,b−x}|h|<\min\{x-a,b-x\} forces x+h∈(a,b)x+h\in(a,b), so q(h)q(h) belongs to Q(η1)Q(\eta_{1}) and to Q(η2)Q(\eta_{2}) and therefore satisfies c−ε<m(η1)≤q(h)≤mˉ(η2)<c+εc-\varepsilon<m(\eta_{1})\le q(h)\le\bar m(\eta_{2})<c+\varepsilon. Thus for every real ε>0\varepsilon>0 there is a real δ>0\delta>0 such that the difference quotient of gg at xx differs from cc by less than ε\varepsilon for all hh with 0<∣h∣<δ0<|h|<\delta and x+h∈Ix+h\in I; that is, gg has a derivative g′(x)=cg'(x)=c at the interior point xx of II, so x∈Dx\in D.

Therefore (a,b)∖D(a,b)\setminus D is contained in the union of ⋂MGM\bigcap_{M}G_{M} and of the sets Eu,vE_{u,v} over the countably many pairs of rationals with 0≤u<v0\le u<v, all of which are null; by the null-set claim it is null. By Step 0, I∖DI\setminus D is null.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…