Each result cited below is universally quantified over the data in its own statement. We work in the setting of The Real Numbers: Standing Notation and Background, which carries the elementary arithmetic, order and absolute-value facts of R used below; natural numbers start at 1, as in Natural Numbers.
Notation. For x∈X and k∈N put Ak(x)={um(x):m∈N, m≥k}. Since ∣uN(x)∣≤C<C+1 for every N, the number C+1>0 witnesses that (uN(x))N∈N is a bounded sequence, and each Ak(x) is nonempty and bounded above by C and below by −C. By the definition of the limit superior and that of the limit inferior,
uˉ(x)=inf{supAk(x):k∈N},u(x)=sup{infAk(x):k∈N}.
Step 0 (Shifting suprema and infima). Let S⊆R be nonempty, let c∈R, and put S+c={s+c:s∈S}. We show: if S is bounded above, then S+c is bounded above and sup(S+c)=supS+c; if S is bounded below, then S+c is bounded below and inf(S+c)=infS+c. Suppose S is bounded above. For s∈S we have s≤supS, hence s+c≤supS+c, so supS+c is an upper bound of S+c. If b is any upper bound of S+c, then s≤b−c for every s∈S, so b−c is an upper bound of S and therefore supS≤b−c, that is, supS+c≤b. Thus supS+c is a least upper bound of S+c in the sense of Upper Bound and Least Upper Bound, and by the uniqueness recorded in The Real Numbers: Standing Notation and Background §bounds it equals sup(S+c). The statement for infima is proved in the same way with all inequalities reversed, using Lower Bound and Greatest Lower Bound in a Totally Ordered Set: infS+c is a lower bound of S+c, and if ℓ is a lower bound of S+c then ℓ−c is a lower bound of S, so ℓ−c≤infS, that is, ℓ≤infS+c.
Consequently, if (bN)N∈N is a bounded sequence of real numbers, say ∣bN∣≤M with M>0, and c∈R, then (bN+c)N∈N is bounded, since ∣bN+c∣≤M+∣c∣ and M+∣c∣>0; and
Nlimsup(bN+c)=NlimsupbN+c,Nliminf(bN+c)=NliminfbN+c.(0.1)
Indeed, writing Bk={bm:m≥k}, the corresponding set for (bN+c) is Bk+c, so by the first part sup(Bk+c)=supBk+c for every k; hence {sup(Bk+c):k∈N}={supBk:k∈N}+c, and the second part gives that its infimum is limsupNbN+c. The identity for the limit inferior follows in the same way, first shifting infBk and then the supremum of {infBk:k∈N}.
Step 1 (Clause 1, bounds). Fix x∈X. From ∣uN(x)∣≤C we get −C≤uN(x)≤C for every N. By claim 2 of Basic Properties of the Limit Inferior and Limit Superior of a Bounded Real Sequence, taken with M=C for the first inequality and with M=−C for the second, we get uˉ(x)≤C and −C≤u(x), and by claim 1 of Basic Properties of the Limit Inferior and Limit Superior of a Bounded Real Sequence we have u(x)≤uˉ(x). Hence −C≤u(x)≤uˉ(x)≤C, and so ∣uˉ(x)∣≤C and ∣u(x)∣≤C.
Step 2 (Clause 2, modulus). Fix x,y∈X. By claim 1 of Metric Space, 0≤d(x,y), so t=ω(d(x,y)) is defined. For every N the hypothesis gives uN(x)−uN(y)≤∣uN(x)−uN(y)∣≤t, that is, uN(x)≤uN(y)+t. The sequence (uN(y)+t)N∈N is bounded by Step 0, so claim 2 of Basic Properties of the Limit Inferior and Limit Superior of a Bounded Real Sequence applies to aN=uN(x) and bN=uN(y)+t, and together with (0.1) it gives
uˉ(x)≤Nlimsup(uN(y)+t)=uˉ(y)+t,u(x)≤Nliminf(uN(y)+t)=u(y)+t.
By claim 3 of Metric Space, d(y,x)=d(x,y), so the same argument with x and y exchanged gives uˉ(y)≤uˉ(x)+t and u(y)≤u(x)+t. Hence ∣uˉ(x)−uˉ(y)∣≤t and ∣u(x)−u(y)∣≤t.
Step 3 (Clause 3, extraction). Fix x∈X and write aN=uN(x), L=uˉ(x) and L′=u(x). For each j∈N the number j1 is positive.
(a) By claim 3 of Basic Properties of the Limit Inferior and Limit Superior of a Bounded Real Sequence, taken with ε=j1, for every j∈N there is Pj∈N with am<L+j1 for every m≥Pj.
(b) Let j,k∈N. Since L is the infimum, and in particular a lower bound, of {supAk′(x):k′∈N}, we have L≤supAk(x), hence L−j1<supAk(x). The set Ak(x) is nonempty and bounded above, so Approximation Property of the Supremum and the Infimum in R §strict-above, applied with S=Ak(x) and b=L−j1, gives an element of Ak(x) exceeding b; that is, there is m∈N with m≥k and L−j1<am.
We now choose N1,N2,… recursively. Let N1 be an index m given by (b) for j=1 and k=P1. Having chosen Nj, let Nj+1 be an index m given by (b) for j+1 in place of j and k=max{Pj+1,Nj+1}. Then Nj+1≥Nj+1>Nj for every j, and Nj≥Pj, so (a) and (b) give L−j1<aNj<L+j1, that is, ∣aNj−L∣<j1 for every j∈N. Given ε>0, claim 3 of The Archimedean Property of the Real Numbers gives J∈N with J1<ε, and then ∣aNj−L∣<j1≤J1<ε for every j≥J. By Limit of a Sequence of Real Numbers, (uNj(x))j∈N converges to uˉ(x).
For the limit inferior we argue symmetrically. (a') By claim 3 of Basic Properties of the Limit Inferior and Limit Superior of a Bounded Real Sequence, for every j there is Pj′∈N with L′−j1<am for every m≥Pj′. (b') For j,k∈N, L′ is the supremum, and in particular an upper bound, of {infAk′(x):k′∈N}, so infAk(x)≤L′<L′+j1; the set Ak(x) is nonempty and bounded below, so Approximation Property of the Supremum and the Infimum in R §strict-below, applied with S=Ak(x) and b=L′+j1, gives m≥k with am<L′+j1. Choosing N1′ by (b') with j=1 and k=P1′, and Nj+1′ by (b') with j+1 and k=max{Pj+1′,Nj′+1}, we obtain N1′<N2′<⋯ with ∣aNj′−L′∣<j1 for every j, and the same Archimedean argument shows that (uNj′(x))j∈N converges to u(x).
Step 4 (Clause 4, uniform convergence). Let K be as in clause 4 and suppose, for a contradiction, that the conclusion fails. Then there is a real ε>0 such that for every N0∈N there exist N∈N with N≥N0 and x∈K with ∣uN(x)−uˉ(x)∣>ε. Fix such an ε; all later choices depend on it.
Choose recursively N1∈N and x1∈K for N0=1, and, having chosen Nj, choose Nj+1 and xj+1∈K for N0=Nj+1. Then N1<N2<⋯, xj∈K, and
∣uNj(xj)−uˉ(xj)∣>εfor every j∈N.(4.1)
Since K is sequentially compact, there are x∗∈K and a strictly increasing sequence (ji)i∈N in N such that (xji)i∈N converges to x∗ in (X,d).
The remaining choices are made in the following order. First, claim 2 of Modulus of Continuity, applied to 3ε>0, gives δ>0 such that ω(t)≤3ε for every t≥0 with t≤δ. Second, by Convergent Sequence in a Metric Space there is i1∈N with d(xji,x∗)<δ for every i≥i1. Third, since x∗∈K we have uˉ(x∗)=u(x∗), so claim 5 of Basic Properties of the Limit Inferior and Limit Superior of a Bounded Real Sequence, taken with a=uˉ(x∗), shows that (uN(x∗))N∈N converges to uˉ(x∗); by Limit of a Sequence of Real Numbers there is P∈N with ∣uN(x∗)−uˉ(x∗)∣<3ε for every N≥P. Finally, put i=max{i1,P}, z=xji and N=Nji.
By Strictly Increasing Sequences of Natural Numbers Dominate Their Index, applied to the strictly increasing sequences (Nj)j∈N and (ji)i∈N, we have N=Nji≥ji≥i≥P. Since i≥i1, d(z,x∗)<δ, and d(x∗,z)=d(z,x∗) by claim 3 of Metric Space; both are nonnegative by claim 1 of Metric Space. Hence the hypothesis on (uN) gives ∣uN(z)−uN(x∗)∣≤ω(d(z,x∗))≤3ε; the choice of P gives ∣uN(x∗)−uˉ(x∗)∣<3ε; and Step 2 gives ∣uˉ(x∗)−uˉ(z)∣≤ω(d(x∗,z))≤3ε. By the triangle inequality for the absolute value,
∣uN(z)−uˉ(z)∣≤∣uN(z)−uN(x∗)∣+∣uN(x∗)−uˉ(x∗)∣+∣uˉ(x∗)−uˉ(z)∣<ε.
But z=xji and N=Nji, so (4.1) with j=ji gives ∣uN(z)−uˉ(z)∣>ε, a contradiction. Hence for every ε>0 there is N0∈N with ∣uN(x)−uˉ(x)∣≤ε for every N≥N0 and every x∈K.