Throughout, λ=λ[0,T] and B=B[0,T] are as in the definition of the Lebesgue space, with λ([0,T])=T by the interval toolkit, and the norm of a class does not depend on the representative used to compute it, by The Lebesgue Space of Square-Integrable Vector-Valued Functions is a Real Inner Product Space. Fix a0∈A, which is possible since A is nonempty.
A normalized representative. We first record: every ξ∈UA has a representative taking all its values in A. Indeed, let u∈L2([0,T];Rm) represent ξ and let N∈B with λ(N)=0 be such that u(t)∈A for t∈[0,T]∖N. Define u~(t)=u(t) for t∈/N and u~(t)=a0 for t∈N; each component u~i is measurable: for a set B in the Borel σ-algebra of R,
(u~i)−1(B)=((ui)−1(B)∖N)∪NB,
where NB=N if a0i∈B and NB=∅ otherwise; here (ui)−1(B)∈B because ui is measurable, N∈B, and B is closed under differences and finite unions by the definition of a σ-algebra. Moreover u~=u off the null set N, so u~∼u, hence u~ is square-integrable and [u~]=ξ; and u~(t)∈A for every t.
Claim 1. Let a∈A and let u be the constant map with value a. Each component ui is constant, hence a simple function on ([0,T],B) and in particular measurable, and ∣u(t)∣2=∣a∣2 for every t, so ∫[0,T]∣u∣2dλ=∣a∣2T<∞. Thus u∈L2([0,T];Rm), and u(t)∈A for every t, so [u]∈UA (take N=∅).
Claim 2. Let ξ∈UA and let u~ be a representative with all values in A. Then ∣u~(t)∣2≤R2 for every t, so by monotonicity of the integral (claim 1 of Linearity and Monotonicity of the Lebesgue Integral)
∥ξ∥L22=∫[0,T]∣u~∣2dλ≤R2T.
Both ∥ξ∥L2 and RT1/2 are nonnegative and their squares satisfy ∥ξ∥L22≤(RT1/2)2, so ∥ξ∥L2≤RT1/2: otherwise RT1/2<∥ξ∥L2, and multiplying this inequality by the two nonnegative numbers in turn would give (RT1/2)2<∥ξ∥L22.
Claim 3. Let ξ,ξ′∈UA and let s be real with 0≤s≤1. Choose representatives u~,u~′ with all values in A. By the operations of The Lebesgue Space of Square-Integrable Vector-Valued Functions on a Compact Interval, su~+(1−s)u~′ is a representative of sξ+(1−s)ξ′, and it is square-integrable because L2([0,T];Rm) is a real vector space with these operations by The Lebesgue Space of Square-Integrable Vector-Valued Functions is a Real Inner Product Space. By the hypothesis of this claim, su~(t)+(1−s)u~′(t)∈A for every t. Hence sξ+(1−s)ξ′∈UA, which is convexity.
Claim 4. The metric open subsets of H determined by dL2 form a topology by Metric Open Sets Form a Topology, and likewise for Rm with the Euclidean distance dE. Let ξ belong to the closure of UA in (H,dL2).
For n∈N let
Bn={v∈L2([0,T];Rm):v(t)∈A for every t∈[0,T] and dL2(ξ,[v])<ι(n)−1},
where ι is the canonical map of R, so that ι(n)−1>0 by claim 3 of Properties of the Canonical Map from the Natural Numbers to an Ordered Field. By Characterization of the Closure in a Metric Space by Open Balls there is an element of UA within distance ι(n)−1 of ξ, and by the normalized-representative paragraph it has a representative with all values in A; hence Bn=∅. The sets Bn are subsets of L2([0,T];Rm), so Axiom of Countable Choice provides a sequence (un)n∈N with un∈Bn for every n.
Then 0≤∥[un]−ξ∥L2<ι(n)−1, and the real sequence (ι(n)−1)n has limit 0 by claim 3 of The Archimedean Property of the Real Numbers; comparing with the constant sequence 0 and using Order Properties of Limits of Real Sequences, the real sequence (∥[un]−ξ∥L2)n has limit 0.
Let u be a representative of ξ. By claim 2 of Completeness of the Lebesgue Space of Square-Integrable Vector-Valued Functions there are natural numbers n1<n2<… and a set N∈B with λ(N)=0 such that (unj(t))j converges to u(t) in (Rm,dE) for every t∈[0,T]∖N. Fix such a t and let ε>0 be real; there is j with dE(u(t),unj(t))<ε, and unj(t)∈A. By Characterization of the Closure in a Metric Space by Open Balls, u(t) lies in the closure of A in (Rm,dE); since A is closed it equals its closure, so u(t)∈A.
Thus u(t)∈A for every t∈[0,T]∖N with λ(N)=0, so ξ=[u]∈UA. Therefore the closure of UA is contained in UA; as it also contains UA and is closed, UA equals its closure and is closed.