Integration, V, p. 121
Denote by λ the Lebesgue measure on Rn.
a) Let C0 be a cube (GT, VI, §1, No. 1) in Rn, and let u be a numerical function on C0 that is integrable (for λ). Let s be a real number such that
s≥λ(C0)1∫C0∣u∣dλ.
Show that there exists a sequence (Ik) of open cubes contained in C0, pairwise disjoint, and satisfying the following conditions:
1∘ ∣u(x)∣≤s almost everywhere in the set C0−⋃kIk;
2∘ if one sets
uk=λ(Ik)1∫Ik∣u(x)∣dλ(x),
then uk≤2ns for all k;
3∘ ∑kλ(Ik)≤s1∫C0∣u(x)∣dλ(x).
(Dividing into two equal intervals each of the projections of C0, among the 2n pairwise disjoint open cubes choose the cubes I1i such that the corresponding 'mean values' u1i are all ≥s, and, because of the choice of s, show that u1i≤2ns. Similarly decompose into 2n cubes each of the cubes of the first subdivision that does not figure among the I1i, which yields a second finite family of cubes I2i chosen among all of the new cubes obtained as those for which the corresponding mean value u2i is ≤s. Continue by recursion. To prove that the condition 1∘ is verified, argue by contradiction, noting that the subset Bm of C0−⋃kIk where ∣u(x)∣≥s+1/m is, up to a negligible set, the intersection of a decreasing sequence (Ur) of open sets, where Ur is the union of the cubes of the r-th subdivision that do not figure among the Ik and that contain a point of Bm.)
b) Let F be the set of integrable numerical functions u on C0 having the following property: for every cube C⊂C0, if mC(u) is the mean value
λ(C)1∫Cu(x)dλ(x),
then ∫C∣u(x)−mC(u)∣dλ(x)≤λ(C); if u∈F, the same is true of u−c for every real number c. For every function u∈F, every number σ>0 and every cube C⊂C0, denote by S(σ,u,C) the set of x∈C such that ∣u(x)−mC(u)∣≥σ. Finally, denote by G(σ) the smallest number such that
λ(S(σ,u,C))≤G(σ)∫C∣u(x)−mC(u)∣dλ(x)
as u runs over F, and C over the set of cubes contained in C0; one has G(σ)≤1/σ.
Show that if σ/2n>s≥1 then
G(σ)≤s1G(σ−2ns).
(One can restrict oneself to proving that, for a u∈F, one has
λ(S(σ,u,C0))≤s1G(σ−2ns)∫C0∣u(x)−mC0(u)∣dλ(x),
and one can suppose, to simplify, that mC0(u)=0. Make use then of the decomposition of C0 into a union of cubes Ik and a set where ∣u(x)∣≤s almost everywhere, on noting that if x∈Ik belongs to S(σ,u,C0), then ∣u(x)−mIk(u)∣≥σ−2ns and consequently
λ(Ik∩S(σ,u,C0))≤G(σ−2ns)∫Ik∣u(x)−mIk(u)∣dλ(x).
c) Deduce from b) that there exist two constants B,b depending only on n, such that, for every function u∈F and every number σ>0, one has λ(S(σ,u,C0))≤Be−bσλ(C0). (Take s=e in b).)