Integration, VII, p. 85
Bourbaki marks this one as harder than the rest.
a) Let f be a positive numerical function defined on R, integrable (for the Lebesgue measure μ on R), bounded, and with compact support. Let γ=supt∈Rf(t). For every w∈R, denote by Uf(w) the set of t∈R such that f(t)≥w; set νf(w)=μ∗(Uf(w)). Show that for every α>1,
∫−∞+∞fα(t)dt=∫0γνf(w)αwα−1dw.
b) Let g be a second numerical function satisfying the same conditions as f, and set δ=supt∈Rg(t). Let h be the function defined on R2 by h(u,v)=f(u)+g(v) if f(u)g(v)=0, and h(u,v)=0 otherwise; finally, set k(t)=supu+v=th(u,v), so that k is positive, integrable, bounded, and has compact support. Show that for every α>1,
∫−∞+∞kα(t)dt≥(γ+δ)α(γα1∫−∞+∞fα(t)dt+δα1∫−∞+∞gα(t)dt).
(Observe that for 0<w<1, one has Uk(γw+δw)⊃Uf(γw)+Ug(δw), and use a) and Exer. 23 d).)
c) Let μn be Lebesgue measure on Rn, A and B two integrable subsets of Rn. Show that (Brunn–Minkowski inequality)
((μn)∗(A+B))1/n≥(μn(A))1/n+(μn(B))1/n.
(Reduce to the case that A and B are compact. Then argue by induction on n using Exer. 23 d), the Lebesgue–Fubini theorem, the inequality proved in b), as well as Hölder’s inequality.)