Bourbaki

Exercise 18, § 6

Integration, IV, p. 139

Bourbaki marks this one as harder than the rest.

Let XX be a locally compact space, μ\mu a positive measure on XX, and ff a μ\mu-measurable numerical function.

a) Let (an)n∈Z(a_n)_{n \in \mathbf{Z}} be a sequence of real numbers, λ\lambda a finite real number; for every n∈Zn \in \mathbf{Z}, set

un(x)=an+⌊λlog⁡∣f(x)∣⌋f(x)if f(x)≠0 and f(x)≠±∞un(x)=0for the other values of x∈X u_n(x) = a_{n+\lfloor \lambda \log |f(x)| \rfloor} f(x) \quad \text{if } f(x) \neq 0 \text{ and } f(x) \neq \pm \infty \\ u_n(x) = 0 \quad \text{for the other values of } x \in X
(where [t][t] denotes the integral part of the finite real number tt). Show that the unu_n are μ\mu-measurable and that, for every finite real number cc such that cλ<1c\lambda < 1,
∑n=−∞+∞∫∗∣un(x)∣ecxdμ(x)≤e∣c∣∫∗∣f(x)∣1−cλdμ(x)⋅(∑n=−∞+∞ecn∣an∣). \sum_{n=-\infty}^{+\infty} \int^* |u_n(x)| e^{cx} d\mu(x) \leq e^{|c|} \int^* |f(x)|^{1-c\lambda} d\mu(x) \cdot \left( \sum_{n=-\infty}^{+\infty} e^{cn}|a_n| \right).

b) Let p′,p′′p', p'' be two finite numbers >0> 0, tt a real number such that 0<t<10 < t < 1, and let pp be the real number defined by 1/p=(1−t)/p′+t/p′′1/p = (1-t)/p' + t/p''. For every number α>0\alpha > 0, set

1/Kα=inf⁡(∑n=−∞+∞∣eαtnan∣p′)(1−t)/p′(∑n=−∞+∞∣e−α(1−t)nan∣p′′)t/p′′, 1/K_\alpha = \inf \left( \sum_{n=-\infty}^{+\infty} |e^{\alpha tn} a_n|^{p'} \right)^{(1-t)/p'} \left( \sum_{n=-\infty}^{+\infty} |e^{-\alpha(1-t)n} a_n|^{p''} \right)^{t/p''},
the infimum being taken over all the absolutely convergent series (an)n∈Z(a_n)_{n \in \mathbf{Z}} of finite real numbers such that ∑n=−∞+∞an=1\sum_{n=-\infty}^{+\infty} a_n = 1. Show that ()
Np(f)≤Kα⋅inf⁡F(u) N_p(f) \leq K_\alpha \cdot \inf F(u)
(
*)
Np(f)≥Kαe−α⋅inf⁡F(u), N_p(f) \geq K_\alpha e^{-\alpha} \cdot \inf F(u),
where u=(un)n∈Zu = (u_n)_{n \in \mathbf{Z}} runs over the set of sequence of functions belonging to Lp′∩Lp′′\mathscr{L}^{p'} \cap \mathscr{L}^{p''} such that the series (un(x))n∈Z(u_n(x))_{n \in \mathbf{Z}} is almost everywhere absolutely convergent with sum f(x)f(x); for every sequence uu having these properties, one sets
F(u)=(∑n=−∞+∞(Np′(eαtnun))p′)(1−t)/p′(∑n=−∞+∞(Np′′(e−α(1−t)nun))p′′)t/p′′ F(u) = \left( \sum_{n=-\infty}^{+\infty} (N_{p'}(e^{\alpha tn} u_n))^{p'} \right)^{(1-t)/p'} \left( \sum_{n=-\infty}^{+\infty} (N_{p''}(e^{-\alpha(1-t)n} u_n))^{p''} \right)^{t/p''}
and in the formulas () and (*) the infimum is taken over the set of sequences u=(un)u = (u_n) having the preceding properties. (To prove (), use Hölder’s inequality; to prove (*), use a) twice with suitable choices of cc and λ\lambda.) In particular, KαK_\alpha is finite for all α>0\alpha > 0.

c) Let p′,p′′,q′,q′′p', p'', q', q'' be finite numbers ≥1\geq 1, tt a number such that 0<t<10 < t < 1, and let

1p=1−tp′+tp′′,1q=1−tq′+tq′′. \frac{1}{p} = \frac{1-t}{p'} + \frac{t}{p''}, \quad \frac{1}{q} = \frac{1-t}{q'} + \frac{t}{q''}.
Let YY be a second locally compact space, ν\nu a positive measure on YY, and let ww be a linear mapping of H(X;R)\mathscr{H}(X; \mathbf{R}) into the vector space (not topological) S(Y,ν;R)\mathscr{S}(Y, \nu; \mathbf{R}) of ν\nu-measurable finite numerical functions on YY. Suppose that: 1∘1^\circ ww maps H(X;R)\mathscr{H}(X; \mathbf{R}) into Lq′(Y;R)∩Lq′′(Y;R)\mathscr{L}^{q'}(Y; \mathbf{R}) \cap \mathscr{L}^{q''}(Y; \mathbf{R}); 2∘2^\circ for every function f∈H(X;R)f \in \mathscr{H}(X; \mathbf{R}),
Nq′(w(f))≤M′Np′(f)andNq′′(w(f))≤M′′Np′′(f). N_{q'}(w(f)) \leq M' N_{p'}(f) \quad \text{and} \quad N_{q''}(w(f)) \leq M'' N_{p''}(f).
From this, conclude that ww also maps H(X;R)\mathscr{H}(X; \mathbf{R}) into Lq(Y;R)\mathscr{L}^q(Y; \mathbf{R}) and that
Nq(w(f))≤M⋅Np(f) N_q(w(f)) \leq M \cdot N_p(f)
with
M≤M′1−tM′′t M \leq {M'}^{1-t} {M''}^t
(M. Riesz’s inequality). (Write ff in the form ∑nun\sum_n u_n as in b), and consider the series with general term w(un)w(u_n); use the inequalities () and (*) of b).)

Solution

No solution has been written yet.

Transcribed from the printed volume. Not translated.