Solution to the Exercises of Lecture 10
University of Salerno February 20, 2015
Exercise 10.1: For this exercise letS be the strip S :={z∈C;−1/2<Rez <1/2}.
Leth:S7→Cbe continuous and holomorphic onS. Assume thathis bounded on∂S, and that there existsα < πsuch that
|h(z)| ≤ceeα|Imz| (z∈S).
Show thathis bounded bykhk∂S.
Solution: Forn∈N, the functionψn(z) :=e−n1(eiβz+e−iβz), withα < β < π, is continuous onSand holomorphic onS. By the maximum principle one obtains
kψnhkS
k ≤ kψnhk∂Sk,
whereSk :={z∈S:|Imz|< k}, k∈N. Observe thatkψnhk{z∈∂Sk:|Imz|=k}→ 0 (k→ ∞). In fact,
|ψnh(z)| ≤c
e−n1(eiβz+e−iβz) eeα|Imz|
=c e−n1(e−βImz+eβImz) cos(β|Rez|)eeα|Imz|
≤c e−cos(β/2)n (e−βImz+eβImz)eeα|Imz|
Hence,
kψnhk{z∈∂Sk:|Imz|=k}≤c e−cos(β/2)n (e−βk+eβk)eeαk
=c e−cos(β/2)n e−βkeeβk(−cos(β/2)n +e(α−β)k)
For big k, we have that e(α−β)k < ε. Thus, choosing ε < cos(β/2)2n , so that e(α−β)k−cos(β/2)n <−ε, we have that, in the limit ask → ∞, the right-hand side of the inequality tends to 0.
Therefore,
kψnhkS ≤ kψnhk∂S. Lettingn→ ∞we obtain the assertion.
Exercise 10.2: Let (Ω, µ) be a measure space. Show that
k1
2(f +g)kpp+k1
2(f−g)kpp 1p
≤2−p1 kfkpp+kgkpp1p
for allf, g∈Lp(µ), 2≤p≤ ∞.
Solution: For 2≤p≤ ∞consider
Tp:Lp(µ)×Lp(µ)−→Lp(µ)×Lp(µ) (f, g) 7→
1
2(f+g),1 2(f −g)
.
We endowLp(µ)×Lp(µ),2≤p≤ ∞, with the norms
k(f, g)kp×p = (kfkpp+kgkpp)1p, f, g∈Lp(µ),2≤p <∞, k(f, g)k∞ = max(kfk∞,kgk∞), f, g∈L∞(µ).
We first estimate the norm of the operatorT onL2(µ)×L2(µ) =L2(µ⊗µ) and L∞(µ)×L∞(µ) =L∞(µ⊗µ). For p= 2 one has
kT(f, g)k2×2 = 1
2kf+g, f−gk2×2
= 1
2 kf+gk22+kf −gk2212
= 1
2 kfk2+ 2(f|g) +kgk2+kfk2−2(f|g) +kgk212
= 1
√2(kfk22+kgk22)12
= 2−12k(f, g)k2×2. And forp=∞, one obtains
kT(f, g)k∞ = 1
2kf+g, f−gk∞
= 1
2max (kf+gk∞,kf−gk∞)
≤ 1
2max (kfk∞+kgk∞,kfk∞+kgk∞)
= 1
2(kfk∞+kgk∞)
≤ max(kfk∞,kgk∞)
= k(f, g)k∞.
Letp∈(2,+∞),p0=∞,p1= 2,M0 = 1,M1= 2−12 andτ= 2p ∈(0,1), then by Riesz-Thorin Theorem, see Corollary 10.6, one has
kT(f, g)kp≤Mτk(f, g)kp, f, g∈S(Ac) (1) withMτ=M01−τM1τ = 2−1p and
Ac={A⊂Ω, measurable:µ(A)<∞}.
By density ofS(Ac) onLp(µ), (1) holds for allf, g∈Lp(µ). Therefore,
k1
2(f+g)kpp+k1
2(f−g)kpp p1
≤2−1p kfkpp+kgkpp1p
, f, g∈Lp(µ).
Exercise 10.3: (a) Letp∈(1,∞),r∈[0,∞). Show that r= inf
α∈Q∩(0,∞)
1
pα1−prp+ (1−1 p)α
. (2)
(b) Let (Ω, µ) be a measure space, and letS∈ L(L2(µ)) be sub-Markovian and substochastic. Show thatS isLp-contractive for allp∈(1,∞).
(c) Let (Ω, µ) be a measure space, and let S ∈ L(L2(µ)) be sub-Markovian, and assume that there existsc >0 such that 1cS is substochastic. Show thatS interpolates to an operatorSp∈ L(Lp(µ)) withkSpk ≤c1p, for 1< p <∞.
Solution: (a) Letp0be the conjugate ofp, i.e. p1+p10 = 1 and letα∈Q∩(0,∞).
Applying Young’s inequality
ab≤ 1 pap+ 1
p0bp0
witha=α1−pp randb=αp−1p , one obtains r=ab ≤ 1
p(α1−pp r)p+
1−1 p
αp−1p p−1p
≤ 1
pα1−prp+
1−1 p
α.
Passing to the infinimum, one gets r≤ inf
α∈Q∩(0,∞)
1
pα1−prp+ (1−1 p)α
.
On the other hand, consider αn = E(nr)
n ∈ Qfor n ≥1, where E(nr) is the integer part ofnr. Since lim
n−→∞αn=r, it follows that
n−→∞lim 1
pα1−pn rp+
1−1 p
αn
=r.
Finally,
r= inf
α∈Q∩(0,∞)
1
pα1−prp+ (1−1 p)α
.
(b) Letu∈L2(µ)∩Lp(µ) andAbe a measurable subset of Ω with finite measure.
Then, by (2) (withr=|u(x)|χA(x)) , it follows that
|u|χA≤1
pα1−p|u|pχA+ (1−1 p)αχA
forα∈Q∩(0,∞). So, by the positivity and theL∞-contractivity ofS, one has S(|u|χA) ≤ 1
pα1−pS(|u|pχA) + (1−1 p)αSχA
≤ 1
pα1−pS(|u|pχA) + (1−1 p)α.
So, passing to the infinimum over α∈Q∩(0,∞), one obtains S(|u|χA)≤(S(|u|pχA))p1.
Let now (An)n∈Nbe an increasing sequence of measurable sets with finite mea- sure such that ∪n∈NAn = Ω. Then, |u|χAn −→ |u| and |u|pχAn −→ |u|p as n −→ ∞. Since S ∈ L(L2(µ)) and S can be extended to a bounded linear operator onL1(µ) (using the fact thatS is substochastic), it follows that
n−→∞lim S(|u|χAn) =S|u|and lim
n−→∞S(|u|pχAn) =S|u|p. Finally one has
S(|u|)≤(S(|u|p))1p, ∀u∈L2(µ)∩Lp(µ). (3) Therefore, by using theL1-contractivity ofSand the fact that|Su| ≤S|u|, one has
Z
Ω
|Su|pdµ≤ Z
Ω
(S|u|)pdµ≤ Z
Ω
S|u|pdµ≤ Z
Ω
|u|pdµ.
Thus,
kSukpp≤ kukpp, ∀u∈L2(µ)∩Lp(µ).
By the density ofL2(µ)∩Lp(µ) inLp(µ), the above estimate shows thatScan be extended to a contractive operator overLp(µ).
(b) Letu∈L2(µ)∩Lp(µ), then from (3) one has (S(|u|))p≤S(|u|p). Thus, Z
Ω
|Su|pdµ≤ Z
Ω
(S|u|)pdµ≤ Z
Ω
S|u|pdµ≤c Z
Ω
|u|pdµ.
Therefore,
kSukpp≤ckukpp, ∀u∈L2(µ)∩Lp(µ).
By the density of L2(µ)∩Lp(µ) inLp(µ), it follows thatS interpolates to an operatorSp∈ L(Lp(µ)) andkSpk ≤c1/p for all 1< p <∞.
Exercise 10.4: Let the hypotheses be as in Exercise 9.5, and additionally b∈C1(Ω). Assume thatω∈Ris such that div b(x)≤ω for allx∈Ω.
(a) Show that kT(t)uk1 ≤eωtkuk1 for u∈L2(Ω)∩L1(Ω) and t ≥0, where T(·) is theC0-semigroup generated by the operator −A.
(b) Compute estimates for kTp(t)k in terms of ω := sup divb fort ≥0, 1≤ p <∞, whereTp(·) is the interpolated semigroup onLp(Ω), analogous to Theorem 10.15(b).
Solution:
(a) In order to show that kT(t)uk1 ≤ eωtkuk1, we show that e−ωtT(t) is substochastic. To do so we use the characterization in Theorem 10.12.(d).
Hence we have to show thataω((u−1)+, u∧1)≥0, whereaωis the form associated withA+ωI.
Let us considerV =H01(Ω) and u∈Cc1(Ω). Observe that if u∈Cc1(Ω) thenu∧1 ∈ V, cf. [1, Proposition 4.11, p.113], and also (u−1)+ ∈ V, sinceu−u∧1 = (u−1)+.
Integrating by parts and using the fact that divb(x)≤ω, we have aω((u−1)+, u∧1) =
Z
Ω
∇(u−1)+· ∇(u∧1)dx+ Z
Ω
b· ∇(u−1)+(u∧1)dx +
Z
Ω
ω(u−1)+(u∧1)dx
≥ Z
Ω
(u−1)+(ω−div b)(u∧1)dx− Z
Ω
(u−1)+b· ∇(u∧1)dx
= Z
Ω
(u−1)+(ω−div b)(u∧1)dx≥0.
(b) We know that kT(t)k∞ ≤ 1 (Exercise 9.5(b)) and kT(t)k1 ≤ eωt with ω= sup div(b). Applying Exercise 10.3.(c) we have
kT(t)kp≤eωtp, t≥0.
References
[1] E.M. Ouhabaz, Analysis of Heat Equations on Domains, London Math.
Soc. Monographs, Vol. 31. Princeton Univ. Press 2004.