QUANTITATIVE ESTIMATES FOR BOUNDED

HOLOMORPHIC SEMIGROUPS
arXiv:2208.14198v3 [math.FA] 8 Feb 2024

TUOMAS HYTÖNEN AND STEFANOS LAPPAS

Abstract. In this paper we revisit the theory of one-parameter semi-

groups of linear operators on Banach spaces in order to prove quantita-
tive bounds for bounded holomorphic semigroups. Subsequently, relying
on these bounds we obtain new quantitative versions of two recent re-
sults of Xu related to the vector-valued Littlewood–Paley–Stein theory
for symmetric diffusion semigroups.

1. Introduction
The theory of Banach valued martingales involving the notions of mar-
tingale type and cotype was introduced and studied in depth by Pisier (we
refer to [10, 11, 12, 13] and Section 4 for more information). His renorming
theorem states that these are geometric properties of the underlying Banach
space, characterized by the existence of an equivalent norm in the space
which is uniformly convex of power type q (for the precise statement of this
result see Theorem 4.7 in Section 4).
On the other hand, first Stein in [14, Chapter IV] proved the follow-
ing result which extends the classical inequality on the Littlewood–Paley
g-function: for every 1 < p < ∞

ˆ ∞ 2 1/2
∂ dt
(1.1) t Tt f ≈ kf kLp (Ω) , ∀f ∈ Lp (Ω),
0 ∂t t Lp (Ω)

where by {Tt }t>0 we denote a symmetric diffusion semigroup (see Definition

4.1 in Section 4) and the equivalence constants depend only on p.
The aforementioned theory about martingale inequalities is closely related
to the vector-valued Littlewood–Paley–Stein theory which was developed in
[8, 15, 16]. In this direction, a Littlewood–Paley theory was first developed
in [15] for functions with values in uniformly convex Banach spaces. In
particular, Xu obtained in [15] the one-sided vector-valued extension of (1.1)
for the classical Poisson semigroup on the torus T.

2020 Mathematics Subject Classification. 47D03 (Primary); 42B25, 46B20.

Key words and phrases. Bounded holomorphic semigroups, Symmetric diffusion semi-
groups, Uniform convexity, Martingale cotype, Littlewood–Paley–Stein theory.
1
2 T. HYTÖNEN AND S. LAPPAS

Martínez–Torrea–Xu [8] characterized, in the vector-valued setting, the va-

lidity of the following one-sided inequality concerning the generalized Little-
wood–Paley g-function associated with a subordinated Poisson symmetric
diffusion semigroup {Pt }t>0 by the martingale type and cotype properties of
the underlying Banach space: for given Banach space X and 1 < q < ∞, X
is of martingale cotype q iff for every 1 < p < ∞
ˆ ∞ q 
∂ dt 1/q
(1.2) t Pt f . kf kLp (Ω;X) , ∀f ∈ Lp (Ω; X).
0 ∂t X t Lp (Ω)

The converse inequality of (1.2) was also treated in [8] iff X is of martingale
type q. Here the subordinated Poisson semigroup {Pt }t>0 is defined by
ˆ ∞ −s
1 e
Pt f = √ √ T t2 f ds,
π 0 s 4s
where {Tt }t>0 is a symmetric diffusion semigroup. This {Pt }t>0 is again a
symmetric diffusion semigroup. Recall that√if A denotes the negative infini-
tesimal generator of {Tt }t>0 , then Pt = e− At .
The question whether (1.2) is also true for a general (not necessarily sub-
ordinated) diffusion semigroup {Tt }t>0 in place of {Pt }t>0 , or even just, in
the special case of the heat semigroup on Rn , was left open in [8] (see [8,
Problem 2 on p. 447]).
This question concerning the heat semigroup on Rn was answered in the
affirmative by Naor and one of us [6], yielding good dependence of various
bounds on the target Banach space X, but also with dependence on the
dimension n of the domain Rn . This was further elaborated by Xu [16],
who extended the previous result to general diffusion semigroups {Tt }t>0
and replaced a concrete bound for the derivative of the heat kernel used in
[6] by an abstract analyticity result from [11]. While giving a form of the
Littlewood–Paley–Stein inequality

ˆ ∞ q 1/q
∂ dt
(1.3) t et∆ f . kf kLq (Rn ;X) ,
0 ∂t Lq (Rn ;X) t
entirely free of the domain’s dimension n, this abstract argument relies on
implicit unquantified bounds, and obscures the dependence of the bound on
the target Banach space X, which was completely explicit in [6].
The main result of this work is Theorem 5.1 in Section 5 which provides a
quantitative version of the main result of [16, Theorem 2 (i)], including ex-
plicit dependence on the target Banach space X and quantitative dimension
free version of (1.3) not only for the heat semigroup on Rn but also for the
general diffusion semigroup {Tt }t>0 (see also [17, 18] for some related work).
The paper is organized as follows: in Sections 2, 3 and 4 we present the
tools for proving our main result in Section 5. To be more precise, we provide
quantitative versions of several classical results in the theory of bounded
holomorphic semigroups, in particular their characterization due to Kato
 BOUNDED HOLOMORPHIC SEMIGROUPS 3

[7]. Having these at our disposal, in Section 4 we quantify the dependence

of the analyticity bound of [11, Remark 1.8(b)] (see also Corollary 4.16)
for extensions of diffusion semigroups {Tt }t>0 to uniformly convex spaces.
As discussed above, this has direct implications to bounds in Littlewood–
Paley–Stein inequalities (1.3) by [6, 16]. In addition, we provide an explicit
dependence of supt>0 kt∂Tt k on the martingale cotype constant mq,X (see
Corollary 4.17), which was the case for the related estimates in [6]. Section
5 is devoted to the proof of Theorem 5.1. Finally, in Section 6 using Corollary
4.17 we obtain a quantitative version of one more result of Xu [17, Theorem
1.4] which we compare with Theorem 5.1.
Notation. Throughout the paper, given a, b ∈ (0, ∞), the notation a . b
means that a ≤ cb for some universal constant c ∈ (0, ∞). The notation
a ≈ b stands for (a . b) ∧ (b . a). If we need to allow for dependence
on parameters, we indicate this by subscripts. Moreover, given a Banach
space X, we denote by Lp (Ω; X) the usual Lp space of strongly measurable
functions from Ω to X and we will use the abbreviation ∂ = ∂/∂t.

2. Preliminaries
We begin by recalling several basic definitions and results from the semi-
group theory:
2.1. Definition (Strongly Continuous Semigroup). Let X be a Banach space.
A family (T (t))t≥0 of bounded linear operators on X is called a strongly
continuous (one-parameter semigroup) (or C0 -semigroup) if
(i) T (0) = I,
(ii) T (t + s) = T (t)T (s) t, s ≥ 0,
(iii) limt↓0 T (t)x = x ∀x ∈ X.
2.2. Remark. A consequence of Definition 2.1 and the uniform boundedness
principle (see also [3, Proposition 5.5]) is the following exponential bounded-
ness of a strongly continuous semigroup (T (t)): there exist constants M ≥ 1
and ω ∈ R such that
kT (t)k ≤ M eωt
for all t ≥ 0.
2.3. Definition (Infinitesimal Generator). Let (T (t)) be a C0 -semigroup.
The infinitesimal generator of (T (t)) is the linear operator defined by
 
T (h)x − x
D(A) := x ∈ X : lim exists ,
h↓0 h
T (h)x − x
Ax := lim .
h↓0 h
2.4. Definition. Let A : D(A) ⊂ X → X be a a closed, linear operator on
some Banach space X. We call
ρ(A) := {λ ∈ C : λ − A : D(A) → X is bijective}
4 T. HYTÖNEN AND S. LAPPAS

the resolvent set and its complement σ(A) := C \ ρ(A) the spectrum of A.
For λ ∈ ρ(A), the inverse
R(λ, A) := (λ − A)−1
is, by the closed graph theorem, a bounded operator on X and will be called
the resolvent of A at the point λ.
2.5. Theorem ([3], Theorem II.3.8 or [9], Theorem I.5.3 and Remark I.5.4).
(Hille–Yosida Generation Theorem) Let A be a linear operator on X and let
M ≥ 1, ω ∈ R. The following conditions are equivalent:
(a) A generates a C0 -semigroup satisfying
kT (t)k ≤ M eωt for t ≥ 0.
(b) A is closed and densely defined, the resolvent set ρ(A) contains the half-
plane {λ ∈ C : Re λ > ω}, and
M
kR(λ, A)n k ≤ for all n ∈ N.
(Re λ − ω)n
2.6. Definition (Open/Closed Sector). Let α ∈ [0, π). We call
Σα := {z ∈ C \ {0} : | arg(z)| < α},
Σα := {z ∈ C : | arg(z)| ≤ α}
the open/closed sector of angle α in the complex plane; the argument is taken
in (−π, π).
2.7. Definition (Holomorphic Semigroup). A family of bounded linear op-
erators (T (z))z∈Σδ ∪{0} is called a holomorphic semigroup (of angle δ) if
(i) T (z1 + z2 ) = T (z1 )T (z2 ) for all z1 , z2 ∈ Σδ .
(ii) The map z 7→ T (z) is holomorphic in Σδ .
(iii) lim T (z)x = x for all x ∈ X and 0 < δ′ < δ.
z→0
z∈Σδ′
If, in addition,
(iv) kT (z)k is bounded in Σδ′ for every 0 < δ′ < δ,
we call (T (z))z∈Σδ ∪{0} a bounded holomorphic semigroup.
2.8. Theorem ([3], IV.1.2). (Resolvent equation) Let A be a closed linear
operator. For λ, µ ∈ ρ(A) one has
R(λ, A) − R(µ, A) = (µ − λ)R(λ, A)R(µ, A).

3. Classical semigroup theory made quantitative

In this section we collect the additional results from semigroup theory
that we will use in Section 4. Most of these results are in principle known,
but difficult to find in the precise quantitative form that we need. For this
reason, we also revisit most of the proofs in order to track the dependence
on the various constants. Besides our application in Section 4, we hope that
 BOUNDED HOLOMORPHIC SEMIGROUPS 5

recording these quantitative formulations might have an independent inter-

est and possible other applications elsewhere. In addition to the standard
references [3, 9], we have benefited from the nice presentation of the classical
theory in [4].
3.1. Lemma ([3], Theorem 4.6; [9], Chapter 2, Theorem 5.2; [4], Theorem
1.1.23). Let (A, D(A)) be a linear operator on X. Suppose that A generates
a bounded strongly continuous semigroup (T (t)) with kT (t)k ≤ M on X,
where M ≥ 1, and there exists a constant C > 0 such that
C
(3.2) kR(r + is, A)k ≤
|s|
for all r > 0 and 0 6= s ∈ R. Then Σ π2 +δ ⊂ ρ(A) for δ = arctan( C1 ) and the
resolvent of A satisfies
√
C2 + M 2
(3.3) kλR(λ, A)k ≤
1−q
for all q ∈ (0, 1) and all λ ∈ C such that | arg(λ)| ≤ π2 + arctan( Cq ).
Proof. This is essentially contained in the proof of (b) ⇒ (a) of [4, Theorem
1.1.23], so we only indicate minor modifications to reach the quantitative
version as stated.
As in [4, page 21], the Hille–Yosida Theorem 2.5 gives the bound
M
kR(λ, A)k ≤ for Re λ > 0.
Re λ
Combined with assumption (3.2), it follows that
 
M C
kR(λ, A)k ≤ min , .
Re λ | Im λ|
Now either Re λ ≥ √ M or | Im λ| ≥
|λ| √ C
|λ| holds and therefore
M 2 +C 2 M 2 +C 2
√
C2 + M 2
(3.4) kR(λ, A)k ≤ for Re λ > 0.
|λ|
(The
√ argument in [4, page
√ 21] is similar but gives the slightly larger constant
2(C + M ) in place of C 2 + M 2 .)
Let then q ∈ (0, 1) and
π q
(∗) Re λ ≤ 0, | arg(λ)| ≤ + arctan( )
2 C
thus | Re λ|/| Im λ| ≤ q/C. For any q ′ ∈ (q, 1), [4, page 22] obtains the
estimates
C 1 1  q2  1
kR(λ, A)k ≤ , ≤ + 1 .
1 − q ′ | Im λ| | Im λ|2 C2 |λ|2
Taking the limit q ′ → q, we conclude that
p √
C q2 + C 2 1 C2 + M 2 1
(3.5) kR(λ, A)k ≤ ≤
1−q C |λ| 1−q |λ|
6 T. HYTÖNEN AND S. LAPPAS

for all λ as in (∗). By combining estimates (3.4) and (3.5) we deduce that
√
C2 + M 2
kλR(λ, A)k ≤
1−q

for all q ∈ (0, 1) and all λ ∈ C such that | arg(λ)| ≤ π

2 + arctan( Cq ). 

For linear operators satisfying the conclusions of Lemma 3.1 and appro-
priate paths γ, the exponential function “etA ” can now be defined via the
Cauchy integral formula. In particular, we write

1
ˆ
(3.6) T (z) := ezA = eµz R(µ, A) dµ,
2πi γ

where z ∈ Σδ and γ is a smooth curve in Σ π2 +δ ⊂ ρ(A) running from

′ ′
∞e−i(π/2+δ ) to ∞ei(π/2+δ ) for some δ′ ∈ (| arg(z)|, δ).

3.7. Lemma ([3], Proposition 4.3 and [9], Chapter 2, Theorem 5.2). Let
(A, D(A)) be a linear operator in X such that Σ π2 +δ ⊂ ρ(A) for δ =
arctan( C1 ), where C ≥ 1, and the resolvent of A satisfies
√
C2 + M 2
(3.8) kµR(µ, A)k ≤
1−q

for all q ∈ (0, 1), some M ≥ 1 and all µ ∈ C such that | arg(µ)| ≤ π2 +
arctan( Cq ). Then, for all z ∈ Σδ , the maps T (z) are bounded linear operators
on X and
√   
C2 + M 2 C
kT (z)k ≤ 8 1 + log ,
1−u 1−u

for all | arg(z)| ≤ arctan( Cu ) and all u ∈ (0, 1).

Proof. Let z ∈ Σ π2 +δ such that | arg(z)| ≤ arctan( Cu ) for some u ∈ (0, 1). We
consider an auxiliary number q ∈ (u, 1), to be chosen later.
We will verify that the integral in (3.6) defining T (z) converges uniformly
in L(X) with respect to the operator norm. Since the functions eµz and
R(µ, A) are analytic in Σ π2 +δ , this integral, if it exists, is by Cauchy’s integral
theorem independent of the particular choice of γ. Hence, we may choose γ
to be the positively oriented boundary of Σ π +arctan( q ) \ B 1 (0). Now, we
2 C |z|
decompose γ in three parts given by the arc e
γ formed by the boundary of the
disk and the two rays going to infinity (this follows by change of variables
 BOUNDED HOLOMORPHIC SEMIGROUPS 7

π q
µ = re±i( 2 +arctan( C )) ):
(3.9) ˆ
1 1
ˆ
µz
e R(µ, A) dµ = eµz R(µ, A) dµ
2πi γ 2πi eγ
ˆ ∞ q
1 i( π
2 +arctan( C )) z π q π q
+ ere R(rei( 2 +arctan( C )) , A)ei( 2 +arctan( C )) dr
2πi 1
|z|
ˆ ∞ q
1 −i( π2 +arctan( C )) z π q π q
− ere R(re−i( 2 +arctan( C )) , A)e−i( 2 +arctan( C )) dr.
2πi 1
|z|

Since 0 < u < q < 1 ≤ C, we have | arg(z)| ≤ arctan( Cu ) < arctan( Cq ) <
arctan(1) = π4 and thus
   
π q π
+ arctan ± arg(z) ∈ ,π .
2 C 2
Hence,
π q π q
Re(re±i( 2 +arctan( C )) z) = Re(re±i( 2 +arctan( C )) |z|ei arg(z) )
   
π q
= r|z| cos + arctan + arg(z)
2 C
    
π q u
≤ r|z| cos + arctan − arctan
2 C C
    
q u
= −r|z| sin arctan − arctan ,
C C
where
    q q
q u dx q−u
ˆ ˆ
C C
arctan − arctan = ≥ dx = .
C C u 1 + x2 u C
C C

Since sin x ≥ π2 x for x ∈ [0, π2 ], it further follows that

π q 2(q − u)
Re(re±i( 2 +arctan( C )) z) ≤ −r|z|ε′ , > 0. ε′ :=
πC
√
Combining this with (3.8) and (3.9), and denoting M f := C 2 +M 2 , we get
1−q

1
ˆ f
M
ˆ
|dµ| M f ∞ ˆ
′ dr
eµz R(µ, A) dµ ≤ eRe(µz) + e−ε r|z|
2πi γ 2π eγ |µ| π 1 r
|z|

f|z| ˆ
M −1 fˆ ∞
M ds
≤ e|z z| |dµ| + e−s
2π γe π ε′ s
f  1 ds ∞ 
f+ M
ˆ ˆ
≤ eM + e−s ds
π ε′ s 1
f  
= eMf + M log 1 + e−1
π ε′
8 T. HYTÖNEN AND S. LAPPAS

  f  
f 1 M 1
=M e+ + log ′
eπ π ε
  
(3.10) f 3 + 1 log 1
≤M .
π ε′
1+Cu
We now choose q = 1+C ∈ (u, 1), so that
C(1 − u) 1−u
1−q = , q−u= .
1+C 1+C
Recalling that C ≥ 1, it follows that
√ √ √
C 2 + M2 1 + C C 2 + M2 C2 + M 2
f=
M = ≤2
1−q C 1−u 1−u
and
1 πC πC(1 + C)
log ′ = log = log
ε 2(q − u) 2(1 − u)
1+C 1
= log π + log C + log + log ,
2 1−u
where log 1+C
2 ≤ log C. Hence
 
1 1 log π 2 1  C
3 + log ′
≤ 3 + + log C + log ≤ 4 + log
π ε π π 1−u 1−u
and finally
   √
f 1 1 C2 + M 2  C 
M 3 + log ′ ≤8 1 + log .
π ε 1−u 1−u
Together with (3.10), this shows that This shows that the integral defining
(T (z)) converges in L(X) absolutely and uniformly for every z ∈ C such that
| arg(z)| ≤ arctan( Cu ), and proves the asserted bound. 
3.11. Remark. As shown in [3, Proposition 4.3] (see also [9, Chapter 2, The-
orem 5.2]) the previous family of bounded linear operators T (z) also satisfies
properties (i), (ii) and (iii) of Definition 2.7.
The following is a variant of [1, Corollary 3.7.12] and [9, Chapter 2, The-
orem 5.2]; except for the quantitative bound, it is stated in the same form
in [4, Lemma 1.1.28]:
3.12. Theorem. Let A be the infinitesimal generator of a strongly continous
semigroup T (t) with kT (t)k ≤ M , M ≥ 1. Suppose that there exists a
constant C ≥ 1 such that is ∈ ρ(A) for |s| > 0 and
C
(3.13) kR(is, A)k ≤ .
|s|
Then A generates a bounded holomorphic semigroup T (z), defined at every
1
z ∈ C such that | arg(z)| . CM , and satisfying the bound
kT (z)k . CM (1 + log(CM )).
 BOUNDED HOLOMORPHIC SEMIGROUPS 9

Proof. The proof follows the idea of the proof of [4, Lemma 1.1.28]. Since
a self-contained proof is not much longer than an indication of the relevant
changes, we give it for the reader’s convenience.
The Hille–Yosida Theorem 2.5 shows that
M
(3.14) λ ∈ ρ(A), kR(λ, A)k ≤ for all λ ∈ C with Re λ > 0.
Re λ
If r > 0 and s 6= 0, the resolvent equation (Theorem 2.8) provides the
identity
R(r + is, A) = (is − (r + is))R(r + is, A)R(is, A) + R(is, A).
Taking the operator norm on both sides and applying (3.13) and (3.14), we
deduce the bound
kR(r + is, A)k ≤ CrkR(r + is, A)k|s|−1 + C|s|−1
e −1 .
≤ CM |s|−1 + C|s|−1 = C(M + 1)|s|−1 = C|s|
where Ce := C(M + 1). Hence, A satisfies the assumptions and therefore the
conclusions of Lemma 3.1. Moreover, by Lemma 3.7 we obtain
p
e2 + M 2 
C
 e 
C
kT (z)k . 1 + log
1−u 1−u
e + M )(1 + log(C))
e 1
. (C for | arg(z)| . ,
e
C
where in the last step we chose u = 21 . Therefore A generates a bounded
holomoprhic semigroup with the stated properties by Remark 3.11. 
The following is a quantitative version of a theorem of Kato [7]. We follow
the proof given in [4, Lemma 1.2.2]:
3.15. Theorem. Let (T (t)) be a strongly continuous semigroup with kT (t)k ≤
M , M ≥ 1. Suppose that there is a complex number ζ with |ζ| = 1 and
K ∈ (0, ∞) such that
(3.16) k(ζI − T (t))xk ≥ kxk/K for x ∈ X and all t > 0.
Then (T (t)) can be extended to a bounded holomorphic semigroup. In par-
ticular,
e + log(C)),
kT (z)k . C(1 e
for every z ∈ C such that | arg(z)| . 1 , where Ce := M 2 K > 0.
e
C

Note that any constant K ∈ (0, ∞) in (3.16) must necessarily satisfy

K ≥ 21 . Indeed, as t → 0, we have T (t)x → x, and hence
kxk/K ≤ k(ζI − T (t))xk ≤ |ζ|kxk + kT (t)xk → 2kxk.
Proof. We will borrow the results of some intermediate steps of the proof
of the implication (c) ⇒ (a) of [4, Lemma 1.2.2]. That proof is set up for
a more general semigroup with a growth bound kT (t)k ≤ M eωt , so we can
10 T. HYTÖNEN AND S. LAPPAS

simply take ω = 0. Moreover, (3.16) is only assumed for t ∈ (0, δ), so we can
take δ = ∞, thus 1/δ = 0. This will simplify some of the formulas of [4].
As in [4], we take θ1 ∈ [0, 2π) and θ2 := 2π − θ1 ∈ (0, 2π] such that
e 1 = e−iθ2 = ζ, where ζ is the number appearing in (3.16). In [4, page 30],
iθ

one then considers α > max{θ1 /δ, ωθ1 }. In our case, this becomes simply
α > 0. In [4, page 31], one further chooses ε > 0, ξ0 > ω, and α1 such that
α1 − ωθ1
− ξ0 > ε.
M Kθ1
For ω = 0, we can take α1 > 0 as small as we like, then ε, ξ0 ∈ (0, 2MαKθ
1
1
),
which clearly satisfy the requirements. The second-to-last paragraph of [4,
page 31] then concludes that
M Kθ1
iα ∈ ρ(A), kR(iα, A)k ≤
α
for all α > α1 . Since we can take α1 > 0 as small as we like, this holds for
every α > 0. As in [4, page 31], the same argument for −α in place of α,
and θ2 in place of θ1 , leads to
M Kθ2
−iα ∈ ρ(A), kR(−iα, A)k ≤ , for all α > 0.
α
A combination of the last two displays then shows that
C
iα ∈ ρ(A), kR(iα, A)k ≤ , for all α ∈ R \ {0},
|α|
where C := M K max{θ1 , θ2 }. Note that
1
π = (θ1 + θ2 ) ≤ max{θ1 , θ2 } ≤ θ1 + θ2 = 2π.
2
Hence the assumptions of Theorem 3.12 are satisfied with M and C as just
defined, and the said theorem implies that (T (t)) can be extended to a
bounded holomorphic semigroup with the estimate
e + log(C)),
kT (z)k . C(1 e

for every z ∈ C such that | arg(z)| . 1 e := M 2 K > 0.

e,
C
where C 

3.17. Remark. A version of Theorem 3.15 is also true (see [4, Lemma 1.2.2]
and [7]), where (3.16) is only assumed for 0 < t < δ, but we only treat the
case stated, since it simplifies the quantitative conclusions and it is the only
case that we need for our applications.
3.18. Corollary. Let (T (t)) be a strongly continuous semigroup with kT (t)k
≤ M , for some M ≥ 1. In addition, suppose that there exists 0 < ε ≤ 2 such
that
(3.19) kT (t) − Ik ≤ 2 − ε for all t > 0.
 BOUNDED HOLOMORPHIC SEMIGROUPS 11

Then (T (t)) can be extended to a bounded holomorphic semigroup in Σθ ,

e = M 2 > 0. In particular,
where θ ≈ 1e and C
C ε
  
M2 M
(3.20) kT (z)k . 1 + log ,
ε ε
1
for every z ∈ C such that | arg(z)| . and
e
C
  
M4 M
(3.21) sup ktT ′ (t)k . 2 1 + log .
t>0 ε ε
Proof. We choose ζ = −1. Since
k(−I − T (t))xk = k − 2x + (I − T (t))xk ≥ 2kxk − k(I − T (t))xk ≥ εkxk,
we deduce from Theorem 3.15 with K = 1ε ≥ 12 that (T (t)) extends to a
bounded holomorphic semigroup which satisfies the following estimate:
  
M2 M
(3.22) kT (z)k . 1 + log ,
ε ε
e= M2
for every z ∈ C such that | arg(z)| . 1e , where C ε > 0. By Cauchy’s
C
integral formula we get
1 T (z)
˛
′
(3.23) T (t) = dz,
2πi (t − z)2
γ

where γ is the boundary of the disk Bt sin(θ) (t) for t > 0 and θ ≈ 1e is the
C
angle of the sector Σθ in which T (z) is holomorphic. Using (3.22) and (3.23)
we deduce that
  
′ 1 M2 M 1
kT (t)k . 1 + log max 2π|t sin(θ)|
2π ε ε z:|t−z|=t sin θ (t − z)2
  
1 M2 M
= 1 + log
t sin(θ) ε ε
Hence,
  
′ 1 M2 M
sup ktT (t)k . 1 + log
t>0 sin(θ) ε ε
  
M4 M
. 2 1 + log ,
ε ε
1 1
where in the last step sin(θ) ≈ e if θ ≈ e. 
C C

3.24. Remark. The first part of Corollary 3.18 was proved by Kato in [7]
under the assumption of (3.19) for 0 < t < δ. The novelty here is the
estimate (3.20) and (3.21).
12 T. HYTÖNEN AND S. LAPPAS

4. Quantitative analyticity of diffusion semigroups in

uniformly convex spaces
In our discussions below we follow [16] but we provide new details con-
cerning the quantitative bounds. Let us recall the following definitions about
symmetric diffusion semigroups, uniform convexity, martingale type and co-
type:
4.1. Definition. Let (Ω, A, µ) be a σ-finite measure space. An operator T
on (Ω, A, µ) is a symmetric Markovian operator if it satisfies the following
properties:
(1) T is a linear contraction on Lp (Ω) for every 1 ≤ p ≤ ∞;
(2) T is positivity preserving and T 1 = 1; and
(3) T is a self-adjoint operator on L2 (Ω).
4.2. Definition. Let (Ω, A, µ) be a σ-finite measure space. By a symmet-
ric diffusion semigroup on (Ω, A, µ) in Stein’s sense [14, Section III.1], we
mean a family {Tt }t>0 of linear maps where each member of this family is
a symmetric Markovian operator and satisfies the following two additional
properties:
(4) Tt Ts = Tt+s ;
(5) limt→0 Tt f = f in L2 (Ω) for every f ∈ L2 (Ω).
4.3. Definition. A Banach space X is said to be uniformly convex if for
every ε ∈ (0, 2] there exists δ ∈ (0, 1] such that, for any vectors x, y in the
closed unit ball (i.e. kxk ≤ 1 and kyk ≤ 1) with kx − yk ≥ ε, one has
kx + yk ≤ 2(1 − δ). In addition, (X, k · k) is said to be uniformly convex
of power type q with 2 ≤ q < ∞ and positive constant δ if the following
inequality holds:
q q
x+y x−y 1
(4.4) +δ ≤ (kxkq + kykq ), ∀x, y ∈ X.
2 2 2
Examples of uniformly convex spaces that satisfy (4.4) with δ = 1 are the
following:
(1) X is a Hilbert space and q = 2. By the Parallelogram law for inner
product spaces it follows:
2 2
x+y x−y 1
+ = (kxk2 + kyk2 ), ∀x, y ∈ X.
2 2 2
(2) X = Lq (Ω) and 2 ≤ q < ∞. We have the following Clarkson’s inequality
(see [5, Corollary 2.29]):
q q
x+y x−y 1
+ ≤ (kxkqLq + kykqLq ), ∀x, y ∈ X.
2 Lq 2 Lq 2
4.5. Definition. Let 1 < q < ∞. A Banach space X is of martingale
cotype q if there exists a positive constant C such that every finite X valued
 BOUNDED HOLOMORPHIC SEMIGROUPS 13

Lq -martingale (fn ) defined on some probability space satisfies the following

inequality:
X
Ekfn − fn−1 kqX ≤ C q sup Ekfn kqX ,
n
n≥1
where E denotes the expectation on the underlying probability space. We
then must have q ≥ 2. X is of martingale type q if the reverse inequality
holds.
4.6. Remark. Following the notation in [12], given a Banach space (X, k · k)
and q ≥ 2 we will denote the martingale cotype q constant of X by mq,X ∈
[1, ∞].
4.7. Theorem ([10], Theorem 3.1). Let q be such that 2 ≤ q < ∞ and let X
be a Banach space. Assume that X is of martingale cotype q, namely:
X
Ekfn − fn−1 kqX ≤ mqq,X sup Ekfn kqX .
n
n≥1

Then there exists an equivalent norm | · | on X such that:

(
∀x, y ∈ X, kxk ≤ |x| ≤ mq,X kxk
(4.8) |x|q +|y|q
and, | x+y q x−y q
2 | +k 2 k ≤ 2 .
In particular, (X, | · |) is uniformly convex of power type q with constant
δ = m−q
q,X .

4.9. Remark. The above result states that if X has martingale cotype q, then
it admits an equivalent norm that satisfies (4.8). The converse to this is also
true and it can be found in [10, Remark 3.1] and [13, Theorem 10.6 (i) =⇒
(iii)]. In particular, if we assume that X is uniformly convex of power type
−1
q with constant δ, then X has martingale cotype q with mq,X ≤ 2δ q . As it
can be seen below in our applications, we could work under the assumption
of uniformly convex of power type q spaces which would imply estimates in
terms of the constant δ.
We will need two results in order to derive bounds for the family of op-
erators {t∂Tt }t>0 on Y = Lp (Ω; X), where 1 < p < ∞, X is a uniformly
convex Banach space or a Banach space of martingale cotype 2 ≤ q < ∞.
The first one is the following Rota’s dilation theorem for positive Markovian
operators, which we state following the formulation in [16, Lemma 10]:
4.10. Lemma ([14], Chapter IV and [16], Lemma 10). Let T = S 2 with
S a symmetric Markovian operator on (Ω, A, µ). Then there exist a larger
e A,
measure space (Ω, eµe) containing (Ω, A, µ), and a σ-subalgebra B of Ae such
that, for every p ∈ [1, ∞) and every Banach space X,
(4.11) T f = EA EB f, ∀f ∈ Lp (Ω, A, µ; X),
where EA is the conditional expectation relative to A (and similarly for EB ).
14 T. HYTÖNEN AND S. LAPPAS

The fact that Rota’s theorem remains valid for functions taking values in
an arbitrary Banach space, as stated above, has been observed e.g. in [8,
after Theorem 2.5]) and [16, the beginning of the proof of Lemma 9]. We
indicate a proof for the reader’s convenience.
Proof. For X ∈ {R, C} (i.e., scalar-valued functions), a more general version
of this lemma is formulated and proved in [14, Chapter IV, Theorem 9], and
restated in the form above as [16, Lemma 10].
PK
Let then X be a general Banach space. If f = k=1 fk ⊗ xk , where
fk ∈ Lp (Ω, A, µ) are scalar-functions and xk ∈ X, the identity follows from
the fact that it holds for each fk by linearity of both sides of (4.11).
For the general case, we note that functions of the form just discussed
are dense in Lp (Ω, A, µ; X). On the other hand, both symmetric Markovian
operators (by Definition 4.1) and conditional expectations (by their well-
known basic properties) are positive linear operators (i.e., they map nonneg-
ative functions to nonnegative functions). Such operators, initially defined
on scalar-valued Lp spaces, have canonical bounded linear extensions to the
corresponding Lp spaces of X-valued functions, for any Banach space X (for
this result, see e.g. [5, Theorem 2.1.3]). Since (4.11) holds for all f in a
dense subspace of Lp (Ω, A, µ; X), and both sides depend continuously on
f ∈ Lp (Ω, A, µ; X), the identity remains valid for all f as stated. 
The following elementary estimate improves the intermediate steps of the
proof given in [16, Lemma 9]:
4.12. Lemma. Let q ∈ (1, ∞) and δ ∈ (0, 1]. If x ∈ [0, 2] satisfies
xq + δ2q (x − 1)q+ ≤ 2q , where (x − 1)+ = max(x − 1, 0),
2δ 1
then x ≤ 2 − ε, where ε := ∈ (0, 1).
1 + 2δ q
Proof. Let us first observe the elementary inequalities
(
α ≤ 1 − αt, α ∈ (0, 1),
(4.13) ∀t ∈ [0, 1] : (1 − t)
≥ 1 − αt, α ∈ (1, ∞).
Now, we proceed with the proof of our lemma. Arguing by contradiction,
we assume that x > 2 − ε, thus x − 1 > 1 − ε > 0, and hence
2q ≥ xq + δ2q (1 − ε)q ≥ xq + δ2q (1 − qε)
by the assumption and (4.13) with α = q > 1. Thus
xq ≤ 2q [1 − δ(1 − qε)],
where qε ∈ (0, 1) and hence δ(1 − qε) ∈ (0, 1). By (4.13) with α = 1q ∈ (0, 1),
we have
 
1 1 2δ
x ≤ 2[1 − δ(1 − qε)] q ≤ 2 1 − δ(1 − qε) = 2 − (1 − qε).
q q
 BOUNDED HOLOMORPHIC SEMIGROUPS 15

Here
2δ 1 2δ 2δ 1
qε = , 1 − qε = , (1 − qε) = = ε.
1 + 2δ 1 + 2δ q 1 + 2δ q
So, assuming that x > 2 − ε, we arrived at x ≤ 2 − ε, a contradiction. Thus
necessarily x ≤ 2 − ε, as claimed. 
The following lemma is well known and immediate from the definition:
4.14. Lemma. If (X, k·k) is uniformly convex of power type q with 2 ≤ q < ∞
and constant δ > 0 appearing in (4.4), then so is Y = Lq (Ω; (X, k · k)).
The following lemma is a quantitative elaboration of [16, Lemma 9]:
4.15. Lemma. Let T = S 2 with S a symmetric Markovian operator on
(Ω, A, µ) and suppose that X is uniformly convex of power type q with 2 ≤
q < ∞ and δ > 0 appearing in (4.4). Then
kI − T k ≤ 2 − ε, on Y = Lq (Ω; X),
2δ 1
where ε := 1+2δ q ∈ (0, 1).
Proof. Notice that by Lemma 4.14, Y = Lq (Ω; X) is also uniformly convex
of power type q with constant δ > 0 and T is a contraction on Y . By Rota’s
dilation lemma (its X-valued version as formulated in Lemma 4.10), we have
T = EA EB Y
.
On the other hand, observe that EA acs as the identity on Y . Hence, for
any λ ∈ C, we have
λ + T = (λEA + EA EB ) Y
= EA (λ + EB ) Y
=: EA (λ + P ) Y ,
where we abbreviated P := EB .
Let y be a unit vector in Y . Using (4.4), we get
q q
λy + P y λy − P y 1
+δ ≤ (|λ|q + 1),
2 2 2
where kλy + P yk ≥ kEA (λy + P y)k ≥ kλy + T yk. On the other hand (noting
that P and EA are contractive projections), we have
λy − P y ≥ P (λy − P y) = λP y − P 2 y = λP y − P y = (λ − 1)P y ,
and hence



λy−P y ≥ |1−λ| P y ≥ |1−λ| λy+P y −|λ| ≥ |1−λ| λy+T y −|λ| .
Also clearly kλy − P yk ≥ 0, and hence in fact


λy − P y ≥ |1 − λ| λy + T y − |λ| + ,


where λy + T y − |λ| + = max(kλy + T yk − |λ|, 0). When kλy + T yk
approaches kλ + T k, we then deduce
 
λ+T q q λ + T − |λ| q 1
+ δ |1 − λ| ≤ (|λ|q + 1).
2 2 + 2
16 T. HYTÖNEN AND S. LAPPAS

In particular, for λ = −1 we obtain

kI − T kq + δ 2q (kI − T k − 1)q+ ≤ 2q .
By Lemma 4.12 for x := kI − T k ∈ [0, 2] this implies
kI − T k ≤ 2 − ε,
2δ 1
where ε := 1+2δ q ∈ (0, 1). 

Corollary 3.18 and Lemma 4.15 imply the following result, which is [11,
Remark 1.8(b)] (see also [16, Lemma 11]):
4.16. Corollary. Suppose that X is uniformly convex of power type q with
2 ≤ q < ∞ and δ > 0 appearing in (4.4). Moreover, suppose that {Tt }t>0 is
a symmetric diffusion semigroup on (Ω, A, µ). Then the extension of {Tt }t>0
to Y = Lq (Ω; X) is analytic. More precisely, we have
  
q q
kT (z)k . 1 + log ,
δ δ
δ
for every z ∈ C such that | arg(z)| . q and {t∂Tt }t>0 is a uniformly bounded
family of operators on Y , namely,
  
q2 q
sup kt∂Tt k . 2 1 + log .
t>0 δ δ
Proof. Applying Lemma 4.15 to T = Tt , we get
kI − Tt k ≤ 2 − ε, for all t > 0.
2δ 1 δ δ
where ε = 1+2δ q ≈ q . Then by choosing M = 1 and ε ≈ q in Corollary 3.18
the proof follows. 
Now, Theorem 4.7 and Corollary 4.16 imply the following estimate of
supt>0 kt∂Tt k in terms of martingale cotype:
4.17. Corollary. Let X be a Banach space of martingale cotype 2 ≤ q < ∞
with cotype constant mq,X . Moreover, suppose that {Tt }t>0 is a symmetric
diffusion semigroup on (Ω, A, µ). Then the extension of {Tt }t>0 to Y =
Lq (Ω; X) is analytic. More precisely, we have
(4.18) kT (z)k . q(mq,X )q+1 (1 + log(q) + q log(mq,X )),
1
for every z ∈ C such that | arg(z)| . qmqq,X
and {t∂Tt }t>0 is a uniformly
bounded family of operators on Y , namely,
(4.19) sup kt∂Tt k . B, B := q 2 (mq,X )2q+1 (1 + log(q) + q log(mq,X )).
t>0

Proof. Assuming that |·| is the equivalent norm on X guaranteed by Theorem

4.7 and Y = Lq (Ω; (X, | · |)) we can deduce by Theorem 4.7 and Corollary
4.16 that
|T (z)| . qmqq,X (1 + log(q) + q log(mq,X ))
 BOUNDED HOLOMORPHIC SEMIGROUPS 17

and
sup |t∂Tt | . (q(mq,X )q )2 (1 + log(q) + q log(mq,X )).
t>0
Hence we conclude that
kT (z)k ≤ mq,X |T (z)|
. q(mq,X )q+1 (1 + log(q) + q log(mq,X ))
and
sup kt∂Tt k ≤ mq,X sup |t∂Tt |
t>0 t>0
. q 2 (mq,X )2q+1 (1 + log(q) + q log(mq,X )).


5. Littlewood–Paley–Stein inequalities: First approach

In this section we will use the quantitative analyticity bounds of the previ-
ous section to obtain the following new quantitative version of [16, Theorem
2 (i)]:
5.1. Theorem. Let X be a Banach space and k a positive integer. If X has
martingale cotype q with 2 ≤ q < ∞, then for every symmetric diffusion
semigroup {Tt }t>0 and for every 1 < p < ∞ we have
ˆ ∞ 1/q
k k q dt
(5.2) t ∂ Tt f X .p,q kk−1 B k+1 mq,X f Lp (Ω;X) ,
0 t Lp (Ω)

for all f ∈ Lp (Ω; X), where B := q 2 (mq,X )2q+1 (1 + log(q) + q log(mq,X )) is

the constant appearing in (4.19) of Corollary 4.17.
If p = q and k = 1, then we have the sharper bound
ˆ ∞ 1/q
q dt
(5.3) t∂Tt f X . Bmq,X f Lq (Ω;X) ,
0 t Lq (Ω)

for all f ∈ Lq (Ω; X) and the implied constant is absolute.

5.4. Remark (see also [16], Remark 3). Applied to the heat semigroup {Ht }t>0
of Rn , the above theorem implies a dimension free estimate for the g-function
associated to {Ht }t>0 :
ˆ ∞ 1/q
q dt
t∂Ht f X .p,q B 2 mq,X f Lp (Rn ;X) , ∀f ∈ Lp (Rn ; X).
0 t n
Lp (R )

when X is of martingale cotype q. In addition, if p = q the same theorem

implies
ˆ ∞ 1/q
q dt
t∂Ht f X . Bmq,X f Lq (Rn ;X) , ∀f ∈ Lq (Rn ; X).
0 t Lq (Rn )
18 T. HYTÖNEN AND S. LAPPAS

Compare this with the following estimate which was obtained in [6, Theorem
17]:
ˆ ∞ 
q dt 1/q √
t∂Ht f Lq (Rn ;Y ) . n · mq,X kf kLq (Rn ;Y ) , ∀f ∈ Lq (Rn ; Y )
0 t
when Y is a Banach space that admits an equivalent norm with modulus of
uniform convexity of power type q. We note that the constant B can be very
large, so that the second bound is typically sharper in moderate dimensions
n, but becomes inferior as n → ∞.
In order to prove Theorem 5.1 the following lemma, due to Naor and one
of us [6, Lemma 24] will play a crucial role in our argument:
5.5. Lemma ([6], Lemma 24). Fix q ∈ [2, ∞), α ∈ (1, ∞) and a Banach
space (X, k · kX ) of martingale cotype q with constant mq,X . Suppose that
{Tt }t>0 is a symmetric diffusion semigroup on a measure space (Ω, X). Then
for any f ∈ Lq (Ω; X) we have
ˆ ∞ 1
q dt q 1
k(Tt − Tαt )f kLq (Ω;X) ≤ (log(α)) q mq,X kf kLq (Ω;X) .
0 t
The following lemma shows Theorem 5.1 in the case p = q and k = 1, 2,
but the bound provided by the lemma is worse than that asserted in the
theorem when k > 2. Thus, some more work will be needed further below
to obtain Theorem 5.1 as stated.
5.6. Lemma (quantitative version of [16], Lemma 13). Let X be a Banach
space of martingale cotype 2 ≤ q < ∞ with cotype constant mq,X and k be
a positive integer. Moreover, suppose that {Tt }t>0 is a symmetric diffusion
semigroup on a measure space (Ω, X). Then for any f ∈ Lq (Ω; X) we have
ˆ ∞ 
k k q dt 1/q
(5.7) t ∂ Tt f Lq (Ω;X) . kk B k mq,X kf kLq (Ω;X) ,
0 t
where B := q 2 (mq,X )2q+1 (1 + log(q) + q log(mq,X )) is the constant appearing
in (4.19) of Corollary 4.17.
5.8. Remark. Later on we will only use this lemma in the case k = 1 and we
will get a different bound form the one appearing in (5.7) for k > 1.
Proof. We will use the idea of the proof of [6, Theorem 17]. By the semigroup
identity ∂Tt+s = ∂Tt Ts and the convergence
1 B
k∂Tt k = kt∂Tt k ≤ −→ 0
t t t→∞
from (4.19), we can expand
∞
X ∞
X



(5.9) ∂Tt f = ∂T2k+1 t − ∂T2k+2 t f = ∂T2k t T2k t − T3·2k t f.
k=−1 k=−1
 BOUNDED HOLOMORPHIC SEMIGROUPS 19

Then by the triangle inequality we can estimate

ˆ ∞
q dt
1/q
t∂Tt f Lq (Ω;X)
0 t
∞  
X ∞
dt 1/q
ˆ
q
≤ t∂T2k t T2k t − T3·2k t f Lq (Ω;X)
0 t
k=−1
(5.10) ∞ ˆ 1/q
X ∞
q dt
−k
= 2 t∂Tt Tt − T3t f Lq (Ω;X)
0 t
k=−1
ˆ ∞ 1/q

q dt
=4 t∂Tt Tt − T3t f Lq (Ω;X)
.
0 t
We are now in a position of using Corollary 4.17 as follows:



(5.11) t∂Tt Tt − T3t f Lq (Ω;X) . B Tt − T3t f Lq (Ω;X) , ∀ t > 0.

where B := q 2 (mq,X )2q+1 (1 + log(q) + q log(mq,X )).

Combining the above inequalities together with Lemma 5.5, we derive
ˆ ∞  ˆ ∞ 
q dt 1/q
q dt 1/q
t∂Tt f Lq (Ω;X) .B Tt − T3t f Lq (Ω;X)
0 t 0 t
1
. B(log(3)) q mq,X f Lq (Ω;X)
. Bmq,X f Lq (Ω;X)
.

This is (5.7) for k = 1. To handle a general k, by the semigroup identity

Tt+s = Tt Ts , we have
 k
k k k t
t ∂ Tt = k ∂T t .
k k

This implies that

(5.12)
ˆ ∞  ˆ ∞ 
q dt 1/q q dt 1/q
tk ∂ k Tt f Lq (Ω;X) = kk (t∂Tt )k f Lq (Ω;X) .
0 t 0 t
Thus, by (5.11), (5.12) and the already proved inequality, we obtain
ˆ ∞ 
k k−1 q dt 1/q
(5.12) . k (t∂Tt ) (t∂Tt )f Lq (Ω;X)
0 t
ˆ ∞ 1/q
q dt
. kk B k−1 t∂Tt f Lq (Ω;X)
0 t
. kk B k mq,X kf kLq (Ω;X) .
The lemma is thus proved. 
To show Theorem 5.1 for any 1 < p < ∞, we will use Stein’s complex
interpolation machinery. To that end, we will need the notion of fractional
20 T. HYTÖNEN AND S. LAPPAS

integrals. For a (nice) function ϕ on (0, ∞) we define

ˆ t
α 1
I ϕ(t) = (t − s)α−1 ϕ(s)ds, t > 0.
Γ(α) 0
The integral on the right hand side is well defined for any α ∈ C with
Re α > 0; moreover, Iα ϕ is analytic in the right half complex plane Re α > 0.
Using integration by parts, Stein showed in [14, Section III.3] that Iα ϕ has
an analytic continuation to the whole complex plane, which satisfies the
following properties:
(1) Iα Iβ ϕ = Iα+β ϕ for any α, β ∈ C;
(2) I0 ϕ = ϕ; and
(3) I−k ϕ = ∂ k ϕ for any positive integer k.
We will apply Iα to ϕ defined by ϕ(s) = Ts f for a given function f in
Lp (Ω; X) and set
Mαt f = t−α Iα ϕ(t) with ϕ(s) = Ts f.
Note that
t
1
ˆ
M1t f = Ts f ds, M0t f = Tt f and M−k k k
t f = t ∂ Tt f for k ∈ N.
t 0

The following lemma is [8, Theorem 2.3]:

5.13. Lemma ([8], Theorem 2.3 and [16], Lemma 14). Let q and X be as in
Theorem 5.1. Then for any 1 < p < ∞ we have
ˆ ∞ 1/q
1 q dt
t∂Mt f X .p,q mq,X f Lp (Ω;X) , ∀ f ∈ Lp (Ω; X).
0 t Lp (Ω)

5.14. Lemma ([16], Lemma 15). Let α and β be complex numbers such that
Re α > Re β > −1. Then for any positive integer k
ˆ ∞ 1 ˆ ∞ 1
k k α q dt q π|Im (α−β)| k k β q dt q
t ∂ Mt f X .Re α,Re β e t ∂ Mt f X
0 t 0 t
on Ω.
Combining Lemma 5.13 and Lemma 5.14 with k = β = 1, we get:
5.15. Lemma ([16], Lemma 16). For any 1 < p < ∞ and α ∈ C with
Re α > 1
ˆ ∞ 1/q
α q dt
t∂Mt f X .p,q,Re α mq,X eπ|Im α| f Lp (Ω;X) ,
0 t Lp (Ω)

for all f ∈ Lp (Ω; X).

Now we provide the following new quantitative version of [16, Lemma 17]
which we will use in the proof of Theorem 5.1:
 BOUNDED HOLOMORPHIC SEMIGROUPS 21

5.16. Lemma ([16], Lemma 17). For any α ∈ C

ˆ ∞ 1/q
α q dt
t∂Mt f X
(5.17) 0 t Lq (Ω)

.Re α (|α| + 1)N (Re α) B 1+N (Re α) mq,X eπ|Im α| f Lq (Ω;X)

,
for all f ∈ Lq (Ω; X), where
(
0, if Re α > 0,
(5.18) N (Re α) :=
smallest n ≥ 0 such that n > − Re α,
and B := q 2 (mq,X )2q+1 (1 + log(q) + q log(mq,X )) is the constant appearing in
(4.19) of Corollary 4.17.
Proof. Combining Lemmas 5.6 and 5.14 with β = 0, we deduce that for a
positive integer k and α ∈ C with Re α > 0
(5.19)
ˆ ∞ 1/q
k k α q dt
t ∂ Mt f X .Re α kk B k mq,X eπ|Im α| f Lq (Ω;X) ,
0 t Lq (Ω)

for all f ∈ Lq (Ω; X).

In order to prove the general case for any α ∈ C, we will use an induction
argument. In particular, we will prove the following estimate for any α ∈ C
with Re α > −n, n ∈ N
(5.20) ktk ∂ k Mαt f kLq (Ω;Lq ((R+ , dt );X)) ≤ φn (k, α) f Lq (Ω;X)
,
t

where
(5.21) φn (k, α) ≤ kk B k C(Re α + n)mq,X eπ|Im α| ψn (k, α),
C(Re α + n) is a positive constant that depends on Re α + n and
(5.22) ψn (k, α) ≤ cn (k + |α|)n B n .
Noting that for any α ∈ C
∂Mαt = −αt−1 Mαt + t−1 Mα−1
t ,
we have
(5.23) tk ∂ k Mα−1
t = (k + α)tk ∂ k Mαt + tk+1 ∂ k+1 Mαt .
Observe that for n = 0, by what is already proved in (5.19) we have
ktk ∂ k Mαt f kLq (Ω;Lq ((R+ , dt );X)) ≤ φ0 (k, α) f Lq (Ω;X)
,
t

where φ0 (k, α) ≤ kk B k C(Re α)mq,X eπ|Im α| ψ0 (k, α) and ψ0 (k, α) = 1.

Now, let us assume that (5.20) holds for some n ∈ N and we prove it for
n + 1. To be more precise, if Re α̃ > −n, where α̃ := α + 1, then using the
induction hypothesis (5.20) and (5.23) we deduce that

(5.24) ktk ∂ k Mα̃−1

t f kLq (Ω;Lq ((R+ , dt );X))
t
22 T. HYTÖNEN AND S. LAPPAS

≤ |k + α̃|ktk ∂ k Mα̃t f kLq (Ω;Lq ((R+ , dt );X)) + ktk+1 ∂ k+1 Mα̃t f kLq (Ω;Lq ((R+ , dt );X))
t t

≤ (|k + α̃|φn (k, α̃) + φn (k + 1, α̃)) f Lq (Ω;X)

= (|k + α + 1|φn (k, α + 1) + φn (k + 1, α + 1)) f Lq (Ω;X)
.
Hence,
φn+1 (k, α) = kk B k C(Reα + n + 1)mq,X eπ|Imα| ψn+1 (k, α)
≤ |k + α + 1|φn (k, α + 1) + φn (k + 1, α + 1)
(5.25)
≤ |k + α + 1|kk B k C(Reα + 1 + n)mq,X eπ|Imα| ψn (k, α + 1)
+ (k + 1)k+1 B k+1 C(Reα + 1 + n)mq,X eπ|Imα| ψn (k + 1, α + 1).
and for ψn+1 (k, α) we have
 k + 1 k
ψn+1 (k, α) ≤ |k + α + 1|ψn (k, α + 1) + (k + 1)Bψn (k + 1, α + 1),
k
where ( k+1 k
k ) ≤ e.
Now, by the induction hypothesis (5.22) for the functions ψn (k, α) and
(5.25) we have
ψn+1 (k, α) ≤ |k + α + 1|cn (k + |α + 1|)n B n
(5.26) + e(k + 1)B · cn (k + 1 + |α + 1|)n B n
≤ cn B n+1 (k + |α| + 2)n+1 (1 + e).
Here
 2 n+1
(k + |α| + 2)n+1 (1 + e) = (k + |α|)n+1 1 + (1 + e)
k + |α|
 2 n+1
(5.27) ≤ (k + |α|)n+1 1 + (1 + e)
max(1, n)
≤ (k + |α|)n+1 c,
since for the minimal nonnegative n > − Re α we have
k + |α| ≥ 1 + |α| ≥ max(1, n).
Now, by the estimate (5.26) for the functions ψn+1 (k, α) and (5.27) we can
conclude
ψn+1 (k, α) ≤ cn B n+1 (k + |α| + 2)n+1 (1 + e)
(5.28)
≤ cn B n+1 (k + |α|)n+1 c = cn+1 B n+1 (k + |α|)n+1 .
This completes the proof of (5.22) by induction. Therefore, by combining
(5.20), (5.21) and (5.22) we obtain the following estimate for any α ∈ C with
Re α > −n
ktk ∂ k Mαt f kLq (Ω;Lq ((R+ , dt );X)) ≤ φn (k, α) f Lq (Ω;X)
t
k k π|Im α| n
≤ k B C(Reα + n)mq,X e c (k + |α|)n B n f Lq (Ω;X)
k k+N (Reα) N (Reα) π|Im α|
.Reα k B (k + |α|) mq,X e f Lq (Ω;X)
,
 BOUNDED HOLOMORPHIC SEMIGROUPS 23

where n = N (Re α) is as in (5.18). In particular for k = 1, we have

(5.17). 

Now we are ready to show Theorem 5.1:

Proof of Theorem 5.1. We will prove by an induction argument the following

more general statement: for any α ∈ C
ˆ ∞ 1/q
k k q dt
(5.29) t ∂ Mαt f X
0 t Lp (Ω)

≤ C(k + |α|)k−1 B k+1+| Re α| mq,X f Lp (Ω;X)

,

for all f ∈ Lp (Ω; X), where C = C(p, q, α) is a positive constant that depends
on p, q and α, and B := q 2 (mq,X )2q+1 (1 + log(q) + q log(mq,X )).
We fix α ∈ C and choose θ ∈ (0, 1), r ∈ (1, ∞), α0 , α1 ∈ C such that the
following hold
1 1−θ θ
= + , α = (1 − θ)α0 + θ α1 ,
(5.30) p q r
Re α1 > 1 and Im α0 = Im α1 = Im α.
Then the classical complex interpolation of vector-valued Lp -spaces (see [2])
implies
 
Lp (Ω; X) = Lq (Ω; X), Lr (Ω; X) θ .
Hence for any f ∈ Lp (Ω; X) with kf kLp (Ω;X) < 1 there exists a continuous
function F from the closed strip {z ∈ C : 0 ≤ Re z ≤ 1} to Lq (Ω; X) +
Lr (Ω; X), which is analytic in the interior and satisfies
F (θ) = f, sup F (iy) Lq (Ω;X)
< 1 and sup F (1 + iy) Lr (Ω;X)
< 1.
y∈R y∈R

Define
2 −θ 2 (1−z)α0 +zα1
Ft (z) = ez t∂Mt F (z).
Viewed as a function of z on the strip {z ∈ C : 0 ≤ Re z ≤ 1}, the function F
takes values in Lq (Ω; Lq (R+ ; X)) + Lr (Ω; Lq (R+ ; X)), where R+ is equipped
with the measure dtt . By the analyticity of M(1−z)α0 +zα1 in z, we see that F
is analytic in the interior of the strip. Moreover, by Lemma 5.16 applied to
(1 − iy)α0 + iyα1 , where real part is Re α0 by (5.30), in place of α we get
ˆ ∞ 1/q
q dt
Ft (iy) X
0 t Lq (Ω)

.Re α0 (|(1 − iy)α0 + iyα1 | + 2)N (Re α0 ) B 1+N (Re α0 ) mq,X

2 −θ 2
× e−y eπ(|Im α|+|Re(α1 −α0 )y|)
.Re α0 G(y)B 1+N (Re α0 ) mq,X ,
24 T. HYTÖNEN AND S. LAPPAS

for all y ∈ R and

 p 1+| Re α0 |
2
G(y) := 2
|α0 | 1 + y + |y||α1 | + 2 e−y e|Re(α1 −α0 )y|) eπ|Im α| .

Notice that y 7→ G(y) is continuous on R and limy→±∞ G(y) = 0. Therefore

it attains some maximum M0 (α0 , α1 , α) = M1 (p, q, α), where the last equal-
ity holds since the choice of α0 and α1 in (5.30) depends only on p, q and α.
Hence
sup kF(iy)kLq (Ω;Lq ((R+ , dt );X)) .p,q,α B 1+N (Re α0 ) mq,X .
y∈R t

Similarly, Lemma 5.15, with −iyα0 + (1 + iy)α1 in place of α, implies

sup kF(1 + iy)kLr (Ω;Lq ((R+ , dt );X)) .p,q,α mq,X .

y∈R t

We then deduce that F(θ) belongs to the complex interpolation space

 
(5.31) Lq (Ω; Lq (R+ ; X)), Lr (Ω; Lq (R+ ; X)) θ

with norm majorized by

.p,q,α (B 1+N (Re α0 ) mq,X )1−θ (mq,X )θ = B (1−θ)(1+N (Re α0 )) mq,X ,

where
n  1 θ o
(1 − θ)(1 + N (Re α0 )) = (1 − θ) 1 + N Re α− Re α1
1−θ 1−θ
(∗)
n 1 θ o
= (1 − θ) 1 + N ( Re α−
1−θ 1−θ
!
n  1 θ o
≤ (1 − θ) 2 + | Re α| +
1−θ 1−θ
2 − 2θ + | Re α| + θ ≤ 2 + | Re α|,

since we can choose Re α1 just slightly larger than 1 to ensure the equality in
(∗). However, the latter space coincides with Lp (Ω; Lq (R+ ; X)) isometrically.
Since
Ft (θ) = t∂Mαt F (θ) = t∂Mαt f,
we get
ˆ ∞ 1/q
q dt
t∂Mαt f X
.p,q,α B 2+| Re α| mq,X f Lp (Ω;X)
,
0 t Lp (Ω)

for all f ∈ Lp (Ω; X). This is (5.29) for k = 1.

 BOUNDED HOLOMORPHIC SEMIGROUPS 25

Now, let us assume that (5.29) holds for some k. Then using (5.23) and
the induction hypothesis (5.29), we get

ktk+1 ∂ k+1 Mαt f kLp (Ω;Lq ((R+ , dt );X))

t

≤ ktk ∂ k Mα−1
t f kLp (Ω;Lq ((R+ , dt );X)) + |k + α|ktk ∂ k Mαt f kLp (Ω;Lq ((R+ , dt );X))
t t

≤ C (k + |α − 1|)k−1 B k+1+| Re α−1|

+|k + α|(k + |α|)k−1 B k+1+| Re α| mq,X f Lp (Ω;X)

= C(k + 1 + |α|)k−1 (1 + k + |α|)B k+2+| Re α| mq,X f Lp (Ω;X)

(k+1)−1 (k+1)+1+| Re α|
= C((k + 1) + |α|) B mq,X f Lp (Ω;X)
.

Thus, we derive (5.29) for any k. Theorem 5.1 corresponds to (5.29) for
α = 0. In particular, we get
ˆ ∞ 1/q
k k q dt
t ∂ Tt f X ≤ C(p, q)kk−1 B k+1 mq,X f Lp (Ω;X) ,
0 t Lp (Ω)

for all f ∈ Lp (Ω; X). Thus, the theorem is completely proved. 

6. Littlewood–Paley–Stein inequalities: Second approach

In this section we obtain an alternative estimate for the constant appearing
in (5.3) of Theorem 5.1 by combining our Corollary 4.17 with another result
of Xu [17, Theorem 1.4]. Before we proceed we need to recall some notions
from [17].
Recall that an operator T on Lp (Ω) (1 ≤ p ≤ ∞) is regular (more precisely,
contractively regular) if
k sup |T (fk )|kp ≤ k sup |fk |kp
k k

for all finite sequences {fk }k≥1 in Lp (Ω).

Recall that {Tt }t>0 is analytic on Lp (Ω; X) if {Tt }t>0 extends to a bounded
analytic function from an open sector Σβ0 = {z ∈ C : | arg(z)| < β0 } to
B(Lp (Ω; X)) for some 0 < β0 ≤ π2 , where B(Y ) denotes the space of bounded
linear operators on a Banach space Y . In this case,
(6.1) Tβ0 = sup{kTz kB(Lp (Ω;X)) : z ∈ Σβ0 } < ∞.

Now, we can state the following result of Xu:

6.2. Theorem ([17], Theorem 1.4). Let X be a Banach space, 2 ≤ q < ∞

and {Tt }t>0 be a strongly continuous semigroup of regular operators on Lp (Ω)
for a single 1 < p < ∞. Assume additionally that {Tt }t>0 satisfies (6.1).
26 T. HYTÖNEN AND S. LAPPAS

′
Let βq = β0 min( pq , pq′ ). If X has martingale cotype q, then
(6.3)
ˆ ∞ 1/q ′
q dt min( pq , pq′ ) 2 ′1+ q1′
t∂Tt f X
. βq−3 Tβ0 max(p q , p )mq,X
0 t Lp (Ω)
× f Lp (Ω;X)
,
for all f ∈ Lp (Ω; X).
1
By choosing p = q, β0 = qmqq,X
, Tβ0 = β0 B, where B = q 2 (mq,X )2q+1 (1 +
log(q) + q log(mq,X )), in Corollary 4.17 we can estimate the constant in (6.3)
in terms of the martingale cotype constant mq,X and exponent q as follows:
. β0−2 Bmq,X f Lp (Ω;X)
(6.4) q 2
= (qmq,X ) Bmq,X f Lp (Ω;X) .

6.5. Remark. In Corollary 4.17 we assume that {Tt }t>0 is as symmetric dif-
fusion semigroup on (Ω, A, µ). Notice that (since any positive contraction
is regular) if {Tt }t>0 is a symmetric diffusion semigroup then {Tt }t>0 is a
strongly continuous semigroup of regular operators on Lp (Ω).
We see that the bound (5.3) that we get by the first approach in the
special case p = q, k = 1 is somewhat better than what we get in (6.4) by
the second approach in the same case. We haven’t traced the depedence
that the second approach would give for p 6= q, since this would require the
analyticity bounds of Corollary 4.17 for p 6= q, while our considerations in
Section 4 are most naturally adapted to the case p = q.
Acknowledgements. Both authors were supported by the Academy of Fin-
land through project Nos. 314829 (“Frontiers of singular integrals”) and
346314 (“Finnish Centre of Excellence in Randomness and Structures”). Also,
the second author would like to thank the Foundation for Education and Eu-
ropean Culture (Founders Nicos and Lydia Tricha), Greece, for their financial
support. We would like to thank the anonymous referee for careful reading
and constructive comments that improved the presentation.
This research was carried out while both authors were working at the
University of Helsinki.

References
[1] W. Arendt, C. J.K. Batty, M. Hieber and F. Neubrander. Vector-valued Laplace
Transforms and Cauchy Problems, Monographs in Mathematics, vol. 96. Birkhäuser,
2011.
[2] J. Bergh and J. Löfström. Interpolation Spaces. Berlin, Springer, 1976.
[3] K. J. Engel and R. Nagel. One-Parameter Semigroups for Linear Evolution Equations,
Graduate Texts in Mathematics, vol. 194. Springer, 2000.
[4] S. Fackler. Holomorphic Semigroups and the Geometry of Banach Spaces. Diploma
Thesis, University of Ulm, Fackler-diplom.pdf, 2011.
 BOUNDED HOLOMORPHIC SEMIGROUPS 27

[5] T. Hytönen, J. van Neerven, M. Veraar, and L. Weis. Analysis in Banach spaces. Vol.
I. Martingales and Littlewood–Paley theory. Springer, Cham, 2016.
[6] T. Hytönen and A. Naor. Heat flow and quantitative differentiation. J. Euro. Math.
Soc., 21(11):3415–3466, 2019.
[7] T. Kato. A Characterization of Holomorphic Semigroups. Proc. Amer. Math. Soc.,
25(3):495–498, 1970.
[8] T. Martínez, J.L. Torrea and Q. Xu. Vector-valued Littlewood–Paley–Stein theory
for semigroups. Adv. Math., 203(2):430–475, 2006.
[9] A. Pazy. Semigroups of Linear Operators and Applications to Partial Differential
Equations, Applied Mathematical Sciences, vol. 44. Springer, 1983.
[10] G. Pisier. Martingales with values in uniformly convex spaces. Israel J. Math., 20:326–
350, 1975.
[11] G. Pisier. Holomorphic semigroups and the geometry of Banach spaces. Ann. of Math.,
115(2):375–392, 1982.
[12] G. Pisier. Probabilistic methods in the geometry of Banach spaces. In Probability and
analysis (Varenna, 1985), volume 1206 of Lecture Notes in Math., pages 167–241.
Springer, Berlin, 1986.
[13] G. Pisier. Martingales in Banach spaces, Cambridge Studies in Advanced Mathemat-
ics, vol. 155. Cambridge University Press, 2016.
[14] E. M. Stein. Topics in Harmonic Analysis Related to the Littlewood–Paley theory,
Ann. Math. Studies. Princeton University Press, 1970.
[15] Q. Xu. Littlewood–Paley theory for functions with values in uniformly convex spaces.
J. Reine Angew. Math., 504:195–226, 1998.
[16] Q. Xu. Vector-valued Littlewood–Paley–Stein Theory for Semigroups II. Int. Math.
Res. Not., 21:7769–7791, 2020.
[17] Q. Xu. Holomorphic functional calculus and vector-valued Littlewood–Paley–Stein
theory for semigroups. Preprint, arXiv:2105.12175, 2021.
[18] Q. Xu. Optimal orders of the best constants in the Littlewood–Paley inequalities. J.
Funct. Anal., 283(6), Paper No. 109570, 37 pp., 2022.

T.H. & S.L.: Department of Mathematics and Statistics, P.O.Box 68 (Pietari

Kalmin katu 5), FI-00014 University of Helsinki, Finland
Email address: tuomas.hytonen@helsinki.fi, vlappas@hotmail.com

T.H.: Department of Mathematics and Systems Analysis, Aalto University,

P.O. Box 11100, FI-00076 Aalto, Finland
Email address: tuomas.p.hytonen@aalto.fi

S.L.: Department of Mathematical Analysis, Faculty of Mathematics and

Physics, Charles University, Sokolovská 83, 186 75 Praha 8, Czech Republic
Email address: stefanos.lappas@matfyz.cuni.cz