A.Lunardi - Interpolation Theory

16
APPUNTI
LECTURE NOTES
Alessandra Lunardi
Dipartimento di Scienze Matematiche,
Fisiche e Informatiche
Università di Parma
Parco Area delle Scienze, 53/A
43124 Parma, Italia
Interpolation Theory
Alessandra Lunardi
Interpolation
Theory

c 2018 Scuola Normale Superiore Pisa
Terza edizione
Seconda edizione: 2009
Prima edizione 1999
isbn 978-88-7642-638-4 (eBook)
DOI 10.1007/978-88-7642-638-4
issn 2532-991X (print)
issn 2611-2248 (online)
Contents
Foreword vii
Introduction ix
Nomenclature xiii
1 Real interpolation 1
1.1 The K -method . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Examples . . . . . . . . . . . . . . . . . . . . . 7
1.1.2 Exercises . . . . . . . . . . . . . . . . . . . . . 18
1.2 The trace method . . . . . . . . . . . . . . . . . . . . . 20
1.2.1 Exercises . . . . . . . . . . . . . . . . . . . . . 26
1.3 Dual spaces of real interpolation spaces . . . . . . . . . 28
1.4 Intermediate spaces and reiteration . . . . . . . . . . . . 30
1.4.1 Examples . . . . . . . . . . . . . . . . . . . . . 37
1.4.2 Applications: the theorems of Marcinkiewicz
and Stampacchia; regularity in elliptic PDE’s . . 39
1.4.3 Exercises . . . . . . . . . . . . . . . . . . . . . 42
2 Complex interpolation 45
2.1 Definitions and properties . . . . . . . . . . . . . . . . . 47
2.1.1 Examples . . . . . . . . . . . . . . . . . . . . . 55
2.1.2 The theorems of Hausdorff-Young and Stein . . 58
2.1.3 Exercises . . . . . . . . . . . . . . . . . . . . . 61
3 Interpolation and domains of operators 63

3.1 Operators with rays of minimal growth . . . . . . . . . . 63
3.1.1 Exercises . . . . . . . . . . . . . . . . . . . . . 69
3.2 Two or more operators . . . . . . . . . . . . . . . . . . 70
3.2.1 Exercises . . . . . . . . . . . . . . . . . . . . . 75
vi Alessandra Lunardi
3.2.2 The sum of two commuting operators . . . . . . 75

3.2.3 Exercises . . . . . . . . . . . . . . . . . . . . . 83
4 Powers of positive operators 85

4.1 Definitions and general properties . . . . . . . . . . . . 85
4.1.1 Exercises . . . . . . . . . . . . . . . . . . . . . 94
4.1.2 Powers of nonnegative operators . . . . . . . . . 95
4.1.3 Exercises . . . . . . . . . . . . . . . . . . . . . 98
4.2 Operators with bounded imaginary powers . . . . . . . . 98
4.2.1 The sum of two commuting operators
with bounded imaginary powers . . . . . . . . . 105
4.2.2 Maximal L p regularity for equations
in UMD spaces . . . . . . . . . . . . . . . . . . 110
4.2.3 Exercises . . . . . . . . . . . . . . . . . . . . . 114
4.3 M-accretive operators in Hilbert spaces . . . . . . . . . 115
4.3.1 Self-adjoint operators in Hilbert spaces . . . . . 122
5 Interpolation and semigroups 129

5.1 Real interpolation between Banach spaces and domains
of generators . . . . . . . . . . . . . . . . . . . . . . . 135
5.1.1 Exercises . . . . . . . . . . . . . . . . . . . . . 140
5.2 Examples and applications . . . . . . . . . . . . . . . . 141
5.3 Regularity in elliptic PDE’s by interpolation . . . . . . . 148
5.4 Regularity in evolution equations by interpolation . . . . 156
6 Analytic semigroups and interpolation 161

6.1 Characterization of real interpolation spaces . . . . . . . 162
6.1.1 Exercises . . . . . . . . . . . . . . . . . . . . . 165
6.2 Generation of analytic semigroups by interpolation . . . 165
6.3 Regularity in abstract parabolic equations . . . . . . . . 167
6.3.1 Exercises . . . . . . . . . . . . . . . . . . . . . 176
6.4 An application: space – time Hölder regularity
in parabolic PDE’s . . . . . . . . . . . . . . . . . . . . 177
A The Bochner integral 181

A.1 Integrals over measurable real sets . . . . . . . . . . . . 181
A.2 L p and Sobolev spaces . . . . . . . . . . . . . . . . . . 183
A.3 Weighted L p spaces . . . . . . . . . . . . . . . . . . . . 185
References 187
Index 195
Foreword
This book is the second edition of my 1999 lecture notes of the courses
on interpolation theory that I delivered at the Scuola Normale in 1998
and 1999. Later I held other lectures on interpolation theory in different
universities, and the original lecture notes were polished and enriched.
In the mathematical literature there are many good books on the sub-
ject, but none of them is very elementary, and in many cases the basic
principles are hidden below great generality.
In these lectures I tried to illustrate the principles of interpolation the-
ory aiming at simplification rather than generality. I reduced the abstract
theory as far as possible, and gave many examples and applications, espe-
cially to operator theory and to regularity in partial differential equations.
Moreover the treatment is self-contained, the only prerequisite being the
knowledge of basic functional analysis.
I would like to thank colleagues and friends who followed parts of
my lectures, read older versions of these notes and gave suggestions and
help for improvements: P. Celada, S. Cerrai, Ph. Clément, G. Da Prato,
C. Villani, J. Zabczyk, A. Zaccagnini.
Foreword to the third edition. The third edition is just a revised ver-
sion of the second one: some proofs were modified or added, new refer-
ences were given, and a number of mistakes and misprints were correc-
ted. Many of them were detected by colleagues who used this book for
their lectures, and told me. In particular, I would like to thank H. Abels,
R. Farwig and J. Voigt for their precious remarks.
Introduction
Let X, Y be two real or complex Banach spaces. By X = Y we mean

that X and Y have the same elements and equivalent norms. By Y ⊂ X
we mean that Y is continuously embedded in X. As usual, we denote
by L(X, Y ) (in short, L(X) if X = Y ) the space of all linear bounded
operators from X to Y , endowed by the norm T L(X,Y ) := sup{T xY :
x X = 1}.
The couple of Banach spaces (X, Y ) is said to be an interpolation
couple if both X and Y are continuously embedded in a Hausdorff to-
pological vector space V. In this case the intersection X ∩ Y is a linear
subspace of V, and it is a Banach space under the norm
v X∩Y := max{v X , vY }.
Also the sum X + Y := {x + y : x ∈ X, y ∈ Y } is a linear subspace of

V. It is endowed with the norm
v X+Y := inf x X + yY .

x∈X, y∈Y, x+y=v
As easily seen, X + Y is isometric to the quotient space (X × Y )/D,

where D = {(x, −x) : x ∈ X ∩ Y }. Since V is a Hausdorff space, then
D is closed, and X + Y is a Banach space. Moreover, x X ≥ x X+Y
and yY ≥ y X+Y for all x ∈ X, y ∈ Y , so that both X and Y are
continuously embedded in X + Y .
The space V is used only to guarantee that X + Y is a Banach space. It
will disappear from the general theory.
If (X,Y) is an interpolation couple, an intermediate space is any Banach
space E such that
X ∩ Y ⊂ E ⊂ X + Y.
An interpolation space between X and Y is any intermediate space such
that for every T ∈ L(X) ∩ L(Y ) (that is, for every linear operator T :
x Alessandra Lunardi
X + Y → X + Y whose restriction to X belongs to L(X) and whose

restriction to Y belongs to L(Y )), the restriction of T to E belongs to
L(E). This implies that there is a constant independent of T such that
T L(E) ≤ C max{T L(X) , T L(Y ) ), as next lemma (taken from [63])
shows.
Lemma 0.1. Let (X, Y ) be an interpolation couple, and let E be an in-

terpolation space between X and Y . Then there is C > 0 such that for
each linear operator T : X + Y → X + Y such that T|X ∈ L(X),
T|Y ∈ L(Y ) we have
T L(E) ≤ C max{T L(X) , T L(Y ) }.
Proof. The space B of the linear operators T : X + Y → X + Y such

that T|X ∈ L(X), T|Y ∈ L(Y ) is easily seen to be a Banach space with
the norm T := max{T L(X) , T L(Y ) }. Indeed, if {Tn } is a Cauchy
sequence, Tn|X converges to some TX in L(X), Tn|Y converges to some TY
in L(Y ), the limit operator T : X + Y → X + Y , T (x + y) = T|X x + T|Y y
is well defined (and obviously linear) in X + Y , and Tn → T in B.
Define a linear operator : B → L(E) by (T ) := T|E . We shall
prove that the graph of is closed, so that is bounded and the statement
follows.
Let Tn → T in B be such that (Tn ) → S in L(E) as n → ∞. Then
for each w ∈ X + Y and for each decomposition w = x + y, with x ∈ X
and y ∈ Y , we have
Tn w−T w X+Y ≤ (Tn −T )x X+(Tn −T )yY ≤ Tn −T (x X+yY )
and taking the infimum over all the decompositions of w we get
Tn w − T w X+Y ≤ Tn − T w X+Y
so that limn→∞ Tn w = T w in X + Y . On the other hand, if w ∈ E then

limn→∞ Tn w = Sw in E and hence in X + Y . Therefore, T w = Sw for
each w ∈ E, i.e. S = (T ).
The general interpolation theory is not devoted to characterize all the

interpolation spaces between X and Y but rather to construct suitable
families of interpolation spaces and to study their properties. The most
known and useful families of interpolation spaces are the real interpola-
tion spaces which will be treated in Chapter 1, and the complex interpol-
ation spaces which will be treated in Chapter 2.
xi Interpolation Theory
Interpolation theory has a wide range of applications. We shall em-

phasize applications to partial differential operators and partial differen-
tial equations, referring to [90, 16] for applications to other fields. In par-
ticular we shall give self-contained proofs of optimal regularity results in
Hölder and in fractional Sobolev spaces for linear elliptic and parabolic
differential equations.
The domains of powers of positive operators in Banach spaces are
not interpolation spaces in general, however in some interesting cases
they coincide with suitable complex interpolation spaces. In any case
the theory of powers of positive operators is very close to interpolation
theory, and there are important connections between them. Therefore
in Chapter 5 we give an elementary treatment of the powers of positive
operators, with particular attention to the imaginary powers.
Nomenclature
BUC(Rn ) the space of the uniformly continuous and bounded func-

tions in Rn , page 11
Lip(Rn ) the space of the Lipschitz continuous and bounded func-
(X, Y )θ, p the real interpolation spaces, page 1
(X, Y )θ the continuous interpolation spaces, page 1
B(I ; Y ) the space of the bounded functions from I to Y , page 42
B sp,q (Rn ) the Besov spaces, page 12
Cb1 (Rn ) the space of bounded functions with bounded derivat-
ives in Rn , page 7
C0∞ (Rn ) the space of the smooth functions in Rn with compact
support, page 58
Cbθ () the space of the θ-Hölder continuous and bounded func-
tions in , page 7
Cbθ (Rn ) the space of the θ-Hölder continuous and bounded func-
Cb (Rn ) the space of the bounded continuous functions in Rn ,
page 7
Hf the Hilbert transform of f , page 108
Hε f the truncated Hilbert transform of f , page 108
Jθ (X, Y ) the class Jθ between X and Y , page 30
K (t, x, X, Y ) the K function, page 1
K θ (X, Y ) the class K θ between X and Y , page 30
L p,q () the Lorentz spaces, page 15
p
L ∗ (I ) the L p spaces for the measure dt/t, page 179
p
L ∗ (I ; X) the vector valued L p spaces for the measure
dt/t, page 180
Sθ the complex sector with angle θ, page 75
W 1, p (a, b; X) the vector valued Sobolev spaces, page 178
W α, p (a, b; X) the vector valued fractional Sobolev spaces, page 179
xiv Alessandra Lunardi
W θ, p () the fractional Sobolev spaces in an open set , page 18

W θ, p (Rn ) the fractional Sobolev spaces in Rn , page 7
[X, Y ]θ the complex interpolation spaces, page 49
BMO() the space of functions with bounded mean oscillation,
page 39
R(A) the range of A, page 96
Chapter 1
Real interpolation
Let (X, Y ) be a real or complex interpolation couple.

p
If I is any interval contained in (0, +∞), L ∗ (I ) is the Lebesgue space
L p with respect to the measure dt/t in I . In particular, L ∞ ∞
∗ (I ) = L (I ).
See Appendix, Section 2.
1.1. The K -method

Definition 1.1. For every x ∈ X + Y and t > 0, set
K (t, x, X, Y ) := inf (a X + tbY ) . (1.1)
x=a+b, a∈X, b∈Y
If there is no danger of confusion, we shall write K (t, x) instead of

K (t, x, X, Y ).
Note that K (1, x) = x X+Y , and for every t > 0 K (t, ·) is a norm
in X + Y , equivalent to the norm of X + Y . Now we define a family of
normed spaces by means of the function K .
Definition 1.2. Let 0 < θ < 1, 1 ≤ p ≤ ∞, and set

(X, Y )θ, p := {x ∈ X + Y : t → t −θ K (t, x) ∈ L ∗p (0, +∞)},
(1.2)
x −θ
(X,Y )θ, p := t K (t, x) L ∗p (0,∞) ;
(X, Y )θ := {x ∈ X + Y : lim+ t −θ K (t, x) = lim t −θ K (t, x) = 0}. (1.3)

t→0 t→+∞
Such spaces are called real interpolation spaces.

The mapping x → x(X,Y )θ, p is easily seen to be a norm in (X, Y )θ, p .
If no confusion may arise, we shall write xθ, p instead of x(X,Y )θ, p .
Since t → K (t, x) is continuous in (0, ∞) for each x ∈ X + Y
(see Exercise 1, Section 1.1.2), then (X, Y )θ is contained in (X, Y )θ,∞ .
Moreover, it is easy to see that (X, Y )θ is a closed subspace of (X, Y )θ,∞ .
It is usually endowed with the (X, Y )θ,∞ -norm.
The spaces (X, Y )θ are also called continuous interpolation spaces.
2 Alessandra Lunardi
Note that K (t, x, X, Y ) = t K (t −1 , x, Y, X) for each t > 0. By the

p
transformation τ = t −1 , which preserves L ∗ (0, ∞), we get
(X, Y )θ, p = (Y, X)1−θ, p , 0 < θ < 1, 1 ≤ p ≤ ∞, (1.4)
and
(X, Y )θ = (Y, X)1−θ . (1.5)
So, pay attention to the order!
Let us consider some particular cases.
(a) Let X = Y . Then X + Y = X, and K (t, x) ≤ min{t, 1}x. There-

fore
X ⊂ (X, X)θ, p , 0 < θ < 1, 1 ≤ p ≤ ∞.
In the next proposition we will see that for any interpolation couple
we have (X, Y )θ, p ⊂ X + Y . So, if X = Y then (X, X)θ, p = X.
(b) If X ∩Y = {0}, then for each x ∈ X +Y there are a unique a ∈ X and
a unique b ∈ Y such that x = a + b, hence K (t, x) = a X + tbY
p
and t → t −θ K (t, x) does not belong to any L ∗ (0, ∞), unless x = 0.
Therefore, (X, Y )θ, p = (X, Y )θ = {0} for every p ∈ [1, ∞], θ ∈
(0, 1).
(c) In the important case where Y ⊂ X we have K (t, x) ≤ x X for
p
every x ∈ X, so that t → t −θ K (t, x) ∈ L ∗ (a, ∞) for all a > 0, and
limt→+∞ t −θ K (t, x) = 0. Therefore, only the behavior near t = 0 of
t −θ K (t, x) plays a role in the definition of (X, Y )θ, p and of (X, Y )θ .
Indeed, one could replace the half-line (0, +∞) by any interval (0, a)
in Definition 1.2, obtaining equivalent norms.
The inclusion properties of the real interpolation spaces are stated below.
Proposition 1.3. For 0 < θ < 1, 1 ≤ p1 ≤ p2 ≤ ∞ we have
X ∩Y ⊂ (X, Y )θ, p1 ⊂ (X, Y )θ, p2 ⊂ (X, Y )θ ⊂ (X, Y )θ,∞ ⊂ X + Y. (1.6)
Moreover,
(X, Y )θ,∞ ⊂ X ∩ Y ,
where X, Y are the closures of X, Y in X + Y .
Proof. As a first step, we show that (X, Y )θ,∞ is contained in X ∩ Y and

that it is continuously embedded in X + Y . For x ∈ (X, Y )θ,∞ we have
K (t, x) = inf a X + tbY ≤ t θ xθ,∞ , t > 0,

x=a+b
3 Interpolation Theory
so that for every n ∈ N (taking t = 1/n) there are an ∈ X, bn ∈ Y such

that x = an + bn , and
1 2
an X + bn Y ≤ θ xθ,∞ .
n n
In particular, x − bn X+Y = an X+Y ≤ an X ≤ 2xθ,∞ n −θ , so that
the sequence {bn } goes to x in X + Y as n → ∞. This implies that
(X, Y )θ,∞ is contained in Y . Arguing similarly (i.e., replacing 1/n by
n and letting n → ∞), or else recalling that (X, Y )θ,∞ = (Y, X)1−θ,∞ ,
we see that (X, Y )θ,∞ is contained also in X. Moreover, by definition
x X+Y = K (1, x). Therefore
x X+Y = K (1, x) ≤ xθ,∞ , ∀x ∈ (X, Y )θ,∞ ,
so that (X, Y )θ,∞ is continuously embedded in X + Y .

We already remarked that the inclusion (X, Y )θ ⊂ (X, Y )θ,∞ is obvi-
ous because K (·, x) is continuous so that t −θ K (t, x) is bounded in every
interval [a, b] with 0 < a < b.
Let us show that (X, Y )θ, p is contained in (X, Y )θ and it is continu-
ously embedded in (X, Y )θ,∞ for p < ∞. For each x ∈ (X, Y )θ, p and
t > 0, recalling that K (·, x) is increasing we get
+∞ 1/ p
−θ −θ p−1
t K (t, x) = (θ p) 1/ p
s ds K (t, x)
t
1/ p (1.7)
+∞
−θ p−1
≤ (θ p)1/ p
s K (s, x) ds p
, t > 0.
t
The right hand side is bounded by (θ p)1/ p xθ, p .Therefore x ∈ (X,Y )θ,∞ ,
and xθ,∞ ≤ (θ p)1/ p xθ, p . Changing θ with 1 − θ and X with Y we
obtain xθ,∞ ≤ ((1 − θ) p)1/ p xθ, p . Therefore,
xθ,∞ ≤ (min{θ, 1 − θ} p)1/ p xθ, p . (1.8)
Moreover letting t → ∞ we get limt→∞ t −θ K (t, x) = 0. To prove that

x ∈ (X, Y )θ we need also that limt→0 t −θ K (t, x) = 0. This can be seen
as follows: since (X, Y )θ, p = (Y, X)1−θ, p then
0 = lim t −(1−θ) K (t, x, Y, X)

t→+∞
= lim t θ K (t −1 , x, X, Y )
t→+∞
= lim+ τ −θ K (τ, x, X, Y ).
τ →0
Let us prove that (X, Y )θ, p1 ⊂ (X, Y )θ, p2 for p1 < p2 < +∞. For
x ∈ (X, Y )θ, p1 we have
+∞ 1/ p2
−θ p2 p2 dt
xθ, p2 = t K (t, x)
0 t
+∞ 1/ p2
−θ p1 p1 −θ p2 − p1 dt
= t K (t, x) (t K (t, x))
0 t
+∞ 1/ p2 ( p2 − p1 )/ p2
−θ p1 p1 dt −θ
≤ t K (t, x) sup t K (t, x)
0 t t>0
= (xθ, p1 ) p1 / p2 (xθ,∞ )1− p1 / p2 ,

and using (1.8) we find
xθ, p2 ≤ [min{θ, 1 − θ} p1 ]1/ p1 −1/ p2 xθ, p1 . (1.9)
Finally, from the inequality K (t, x) ≤ min{1, t} x X∩Y for every x ∈
X ∩ Y it follows immediately that X ∩ Y is continuously embedded in
(X, Y )θ, p for 0 < θ < 1, 1 ≤ p ≤ ∞.
The statement is so completely proved.
The first part of the proof of Proposition 1.3 shows the connection
between interpolation theory and approximation theory. Indeed, the se-
quence bn in the proof consists of elements of Y and converges to x
in X + Y . The rate of convergence of bn and the rate of blowing up
of bn Y are described precisely by the fact that x ∈ (X, Y )θ,∞ (or
x ∈ (X, Y )θ, p ⊂ (X, Y )θ,∞ ). In particular, if x ∈ (X, Y )θ,∞ then
x − bn X+Y ≤ const. n −θ , and bn Y ≤ const. n 1−θ .
If Y ⊂ X other embeddings hold.
Proposition 1.4. If Y ⊂ X, for 0 < θ1 < θ2 < 1 we have
(X, Y )θ2 ,∞ ⊂ (X, Y )θ1 ,1 . (1.10)
Therefore, (X, Y )θ2 , p ⊂ (X, Y )θ1 ,q for every p, q ∈ [1, ∞].
Proof. For x ∈ (X, Y )θ2 ,∞ we have, using the inequalities K (t, x) ≤
x X for t ≥ 1 and K (t, x) ≤ t θ2 xθ2 ,∞ for 0 < t ≤ 1,
1 +∞
−θ1 −1
xθ1 ,1 = t K (t, x)dt + t −θ1 −1 K (t, x)dt
0 1
1 +∞
−θ1 −1 θ2
≤ t xθ2 ,∞ t dt + t −θ1 −1 x X dt (1.11)
0 1
1 1
≤ xθ2 ,∞ + x X ,
θ2 − θ1 θ1
and the statement follows since (X, Y )θ2 ,∞ ⊂ X + Y = X because

Y ⊂ X.
Note that (1.10) is not true in general. See next Example 1.11.
Proposition 1.5. For all θ ∈ (0, 1), p ∈ [1, ∞], (X, Y )θ, p is a Banach
space. For all θ ∈ (0, 1), (X, Y )θ is a Banach space, endowed with the
norm of (X, Y )θ,∞ .
Proof. Let {xn }n∈N be a Cauchy sequence in (X, Y )θ, p . By the continuous
embedding (X, Y )θ, p ⊂ X + Y , {xn }n∈N is a Cauchy sequence in X + Y
too, so that it converges to an element x ∈ X + Y .
Let us estimate x n − xθ, p . Fix ε > 0, and let xn − xm θ, p ≤ ε for
n, m ≥ n ε . Since y → K (t, y) is a norm in X + Y , for every n, m ∈ N
and t > 0 we have K (t, xn − x) ≤ K (t, xn − xm ) + K (t, xm − x), so that
t −θ K (t, xn −x) ≤ t −θ K (t, xn −xm )+t −θ max{t, 1}xm −x X+Y . (1.12)
Let p = ∞. Then for every t > 0 and n, m ≥ n ε
t −θ K (t, xn − x) ≤ ε + t −θ max{t, 1}xm − x X+Y .
Letting m → +∞, we find t −θ K (t, xn − x) ≤ ε for every t > 0. This

implies that x ∈ (X, Y )θ,∞ and that xn → x in (X, Y )θ,∞ . Therefore
(X, Y )θ,∞ is complete.
It is easy to see that (X, Y )θ is a closed subspace of (X, Y )θ,∞ . Since
(X, Y )θ,∞ is complete, then also (X, Y )θ is complete.
Let now p < ∞. Then
1/δ 1/ p
−θ p−1
xn − xθ, p = lim t K (t, xn − x) dt
p
.
δ→0 δ
Due again to (1.12), for every δ ∈ (0, 1) we get, for n, m ≥ n ε ,

1/δ 1/ p
−θ p p dt
t K (xn − x) ≤ xn − xm θ, p
δ t
1/δ 1/ p
−θ p dt
+xm −x X+Y t max{t, 1} ≤ ε+C(δ, p)xm − x X+Y .
δ t
Letting first m → ∞ and then δ → 0 we get x ∈ (X, Y )θ, p and xn → x
in (X, Y )θ, p . So, (X, Y )θ, p is complete.
The spaces (X, Y )θ, p and (X, Y )θ are interpolation spaces, as a con-
sequence of the following important theorem.
Theorem 1.6. Let (X 1 , Y1 ), (X 2 , Y2 ) be interpolation couples. If T ∈

L(X 1 ,X 2 )∩L(Y1 ,Y2 ), then T ∈ L((X 1 ,Y1 )θ, p , (X 2, Y2 )θ, p )∩L((X 1 ,Y1 )θ ,
(X 2 ,Y2 )θ ) for every θ ∈ (0, 1) and p ∈ [1, ∞]. Moreover,
T L((X 1 ,Y1 )θ, p ,(X 2 ,Y2 )θ, p ) ≤ (T L(X 1 ,X 2 ) )1−θ (T L(Y1 ,Y2 ) )θ . (1.13)
Proof. If T is the null operator, there is nothing to prove. If T = 0, either

T L(X 1 ,X 2 ) = 0 or T L(Y1 ,Y2 ) = 0. Assume that T L(X 1 ,X 2 ) = 0. Let
x ∈ (X 1 , Y1 )θ, p : then for every a ∈ X 1 , b ∈ Y1 such that x = a + b and
for every t > 0 we have

T L(Y1 ,Y2 )
T a X 2 + tT bY2 ≤ T L(X 1 ,X 2 ) a X 1 + t bY1 ,
T L(X 1 ,X 2 )
so that, taking the infimum over all a, b, we get

T L(Y1 ,Y2 )
K (t,T x,X 2 ,Y2 ) ≤ T L(X 1 ,X 2 ) K t , x,X 1 ,Y1 , t > 0. (1.14)
T L(X 1 ,X 2 )
T
L(Y1 ,Y2 )
Setting s = t T L(X ,X )
we get T x ∈ (X 2 , Y2 )θ, p , and
1 2
θ
T L(Y1 ,Y2 )
T x(X 2 ,Y2 )θ, p ≤ T L(X 1 ,X 2 ) x(X 1 ,Y1 )θ, p ,
T L(X 1 ,X 2 )
and (1.13) follows. From (1.14) it follows also that
limt→0 t −θ K (t, x, X 1 , Y1 ) = limt→∞ t −θ K (t, x, X 1 , Y1 ) = 0 ⇒
⇒ limt→0 t −θ K (t, T x, X 2 , Y2 ) = limt→∞ t −θ K (t, T x, X 2 , Y2 ) = 0,
that is, T maps (X 1 , Y1 )θ into (X 2 , Y2 )θ .

In the case where T L(X 1 ,X 2 ) = 0 we get the result either replacing
everywhere T L(X 1 ,X 2 ) by ε > 0 and then letting ε → 0, or else repla-
cing X i by Yi for i = 1, 2 and θ by 1 − θ (see (1.4) and (1.5)).
In the papers [17, 27] it was shown that if in addition T : X 1 → X 2 , or

T : Y1 → Y2 is compact, then T : (X 1 , Y1 )θ, p → (X 2 , Y2 )θ, p is compact
for every θ ∈ (0, 1) and p ∈ [1, +∞).
Taking X 1 = X 2 = X, Y1 = Y2 = Y , it follows that (X, Y )θ, p and
(X, Y )θ are interpolation spaces. Another important consequence is the
next corollary.
Corollary 1.7. Let (X, Y ) be an interpolation couple. For 0 < θ < 1,

1 ≤ p ≤ ∞ there is c(θ, p) such that
θ
y(X,Y )θ, p ≤ c(θ, p)y1−θ
X yY , y ∈ X ∩ Y. (1.15)
Proof. Set K = R or K = C, according to the fact that X, Y are real

or complex Banach spaces. Let y ∈ X ∩ Y , and define T by T (λ) =
λy for each λ ∈ K. Then T L(K,X) = y X , T L(K,Y ) = yY ,
and T L(K,(X,Y )θ, p ) = y(X,Y )θ, p . The statement follows now taking
X 1 = Y1 = K and X 2 = X, Y2 = Y in Theorem 1.6, and recalling that
(K, K)θ, p = K.
Another more direct proof is the following: for y ∈ X ∩ Y \ {0}, we
have K (t, y) ≤ min{y X , tyY }, so that
y X
t≤ ⇒ K (t,y) ≤ tyY ⇒ t −θ K (t,y) ≤ t 1−θ yY ≤ y1−θ θ
X yY ,
yY
and
y X
t≥ ⇒ K (t, y) ≤ y X
yY

yY θ
⇒ t −θ K (t, y) ≤ y X = y1−θ θ
X yY .
y X
θ
Therefore y(X,Y )θ,∞ = supt>0 t −θ K (t, y) ≤ y1−θ
X yY , and the state-
ment follows for p = +∞ with constant c(θ, ∞) = 1. For p < +∞ we
already know that (X, Y )θ, p is continuously embedded in (X, Y )θ,∞ , and
the proof is complete.
1.1.1. Examples
Let us see some basic examples. Cb (Rn ) is the space of the bounded con-
tinuous functions in Rn , endowed with the sup norm · ∞ ; Cb1 (Rn ) is the
subset of the continuously differentiablefunctions with bounded deriv-
n
atives, endowed with the norm f ∞ + i=1 Di f ∞ . For θ ∈ (0, 1),
θ
Cb (R ) is the set of the bounded and uniformly θ-Hölder continuous
n
functions, endowed with the norm
| f (x) − f (y)|
f Cbθ = f ∞ + [ f ]C θ = f ∞ + sup .
x= y |x − y|θ
For θ ∈ (0, 1), p ∈ [1, ∞), W θ, p (Rn ) is the space of all f ∈ L p (Rn )
such that
1/ p
| f (x) − f (y)| p
[ f ]W θ, p = dx dy < ∞.
Rn ×Rn |x − y|θ p+n
It is endowed with the norm · L p + [ · ]W θ, p .

Example 1.8. For 0 < θ < 1, 1 ≤ p < ∞ we have
(Cb (Rn ), Cb1 (Rn ))θ,∞ = Cbθ (Rn ), (1.16)
(L p (Rn ), W 1, p (Rn ))θ, p = W θ, p (Rn ), (1.17)

with equivalence of the respective norms.
Proof. Let us prove the first statement. Let f ∈ (Cb (Rn ), Cb1 (Rn ))θ,∞ .
Since (X, Y )θ,∞ ⊂ X + Y , we have f ∞ ≤ const. f θ,∞ . In our case,
since · ∞ ≤ · Cb1 , the constant is 1. Indeed, for every decomposition
f = a + b, with a ∈ Cb (Rn ), b ∈ Cb1 (Rn ) we have f ∞ ≤ a∞ +
b∞ , so that
f ∞ ≤ K (1, f, Cb (Rn ), Cb1 (Rn )) ≤ f θ,∞ .
Moreover for x = y and again for every decomposition f = a + b, with

a ∈ Cb (Rn ), b ∈ Cb1 (Rn ), we have
| f (x) − f (y)| ≤ |a(x) − a(y)| + |b(x) − b(y)| ≤ 2a∞ + bCb1 |x − y|,
so that, taking the infimum over all the decompositions,
| f (x) − f (y)| ≤ 2K (|x − y|, f, Cb (Rn ), Cb1 (Rn )) ≤ 2|x − y|θ f θ,∞ .
Therefore f is θ-Hölder continuous and f Cbθ = f ∞ + [ f ]C θ ≤

3 f θ,∞ .
Conversely, let f ∈ Cbθ (Rn ). Let ϕ ∈ C ∞ (R
) be a nonnegative func-
n
tion with support in the unit ball, such that Rn ϕ(x)dx = 1. For every
t > 0 set

1 x−y
bt (x) = n f (y)ϕ dy, ; at (x) = f (x)−bt (x), x ∈ Rn . (1.18)
t Rn t
Then
1 y
at (x) = n ( f (x) − f (x − y))ϕ dy
t Rn t
so that

1 θ θ
at ∞ ≤ [ f ]C θ n |y| ϕ(y/t)dy = t [ f ]C θ |w|θ ϕ(w)dw.
t Rn Rn
Moreover, bt ∞ ≤ f ∞ , and for each i = 1, . . . , n

1 x−y
Di bt (x) = n+1 f (y)Di ϕ dy, x ∈ Rn .
t Rn t

Since Rn Di ϕ((x − y)/t)dy = 0, we get

1 y
Di bt (x) = n+1 ( f (x − y) − f (x))Di ϕ dy, x ∈ Rn , (1.19)
t Rn t
which implies

θ−1
Di bt ∞ ≤ t [ f ]C θ |w|θ |Di ϕ(w)|dw.
Rn
Therefore,
t −θ K (t, f ) ≤ t −θ (at ∞ + tbt Cb1 ) ≤ C f C θ , 0 < t ≤ 1.
For t ≥ 1 (see remark (c) after Definition 1.1), we can take at = f ,

bt = 0 which implies
t −θ K (t, f ) ≤ t −θ f ∞ ≤ f ∞ , t ≥ 1.
The embedding Cbθ (Rn ) ⊂ (Cb (Rn ), Cb1 (Rn ))θ,∞ follows.
The proof of the second statement is similar. We recall that for every
b ∈ W 1, p (Rn ) and h ∈ Rn \ {0} we have (see any textbook about Sobolev
spaces, e.g., [12, Chapter X])
1/ p
|b(x + h) − b(x)| p
dx ≤ |Db| L p ,
Rn |h|
where we set as usual Db = (bx1 , . . . , bxn ).
For every f ∈ (L p (Rn ), W 1, p (Rn ))θ, p and h ∈ Rn let a = a(h) ∈
L (Rn ), b = b(h) ∈ W 1, p (Rn ) be such that f = a + b, and
p
a L p + |h| bW 1, p ≤ 2K (|h|, f ).
Then

| f (x + h) − f (x)| p p−1 |a(x + h) − a(x)|
p
|b(x + h) − b(x)| p
≤2 +
|h|θ p+n |h|θ p+n |h|θ p+n
so that

| f (x + h) − f (x)| p
dx
Rn |h|θ p+n

|a(x + h) − a(x)| p |b(x + h) − b(x)| p
≤2 p−1
+ dx
Rn |h|θ p+n |h|θ p+n
p p
a L p p−1 |h| |Db| L p
p
≤ 22 p−1 + 2
|h|θ p+n |h|θ p+n
≤ C p |h|−θ p−n (a L p + |h| bW 1, p ) p

p
≤ C p |h|−θ p−n K (|h|, f ) .
It follows that

| f (x + h) − f (x)| p
dx dh ≤ C p |h|−θ p−n K (|h|, f ) p dh
Rn ×Rn |h|θ p+n Rn
∞
K (r, f ) p
= Cp dr dσn−1
0 r θ p+1 ∂ B(0,1)
p
= C p,n f θ, p .
Therefore,(L p (Rn ),W 1, p (Rn ))θ, p is continuously embedded inW θ, p (Rn ).

To be precise, we have estimated so far only the seminorm [ f ]θ, p . But we
already know that each (X, Y )θ, p is continuously embedded in X + Y ; in
our case X + Y = X = L p (Rn ) so that we have also f L p ≤ C f θ, p .
To prove the other embedding, for each f ∈ W θ, p (Rn ) define at and
bt by (1.18). Then
p
p 1 x−y
at L p = | f (y) − f (x)| n ϕ dy dx
Rn Rn t t

1 x−y
≤ | f (y) − f (x)| p n ϕ dy dx.
Rn Rn t t
where for p > 1 we applied the Hölder inequality to the product
| f (x) − f (y)|(t −n ϕ((x − y)/t))1/ p · (t −n ϕ((x − y)/t))1−1/ p .

So, we get
∞
p dt
t −θ p at L p
t
0
∞
1 x−y dt
≤ t −θ p | f (y) − f (x)| p ϕ dydx
0 Rn ×Rn tn t t
∞
−θ p 1 x − y dt
= | f (y) − f (x)| p
t ϕ dy dx
Rn ×Rn 0 tn t t
∞
−θ p 1 x − y dt
= | f (y) − f (x)| p
t ϕ dy dx
Rn ×Rn |x−y| tn t t

ϕ∞ | f (y) − f (x)| p p
≤ dy dx = C[ f ]W θ, p .
θp + n Rn ×Rn |y − x|θ p+n
Using (1.19) and arguing similarly, we get

∞ p−1
(1−θ) p p dt C Di ϕ∞ p
t Di bt L p ≤ i [ f ]W θ, p ,
0 t θp + n

with Ci = Rn |Di ϕ(y)|dy. Moreover, bt L p ≤ f L p ϕ L 1 = f L p ,
so that 1
dt 1
t (1−θ) p bt L p
p p
≤ f L p .
0 t (1 − θ) p
p
Therefore, t K (t, f, L p , W 1, p ) ≤ t −θ at L p +t 1−θ bt W 1, p ∈ L ∗ (0, 1),
−θ
with norm estimated by C f W θ, p . Since W 1, p (Rn ) ⊂ L p (Rn ), this is

enough to guarantee that f ∈ (L p (Rn , W 1, p (Rn ))θ, p (see again remark
(c) after Definition 1.1) and the second part of the statement follows.
The present proof is a bit long, but essentially elementary and self-
contained. We will see a much shorter proof in Section 5.2, as an applic-
ation of the theory developed in Chapter 5.
Note that the proof of (1.16) yields also
(L ∞ (Rn ), Lip(Rn ))θ,∞ = (BUC(Rn ), BUC1 (Rn ))θ,∞ = Cbθ (Rn ),
where we use obvious notation: Lip(Rn ) is the space of the Lipschitz

continuous and bounded functions, endowed with its natural norm f →
f ∞ +supx= y | f (x)− f (y)|/|x − y|, and BUC(Rn ), BUC1 (Rn ), are the
subspaces of Cb (Rn ), Cb1 (Rn ) consisting of uniformly continuous func-

tions (resp., functions with uniformly continuous first order derivatives).
Characterizations similar to (1.16) and (1.17) hold if Rn is replaced by
a good open set ⊂ Rn . Below we give the details of the generalization
of (1.16), leaving the generalization of (1.17) as an exercise.
If is any open set in Rn , the Hölder spaces Cbθ (), 0 < θ < 1, are
defined as in the case = Rn : they consist of the bounded functions f
such that
| f (x) − f (y)|
[ f ]C θ := sup < +∞,
x,y∈; x = y |x − y|θ
and they are endowed with the norm · ∞ + [ · ]C θ .
Example 1.9. Let ∈ Rn be an open set with the following property:
there exists an extension operator E such that E ∈ L(Cb (), Cb (Rn ))
∩ L(Cb1 (), Cb1 (Rn )), and also, for some θ ∈ (0, 1), E ∈ L(Cbθ (),
Cbθ (Rn )) (by extension operator we mean that E f | = f (x), for all f ∈
Cb ()). Then
(Cb (), Cb1 ())θ,∞ = Cbθ ().
Proof. Theorem 1.6 implies that
E ∈ L((Cb (), Cb1 ())θ,∞ , (Cb (Rn ), Cb1 (Rn ))θ,∞ ).
We know already that (Cb (Rn ), Cb1 (Rn ))θ,∞ = Cbθ (Rn ). Therefore, for
every f ∈ (Cb (), Cb1 ())θ,∞ the extension E f is in Cbθ (Rn ) and
E f Cbθ (Rn ) ≤ C f (Cb (),Cb1 ())θ,∞ .
Since f = E f | , then f ∈ Cbθ () and f Cbθ () ≤ C f (Cb (),Cb1 ())θ,∞ .
Conversely, if f ∈ Cbθ () then E f ∈ Cbθ (Rn ) = (Cb (Rn ), Cb1 (Rn ))θ,∞ .
The retraction operator Rg = g| belongs obviously to
L(Cb (Rn ), Cb ()) ∩ L(Cb1 (Rn ), Cb1 ()).
Again by Theorem 1.6, f = R(E f ) belongs to (Cb (), Cb1 ())θ,∞ , with
norm not exceeding CE f Cbθ (Rn ) ≤ C f Cbθ () .
Such a good extension operator exists if is an open set with uni-

formly C 1 boundary. ∂ is said to be uniformly C 1 if there are N ∈ N
and a (at most) countable set of balls Bk whose interior parts cover ∂,
such that the intersection of more than N of these balls is empty, and dif-
feomorphisms ϕk : Bk → B(0, 1) ⊂ Rn such that ϕk (Bk ∩ ) = {y ∈
B(0, 1) : yn ≥ 0}, and ϕk C 1 + ϕk−1 C 1 are bounded by a constant
independent of k. (In particular, each bounded with C 1 boundary has

uniformly C 1 boundary).
It is sufficient to construct E when = Rn+ . The construction of E
for any open set with uniformly C 1 boundary will follow by the usual
method of locally straightening the boundary and using a partition of 1.
If = Rn+ = {x = (x , xn ) ∈ Rn : xn > 0} we may use the reflection
method: we set

 f (x), xn ≥ 0,
E f (x) =

α1 f (x , −xn ) + α2 f (x , −2xn ), xn < 0,
where α1 , α2 satisfy the continuity condition α1 + α2 = 1 and the differ-

entiability condition −α1 − 2α2 = 1, that is α1 = 3, α2 = −2.
Then E belongs to
L(Cb (Rn+ ), Cb (Rn )) ∩ L(Cb1 (Rn+ ), Cb1 (Rn )) ∩ L(Cbθ (Rn+ ), Cbθ (Rn ))
for every θ ∈ (0, 1).

For another important generalization of Example 1.8 we introduce the
Besov spaces B sp,q (Rn )
B sp,q (Rn ) is defined as follows. If s is not an integer, let [s] and {s}
be the integral and the fractional parts of s, respectively. For p, q < ∞,
B sp,q (Rn ) consists of the functions f ∈ W [s], p (Rn ) such that
dh
q/ p 1/q
α α
[ f ] B sp,q := |D f (x +h)−D f (x)| d x p
|α|=[s] Rn |h|n+{s}q Rn
is finite (for [s] = 0, we set as usual W 0, p (Rn ) = L p (Rn )). In particular,

for p = q we have B sp, p (Rn ) = W s, p (Rn ).
For q = ∞, B sp,∞ (Rn ) consists of the functions f ∈ W [s], p (Rn ) such
that
1/ p
1 α α
[ f ] B sp,∞ := sup |D f (x + h) − D f (x)| dx p
|α|=[s] h∈R \{0}

n |h|{s} Rn
is finite.
For p = +∞ the L p norms of D α f (· + h) − D α f are replaced by sup
s
norms. In particular, B∞,∞ (Rn ) = Cbs (Rn ).
If s = k ∈ N, for p, q < ∞ the space B kp,q (Rn ) consists of the

functions f ∈ W k−1, p (Rn ) such that

[ f ] B kp,q :=
|α|=[s]−1
q/ p1/q
dh α α α
|D f (x +2h)−2D f (x +h)+ D f (x)| dx p
Rn |h|n+q Rn
is finite. For p = +∞ or q = +∞ we modify as above. In particular,

1
B∞,∞ (Rn ) :=

| f (x + 2h) − 2 f (x + h) + f (x)|
f ∈ Cb (R ) : suph=0, x∈Rn
n
<∞ .
|h|
It is easy to see that any Lipschitz continuous and bounded function be-
1
longs to B∞,∞ (Rn ), but the elements of B∞,∞
1
(Rn ) are not necessarily
Lipschitz continuous.
Example 1.10. For 0 < θ < 1, m ∈ N we have
(Cb (Rn ), Cbk (Rn ))θ,∞ = B∞,∞
mθ
(Rn ).
In particular, if mθ is not integer,
(Cb (Rn ), Cbk (Rn ))θ,∞ = Cbmθ (Rn ).
For 1 ≤ p, q < ∞, m ∈ N we have
(L p (Rn ), W m, p (Rn ))θ,q = B mθ
p,q (R ).
n
A part of these statements will be proved in Chapter 5, with the aid of

semigroup theory (Examples 5.14, 5.15). Some particular cases are left
as exercises in this chapter (Example 6, Section 1.1.2, Example 1, Sec-
tion 1.4.3). For the complete proof see [90, Section 2.3, 2.4] for p > 1,
and [50] for p = 1.
Similar results hold if Rn is replaced by a regular domain ⊂ Rn . The
Besov spaces B sp,q () are defined as the restrictions to of the functions
belonging to B sp,q (Rn ). See again [50, 90].
Now we turn to real interpolation between L p spaces. We will obtain
Lorentz spaces.
Let (, µ) be a σ -finite measure space. To define the Lorentz spaces
L p,q () we introduce the rearrangements, as follows. For every measur-
able f : → R or f : → C set
m(σ, f ) = µ{x ∈ : | f (x)| > σ }, σ ≥ 0,
and
f ∗ (t) = inf{σ : m(σ, f ) ≤ t}, t ≥ 0.
Both m(·, f ) and f ∗ are nonnegative, decreasing (i.e. nonincreasing),
right continuous, and f ∗ , | f | are equi-measurable, that is for each σ0 > 0
we have
|{t > 0 : f ∗ (t) > σ0 }| = m(σ0 , f ) = µ{x ∈ : | f (x)| > σ0 },
and consequently |{t > 0 : f ∗ (t) ∈ [σ1 , σ2 ]}| = µ{x ∈ : | f (x)| ∈
[σ1 , σ2 ]}, etc. Therefore, setting E := {x ∈ : | f (x)| > σ }, with
σ > 0, we have
µ(E)
∗
| f (x)|µ(dx) = f (t)dt = f ∗ (t)dt,
E {t>0: f ∗ (t)>σ } 0
and ∞
| f (x)| µ(dx) =
p
( f ∗ (t)) p dt, p ≥ 1, (1.20)
0
sup ess | f | = f (0) = sup ess f ∗ .
∗
(1.21)
f ∗ is called the nonincreasing rearrangement of f onto (0, ∞).
The Lorentz spaces L p,q () (1 ≤ p ≤ ∞, 1 ≤ q ≤ ∞) are defined
by

L () := f ∈ L 1 ()+ L ∞ () : f L p,q
p,q
∞ 1/q (1.22)
1/ p ∗ q dt
:= (t f (t)) <∞ ,
0 t
for q < ∞, and
L p,∞ () := { f ∈ L 1 () + L ∞ () : f L p,∞
(1.23)
:= sup t 1/ p f ∗ (t) < ∞}.
t>0
(For p = ∞ we set as usual 1/∞ = 0). By the definition of f ∗ we have

also

∞

1/ p
p,∞
() = f ∈ L () + L () : sup σ m(σ, f )
1
< ∞ . (1.24)
σ >0
The spaces L p,∞ () are also called Marcinkiewicz spaces.

Note that in general · L p,q is not a norm but only a quasi-norm, i.e. the
triangle inequality is replaced by f + g ≤ C( f + g). Moreover,
by (1.20),
L p, p () = L p (), 1 ≤ p ≤ ∞.
Example 1.11. Let (, µ) be a σ -finite measure space.

Then (L 1 (), L ∞ ()) is an interpolation couple. For 0 < θ < 1,
1 ≤ q ≤ ∞ we have
1
(L 1 (), L ∞ ())θ,q = L 1−θ ,q (). (1.25)
Proof. Let V be the space of all measurable, a.e. finitely valued (real
or complex) functions defined in . V is a linear topological Hausdorff
space under convergence in measure on each measurable E ⊂ with fi-
nite measure µ(E). Both L 1 () and L ∞ () are continuously embedded
in V. Therefore (L 1 (), L ∞ ()) is an interpolation couple.
The proof of (1.25) relies on the equality
t
∞
K (t, f, L (), L ()) =
1
f ∗ (s)ds, t > 0. (1.26)
0
Once (1.26) is established, (1.25) follows easily. Indeed, since f ∗ is

decreasing then K (t, f ) ≥ t f ∗ (t), so that for q < ∞
∞ ∞
−θq q dt dt
t −θq+q f ∗ (t)q
−θ q
t K (t, f ) L q = t K (t, f ) ≥
∗
0 t 0 t
q
= f L 1/(1−θ),q () ,
and similarly, for q = ∞
sup t −θ K (t, f ) ≥ sup t 1−θ f ∗ (t) = f L 1/(1−θ),∞ () .

t>0 t>0
The opposite inequality follows from the Hardy-Young inequality (A.10)

(i) for q < ∞:
∞ t
−θ q −θq ∗ ds q dt
t K (t, f ) L q = t s f (s)
∗
0 0 s t

1 ∞ (1−θ)q ∗ ds q
≤ q s ( f (s))q = f L 1/(1−θ),q () ,
θ 0 s
and from the obvious inequality

t
−θ −θ ds 1−θ ∗ 1
t K (t, f ) L ∞ ≤ t s f (s) L ∞ = f L 1/(1−θ),∞ ()
0 s
1−θ θ
for q = ∞.
Let us prove (1.26). To prove ≤, for every f ∈ L 1 () + L ∞ () and

t > 0, x ∈ we set


 f (x)
 = f (x) − f ∗ (t) , if | f (x)| > f ∗ (t),
a(x) | f (x)|


= 0 otherwise,
b(x) = f (x) − a(x).

The function a is defined in such a way that |a(x)| = | f (x)| − f ∗ (t) if
| f (x)| > f ∗ (t), |a(x)| = 0 if | f (x)| ≤ f ∗ (t). Then

a L 1 = (| f (x)| − f ∗ (t))µ(dx),
E
where E = {x ∈ : | f (x)| > f ∗ (t)} has measure µ(E) = |{s > 0 : f ∗ (s) >
f ∗ (t)}| (because | f | and f ∗ are equi-measurable) = m( f ∗ (t), f ) ≤ t,
and f ∗ is constant in [µ(E), t]. Therefore,
µ(E) t
a L 1 = ( f ∗ (s) − f ∗ (t))ds ≤ ( f ∗ (s) − f ∗ (t))ds.
0 0
Moreover, 
 = | f (x)| if | f (x)| ≤ f ∗ (t),
|b(x)|

= f ∗ (t) if | f (x)| > f ∗ (t),
so that t
1
∗
|b(x)| ≤ f (t) = f ∗ (t)ds, x ∈ .
t 0
Therefore,
t
∞
K (t, f, L , L ) ≤ a L 1 + tb L ∞ ≤
1
f ∗ (s)ds.
0
To prove the opposite inequality we use the fact that for every decompos-
ition f = a + b we have (see Exercise 9, Section 1.1.2)
f ∗ (s) ≤ a ∗ ((1 − ε)s) + b∗ (εs), s ≥ 0, 0 < ε < 1.
Then, if a ∈ L 1 (), b ∈ L ∞ (), and a + b = f we have
t t t
∗ ∗
f (s)ds ≤ a ((1 − ε)s)ds + b∗ (εs)ds
0 0 0
∞
1 ∗ 1∗
≤ a (τ )dτ + tb (0) = |a(x)|µ(dx) + t sup ess |b|.
1−ε 0 1−ε
Letting ε → 0 we get
t
f ∗ (s)ds ≤ a L 1 + tb L ∞ ,
0
so that t
K (t, f, L 1 (), L ∞ ()) ≤ f ∗ (s)ds,
0
and the statement follows.
A detailed treatment of Lorentz spaces from the interpolation point of
view, with several applications and historical remarks, is in the lecture
notes [89].
1.1.2. Exercises
1. Prove that for every x ∈ X + Y , t → K (t, x) is concave (and hence
continuous) in (0, ∞).
2. Prove that x → xθ, p is a norm in (X, Y )θ, p , for each θ ∈ (0, 1)
and p ∈ [1, ∞].
3. Take θ = 0 in Definition 1.1 and show that (X, Y )0, p = (X, Y )0 =
{0}, for all p ∈ [1, ∞). Show that X ⊂ (X, Y )0,∞ .
Take θ = 1 in Definition 1.1 and show that (X, Y )1, p = (X, Y )1 =
{0}, for all p ∈ [1, ∞). Show that Y ⊂ (X, Y )1,∞ .
4. Following the method of Example 1.8 show that (Cb (Rn),Cb1 (Rn))1,∞=
Lip(Rn ), and that for 1< p < ∞,(L p (Rn ),W 1, p (Rn ))1,∞ = W 1, p (Rn ).
Show that (C0 (Rn ), C01 (Rn ))θ,∞ = C0θ (Rn ) for each θ ∈ (0, 1). Here
C0 (Rn ) is the subspace of Cb (Rn ) consisting of the functions f such
that lim|x|→∞ f (x) = 0, and C01 (Rn ) := { f ∈ Cb1 (Rn ) : f, Di f ∈
C0 (Rn ), i = 1, . . . n}.
5. Show that for 0 < θ < 1, Cb1 (Rn ) is not dense in Cbθ (Rn ). Show that
the continuous interpolation space (Cb (Rn ), Cb1 (Rn ))θ is the space of
the “little-Hölder continuous” functions h θ (Rn ), consisting of those
bounded functions f such that
| f (x + h) − f (x)|
lim sup = 0.
h→0 x∈Rn |h|θ
6. Following the method of Example 1.8 show that
(L p (Rn ), W 1, p (Rn ))θ,q = B θp,q (Rn ).

7. Let be an open set in Rn such that there exists an extension oper-

ator E ∈ L(L p (),L p (Rn )) ∩ L(W θ, p (),W θ, p (Rn )) ∩ L(W 1, p (),
W 1, p (Rn )), for some p ∈ [1, ∞) and θ ∈ (0, 1). Show that
(L p (), W 1, p ())θ, p = W θ, p ().
In particular, if I ⊂ R is any interval,
(L p (I ), W 1, p (I ))θ, p = W θ, p (I ).
Show that if n > 1 and has uniformly C 1 boundary such extension

operator E does exist (see the remarks after Example 1.9).
The space W θ, p () is usually defined as the set of the restrictions to
of the functions in W θ, p (Rn ), or else as the set of the functions
f ∈ L p () such that
1/ p
| f (x) − f (y)| p
[ f ]W θ, p = dx dy < ∞.
× |x − y|θ p+n
If the boundary of is uniformly C 1 , these notions do coincide.

8. Let I ⊂ R be any interval, and let X be any Banach space. Show
that for 0 < θ < 1 and 1 ≤ p < ∞
(C b (I ; X), Cb1 (I ; X))θ,∞ = Cbθ (I ; X),

(L p (I ; X), W 1, p (I ; X))θ, p = W θ, p (I ; X).
9. Let (, µ) be any measure space. Prove that for each a, b ∈

L 1 () + L ∞ () we have
(a + b)∗ (s) ≤ a ∗ ((1 + ε)s) + b∗ (εs), s ≥ 0, 0 < ε < 1.
(This is used in Example 1.11). Hint: show preliminarily that
m(σ0 + σ1 , a + b) ≤ m(σ0 , a) + m(σ1 , b), σ0 , σ1 ≥ 0.
10. Prove that for each x ∈ X + Y the function K (·, x) satisfies
t
K (t, x) ≤ K (s, x) 0 < s < t.
s
1.2. The trace method

In this section we describe another construction of the real interpolation
spaces, which will be useful for proving other properties, and will let us
see the connection between interpolation theory and trace theory.
We shall use L p and Sobolev spaces of functions with values in Banach
spaces, whose definitions and elementary properties are in Appendix A.
Definition 1.12. For 0 < θ < 1 and 1 ≤ p ≤ ∞ define V ( p, θ, Y, X) as
the set of all functions u : R+ → X + Y such that u ∈ W 1, p (a, b; X + Y )
for every 0 < a < b < ∞, and
p
t → u θ (t) := t θ u(t) ∈ L ∗ (0, +∞; Y ),
p
t → wθ (t) := t θ u (t) ∈ L ∗ (0, +∞; X),
with norm
uV ( p,θ,Y,X) := u θ L ∗p (0,+∞;Y ) + wθ L ∗p (0,+∞;X) .
Moreover, for p = +∞ define a subspace of V (∞, θ, Y, X), by

V0 (∞, θ, Y, X) := u ∈ V (∞, θ, Y, X) : lim t θ u(t)Y
t→0
θ

= lim t u (t) X = 0 .
t→0
It is not difficult to see that V ( p, θ, Y, X) is a Banach space endowed

with the norm · V ( p,θ,Y,X) , and that V0 (∞, θ, Y, X) is a closed subspace
of V (∞, θ, Y, X). Moreover any function belonging to V ( p, θ, Y, X)
has a X + Y -valued continuous extension
t at t = 0. Indeed, for 0 < s < t
from the equality u(t) − u(s) = s u (σ )dσ it follows, for 1 < p < ∞,
t 1/ p t 1/ p
θ−1/ p p −(θ−1/ p) p
u(t)−u(s) X+Y ≤ σ u (σ ) X dσ σ dσ
s s
−1/ p
≤ uV ( p,θ,Y,X) [ p (1 − θ)] (t p (1−θ) − s p (1−θ) )1/ p ,
with p = p/( p − 1). Arguing similarly, one sees that also if p = 1 or
p = ∞, then u is uniformly continuous with values in X + Y near t = 0.
With the aid of Corollary A.13 we are able to characterize the real
interpolation spaces as trace spaces.
Proposition 1.13. For (θ, p) ∈ (0, 1) × [1, +∞], (X, Y )θ, p is the set of
the traces at t = 0 of the functions in V ( p, 1 − θ, Y, X), and the norm
xθ,
Tr
p = inf{uV ( p,1−θ,Y,X) : x = u(0), u ∈ V ( p, 1 − θ, Y, X)}
is an equivalent norm in (X, Y )θ, p . Moreover, for 0 < θ < 1, (X, Y )θ is

the set of the traces at t = 0 of the functions in V0 (∞, 1 − θ, Y, X).
Proof. Let x ∈ (X, Y )θ, p . We need to define a function u ∈ V ( p, 1 −

θ, Y, X) such that u(0) = x.
For every t > 0 there are at ∈ X, bt ∈ Y such that at X + tbt Y ≤
2K (t, x). It holds t 1−θ bt Y ≤ 2t −θ K (t, x), and the function t →
p
t −θ K (t, x) is in L ∗ (0, +∞). Moreover, we already know (see the proof
of Proposition 1.3) that limt→0 bt = x in X + Y . So, the function t → bt
looks a good candidate for u. But in general it is not measurable with
values in X + Y , let alone locally W 1, p . So we have to modify it, and we
proceed as follows.
For every k ∈ Z let ak ∈ X, bk ∈ Y be such that ak + bk = x, and
ak X + 2k bk Y ≤ 2K (2k , x).
For t > 0 set

+∞
+∞
u(t) = bk χ 2k ,2k+1 (t) = (x − ak ) χ 2k ,2k+1 (t),
( ] ( ]
k=−∞ k=−∞
where χ I denotes the characteristic function of the interval I , and

t
1
v(t) = u(s)ds.
t 0
We already remarked that x − bk X+Y → 0 as k → −∞. (How-

ever, we repeat the argument: since (X, Y )θ, p is contained in (X, Y )θ,∞
then t −θ K (t, x) is bounded, so that limt→0 K (t, x) = 0. Therefore,
limk→−∞ ak X = 0, so that x − bk X+Y ≤ ak X → 0 as k → −∞).
It follows that x = limt→0 u(t) = limt→0 v(t) in X + Y . Moreover,
recalling that K (·, x) is increasing, we get

+∞
|t 1−θ u(t)Y ≤ t −θ tχ 2k ,2k+1 (t)21−k K (2k , x) ≤ 4t −θ K (t, x), (1.27)
( ]
k=−∞
p
so that t → t 1−θ u(t) ∈ L ∗ (0, +∞; Y ). By Corollary A.13, t →
p
vθ (t) := t 1−θ v(t) belongs to L ∗ (0, +∞; Y ), and
vθ L ∗p (0,+∞;Y ) ≤ 4θ −1 xθ, p .
On the other hand,

t
+∞
1
v(t) = x − χ 2k ,2k+1 (s)ak ds,
t 0 k=−∞ ( ]
so that v is differentiable almost everywhere with values in X, and

1 t 1

v (t) = 2 g(s)ds − g(t), t ∈ / 2Z ,
t 0 t

where g(t) := +∞ k=−∞ χ(2k ,2k+1 ] (t)ak is such that

+∞
g(t) X ≤ χ 2k ,2k+1 (t)2K (2k , x) ≤ 2K (t, x).
( ]
k=−∞
It follows that
t 1−θ v (t) X ≤ t −θ sup g(s) X + t −θ g(t) X
0<s<t (1.28)
−θ
≤ 4t K (t, x), t > 0.
p
Then t → wθ (t) := t 1−θ v (t) belongs to L ∗ (0, +∞; X), and
wθ L ∗p (0,+∞;X) ≤ 4xθ, p .
Therefore, x is the trace at t = 0 of a function v ∈ V ( p, 1 − θ, Y, X),
and
xθ,
Tr
p ≤ 4(1 + 1/θ)xθ, p .
If x ∈ (X, Y )θ , then limt→0 t 1−θ u(t)Y = 0 by (1.27), and consequently

limt→0 t 1−θ v(t)Y = 0. By (1.28), we have also limt→0 t −θ g(t) X
= 0, so that limt→0 t 1−θ v (t) X = 0. Then v ∈ V0 (∞, 1 − θ, Y, X).
Conversely, let x be the trace at t = 0 of a function u ∈ V ( p, 1 −
θ, Y, X). Then
t
x = x − u(t) + u(t) = − u (s)ds + u(t), t > 0,
0
so that
t
1
−θ
K (t, x) ≤ t u (s)ds

+ t u(t)Y , t > 0.
1−θ 1−θ
t t (1.29)
0 X
p
Corollary A.13 implies now that t → t −θ K (t, x) belongs to L ∗ (0, +∞),
so that x ∈ (X, Y )θ, p , and
1
xθ, p ≤
xθ,
Tr
p.
θ
If x is the trace of a function u ∈ V0 (∞, 1−θ, Y, X), possibly multiplying
by a function ζ ∈ C 1 ((0, +∞); R) such that ζ ≡ 1 in (0, 1), ζ ≡ 0 in
(2, +∞), we may assume without loss of generality that u vanishes for t
large. Then, by (1.29), limt→0 t −θ K (t, x) = limt→∞ t −θ K (t, x) = 0, so
that x ∈ (X, Y )θ .
Corollary 1.14. Let 1 < p < ∞. Then

(i) (X, Y )1−1/ p, p is the set of the traces at t = 0 of the functions u ∈
W 1, p (0, ∞; X) ∩ L p (0, ∞; Y );
(ii) W 1, p (0, ∞; X) ∩ L p (0, ∞; Y ) ⊂ Cb ([0, ∞); (X, Y )1−1/ p, p ).
Proof. Statement (i) follows from Proposition 1.13, taking θ = 1 − 1/ p.
Since for u ∈ W 1, p (0, ∞; X) ∩ L p (0, ∞; Y ) and t ≥ 0 the function
s → u(t + s) belongs to W 1, p (0, ∞; X) ∩ L p (0, ∞; Y ), still from Pro-
position 1.13 we obtain that u(t) ∈ (X, Y )1−1/ p, p for each t ≥ 0 and by
estimate (1.29) we have
1
u(t)(X,Y )1−1/ p, p ≤ (uW 1, p (0,∞;X) + u L p (0,∞;Y ) ), t ≥ 0,
1 − 1/ p
so that u is bounded with values in (X, Y )1−1/ p, p . To prove that it is
continuous we approach it in W 1, p (0, ∞; X) ∩ L p (0, ∞; Y ) by a se-
quence of functions u n belonging to W 1, p (0, ∞; X) ∩ L p (0, ∞; Y ) ∩
C([0, +∞); Y ). For instance, we may extend u to the whole of R by
reflection and then we approximate the extension by convolution with
smooth mollifiers. Being continuous with values both in X and in Y ,
each u n is continuous with values in (X, Y )1−1/ p, p . By the above estim-
ate, supt≥0 u n (t) − u(t)(X,Y )1−1/ p, p goes to 0 as n → ∞, and statement
(ii) follows.
The next example shows an important connection between interpolation
theory and trace theory.
Example 1.15. Let Rn+1 + denote the upper half-space {(x, t) ∈ R × R :
n
t > 0}. Then for 1 < p < ∞, W 1−1/ p, p

(R ) is the space of the traces at
n
(x, 0) of the functions (x, t) → v(x, t) ∈ W 1, p (Rn+1

+ ).
Proof. We already know that W 1−1/ p, p (Rn ) = (L p (Rn ),W 1, p (Rn ))1−1/ p, p ,
see Example 1.8. By Proposition 1.13, W 1−1/ p, p (Rn ) is the space of the
traces at t = 0 of the functions v ∈ V ( p, 1p , W 1, p (Rn ), L p (Rn )). On the
other hand, v ∈ V ( p, 1p , W 1, p (Rn ), L p (Rn )) if and only if the function
v(x, t) := v(t)(x) is in W 1, p (Rn+1
+ ). Indeed, concerning measurability,
it is possible to see that a function w : (0, +∞) → L p (Rn ) (resp., w :
(0, +∞) → W 1, p (Rn )) is measurable in the sense of Definition A.1 iff
(t, x) → w(t)(x) is measurable (resp., (t, x) → w(t)(x) and (t, x) →
Di w(t)(x) are measurable for all i = 1, . . . n). Concerning estimates,
p
the condition t → t 1/ p v(t) ∈ L ∗ ((0, +∞), W 1, p (Rn )) means that the
function v(t, x) := v(t)(x) satisfies
+∞ n
|v(x, t)| +
p
|vxi (x, t)| dx dt < ∞,
p
0 Rn i=1
p
and the condition t → t 1/ p v (t) ∈ L ∗ ((0, +∞), L p (Rn )) means that
+∞
|vt (x, t)| p dx dt < ∞.
0 Rn
Therefore, v belongs to V ( p, 1p ,W 1, p (Rn ),L p (Rn )) iff (x, t) → v(t)(x)

belongs to W 1, p (Rn+1
+ ).
In particular, choosing p = 2 we obtain that H 1/2 (Rn ) := W 1/2,2 (Rn )

is the space of the traces at (x, 0) of the functions (x, t) → v(x, t) ∈
H 1 (Rn+1
+ ).
Remark 1.16. Two technical remarks about trace spaces, used in the
next proofs.
Remark 1. By Proposition 1.13, if x ∈ (X, Y )θ, p or x ∈ (X, Y )θ ,
then x is the trace at t = 0 of a function u belonging to L p (a, b; Y ) ∩
W 1, p (a, b; X +Y ) for 0 < a < b. But it is possible to find a more regular
function v ∈ V ( p, 1 − θ, Y, X) (or v ∈ V0 (∞, 1 − θ, Y, X)) such that
v(0) = x. For instance we may take

1 t
v(t) = u(s)ds, t ≥ 0.
t 0
Then v ∈ W 1, p (a, b; Y ) ∩ W 2, p (a, b; X + Y ) for 0 < a < b, v(0) = x,
p
and moreover t → t 1−θ v(t) belongs to L ∗ (0, +∞; Y ), t → t 2−θ v (t)
p p
belongs to L ∗ (0, +∞; Y ), and t → t 1−θ v (t) belongs to L ∗ (0, +∞; X),
with norms estimated by const. uV ( p,1−θ,Y,X) .
Even better, choose any smooth nonnegative function ϕ : R+ → R,
∞
with compact support and 0 s −1 ϕ(s)ds = 1, and set
∞ ∞
t dτ t ds
v(t) = ϕ u(τ ) = ϕ(s)u .
0 τ τ 0 s s
Then v ∈ C ∞ (R+ ; Y ), v ∈ C ∞ (R+ ; X), v(0) = x, and
t → t n−θ v (n) (t) ∈ L ∗p (0, +∞; X), n ∈ N,
t → t n+1−θ v (n) (t) ∈ L ∗p (0, +∞; Y ), n ∈ N ∪ {0},
with norms estimated by c(n)uV ( p,1−θ,Y,X) . If in addition p = ∞ and
x ∈ (X, Y )θ then
lim t n−θ v (n) (t) X = 0, n ∈ N,
t→0
lim t n+1−θ v (n) (t)Y = 0. n ∈ N ∪ {0}.

t→0
Remark 2. Let x ∈ X + Y be the trace at t = 0 of a function v ∈

V ( p, 1 − θ, Y, X). Fix any ϕ ∈ C0∞ ([0, +∞)) such that ϕ ≡ 1 in a
right neighborhood of 0, say in (0, 1]. The function t → ϕ(t)v(t) is
in V ( p, 1 − θ, Y, X), its norm does not exceed CvV ( p,1−θ,Y,X) , with
C depending only on ϕ, and its trace at t = 0 is still x. Moreover, it
has compact support in [0, +∞). This shows that in the definition of
the trace spaces one could consider just the subset of V ( p, 1 − θ, Y, X)
consisting of the functions with compact support, obtaining an equivalent
trace space (i.e., the same space with an equivalent norm).
By means of the trace method it is easy to prove some important dens-
ity properties.
Proposition 1.17. Let 0 < θ < 1. For 1 ≤ p < ∞, X ∩ Y is dense in
(X, Y )θ, p . For p = ∞, (X, Y )θ is the closure of X ∩ Y in (X, Y )θ,∞ .
Proof. Let p < ∞, and let x ∈ (X, Y )θ, p . By Remark 1.16(1), x = v(0),
where v ∈ C ∞ ((0, ∞); X ∩ Y ) ∩ V ( p, 1 − θ, Y, X), and moreover t →
p
t 2−θ v ∈ L ∗ (0, +∞; Y ). Set
xε = v(ε), ∀ε > 0.
Then xε ∈ X ∩ Y , and we shall show that xε → x in (X, Y )θ, p .
We have xε − x = z ε (0), where
z ε (t) = (v(ε) − v(t))χ[0,ε] (t).
It is not hard to check that z ε ∈ W 1, p (a, b; X) for 0 < a < b < ∞, and
that z ε (t) = −v (t)χ(0,ε) (t). It follows that
lim t 1−θ z ε (t) L ∗p (0,+∞;X) = 0.
ε→0
Moreover, due to the equality

+∞
z ε (t) = χ(0,ε) (s)v (s)ds,
t
we get, using the Hardy-Young inequality (A.10)(ii),

+∞ 1/ p
(1−θ) p dt
(t z ε (t)Y )
0 t
+∞ +∞ p 1/ p
(1−θ) p ds dt
≤ t χ(0,ε) (s)sv (s)Y
0 t s t
+∞ 1/ p
1 (2−θ) p ds
≤ χ(0,ε) (s)s v (s)Y ,
1−θ 0 s
p
so that t → t 1−θ z ε (t) ∈ L ∗ (0, +∞; Y ) for every ε, and
lim t 1−θ z ε (t) L ∗p (0,+∞;Y ) = 0.

ε→0
Therefore, z ε → 0 in V ( p, 1 − θ, Y, X) as ε → 0, which means that

x ε − xθ,
Tr
p → 0 as ε → 0. From Proposition 1.13 we get limε→0 x ε −
xθ, p = 0.
Let now x ∈ (X, Y )θ . Due again to Remark 1.16, x is the trace at
t = 0 of a function v ∈ V0 (∞, 1 − θ, Y, X), such that t → t 2−θ v (t)
∈ L ∞ (0, +∞; Y ) and limt→0 t 2−θ v (t)Y = 0. Let xε , z ε be defined as
above. Then limt→0 t 1−θ v (t) X = 0, so that
sup t 1−θ z ε (t) X = sup t 1−θ v (t) X → 0, as ε → 0,

t>0 0<t≤ε
and
sup t 1−θ z ε (t)Y = sup t 1−θ v(ε)−v(t)Y ≤ 2 sup s 1−θ v(s)Y → 0,

t>0 0<t≤ε 0<s≤ε
as ε → 0. Arguing as before, it follows that xε − xθ,∞ → 0 as

ε → 0.
1.2.1. Exercises
1. Prove the statements of Remark 1.16.
2. A regularized K -method.
(a) Prove that (X, Y )θ, p is the set of the elements x ∈ X + Y such
that x = u(t) + v(t) for almost all t > 0, with t → t −θ u(t) ∈
p p
L ∗ (0, ∞; X), t → t 1−θ v(t) ∈ L ∗ (0, ∞; Y ), and the norm
x → inf t −θ u(t) L ∗p (0,∞;X) + t 1−θ v(t) L ∗p (0,∞;Y )

x=u(t)+v(t)
is equivalent to the norm of (X, Y )θ, p . Prove that (X, Y )θ is the

set of the elements x ∈ X +Y such that x = u(t)+v(t) for almost
all t > 0, with t → t −θ u(t) ∈ L ∞ (0, ∞; X), t → t 1−θ u(t) ∈
L ∞ (0, ∞; Y ), and limt→0 t −θ u(t) X = limt→∞ t −θ u(t) X =
0, limt→0 t 1−θ v(t)Y = limt→∞ t 1−θ v(t)Y = 0.
(b) Using (a), prove that (X, Y )θ, p is the set of the elements x ∈ X +Y
such that x = u(t) + v(t) for almost all t > 0, with t → t θ u(t) ∈
p p
L ∗ (0, ∞; X), t → t θ−1 v(t) ∈ L ∗ (0, ∞; Y ), and the norm
x → inf t θ u(t) L ∗p (0,∞;X) + t θ−1 v(t) L ∗p (0,∞;Y )

x=u(t)+v(t)
is equivalent to the norm of (X, Y )θ, p . Prove that (X, Y )θ is the

set of the elements x ∈ X +Y such that x = u(t)+v(t) for almost
all t > 0, with t → t θ u(t) ∈ L ∞ (0, ∞; X), t → t θ−1 v(t) ∈
L ∞ (0, ∞; Y ), limt→0 t θ u(t) X = limt→∞ t θ u(t) X = 0 and
limt→0 t θ−1 v(t)Y = limt→∞ t θ−1 v(t)Y = 0.
3. The mean method.
(a) Prove that (X, Y )θ, p is the set of the elements x ∈ X + Y such
∞ p
that x = 0 u(t)dt/t, with t → t −θ u(t) ∈ L ∗ (0, ∞; X), t →
p
t 1−θ u(t) ∈ L ∗ (0, ∞; Y ), and the norm
x → inf t −θ u(t) L ∗p (0,∞;X) + t 1−θ u(t) L ∗p (0,∞;Y )
x= 0∞ u(t)dt/t
is equivalent to the norm of (X,Y )θ, p . Prove that

∞(X,Y )θ is the set
of the elements x ∈ X + Y such that x = 0 u(t)dt/t, with
t → t −θ u(t) ∈ L ∞ (0, ∞; X), t → t 1−θ u(t) ∈ L ∞ (0, ∞; Y ),
and lim t −θ u(t) X = lim t −θ u(t) X = 0, limt→0 t 1−θ u(t)Y =
t→0 t→∞
lim t 1−θ u(t)Y = 0.
t→∞
∞
(Hint: use Remark 1.16 to write any x ∈ (X,Y)θ, p as x = − v(t)dt,
0
as in the proof of next Proposition 1.20).
(b) Using (a), prove that (X, Y )θ, p is the set of the elements x ∈ X +Y
∞ p
such that x = 0 u(t)dt/t, with t → t θ u(t) ∈ L ∗ (0, ∞; X),
p
t → t θ−1 u(t) ∈ L ∗ (0, ∞; Y ), and the norm
x → inf t θ u(t) L ∗p (0,∞;X) + t θ−1 v(t) L ∗p (0,∞;Y )
x= 0∞ u(t)dt/t
is equivalent to the norm of (X, Y )θ, p . Prove that

∞ (X, Y )θ is the
set of the elements x ∈ X + Y such that x = 0 u(t)dt/t, with
lim t θ u(t) X = lim t θ u(t) X = 0, and limt→0 t θ−1 u(t)Y =
t→0 t→∞
lim t θ−1 u(t)Y = 0.
t→∞
4. Using Proposition 1.13 and the inclusion L p (I ; L p (J ; X)) ⊂ L p (I ×

J, X) for arbitrary real intervals I , J and Banach space X (if we
identify any function f ∈ L p (I ; L p (J ; X)) with the function (t,s) →
f (t)(s), t ∈ I , s ∈ J ), show that for any θ ∈ (0, 1), p ∈ [1, ∞], and
for any interval I ⊂ R,
(L p (I ; X), L p (I ; Y ))θ, p ⊂ L p (I ; (X, Y )θ, p ).
(The opposite inclusion is also true, see [69, Chapter VII].)
1.3. Dual spaces of real interpolation spaces

Since both X and Y are continuously embedded in (X ∩ Y ) , then taking
V = (X ∩ Y ) we see that (X , Y ) is an interpolation couple.
The main result is the following.
Theorem 1.18. If X ∩ Y is dense in X and in Y , then for each θ ∈ (0, 1)
((X, Y )θ, p ) = (X , Y )θ, p , if p < ∞,

((X, Y )θ ) = (X , Y )θ,1 .
We will not give all the details of the proof, but just a sketch.
Step 1. The embeddings ⊃ are easy. If f ∈ (X , Y )θ, p with p ∈
(1, +∞] then by Exercise 3(a), Section 1.2.1, we have
∞
dt
f = ϕ(t) ,
0 t
p p
where t → t −θ ϕ(t) ∈ L ∗ (0, ∞; X ), t → t 1−θ ϕ(t) ∈ L ∗ (0, ∞; Y ),
and
t −θ ϕ(t) L p (0,∞;X ) + t 1−θ ϕ(t) L p (0,∞;Y ) ≤ C1 f (X ,Y )θ, p .

∗ ∗

Since f ∈ X + Y , then f ∈ (X ∩ Y ) . For each x ∈ X ∩ Y , write
x = u(t) + v(t), t > 0,

p p
with t → t θ u(t) ∈ L ∗ (0, +∞; X), t → t θ−1 v(t) ∈ L ∗ (0, +∞; Y ), and
t θ u(t) L ∗p (0,+∞;X) + t θ−1 v(t) L ∗p (0,+∞;Y ) ≤ C2 x(X,Y )θ, p
(see Exercise 2(b), Section 1.2.1). Then,

∞ ∞
dt dt
| f (x)| ≤ |ϕ(t)(u(t))| + |ϕ(t)(v(t))|
0 t 0 t
∞ ∞
dt dt
≤ t −θ ϕ(t) X t θ u(t) X + t 1−θ ϕ(t)Y t θ−1 v(t))Y
0 t 0 t
≤ t −θ ϕ(t) L p (X ) t θ u(t) L ∗p (X) + t 1−θ ϕ(t) L p (Y ) t θ−1 v(t) L ∗p (Y )

∗ ∗
≤ C1 C2 f (X ,Y )θ, p x(X,Y )θ, p .
Since X ∩ Y is dense in (X, Y )θ, p if p < ∞ and in (X, Y )θ if p = ∞, f

has a continuous extension to (X, Y )θ, p if p < ∞, to (X, Y )θ if p < ∞.
Step 2. The proof of other embeddings is a bit more complicated, al-

though it relies on similar arguments.
Let f belong to the dual space of (X, Y )θ, p with p < ∞.
For x ∈ (X, Y )θ, p let us consider all the couples of functions (
u ,
v) ∈
p p
L ∗ (0, ∞; X) × L ∗ (0, ∞; Y ) such that t θ u (t) + t θ−1
v (t) = x for a.e.
t > 0. Such couples do exist by Exercise 2(a), Section 1.2.1, and there is
C > 0 such that
xθ, p ≤ C inf (

u L ∗p (0,∞;X) +
v L ∗p (0,∞;Y ) ).
(
u ,
v)
The set of all such couples (

u ,
v ) when x runs in (X, Y )θ, p is a closed
p p
subspace S of L ∗ (0, ∞; X) × L ∗ (0, ∞; Y ). The functional

f (
u ,
v ) := f (x)
is well defined and continuous in S, since for each (

u ,
v ) ∈ S we have
|
f ( v )| ≤ f ((X,Y )θ, p ) xθ, p
u ,
≤ C f ((X,Y )θ, p ) (
u L ∗p (0,∞;X) +
v L ∗p (0,∞;Y ) ).
By the Hahn-Banach Theorem it has an extension (still called f ) to the

p p
whole L ∗ (0, ∞; X) × L ∗ (0, ∞; Y ), with the same norm. Now, assume
p p
that the dual spaces of L ∗ (0, ∞; X) and of L ∗ (0, ∞; Y ) are isomorphic

to L ∗ (0, ∞; X ) and to L ∗ (0, ∞; Y ), respectively1 . Then
p p
f has the
representation
∞ ∞
dt dt
f (u, v) = g(t)(u(t)) + h(t)(v(t)) , (1.30)
0 t 0 t
p p
with g ∈ L ∗ (0, ∞; X ) and h ∈ L ∗ (0, ∞; Y ). Taking x ∈ X ∩ Y ,
ϕ ∈ C0∞ (0, ∞), and setting u(t) = ϕ(t)x, v(t) = −tϕ(t)x, we get
t θ u(t) + t θ−1 v(t) ≡ 0, so that
f (u, v) = 0 which means
∞
dt
ϕ(t)(g(t) − th(t))x = 0,
0 t
and therefore g(t) − th(t) = 0 a.e. for t > 0 (this also implies that

g ∈ L p (a, b; Y ) and h ∈ L p (a, b; X ) for 0 < a < b < ∞).

1 The spaces X such that L p (0, ∞; X) is isomorphic to L p (0, ∞; X ) may be characterized in
∗ ∗
terms of the Radon–Nikodym property of X , see [46, 40]. In particular, if X is separable, or if X

is reflexive, the isomorphism holds.
Still for x ∈ X ∩ Y , let now x = u(t) + v(t) for t > 0, with u, v

as in Step 1. Then, setting u (t) = t −θ u(t), v (t) = t 1−θ v(t), we have
θ θ−1
x = t u (t) + t v (t) for a.e. t > 0, so that
∞

v (t) dt
f (x) = f (u ,v) = g(t)(u (t)) + th(t)
0 t t
∞

v (t) dt
= g(t) u (t) +
0 t t
∞ ∞
−θ dt dt
= t g(t)(u(t) + v(t)) = t −θ g(t)x
0 t 0 t
∞
dt
:=
g (t)x
0 t
p
where t → t θ
g (t) = g(t) ∈ L ∗ (0, ∞; X ) and t → t θ−1
g (t) = h(t) ∈
p
L ∗ (0, ∞; Y ). Therefore,
1 ∞
dt dt
f (x) =
g (t)x +
g (t)x := ( f 1 + f 2 )x, x ∈ X ∩ Y,
0 t 1 t
where f 1 , f 2 are extended by continuity to all x ∈ Y , x ∈ X respectively.
Then f ∈ X + Y ; moreover by Exercise 3(b), Section 1.2.1, f belongs
to (X , Y )θ, p and
f (X ,Y )θ, p ≤ C (t θ
g (t) L p (0,∞;X ) + t θ−1
g (t) L p (0,∞;Y ) )
∗ ∗
= C (g L p (0,∞;X ) + h L p (0,∞;Y ) )

∗ ∗
≤ C f ((X,Y )θ, p ) .
Step 3. In general, we cannot guarantee that f has the representation

(1.30). The proof is achieved by the same procedure, using discrete ana-
logues of the characterizations of Exercises 2 and 3, Section 1.2.1, where
series replace integrals: since the dual space of the sequence space l p (Z )
is isomorphic to l p (Z ) for p < ∞ and the dual space of c0 (Z ) is iso-
morphic to l1 (Z ) for any Banach space Z , the difficulty is overcome. For
the details, see [90, Section 1.11.2].
1.4. Intermediate spaces and reiteration

Let us introduce two classes of intermediate spaces for the interpolation
couple (X, Y ).
Definition 1.19. Let 0 ≤ θ ≤ 1, and let E be a Banach space such that

X ∩ Y ⊂ E ⊂ X + Y.
(i) We say that E belongs to the class Jθ between X and Y if there is a
constant c such that
θ
x E ≤ cx1−θ
X xY , ∀x ∈ X ∩ Y.
In this case we write E ∈ Jθ (X, Y ).

(ii) We say that E belongs to the class K θ between X and Y if there is
k > 0 such that
K (t, x) ≤ k t θ x E , ∀x ∈ E, t > 0.
In this case we write E ∈ K θ (X, Y ).
If θ ∈ (0, 1) this means that E is continuously embedded in (X,Y )θ,∞ .
A useful characterization of Jθ (X, Y ) is the following.

Proposition 1.20. Let 0 < θ < 1, and let E be a Banach space such that
X ∩ Y ⊂ E ⊂ X + Y . The following statements are equivalent:
(i) E belongs to the class Jθ between X and Y ,
(ii) (X, Y )θ,1 ⊂ E.
Proof. The implication (ii) ⇒ (i) is a straightforward consequence of

Corollary 1.7, with p = 1. Let us show that (i) ⇒ (ii). For every x ∈
(X, Y )θ,1 let u ∈ V (1, 1 − θ, Y, X) be such that u(t) vanishes for large t,
u(0) = x, and set
1 t
v(t) = u(s)ds.
t 0
By Remark 1.16, t → t 2−θ v (t) belongs to L 1∗ (0, +∞; Y ), and t →
t 1−θ v (t) belongs to L 1∗ (0, +∞; X). Moreover v(0) = x, v(+∞) = 0,
so that +∞
x =− v (t)dt.
0
Let c be such that y E ≤ cyθY y1−θ

X for every y ∈ X ∩ Y . Then
v (t) E ≤ cv (t)θY v (t)1−θ −1 2−θ
X = ct t v (t)θY t 1−θ v (t)1−θ
X .
Since t → t 2−θ v (t)θY belongs to L ∗ (0, +∞) and t → t 1−θ v (t)1−θ

1/θ
X
1/(1−θ)
belongs to L ∗ (0, +∞), by the Hölder inequality v ∈ L 1 (0,+∞; E),
and
x E ≤ c(t 2−θ v (t) L 1∗ (0,∞;Y ) )θ (t 1−θ v (t) L 1∗ (0,∞;X) )1−θ
≤ const. xθ,1 .
By Definition 1.19 and Proposition 1.17, if 0 < θ < 1 a space E belongs

to K θ (X, Y ) ∩ Jθ (X, Y ) if and only if
(X, Y )θ,1 ⊂ E ⊂ (X, Y )θ,∞ .
In particular, (X, Y )θ, p and (X, Y )θ are in K θ (X, Y ) ∩ Jθ (X, Y ), for

every p ∈ [1, ∞]. We shall see in chapter 2 that also the complex in-
terpolation spaces [X, Y ]θ are in K θ (X, Y ) ∩ Jθ (X, Y ).
But there are also intermediate spaces belonging to K θ (X,Y )∩ Jθ (X,Y )
which are not interpolation spaces.
Example 1.21. Cb1 (Rn ) is in J1/2 (Cb (Rn ), Cb2 (Rn )) ∩ K 1/2 (Cb (Rn ),
Cb2 (Rn )), but it is not an interpolation space between Cb (Rn ) and Cb2 (Rn ).
Proof. From the inequalities (i = 1, . . . , n)
1
| f (x + hei ) − f (x) − Di f (x)h| ≤ Dii f ∞ h 2 , x ∈ Rn , h > 0,
2
we get
| f (x + hei ) − f (x)| 1
|Di f (x)| ≤ + Dii f ∞ h, x ∈ Rn , h > 0,
h 2
so that
2 f ∞ 1
Di f ∞ ≤ + Dii f ∞ h, h > 0.
h 2
Taking the minimum for h ∈ (0, +∞) we get
Di f ∞ ≤ 2( f ∞ )1/2 (Dii f ∞ )1/2 , f ∈ Cb2 (Rn ),
so that

n
f Cb1 ≤ ( f ∞ )1/2 ( f ∞ )1/2 + 2 i=1 (Dii f ∞ )1/2
≤ C( f ∞ )1/2 ( f Cb2 )1/2 .
This implies that Cb1 (Rn ) belongs to J1/2 (Cb (Rn ), Cb2 (Rn )). To prove that
it belongs also to K 1/2 (Cb (Rn ), Cb2 (Rn )), namely that it is continuously
embedded in (Cb (Rn ), Cb2 (Rn ))1/2,∞ , we argue as in Example 1.8: for
every f ∈ Cb1 (Rn ) and t > 0 the functions at , bt defined in (1.18) are
easily seen to satisfy
at ∞ ≤ Ct[ f ]Lip , bt Cb1 ≤ C f Cb1 , Di j bt ∞ ≤ Ct −1 [ f ]Lip .

Therefore, K (t, f,Cb (Rn ),Cb2 (Rn )) ≤ at 1/2 ∞ + tbt 1/2 Cb2 ≤ C(t 1/2 +
t) f Cb1 so that Cb1 (Rn ) is in K 1/2 (Cb (Rn ), Cb2 (Rn )).
But Cb1 (Rn ) is not a real interpolation space between Cb (Rn ) and
Cb2 (Rn ), even for n = 1.
Indeed, consider the family of operators
1
x
(Tε f )(x) = ( f (y) − f (0)) dy, x ∈ R.
−1 x2 + y 2 + ε2
It is easy to see that Tε L(Cb (R)) and Tε L(Cb2 (R)) are bounded by a con-
stant independent of ε. Indeed, for every continuous and bounded f ,
1
|x|
|(Tε f )(x)| ≤ 2 f ∞ dy ≤ 2π f ∞ ,
−1 x2 + y 2 + ε2
1
−x 2 + y 2 + ε2
(Tε f ) (x) = ( f (y) − f (0))dy,
−1 (x 2 + y 2 + ε2 )2
and for every f ∈ Cb1 (R),

1
−2x(−x 2 + 3y 2 + 3ε2 ) y

(Tε f ) (x) = f (s)ds dy
−1 (x 2 + y 2 + ε2 )3 0

1
−2x(−x 2 + 3y 2 + 3ε2 ) y
= ( f (s) − f (0))ds dy,
−1 (x 2 + y 2 + ε2 )3 0
so that, if f ∈ Cb2 (R),

1
x 2 + 3y 2 + 3ε2
|(Tε f ) (x)| ≤ |x| dy f ∞ ≤ 3π f ∞ .
−1 (x 2 + y 2 + ε2 )2
On the contrary, choosing f ε (x) = (x 2 + ε2 )1/2 η(x), with η ∈ C0∞ (R),

η ≡ 1 in [−1, 1], we get

1
(y 2 + ε2 )1/2 − ε 1 1/ε
(s 2 + 1)1/2 − 1
(Tε f ε ) (0) = dy = ds
−1 y +ε
2 2 ε −1/ε s2 + 1
so that limε→0 (Tε f ε ) (0) = +∞, while the Cb1 norm of f ε is bounded by
a constant independent of ε. Therefore Tε L(Cb1 (R)) blows up as ε → 0.
This shows that there does not exist any constant C such that
T L(Cb1 (R)) ≤ C(T L(Cb2 (R)) + T L(Cb (R)) )

for all T ∈ L(Cb2 (R)) ∩ L(Cb (R)), and therefore Cb1 (R) is not an inter-
polation space between Cb (R) and Cb2 (R) by Lemma 0.1.
This counterexample is due to Mitjagin and Semenov [78], it shows
also that C 1 ([−1, 1]) is not a real interpolation space between C([−1, 1])
and C 2 ([−1, 1]), and it may be obviously adapted to show that for any
dimension n, Cb1 (Rn ) is not a real interpolation space between Cb (Rn )
and Cb2 (Rn ).
Remark 1.22. Arguing similarly one sees that Cbk (Rn ) is in
J1/2 (Cbk−1 (Rn ), Cbk+1 (Rn )) ∩ K 1/2 (Cbk−1 (Rn ), Cbk+1 (Rn )),
for every k ∈ N. It follows easily that for m 1 < k < m 2 ∈ N, Cbk (Rn ) be-
longs to J(k−m 1 )/(m 2 −m 1 ) (Cbm 1 (Rn ), Cbm 2 (Rn )). For instance, knowing that
Cb1 (Rn ) is in J1/2 (Cb (Rn ), Cb2 (Rn )) and that Cb2 (Rn ) is in J1/2 (Cb1 (Rn ),
Cb3 (Rn )) one gets, for every f ∈ Cb3 (Rn ),
1/2 1/2 1/2
f Cb1 ≤ C f 1/2
∞ f C 2 ≤ C f ∞ ( f C 1 f C 3 )
1/2 1/2
b b b
3/4 1/2 1/4

so that f C 1 ≤ C f ∞ f C 3 , which implies
b b
f Cb1 ≤ C f 2/3
1/3
∞ f C 3
b
that is, Cb1 (Rn ) belongs to J1/3 (Cb (Rn ), Cb3 (Rn )).
Now we are able to state the Reiteration Theorem. It is one of the main
tools of general interpolation theory.
Theorem 1.23. Let 0 ≤ θ0 < θ1 ≤ 1. Fix θ ∈ (0, 1) and set ω =
(1 − θ)θ0 + θθ1 . The following statements hold true.
(i) If E i belong to the class K θi (i = 0, 1) between X and Y , then
(E 0 , E 1 )θ, p ⊂ (X, Y )ω, p , ∀ p ∈ [1, ∞], (E 0 , E 1 )θ ⊂ (X, Y )ω .
(ii) If E i belong to the class Jθi (i = 0, 1) between X and Y , then
(X, Y )ω, p ⊂ (E 0 , E 1 )θ, p , ∀ p ∈ [1, ∞], (X, Y )ω ⊂ (E 0 , E 1 )θ .
Consequently, if E i belong to K θi (X, Y ) ∩ Jθi (X, Y ), then
(E 0 , E 1 )θ, p = (X, Y )ω, p , ∀ p ∈ [1, ∞], (E 0 , E 1 )θ = (X, Y )ω ,

Proof. Let us prove statement (i). Let ki be such that
K (t, x) ≤ ki t θi x Ei , x ∈ E i , i = 0, 1.
For each x ∈ (E 0 , E 1 )θ, p , let a ∈ E 0 , b ∈ E 1 be such that x = a + b.

Then
K (t, x, X, Y ) ≤ K (t, a, X, Y )+K (t, b, X, Y ) ≤ k0 t θ0 a E0 +k1 t θ1 b E1 .
Since a and b are arbitrary, it follows that
K (t, x, X, Y ) ≤ max{k0 , k1 }t θ0 K (t θ1 −θ0 , x, E 0 , E 1 ).
Consequently,
t −ω K (t, x, X, Y ) ≤ max{k0 , k1 }t −θ(θ1 −θ0 ) K (t θ1 −θ0 , x, E 0 , E 1 ). (1.31)
By the change of variable s = t θ1 −θ0 we obtain that t → t −ω K (t, x, X, Y )

p
belongs to L ∗ (0, +∞), so that x is in (X, Y )ω, p , and
x(X,Y )ω, p ≤ max{k0 , k1 }(θ1 − θ0 )−1/ p x(E0 ,E1 )θ, p , if p < ∞,
x(X,Y )ω,∞ ≤ max{k0 , k1 }x(E0 ,E1 )θ, p , if p = ∞.
If x ∈ (E 0 , E 1 )θ , by (1.31) we get
lim t −ω K (t, x, X, Y ) ≤ max{k0 , k1 } lim s −θ K (s, x, E 0 , E 1 ) = 0,

t→0 s→0
and
lim t −ω K (t, x, X, Y ) ≤ max{k0 , k1 } lim s −θ K (s, x, E 0 , E 1 ) = 0,

t→+∞ s→+∞
so that x ∈ (X, Y )ω .
Let us prove statement (ii). By Proposition 1.13 and Remark 1.16(1),
every x ∈ (X, Y )ω, p is the trace at t = 0 of a regular function v : R+ →
p
X ∩ Y such that v(+∞) = 0, t → t 1−ω v (t) belongs to L ∗ (0, +∞, X),
2−ω p
t → t v (t) belongs to L ∗ (0, +∞, Y ), and
t 1−ω v (t) L ∗p (0,+∞,X) + t 2−ω v (t) L ∗p (0,+∞,Y ) ≤ kx(X,Y

Tr
)ω, p ,
with k independent of x and v. We shall show that the function
g(t) = v(t 1/(θ1 −θ0 ) ), t > 0,
belongs to V ( p, 1 − θ, E 0 , E 1 ): since g(0) = x, this will imply, through

Proposition 1.13, that x ∈ (E 0 , E 1 )θ, p .
To this aim we preliminarily estimate v (t) Ei , i = 0, 1. Let ci be

such that
θi
y Ei ≤ ci y1−θ
X yY
i
y ∈ Y, i = 0, 1.
Then
ci
v (s) Ei ≤ θ +1−ω
s 1−ω v (s)1−θ
X s
i
v (s)θYi ,
2−ω
i = 0, 1,
s i
so that from the equalities
θ0 + 1 − ω = 1 − θ(θ1 − θ0 ), θ1 + 1 − ω = 1 + (1 − θ)(θ1 − θ0 ),
we get

(i) s 1−θ(θ1 −θ0 ) v (s) L ∗p (0,+∞;E0 ) ≤ c0 k x(X,Y
Tr
)ω, p ,
(1.32)
(ii) s 1+(1−θ)(θ1 −θ0 ) v (s) L ∗p (0,+∞;E1 ) ≤ c1 k x(X,Y
Tr
)ω, p .
∞
From the equality v(t) = − t v (s)ds and 1.32(ii), using the Hardy-
Young inequality (A.10)(ii) if p < ∞, we get
c1 k
t (1−θ)(θ1 −θ0 ) v(t) L ∗p (0,+∞;E1 ) ≤ x(X,Y
Tr
)ω, p .
(1 − θ)(θ1 − θ0 )
p
It follows that t → t 1−θ g(t) ∈ L ∗ (0, +∞; E 1 ), and
t 1−θ g(t) L ∗p (0,+∞;E1 ) ≤ (θ1 − θ0 )−1/ p t (1−θ)(θ1 −θ0 ) v(t) L ∗p (0,+∞;E1 ) .
Moreover g (t) = (θ1−θ0 )−1 t −1+1/(θ1 −θ0 ) v (t 1/(θ1 −θ0 ) ), so that, by (1.32)(i),
p
t → t 1−θ g (t)= (θ1−θ0 )−1 t (1−θ(θ1−θ0 ))/(θ1−θ0 ) v (t 1/(θ1 −θ0 ) ) ∈ L ∗ (0,+∞; E 0 ),
and
t 1−θ g (t) L ∗p (0,+∞;E0 ) ≤ (θ1 − θ0 )−1−1/ p t 1−θ(θ1 −θ0 ) v (t) L ∗p (0,+∞;E0 ) .
Therefore, g ∈ V ( p, 1−θ, E 0 , E 1 ), so that x = g(0) belongs to (E 0 , E 1 )θ, p ,

and
x(E
T
0 ,E 1 )θ, p
≤ (θ1 − θ0 )−1−1/ p kx(X,Y
T
)ω, p .
If x ∈ (X, Y )ω , then (1.32)(i) has to be replaced by
lim s 1−θ(θ1 −θ0 ) v (s) E0 = 0,

s→0
so that
t −θ+1/(θ1 −θ0 ) 1/(θ1 −θ0 )
lim t 1−θ g (t) E0 = lim v (t ) E0 = 0.
t→0 t→0 θ1 − θ0
Similarly, (1.32)(ii) has to be replaced by

lim s 1+(1−θ)(θ1 −θ0 ) v (s) E1 = 0.
s→0
Using the equality

ε1/(θ1 −θ0 ) +∞
t 1−θ
t 1−θ
g(t) = t 1−θ
v (s)ds + 1−θ ε 1−θ
v (s)ds ,
t 1/(θ1 −θ0 ) ε ε1/(θ1 −θ0 )
which holds for 0 < t < ε, one deduces that limt→0 t 1−θ g(t) E1 = 0.
A first consequence of the Reiteration Theorem is the following char-
acterization of real interpolation spaces between couples of real interpol-
ation spaces.
Corollary 1.24. By Proposition 1.2.3, the spaces (X, Y )θ, p and (X, Y )θ
are in K θ (X, Y ) ∩ Jθ (X, Y ) for 0 < θ < 1 and 1 ≤ p ≤ ∞. The
Reiteration Theorem yields
((X, Y )θ0 ,q0 , (X, Y )θ1 ,q1 )θ, p = (X, Y )(1−θ)θ0 +θθ1 , p ,
((X, Y )θ0 , (X, Y )θ1 ,q )θ, p = (X, Y )(1−θ)θ0 +θθ1 , p ,
((X, Y )θ0 ,q , (X, Y )θ1 )θ, p = (X, Y )(1−θ)θ0 +θθ1 , p ,
for 0 < θ0 < θ1 < 1, 1 ≤ p, q ≤ ∞. Moreover, since X belongs to
K 0 (X, Y ) ∩ J0 (X, Y ), and Y belongs to K 1 (X, Y ) ∩ J1 (X, Y ) between
X and Y , then
((X, Y )θ0 ,q , Y )θ, p = (X, Y )(1−θ)θ0 +θ, p , ((X, Y )θ0 , Y )θ = (X, Y )(1−θ)θ0 +θ ,
and
(X, (X, Y )θ1 ,q )θ, p = (X, Y )θ1 θ, p , (X, (X, Y )θ1 )θ = (X, Y )θ1 θ ,
for 0 < θ0 , θ1 < 1, 1 ≤ p, q ≤ ∞.
1.4.1. Examples
The following examples are immediate consequences of Examples 1.8,
1.9, 1.11 and of Corollary 1.24.
Example 1.25. Let 0 ≤ θ1 < θ2 ≤ 1, 0 < θ < 1. Then
(Cbθ1 (Rn ), Cbθ2 (Rn ))θ,∞ = Cb(1−θ)θ1 +θθ2 (Rn ).
If is an open set in Rn with uniformly C 1 boundary, then
(Cbθ1 (), Cbθ2 ())θ,∞ = Cb(1−θ)θ1 +θθ2 ().
Example 1.26. Let 0 ≤ θ1 < θ2 ≤ 1, 0 < θ < 1, 1 ≤ p < ∞. Then

(W θ1 , p (Rn ), W θ2 , p (Rn ))θ, p = W (1−θ)θ1 +θθ2 , p (Rn ).
If is an open set in Rn with uniformly C 1 boundary, then
(W θ1 , p (), W θ2 , p ())θ,∞ = W (1−θ)θ1 +θθ2 , p ().
Example 1.27. Let (, µ) be a σ -finite measure space. Then for 1 ≤
p0 < p1 ≤ ∞ and 0 < θ < 1, 1 ≤ q ≤ ∞ we have
1 1−θ θ
(L p0 (), L p1 ())θ,q = L p,q (), = + .
p p0 p1
In particular, if p0 < q < p1 , then

p0 p0 −1
(L (), L ())θ,q = L
p0 p1 q,q
() = L () for θ = 1−
q
1− .
q p1
For 1 < p0 < p1 < ∞ and 0 < θ < 1, 1 ≤ q0 ≤ q1 ≤ ∞ we have
(L p0 ,q0 (), L p1 ,q1 ())θ,q = L p,q (),
with
1 1−θ θ
= + .
p p0 p1
Proof. The spaces L pi () are in J1−1/ pi (L 1(),L ∞())∩K 1−1/ pi(L 1 (),
L ∞ ()) for i = 1, 2. This is obvious for pi = 1 or pi = ∞, and it is
true also for pi ∈ (1, ∞), since
L pi () = L pi , pi () = (L 1 (), L ∞ ())1−1/ pi , pi
by Example 1.11. The first statement follows by reiteration.
The same argument yields the characterization of the interpolation
spaces between Lorentz spaces, for 1 ≤ p0 < p1 ≤ ∞.
Examples 1.26, 1.27 have several generalizations. It is possible to cha-
racterize real interpolation spaces between L p spaces or Sobolev spaces
with weights (e.g., [90, Section 1.18.5, Chapter 3], [5]), and between L p
spaces and fractional Sobolev and Besov spaces with variable exponents
(e.g., [4, 62]).
In the case 1/q = (1−θ)/p0 +θ/ p1 , the equality (L p0 (), L p1 ())θ,q =
L () from Example 1.27 may be generalized to (L p0 ( ; X),
q
L p1 (; Y ))θ,q = L q (; (X, Y )θ,q ) for every interpolation couple (X, Y ).
It holds for p0 = p1 too. The particular case where is a real in-
terval may be useful in the study of evolution equations. But for q =
((1 − θ)/ p0 + θ/ p1 )−1 such equality is unknown in general, even for
p0 = p1 . See [26].
1.4.2. Applications: the theorems of Marcinkiewicz

and Stampacchia; regularity in elliptic PDE’s
Let (, µ), (, ν) be two σ -finite measure spaces. Traditionally, a linear
operator T : L 1 () + L ∞ () → L 1 () + L ∞ () is called of weak type
( p, q) if there is M > 0 such that
sup σ (ν{y ∈ : |T f (y)| > σ })1/q ≤ M f L p () ,
σ >0
for all f ∈ L p (). This is equivalent to say that the restriction of T to

L p () is a bounded operator from L p () to L q,∞ (). Indeed, by (1.24),
sup σ (ν{y ∈ : |g(y)| > σ })1/q = sup t 1/q g ∗ (t) = g L q,∞ .
σ >0 t>0
T is said to be of strong type ( p, q) if its restriction to L p () is a bounded

operator from L p () to L q ().
Since L q () = L q,q () ⊂ L q,∞ (), then any operator of strong type
( p, q) is also of weak type ( p, q).
Theorem 1.28. Let T : L 1 () + L ∞ () → L 1 () + L ∞ () be of weak
type ( p0 , q0 ) and ( p1 , q1 ), with constants M0 , M1 respectively, and
1 ≤ p0 , p1 ≤ ∞, 1 < q0 , q1 ≤ ∞,
q0 = q1 , p0 ≤ q0 , p1 ≤ q1 .
Fixed any θ ∈ (0, 1) define p and q by
1 1−θ θ 1 1−θ θ
= + , = + .
p p0 p1 q q0 q1
Then T is of strong type ( p, q), and there is C independent of θ such that
T f L q () ≤ C M01−θ M1θ f L p () , f ∈ L p ().
Proof. For i = 1, 2, T is bounded from L pi () to L qi ,∞ (), with
norm not exceeding Mi . Theorem 1.6 implies that T is bounded from
(L p0 (), L p1 ())θ, p to (L q0 ,∞ (), L q1 ,∞ ())θ, p , and
T L((L p0 (),L p1 ())θ, p ,(L q0 ,∞ (),L q1 ,∞ ())θ, p ) ≤ M01−θ M1θ .
On the other hand, we know from Example 1.27 that
(L p0 (), L p1 ())θ, p = L p, p () = L p (),
and
(L q0 ,∞ (), L q1 ,∞ ())θ, p = L q, p ().
Since p0 ≤ q0 and p1 ≤ q1 then p ≤ q, so that L q, p () ⊂ L q,q () =
L q (). It follows that T is continuous from L p () to L q (), and
T L(L p (),L q ()) ≤ const. M01−θ M1θ .
Theorem 1.28 is slightly less general than the complete Marcinkiewicz

Theorem, which holds also for q0 or q1 = 1.
Since every T of strong type ( p, q) is also of weak type ( p, q) we may
recover a part of the Riesz–Thorin Theorem from Theorem 1.28: we get
that if T is of strong type ( pi , qi ), i = 0, 1, with pi , qi subject to the
restrictions of Theorem 1.28, then T is of strong type ( p, q). The full
Riesz–Thorin Theorem will be proved in Chapter 2.
Let f be a locally integrable function. For each measurable subset
A ⊂ with measure µ(A) ∈ (0, +∞) we define the mean value of f on
A by
1
fA = f (x)µ(d x).
µ(A) A
The space BMO() (BMO stands for bounded mean oscillation) consists
of those locally integrable functions f such that

1
sup | f (x) − f A |µ(dx) < ∞.
0<µ(A)<∞ µ(A) A
If µ() < ∞ we see immediately that L ∞ () ⊂ BMO() ⊂ L 1 ().

It is possible to show that if ⊂ Rn is a Lipschitz bounded domain and
µ is the Lebesgue measure, then BMO() is contained in each L p (),
1 ≤ p < ∞, and that the norms
1/ p
1
f L p + [ f ]BMO, p = f L p + sup | f (x) − f A | dx
p
,
A µ(A) A
1≤ p<∞
are equivalent in BMO(). This was proved by John and Nirenberg in

the well known paper [58] in the case where is a cube. For such ’s
we may extend the result of Example 1.11 to the interpolation couple
(L 1 (), BMO()).
Example 1.29. Let be a bounded domain in Rn with Lipschitz con-
tinuous boundary. Then for 0 < θ < 1, 1 ≤ q ≤ ∞,
1
(L 1 (), BMO())θ,q = L 1−θ ,q ().
The first characterization of real interpolation spaces between L 1 and

BMO is in [53], with the Lebesgue measure in Rn . The method of [53]
was used later in [67] for more general measure spaces (, µ).
As a consequence we get the main part of the Stampacchia interpola-
tion theorem.
Theorem 1.30. Let be a bounded domain in Rn with Lipschitz con-

tinuous boundary, let 1 ≤ r < ∞ and let T ∈ L(L r ()) ∩ L(L ∞ (),
BMO()), or else T ∈ L(L r ()) ∩ L(BMO()). Then T ∈ L(L p ())
for every p ∈ (r, ∞), and
r/ p 1−r/ p
T L(L p ) ≤ CT L(L r ) T L(L ∞ ,BMO)
in the first case,

r/ p 1−r/ p
T L(L p ) ≤ CT L(L r ) T L(BMO)
in the second case.
Proof. In the first case, Theorem 1.6 implies that T is bounded from
(L r , L ∞ )θ, p to (L r , BMO)θ, p for every θ ∈ (0, 1) and p ∈ [1, ∞], and
θ
T L((L r ,L ∞ )θ, p ,(L r ,BMO)θ, p ) ≤ T 1−θ
L(L r ) T L(L ∞ ,BMO) .
r
By Example 1.27, (L r (), L ∞ ())θ, p = L 1−θ , p (). By Example 1.29
and reiteration, we still have
r
(L r (), BMO())θ, p = L 1−θ , p ().
Taking θ = 1 − r/ p (so that r/(1 − θ) = p) gives the first statement

through the equality L p, p () = L p (). The proof of the second state-
ment is similar.
Campanato and Stampacchia ([21]) used the above interpolation theorem

to prove optimal regularity results for elliptic boundary value problems,
as follows.
Let
n
A= Di (ai j (x)D j )
i, j=1
be an elliptic operator with L ∞ coefficients in a bounded open set ⊂

Rn with regular boundary. If f j , j = 0, . . . , n are in L 2 (), a weak
solution of the Dirichlet problem

n

 Au = f 0 + D j f j in ,
j=1


u = 0 in ∂
is any u ∈ H01 () such that for every ϕ ∈ C0∞ () it holds

n
ai j (x)D j u(x)Di ϕ(x)dx
i, j=1

n
=− f 0 (x)ϕ(x)dx + f j (x)D j ϕ(x)dx.
j=1
Using the Lax-Milgram theorem, it is not hard to see that the Dirichlet
problem has a unique weak solution u, and u H 1 ≤ C nj=0 f j L 2 .
This is a first step in several basic courses in PDE’s; see, e.g., [12, Chapter
IX] or [1]. The second step is a regularity theorem: if f j = 0 for each
j + 1, . . . , n and the coefficients ai j ∈ C 1 (), then u ∈ H 2 () and
u H 2 ≤ C f 0 L 2 . See again [12] or [1].
Moreover Campanato in [20, Theorem 16.II] was able to prove the
following:
(i) if the coefficients ai j are in C α () for some α ∈ (0, 1) and the func-
tions f j are in BMO() for j = 0, . . . , n, then
each derivative Di u
belongs to BMO(), and Di uBMO,2 ≤ C nj=0 f j BMO,2 ;
(ii) if the coefficients ai j are in C 1+α () for some α ∈ (0, 1), f j = 0 for
j = 1, . . . , n and f 0 ∈ BMO(), then u ∈ H 2 () satisfies the equa-
tion a.e., each second order derivative Di j u belongs to BMO(), and
Di j uBMO,2 ≤ C f 0 BMO,2 .
In case (i), applying Theorem 1.30 with r = 2 to the operators Ti , i =

1, . . . , n, defined by Ti ( f 0 , . . . , f n ) = Di u, u being the solution of the
Dirichlet problem, we get that if the f j ’s are in L p (), 2 <p < ∞, then
each derivative Di u belongs to L p (), and Di u L p ≤ C nj=0 f j L p .
In case (ii), applying again Theorem 1.30 with r = 2 to the oper-
ators Ti j , i, j = 1, . . . , n, defined by Ti j f 0 = Di j u, we get that if
f 0 ∈ L p (), 2 < p < ∞, then each derivative Di j u belongs to L p (),
and Di j u L p ≤ C f 0 L p .
1.4.3. Exercises
1. Show that for 0 < θ < 1, θ = 1/2,
(Cb (Rn ), Cb2 (Rn ))θ,∞ = Cb2θ (Rn ),
and
(Cb (), Cb2 ())θ,∞ = Cb2θ (),
if ⊂ Rn is any open set with uniformly C 2 boundary. (Hint: prove

that (Cb1 (Rn ), Cb2 (Rn ))α,∞ = Cb1+α (Rn ) using Example 1.8, then use
the Reiteration Theorem with E = Cb1 (Rn )).
2. Show that for k ∈ N and 0 ≤ α < k < β, Cbk (Rn ) belongs to the
β
class J(k−α)/(β−α) between Cbα (Rn ) and Cb (Rn ).
3. Let be a bounded and measurable subset of Rn with the following
property: there exists C > 0 such that |∩ B(x, r)| ≥ Cr n , for each
x ∈ and r < 1. Prove that L ∞ () belongs to the class Jn/(n+α)
between L 1 () and C α (). (Hint: for r < 1 write f (x) = | ∩
B(x, r)|−1 ∩B(x,r) f (x)dy, subtract and add f (y) in the integral
and estimate, to get f ∞ ≤ C(r α f C α + r −n f L 1 ); for r ≥ 1
this estimate is obvious. Then minimize over r > 0).
4. Let 0 < α < 1, and let I ⊂ R be any interval. B(I ; Y ) denotes the
space of the bounded functions from I to Y , endowed with the sup
norm.
(i) Prove that for 0 ≤ θ ≤ 1
B(I ; Y ) ∩ C bθ (I ; X) ⊂ Cbαθ (I, (X, Y )1−α,1 ),
and that there exists C > 0 such that

α
uCbαθ (I,(X,Y )1−α,1 ) ≤ Cu1−α
B(I,Y ) uC θ (I,X) ,
b
for each u ∈ B(I ; Y )∩Cbθ (I ; X). (For θ = 0 we mean Cb0 (I ; X) =

Cb (I ; X)). Show that for θ = 1, the space Cb1 (I ; X) may be
replaced by the space Lip(I ; X).
(ii) Prove that for 0 ≤ θ1 , θ2 ≤ 1,
Cbθ1 (I ; Y ) ∩ Cbθ2 (I ; X) ⊂ Cb(1−α)θ1 +αθ2 (I, (X, Y )α,1 ),
and that there exists C > 0 such that
uC (1−α)θ1 +αθ2 (I,(X,Y ) ≤ C(uC θ1 (I ;Y ) )1−α (uC θ2 (I ;X) )α ,

b α,1 ) b b
for each u ∈ Cbθ1 (I ; Y ) ∩ Cbθ2 (I ; X).

Deduce that if E ∈ J1−α (X,Y), the same results hold with (X,Y )1−α,1
replaced by E.
5. Let α ∈ (0, 1), and let I be any halfline, or I = R.
(i) Prove that for every u ∈ C 1+α (I ; X) ∩ C α (I ; Y ), u is bounded
with values in (X, Y )α,∞ .
(ii) Prove that
C 1+α (I ; X) ∩ C α (I ; Y ) ⊂ Lip(I ; (X, Y )α,∞ ).
(iii) Prove that for 0 < θ < 1, θ = α, and for each p ∈ [1, ∞],
C 1+α (I ; X) ∩ C α (I ; Y ) ⊂ C 1+α−θ (I ; (X, Y )θ, p ).
Let again α ∈ (0, 1).
(iv) Let u ∈ C 1+α ([0, 1]; X) ∩ C α ([0, 1]; Y ). Find γ1 , γ2 ∈ R

such that the function u defined by u (t) = u(t) for 0 ≤ t ≤
1,
u (t) = γ1 u(−t) + γ2 u(−t/2) for −1 ≤ t < 0, belongs to
C 1+α ([−1, 1]; X) ∩ C α ([−1, 1]; Y ). (This is called generalized
reflection method).
(v) Prove that statements (i), (ii), (iii) hold for any interval I ⊂ R.
1
Hints: (i) for t, t + h ∈ I write u (t) = ah (t) + bh (t), ah (t) = 0 (u (t) −
u (t + σ h))dσ , bh (t) = (u(t + h) − u(t))/ h; (ii) use the same decompos-
ition for u(t) − u(s) (note that boundedness of u does not imply meas-
urability, so that (ii) does not follow trivially from (i) by integration); to
prove (iii) use (ii) and reiteration; to prove (v) extend u to I (if I is not
closed) by continuity, then extend it to a bigger interval by generalized
reflection around one of the endpoints of I , then multiply by a smooth
cutoff function to extend it to a halfline; apply the results of (i), (ii), (iii)
to the final extension.
Chapter 2
Complex interpolation
The complex interpolation method is due to Calderon [18]. It works in

complex interpolation couples. It may sound “artificial” compared to the
more “natural” real interpolation method of chapter 1, see next Defin-
itions 2.1 and 2.3. It is in fact an abstraction and a generalization of
the method used in the proof of the Riesz–Thorin interpolation theorem,
which we show below.
The theorem of Riesz–Thorin. Let (, µ), (, ν) be σ -finite measure
spaces. Let 1 ≤ p0 , p1 , q0 , q1 ≤ ∞ and let T : L p0 () + L p1 () →
L q0 () + L q1 () be a linear operator such that
T ∈ L(L p0 (), L q0 ()) ∩ L(L p1 (), L q1 ()).
Then
T ∈ L(L pθ (), L qθ ()), 0 < θ < 1,
with
1 1−θ θ 1 1−θ θ
= + , = + , (2.1)
pθ p0 p1 qθ q0 q1
and setting Mi = T L(L pi (),L qi ()) , i = 0, 1, then
T L(L pθ (),L qθ ()) ≤ M01−θ M1θ .
In the case that some of the pi ’s or the qi ’s is ∞ the statement still holds
if we set as usual 1/∞ = 0.
Proof. The proof is easy if p0 = p1 . If also q0 = q1 there is nothing

to prove; if q0 = q1 we have (see Exercise 7, Section 2.1.3) L q0 () ∩
L q1 () ⊂ L qθ () and
g L qθ () ≤ (g L q0 () )1−θ (g L q1 () )θ , g ∈ L q0 () ∩ L q1 (),
and this implies the statement.

From now on we assume that p0 = p1 , so that in particular pθ < ∞.

We recall that the set of the simple functions (:= finite linear combin-
ations of characteristic functions of measurable sets with finite measure)
a : → C is dense in L p () for every p ∈ [1, +∞), and the set of the
simple functions b : → C is dense in L q (), for every q ∈ [1, +∞).
Moreover, for each measurable function f : → C we have

1
f L q () = sup f (x)b(x) ν(dx)
b L q ()
where b ≡ 0 varies in the set of the simple functions from to C.

Let a : → C be a simple function. Then a ∈ L p () for each
p ∈ [1, +∞]; in particular a ∈ L pθ (). To estimate T a L qθ () we use
the above characterization of the L qθ -norm, estimating the integral

(T a)(x)b(x) ν(dx)

where b : → C is any simple function. To this aim, for every z ∈ S :=

{z = x + i y ∈ C : 0 ≤ x ≤ 1}, we define



 pθ 1−z z
p0 + p1
a(x)
 |a(x)| , if x ∈ , a(x) = 0,
f (z)(x) = |a(x)|


 0, if x ∈ , a(x) = 0;



 qθ 1−z
q0
+ z b(x)
 |b(x)| q1
, if x ∈ , b(x) = 0,
g(z)(x) = |b(x)|



0, if x ∈ , b(x) = 0.
if qθ = ∞ (i.e., (q0 , q1 ) = (1, 1)), and
g(z)(x) = b(x),
if qθ = ∞.1
Then for each z ∈ S, f (z) ∈ L p () for every p, and g(z) ∈ L q () for
every q. In particular, f (z) ∈ L p0 () ∩ L p1 () so that T f (z) ∈ L q0 ()
∩ L q1 (), and the function

F(z) = T f (z) g(z) ν(dx)

1 We recall that q is the conjugate exponent of q ∈ [1, ∞] defined by 1/q + 1/q = 1.

is well defined, holomorphic

in the interior of S, continuous and bounded
in S, and F(θ) = (T a)(x)b(x) ν(dx) is the integral that we want to
estimate. For every t ∈ R we have
|F(it)| ≤ T f (it) L q0 () g(it) L q0 ()
p /p qθ /q0
≤ T L(L p0 (),L q0 ()) a Lθpθ ()
0
b ,
L qθ ()
|F(1 + it)| ≤ T f (1 + it) L q1 () g(1 + it) L q1 ()
p /p qθ /q1
≤ T L(L p1 (),L q1 ()) a Lθpθ ()
1
b .
L qθ ()
By the three lines theorem (see Exercise 2, Section 2.1.3) we get

T a b ν(dx) = |F(θ)| ≤ (sup |F(it)|)1−θ (sup |F(1 + it)|)θ

t∈R t∈R
θ
≤ T 1−θ
L(L p0 (),L q0 ()) T L(L p1 (),L q1 ()) a L θ () b L qθ () .
p
Since T a L qθ () is the supremum of |F(θ)|/b L qθ () when b ≡ 0 runs

in the set of the simple functions on , we get
θ
T a L qθ () ≤ T 1−θ
L(L p0 (),L q0 ()) T L(L p1 (),L q1 ()) a L θ () ,
p
for every simple function a : → C. Since the set of such a’s is dense
in L pθ () the statement follows.
Taking in particular pi = qi , we get that if T ∈ L(L p0 (), L p0 ()) ∩
L(L p1 (), L p1 ()) then T ∈ L(L r (), L r ()) for every r ∈ [ p0 , p1 ].
The crucial part of the proof is the use of the three lines theorem for
the function F. The explicit expression of F is not important; what is
important is that F is holomorphic in the interior of S, continuous and
bounded in S, that F(θ) leads to the norm T a L qθ () , and that the beha-
vior of F in iR and in 1+iR is controlled. Banach space valued functions
of this type are precisely those used in the construction of the complex
interpolation spaces.
2.1. Definitions and properties

Throughout the section we shall use the maximum principle for holo-
morphic functions with values in a complex Banach space X: if is a
bounded open subset of C and f : → X is holomorphic in and

continuous in , then f (ζ ) X ≤ max{ f (z) X : z ∈ ∂}, for every
ζ ∈ . This is well known if X = C, and may be recovered if X is a
general Banach space by the following argument. For every ζ ∈ let
x ∈ X be such that f (ζ ) X = f (ζ ), x and x X = 1. Applying
the maximum principle to the complex function z → f (z), x we get
f (ζ ) X = | f (ζ ), x | ≤ max{| f (z), x | : z ∈ ∂}
≤ max{ f (z) X : z ∈ ∂}.

The maximum principle holds also for functions defined in strips. Deal-
ing with complex interpolation, we shall consider the strip
S := {z = x + i y ∈ C : 0 ≤ x ≤ 1}.
If f : S → X is holomorphic in the interior of S, continuous and
bounded in S, then for each ζ ∈ S
f (ζ ) X ≤ max{sup f (it) X , sup f (1 + it) X }.
t∈R t∈R
See Exercise 1, Section 2.1.3.

Let (X, Y ) be an interpolation couple of complex Banach spaces.
Definition 2.1. Let S be the strip {z = x + i y ∈ C : 0 ≤ x ≤ 1}.
F(X, Y ) is the space of all functions f : S → X + Y such that
(i) f is holomorphic in the interior of S and continuous and bounded up
to its boundary, with values in X + Y ;
(ii) t → f (it) ∈ C b (R; X), t → f (1 + it) ∈ Cb (R; Y ), and
f F (X,Y ) = max{sup f (it) X , sup f (1 + it)Y } < ∞.
t∈R t∈R
F0 (X, Y ) is the subspace of F(X, Y ) consisting of the functions f : S →

X + Y such that
lim f (it) X = 0, lim f (1 + it)Y = 0.
|t|→∞ |t|→∞
Then F(X, Y ) and F0 (X, Y ) are Banach spaces, continuously embedded

in Cb (S, X + Y ). Indeed, if { f n } is a Cauchy sequence, the maximum
principle gives, for all z ∈ S,
f n (z) − f m (z) X+Y
≤ max{supt∈R ( f n − f m )(it) X+Y , supt∈R ( f n − f m )(1 + it) X+Y }
≤ max{supt∈R ( f n − f m )(it) X , supt∈R ( f n − f m )(1 + it)Y }.

Therefore for every z ∈ S there exists f (z) = limn→∞ f n (z) in X + Y ;

the limit is uniform for z ∈ S so that f ∈ F(X, Y ). Since t → f n (it)
converges in Cb (R; X) and t → f n (1 + it) converges in Cb (R; Y ), then
f n converges to f in F(X, Y ). Moreover, since F0 (X, Y ) is closed in
F(X, Y ), then F0 (X, Y ) is a Banach space too.
An important technical lemma about the space F0 (X, Y ) is the follow-
ing one. Its intricate proof is due to Calderon [18, page 132–133]. A
very detailed proof is in the book of Krein, Petunin, and Semenov [63,
page 217–220].
2 +λz
Lemma 2.2. The linear hull of the functions eδz a, δ > 0, λ ∈ R,
a ∈ X ∩ Y , is dense in F0 (X, Y ).
The complex interpolation spaces [X, Y ]θ are defined through the traces
on the real axis of the functions in F(X, Y ).
Definition 2.3. For every θ ∈ [0, 1] set
[X, Y ]θ := { f (θ) : f ∈ F(X, Y )},
a[X,Y ]θ := inf f F (X,Y ) .

f ∈F (X,Y ), f (θ)=a
[X, Y ]θ is isomorphic to the quotient space F(X, Y )/Nθ , where Nθ is

the subset of F(X, Y ) consisting of the functions which vanish at z = θ.
Since Nθ is closed, the quotient space is a Banach space and so is [X, Y ]θ .
Some immediate consequences of the definition are listed below.
(i) For every θ ∈ (0, 1),
[X, Y ]θ = [Y, X]1−θ .
(ii) We get an equivalent definition of [X,Y]θ replacing the space F(X,Y)

by the space F0 (X, Y ). Indeed, for each f ∈ F(X, Y ) and δ > 0
2
the function f δ (z) = eδ(z−θ) f (z) is in F0 (X, Y ), f δ (θ) = f (θ),
2 2
and f δ F (X,Y ) ≤ max{eδθ , eδ(1−θ) } f F (X,Y ) . Letting δ → 0 we
obtain also
inf f F (X,Y ) = inf f F (X,Y ) .

f ∈F (X,Y ), f (θ)=a f ∈F0 (X,Y ), f (θ)=a
(iii) If Y = X then [X, X]θ = X, with identical norms (see Exercise 3,

Section 2.1.3).
(iv) For every t ∈ R and for every f ∈ F(X, Y ), then f (θ + it) ∈
[X, Y ]θ for each θ ∈ (0, 1), and f (θ + it)[X,Y ]θ = f (θ)[X,Y ]θ
(this is easily seen replacing the function f by g(z) = f (z + it)).
Finally, from Lemma 2.2 it follows that X ∩ Y is dense in [X, Y ]θ for

every θ ∈ (0, 1). In the present chapter, this fact will be used only in
Example 2.11.
Proposition 2.4. Let 0 < θ < 1. Then
X ∩ Y ⊂ [X, Y ]θ ⊂ X + Y.
Proof. Let a ∈ X ∩ Y . The constant function f (z) = a belongs to

F(X, Y ), and
f F (X,Y ) ≤ max{a X , aY }.
Therefore, a = f (θ) ∈ [X, Y ]θ and a[X,Y ]θ ≤ a X∩Y .
The embedding [X, Y ]θ ⊂ X + Y follows again from the maximum
principle: if a = f (θ) with f ∈ F(X, Y ) then
a X+Y ≤ max{sup f (it) X+Y , sup f (1 + it) X+Y }

t∈R t∈R
≤ max{sup f (it) X , sup f (1 + it)Y } = f F (X,Y )
t∈R t∈R
so that a X+Y ≤ a[X,Y ]θ .

Remark 2.5. Let V(X, Y ) be the linear hull of the functions of the type
ϕ(z)x, with ϕ ∈ F0 (C, C) and x ∈ X ∩Y . If a ∈ X ∩Y , its [X, Y ]θ -norm
may be obtained also as
a[X,Y ]θ = inf f F (X,Y ) . (2.2)

f ∈V (X,Y ), f (θ)=a
Proof. For each ε > 0 let f 0 ∈ F0 (X, Y ) be such that f 0 (θ) = a and
f 0 F (X,Y ) ≤ a[X,Y ]θ + ε. Let z → r(z) be a function continuous in
S and holomorphic in the interior of S with values in the unit disk, and
such that r(θ) = 0, r (θ) = 0, r(z) = 0 for z = θ. For instance, we may
take
z−θ
r(z) = , z ∈ S.
z+θ
Set 2
f 0 (z) − e(z−θ) a
f 1 (z) = , z ∈ S.
r(z)
Then f 1 ∈ F0 (X, Y ), so that by Lemma 2.2 there exists a function

n
f 2 (z) = exp(δk z 2 + γk z)xk ,
k=1
with δk > 0, γk ∈ R, xk ∈ X ∩ Y , such that f 1 − f 2 F (X,Y ) ≤ ε. Set

2
f (z) = e(z−θ) a + r(z) f 2 (z), z ∈ S.
Then f ∈ V(X, Y ) and
f F (X,Y ) ≤ f 0 F (X,Y ) + f − f 0 F (X,Y ) ≤ a[X,Y ]θ + 2ε.
Remark 2.5 will be used in Theorem 2.7 and in Theorem 4.17.

We prove now that the spaces [X, Y ]θ are interpolation spaces.
Theorem 2.6. Let (X 1 , Y1 ), (X 2 , Y2 ) be complex interpolation couples.

If a linear operator T belongs to L(X 1 , X 2 ) ∩ L(Y1 , Y2 ), then the re-
striction of T to [X 1 , Y1 ]θ belongs to L([X 1 , Y1 ]θ , [X 2 , Y2 ]θ ) for every
θ ∈ (0, 1). Moreover,
T L([X 1 ,Y1 ]θ ,[X 2 ,Y2 ]θ ) ≤ (T L[X 1 ,X 2 ] )1−θ (T L[Y1 ,Y2 ] )θ . (2.3)
Proof. First let T L(X 1 ,X 2 ) = 0 and T L(Y1 ,Y2 ) = 0. If a ∈ [X 1 , Y1 ]θ ,

let f ∈ F(X 1 , Y1 ) be such that f (θ) = a. Set
z−θ
T L(X 1 ,X 2 )
g(z) = T f (z), z ∈ S.
T L(Y1 ,Y2 )
Then g ∈ F(X 2 , Y2 ), and
g(it) X 2 ≤ (T L(X 1 ,X 2 ) )1−θ (T L(Y1 ,Y2 ) )θ f (it) X 1 ,
g(1 + it)Y2 ≤ (T L(X 1 ,X 2 ) )1−θ (T L(Y1 ,Y2 ) )θ f (1 + it)Y1 ,
so that gF (X 2 ,Y2 ) ≤ (T L(X 1 ,X 2 ) )1−θ (T L(Y1 ,Y2 ) )θ f F (X 1 ,Y1 ) . There-
fore T a = g(θ) ∈ [X 2 , Y2 ]θ , and
T a[X 2 ,Y2 ]θ ≤ gF (X 2 ,Y2 ) ≤ (T L(X 1 ,X 2 ) )1−θ (T L(Y1 ,Y2 ) )θ f F (X 1 ,Y1 ) .
Taking the infimum over all f ∈ F(X 1 , Y1 ) we get
T a[X 2 ,Y2 ]θ ≤ (T L(X 1 ,X 2 ) )1−θ (T L(Y1 ,Y2 ) )θ |a[X 1 ,Y1 ]θ .
If either T L(X 1 ,X 2 ) or T L(Y1 ,Y2 ) vanishes, replace it by ε > 0 in the

definition of g and then let ε → 0 to get the statement.
If both T L(X 1 ,X 2 ) and T L(Y1 ,Y2 ) vanish, set g(z) = T f (z).
As in the case of real interpolation, it is natural to ask whether T :

[X 1 , Y1 ]θ → [X 2 , Y2 ]θ is compact when T : X 1 → X 2 or T : Y1 → Y2 is
compact. There are answers in special cases but the question is still open
in general. See the review paper [29].
Theorem 2.6 has an interesting extension to linear operators depending
on z ∈ S.
Theorem 2.7. Let (X 1 , Y1 ), (X 2 , Y2 ) be complex interpolation couples.
For every z ∈ S let Tz ∈ L(X 1 ∩ Y1 , X 2 + Y2 ) be such that for every
x ∈ X 1 ∩ Y1 the function z → Tz x is holomorphic in the interior of
S and continuous and bounded in S with values in X 2 + Y2 , and the
functions t → Tit x, t → T1+it x are continuous in R with values in X 2 ,
Y2 , respectively. Assume in addition that there exist real numbers M0 ,
M1 ≥ 0 such that
Tit x X 2 ≤ M0 x X 1 , T1+it xY2 ≤ M1 xY1 , x ∈ X 1 ∩ Y1 , t ∈ R.
Then for every θ ∈ (0, 1) we have Tθ x ∈ [X 2 , Y2 ]θ and
Tθ x[X 2 ,Y2 ]θ ≤ M01−θ M1θ x[X 1 ,Y1 ]θ
so that Tθ has an extension belonging to L([X 1 , Y1 ]θ , [X 2 , Y2 ]θ ) (which

we still call Tθ ) satisfying
Tθ L([X 1 ,Y1 ]θ ,[X 2 ,Y2 ]θ ) ≤ M01−θ M1θ . (2.4)
Proof. The proof is just a modification of the proof of Theorem 2.6.

Assume first that M0 and M1 are positive. For every a ∈ X 1 ∩ Y1 let
f ∈ V(X 1 , Y1 ) be such that f (θ) = a (see Remark 2.5), and set

M0 z−θ
g(z) = Tz f (z), z ∈ S.
M1
Then g ∈ F(X 2 , Y2 ) and
g(it) X 2 ≤ M01−θ M1θ f (it) X 1 ,
g(1 + it)Y2 ≤ M01−θ M1θ f (1 + it)Y1 ,
so that gF (X 2 ,Y2 ) ≤ M01−θ M1θ f F (X 1 ,Y1 ) . Therefore Tθ a = g(θ) be-

longs to the space [X 2 , Y2 ]θ , and
Tθ a[X 2 ,Y2 ]θ ≤ M01−θ M1θ inf f F (X 1 ,Y1 ) .

f ∈V (X 1 ,Y1 ), f (θ)=a
On the other hand by Remark 2.5 we have
a[X 1 ,Y1 ]θ = inf f F (X 1 ,Y1 ) ,

f ∈V (X 1 ,Y1 ), f (θ)=a

If either M0 or M1 vanishes, replace it by ε in the definition of g, and
then let ε → 0. If both M0 and M1 vanish, define g by g(z) = Tz f (z)
and follow the above arguments.
Let us come back to Theorem 2.6 and to its consequences. The same
proof of Corollary 1.7 (through the equality [C, C]θ = C with the same
norm) yields
Corollary 2.8. For every θ ∈ (0, 1) we have

θ
y[X,Y ]θ ≤ y1−θ
X yY , ∀y ∈ X ∩ Y. (2.5)
Therefore, [X, Y ]θ ∈ Jθ (X, Y ). By Proposition 1.20, this means that

(X, Y )θ,1 ⊂ [X, Y ]θ . It is also true that [X, Y ]θ ⊂ (X, Y )θ,∞ ; to prove
this embedding we use the following lemma, which gives a Poisson for-
mula for holomorphic functions in a strip with values in Banach spaces.
Lemma 2.9. For every bounded f : S → X which is continuous in S

and holomorphic in the interior of S we have
f (z) = f 0 (z) + f 1 (z), z = x + i y ∈ S,
where

f (it)
f 0 (z) = eπ(y−t) sin(π x) dt,
R sin (π x) + (cos(π x) − exp(π(y − t)))2
2

f (1 + it)
f 1 (z) = eπ(y−t) sin(π x) dt.
R sin (π x) + (cos(π x) + exp(π(y − t)))2
2
(2.6)
Sketch of the proof. Let first X = C. Then (2.6) may be obtained using
the Poisson formula for the unit circle,

1 1 − |ζ |2
f (ζ ) = f (λ) dλ, |ζ | < 1,
2π |λ|=1 |ζ − λ|2
(which holds for every f which is holomorphic in the interior and con-
tinuous up to the boundary), and the conformal mapping
eπi z − i
ζ (z) = , z ∈ S,
eπi z + i
which transforms S into the unit circle. If X is a general Banach space
(2.6) follows as usual, considering the complex functions z → f (z), x
for every x ∈ X , and applying (2.6) to each of them.
Proposition 2.10. For each θ ∈ (0, 1), [X, Y ]θ ∈ K θ (X, Y ), that is

[X, Y ]θ is continuously embedded in (X, Y )θ,∞ .
Proof. Let a ∈ [X, Y ]θ . For every f ∈ F(X, Y ) such that f (θ) = a split
a = f 0 (θ) + f 1 (θ) according to (2.6).
Note that for z = x + i y ∈ S we have

1
0< eπ(y−t) sin(π x) 2 dt < 1,
R sin (π x) + (cos(π x) − exp(π(y − t)))2

1
0< eπ(y−t) sin(π x) 2 dt < 1.
R sin (π x) + (cos(π x) + exp(π(y − t)))2
Indeed, both kernels are positive so that both integrals are positive;
moreover if f is holomorphic in S, continuous and bounded in S and
f ≡ 1 on iR and on 1 + iR then f ≡ 1 in S, so that the sum of the
integrals is 1.
Therefore for every z ∈ S, f 0 (z) ∈ X, f 0 (z) X ≤ supτ ∈R f (iτ ) X ,
and f 1 (z) ∈ Y , f 1 (z)Y ≤ supτ ∈R f (1 + iτ )Y . Then for each t > 0
we get
K (t, a) ≤ f 0 (θ) X + t f 1 (θ)Y ≤ sup f (iτ ) X + t sup f (1 + iτ )Y .

τ ∈R τ ∈R
The function g(z) = t θ−z f (z) is in F(X, Y ), and g(θ) = a. Applying

the above estimate to g we get
K (t, a) ≤ sup f (iτ ) X t θ + t sup f (1 + iτ )Y t θ−1 ≤ 2t θ f F (X,Y ) .

τ ∈R τ ∈R
Since f is arbitrary,
K (t, a) ≤ 2t θ a[X,Y ]θ ,

By Corollary 2.8 and Proposition 1.3, [X, Y ]θ ∈ Jθ (X, Y ) ∩ K θ (X, Y ).

This implies, through Proposition 1.4, that [X, Y ]θ2 ⊂ [X, Y ]θ1 for
θ1 < θ2 whenever Y ⊂ X.
This also allows to use the Reiteration Theorem to characterize the real
interpolation spaces between complex interpolation spaces. We get, for
0 < θ1 < θ2 < 1, 0 < θ < 1, 1 ≤ p ≤ ∞,
([X, Y ]θ1 , [X, Y ]θ2 )θ, p = (X, Y )(1−θ)θ1 +θθ2 , p .
Further reiteration properties are the following. Calderon ([18]) showed

that if one of the spaces X, Y is continuously embedded in the other one,
or if X, Y are reflexive and X ∩ Y is dense both in X and in Y , then
[[X, Y ]θ1 , [X, Y ]θ2 ]θ = [X, Y ](1−θ)θ1 +θθ2 ,
Lions ([68]) proved that if X and Y are reflexive, then for 0 < θ1 < θ2 <
1, 0 < θ < 1, 1 < p < ∞,
[(X, Y )θ1 , p , (X, Y )θ2 , p ]θ = (X, Y )(1−θ)θ1 +θθ2 , p .
The question whether [X, Y ]θ coincides with some (X, Y )θ, p has no gen-
eral answer. We will see at the end of Chapter 4 that if X and Y are
Hilbert spaces then
[X, Y ]θ = (X, Y )θ,2 , 0 < θ < 1,
but in the non hilbertian case there are no general rules. See next Ex-
amples 2.11 and 2.12.
Concerning duality, Calderon ([18]) showed that if X ∩ Y is dense in
X and in Y , and one of the spaces X, Y is reflexive, then
([X, Y ]θ ) = [X , Y ]θ , θ ∈ (0, 1).
Without reflexivity, the embedding ⊃ still holds. See Exercise 6, Sec-

tion 2.1.3.
2.1.1. Examples
Example 2.11. Let (, µ) be a measure space with σ -finite measure,
and let 1 ≤ p0 , p1 ≤ ∞, 0 < θ < 1. Then
1 1−θ θ
[L p0 (), L p1 ()]θ = L p (), = + ,
p p0 p1
and their norms coincide. (In the case p0 = ∞ or p1 = ∞ the statement
is correct if we set as usual 1/∞ = 0).
Proof. The proof follows the proof of the Riesz–Thorin theorem at the
beginning of the chapter.
Let a : → C be a simple function; we may assume without loss of
generality that a L p = 1. For z ∈ S set

p 1−z z
p0 + p1
a(x)
f (z)(x) := |a(x)| , if x ∈ , a(x) = 0,
|a(x)|
f (z)(x) := 0, if x ∈ , a(x) = 0.
Then f is continuous in S and holomorphic in the interior of S with
values in L p0 () + L p1 (), and for each t ∈ R we have
p
p/ p0
| f (it)(x)| = |a(x)| p0 , f (it) L p0 ≤ a L p = 1,
p
p/ p1
| f (1 + it)(x)| = |a(x)| ,p1
f (1 + it) L p1 ≤ a L p = 1.
Moreover, t → f (it) is continuous with values in L p0 () and t →
f (1 + it) is continuous with values in L p1 (). Therefore, f belongs to
F(L p0 (), L p1 ()) and f F (L p0 ,L p1 ) ≤ 1. Since f (θ) = a, then
a[L p0 ,L p1 ]θ ≤ 1 = a L p . (2.7)
Since p < +∞, the set S() of the simple functions is dense in L p ();
therefore inequality (2.7) may be extended to every a ∈ L p () and it
shows that L p () is continuously embedded in [L p0 , L p1 ]θ .
To prove the other inclusion we remark that

1 = a L p = sup a(x)b(x)µ(dx) : b ∈ S(), b L p = 1 .

For every b ∈ S() with b L p = 1 set, as before,

p 1−z
+ z b(x)
g(z)(x) := |b(x)| p0 p1 , if x ∈ , b(x) = 0,
|b(x)|
g(z)(x) := 0, if x ∈ , b(x) = 0,
and define, for every f ∈ F(L , L p1 ) such that f (θ) = a,
p0

F(z) := f (z)(x)g(z)(x)µ(dx), z ∈ S.

Then F is holomorphic in the interior of S and continuous in S, so that

the maximum principle (see exercise 2, Section 2.1.3) implies
|F(z)| ≤ max{sup |F(it)|, sup |F(1 + it)|}, z ∈ S.

t∈R t∈R
But |F(it)| and |F(1 + it)| may be easily estimated:

p / p0
|F(it)| ≤ f (it) L p0 g(it) L p0 = f (it) L p0 b L p = f (it) L p0 ,
|F(1 + it)| ≤ f (1 + it) L p1 g(1 + it) L p1
p / p1
= f (1 + it) L p1 b L p = f (1 + it) L p1 ,
so that
|F(z)| ≤ max{supt∈R f (it) L p0 , supt∈R f (1 + it) L p1 }
≤ f F (L p0 ,L p1 ) , z ∈ S.
Therefore,

a(x)b(x)µ(dx) = |F(θ)| ≤ f F (L p0 ,L p1 ) .

Since b is arbitrary,
a L p ≤ f F (L p0 ,L p1 ) .
Since f is arbitrary,
a L p ≤ a[L p0 ,L p1 ]θ . (2.8)
This implies that [L p0 , L p1 ]θ is continuously embedded in L p (). Es-
timates (2.7) and (2.8) show that [L p0 , L p1 ]θ and L p () have the same
norm.
Example 2.12. For 0 < θ < 1, 1 < p < ∞, m ∈ N,
[L p (Rn ), W m, p (Rn )]θ = H mθ, p (Rn ).
The proof is in [90, Section 2.4.2].

We recall that for s > 0
H s, p (Rn ) := { f ∈ L p (Rn ) : f H s, p := F −1 (1 + |x|2 )s/2 F f L p < ∞},
where F is the Fourier transform. It is known that if s = k is integer then
H k, p (Rn ) = W k, p (Rn ), k ∈ N.
Moreover it is known that

B sp, p (Rn ) ⊂ H s, p (Rn ) ⊂ B sp,2 (Rn ), 1 < p ≤ 2,
B sp,2 (Rn ) ⊂ H s, p (Rn ) ⊂ B sp, p (Rn ), 2 ≤ p < ∞,

and the inclusions are strict if p = 2. See [90, Section 2.3.3]. We recall
(Example 1.10) that (L p (Rn ), W m, p (Rn ))θ, p = B mθ
p, p (R ). Therefore
n
[L p (Rn ), W m, p (Rn )]θ = (L p (Rn ), W m, p (Rn ))θ, p

unless p = 2.
Instead, the complex interpolation spaces [Cb (Rn ), Cbm (Rn )]θ are not
known. In the paper [28] it is shown that [C([0, 1]), Lip([0, 1])]θ is not a
Hölder or little-Hölder space.
Examples 2.11 and 2.12 have several generalizations. The paper [83]
and its longer version [84] contain lots of results about complex interpo-
lation between weighted L p , Besov and Lizorkin-Triebel spaces. Charac-
terizations of complex interpolation between L p spaces and Besov spaces
with variable exponents are in [36, 4].
2.1.2. The theorems of Hausdorff–Young and Stein

Applying Theorem 2.6 to the spaces X 1 = L p0 (), X 2 = L q0 (), Y1 =
L p1 (), Y2 = L q1 () and recalling Example 2.11 we get the Riesz–
Thorin Theorem, as stated at the beginning of the chapter. However,
the proof of Example 2.11 is modeled on the proof of the Riesz–Thorin
Theorem, so that this has not to be considered an alternative proof.
An important application of the Riesz–Thorin theorem (or, equival-
ently, of Theorem 2.6 and Example 2.11) is the theorem of Hausdorff
and Young on the Fourier transform in L p (Rn ). We set, for every f ∈
L 1 (Rn ),

1
(F f )(k) := e−ix,k f (x)dx, k ∈ Rn .
(2π)n/2 Rn
As easily seen, F f L 2 = f L 2 for every f ∈ C0∞ (Rn ) (the space of
the smooth functions in Rn with compact support), so that F is canonic-
ally extended to an isometry (still denoted by F) to L 2 (Rn ).
Theorem 2.13. If 1 < p ≤ 2, F is a bounded operator from L p (Rn ) to

L p (Rn ), p = p/( p − 1), and
1
FL(L p ,L p ) ≤ .
(2π)n(1/ p−1/2)
Proof. Since FL(L 1 ,L ∞ ) ≤ (2π)−n/2 and FL(L 2 ) = 1, by the Riesz–
Thorin theorem F ∈ L(L pθ , L qθ ) for every θ ∈ (0, 1), pθ and qθ being
defined by (2.1): pθ = 2/(1 + θ), qθ = 2/(1 − θ) = pθ . Moreover,
2/ p−1
θ 1
FL(L pθ ,L pθ ) ≤ FL(L 1 ,L ∞ ) FL(L 2 ) ≤
1−θ
. (2.9)
(2π)n/2
When θ runs in (0, 1), pθ = 2/(1 + θ) runs in (1, 2), and the statement
is proved.
A useful generalization of the Riesz-Thorin theorem is the Stein inter-
polation theorem. It is obtained by applying Theorem 2.7 to the interpol-
ation couples (L p0 (), L p1 ()), (L q0 (), L q1 ()), using the character-
ization of Example 2.11.
Theorem 2.14. Let (, µ), (, ν) be σ -finite measure spaces. Let p0 ,
p1 , q0 , q1 ∈ [1, ∞] and define as usual pθ , qθ by
1 1−θ θ 1 1−θ θ
= + , = + , 0 < θ < 1.
pθ p0 p1 qθ q0 q1
Assume that for every z ∈ S, Tz : L p0 () ∩ L p1 () → L q0 () + L q1 ()
is a linear operator such that
(i) for each f ∈ L p0 () ∩ L p1 (), z → Tz f is holomorphic in the
interior of S, continuous and bounded in S, with values in L q0 () +
L q1 ();
(ii) for each f ∈ L p0 ()∩L p1 (), t → Tit f is continuous and bounded
in R with values in L q0 (), t → T1+it f is continuous and bounded
in R with values in L q1 ();
(iii) there are M0 , M1 > 0 such that for each f ∈ L p0 () ∩ L p1 (),
sup T (t) f L q0 () ≤ M0 f L p0 () ,
t∈R
sup T (t) f L q1 () ≤ M1 f L p1 () .

t∈R
Then for each θ ∈ (0, 1) and for each f ∈ L p0 () ∩ L p1 () we have
Tθ f L qθ () ≤ M01−θ M1θ f L pθ () .
Therefore, Tθ may be extended to a bounded operator (which we still call

Tθ ) from L pθ () to L qθ (), with pθ and qθ defined in (2.1), and
Tθ L(L pθ (),L qθ ()) ≤ M01−θ M1θ .
If ( p0 , p1 ) = (∞, ∞), Theorem 2.14 has a slightly sharper version,

stated below, obtained modifying the direct proof of the Riesz–Thorin
theorem. We recall that if (, µ) is a measure space, a simple function
is a (finite) linear combination of characteristic functions of measurable
sets with finite measure.
Theorem 2.15. Let (, µ), (, ν) be σ -finite measure spaces. Assume
that for every z ∈ S, Tz is a linear operator defined in the set of the simple
functions on , with values into measurable functions on , such that for
every couple of simple functions a : → C and b : → C, the product
Tz a · b is integrable on and

z → (Tz a)(x)b(x)ν(dx),

is continuous and bounded in S, holomorphic in the interior of S.

Assume moreover that for some p j , q j ∈ [1, +∞], j = 0, 1, ( p0 , p1 ) =
(∞, ∞) we have
Tit a L q0 () ≤ M0 a L p0 () , T1+it a L q1 () ≤ M0 a L p1 () , t ∈ R,
for every simple function a. Then for each θ ∈ (0, 1), Tθ may be extended
to a bounded operator (which we still call Tθ ) from L pθ () to L qθ (),
with pθ and qθ defined in (2.1), and
Tθ L(L pθ (),L qθ ()) ≤ M01−θ M1θ .
Proof. The proof is just a modification of the proof of the Riesz–Thorin

theorem. For every couple of simple functions a : → C, b : → C,
we apply the three lines theorem to the function

F(z) = Tz f (z) g(z) ν(dx)

where f and g are defined as in the proof of the Riesz–Thorin theorem,

i.e.

p 1−z + z a(x)
f (z)(x) = |a(x)| θ p0 p1 , if x ∈ , a(x) = 0;
|a(x)|
f (z)(x) = 0, if x ∈ , a(x) = 0,
and

qθ 1−z
+ z b(x)
q0
g(z)(x) = |b(x)| q1
, if x ∈ , b(x) = 0,
|b(x)|
g(z)(x) = 0, if x ∈ , b(x) = 0,
if qθ < ∞,
g(z)(x) = b(x)
if qθ = ∞ (i.e. q0 = q1 = 1). We get

(Tθ a)(x)b(x) ν(dx) = |F(θ)| ≤ M 1−θ M θ a L pθ () b q ,
0 1 L θ ()

so that
T a L qθ () ≤ M01−θ M1θ a L pθ () ,
for every simple a defined in . The assumption ( p0 , p1 ) = (∞, ∞)
implies that pθ < ∞. Since the simple functions are dense in L pθ () the
statement follows.
2.1.3. Exercises
1. The maximum principle for functions defined on a strip. Let f : S →
X be holomorphic in the interior of S, continuous and bounded in S.
Prove that for each ζ ∈ S
f (ζ ) ≤ max{sup f (it), sup f (1 + it)}.

t∈R t∈R
(Hint: for each ε ∈ (0, 1) let z 0 be such that f (z 0 ) ≥ f ∞ (1 − ε);

consider the functions f δ (z) = exp(δ(z − z 0 )2 ) f (z), and apply the
maximum principle in the rectangle [0, 1] × [−M, M] with M large).
2. The three lines theorem. Let f : S → X be holomorphic the interior
of S, continuous and bounded in S. Show that
f (θ) X ≤ (sup f (it) X )1−θ (sup f (1 + it) X )θ , 0 < θ < 1.

t∈i R t∈i R
This estimate implies that if f vanishes in iR or in 1 + iR then f

vanishes in S.
(Hint: apply the maximum principle of exercise 1 to ϕ(z) = eλz f (z)
and then choose λ > 0 properly).
3. Show that [X, X]θ = X, with identical norms.
4. If (X, Y ) is a complex interpolation couple, for every γ ∈ R define
the space Fγ (X, Y ) as F(X, Y ) with (ii) of Definition 2.1 replaced by
the conditions
sup e−γ |z| f (z) X+Y < ∞,
z∈S
and
f Fγ (X,Y )= max{sup e−γ |t| f (it) X , sup e−γ |1+it| f (1+it)Y } < ∞.
t∈R t∈R
Prove that for each θ ∈ (0, 1), [X, Y ]θ coincides with the space of
the traces at z = θ of the functions f ∈ Fγ (X, Y ), and its norm is
equivalent to x → inf{ f Fγ (X,Y ) : f (θ) = x}.
5. Let (X, Y ) be any complex interpolation couple. Set [X, Y ]0 := X,
[X, Y ]1 := Y . Show that for 0 ≤ θ0 < θ1 ≤ 1 and 0 < θ < 1 we have
[X, Y ](1−θ)θ0 +θθ1 ⊂ [[X, Y ]θ0 , [X, Y ]θ1 ]θ .
(Hint: if x = f ((1 − θ)θ0 + θθ1 ) set g(z) = f ((1 − z)θ0 + zθ1 )).
6. Show that if X ∩ Y is dense in X and in Y , then for each θ ∈ (0, 1)
we have

[X, Y ]θ ⊂ [X , Y ]θ .
(Hint: for x ∈ X ∩ Y and x ∈ X ∩ Y write x = f (θ), with f and
f of the special type given by Lemma 2.2, and estimate |x (x)| using
the three lines theorem of exercise 2. Then extend using density.)
7. Let (, ν) be any measure space, let q0 = q1 ∈ [1, +∞], and for
every θ ∈ (0, 1) define qθ by
1 1−θ θ
= + .
qθ q0 q1
Prove that L q0 () ∩ L q1 () ⊂ L qθ (), and
g L qθ () ≤ (g L q0 () )1−θ (g L q1 () )θ , g ∈ L q0 () ∩ L q1 (),
for every g ∈ L q0 ()∩ L q1 (). (This result has been used in the proof
of the Riesz-Thorin Theorem.)
Chapter 3
Interpolation and domains of operators
3.1. Operators with rays of minimal growth

Let X be a real or complex Banach space with norm · . In this section
we consider a linear operator A : D(A) ⊂ X → X such that
ρ(A) ⊃ (0, ∞), ∃M : λR(λ, A)L(X) ≤ M, λ > 0. (3.1)
Since ρ(A) is not empty, then A is a closed operator, so that D(A) is a

Banach space with the graph norm x D(A) = x + Ax. Moreover
for every m ∈ N also Am is a closed operator (see Exercise 1, Section
3.1.1).
This section is devoted to the study of the real interpolation spaces
(X, D(A))θ, p , and, more generally, (X, D(Am ))θ, p . Since the graph norm
of D(Am ) is equivalent to the graph norm of D(B m ), B = eit (A + ωI )
for every t, ω ∈ R, the case of an operator B satisfying
ρ(B) ⊃ {λeiθ : λ > λ0 }, ∃M : λR(λeiθ , B)L(X) ≤ M, λ > λ0
for some θ ∈ [0, 2π), λ0 ≥ 0, may be easily reduced to this one. The
half-line r = {λeiθ : λ > λ0 } is called a ray of minimal growth of the
resolvent of B. See [2, Def. 2.1].
Proposition 3.1. Let A satisfy (3.1). Then
(X, D(A))θ, p = {x ∈ X : λ → ϕ(λ) := λθ AR(λ, A)x ∈ L ∗p (0, +∞)}
and the norms xθ, p and
x∗θ, p = x + ϕ L ∗p (0,+∞)
are equivalent.
Proof. Let x ∈ (X, D(A))θ, p . Then if x = a + b with a ∈ X, b ∈ D(A),

for every λ > 0 we have
λθ AR(λ, A)x ≤ λθ AR(λ, A)a + λθ R(λ, A)Ab
≤ (M + 1)λθ a + Mλθ−1 Ab
≤ (M + 1)λθ (a + λ−1 b D(A) ),
so that
λθ AR(λ, A)x ≤ (M + 1)λθ K (λ−1 , x).
With the change of variables λ → λ−1 we see that the right hand side
p
belongs to L ∗ (0, ∞), with norm equal to (M + 1)xθ, p . Therefore
p
ϕ(λ) = λθ AR(λ, A)x is in L ∗ (0, ∞), and
x∗θ, p ≤ (M + 1)xθ, p .
p
Conversely, if ϕ ∈ L ∗ (0, ∞), set for every λ ≥ 1
x = λR(λ, A)x − AR(λ, A)x,
so that
λθ K (λ−1 , x) ≤ λθ (AR(λ, A)x + λ−1 λR(λ, A)x D(A) )
= λθ (2AR(λ, A)x + R(λ, A)x).
p
The right hand side belongs to L ∗ (1, ∞), with norm estimated by
1/ p
1
2x∗θ, p + M x, p < ∞,
(1 − θ) p
2x∗θ,∞ + Mx, p = ∞.
p
It follows that t → t −θ K (t, x) ∈ L ∗ (0, 1), and hence x ∈ (X,D(A))θ, p
and
xθ, p ≤ C p (x∗θ, p + x).
The following notation is widely used.

Definition 3.2. For 0 < θ < 1, 1 ≤ p ≤ ∞ we set
D A (θ, p) = (X, D(A))θ, p .
For 0 < θ < 1, k ∈ N, 1 ≤ p ≤ ∞ we set
D A (θ + k, p) = {x ∈ D(Ak ) : Ak x ∈ D A (θ, p)},
x D A (θ+k, p) = x + Ak x D A (θ, p) ,
that is, D A (θ + k, p) is the domain of the part of Ak in D A (θ, p).
From the definition we get easily

Corollary 3.3. For 0 < θ < 1, 1 ≤ p ≤ ∞,
D A (θ + 1, p) = (D(A), D(A2 ))θ, p
and, more generally,
D A (θ + k, p) = (D(Ak ), D(Ak+1 ))θ, p .
Proof. It is sufficient to remark that (I − A)k is an isomorphism from

D(Ak ) to X, and also from D(Ak+1 ) to D(A). By the interpolation
theorem 1.6, it is an isomorphism between (D(Ak ), D(Ak+1 ))θ, p and
(X, D(A))θ, p , and the statement follows.
It is also important to characterize the real interpolation spaces between
X and D(A2 ), or more generally, between X and D(Am ) for m ∈ N. The
following proposition is useful.
Proposition 3.4. Let A satisfy (3.1). Then
D(A) ∈ J1/2 (X, D(A2 )) ∩ K 1/2 (X, D(A2 )).
Proof. Let us prove that D(A) ∈ J1/2 (X, D(A2 )). For every x ∈ D(A) it
holds
lim λR(λ, A)x = lim R(λ, A)Ax + x = x.
λ→∞ λ→∞
Setting f (σ ) = σ R(σ, A)x for σ > 0, we have
f (σ ) = R(σ, A)x − σ R(σ, A)2 x = R(σ, A)(I − σ R(σ, A))x

= −R(σ, A)2 Ax
and f (+∞) = x, so that
∞
x − λR(λ, A)x = − R(σ, A)2 Ax dσ, λ > 0,
λ
and if x ∈ D(A2 ),
∞
Ax = λAR(λ, A)x − R(σ, A)2 A2 x dσ, λ > 0.
λ
Therefore,
M2 2
Ax ≤ λ(M + 1)x + A x, λ > 0.
λ
Taking the infimum for λ ∈ (0, ∞) we get
Ax ≤ 2M(M + 1)1/2 x1/2 A2 x1/2 , x ∈ D(A2 ), (3.2)
so that
1/2
x D(A) ≤ Cx1/2 x D(A2 ) , x ∈ D(A2 ),
that is, D(A) ∈ J1/2 (X, D(A2 )).
Let us prove that D(A) ∈ K 1/2 (X, D(A2 )). For every x ∈ D(A) split
x as
x = −R(λ, A)Ax + λR(λ, A)x, λ > 0,
where
M
R(λ, A)Ax ≤ x D(A) ,
λ
λR(λ, A)x D(A2 ) = λR(λ, A)x + λAR(λ, A)Ax

≤ Mx + λ(M + 1)Ax
so that, setting t = λ−2 ,
K (t, x, X, D(A2 )) ≤ R(t −1/2 , A)Ax + tt −1/2 R(t −1/2 , A)x D(A2 )
≤ Mt 1/2 x D(A) + Mtx
+(M + 1)t 1/2 Ax, t >0
which implies that t → K (t,x,X,D(A2 )) is bounded in (0,1] by (2M +

1)x D(A) . Since it is bounded by x in (1,∞), then x∈ (X,D(A2 ))1/2,∞
and
x(X,D(A2 ))1/2,∞ ≤ (2M + 1)x D(A) .
But in general D(A) is not an interpolation space between X and D(A2 ).

As a counterexample we may take X = Cb (R), A : Cb1 (R) → X, Au =
u . See Example 1.21.
Using Proposition 3.4 and Reiteration Theorem we get a useful char-
acterization of (X, D(A2 ))θ, p .
Proposition 3.5. Let A satisfy (3.1). Then for θ = 1/2
(X, D(A2 ))θ, p = D A (2θ, p).

Proof. Taking into account that X belongs to J0 (X,D(A2))∩K 0 (X,D(A2 ))

and that D(A) belongs to J1/2 (X, D(A2))∩ K 1/2 (X, D(A2 )), and applying
the Reiteration Theorem with E 0 = X, E 1 = D(A) we get
D A (α, p) = (X, D(A))α, p = (X, D(A2 ))α/2, p , 0 < α < 1,
and setting α = 2θ the statement follows for 0 < θ < 1/2. Taking into
account that D(A2 ) belongs to J1 (X, D(A2 )) ∩ K 1 (X, D(A2 )) and that
D(A) belongs to J1/2 (X, D(A2 )) ∩ K 1/2 (X, D(A2 )), and applying the
Reiteration Theorem with E 0 = D(A), E 1 = D(A2 ) we get
D A (α + 1, p) = (D(A), D(A2 ))α, p = (X, D(A2 ))(α+1)/2, p , 0 < α < 1,
and setting α + 1 = 2θ the statement follows for 1/2 < θ < 1.

Another characterization, which holds also for θ = 1/2, is the follow-
ing one.
Proposition 3.6. Let A satisfy (3.1). Then for 0 < θ < 1, 1 ≤ p ≤ ∞
(X, D(A2 ))θ, p = {x ∈ X : λ →

ϕ (λ) = λ2θ (AR(λ, A))2 x ∈ L ∗p (0,∞)},
and the norms x(X,D(A2 ))θ, p and
xθ, p = x +
ϕ L ∗p (0,∞)
are equivalent.
Proof. The proof is very close to the proof of Proposition 3.1.
Let x ∈ (X, D(A2 ))θ, p . Then if x = a + b with a ∈ X, b ∈ D(A2 ),
for every λ > 0 we have
λ2θ (AR(λ, A))2 x ≤ λ2θ (AR(λ, A))2 a + λ2θ R(λ, A)2 A2 b

≤ (M + 1)2 λ2θ a + M 2 λ2θ−2 A2 b
≤ (M + 1)2 λ2θ (a + λ−2 b D(A2 ) ),
so that
λ2θ (AR(λ, A))2 x ≤ (M + 1)2 λ2θ K (λ−2 , x).
p
We know that λ−θ K (λ, x) ∈ L ∗ (0, ∞). With the change of variable
p
ξ = λ−2 we get λ → λ2θ K (λ−2 , x) ∈ L ∗ (0, ∞), with norm equal to
p
2−1/ p xθ, p . Therefore
ϕ (λ) = λ2θ (AR(λ, A))2 x is in L ∗ (0, ∞), and
xθ, p ≤ 2−1/ p (M + 1)2 xθ, p .
(The formula is true also for p = ∞ if we set 1/∞ = 0).

p
Conversely, if ϕ̃ ∈ L ∗ (0, ∞), from the obvious identity
x = λ2 R(λ, A)2 x − 2λAR(λ, A)2 x + A2 R(λ, A)2 x,
where
λAR(λ, A)2 x = λ(λ − A)AR(λ, A)3 x
= AR(λ, A)λ2 R(λ, A)2 x − λR(λ, A)A2 R(λ, A)2 x
we get
x = (I −2AR(λ,A))λ2 R(λ,A)2 x +(2λR(λ,A)+I )A2 R(λ,A)2 x, λ ≥ 1,
where
(I − 2AR(λ, A))λ2 R(λ, A)2 x D(A2 )

= (I − 2AR(λ, A))λ2 R(λ, A)2 x
+ (I − 2AR(λ, A))λ2 A2 R(λ, A)2 x
≤ (2M + 3)M 2 x + (2M + 3)λ2 A2 R(λ, A)2 x
and
(2λR(λ, A) + I )A2 R(λ, A)2 x ≤ (2M + 1)A2 R(λ, A)2 x.
Therefore,
λ2θ K (λ−2 , x, X, D(A2 ))

≤ λ2θ ((2λR(λ, A) + I )A2 R(λ, A)2 x
+ λ−2 (I − 2AR(λ, A))λ2 R(λ, A)2 x D(A2 ) )
≤ (4M + 4)λ2θ A2 R(λ, A)2 x + (2M + 3)M 2 λ2θ−2 x.
p
The right hand side belongs to L ∗ (1, ∞), with norm estimated by
1/ p
1
(4M + 3)xθ, p + (2M + 2)M 2
x,
(2 − 2θ) p
which is true also for p = ∞ with the convention (1/∞)1/∞ = 1. It

p
follows that t → t −θ K (t, x, X, D(A2 )) ∈ L ∗ (0, 1), and hence x ∈
(X, D(A2 ))θ, p and
x(X,D(A2 ))θ, p ≤ C p (xθ, p + x).

Propositions 3.4 and 3.5 may be generalized as follows.

Proposition 3.7. Let A satisfy (3.1), and let r, m ∈ N, r > m. Then
D(Ar ) ∈ Jr/m (X, D(Am )) ∩ K r/m (X, D(Am )).
Proposition 3.8. Let A satisfy (3.1), and let m ∈ N. Then for θ ∈ (0, 1)
such that θm ∈
/ N, and for 1 ≤ p ≤ ∞
(X, D(Am ))θ, p = D A (mθ, p).
See [90, Section 1.1.4.3], also for integer θm.
3.1.1. Exercises
1. Let A : D(A) ⊂ X → X satisfy (3.1). Prove that for every m ∈ N,
Am is a closed operator. (Hint: use estimate (3.2)).
2. Let A : D(A) ⊂ X → X satisfy (3.1), and let 0 < θ < 1. Prove that
(X, D(A))θ = {x ∈ X : lim λθ AR(λ, A)x = 0}.
λ→∞
3. Prove Proposition 3.7. Hint: to show that D(Ar ) ∈ K r/m (X, D(Am ))
prove preliminarily that D(Ar ) ∈ K 1/(m−r) (D(Ar−1 ), D(Am )), using
a procedure similar to the one of Proposition 3.4, and then argue by
reiteration.
4. Prove Proposition 3.8.
5. Let a < b, 1 ≤ p ≤ ∞, and set X := L p (a, b; X), D(B) := {u ∈
W 1, p (a, b; X) : u(a) = 0}, or X := C([a, b]; X), D(B) := {u ∈
C 1 ([a, b]; X) : u(a) = 0}, Bu = u . Show that the resolvent set of
B is the whole complex plane, that
t
(λI − B)−1 f (t) = − eλ(t−s) f (s)ds, a < t < b,
a
and that each half-line {re : r > 0} with θ ∈ (π/2, 3π/2) is a ray
iθ
of minimal growth for the resolvent of B.

6. Let A satisfy (3.1), let a < b, 1 ≤ p < ∞, and define the multiplic-
ation operator by A in the space X = L p (a, b; X),
D(A) = L p (a, b; D(A)), (Au)(t) = Au(t), a < t < b.
Prove that A satisfies (3.1) too, and show that DA (θ, p) = L p (a, b;
D A (θ, p)) for each θ ∈ (0, 1). If X = C([a, b]; X) and D(A) =
C([a, b]; D(A)), prove that DA (θ, ∞) = X ∩ B([a, b]; D A (θ, ∞))
(the latter is the space of the bounded functions with values in
D A (θ, ∞)).
3.2. Two or more operators

Let us consider now two operators A : D(A) → X, B : D(B) → X,
both satisfying (3.1). Throughout the section we shall assume that A and
B commute, in the sense that
R(λ, A)R(µ, B) = R(µ, B)R(λ, A), λ, µ > 0. (3.3)
It follows that D(Ak B h ) = D(B h Ak ) for all natural numbers h, k, and

that Ak B h x = B h Ak x for every x in D(Ak B h ).
Definition 3.9. For every m ∈ N set

m
m
K m := D(A j B m− j ), x K m := x + A j B m− j x.
j=0 j=0
The main result of the section is the following.
Theorem 3.10. Let m ∈ N, p ∈ [1, ∞] and θ ∈ (0, 1) be such that mθ

is not integer, and set k = [mθ], σ = {mθ}. Then we have
(X, K m )θ, p = {x ∈ K k : A j B k− j x ∈ D A (σ, p) ∩ D B (σ, p), j = 0, . . . , k},
and the norms

x → x(X,K m )θ, p ,

k
x → x + (A j B k− j x D A (σ, p) + A j B k− j x D B (σ, p) )
j=0
are equivalent.
We shall give the details of the proof only for m = 1 and m = 2,

leaving the rest as an exercise. We begin with m = 1.
Proposition 3.11. For every p ∈ [1, ∞] and θ ∈ (0, 1) we have
(X, K 1 )θ, p = D A (θ, p) ∩ D B (θ, p),
and the norms
x → x(X,K 1 )θ, p , x → x D A (θ, p) + x D B (θ, p)
are equivalent.
Proof. The embedding (X, K 1 )θ, p ⊂ D A (θ, p) ∩ D B (θ, p) is obvious,

since K 1 = D(A) ∩ D(B) is continuously embedded both in D(A) and
in D(B).
Let x ∈ D A (θ, p) ∩ D B (θ, p). We recall (see Proposition 3.1) that the
functions
λ → λθ AR(λ, A)x, λ → λθ B R(λ, B)x, λ > 0,

p
are in L ∗ (0,∞) and their norms are bounded byCx D A (θ,p) ,Cx D B (θ,p) ,
respectively.
For every λ > 0 set
v(λ) = λ2 R(λ, A)R(λ, B)x, λ > 0, (3.4)
and split x = x − v(λ) + v(λ). It holds
v(λ) − x ≤ λR(λ, A)(λR(λ, B)x − x) + λR(λ, A)x − x
≤ MB R(λ, B)x + AR(λ, A)x,
and
v(λ) K 1 = v(λ) + Av(λ) + Bv(λ)
≤ M 2 x + λMAR(λ, A)x + λMB R(λ, B)x.
Therefore, for λ ≥ 1
λθ K (λ−1 , x, X, K 1 ) ≤ 2Mλθ (AR(λ, A)x

+ B R(λ, B)x) + M 2 λθ−1 x,
p
so that λ → λθ K (λ−1 , x, X, K 1 ) ∈ L ∗ (1, ∞), with norm estimated by
const. (x +x D A (θ, p) + x D B (θ, p) ). Then λ → λ−θ K (λ, x, X, K 1 ) ∈
p
L ∗ (0, 1), with the same norm, and the statement follows.
For m = 2, the proof is in three steps. The first step is Proposition 3.11,
that characterizes (X, K 1 )θ, p . As a second step (Proposition 3.12) we
characterize (K 1 , K 2 )θ, p . Then we show (Proposition 3.13) that K 1 be-
longs to J1/2 ∩ K 1/2 between X and K 2 . The final statement will follow
from the Reiteration Theorem.
Proposition 3.12. For every p ∈ [1, ∞] and θ ∈ (0, 1) we have
(K 1 , K 2 )θ, p = {x ∈ K 1 : Ax, Bx ∈ D A (θ, p) ∩ D B (θ, p)}
= D A (θ + 1, p) ∩ D B (θ + 1, p),
and the norms

x → x(K 1 ,K 2 )θ, p ,
x → x + Ax D A (θ, p) + Ax D B (θ, p) + Bx D A (θ, p) + Bx D B (θ, p) ,
x → x D A (θ+1, p) + x D B (θ+1, p)

are equivalent.
Proof. Let us prove the embeddings ⊂. Since K 1 ⊂ D(A) and K 2 ⊂
D(A2 ) then (K 1 , K 2 )θ, p ⊂ (D(A), D(A2 ))θ, p = D A (θ + 1, p). Sim-
ilarly, (K 1 , K 2 )θ, p ⊂ D B (θ + 1, p). It remains to show that each x ∈
(K 1 , K 2 )θ, p is such that Ax ∈ D B (θ, p) and Bx ∈ D A (θ, p). For every
a ∈ K 1 , b ∈ K 2 such that x = a + b we have
λθ B R(λ, B)Ax ≤ λθ B R(λ, B)Aa + λθ B R(λ, B)Ab
≤ λθ (M + 1)Aa + Mλθ−1 B Ab
≤ (M + 1)λθ (a K 1 + λ−1 b K 2 )
so that
λθ B R(λ, B)Ax ≤ (M + 1)λθ K (λ−1 , x, K 1 , K 2 ), λ > 0.
p
It follows that λ → λθ B R(λ, B)Ax belongs to L ∗ (0, ∞) with norm
bounded by (M + 1)x(K 1 ,K 2 )θ, p , and the embedding ⊂ is proved.
The proof of the embedding
{x ∈ K 1 : Ax, Bx ∈ D A (θ, p) ∩ D B (θ, p)} ⊂ (K 1 , K 2 )θ, p
is similar to the corresponding proof in Proposition 3.11, and is omitted.
Let us prove that D A (θ + 1, p) ∩ D B (θ + 1, p) ⊂ {x ∈ K 1 : Ax, Bx ∈
D A (θ, p) ∩ D B (θ, p)}. We have only to show that if x ∈ D A (θ + 1, p) ∩
D B (θ + 1, p) then Ax ∈ D B (θ, p) and Bx ∈ D A (θ, p). Indeed, for each
λ > 0 we have
λθ B 2 R(λ, B)2 Ax
≤ λθ+1 B 2 R(λ, B)2 R(λ, A)Ax + λθ B 2 R(λ, B)2 AR(λ, A)Ax
≤ λθ M(M + 1)B R(λ, B)Bx + λθ (M + 1)2 AR(λ, A)Ax

p
so that λ → λθ B 2 R(λ, B)2 Ax ∈ L ∗ (0, ∞) with norm bounded by
M(M + 1)Bx∗D B (θ, p) + (M + 1)2 Ax∗D A (θ, p) .
By Proposition 3.6, Ax ∈ D B 2 (θ/2 p). This space coincides with

D B (θ, p) by Proposition 3.5. So, Ax ∈ D B (θ, p) and Ax D B (θ, p) ≤
C(x D A (θ+1, p) + x D B (θ+1, p) ). Similarly, Bx ∈ D A (θ, p) and
Bx D A (θ, p) ≤ C(x D B (θ+1, p) + x D A (θ+1, p) ).
In the last part of the proof of Proposition 3.12 we have shown that
if x ∈ D A (θ + 1, p) ∩ D B (θ + 1, p) then Ax ∈ D B (θ, p) and Bx ∈
D A (θ, p), a sort of “mixed regularity” result. However it is not true in
general that D(A2 ) ∩ D(B 2 ) ⊂ D(AB).
For instance, let A be the realization of ∂/∂ x and let B be the real-
ization of ∂/∂ y in X := C b (R2 ). Then D A (θ + 1, ∞) consists of the
functions f ∈ X such that x → f (x, y) ∈ Cbθ+1 (R), uniformly with
respect to y ∈ R, and similarly D B (θ + 1, ∞) consists of the functions
f ∈ X such that y → f (x, y) ∈ Cbθ+1 (R), uniformly with respect to
x ∈ R. Proposition 3.12 states that if f ∈ D A (θ + 1, ∞) ∩ D B (θ + 1, ∞)
then ∂ f /∂ x ∈ D B (θ, ∞), that is it is Hölder continuous also with re-
spect to y, and ∂ f /∂ y ∈ D A (θ, ∞), that is it is Hölder continuous also
with respect to x. On the other hand, it is known that in this example
D(A2 ) ∩ D(B 2 ) is not embedded in D(AB), because there are functions
u such that ∂ 2 u/∂ x 2 , ∂ 2 u/∂ y 2 are bounded and continuous but ∂ 2 u/∂ x∂ y
is not.
Next step consists in proving that K 1 belongs to J1/2 (X, K 2 ) and to
K 1/2 (X, K 2 ).
Proposition 3.13.
K 1 ∈ J1/2 (X, K 2 ) ∩ K 1/2 (X, K 2 ).
Proof. We already know that D(A) and D(B) belong to J1/2 (X, D(A2 ))
and to J1/2 (X, D(B 2 )), respectively. Therefore there is C > 0 such that
x K 1 ≤ x D(A) + x D(B)
1/2 1/2
≤ Cx1/2 (x D(A2 ) + x D(B 2 ) )
1/2
≤ C x1/2 x K 2 ,
which means that K 1 ∈ J1/2 (X, K 2 ).
To prove that K 1 ∈ K 1/2 (X, K 2 ), for every x ∈ K 1 we split again
x = x − v(λ) + v(λ) for every λ > 0, where v is the function defined in
(3.4). Then
v(λ) − x ≤ λR(λ, A)(λR(λ, B)x − x) + λR(λ, A)x − x
= λR(λ, A)R(λ, B)Bx + R(λ, A)Ax
≤ λ−1 (M(M + 1)Bx + MAx),
and
v(λ) K 2 = v(λ) + A2 v(λ) + ABv(λ) + B 2 v(λ)
≤ M 2 x + 2M(M + 1)λ(Ax + Bx).
Setting λ = t −1/2 we deduce that

t −1/2 K (t, x, X, K 2 ) ≤ t −1/2 (x − v(t −1/2 ) + tv(t −1/2 ) K 2 )
≤ C(x K 1 + t 1/2 x),
is bounded in (0, 1). We know already that t → t −1/2 K (t, x, X, K 2 )

is bounded in [1, ∞). Therefore K 1 is in the class K 1/2 between X
and K 2 .
The Reiteration Theorem and Propositions 3.11, 3.13 yield now
Proposition 3.14. Let p ∈ [1, ∞], θ ∈ (0, 1), θ = 1/2. Then
(X, K 2 )θ, p = D A (2θ, p) ∩ D B (2θ, p),
and for θ > 1/2 we have also
(X, K 2 )θ, p = {x ∈ K 1 : Ax, Bx ∈ D A (2θ − 1, p) ∩ D B (2θ − 1, p)},
Proof. For θ < 1/2 we apply the Reiteration Theorem with Y = K 2 ,
E 0 = X, E 1 = K 1 , and the statement follows from Proposition 3.11. For
θ > 1/2 we apply the Reiteration Theorem with Y = K 2 , E 0 = K 1 ,
E 1 = K 2 , and the statement follows from Proposition 3.12.
Theorem 3.10 in its full generality may be proved by recurrence. The
main ingredients are the following propositions, whose proofs are similar
to the proofs of Propositions 3.12 and 3.13, respectively.
Proposition 3.15. For every k ∈ N, p ∈ [1, ∞] and θ ∈ (0, 1) we have
(K k ,K k+1 )θ, p = {x ∈ K k : A j B k− j x ∈ D A (θ, p) ∩ D B (θ, p), j = 0, . . . , k}
and the norms
x → x(K k ,K k+1 )θ, p ,

k
x → x + (A j B k− j x D A (θ, p) + A j B k− j x D B (θ, p) )
j=0
are equivalent.
Proposition 3.16. For every k ∈ N ∪ {0},

K k+1 ∈ J1/2 (K k , K k+2 ) ∩ K 1/2 (K k , K k+2 ).
More generally,
K k+1 ∈ J1/s (K k , K k+s ) ∩ K 1/s (K k , K k+s ).
The results and the procedures of this section are easily extended to the
case of a finite number of operators A1 , . . . , An .
3.2.1. Exercises
1. Prove that if R(λ0 , A)R(µ0 , B) = R(µ0 , B)R(λ0 , A) for some λ0 ,
µ0 > 0 then (3.3) holds.
2. Prove Proposition 3.15.
3. Prove Proposition 3.16. Hint: for the first statement, follow step by
step the proof of Proposition 3.13; for the second statement replace
v(λ) by w(λ) = λ2s R(λ, A)s R(λ, B)s x.
4. Prove Theorem 3.10 by recurrence on m, using the procedure of Pro-
position 3.14 and the results of Propositions 3.15 and 3.16.
3.2.2. The sum of two commuting operators
Here X is a complex Banach space, and A : D(A) → X, B : D(B) → X
are linear operators satisfying (3.1) and (3.3). For simplicity, we also
assume that 0 ∈ ρ(A) ∩ ρ(B).
We shall study the closability and the invertibility of the operator A +
B, following the approach of Da Prato and Grisvard [30].
From the properties of the resolvent operators, it follows easily that the
resolvent sets of A and B contain the sector {λ ∈ C : Re λ > 0 |Im λ| <
Re λ/M}, and that every half-line {reiθ : r > 0} with |θ| < arctan 1/M
is a ray of minimal growth of the resolvent of A and B (see next Lemma
4.2). However, we need that A and B satisfy a stronger condition. For
θ ∈ (0, π] let us denote by Sθ the complex sector
Sθ := {λ ∈ C : λ = 0, | arg λ| < θ}.
The stronger assumption is the following: there are angles θ A , θ B > 0
and positive numbers M A , M B such that

(a) ρ(A) ⊃ Sθ A , λR(λ, A) ≤ M A , ∀λ ∈ Sθ A ,
(b) ρ(B) ⊃ Sθ B , λR(λ, B) ≤ M B , ∀λ ∈ Sθ B ,
(3.5)
θ A + θ B > π.
Since θ A + θ B > π, then either θ A > π/2 or θ B > π/2, or both. To

be definite, assume that θ A > π/2. Then A is a sectorial operator and it
generates an analytic semigroup. Sectorial operators and analytic semig-
roups are treated in detail in Chapter 6; however we will not need the
specific techniques of analytic semigroups here.
Assumption (3.5) implies that for every θ ∈ (π − θ B , θ A ) the path
γ := {reiθ : r > 0} ∪ {re−iθ : r > 0} (3.6)
is contained in ρ(A) ∩ ρ(−B). See Figure 3.1.

é 

Figure 3.1. The curve γ .
Moreover, assumption (3.3) implies that
(A − λI )−1 (B + λI )−1 = (B + λI )−1 (A − λI )−1
for arg λ ∈ (π − θ B , θ A ).
Proposition 3.17. Let A, B satisfy (3.3) and (3.5), with θ A > π/2. As-
sume in addition that 0 ∈ ρ(A) ∩ ρ(B) and that D(A) or D(B) is dense
in X. Then A + B : D(A) ∩ D(B) → X is closable, and its closure
A + B is invertible with inverse S given by

1
S=− (A − λI )−1 (B + λI )−1 dλ (3.7)
2πi γ
where γ is the path defined in (3.6), positively oriented from ∞e−iθ to

∞eiθ , with any θ ∈ (π − θ B , θ A ).
Proof. From assumption (3.5) it follows that S is a bounded operator, and

since λ → (A − λI )−1 (B + λI )−1 is holomorphic for | arg λ| ∈ (π −
θ B , θ A ) and in a neighborhood of 0, S is independent of θ ∈ (π − θ B , θ A ).
The proof is in three steps: (i) we show that S is a left inverse of
A + B; (ii) we show that for every α ∈ (0, 1) S maps D A (α, ∞) (resp.
D B (α, ∞)) to D(A) ∩ D(B) and (A + B)Sx = x for each x ∈ D A (α, ∞)
(resp. x ∈ D B (α, ∞)); (iii) using steps (i) and (ii) we show in a standard
way that if D(A) or D(B) is dense, then A + B is closable and S =
−1
A+B .
(i) For every x ∈ D(A) ∩ D(B) we have
(B + λI )−1 (A − λI )−1
(A −λI )−1 (B +λI )−1 (Ax + Bx) = − Bx + Ax.
λ λ
To avoid singularities near λ = 0 we modify a bit the path γ , replacing it
by
γε = {reiθ : r > ε} ∪ {re−iθ : r > ε} ∪ {εeiη : η ∈ [θ, 2π − θ]},
still positively oriented from ∞e−iθ to ∞eiθ , with ε > 0 small enough,
such that the closed ball centered at 0 with radius ε is contained in the
resolvents sets of A and B.

é

é
Figure 3.2. The curve γε .
Then we have

1 (B + λI )−1 1 (A − λI )−1
S(Ax +Bx) = dλ Bx − dλ Ax.
2πi γε λ 2πi γε λ
The first integral vanishes, because λ → λ−1 (B + λI )−1 is holomorphic

for λ = 0, |π − arg λ| < θ B , and its norm does not exceed const.|λ|−2 .
Indeed, the integral over γε is the limit as n → ∞ of the integrals over
γε,n := {reiθ : ε < r < n}∪{re−iθ : ε < r < n}∪{εeiη : η ∈ [θ, 2π −θ]}
∪ {neiη : η ∈ [θ, 2π − θ]}, oriented counterclockwise, that vanish for
every n.

é é
Figure 3.3. The curve γε,n .

(A−λI )−1
1
The second integral, 2πi γε λ
dλ, is equal to −I . Indeed, it is the
limit as as n → ∞ of the integrals over γε,n := {reiθ : ε < r <
n}∪{re−iθ : ε < r < n} ∪ {εeiη : η ∈ [θ, 2π −θ]} ∪ {neiη : η ∈ [−θ, θ]},
oriented counterclockwise, that are equal to −I for every n.
It follows that S(Ax + Bx) = x.
(ii). Fix x ∈ D A (α, ∞). Then for each λ ∈ γ we have
A(A − λI )−1 (B + λI )−1 x = (B + λI )−1 A(A − λI )−1 x (3.8)
whose norm is bounded by a constant independent of λ for λ close to 0,
and by M B (M A +1)x D A (α,∞ |λ|−θ−1 for λ = 0, by assumption (3.5) and
Proposition 3.1. Therefore, Sx ∈ D(A). To prove that Sx ∈ D(B) we
modify again γ to γε as before, and for each λ ∈ γε we use the identity
1
(A − λI )−1 (B + λI )−1 x = (B + λI )−1 (A(A − λI )−1 x − x)
λ
that implies

1 1
Sx = − (B + λI )−1 A(A − λI )−1 x dλ, (3.9)
2πi γε λ

é
Figure 3.4. The curve

γε,n .
)−1
since the integral γε (B+λI
λ
dλ vanishes, as we already remarked. For
each λ ∈ γε , the norm of λ B(B + λI )−1 A(A − λI )−1 x is bounded by
1
(M B + 1)(M A + 1)x D A (α,∞) |λ|−θ−1 , again by assumption (3.5) and

Proposition 3.1. Therefore, Sx ∈ D(B).
By (3.8) and (3.9) we get

1 B
(A + B)Sx = − I+ (B + λI )−1 A(A − λI )−1 x dλ
2πi γε λ

1 A(A − λI )−1
=− x dλ
2πi γε λ

1 A(A − λI )−1
= limn→∞ − x dλ
2πi γε,n
λ
= x.
If x ∈ D B (α, ∞), similar arguments (left as exercise) lead to Sx ∈

D(A) ∩ D(B) and (A + B)Sx = x.
(iii) Let us show that A + B is closable. Let xn ∈ D(A) ∩ D(B) be such
that xn → 0, (A + B)xn → y as n → ∞. Since S is a left inverse of
A + B then
0 = lim xn = lim S(A + B)xn = Sy.

n→∞ n→∞
Since A−1 y ∈ D(A) ⊂ D A (α, ∞) for each α ∈ (0, 1), then by step (ii)
S A−1 y ∈ D(A) ∩ D(B) and
A−1 y = (A + B)S A−1 y = (A + B)A−1 Sy = 0,
so that y = 0 and A + B is closable.

As a last step we prove that A + B is invertible and that its inverse is
S. If x ∈ D(A + B) there is a sequence xn ∈ D(A) ∩ D(B) such that
xn → x and (A + B)xn → A + B x as n → ∞. Then
x = lim xn = lim S(A + B)xn = S A + B x,

n→∞ n→∞
which means that S is a left inverse of A + B. Now, in the case that D(A)
(resp. D(B)) is dense in X, for x ∈ X let xn ∈ D(A) (resp., xn ∈ D(B))
be such that xn → x as n → ∞. By step (ii) Sxn ∈ D(A) ∩ D(B),
limn→∞ Sxn = Sx and limn→∞ (A + B)Sxn = limn→∞ xn = x. This
implies that Sx ∈ D(A + B) and A + B Sx = x, so that S is also a right
inverse of A + B.
Under the assumptions of Proposition 3.17 the operator A + B is not

closed in general. Equivalently, S does not map X into D(A) ∩ D(B) in
general. It is possible to prove that it maps X into (X, D(A2 ))1/2,∞ and
into (X, D(B 2 ))1/2,∞ , see exercises. Such spaces are not very far from
D(A) and D(B), respectively, but still they have weaker topologies, in
general. See Proposition 3.4.
However, in the next theorem we shall see that S maps continuously
D A (α, p) into D A (α + 1, p), and it maps continuously D B (α, p) into
D B (α + 1, p), for each α ∈ (0, 1) and p ∈ [1, ∞]. Therefore, replacing
X by Y = D A (α, p) or by Y = D B (α, p), the part of A + B in Y is
invertible with bounded inverse S|Y .
Theorem 3.18. Under the assumptions of Proposition 3.17, for each α ∈

(0, 1) and p ∈ [1, ∞] and for each x ∈ D A (α, p) ∪ D B (α, p), Sx ∈
D(A) ∩ D(B) and (A + B)Sx = x. Moreover, there is C = C(α, p) > 0
such that
(i) if x ∈ D A (α, p), then ASx and B Sx belong to D A (α, p), with
ASx D A (α, p) + B Sx D A (α, p) ≤ Cx D A (α, p) , (3.10)
(ii) if x ∈ D B (α, p), then ASx and B Sx belong to D B (α, p), with
ASx D B (α, p) + B Sx D B (α, p) ≤ Cx D B (α, p) . (3.11)

Proof. In Step (ii) of the proof of Proposition 3.17 we already showed

that Sx ∈ D(A) ∩ D(B) for each x ∈ D A (α, p) ∪ D B (α, p), and (A +
B)Sx = x. We have to prove statements (i) and (ii).
Let x ∈ D B (α, p). Let γ be the curve defined by 3.6 (see Fig. 3.1).
For ξ > 0 we have

−1 1
(B − ξ I ) Sx = − (A − λI )−1 (B − ξ I )−1 (B + λI )−1 x dλ
2πi γ

1 (A − λI )−1
=− ((B −ξ I )−1 x − (B +λI )−1 x)dλ
2πi γ λ+ξ

1 1
= (A − λI )−1 (B + λI )−1 x dλ
2πi γ λ+ξ

since γ λ+ξ (A
1
− λI )−1 dλ = 0. Therefore,
B(B − ξ I )−1 Sx = Sx + ξ(B − ξ I )−1 Sx

1
=− (A − λI )−1 (B + λI )−1 x dλ
2πi γ

1 −1 −1
−ξ (A − λI ) (B + λI ) x dλ
γ λ+ξ

1 λ
=− (A − λI )−1 (B + λI )−1 x dλ
2πi γ λ+ξ
which implies

−1 1 λ
B(B − ξ I ) B Sx = − (A − λI )−1 B(B + λI )−1 x dλ.
2πi γ λ+ξ
For λ = re±iθ ∈ γ we have
λ MA
(A − λI )−1 B(B + λI )−1 x ≤ B(B + re±iθ I )−1 x
λ+ξ r| cos θ| + ξ
so that, setting ϕ± (r) = r α B(B + re±iθ I )−1 x,

α −1 MA ∞ ξ α 1 dr
ξ B(B−ξ I ) B Sx ≤ (ϕ− (r)+ϕ+ (r)) .
2π 0 r | cos θ| + ξ/r r
The right hand side may be seen as the multiplicative convolution (see
Appendix) between g(ξ ) := M A ξ α /2π(ξ + | cos θ|), that belongs to
p
L 1∗ (0, ∞), and ϕ− + ϕ+ , that belongs to L ∗ (0, ∞) by Proposition 3.1.
p
Therefore, it belongs to L ∗ (0, ∞) with norm estimated by Cx D B (α, p) ,
and so does the function ξ → ξ α B(B − ξ I )−1 B Sx. By Proposition
3.1, B Sx ∈ D B (α, p). Since ASx = x − B Sx, ASx ∈ D B (α, p) and
(3.11) follows.
If x ∈ D A (α, p) the proof is similar, and it is left as exercise.
Theorem 3.18 may be applied to evolution equations in general Banach
space X, 
 u (t) = Au(t) + f (t), 0 < t < T,
(3.12)

u(0) = 0,
where A satisfies (3.5)(a) with θ A > π/2 (so that A is s sectorial operator,
see Chapter 6), and 0 ∈ ρ(A). Problem (3.12) is seen as the equation
Au + Bu = − f
in the space X := L p (0, T ; X), 1 ≤ p, ∞, where A and B are defined

by
A : L p (0, T ; D(A)) → X , (Au)(t) = Au(t),
B : D(B) = {u ∈ W 1, p (0, T ; X) : u(0) = 0} → X , (Bu)(t) = −u (t).
Then D(B) is dense in X . For 0 < α < 1, the space DA (α, p) coin-
cides with L p (0, T ; D A (α, p)) by Exercise 6, Section 3.1.1. The space
DB (α, p) coincides with W α, p (0, T ; X), if 0 < α < 1/ p, and with
{u ∈ W α, p (0, T ; X); u(0) = 0}, if 1/ p < α < 1. This was shown in
[30] using the characterization of next Proposition 5.7, since B is the in-
finitesimal generator of the strongly continuous semigroup et B defined by

 u(s − t), t ≤ s ≤ T,
tB
(e u)(s) :=

0, 0 ≤ s ≤ t.
Theorem 3.18 yields that if f ∈ L p (0, T ; D A (α, p)) or f ∈ DB (α, p),

equation Au + Bu = − f has a unique solution u ∈ D(A) ∩ D(B), and
Au , Bu belong to L p (0, T ; D A (α, p)) or to DB (α, p), respectively. This
implies that problem (3.12) has a unique solution u, such that both u and
Au belong to L p (0, T ; D A (α, p)) or to W α, p (0, T ; X), respectively.
If A satisfies (3.5)(a) with θ A > π/2 but 0 ∈ / ρ(A) the result still holds,
it is sufficient to replace A by A − ωI and u, f by v(t) := u(t)e−ωt ,
g(t) = f (t)e−ωt with ω > 0.
This is an example of maximal regularity in evolution equations, the

simplest one that can be obtained from Theorem 3.18. Maximal regu-
larity means that the solution u is such that both u and Au belong to the
same space as the datum f . In Section 4.2.1 and in Chapter 6 we shall
see other types of maximal regularity in evolution equations.
3.2.3. Exercises
1. Complete Step (ii) of the proof of Proposition 3.17 showing that for
each x ∈ D B (α, ∞), Sx ∈ D(A) ∩ D(B).
2. Prove that for every x ∈ X, Sx ∈ (X,D(A2 ))1/2,∞ ∩ (X, D(B 2 ))1/2,∞ .
Hint: use Proposition 3.6, showing preliminarily that for every ξ > 0
we have
A2 (A − ξ I )−2 Sx

1 2ξ ξ2
=− 1+ + (A−λI )−1 (B +λI )−1 x dλ
2πi γ ξ −λ (ξ −λ)2
and, similarly,
B 2 (B − ξ I )−2 Sx

1 2ξ ξ2
=− 1− + (A−λI )−1 (B +λI )−1 x dλ.
2πi γ ξ −λ (ξ −λ)2
3. Complete the proof of Theorem 3.18, showing that if x ∈ D A (α, p)

then ASx and B Sx belong to D A (α, p).
Chapter 4
Powers of positive operators
The powers (with real or complex exponents) of positive operators are

important tools in the study of partial differential equations. The theory of
powers of operators is very close to interpolation theory, even if in general
the domain of a power of a positive operator is not an interpolation space.
Although the theory of powers of operators can be seen as a particular
case of the Functional Calculus (e.g., [51, 52]), here we give simple self-
contained proofs, according to the spirit of these lecture notes.
Through the whole chapter X is a complex Banach space.
4.1. Definitions and general properties

Definition 4.1. A linear operator A : D(A) ⊂ X → X is called a pos-
itive operator if the resolvent set of A contains (−∞, 0] and there is
M > 0 such that
M
R(λ, A)L(X) ≤ , λ ≤ 0. (4.1)
1 + |λ|
Note that A is a positive operator iff (−∞, 0) is a ray of minimal growth
for the resolvent R(λ, A) and 0 ∈ ρ(A). So, if A is a positive operator,
then −A satisfies (3.1) so that all the results of Section 3.1 are applicable.
Examples of unbounded positive operators are readily given: for in-
stance, the realization of the first order derivative with Dirichlet boundary
condition at x = 0 in C([0, 1]) or in L p (0, 1), 1 ≤ p ≤ ∞ is positive.
More generally, if A is the generator of a strongly continuous or analytic
semigroup T (t) such that T (t) ≤ Me−ωt for some ω > 0, then −A
is a positive operator. This is an immediate consequence of the already
mentioned resolvent formula
∞
−R(−λ, −A) = R(λ, A) = e−λt T (t)dt, λ > −ω.
0
This section is devoted to the construction and to the main properties of

the powers A z , where z is an arbitrary complex number.
If A : X → X is a bounded positive operator the powers A z are readily

defined by
1
Az = λz R(λ, A)dλ,
2πi γ
where γ is any piecewise smooth curve surrounding σ (A), avoiding
(−∞, 0], with index 1 with respect to every element of σ (A). Several
properties of A z follow easily from the definition: for instance, z → A z
is holomorphic with values in L(X); if z = k ∈ Z then A z defined above
coincides with Ak ; for each z, w ∈ C we have A z Aw = Aw A z = A z+w ;
(A−1 )z = A−z , etc.
In the case where A is unbounded the theory is much more complic-
ated. To define A z we shall use an elementary but important spectral
property, stated in the next lemma.
Lemma 4.2. Let A be a positive operator. Then the resolvent set of A
contains the set
= {λ ∈ C : Re λ ≤ 0, |Im λ| < (|Re λ| + 1)/M} ∪ {λ ∈ C : |λ| < 1/M},
where M is the number in formula (4.1), and for every θ0 ∈ (0,arctan1/M),

r0 ∈ (0, 1/M) there is M0 > 0 such that
M0
R(λ, A) ≤
1 + |λ|
for all λ ∈ C with |λ| ≤ r0 , and for all λ ∈ C with Re λ < 0 and
|Im λ|/|Re λ| ≤ tan θ0 .
Proof. It is sufficient to recall that for every λ0 ∈ ρ(A) the resolvent set
ρ(A) contains the open ball centered at λ0 with radius 1/R(λ0 , A), and
that for |λ − λ0 | < 1/R(λ0 , A) it holds

∞
R(λ, A) = (−1)n (λ − λ0 )n R(λ0 , A)n+1 .
n=0
The union of the balls centered at λ0 ∈ (−∞,0] with radius R(λ0 ,A)−1
contains the set , and the estimate follows easily.
For θ ∈ (π/2, π), r > 0, let γr, θ be the curve defined by γr, θ =
−γr,(1)θ − γr,(2)θ + γr,(3)θ , where γr,(1)θ , γr,(3)θ are the half lines parametrized re-
spectively by z = ξ eiθ , z = ξ e−iθ , ξ ≥ r, and γr,(2)θ is the arc of circle
parametrized by z = reiη , −θ ≤ η ≤ θ. See the figure.
Now we are ready to define A z for Re z < 0 through a Dunford integ-
ral.
0 r
r,
Figure 4.1. The curve γr, θ .
Definition 4.3. Fix any r ∈ (0, 1/M), θ ∈ (π − arctan 1/M, π). For Re
α < 0 set
α 1
A = λα R(λ, A)dλ. (4.2)
2πi γr,θ
Since λ → λα R(λ, A) is holomorphic in \ (−∞, 0] with values in
L(X), the integral is an element of L(X) independent of r and θ. Writing
down the integral we get
∞
1
α
A = ξ α (−eiθ(α+1) R(ξ eiθ , A) + e−iθ(α+1) R(ξ e−iθ , A))dξ
2πi r
θ
r α+1
− eiη(α+1) R(reiη , A)dη
2π −θ
(4.3)
for every r ∈ (0, 1/M), θ ∈ (π − arctan 1/M, π).
Of course formula (4.3) may be reworked to get simpler expressions
for Aα . For instance, if −1 < Re α < 0 we may let r → 0, θ → π to get

α sin(πα) ∞ α
A x =− ξ (ξ I + A)−1 x dξ. (4.4)
π 0
Note that for every a > 0 and α ∈ (−1, 0) we have

sin(πα) ∞ ξ α
aα = − dξ, (4.5)
π 0 ξ +a
which agrees with (4.4) of course, and will be used later.
From the definition it follows immediately that the function z → A z is
holomorphic in the half plane Re z < 0, with values in L(X). Its behavior
near the imaginary axis is not obvious, but it is of great importance in the
developments of the theory and will be discussed in the next section.
Let us see some basic properties of the operators A α .
Proposition 4.4. The following statements hold true.
(i) For α = −n, n ∈ N, the operator defined in (4.2) coincides with

A−n = n-th power of the inverse of A.
(ii) For Re z < −k, k ∈ N, the range of A z is contained in the domain
D(Ak ), and
Ak A z x = Ak+z x, x ∈ X.
(iii) For Re z < 0 and x ∈ D(Ak ), k ∈ N, Az x ∈ D(Ak ), and
A z Ak x = Ak A z x.
(iv) For Re z 1 , Re z 2 < 0 we have
A z1 A z2 = A z1 +z2 .
Proof. (i) Let α = −n. It is easy to see that

1 1
λ−n R(λ, A)dλ = lim λ−n R(λ, A)dλ,
2πi γr,θ k→∞ 2πi γ
k
with γk as in figure.
é
Figure 4.2. The curve γk .
For every k ∈ N the function λ → R(λ, A) is holomorphic in the

bounded region surrounded by γk . For every k ∈ N we have

1 1 d n−1
λ−n R(λ, A)dλ = − R(λ, A)λ=0 = A−n ,
2πi γk (n − 1)! dλn−1
and letting k → ∞,

1
λ−n R(λ, A)dλ = A−n .
2πi γr, θ
(ii) Let k = 1, Re z < −1. Then, since
λz AR(λ, A) = λz (λR(λ, A) − I ) ≤ (M0 + 1)|λ|Re z ,
the integral defining A z is in fact an element of L(X, D(A)), and

1 1 1
A λ R(λ, A)dλ =
z
λ z+1
R(λ, A)dλ − λz dλI.
2πi γr,θ 2πi γr,θ 2πi γr,θ
But the last integral vanishes, so that A · A z = A1+z , and the statement is
proved for k = 1. The statement for any k follows arguing by recurrence.
Statement (iii) is obvious because Ak commutes with R(λ, A) on D(Ak ),
and this implies that Ak commutes with A z on D(Ak ).
(iv) Let θ1 < θ2 < π, 1/M > r1 > r2 > 0, so that γr1 ,θ1 is on the right
hand side of γr2 ,θ2 . Then

1
Az1 Az2 = λz1 R(λ, A)dλ w z2 R(w, A)dw
(2πi)2 γr1 ,θ1 γr2 ,θ2

1 R(λ, A) − R(w, A)
= λz 1 w z 2 dλ dw
(2πi)2 γr1 ,θ1 ×γr2 ,θ2 w−λ

1 w z2
= λz1 R(λ, A)dλ dw
(2πi)2 γr1 ,θ1 γr2 ,θ2 w−λ

1 λz 1
− w R(w, A)dw
z2
dλ
(2πi)2 γr2 ,θ2 γr1 ,θ1 w−λ

1
= λz1 +z2 R(λ, A)dλ = Az1 +z2 .
2πi γr1 ,θ1
Statement (iv) of the proposition implies immediately that A z is one to

one. Indeed, if A z x = 0 and n ∈ N is such that −n < Re z, then
A−n x = A−n−z A z x = 0, so that x = 0. Therefore it is possible to define
Aα if Re α > 0 as the inverse of A−α . But in this way the powers Ait ,
t ∈ R, remain undefined. So we give a unified definition for Re α ≥ 0.
Definition 4.5. Let 0 ≤ Re α < n, n ∈ N. We set
D(Aα ) = {x ∈ X : Aα−n x ∈ D(An )}, Aα x = An Aα−n x.

From Proposition 4.4 it follows that the operator Aα is independent of

n: indeed, if n, m > Re α, then Aα−m x = An−m Aα−n x both for n < m
(by Proposition 4.4(iv)) and for n > m (by Proposition 4.4(ii), taking z =
α −n and k = n −m), so that Aα−m x ∈ D(Am ) iff An−m Aα−n x ∈ D(Am )
i.e. Aα−n x ∈ D(An ).
For α = 0 we get immediately A0 = I . Moreover for Re α > 0 we get
D(Aα ) = A−α (X); Aα = (A−α )−1 .
Indeed, Aα−n x ∈ D(An ) iff there is y ∈ X with Aα−n x = A−n y. Such a
y is obviously unique, and Aα x = y by definition. Moreover A−n A−α y
= A−α A−n y = A−α Aα−n x = A−n x so that x = A−α y is in the range of
A−α and Aα = (A−α )−1 .
Let Re α > 0. Since Aα has a bounded inverse, then it is a closed
operator, so that D(Aα ) is a Banach space endowed with the graph norm.
Again, since Aα has a bounded inverse, its graph norm is equivalent to
x → Aα x,
which is usually considered the canonical norm of D(Aα ).
If Re α = 0, α = it with t ∈ R, Ait is the inverse of A−it in the sense
that for each x ∈ D(Ait ), Ait x ∈ D(A−it ) and A−it Ait x = x. Indeed, if
x ∈ D(Ait ) then Ait−1 x ∈ D(A), and Ait x = A(Ait−1 x) by definition.
Therefore A−1−it Ait x = A−1−it A · Ait−1 x = A · A−1−it Ait−1 x = A ·
A−2 x ∈ D(A), which implies that Ait x ∈ D(A−it ) and A−it Ait x = x.
But in general the operators Ait are not bounded, see next Exam-
ple 4.12. However, they are closed operators, because A−1+it is bounded
and A is closed (see next Exercise 7, Section 4.1.1). Therefore also
D(Ait ) is a Banach space under the graph norm.
From the definition it follows easily that for 0 ≤ Re α < n ∈ N, the
domain D(An ) is continuously embedded in D(Aα ): indeed for each x ∈
D(An ), Aα−n x ∈ D(An ) by Proposition 4.4(iii), and Aα x = An Aα−n x =
Aα−n An x so that Aα x ≤ Aα−n An x. This property is generalized
in the next theorem.
Theorem 4.6. Let α, β ∈ C be such that Re β < Re α. Then D(Aα ) ⊂
D(Aβ ), and for every x ∈ D(Aα ),
Aβ x = Aβ−α Aα x.
Moreover for each x ∈ D(Aα ), Aβ x ∈ D(Aα−β ) and
Aα−β Aβ x = Aα x.
Conversely, if x ∈ D(Aβ ) and Aβ x ∈ D(Aα−β ), then x ∈ D(Aα ) and
again Aα−β Aβ x = Aα x.
Proof. The embedding D(Aα ) ⊂ D(Aβ ) is obvious if Re β < 0; it has

to be proved for Re β ≥ 0.
If x ∈ D(Aα ), A−n+α x ∈ D(An ) for n > Re α. Therefore, by Pro-
position 4.4(iii) we have A−n+β x = Aβ−α A−n+α x ∈ D(An ), so that
x ∈ D(Aβ ), and
Aβ x = An Aβ−α A−n+α x = Aβ−α Aα x.
Since Aβ−α is a bounded operator, Aβ x ≤ Aβ−α L(X) Aα x, and
D(Aα ) is continuously embedded in D(Aβ ).
Let again x ∈ D(Aα ), and let n > max {Re α, Re (α − β)}. Then
A−n+α−β Aβ x = A−n+α−β Aβ−α Aα x = A−n Aα x ∈ D(An ),
so that Aβ x ∈ D(Aα−β ) and Aα−β Aβ x = Aα x.

Let now x ∈ D(Aβ ) be such that Aβ x ∈ D(Aα−β ), and fix n >
max{ Re α, Re α − β}. Then
Aα−2n x = Aα−n−β A−n+β x = Aα−n−β A−n Aβ x = A−n Aα−n−β Aβ x
is in D(A2n ), so that x ∈ D(Aα ) and Aα x = A2n Aα−2n x = Aα−β Aβ x.
The condition Re β < Re α is essential in the above theorem when Re

α > 0. In fact for every α > 0, t ∈ R we have D(Aα ) ⊂ D(Aα+it ) if and
only if Ait is bounded. See Exercise 2, Section 4.1.1.
Now we give some representation formulas for Aα x when x ∈ D(Aα ).
We consider first the case where 0 < Re α < 1. Taking n = 1 in the
definition, we see that x ∈ D(Aα ) if and only if Aα−1 x ∈ D(A). Letting
r → 0 and θ → π in the representation formula (4.3) for Aα−1 x (i.e.,
using formula (4.4) with α replaced by α − 1) we get

α−1 sin(πα) ∞ α−1
A x= ξ (ξ I + A)−1 x dξ. (4.6)
π 0
∞
Therefore x ∈ D(Aα ) if and only if the integral 0 ξ α−1 (ξ I + A)−1 x dξ
is in the domain of A, and in this case
∞
sin(πα)
Aα x = A ξ α−1 (ξ I + A)−1 x dξ
π 0
∞ (4.7)
1
= A ξ α−1 (ξ I + A)−1 x dξ ,
(α)(1 − α) 0
which is the well-known Balakrishnan formula.

Another important representation formula holds for −1 < Re α < 1. The

starting point is again formula (4.3) for Aα−1 x. We let θ → π and then
we integrate by parts in the integrals between r and ∞, getting

sin(πα) ∞ α sin(πα)
α−1
A x = ξ (ξ I + A)−2 x dξ − r α (r I + A)−1 x
πα r πα

r α π iηα iη
− e (re I + A)−1 x dη
2π −π
(with (sin(πα))/(πα) replaced by 1 if α = 0) and letting r → 0 we get

(both for Re α ∈ (0, 1) and for Re α ∈ (−1, 0])
∞
α−1 1
A x= ξ α (ξ I + A)−2 x dξ. (4.8)
(1 − α)(1 + α) 0
∞
Therefore x ∈ D(Aα ) if and only if the integral 0 ξ α (ξ I + A)−2 x dξ is
in the domain of A, and in this case
∞
1
Aα x = A ξ α (ξ I + A)−2 x dξ. (4.9)
(1 + α)(1 − α) 0
The most general formula of this type may be found as usual in the book
of Triebel: for n ∈ N ∪ {0}, m ∈ N, −n < Re α < m − n we have
∞
α (m)
A x= A m−n
t α+n−1 (t I + A)−m x dt
(α + n)(m − n − α) 0
for every x ∈ D(Aα ). See [90, Section 1.5.1].

We already know that the domain D(A) is continuously embedded in
D(Aα ) for Re α ∈ [0, 1). With the aid of the representation formulas
(4.7) and (4.9) we are able to prove more precise embedding properties
of D(Aα ).
Proposition 4.7. For 0 < Re α < 1, D(Aα ) belongs to JRe α (X, D(A)) ∩
K Re α (X, D(A)), i.e.
(X, D(A))Re α,1 ⊂ D(Aα ) ⊂ (X, D(A))Re α,∞ .
Proof. The embedding (X, D(A))Re α,1 ⊂ D(Aα ) is easy, because for
ξ > 0 we have
Aξ α−1 (ξ I + A)−1 x = ξ Re α−1 A(ξ I + A)−1 x

and for every x ∈ (X, D(A))Re α,1 the function ξ → ξ Re α A(A+ξ I )−1 x
is in L 1∗ (0, ∞) by Proposition 3.1. Using the representation formula (4.6)
for Aα−1 x, we get Aα−1 x ∈ D(A), i.e. x ∈ D(Aα ) and by (4.7)
∞
1 dξ
Aα x ≤ ξ Re α A(A + ξ I )−1 x
|(α)(1 − α)| 0 ξ
≤ Cx(X,D(A))Re α,1 .
Let now x ∈ D(Aα ). Then x = A−α y, with y = Aα x, so that x =
A · A−α−1 y. We use the representation formula (4.8) for A−α−1 y, that
gives ∞
A
x= t −α (A + t I )−2 y dt.
(1 − α)(1 + α) 0
On the other hand, by Proposition 3.1 we have
x ∈ (X, D(A))Re α,∞ ⇐⇒ sup λRe α A(A + λI )−1 x < +∞,
λ>0
and in this case

x(X,D(A))Re α,∞ ≤ C(α) sup λRe α A(A + λI )−1 x.
λ>0
So, we estimate
supλ>0 λRe α A(A + λI )−1 x
Re α 2
λ A (A + λI )−1 ∞ −α

= supλ>0 t (A + t I ) y dt
−2
(1 − α)(1 + α) 0 .
For every λ > 0 we have

Re α 2
∞
λ A (A + λI )−1 t −α
(A + t I ) y dt
−2

0
λ
M
≤ λRe α t −Re α (M + 1)2 y dt
1+λ 0
∞
Re α M(M + 1)
+λ (M + 1) t −Re α ydt
λ 1+t
≤ Cy
so that x ∈ (X, D(A))Re α,∞ and
x(X,D(A))Re α,∞ ≤ C y = C Aα x,
which implies that D(Aα ) ⊂ (X, D(A))Re α,∞ .
Remark 4.8. Arguing similarly (using formula (4.9) instead of (4.7)) we

see easily that for every ε ∈ (0, 1) and t ∈ R, (X, D(A))ε,1 is contained
in D(Ait ). Indeed, since ξ it A(ξ I + A)−2 x ≤ M(1 + ξ )−1 A(ξ I +
A)−1 x, then the function ξ → ξ it A(ξ I + A)−2 x is in L 1 (0, ∞) for
∞
each x in (X, D(A))ε,1 , so that the integral 0 ξ it (ξ I + A)−2 x dξ belongs
to the domain of A. Therefore, for every ε ∈ (0, 1) and p ∈ [1, ∞], t ∈
R, (X, D(A))ε, p and (X, D(A))ε are continuously embedded in D(Ait )
(because they are continuously embedded in (X, D(A))ε/2,1 ).
Remark 4.9. Let 0 < α < 1. It is possible to show that in its turn Aα is
a positive operator, and that

α 1 R(z, A)
R(λ, A ) = dz, λ ≤ 0. (4.10)
2πi γr,θ λ − z α
(see Exercise 6, Section 4.1.1). Using the above formula for the resolvent,
one shows that −Aα is a sectorial operator. This may be surprising, since
−A is not necessarily sectorial. This also may help in avoiding mistakes
driven by “intuition”. Consider for instance the case where X = L 2 (0, π)
and A is the realization of −d 2 /dx 2 with Dirichlet boundary condition,
i.e. D(A) = H 2 (0, π) ∩ H01 (0, π) and Au = −u . One could think
that A1/2 is a realization of i d/dx with some boundary condition, but
this cannot be true because such operators are not sectorial. See next
Example 4.34.
4.1.1. Exercises
In exercises 1–8, A is a positive operator in general Banach space X.
1. Let Y be any interpolation space between X and D(A). Prove that

the part of A in Y is a positive operator.
2. Let Re α > 0, t ∈ R \ {0}. Show that D(A α ) ⊂ D(Aα+it ) if and
only if Ait is bounded.
3. Prove that (4.8) holds for −1 < Re α < 1.
4. Show that for any α ∈ C with Re α > 1, Re α ∈ / N, D(Aα) belongs to
J{Reα} (D(A [Reα]
),D(A [Reα]+1
)) and to K {Reα} (D(A[Reα] ),D(A[Reα]+1 )),
where {Re α} and [Re α] are the fractional part and the integral part
of Re α, respectively.
5. Prove that if D(A) is compactly embedded in X, then the operator
Aα is compact for every α with negative real part, and therefore
D(Aα ) is compactly embedded in X for every α with positive real
part.
6. Prove that for every α > 0, Aα is a positive operator, and that (4.10)
holds.
7. Prove that if : D() ⊂ X → X is a linear closed operator, and
B : X → X is a linear bounded operator, then C : D(C) = {x ∈
X : Bx ∈ D()} → X, C x = Bx, is a closed operator. (This is
used to check that Ait is a closed operator for each t ∈ R).
8. Prove that if t, s ∈ R and x ∈ D(Ais ) ∩ D(Ai(t+s) ) then Ais x ∈
D(Ait ) and Ait Ais x = Ai(t+s) x.
9. Prove that A−1 is a nonnegative operator, and that for 0 < Re α < 1
A−α = (A−1 )α .
10. Let A be a positive operator in a Hilbert space H . Show that for each
α ∈ C, (A∗ )α = (Aα )∗ , so that if α is real, then (A∗ )α = (Aα )∗ .
11. Let ⊂ Rn be an open set, let 1 ≤ p ≤ ∞, and set X = L p ()
(resp. X = Cb ()). Let a : → R be a measurable (resp. continu-
ous) function, such that ess inf a(x) > 0. Prove that the multiplica-
tion operator f → a · f , with domain D(A) = { f ∈ X : a f ∈ X},
is positive, and that A z is the multiplication operator by a z .
12. Let A, B be two positive operators such that R(λ, A)R(µ, B) =
R(µ, B) R(λ, A) for some/all λ, µ < 0. Prove that for each z ∈ C
we have
D(AB)z = {x ∈ D(B z ) : B z x ∈ D(A z )} = {x ∈ D(Az ) : A z x ∈ D(B z )}
and
(AB)z x = A z B z x = B z A z x, x ∈ D(AB)z .
4.1.2. Powers of nonnegative operators

A part of the theory of powers of positive operators may be extended to
nonnegative operators.
Definition 4.10. A linear operator A : D(A) ⊂ X → X is called non-
negative if the resolvent set of A contains (−∞, 0) and there is M > 0
such that
M
(λI + A)−1 ≤ , λ > 0.
λ
In other words, A is a nonnegative operator iff (−∞, 0) is a ray of min-
imal growth of the resolvent of A.
An important example of nonnegative operator is the realization A of

− (the Laplace operator) in L p (Rn ), 1 ≤ p ≤ ∞. But A is not positive
because 0 ∈ σ (A). However if p < ∞ then A is one to one. See
Exercise 3, Section 4.1.3.
If 0 ∈ σ (A) but A is one to one, it is still possible to define A z for −1 <
Re z < 1.
Let −1 < Re z < 1, z = 0, and define an operator Bz on D(A)∩ R(A)1
by

sin(π z) x A−1 x
Bz x = −
π z 1+z
1 ∞
−1 −1 −1
+ ξ z+1
(ξ I + A) A x dξ + ξ z−1
(ξ I + A) Ax dξ
0 1
(4.11)
for each x ∈ D(A) ∩ R(A) (note that in the case where 0 ∈ ρ(A), Bz x
coincides with A z x since formula (4.11) is obtained easily from (4.9)).
Then one checks that Bz : D(A) ∩ R(A) → H is closable, and defines
Az as the closure of Bz .
Another way to define Aα for 0 < α < 1, even if A is not one to one,
is the following: for λ > 0 one defines the candidate to be (λI + Aα )−1
by
∞
sin(πα) ξα
Rλ := (ξ I + A)−1 dξ, (4.12)
π 0 λ2 + 2λξ α cos(πα) + ξ 2α
then one checks that Rλ is invertible for every λ > 0 and Rλ Rµ = (Rµ −
Rλ )/(λ − µ). Therefore there exists a unique closed operator B such that
Rλ = (λI + B)−1 for λ > 0, and we set Aα = B. Note that in the
case where 0 ∈ ρ(A) the above formula coincides with the expression of
(λI + Aα )−1 obtained from (4.10) letting r → 0 and θ → π.
From the representation formula (4.12) it follows that
lim (λI + (εI + A)α )−1 = (λI + Aα )−1 , in L(X),

ε→0
which will be used in the proof of next lemma. In its turn, Lemma 4.11
will be used in the proof of Theorem 4.28.
1 By R(A) = {Ax : x ∈ D(A)} we denote the range of the operator A.

Lemma 4.11. Let A : D(A) ⊂ X → X be any nonnegative densely

defined operator. Then D(A + εI )α = D(Aα ) for each α ∈ [0, 1], and
there is C independent of ε such that
(A + εI )α x − Aα x ≤ Cεα x, x ∈ D(Aα ).
Proof. Let 0 < α < 1. For 0 < η < ε, A + εI and A + ηI are positive
operators. Using the Balakrishnan formula (4.7), for each x ∈ D(A) and
δ > 0 we get
(A + εI )α x − (A + ηI )α x =
δ ∞
sin(πα)
= + ξ α [((ξ + ε)I + A)−1 ((ξ + η)I + A)−1 x] dξ
π 0 δ
sin(πα)
= (I1 + I2 ),
π
δ
I1 = ξ α−1 (−xi[((ξ + ε)I + A)−1 + ξ((ξ + η)I + A)−1 )]x dξ
0
δ
= (ε − η) ξ α ((ξ + ε)I + A)−1 ((ξ + η)I + A)−1 x dξ
0
for every x ∈ D(A) and δ > 0. Therefore,

(A + εI )α x − (A + ηI )α x
∞ δ
sin(πα) α−2 α−1
≤ (ε − η)M 2
ξ dξ + 2(1 + M) ξ dξ x
π δ 0

sin(πα) ε − η 2 α−1 2(1 + M) α
= M δ + δ x
π 1−α α
for every x ∈ D(A) and δ > 0. Taking δ = (ε − η) we get
(A + εI )α x − (A + ηI )α x ≤ C(α)(ε − η)α x.
Therefore, for every x ∈ D(A) the function ε → (A+εI )α x is uniformly
continuous, so that there exists the limit limε→0 (A + εI )α x := Bx. Let-
ting η → 0 in the above estimate we find
(A + εI )α x − Bx ≤ C(α)εα x (4.13)
for each ε > 0, for each x ∈ D(A). But D(A) is a core of (A + εI )α
(that is, D(A) is dense in the domain of (A + εI )α ): indeed, for every
y ∈ D((A + εI )α ) the sequence yn = n(n I + A)−1 y is in D(A), yn → y

and (A + εI )α yn = n(n I + A)−1 (A + εI )α y → (A + εI )α y as n → ∞,
because D(A) is dense in X. This implies that B is closable, its closure
B has domain D((A + εI )α ), and inequality (4.13) holds also for B. So,
(A + εI )α x → Bx uniformly for x ∈ D(B), x ≤ 1, and this implies
that (−∞, 0) ⊂ ρ(B), and (λI + (A + εI )α )−1 → (λI + B)−1 as ε → 0
in L(X). Since (λI + (A + εI )α )−1 → (λI + Aα )−1 , then B = Aα .
4.1.3. Exercises
1. Let A be a densely defined nonnegative operator such that R(A) is
dense in X. Show that R(A) ∩ D(A) is dense in X, and that the
operators Bz : D(A) ∩ R(A) → X defined in (4.11) are closable.
2. Let A be a nonnegative one to one operator, and set Aε = (εI +
A)(ε A + I )−1 for ε > 0. Show that (i) ρ(Aε ) ⊃ (−∞, 0], (ii)
(λI + Aε )−1 → (λI + A)−1 for λ > 0, Aε x → x for x ∈ D(A),
A−1 −1
ε x → A x for x ∈ R(A) as ε → 0, (iii) Aε x → A x for
z z
x ∈ D(A) ∩ R(A) as ε → 0.
3. Prove that the realization A of − in L p (Rn ), 1 ≤ p ≤ ∞, is a
nonnegative operator. Prove that if p < ∞ A is one to one, and that
if p = ∞ the kernel of A consists of the constant functions.
Hint: denoting by T (t) the Gauss-Weierstrass
√ semigroup, use the
estimates Di T (t)L(L p ) ≤ C/ t to show that if f = 0 then
Di f = Di T (t) f vanishes for every i = 1, . . . , n, so that f is con-
stant.
4.2. Operators with bounded imaginary powers

Let again A be a positive operator in a complex Banach space X. We
know that the operators A z are bounded if Re z < 0 and unbounded
in general if Re z > 0 (this is because D(A z ) ⊂ (X, D(A))Re z,∞ by
Proposition 4.7). Moreover Remark 4.8 tells us that for every t ∈ R,
the domain D(Ait ) is not very far from the underlying space X because
all the interpolation spaces (X, D(A))ε, p are continuously embedded in
D(Ait ). So, it is natural to ask whether D(Ait ) = X and Ait is a bounded
operator also for t = 0. The general answer is “no”, as the following
example shows.
Example 4.12. Let S be the shift operator, S(ξ1 ,ξ2 ,ξ3 , ...) = (0,ξ1 ,ξ2 , ...),
in X = c0 = the space of all complex valued sequences ξn such that
limn→∞ ξn = 0, endowed with the sup norm, and set A := (I − S)−1 .
Then A is a positive operator, and Ait is unbounded for every t = 0.
Proof. It is easily seen that the domain ∞ of A is the subset of c0 consisting

of the sequences ξ = {ξn } such that n=1 ξn = 0 (which is dense in c0 ),
and
A(ξ1 , ξ2 , ξ3 , . . .) = (ξ1 , ξ1 + ξ2 , ξ1 + ξ2 + ξ3 , . . .).
An easy computation shows that for λ > 0
∞ k−1
−1 I 1 λ
(λI + A) = − Sk ,
λ + 1 (λ + 1)2 k=1 λ + 1
so that
∞ k−1
1−1 1 λ 2
(λI + A) ≤ 1+ ≤ ,
λ+1 λ + 1 k=1 λ + 1 λ+1
which implies that A is a positive operator. Replacing in (4.2), for Re

α < 0, and then also for Re α = 0, we get
α(α + 1) 2 α(α + 1)(α + 2) 3

Aα = I + αS + S + S + ...
2 3!
So, for α = it the n-th component of Ait ξ is
it (it + 1) it (it + 1) · . . . · (it + n − 2)

ξn + it ξn−1 + ξn−2 + . . . + ξ1 .
2 (n − 1)!
Fix any n ∈ N and define a sequence ξ ∈ D(A) as follows:
1 1 2
ξn = 0, ξn−1 = , ξn−2 = , ξn−3 = ,...,
i i(it + 1) i(it + 1)(it + 2)
(n − 2)!
ξ1 = ,
i(it + 1) · . . . · (it + n − 2)
∞
while for k > n ξk is arbitrary, subject only to |ξk | ≤ 1 and k=1 ξk = 0.
Then we get

1 1 1
A ξ ≥ |(A ξ )n | = |t| 1 + + + . . . +
it it
2 3 n−2
while the norm of ξ is 1. Letting n → ∞ we obtain supξ ∈D(A), ξ =1

Ait ξ = +∞, so that Ait is unbounded.
Let us see the behavior of A z for Re z < 0, z close to the imaginary

axis. From the representation formula (4.4) we get easily, using (4.5),
∞ Re z
M ξ
A ≤ | sin(π z)|
z
dξ
π 0 ξ +1
(4.14)
| sin(π z)|
≤M , Re z ∈ (−1,0).
| sin(πRe z)|
In particular, for real z = −α, with 0 < α < 1 we get A−α ≤ M,

so that A−α is bounded on the real interval (−1, 0), and hence it is
bounded on any real interval (−a, 0), a > 0. But in general A z may be
unbounded in other subsets of the left half-plane such that ∩iR = ∅.
However, if the operator Ait is bounded for t ∈ I ⊂ R and Ait ≤ C
for every t ∈ I , then for z = −α + it we have
A−α+it ≤ A−α Ait ≤ MC, 0 < α < 1/2, t ∈ I.
A sort of converse of the above considerations is in the next lemma. It

gives a simple (but hard to be checked) sufficient condition for Ait to be
bounded.
Lemma 4.13. Let A be a positive operator with dense domain D(A).

Assume that there are a set ⊂ {z ∈ C : Re z < 0} and a constant
C > 0 such that ∩ iR = ∅ and A z ≤ C for z ∈ . Then for every
t ∈ R such that it ∈ , Ait is a bounded operator and Ait ≤ C.
Proof. For every x ∈ D(A) the function z → Az x is continuous for

Re z < 1, so that limz→it, z∈ A z x = Ait x. Since D(A) is dense and
Az ≤ C for z ∈ , it follows that for every x ∈ X there exists the
limit limz→it, z∈ A z x. Denoting such a limit by T x, we get T x ≤
Cx. Then Ait is a closed operator which coincides with the bounded
operator T on a dense subset. This implies that Ait = T so that Ait is
bounded.
The most popular examples of positive unbounded operators with

bounded imaginary powers are m-accretive positive operators in Hilbert
spaces, which will be discussed in the next section. Self-adjoint positive
operators belong to this class.
A simple example in non Hilbert spaces is the realization of the de-
rivative in L p (0, T ), with domain {u ∈ W 1, p (0, T ) : u(0) = 0}, for
1 < p < ∞: see Proposition 4.23 for a generalization.
Another class of operators with bounded imaginary powers is the fol-
lowing ([37]).
Proposition 4.14. Let A : D(A) → X be any positive operator, and let

0 < θ < 1, 1 ≤ p ≤ ∞. Then the parts of A in (X, D(A))θ, p and in
(X, D(A))θ have bounded imaginary powers.
Proof. It is not hard to check (see Exercise 1, Section 4.2.1) that the part
of A in any of the above spaces is still a positive operator.
First we consider the case p = ∞. Let Aθ : D A (θ + 1, ∞) →
D A (θ, ∞) be the part of A in D A (θ, ∞). We already know (Remark 4.8)
that D A (θ, ∞) = (X, D(A))θ,∞ is contained in D(Ait ). Therefore for
every x ∈ D A (θ, ∞), Ait−1 x belongs to the domain D(A). To obtain an
estimate for Ait x = A(Ait−1 x) we use the representation formula
(4.8) for Ait−1 x,
∞
1
A x =
it−1
ξ it (A + ξ I )−2 x dξ
(1 − it)(1 + it) 0

sin(πit) ∞ it
= ξ (A + ξ I )−2 x dξ,
πit 0
which gives
∞
eπt − e−πt M
A x ≤
it
ξ θ A(A + ξ I )−1 x dξ
2πt 0 ξ θ (1+ ξ)
eπt − e−πt
≤ C(θ) x(X,D(A))θ,∞ .
t
We prove now that in fact Ait−1 x belongs to D A (θ + 1, ∞). We use again
the representation formula (4.8) for Ait−1 x, which implies that for every
λ>0
λθ A(λI + A)−1 A(Ait−1 x)
λ
eπt −e−πt

≤
θ
λ (λI + A) −1
ξ it−θ
A(A+ξ I ) ξ A(A+ξ I ) x dξ
−1 θ −1

2πt 0
∞
eπt −e−πt
θ −1 −1 θ −1

+ λ A(λI + A) ξ it−θ
(A+ξ I ) ξ A(A+ξ I ) x dξ
2πt λ

−θ
eπt −e−πt Mλθ (M + 1)λ1−θ θ Mλ
≤ + (M + 1)λ x(X,D(A))θ,∞
2πt λ+1 1−θ θ
eπt − e−πt
≤ C x(X,D(A))θ,∞ .
t
Therefore, Ait−1 x belongs to D A (θ + 1, ∞), which is the domain of Aθ

in D A (θ, ∞). It follows that x is in the domain of Aitθ , and
eπt − e−πt
Aitθ x(X,D(A))θ,∞ ≤ C x(X,D(A))θ,∞ .
t
The rest of the statement follows by interpolation: knowing that for 0 <
θ1 < θ2 < 1 the part of Ait in (X, D(A))θ1 ,∞ and in (X, D(A))θ2 ,∞
is a bounded operator, from Theorem 1.6 it follows that for every θ ∈
(θ1 , θ2 ) the part of Ait in (X, D(A))θ, p and in (X, D(A))θ is a bounded
operator.
Another important result is proved through the so called “transference

principle”. See Coifman–Weiss [25].
Theorem 4.15. Let (, µ) be a σ -finite measure space, and let 1 < p <
∞. If A is a positive operator in L p (, µ) such that (λI + A)−1 ≤ 1/λ
and (λI + A)−1 is positivity preserving for λ > 0 (i.e. f (x) ≥ 0 a.e.
implies ((λI + A)−1 f )(x) ≥ 0 a.e), then the operators Ait are bounded
in L p (, µ), and there is C > 0 such that
Ait ≤ C(1 + t 2 )eπ|t|/2 , t ∈ R.
Let us come back to the general theory. By Theorem 4.6, if the operators
Ait are bounded for any t in a small neighborhood of 0, then they are
bounded for every t ∈ R. Moreover, if Ait ≤ C for −δ ≤ t ≤ δ, then
there exists C , γ > 0 such that Ait ≤ C eγ |t| for every t ∈ R.
Lemma 4.16. Let A be a positive operator such that Ait ∈ L(X) for
every t ∈ R, and t → Ait is locally bounded. Then for every x ∈ D(A)
the function z → A z x is continuous in the closed half-plane Re z ≤ 0.
Proof. If x ∈ D(A) then z → A z x is holomorphic for Re z < 1, so that

it is obviously continuous for Re z ≤ 0. We have already remarked that
(4.14) implies Aα ≤ M for −1/2 < α < 0, so that for z = α + iβ,
−1/2 < α < 0, we get
A z ≤ MAiβ .
In particular, for every t ∈ R and r > 0 small enough the norm A z − Ait
is bounded in the half circle {z : |z − it| ≤ r, Re z ≤ 0}, by a constant
independent of z. It follows that for every x ∈ D(A), limz→it A z x =
Ait x.
Note that if x ∈
/ D(A) the function z → A z x cannot be continuous in
(−∞, 0]. Indeed, by Proposition 4.7, A−t x ∈ (X, D(A))t,∞ for 0 < t <
1, and (X, D(A))t,∞ is contained in D(A). Therefore, if t → A−t x is
continuous in (−∞, 0] then x ∈ D(A).
The family of operators {Ait : t ∈ R} plays an important role also in
the interpolation properties of the domains D(A z ).
Theorem 4.17. Let A be a positive operator with dense domain such that
for every t ∈ R Ait ∈ L(X), and there are C, γ > 0 such that
Ait ≤ Ceγ |t| , t ∈ R.
Then for 0 ≤ Re α < Re β
[D(Aα ), D(Aβ )]θ = D(A(1−θ)α+θβ ).
Proof. By Theorem 4.6 we may assume that α = 0 without loss of gen-

erality. Moreover since Ait is bounded for every t ∈ R, then D(Aβ ) =
D(ARe β ) for Re β > 0. See Exercise 2, Section 4.2.1. So we may also
assume that β ∈ (0, ∞).
Let x ∈ D(Aθβ ), and set
2
f (z) := e(z−θ) A−(z−θ)β x, 0 ≤ Re z ≤ 1.
2
(The factor e(z−θ) is here just to make f bounded. If γ = 0 we can take
f (z) = A−(z−θ)β x).
Let us prove that f ∈ F(X, D(Aβ )). f is obviously holomorphic in
the strip Re z ∈ (0, 1) and continuous up to Re z = 1 with values in
X. Since D(A) is dense in X, f is also continuous up to Re z = 0 with
values in X. Indeed, A−(z−θ)β x = A−zβ Aθβ x, and we know from Lemma
4.16 that w → Aw y is continuous with values in X for Re w ≤ 0 for
every y ∈ D(A) = X. Similarly, t → f (1 + it) is continuous with
values in D(Aβ ). f is also bounded, since
A−(z−θ)β x = A−βiIm z A−βRe z Aθβ x ≤ A−βRe z Ceγβ|Im z| Aθβ x
Therefore, f ∈ F(X, D(Aβ )). Since f (θ) = x, then x ∈ [X, D(Aβ )]θ
and
2 +θ 2
x[X,D(Aβ )]θ ≤ max{supt∈R e−t A−(it−θ)β x,
2 +(1−θ)2
supt∈R e−t A−(1+it−θ)β x D(Aβ ) }
≤ C Aθβ x.
It follows that D(Aθβ ) is continuously embedded in [X, D(Aβ )]θ .

Conversely, let x ∈ D(Aβ ), and let f ∈ F(X, D(Aβ )) be such that
f (θ) = x.
The function
2
F(z) = e(z−θ) A zβ f (z),
is continuous with values in X both for Re z = 0 and for Re z = 1, and
we have
2 +θ 2
supt∈R F(it) ≤ supt∈R e−t Ceγβ|t| supt∈R f (it)
(4.15)

≤ C f F (X,D(Aβ )) ,
2 +(1−θ)2
supt∈R F(1 + it) ≤ supt∈R e−t Ceγβ|t| supt∈R Aβ f (1 + it)
≤ C f F (X,D(Aβ )) ,
(4.16)
so that F is bounded with values in X for Re z = 0 and for Re z = 1.
If F would be holomorphic in the interior of S and continuous in S, we
could apply the maximum principle to get “Aθ x ≤ C f F (X,D(Aβ )) ”.
But in general F is not even defined in the interior of S, because f has
values in X and not in the domain of some power of A. Therefore, we
have to modify this procedure.
By Remark 2.5,
x[X,D(Aβ )]θ = inf{ f F (X,D(Aβ )) : f ∈ V(X, D(Aβ )), f (θ) = x}.
Let f ∈ V(X, D(Aβ )) be such that f (θ) = x. The function
2
F(z) = e(z−θ) A zβ f (z), 0 ≤ Re z ≤ 1,
is now well defined and holomorphic for Re z ∈ (0, 1), continuous with
values in X up to Re z = 0, Re z = 1, and bounded with values in X. By
the maximum principle (see Exercise 1, Section 2.1.3),
A θβ x = F(θ) ≤ max{supt∈R F(it), supt∈R F(1 + it)}
≤ C f F (X,D(Aβ )) ,
where the last inequality follows from estimates (4.15) and (4.16).
Taking the infimum over all the f ∈ V(X, D(Aβ )) such that f (θ) = x
we get
Aθβ x ≤ C x[X,D(Aβ )]θ , x ∈ D(Aβ ).
Since D(Aβ ) is dense in [X, D(Aβ )]θ it follows that [X, D(Aβ )]θ is con-
tinuously embedded in D(Aθβ ).
4.2.1. The sum of two commuting operators with bounded imaginary

powers
We consider now two positive operators, A : D(A) ⊂ X → X, B :
D(B) ⊂ X → X, having bounded imaginary powers such that
Ait ≤ Meγ A |t| , B it ≤ Meγ B |t| , t ∈ R. (4.17)
We also assume that A and B commute in the resolvent sense,
R(λ, A)R(µ, B) = R(µ, B)R(λ, A), λ, µ ≤ 0. (4.18)
Our assumptions imply immediately (through formula (4.2)) that A z B w =

B w A z for Re w, Re z < 0 and also (less immediately but easily) for Re
w, Re z ≤ 0.
We shall study the closability and the invertibility of the operator A +
B, following the approach of Dore and Venni [38].
Proposition 4.18. Let A and B be positive operators satisfying (4.17)
and (4.18). Assume in addition that γ A + γ B < π and that D(A) or
D(B) is dense in X. Then A + B : D(A) ∩ D(B) → X is closable, and
its closure A + B is invertible with inverse S given by

1 a+i∞ A−z B z−1 1 a+i∞ B z−1 A−z
S= dz = dz (4.19)
2i a−i∞ sin(π z) 2i a−i∞ sin(π z)
with any a ∈ (0, 1). Moreover, S is a left inverse of A + B.
Proof. Since γ A + γ B < π the norm of the operator A−z B z−1 / sin(π z)
in the integral decays exponentially as |Im z| → ∞. Therefore S is a
bounded operator, and since z → A−z B z−1 / sin(π z) is holomorphic in
the strip Re z ∈ (0, 1), S is independent of a ∈ (0, 1).
The proof is in three steps: (i) we show that S is a left inverse of A + B;
(ii) we show that if D(A) (resp. D(B)) is dense, then for every ε ∈ (0, 1)
S maps D(A1−ε ) (resp. D(B 1−ε )) to D(A) ∩ D(B) and (A + B)Sx = x
for each x ∈ D(A1−ε ) (resp. x ∈ D(B 1−ε )); (iii) using steps (i) and (ii)
−1
we show in a standard way that A + B is closable and S = A + B .
(i) For every x ∈ D(A) ∩ D(B) it holds

1 a+i∞ B z−1 A1−z x A−z B z x
S(Ax + Bx) = + dz
2i a−i∞ sin(π z) sin(π z)

1 a−1+i∞
B z A−z x 1 a+i∞
A−z B z x
=− dz + dz.
2i a−1−i∞ sin(π z) 2i a−i∞ sin(π z)
The function g : = {z ∈ C : −1 < Re z < 1} → X defined by

 z −z
 B A x, −1 < Re z ≤ 0,
g(z) =

A−z B z x, 0 ≤ Re z < 1,
is holomorphic in the strips −1 < Re z < 0, 0 < Re z < 1 and continuous

up to the imaginary axis (see Exercise 3, Section 4.1.3). Therefore it is
holomorphic in the whole strip . It follows that z → g(z)/ sin(π z) is
holomorphic in \ {0}. Moreover it has a simple pole at z = 0, with
residue x/π, and it decays exponentially, uniformly for −1 < Re z < 1,
as | Im z| → ∞. It follows that

g(z)
S(A + B)x = πRes , 0 = x.
sin(π z)
(ii) Assume for instance that that D(A) is dense. Fix x ∈ D(A1−ε ),
with 0 < ε < 1 and choose a = ε in the definition of S. The function
z → A1−z B z−1 x/ sin(π z) is well defined because B z−1 maps D(A1−ε ) =
D(A1−z ) into itself, and it is integrable over ε + iR, since
A1−z B z−1 x = Aε−z B z−1 A1−ε x ≤ Ce(γ A +γ B )|Im z| A1−ε x.
Therefore, Sx ∈ D(A) and

a+i∞ a+i∞
1 A1−z B z−1 x 1 B z−1 A1−z x
ASx = dz = dz.
2i a−i∞ sin(π z) 2i a−i∞ sin(π z)
To show that Sx ∈ D(B), we remark that the map z → B z−1 A−z x/ sin(π z)
is holomorphic for ε−1 < Re z < 1, z = 0, continuous up to Re z = ε−1
(because D(A) is dense, see Lemma 4.16), and it decays exponentially
as |Im z| → ∞. Therefore, we may shift the vertical line Re z = a to Re
z = ε − 1 in the definition of S, to get
ε−1+i∞ z−1 −z
1 B z−1 A−z x B A x
Sx = dz + πRes ,0
2i ε−1−i∞ sin(π z) sin(π z)
ε−1+i∞
1 B z−1 A−z x
= dz + B −1 x.
2i ε−1−i∞ sin(π z)
As easily seen, the integral defines an element of D(B). Therefore Sx ∈

D(B) and
1 ε−1+i∞ B z A−z x
B Sx = dz + x
2i ε−1−i∞ sin(π z)
ε+i∞
1 B z−1 A1−z x
=− dz + x
2i ε−i∞ sin(π z)
= −ASx + x.
(iii) Let us show that A + B is closable. Let xn ∈ D(A) ∩ D(B) be such
that xn → 0, (A + B)xn → y as n → ∞. Since S is a left inverse of
A + B then
0 = lim xn = lim S(A + B)xn = Sy.

n→∞ n→∞
Since A−1 y ∈ D(A) ⊂ D(A1−ε ) for each ε > 0, then by step (ii)
S A−1 y ∈ D(A) ∩ D(B) and
A−1 y = (A + B)S A−1 y = (A + B)A−1 Sy = 0,
so that y = 0 and A + B is closable.

As a last step we prove that A + B is invertible and its inverse is S. If
x ∈ D(A + B) there is a sequence xn ∈ D(A) ∩ D(B) such that xn → x
and (A + B)xn → A + B x as n → ∞. Then
x = lim xn = lim S(A + B)xn = S A + Bx,

n→∞ n→∞
which means that S is a left inverse of A + B. Now assume again that

D(S) is dense in X, and for x ∈ X let x n ∈ D(A) be such that xn → x
as n → ∞. By step (ii) Sxn ∈ D(A) ∩ D(B), limn→∞ Sxn = Sx
and limn→∞ (A + B)Sxn = limn→∞ xn = x. This implies that Sx ∈
D(A + B) and A + B Sx = x, so that S is also a right inverse of A + B.
Under the assumptions of Proposition 4.18 the operator A + B is not

closed in general. If we knew that A + B is closed, it would follow that
A + B is invertible with bounded inverse. In the next proposition we use
this fact to get information on the resolvent set of a positive operator with
bounded imaginary powers.
Proposition 4.19. Let A be a positive operator with bounded imaginary
powers, satisfying
Ait ≤ Ceγ A |t| , t ∈ R.
If 0 ≤ γ A < π, the resolvent set of A contains the sector {λ ∈ C :

|arg (λ − π)| < π − γ A }.
Proof. We apply Proposition 4.18 to the operators A, B = −λI for every
λ in the sector. If −λ = ρeiθ with ρ > 0, θ ∈ (−π, π), then B it =
(−λ)it I = e−θt , so that γ B = θ = arg (−λ). Since D(B) = X
and A + B = A − λI is closed, the statement follows from Proposition
4.18.
Under a further suitable assumption on the space X, the conditions of
Proposition 4.18 are sufficient for A + B to be closed. Such assumption
has several equivalent formulations.
The “geometric” formulation is the following: there exists a symmetric
function ζ : X × X → R which is convex in each variable, such that
ζ(0, 0) > 0, and ζ(x + y) ≤ x + y whenever x ≤ 1 ≤ y. In this
case the space X is called ζ -convex.
The “probabilistic” formulation is the following: there are p ∈ (1, ∞),
C > 0 such that for every probability space (, F, P) and for every
martingale {u k : → X} with respect to any filtration {Fk }, for each
choice of εk ∈ {−1, 1}, and for each n ∈ N we have

n n

εk (u k − u k−1 ) ≤ C (u k − u k−1 )
k=0 p k=0 p
L (;X) L (;X)
(where we set u −1 = 0). In this case X is called UMD space, or to have

the property of unconditionality of martingale differences.
The formulation which is useful here is the following: for some p ∈
(1, ∞) and ε > 0, the truncated Hilbert transform

1 f (t − y)
(Hε f )(t) = dy, t ∈ R
π |y|≥ε y
is a bounded operator from L p (R; X) to itself. Then it is possible to
show that this is true for every p ∈ (1, ∞) and ε > 0. Moreover for each
f ∈ L p (R; X) there exists the limit
lim Hε f
ε→0
in L p (R; X) and a.e. pointwise. Such a limit is denoted by H f and it is

called the Hilbert transform of f .
Equivalence of the above properties is not trivial. Bourgain [10] showed
that any UMD space has the Hilbert transform property. The converse
was proved by Burkholder [15], who also proved in [14] that X is ζ -
convex iff it is a UMD space.
Theorem 4.20. Let X be a UMD space, and let A, B be densely defined2

positive operators in X satisfying the assumptions of Proposition 4.18.
Then A + B : D(A) ∩ D(B) → X is closed, and 0 ∈ ρ(A + B).
Proof. After Proposition 4.18, we have only to show that S maps X into
D(A) ∩ D(B). This obviously implies that S is a right inverse of A +
B, since by Proposition 4.18 S is a right inverse of A + B. Again by
Proposition 4.18, S is a left inverse of A + B. Therefore S is the inverse
of A + B and 0 ∈ ρ(A + B).
Let us show that S maps X into D(B). For 0 < ε < 1/2 we have

1 A−z B z−1 x
Sx = dz
2i ε sin(π z)
where ε is the curve {is : |s| ≥ ε} ∪ {z ∈ C : |z| = ε}, Re z ≥ 0,
oriented with increasing imaginary part. So,

1 A−is B is−1 x 1 π/2 εeiθ iθ iθ
Sx = ds + A−εe B εe −1 x dθ
2 |s|≥ε sin(πis) 2 −π/2 sin(πεeiθ )
= I1,ε + I2,ε .
Since D(A) is dense, I2,ε goes to B −1 x/2 as ε → 0. Moreover I1,ε is in

D(B) for every ε. We reach our goal if we show that I1,ε converges in
D(B) as ε → 0, i.e. if B I1,ε converges in X as ε → 0. Indeed, in that
case we have
Sx = B −1 x/2 + lim I1,ε ∈ D(B).
ε→0
Let us split B I1,ε into the sum

1 A−is B is x 1 A−is B is x
B I1,ε = ds + ds
2 |s|≥1 sin(πis) 2 ε≤|s|≤1 πis

1 −is 1 1
+ A is
B x − ds.
2 ε≤|s|≤1 sin(πis) πis
The first term is independent of ε, the third one is easily seen to converge
as ε → 0 because 1/(sin(πis)) − 1/(πis) is bounded. The second term
is equal to 1/(2i)Hε f (0) where
f (s) = χ(−1,1) (s)Ais B −is x, s ∈ R.
2 Every UMD space X is reflexive, and therefore by [61] all positive operators in X are densely
defined.
Since f ∈ L p (R; X) for each p > 1 and X is ζ -convex, the truncated

Hilbert transform of f converges in X for almost every t ∈ R. Let us
prove that 0 is one of such t’s. Fix once and for all t ∈ (0, 1) such that
Hε f (t) converges. Then

Ai(t−s) B i(s−t) x
π(Hε f )(0) = A−it B it ds = A−it B it ·
ε≤|s|≤1 s
t+1 i(t−s) i(s−t)
f (t −s) t−1
Ai(t−s) B i(s−t) x A B x
· ds + ds − ds
|s|≥ε s −1 s 1 s
converges as ε → 0. This concludes the proof that Sx ∈ D(B). In the
same way one shows that Sx ∈ D(A), and the statement follows.
Remark 4.21. ([87]) Assume that two positive operators A, B satisfy
(4.18) and that A + B : D(A + B) = D(A) ∩ D(B) → X is a pos-
itive operator. Then Aα B 1−α (A + B)−1 is a bounded operator for each
α ∈ (0, 1).
Indeed, (A(A + B)−1 )α and (B(A + B)−1 )1−α are bounded, since they
are powers of bounded operators, moreover by Exercise 12, Section 4.1.1,
we have (A(A + B)−1 )α = Aα (A + B)−α , and (B(A + B)−1 )1−α =
B 1−α (A + B)1−α . Therefore,
Aα B 1−α (A + B)−1 = Aα (A + B)−α B 1−α (A + B)1−α ∈ L(X).
Proposition 4.18 and Theorem 4.20 have to be compared with Propos-
ition 3.17 and Theorem 3.18 , respectively. In Section 3.2.2 X is any
Banach space, and no assumptions on the powers of A and B are made.
An invertibility result is proved for the parts of A + B in the spaces
D A (θ, p) and D B (θ, p). Here we have stronger assumptions that yield
invertibility in X.
4.2.2. Maximal L p regularity for evolution equations in UMD spaces

The main application – at least, from our point of view – of the theorem
of Dore and Venni is to evolution equations in a UMD space X,

 u (t) + Au(t) = f (t), 0 < t < T,
(4.20)

u(0) = 0,
with f ∈ L p (0, T ; X) for some p ∈ (1, ∞), T > 0. Here we assume that
A : D(A) ⊂ X → X is a positive operator having bounded imaginary
powers, and
Ais ≤ Ceγ A |s| , s ∈ R.
We read problem (4.20) as the equation
Au + Bu = f
in the space X := L p (0, T ; X), where A and B are defined by
A : L p (0, T ; D(A)) → X , (Au)(t) = Au(t),
B : D(B) = {u ∈ W 1, p (0, T ; X) : u(0) = 0} → X , (Bu)(t) = u (t).

By Exercise 4, Section 4.2.3, A is positive and has bounded imaginary
powers, estimated by
Ais ≤ Ceγ A |s| , s ∈ R.
By Exercise 5, Section 3.1.1, B is a positive operator. To show that it has

bounded imaginary powers we shall use an following extension of the
Mikhlin Theorem to function with values in UMD Banach spaces, proved
in [93]. We recall that the Fourier and the inverse Fourier transforms of a
function f ∈ L 1 (R; X) are defined by

1
F f (ξ ) := √ e−iξ t f (t) dt, ξ ∈ R,
2π R
F −1 f := F( f (−·)).
Theorem 4.22. Let X be a UMD space, and let 1 < p < ∞. If m :

R → C is a differentiable function such that
| | |m| | | := max{sup |m(ξ )|, sup |ξ m (ξ )|} < ∞,

ξ ∈R ξ ∈R
then m is a multiplier in L p (R; X), i.e. f → F −1 (m · F f ) is a bounded

operator in L p (R; X). Moreover,
F −1 mFL(L p (R;X)) ≤ c p | | |m| | |.
Proposition 4.23. The operator B defined above has bounded imaginary

powers, satisfying the estimate
Bit ≤ C(1 + |t|)eπ|t|/2 , t ∈ R. (4.21)
Proof. We shall show that for 0 < ε < 1 and 0 < Re z < 1 we have
(εI + B)−z L(X ) ≤ C(1 + |z|)ε−Re z eπ|Im z|/2 , (4.22)

with C independent of z and ε. Then the statement will be proved letting

first Re z to 0 and then ε to 0.
First step: estimate (4.22).
Formula (4.6) and Exercise 5, Section 3.1.1 give (recalling that sin(π(1−z))
π
=
1
(z)(1−z)
)
t
−z sin(π(1 − z)) ∞ −z −(λ+ε)(t−s)
((εI +B) f )(t) = λ e f (s)ds dλ
π 0 0
t ∞
1 −(λ+ε)(t−s) −z
= f (s) e λ dλ ds
(z)(1 − z) 0 0
t
1
= (t − s)z−1 e−ε(t−s) f (s)ds
(z) 0
= (ψε,z f )(t)
where
t z−1 e−εt
ψε,z (t) := 11(0,∞) (t)
(z)
and f is the null extension of f to the whole R.
If f ∈ C0∞ ((0, T ); X) (the space of X-valued smooth functions with
support contained in (0, T )) then f ∈ C0∞ (R; X), so that (εI + B)−z f is
the restriction to (0, T ) of F −1 (m z,ε · F
f ), where m z,ε := Fψε,z is given
by
∞
1 −iξ t z−1 −εt (ε + iξ )−z
m ε,z (ξ ) = √ e t e dt = √ e−λ λz−1 dλ
2π(z) 0 2π(z) R
and R is the half-line {t (ε + iξ ) : t ≥ 0}. As easily seen,
∞
−λ z−1
e λ dλ = e−λ λz−1 dλ = (z),
R 0
so that
(ε + iξ )−z
m ε,z (ξ ) = √ , ξ ∈ R. (4.23)
2π
√
Now, |m ε,z (ξ )| ≤ ε−Re z eπ|Imz|/2 / 2π and m ε,z (ξ) = −i z(ε+iξ )−1 m ε,z (ξ),
so that
1
| | |m ε,z | | | ≤ √ (1 + |z|)ε−Re z eπ|Im z|/2 .
2π
Theorem 4.22 implies that m ε,z is a Fourier multiplier in L p (R; X) and
cp
F −1 m ε,z · FL(L p (R;X)) ≤ √ (1 + |z|)ε−Re z eπ|Im z|/2 .
2π
Therefore,
(εI + B)−z f X ≤ F −1 m ε,z · F
f ) L p (R;X)
≤ C(1 + |z|)ε−Re z eπ|Im z|/2 f X ,

with C independent of z, ε, and f . Since C0∞ ((0, T ); X) is dense in X ,
estimate (4.22) follows.
Second step: Re z → 0.
Fix t ∈ R. Letting z = α − it in (4.22), with α < 0, t ∈ R, we see that
the norm (εI + B)−α+it L(X ) stays bounded as α → 0+ . By Lemma
4.13, (εI + B)it is a bounded operator and
(εI + B)it L(X ) ≤ C(1 + |t|)eπ|t|/2 , t ∈ R. (4.24)
Third step: ε → 0.
By formula (4.9), for 0 < ε < 1 and f ∈ D(B) we have
∞
1
(εI + B) f =
it
ξ it ((ξ + ε)I + B)−2 B f dξ,
(1 + it)(1 − it) 0
so that (εI + B)it f goes to Bit f as ε → 0+ . Using estimate (4.24) we
obtain
Bit f X = lim+ (εI + B)it f X ≤ C(1 + |t|)eπ|t|/2 , t ∈ R.
ε→0
Since D(B) is dense in D(Bit ) and Bit is closed, then Bit is bounded and
estimate (4.21) holds.
A note about the proof: we introduced the operators (εI +B)−z instead
of using B −z from the very beginning, because the above computation of
m ε,z is immediate. Formula (4.23) holds also at ε = 0, but the proof
requires some effort.
Therefore, if γ A < π/2 it is possible to apply Theorem 4.20 to equation
(4.20), seen as the equation Au + Bu = f in X = L p (0, T ; X). We
obtain the following maximal regularity result.
Theorem 4.24. Let A be a positive operator with bounded imaginary
powers satisfying
Ait L(X) ≤ Ceγ A |t| , t ∈ R,
for some C > 0 and γ A < π/2. Let 1 < p < ∞ and T > 0. Then
for every f ∈ L p (0, T ; X) problem (4.20) has a unique solution u ∈
W 1, p (0, T ; X) ∩ L p (0, T ; D(A)), which depends continuously on f .
For instance, if A is the realization of − in L p (), 1 < p < ∞, with

Dirichlet boundary condition:
1, p
D(A) = W 2, p () ∩ W0 (), Au = −u,
being an open bounded set in Rn with regular boundary, then A is a

positive operator with bounded imaginary powers, and γ A < π/2 by [82]
(note that the estimate of Theorem 4.15 is not enough, because it does
not imply that γ A < π/2). We get that for each f ∈ L p ((0, T ) × ) the
problem


 u t (t, x) = u(t, x) + f (t, x), 0 < t < T, x ∈ ,



u(t, x) = 0, 0 < t < T, x ∈ ∂,





u(0, x) = 0, x ∈ ,
has a unique solution u ∈ W p1,2 ((0, T ) × ), i.e. u, u t , Di u, Di j u ∈

L p ((0, T ) × ) for i, j = 1, . . . , n.
Problems with nonzero initial data,

 v (t) + Av(t) = f (t), 0 < t < T,
(4.25)

v(0) = v0 ,
are easily reduced to (4.20), since any solution may be split as the sum
v = u + w, where u is the solution of (4.20) and w is a solution to

 v (t) + Av(t) = 0, 0 < t < T,
(4.26)

v(0) = v0 ,
We shall see in Chapter 6 that problem (4.26) has a unique solution v.

From Corollary 1.14 it follows that v ∈ L p ((0,T );D(A))∩W 1, p ((0,T );X)
iff v0 ∈ D A (1 − 1/ p, p). Therefore, we can conclude that for every
f ∈ L p ((0, T ); X) and v0 ∈ D A (1−1/ p, p) problem (4.25) has a unique
solution v ∈ L p ((0, T ); D(A)) ∩ W 1, p ((0, T ); X).
4.2.3. Exercises
1. Improve the estimate of Proposition 4.14 showing that for each β <
arctan 1/M there is C such that
Ait ≤ Ce(π−β)|t| , t ∈ R.
Hint: instead of using formula (4.9), modify formula (4.3) for Ait−1 x
letting only r → 0 and leaving θ < arctan 1/M fixed.
2. Let A, B be two positive operators with bounded imaginary powers,
satisfying (4.18). Show that A z B w = B w A z for Re z, Re w ≤ 0.
Show that for Re z ≤ 0, A z maps D(B) into itself, and A z Bx =
B Az x for each x ∈ D(B). Show that for Re z ≤ 0, A z maps D(B)
into itself.
3. Let A, B be two positive operators with bounded imaginary powers,
satisfying (4.17) and (4.18). Show that for every x ∈ D(A) ∩ D(B)
the functions z → B z A−z x, −1 < Re z ≤ 0, and z → A−z B z x,
0 ≤ Re z < 1, are continuous (this is used in the proof of Proposition
4.18).
4. Let A be a positive operator with bounded imaginary powers in a
Banach space X. Show that for each a, b and for each p ∈ [1, ∞]
the operator in the space L p (a, b; X) defined by
A : D(A) := { f ∈ L p (a, b; X) : f (a) = 0} → L p (a, b; X),
(A) f (t) := A f (t)
has bounded imaginary powers, given by
(Ais f )(t) = Ais f (t), s ∈ R, a < t < b.
5. Let X be a UMD space, and let 1 < p < ∞. Let D(C) :=

W 1, p (R; X) and C : D(C) → L p (R; X), Cu = u + u. Show
that C is positive and has bounded imaginary powers.
4.3. M-accretive operators in Hilbert spaces

Throughout this section H is a complex Hilbert space, and A : D(A) ⊂
H → H is a linear operator satisfying
D(A) = H, ρ(A) ⊃ (−∞, 0), (λI + A)−1 ≤ 1/λ, λ > 0. (4.27)
Therefore, A is a nonnegative operator. Moreover it satisfies the resolvent
estimate with constant M = 1, and this is not a mere notational simpli-
fication but it is a crucial assumption.
Due to the Hille-Yosida Theorem, assumption (4.27) is equivalent to
the hypothesis that −A is the infinitesimal generator of a contraction
semigroup e −t A . Therefore, for each x ∈ D(A),
e−t A x − x
Re −Ax, x = lim Re , x ≤ 0.
t→0 t
Any operator B : D(B) ⊂ H → H satisfying
Re Bx, x ≥ 0, x ∈ D(B)
is called accretive. Therefore, A is accretive.

It is possible to show that A is m-accretive (maximal accretive), in the
sense that it has no proper accretive extension, and conversely, any closed
m-accretive operator satisfies (4.27). So, operators satisfying (4.27) are
often referred in the literature as m-accretive operators.
It is easy to see that A satisfies (4.27) iff A ∗ does. Moreover, if A
satisfies (4.27) and it is one to one, then the range of A is dense in H .
We shall describe the approach of Kato [59, 60] to study the imaginary
powers of A and the relationship between the domains of A α and (A∗ )α .
First we consider m-accretive bounded operators satisfying in addition
Re Ax, x ≥ δx2 , x ∈ H. (4.28)
for some δ > 0. The case of unbounded operators will be reduced to this
one, through the use of the Yosida approximations n A(n I + A)−1 .
Assumption (4.28) implies that the resolvent set of A contains (−∞,δ),
and (λI + A)−1 ≤ (λ + δ)−1 for every λ ≥ 0. Therefore A is a positive
bounded operator, so that for each z ∈ C the complex powers A z are
defined by
1
Az = λz R(λ, A)dλ, (4.29)
2πi γ
where γ is any regular curve surrounding σ (A) with index 1 with respect
to every point of σ (A), and avoiding (−∞, 0]. Moreover we have
(Aα )∗ = (A∗ )α , α ∈ C.
See Exercise 10, Section 4.1.1.

We will need the following lemma.
Lemma 4.25. Let A be a bounded m-accretive operator satisfying (4.28).
Then for real β ∈ [−1, 1], Aβ satisfies
Re Aβ x, x ≥ δ β x2 , 0 ≤ β ≤ 1, x ∈ H, (4.30)
Re Aβ x, x ≥ (δA−2 )−β x2 , −1 ≤ β ≤ 0, x ∈ H.(4.31)
Proof. For 0 < β < 1 we use the Balakrishnan formula (4.7), which
implies

β sin(πβ) ∞ β−1
A x, x = ξ A(ξ I + A)−1 x, xdξ.
π 0
Recalling that
ReA(ξ I + A)−1 x, x = Re(I − ξ(ξ I + A)−1 )x, x
≥ x2 − ξ |(ξ I + A)−1 x, x|

ξ
≥ x2 − x2
ξ +δ
δ
= x2 ,
ξ +δ
we obtain, through (4.5),

∞
β δ sin(πβ) ξ β−1
ReA x, x ≥ x2 dξ = δ β x2 ,
π 0 ξ +δ
i.e. (4.30) holds. Moreover,
ReA−1 x, x = ReA−1 x, A A−1 x ≥ δA−1 x2 ≥ δA−2 x2 ,
so that (4.31) holds for β = −1. But using again the representation
formula (4.29) we see easily that
Aβ = (A−1 )−β , β ∈ C,
so that (4.31) holds for every β ∈ [−1, 0].
Theorem 4.26. Let A : H → H be a bounded operator. Assume that

there exists δ > 0 such that (4.28) holds. Then for each α ∈ [0, 1/2) and
for each x ∈ H
(A∗ )α x ≤ cα Aα x, Aα x ≤ cα (A∗ )α x,
with cα = tan π(1 + 2α)/4. Moreover,
Ait ≤ eπ|t|/2 , t ∈ R.
Proof. Let us introduce the “real part” and the “imaginary part” of Aα
defined by
Aα + (A∗ )α Aα − (A∗ )α
Hα = , Kα = ,
2 2i
i.e.
Aα = Hα + i K α , (A∗ )α = Hα − i K α .
Then for every x ∈ H

Hα x2 − K α x2 = Re Aα x, (A∗ )α x = Re Aα+α x, x.
If −1/2 ≤ Re α ≤ 1/2, then Aα+α still satisfies (4.28) with δ replaced by
the constant δα = min{δ α , (δA−2 )α } by Lemma 4.25. Therefore,
Hα x2 ≥ K α x2 + δα x2 , x ∈ H. (4.32)
This implies that Hα is one to one. Since Hα∗ = (Aα + (A∗ )α )/2, and A∗
satisfies the assumptions of the theorem, also Hα∗ is one to one. Therefore
Hα is invertible, and since K α y ≤ Hα y for every y because of (4.32),
then K α Hα−1 x ≤ Hα Hα−1 x = x for each x, so that
K α Hα−1 ≤ 1, −1/2 ≤ Re α ≤ 1/2.
Now we improve this estimate applying the maximum principle to the
function
K α Hα−1
α → (α) = , −1/2 ≤ Re α ≤ 1/2.
tan(πα/2)
Such a function is holomorphic in the whole strip (even at α = 0, because
K 0 = 0) and bounded with values in L(H ); moreover | tan(πα/2)| = 1 if
Re α = ±1/2 so that (α) ≤ 1 on the boundary of the strip. Therefore
(α) ≤ 1 for each α in the strip, i.e.
K α Hα−1 ≤ tan(πα/2), −1/2 ≤ Re α ≤ 1/2. (4.33)
This is an important improvement of K α H α−1 ≤ 1, because
| tan(πα/2)| < 1 for |Re α| < 1/2, so that I ± i K α Hα−1 is invertible
with bounded inverse.
Since Aα = Hα + i K α , (A∗ )α = Hα − i K α , we get for |Re α| < 1/2
1 + | tan(πα/2)|
(A∗ )α A−α = (I − i K α Hα−1 )(I + i K α Hα−1 )−1 ≤ .
1 − | tan(πα/2)|
(4.34)
Therefore for real α ∈ [0, 1/2)
1 + tan(πα/2) α π(1 + 2α) α
(A∗ )α x ≤ A x = tan A x, x ∈ H,
1 − tan(πα/2) 4
and we get a similar inequality exchanging A with A∗ . So, the first state-
ment is proved.
Moreover, taking α = −it with real t, in (4.34) we get
1 + | tan(πit/2)|
A−it x2 = (A∗ )it A−it x, x ≤ x2 = eπ|t| x2 ,
1 − | tan(πit/2)|
and the last statement follows.
Theorem 4.26 is the starting point to show several properties of the

powers of general m-accretive operators. The first property concerns the
equivalence of the domains D(Aα ), D((A∗ )α ) for 0 ≤ α < 1/2. But we
need a preliminary lemma.
Lemma 4.27. Let A : D(A) ⊂ H → H satisfy (4.27). For every n ∈ N
set
A −1
Jn = I + = n(n I + A)−1 .
n
Then Jn is a bounded nonnegative operator, and for every α ∈ [0, 1]
Jnα ≤ 1,
lim J α x = x, x ∈ H.
n→∞ n
Proof. As easily seen, for every λ > 0 λI + Jn is invertible with (bounded)

inverse
(λI + Jn )−1 = (n I + A)(n(λ + 1) + λA)−1 .
For every ε > 0 the operator I + ε A is positive. The representation
formula (4.6) gives

−α sin(πα) ∞ −α
(I + A/n) = ξ (ξ I + I + A/n)−1 dξ,
π 0
where (ξ I + I + A/n)−1 ≤ 1/(ξ + 1), so that due to (4.5)

−α sin(πα) ∞ 1
(I + ε A) ≤ dξ = 1.
π 0 ξ (ξ + 1)
−α
It follows (see Exercise 9, Section 4.1.1)

(I + A/n)−α = ((I + A/n)−1 )−α = Jnα ≤ 1,
and by the dominated convergence theorem
lim J α x = lim (I + A/n)−α x
n→∞ n n→∞

sin(πα) ∞ ξ −α
= x dξ = x, x ∈ H.
π 0 ξ +1
Theorem 4.28. Let A : D(A) ⊂ H → H be any m-accretive operator.

Then for every α ∈ [0, 1/2), D(Aα ) = D((A∗ )α ) and for each x in the
common domain
π(1 + 2α) α π(1 + 2α)
(A∗ )α x ≤ tan A x, Aα x ≤ tan (A∗ )α x.
4 4
Proof. We know already (Theorem 4.26) that the statement is true if A is

bounded and satisfies (4.28) for some δ > 0. As a second step, we prove
that the statement is true if A satisfies (4.27) and 0 ∈ ρ(A). Then it will
follow easily that the statement is true in the general case.
Let 0 ∈ ρ(A). Let us consider the Yosida approximations,
An = A Jn = n A(n I + A)−1 , n ∈ N.
It is not hard to see that the operators An are bounded (with An ≤ n),
m-accretive (with (ξ I + An )−1 = n 2 /(n + ξ )2 (nξ/(n + ξ )I + A)−1 +
I /(n + ξ )), and satisfy (4.28) since
1 1
An x, x = A Jn x, (n I + A)Jn x = A Jn x, Jn x + An x2
n n
and An is invertible, with A−1

n = A
−1
+ I /n. Therefore by Theorem 4.26
we have 
 π(1 + 2α) α
 ∗ α
 (An ) ≤ tan An ,
 4
(4.35)

 π(1 + 2α)

 An ≤ tan
α ∗ α
(An ) ,
4
for every n ∈ N.
Note that since I + A/n is a positive operator, then (I + A/n)−α is well
defined for every α ≥ 0, and by Exercise 9, Section 4.1.1, (I + A/n)−α =
Jnα for 0 ≤ α ≤ 1. Therefore,
Jnα = (A−1 An )α = Aαn A−α = A−α Aαn ,
as it is easy to check. Therefore for every x ∈ D(Aα ), Jnα x ∈ D(Aα ),

and we have
Aαn x = Aα Jnα x = Jnα Aα x, 0 ≤ α ≤ 1.
Now we use Lemma 4.27, which states that Jnα ≤ 1, and limn→∞ Jnα x =
x for each x. We get
(i) Aαn x ≤ Aα x, (ii) lim Aαn x = Aα x, x ∈ D(Aα ). (4.36)

n→∞
Using (4.35) for 0 ≤ α < 1/2 and then (4.36)(i) we get
π(1 + 2α) α π(1 + 2α) α

(A∗n )α x ≤ tan An x ≤ tan A x, (4.37)
4 4
so that the sequence (A∗n )α x is bounded for each x ∈ D(Aα ). Moreover,

for every y ∈ D(Aα ) we have, due to (4.36)(ii),
(A∗n )α x, y = x, Aαn y → x, Aα y, n → ∞.
Since D(Aα ) is dense in H (because D(A) is dense and D(Aα ) ⊃ D(A)),
and (A∗n )α x is bounded by (4.37), then (A∗n )α x converges weakly to
some w ∈ H . Such a w satisfies w, y = x, Aα y for every y ∈ D(Aα ),
and this implies that x ∈ D((Aα )∗ ) = D((A∗ )α ) and w = (A∗ )α x.
Therefore, the domain of Aα is contained in the domain of (A∗ )α . Ex-
changing the roles of A and A∗ we get that D(Aα ) = D((A∗ )α ), and
limn→∞ (A∗n )α x = (A∗ )α x for each x in the common domain. Letting
n → ∞ in (4.37), and in the similar estimate with A∗ in the place of A,
we get the claimed estimates.
Now we consider the case where 0 ∈ σ (A). For each ε > 0 the
operator A + εI satisfies (4.27) and 0 belongs to its resolvent set, so
D((A + εI )α ) = D((A∗ + εI )α ) for 0 ≤ α < 1/2, and for every x in the
common domain we have
π(1 + 2α)
(A∗ + εI )α x ≤ tan (A + εI )α x,
4
π(1 + 2α)
(A + εI )α x ≤ tan (A∗ + εI )α x.
4
By Lemma 4.11, D((A + εI )α ) = D(Aα ) and (A + εI )α x → Aα x for
each x ∈ D(Aα ), and the same holds with A replaced by A∗ . Letting
ε → 0 the statement follows.
Let us consider now the imaginary powers Ait , t ∈ R.
Theorem 4.29. Assume that A : D(A) ⊂ H → H is m-accretive and
that 0 ∈ ρ(A). Then the imaginary powers Ait , t ∈ R, are bounded
operators, and
Ait ≤ eπ|t|/2 , t ∈ R.
Proof. Theorem 4.26 implies that the statement is true if A is bounded
and satisfies (4.28).
Next step is to show that the statement is true if A is bounded, m-
accretive and one to one. This is done considering the operators A + εI
with ε > 0, which satisfy (4.28) with δ = ε, and letting ε → 0. Indeed,
since (A + εI )x → x and (ξ I + A + εI )−1 x → (A + ξ I )−1 x for each
x ∈ H , and (A + εI )−1 x → A−1 x for each x ∈ R(A) as ε → 0, we may
let ε → 0 in formula (4.11) getting
Ait x = lim (A + εI )it x ≤ eπ|t|/2 x, t ∈ R.
ε→0
Since Ait is closed, we may conclude that D(Ait ) = H and Ait ≤

eπ|t|/2 for every t ∈ R.
The fact that (A + εI )−1 x → A−1 x for each x ∈ R(A) as ε → 0 may
be proved as follows: from the equality
A((A + εI )−1 y − A−1 y) + ε((A + εI )−1 y − A−1 y) = −εy,
which holds for each y ∈ H , we obtain
A((A + εI )−1 y − A−1 y)2 + ε2 (A + εI )−1 y − A−1 y ≤ ε2 y2 ,
so that 0 = limε→0 A((A + εI )−1 y − A−1 y) = limε→0 (A + εI )−1 Ay −
A−1 Ay, i.e. limε→0 (A + εI )−1 x = A−1 x for every x = Ay in the range
of A.
In the final step we consider a general m-accretive operator A with
0 ∈ ρ(A). Therefore A−1 is bounded, m-accretive and one to one. It is
not hard to see that
Ait = (A−1 )−it , t ∈ R.
But we already know that (A−1 )−it is bounded, and its norm does not
exceed eπ|t|/2 . The statement follows.
Theorems 4.29 and 4.17 yield the next corollary.
Corollary 4.30. If A : D(A) ⊂ H → H is m-accretive and 0 ∈ ρ(A)
then for 0 ≤ Re α < Re β we have
[D(Aα ), D(Aβ )]θ = D(A(1−θ)α+θβ ).
4.3.1. Self-adjoint operators in Hilbert spaces

Here H is again a complex Hilbert space, and A : D(A) ⊂ H → H is
a positive self-adjoint operator. The common definition of positivity for
self-adjoint operators looks weaker than ours (Definition 4.1): usually, a
self-adjoint operator is called positive if there is δ > 0 such that
Ax, x ≥ δx2 , x ∈ D(A). (4.38)
Next lemma shows that the definitions are equivalent.
Lemma 4.31. Let A be a self-adjoint operator. If A satisfies (4.38), then
ρ(A) contains (−∞, δ) and
1
R(λ, A) ≤ , λ < δ. (4.39)
δ−λ
Conversely, if ρ(A) contains (−∞, δ) and (4.39) holds, then A satisfies
(4.38).
Proof. Let A satisfy (4.38) and let λ < δ, x ∈ D(A). Then

(λI − A)x2 = (λ − δ)x + (δ I − A)x2
= (λ − δ)2 x2 + 2(λ − δ)x, δx − Ax + (δ I − A)x2 ,

so that
(λI − A)x2 ≥ (λ − δ)2 x2 , (4.40)
and therefore λI − A is one to one. We prove that it is also onto, showing
that its range is both closed and dense in H . Let xn ∈ D(A) be such that
λxn − Axn converges. From the inequality (4.40) we get
(λI − A)(xn − xm )2 ≥ (λ − δ)2 xn − xm 2 , n, m ∈ N,
so that xn is a Cauchy sequence, λx n is a Cauchy sequence, and Axn is
a Cauchy sequence. Then x n and Axn converge; let x, y be their limits.
Since A is self-adjoint, then it is closed (we recall that each adjoint oper-
ator is closed), so that x ∈ D(A), Ax = y, and λx n − Axn converges to
λx − Ax ∈ R(λI − A). Therefore, the range of λI − A is closed.
Let y be orthogonal to the range of (λI − A). Then for each x ∈ D(A)
we have y, λx − Ax = 0, so that y ∈ D(A∗ ) = D(A) and λy − A∗ y =
λy − Ay = 0. Since λI − A is one to one, it follows y = 0. Therefore,
the range of (λI − A) is dense.
Let us estimate R(λ, A). For x ∈ H , let u = R(λ, A)x. From the
equality x, u = λu − Au, u it follows x, u ≤ (λ − δ)u2 ≤ 0, so
that
(δ − λ)u2 ≤ |x, u| ≤ x u
which implies R(λ, A) ≤ (δ − λ)−1 .
Conversely, assume that ρ(A) contains (−∞, δ) and that (4.39) holds.
Then the resolvent set of the operator B : D(B) = D(A) → H , Bx :=
−Ax + δx contains (0, +∞) and R(λ, B) = −R(δ − λ, A) so that
R(λ, B) ≤ 1 for λ > 0. Therefore, B is the infinitesimal generator
of a contraction semigroup et B . For each x ∈ D(B) we have
1 d tB 2
Bx, x = lim+ et B Bx, et B x = lim+ e x ≤ 0,
t→0 t→0 2 dt
because et B ≤ 1 for t > 0 implies that t → et B x2 is a decreasing
function. Since B = −A + I , the inequality Bx, x ≤ 0 for each
x ∈ D(B) is equivalent to (4.38).
In this case it is possible to define the powers Az for z ∈ C through the
spectral decomposition of A.
Theorem 4.32. Let A : D(A) ⊂ H be a self-adjoint operator. There

exists a unique family of self-adjoint projections E λ ∈ L(H ), λ ∈ R,
with the following properties:
(i) E λ E µ = E µ E λ = E λ ≤ E µ , for λ < µ,
(ii) lim+ E λ+t x = E λ x, λ ∈ R, x ∈ H ,
t→0
(iii) lim E λ x = 0, limλ→+∞ E λ x = x, x ∈ H ,
λ→−∞
such that
+∞
D(A) = x ∈ H : λ dE λ x < ∞ ,
2 2
−∞
+∞
Ax, y = µ dE µ x, y, x ∈ D(A), y ∈ H.
−∞
The above integrals are meant as improper integrals of Stieltjes integrals.

Indeed, λ → E λ x2 = E λ x, x is nonnegative and nondecreasing, and
E λ x, y = E λ (x + y)/22 − E λ (x − y)/22 +i E λ (x + i y)/22
−i E λ (x − i y)/22 , so that λ → E λ x, y is of bounded variation.
The family {E λ : λ ∈ R} is called the spectral resolution or spectral
decomposition of A. See, e.g., [80, Chapter VIII, Section 120–121].
If f : R → C is continuous, the operator f (A) is defined by
+∞
D( f (A)) = x ∈ H : | f (λ)| dE λ x < ∞ ,
2 2
−∞
+∞
f (A)x, y = f (µ)dE µ x, y, x ∈ D( f (A)), y ∈ H.
−∞
It is possible to show (see [80, Chapter IX, Section 126–128]) that for
every λ ∈ ρ(A) we have
+∞
1
R(λ, A)x, y = dE µ x, y, y ∈ H,
−∞ λ − µ
and that this definition of f (A) agrees with the usual definition in the
case where f is a power with integer exponent, i.e.
+∞
D(Ak ) = x ∈ H : λ2k dE λ x2 < ∞ , k ∈ Z,
−∞
+∞
Ak x, y = µk dE µ x, y, k ∈ Z, x ∈ D(Ak ), y ∈ H.
−∞
The function E λ is constant on each interval contained in the resolvent

set of A. In the case of a positive operator the spectrum of A is contained
in (0, ∞), so that all the above integrals are in fact integrals over (0, ∞).
Moreover, we can prove that this definition agrees with the definition of
the powers A z for any z ∈ C.
Theorem 4.33. If A is self-adjoint and positive, for every z ∈ C we have
+∞
D(A ) = x ∈ H : A x =
z z 2
λ dE λ x < ∞ , (4.41)
2Rez 2
0
and +∞
Az x = λz d E λ x, x ∈ D(A z ). (4.42)
0
Proof. Let Re z < 0. Then

∞
1 1 1
A =z
λ R(λ, A)dλ =
z
λ z
d E µ dλ
2πi γr,θ 2πi γr,θ 0 λ−µ
∞ ∞
1 λz
= dλ d E µ = µz d E µ ,
0 2πi γr,θ λ−µ 0
and the statement holds for Re z < 0. If Re z ≥ 0 let n > Re z, n ∈ N.

By definition, x ∈ D(A z ) if and only if
+∞
A x=
z−n
λz−n d E λ x ∈ D(An ).
0
+∞
For each y ∈ H , y ∈ D(An ) iff 0 µ2n dE µ y2 < ∞. Therefore,
x ∈ D(A z ) iff
+∞ +∞ ∞
µ dE µ A x =
2n z−n 2
µ dE µ
2n
λz−n d E λ x2
0 0 0
+∞ µ
= µ d
2n
λz−n E λ x2
0 0
+∞
= µ2Re z dE µ x2 < ∞
0
and in this case

+∞ +∞
A x, y =
z
µ dµ
n
λz−n dλ E λ x, E µ y
0 0
+∞ µ +∞
= µn dµ λz−n dλ E λ x, y = µz dµ E µ x, y.
0 0 0
It follows from the definition that if f is bounded, then f (A) is a bounded

operator, and f (A)L(H ) ≤ f ∞ . If A satisfies (4.38), then σ (A) ⊂
[δ, +∞), so that defining f (λ) = λit for λ ≥ δ and extending f to
a continuous bounded function to the whole R, we obtain that Ait is a
bounded operator, and Ait ≤ 1, for every real t.
Example 4.34. Let be a bounded open set with C 2 boundary, and let
H = L 2 (), A : D(A) = H 2 () ∩ H01 (), Au = −u. It is well
known that σ (A) consists of a sequence of positive eigenvalues, each of
them with finite dimensional eigenspace, and there exists an orthonormal
basis of H consisting of eigenfunctions of A, (e.g., [12, Chapter IX]).
Denoting such a basis by {en }n∈N and denoting by λn the eigenvalue with
eigenfunction en , we have

Eλu = u, en en .
n: λn ≤λ
For every α with positive real part, the domain of Aα consists of those
u ∈ L 2 () such that
∞
λn2Re α |u n |2 < ∞,
n=1
where u n = u, en L 2 , and

∞
Aα u = λαn u n en , u ∈ D(Aα ).
n=1
√
Moreover, {en / λn : n ∈ N} is an orthonormal basis of H0 (), with
1
respect to the scalar product Du, Dv H 1 := u(x)v(x)dx (see, e.g.,

0
[12, Section IX.8]).
In the particular case = (0, π) it is easy to see that

en (x) = 2/π sin(nx), λn = n 2 .
∞ 4
Therefore, u ∈ D(A) ∞= H2 (0, π)∩ H0 (0, π) iff n=1 n |u n | < ∞,
2 1 2
and
in this
∞ 2 case Au = n=1 n u e
n n . Taking α = 1/2, we get u ∈
∞ D(A 1/2
) iff
n=1 n |u n | < ∞, that is iff u ∈ H0 (0, π), and A u = n=1 nu n en .
2 1 1/2
Let us come back to the general theory. If A satisfies (4.38), then (4.41)
– (4.42) can be taken as a definition of A z , and all the properties of A z
could be deduced, without invoking any result of the previous sections.
For instance, it follows immediately that for every x ∈ D(Aα ) the func-
tion z → A z x is holomorphic in the half-plane Re z < Re α, that D(A z )
depends only on Re z, and if Re z 1 < Re z 2 then D(A z1 ) ⊃ D(A z2 ).
Most important, we already remarked that for every t ∈ R the operator

Ait is bounded, and Ait ≤ 1. By Theorem 4.17, this implies that
[D(Aα ),D(Aβ )]θ = D(A(1−θ)α+θβ ) for 0 ≤ Re α < Re β. Moreover, these
spaces coincide also with the real interpolation spaces (D(Aα ),D(Aβ ))θ,2 ,
as the next theorem 4.36 shows. For its proof we need a lemma.
Lemma 4.35. Let L : D(L) ⊂ H be a self-adjoint positive operator.

Then i L generates a strongly continuous group of contractions eit L in H .
Proof. Let us prove that the resolvent set of i L contains R \ {0}.

Since L is self-adjoint, then L x, x ∈ R for each x ∈ D(L), so that Re
i L x, x = 0 for each x ∈ D(L) = D(i L). This implies that for every
λ > 0, λI − i L and −λI − i L are one to one, and (λI − i L)−1 x ≤
x/λ, for each x ∈ (λI − i L)(H ), (λI + i L)−1 x ≤ x/λ, for each
x ∈ (λI +i L)(H ). Indeed for λ > 0 and y ∈ D(L), setting λy −i L y = x
we have
λy2 ≤ Re λy2 − Re i L y, y = Re x, y ≤ x y
so that y ≤ x/λ. Replacing i L by −i L we get a similar estimate:

so, both λI − i L and −λI − i L are one to one.
To prove that λI − i L and −λI − i L are onto it is enough to check that
their ranges are dense in H . Let y be orthogonal to the range of λI − i L,
then λx − i L x, y = 0 for each x ∈ D(L), so that
y ∈ D(λI − (i L) ), λy − (i L) y = λy + i L y = 0,
and since λI + i L is one to one, y = 0. It follows that the range of

λI − i L is dense in H . Similarly one shows that the range of λI + i L is
dense in H . Consequently, R \ {0} ⊂ ρ(i L) and R(λ, i L) ≤ 1/λ, and
this implies that the statement holds.
Theorem 4.36. Let A be a self-adjoint operator satisfying (4.38), and let

α, β ∈ C with Re α ≥ 0, Re β ≥ 0. Then for every θ ∈ (0, 1)
[D(Aα ), D(Aβ )]θ = (D(Aα ), D(Aβ ))θ,2 = D(A(1−θ)α+θβ ).
Proof. We already know that [D(Aα ), D(Aβ )]θ = D(A(1−θ)α+θβ ). It is

sufficient to prove that for real β > 0 we have
(H, D(Aβ ))θ,2 = D(Aβθ ),
and the general statement will follow by interpolation and reiteration.

Since Aβ is in its turn a positive self-adjoint operator, by Lemma 4.35

i Aβ generates a strongly continuous group of operators
β
T (t) = eit A , t ∈ R.
Now we borrow a result from next chapter. By Proposition 5.7, the

space (H, D(Aβ ))θ,2 consists of the elements x such that t → ψ(t) :=
t −θ T (t)x − x ∈ L 2∗ (0, ∞). We have
∞
dt
ψ L 2 (0,∞) =
2
t −2θ T (t)x − x2
∗
0 t
∞ ∞
−2θ β dt
= t |eitλ − 1|2 dE λ x2
0 0 t
∞ ∞ β
|eitλ − 1|2 dt
= dE λ x2
0 0 t 2θ t
∞ ∞
|eiτ − 1|2
= dτ λ2θβ dE λ x2
0 τ 2θ+1 0
= CAβθ x2 ,
so that (H, D(Aβ ))θ,2 = D(Aβθ ), with equivalence of the norms.
An important corollary about interpolation in Hilbert spaces follows.
Corollary 4.37. Let H1 , H2 be Hilbert spaces, with H2 ⊂ H1 , H2 dense

in H1 . Then for every θ ∈ (0, 1),
[H1 , H2 ]θ = (H1 , H2 )θ,2 .
Proof. It is known (see e.g. [80, Chapter VIII, Section 124]) that there
exists a self-adjoint positive operator A in H1 such that D(A) = H2 . The
statement is now a consequence of Theorem 4.36.
Chapter 5
Interpolation and semigroups
We already mentioned semigroups here and there in the previous chapters.

Indeed, dealing with positive operators it is natural to shift to semigroup
theory because of the strong links between A and e −t A , when the lat-
ter exists. Moreover several interpolation theorems have easy and clean
proofs through semigroups.
For the general theory of semigroups in Banach space we refer to [41]
(see [42] for a simplified version). For analytic semigroups, to [71]. Here
we recall just the definitions and some properties which will be used in
the sequel.
A semigroup (of linear operators) in a Banach space X is a family of
linear operators in L(X) such that T (0) = I and T (t + s) = T (t)T (s)
for all t, s ≥ 0. A semigroup is called strongly continuous if T (·)x :
[0, ∞) → X is continuous for each x ∈ X. If T (t) is a strongly continu-
ous semigroup, its infinitesimal generator A is defined by

T (h)x − x T (h)x − x
D(A) = x ∈ X : ∃ lim+ , Ax = lim+ .
h→0 h h→0 h
The most important result about strongly continuous semigroups is the
Hille-Yosida Theorem that we already mentioned in the previous chapters,
and that characterizes the infinitesimal generators of strongly continuous
semigroups.
Theorem 5.1. Let A : D(A) → X be a linear operator, and let M > 0,
ω ∈ R. Then A is the infinitesimal generator of a semigroup T (t) such
that T (t)L(X) ≤ Meωt for each t > 0 if and only if
(i) D(A) is dense in X,
(ii) ρ(A) ⊃ (ω, +∞),
(iii) for each n ∈ N and λ > ω, (λI − A)−n L(X) ≤ M(λ − ω)−n .
An important class of semigroups are the analytic semigroups, generated
by sectorial operators. They are not necessarily strongly continuous. A
linear operator A : D(A) ⊂ X → X is called sectorial if there are

constants ω ∈ R, β ∈ (π/2, π), M > 0 such that

 (i) ρ(A) ⊃ Sβ,ω := {λ ∈ C : λ = ω, |arg(λ − ω)| < β},

(5.1)

 (ii) R(λ, A)L(X) ≤ M ∀λ ∈ Sβ,ω .
|λ − ω|
(5.1) allows us to define a semigroup et A in X through the Dunford in-
tegral
1
et A = etλ R(λ, A)dλ, t > 0,
2πi ω+γr,η
where r > 0, η ∈ (π/2, β), and γr,η is the curve {λ ∈ C : |argλ| =
η, |λ| ≥ r} ∪ {λ ∈ C : |argλ| ≤ η, |λ| = r}, oriented counterclockwise.
We also set e0A = I .
(A )
+r
r, +
Figure 5.1. The curve ω + γr,η .
It is possible to show that the function t → et A is analytic in (0, +∞)

with values in L(X) (in fact, with values in L(X, D(Am )) for every m),
so that et A is called the analytic semigroup generated by A and A is
called the generator of et A . One sees easily that for x ∈ X there ex-
ists limt→0 et A x if and only if x ∈ D(A), and in this case the limit is x.
Therefore et A is strongly continuous if and only if D(A) is dense in X.
It is worth to remark that if a semigroup T (t) is differentiable for t >
0 with values in L(X), with T (t) ≤ M exp(ωt) for some M ≥ 0,
ω ∈ R, and either T (t) is strongly continuous, or there exists t0 > 0
such that T (t0 ) is one to one, then there exists a unique sectorial operator
A : D(A) → X such that T (t) = et A . If T (t) is strongly continuous,
then A coincides with the infinitesimal generator.
The following basic examples will be considered in the sequel.
Example 5.2. Translation semigroups.

Let {e1 , . . . , en } be the canonical basis of Rn , and for i = 1, . . . , n set
(Ti (t) f )(x) := f (x + tei ), x ∈ Rn , (5.2)
for each function f : Rn → C. It is easy to see that each Ti is strongly
continuous in L p (Rn ), 1 ≤ p < ∞, in BUC(Rn ), and it is not strongly
continuous in L ∞ (Rn ), in Cb (Rn ), in Cbα (Rn ) for 0 < α < 1. It is not
analytic in any of these spaces.
The infinitesimal generator Ai in X = L p (Rn ) and in X = BUC(Rn )
is the realization of the partial derivative ∂/∂ xi in X, and precisely
D(Ai ) = { f ∈ L p (Rn ) : ∃Di f ∈ L p (Rn )}, Ai f = Di f
if X = L p (Rn ), 1 ≤ p < ∞, and
D(Ai ) = { f ∈ BUC(Rn ) : ∃Di f ∈ BUC(Rn )}, Ai f = Di f
if X = BUC(Rn ).
Example 5.3. The Gauss-Weierstrass semigroup.

1 |y|2
T (t) f (x) := e− 4t f (x − y)dy, t >0 (5.3)
(4πt)n/2 Rn
is well defined in the spaces L p (Rn ), 1 ≤ p ≤ +∞, in Cbk (Rn ), in

Cbk+α (Rn ), for k ∈ N and α ∈ (0, 1), and in many other functional spaces.
It is not hard to see that it is strongly continuous in L p (Rn ) for 1 ≤
p < +∞, and in BUC(Rn ), and that it is not strongly continuous in
L ∞ (Rn ), in Cb (Rn ) and in Cbα (Rn ) for 0 < α < 1. T (t) is called also
heat semigroup because for a large class of initial data f , the function
u(t, x) = (T (t) f )(x) is a solution to the heat equation

 u t (t, x) = u(t, x), t > 0, x ∈ Rn ,

u(0, x) = f (x), x ∈ Rn .
As easily seen, for every multi-index β = (β1 , . . . , βn ) ∈ (N ∪ {0})n we
have
C
D β T (t) f ∞ ≤ |β|/2 f ∞ , t > 0, f ∈ L ∞ (Rn ). (5.4)
t
These estimates follow directly from the representation formula (5.3),
and for |β| = 2 they imply that T (t) is analytic in L ∞ (Rn ) and in its
subspaces Cb (Rn ), BUC(Rn ), because

d
T (t) f = T (t) f ∞ ≤ C f ∞ , t > 0, f ∈ L ∞ (Rn ).
dt t
∞
Moreover, since the derivatives Di commute with T (t) on Cb1 (Rn ), for
every k ∈ N we have
C
D β T (t) f ∞ ≤ f C h (Rn ) , t > 0, f ∈ Cbh (Rn ), (5.5)
t (|β|−h)/2 b
n
if |β| = i=1 βi ≥ h. So (taking |β| = h + 2), T (t) is analytic also in
Cbh (Rn ).
Similarly,
C
D β T (t) f L p (Rn ) ≤ f W h, p (Rn ) , t > 0, f ∈ W h, p (Rn ), (5.6)
t (|β|−h)/2
for |β| ≥ h. Taking as before |β| = h + 2, we obtain that T (t) is analytic
in L p (Rn ) and in W h, p (Rn ) for each p ≥ 1 and h ∈ N.
Example 5.4. The Ornstein-Uhlenbeck semigroup is defined by
−1/2 2
1 |K t y|
T (t) f (x) = e− 2t f (et B x − y)dy, t > 0,
(2πt) (det K t )1/2
n/2
Rn
(5.7)
where Q is a symmetric and positive definite matrix, B = 0 is a n × n
matrix, and
1 t s B s B∗
Kt = e Qe ds, (5.8)
t 0
∞ n n
with es B = n=0 s B /n!. This semigroup is well defined in all the
above functional spaces, it is strongly continuous in L p (Rn ) for 1 ≤ p <
+∞, if B = 0 it is not strongly continuous in L ∞ (Rn ), in Cb (Rn ), in
BUC(Rn ), and in Cbα (Rn ).
For a large class of initial data f , the function u(t, x) = (T (t) f )(x) is
the solution to problem

u t (t, x) = Au(t, x), t > 0, x ∈ Rn ,
u(0, x) = f (x), x ∈ Rn .
where A is the Ornstein-Uhlenbeck operator defined by
1 n n
(A)u(x) = qi j Di j u(x) + bi j xi D j u(x)
2 i, j=1 i, j=1 (5.9)
1
= Tr(Q D 2 u)(x) + Bx, Du(x).
2
Even if the derivatives do not commute with T (t), we have a simple com-
mutation formula,
∗
D(T (t) f ) = et B T (t)(D f ), t > 0, f ∈ Cb1 (Rn ),
that allows to prove estimates similar to (5.5),
Ceε(|β|−h)t
D β T (t) f ∞ ≤ f C h (Rn ) , t > 0, f ∈ Cbh (Rn ), (5.10)
t (|β|−h)/2 b
for any multi-index β with |β| ≥ h. Here ε depends on the matrix B, we

may take ε = 0 if all the eigenvalues of B have negative real part.
However, (5.10) does not imply that T (t) is analytic in L ∞ (Rn ) or in
Cb (Rn ), because some coefficients of A are unbounded, and from estim-
ate T (t) f Cb2 (Rn ) ≤ C f ∞ /t we cannot deduce that AT (t) f ∞ ≤
C f ∞ /t. In fact, it is possible to show that for each p ∈ [1, ∞], T (t)
is not analytic in L p (Rn ).
In the sequel, we shall use two basic properties of strongly continu-
ous and of analytic semigroups. First, the representation formula for the
resolvent of the generator A of T (t):
∞
R(λ, A) f = e−λt T (t) f dt, f ∈ X, Re λ > ω, (5.11)
0
provided T (t) ≤ Meωt for some M, ω, and for all t > 0; and second,
the variation of constants formula
t t
u(t) = T (t − s) f (s)ds = T (s) f (t − s)ds, (5.12)
0 0
that represents the (mild) solution to the forward Cauchy problem

u (t) = Au(t) + f (t), 0 < t < T,
(5.13)
u(0) = 0.
Precisely, we have the following results.

Proposition 5.5. Let T (t) be a strongly continuous or analytic semig-
roup such that T (t) ≤ Meωt for every t ≥ 0, and let A be its generator.
Then the resolvent set ρ(A) of A contains the half-plane := {λ ∈ C :
Re λ > ω}, and (5.11) holds for λ ∈ .
However, formula (5.11) is still meaningful in several cases in which
T (t) is not strongly continuous or analytic, and it may be used to define
a generalized notion of infinitesimal generator, as follows.
Example 5.6. Let T (t) be a semigroup in the space X = Cb () or X =

Cb (), where is an open set in Rn , such that T (t) ≤ Meωt and such
that for each f ∈ X the function (t, x) → (T (t) f )(x) is continuous in
[0, +∞) × (respectively, in [0, +∞) × ). For Re λ > ω the operator
R(λ) given by
+∞
(R(λ) f )(x) = e−λt (T (t) f )(x)dt
0
belongs to L(X). Due to the semigroup property, it satisfies the resolvent

identity
R(λ) − R(µ) = (µ − λ)R(λ)R(µ).
Moreover, R(λ) is one to one for all λ > ω. Indeed, if R(λ0 ) f is the null
function for some λ0 , then it vanishes for all λ because of the resolvent
identity. Then for each x ∈ the Laplace transform of the continuous
function t → (T (t) f )(x) is zero, so that (T (t) f )(x) = 0 for each t ≥ 0,
in particular taking t = 0 we get f (x) = 0.
The general spectral theory implies that there exists a unique closed
operator A whose resolvent set contains the half-plane {λ ∈ C : Re λ >
ω} and R(λ) = R(λ, A), for Re λ > ω. Such operator A may still be
called generator of T (t).
Under the additional assumption that T (t) preserves pointwise con-
vergence for bounded sequences ( f n ) in X (i.e., if ( f n ) is a bounded
sequence that converges pointwise to f ∈ X and t > 0, then T (t) f n
converges pointwise to T (t) f ), it is possible to see that for each f ∈
D(A), setting u(t, x) := (T (t) f )(x) we have u t (t, x) = Au(t, x) =
(T (t)A f )(x) for t ≥ 0, x ∈ (resp., x ∈ ). See the paper [79].
The above mentioned semigroups T (t) (translation semigroups, Gauss-
Weierstrass semigroup, Ornstein-Uhlenbeck semigroup) satisfy the con-
tractivity estimate T (t)L(Cb (Rn )) ≤ 1, they preserve pointwise conver-
gence of bounded sequences, and for each continuous and bounded f the
function (t, x) → (T (t) f )(x) is continuous in [0, +∞) × Rn .
In all these cases (strongly continuous semigroups, analytic semi-
groups, the semigroups of Example 5.6), since AR(λ, A) = λR(λ, A) −
I , then
∞
AR(λ, A) f = λe−λt (T (t) − I ) f dt, f ∈ X, Re λ > ω, (5.14)
0
Concerning formula (5.12), it is not hard to see that if T (t) is a strongly

continuous or analytic semigroup, f ∈ C([0, T ]; X), and (5.13) has a
solution u ∈ C 1 ([0, T ]; X) ∩ C([0, T ]; D(A)), then u is given by for-
mula (5.12).
Whenever (5.12) makes sense, the function u defined there is called mild
solution to (5.13).
5.1. Real interpolation between Banach spaces and domains

of generators
Proposition 5.7. Let A generate a bounded semigroup T (t). Then
(X, D(A))θ, p = {x ∈ X : t → ψ(t) := t −θ T (t)x − x ∈ L ∗p (0, ∞)}
and the norms xθ, p and
x∗∗
θ, p := x + ψ L ∗ (0,∞)
p
are equivalent.
Proof. Recall that for every b ∈ D(A) we have
t t
T (t)b − b = AT (s)b ds = T (s)Ab ds, t > 0.
0 0
Let x ∈ (X, D(A))θ, p . Then if x = a + b with a ∈ X, b ∈ D(A), for

every t > 0 we have
t −θ T (t)x − x ≤ t −θ (T (t)a − a + T (t)b − b)
≤ t −θ ((M + 1)a + t MAb) ≤ t −θ (M + 1)(a + tb D(A) ).

Taking the infimum over all decompositions x = a + b we get
t −θ T (t)x − x ≤ (M + 1)t −θ K (t, x).

p
Therefore ψ(t) = t −θ T (t)x − x ∈ L ∗ (0, ∞) and
x∗∗
θ, p ≤ (M + 1)xθ, p .
p
Conversely, if ψ ∈ L ∗ (0, ∞) let us use (5.14) to get
∞
θ T (t)x − x dt
λ AR(λ, A)x ≤ λθ+1 t θ+1 e−λt ,
0 tθ t
that is, ϕ is the multiplicative convolution between the functions f (t) =
t θ+1 e−t and ψ(t) = t −θ T (t)x − x. Since f ∈ L 1∗ (0, ∞) and ψ ∈
p p
L ∗ (0, ∞), then ϕ ∈ L ∗ (0, ∞) and ϕ L ∗p (0,∞) ≤ f L 1∗ (0,∞) ψ L ∗p (0,∞) ,
so that
x∗θ, p ≤ (θ + 1)x∗∗ θ, p ,

We already know from Proposition 3.5 that (X, D(A2 ))θ, p = D A (2θ, p)
for θ = 1/2. Hence, by Proposition 5.7, for each x ∈ (X, D(A2 ))θ, p with
p
θ < 1/2 the function t → t −2θ T (t)x − x belongs to L ∗ (0, ∞), and
−2θ
consequently (since T (t) − I L(X) is bounded) t → t (T (t) − I )2 x
p
belongs to L ∗ (0, ∞).
This condition is in fact a characterization of the spaces (X, D(A2 ))θ, p
for every θ ∈ (0, 1), as the next proposition shows.
Proposition 5.8. Under the assumptions of Proposition 5.7, for every

θ ∈ (0, 1) and p ∈ [1, ∞] we have
:= t −2θ (T (t) − I )2 x ∈ L ∗p (0, ∞)}

(X, D(A2 ))θ, p = {x ∈ X : t → ψ(t)
and the norm

L p (0,∞)
x → xθ, p := x + ψ ∗
is equivalent to the norm of (X, D(A2 ))θ, p .
Proof. Recall that for every b ∈ D(A2 ) we have

t
(T (t) − I ) b = (T (t) − I )
2
T (σ )Ab dσ
0
t t
= T (s + σ )A2 b ds dσ, t > 0,
0 0
so that
(T (t) − I )2 b ≤ t 2 MA2 b.
Let x ∈ (X, D(A2 ))θ, p . Then if x = a + b with a ∈ X, b ∈ D(A2 ), for

every t > 0 we have
t −2θ (T (t) − I )2 x ≤ t −2θ ((T (t) − I )2 a + (T (t) − I )2 b)
≤ t −2θ ((M + 1)2 a + t 2 M 2 A2 b)
so that
t −2θ (T (t) − I )2 x ≤ (M + 1)2 t −2θ K (t 2 , x, X, D(A2 )).
= t −2θ (T (t) − I )2 x ∈ L ∗p (0, ∞) and

Therefore ψ(t)
xθ, p ≤ 2−1/ p (M + 1)2 x(X,D(A2 ))θ, p .


Conversely, let x be such that ψ(t)
p
∈ L ∗ (0, ∞). Then from (5.14) it
follows that
∞ ∞
(AR(λ,A)) x = λ
2 2
e−λ(t+s) (T (t +s) −T (t) −T (s) + I )x ds dt
0 0
∞ 2u
−2λu
= 2λ 2
e (T (2u) −T (t) −T (2u −t) + I )x dt du
0 0
∞ 2u
−2λu
= 2λ 2
e (T (2u) − 2T (u) + I )x dt dt du
0 0
∞ 2u
−2λu
+ 2λ 2
e du (2T (u) −T (t) −T (2u −t))x dt.
0 0
The first integral is nothing but

∞
4λ2 ue−2λu (T (u) − I )2 x du.
0
To rewrite the second one we note that

2u 2u
(2T (u) − T (t) − T (2u − t))x dt = (2T (u) − 2T (t))x dt
0 0
u 2u
= + (2T (u) − 2T (t))x dt
0 u
u 2u
= 2(T (u − t) − I )T (t)x dt + 2T (u)(I − T (t − u))x dt
0 u
u u
= 2(T (s) − I )T (u − s)x ds + 2T (u)(I − T (s))x ds
0 0
u
=2 (T (s) − I )(T (u − s) − T (u))x ds
0
u
= −2 (T (s) − I )2 T (u − s)x ds.
0
Therefore,

λ2θ (AR(λ, A))2 x ≤ 4|( f ψ)(λ)| 1 )(λ)|,
+ 2|( f ψ
where stands for the multiplicative convolution and

τ
2+2θ −2τ M
f (τ ) = τ e , ψ1 (τ ) = 1+2θ (T (s) − I )2 xds.
τ 0
Let us remark now that the Hardy-Young inequality (A.10)(i) implies that
−α p
that t → t z(t) ∈ L ∗ (0, ∞) the same is true for
if a function z is such
−1 t
its mean v(t) = t 0 z(s)ds, with
1
t → t −α v(t) L ∗p (0,∞) ≤ t → t −α z(t) L ∗p (0,∞) ,
(α + 1)
and this is easily seen to hold also for p = ∞. Therefore, ψ 1 ∈ L ∗p (0, ∞)
and ψ L p (0,∞) . It follows that the function ϕ
1 L p (0,∞) ≤ (2θ + 1)−1 ψ
∗ ∗
p
defined by ϕ(λ) := λ2θ (AR(λ, A))2 x belongs to L ∗ (0, ∞), and
L p (0,∞) + ψ
ϕ L ∗p (0,∞) = x∗θ, p ≤ 4 f L 1∗ (0,∞) (ψ 1 L p (0,∞) ),
∗ ∗
so that x ∈ (X, D(A2 ))θ, p by Proposition 3.6, and x(X,D(A2 ))θ, p ≤

L p (0,∞) ).
C(x + ψ ∗
Remark 5.9. In the proof of Proposition 5.7 we have not used the fact
that T (t) is strongly continuous or analytic. The proof works also for the
semigroups in the space C b () described in Example 5.6.
Propositions 5.7 and 5.8 may be generalized as follows. A proof is in
[90, Section 1.13.2].
Proposition 5.10. Under the assumptions of Proposition 5.7, for every
θ ∈ (0, 1), p ∈ [1, ∞], and k ∈ N we have
(X, D(Ak ))θ, p = {x ∈ X : t → t −kθ (T (t) − I )k x ∈ L ∗p (0, ∞)}
and the norms · θ, p and
x + t → t −kθ (T (t) − I )k x L ∗p (0,∞)
are equivalent.
Remark 5.11. If θk = m is integer, the domain of Am does not co-
incide in general with any interpolation space (X, D(Ak ))θ, p ; we have
only the embeddings (X, D(Ak ))θ,1 ⊂ D(Am ) ⊂ (X, D(Ak ))θ,∞ (Pro-
position 3.7). The only important case in which D(Am ) coincides with
an interpolation space is given by Theorem 4.17: if −A+λI has bounded
imaginary powers for some λ ∈ R then D((−A +λI )m ) = [X, D((−A +
λI )k )]θ and hence D(Am ) = [X, D(A)k ]θ . We recall that this is true if X
is a Hilbert space and −A is m-accretive (i.e., A is m-dissipative), since
−A + λI satisfies the assumptions of Theorem 4.29 for every λ > 0. In
this case we have also D(Am ) = (X, D(Ak ))θ,2 by Corollary 4.37.
A useful embedding result in applications to operator theory and partial

differential equations is the following.
Theorem 5.12. Let A : D(A) ⊂ X be the generator of a bounded semi-
group T (t). Assume moreover that there exists a Banach space E ⊂ X
and m ∈ N, 0 < β < 1, C > 0 such that
C
T (t)L(X,E) ≤ , t > 0,
t mβ
and that t → T (t)x is measurable with values in E, for each x ∈ X.
Then D(Am ) ⊂ E and E ∈ Jβ (X, D(Am )), so that
(X, D(Am ))θβ, p ⊂ (X, E)θ, p , θ ∈ (0, 1), p ∈ [1, ∞].
Proof. Let x ∈ D(Am ), λ > 0 and set (λI − A)m x = y. Then x =

(R(λ, A))m y so that
∞
(−1)m−1 d m−1 1
x= R(λ, A)y = e−λs s m−1 T (s)y ds.
(m − 1)! dλm−1 (m − 1)! 0
The function s → e−λs s m−1 T (s)y is in L 1 (0, ∞; E), so that x ∈ E and
∞
C
x E ≤ e−λs s m(1−β)−1 dsy
(m − 1)! 0
C(m(1 − β)) mβ−m
= λ y
(m − 1)!
m
C(m(1 − β)) mβ−m
m m−r r r

= λ λ (−1) A x
(m − 1)! r=0
r

m
≤ C λmβ−r Ar x.
r=0
Let us recall (Proposition 3.7) that D(Ar ) belongs to Jr/m (X, D(Am )) so
r/m 1−r/m
that there is C such that x D(Ar ) ≤ Cx D(Am ) x X . Using such

inequalities and then ab ≤ C(a + b ) with p = m/r, p = m/(m − r)
p p
we get
x E ≤ Cλmβ (λ−m x D(Am ) + x), λ > 0,
so that taking the minimum for λ > 0
β
x E ≤ Cx1−β x D(Am )
and the statement holds.

Corollary 5.13. Let X be a Hilbert space, let A : D(A) → X be m-

dissipative, and let the assumptions of Theorem 5.12 hold with mβ > 1.
Then
D(A) ⊂ (X, E)1/mβ,2 .
Proof. Applying Theorem 5.12 with θ = 1/mβ gives
(X, D(Am ))1/m,2 ⊂ (X, E)1/mβ,2 .
On the other hand, Remark 5.11 yields
(X, D(Am ))1/m,2 = D(A)
5.1.1. Exercises
In next exercises, A is the generator of a (not necessarily bounded) semig-
roup T (t) in a Banach space X.
1. Prove that for each θ ∈ (0, 1), p ∈ [1, ∞], and a > 0 we have
(X, D(A))θ, p = {x ∈ X : t → t −θ T (t)x − x ∈ L ∗p (0, a)}
and that the norms

a 1/ p
dt
x → x + t −θ p T (t)x − x p
0 t
are equivalent to the norm of (X, D(A))θ, p .

2. Prove that for each θ ∈ (0, 1), p ∈ [1, ∞], and a > 0 we have
(X, D(A2 ))θ, p = {x ∈ X : t → t −2θ (T (t) − I )2 x ∈ L ∗p (0, a)}
and that the norms

a 1/ p
dt
x → x + t −2θ p (T (t) − I )2 x p
0 t
are equivalent to the norm of (X, D(A))θ, p .

3. Prove that the conclusions of Theorem 5.12 and of Corollary 5.13
still hold if the assumption T (t)L(X,E) ≤ Ct −mβ is replaced by
T (t)L(X,E) ≤ Ct −mβ eωt for some ω > 0.
4. Let θ ∈ (0, 1). Prove that x ∈ (X, D(A))θ iff limt→0 t −θ T (t)x −
x = 0.
5. Let 0 < α < 1. Prove that t → T (t)x ∈ C 1+α ([0, 1]; X) if and only
if t → T (t)x ∈ C α ([0, 1]; D(A)), if and only if x ∈ D A (α + 1, ∞).
6. Let X be any Banach space, let T > 0 and set X := L p (0, T ; X),
1 ≤ p < ∞, or X := {u ∈ C([0, T ]; X) : u(0) = 0}. Show that the
operator B defined by D(B) := {u ∈ X : ∃u ∈ X }, Bu = −u , is
the infinitesimal generator of the strongly continuous semigroup et B
defined by

 u(s − t), t ≤ s ≤ T,
tB
(e u)(s) :=

0, 0 ≤ s ≤ t,
and show that if X := L p (0, T ; X) then DB (θ, p) = W θ, p (0, T ; X),

if 0 < θ < 1/ p, DB (θ, p) = {u ∈ W θ, p (0, T ; X); u(0) = 0}, if
1/ p < θ < 1, while if X := {u ∈ C([0, T ]; X) : u(0) = 0} then
DB (θ, ∞) = {u ∈ C θ ([0, T ]; X); u(0) = 0}.
5.2. Examples and applications

Example 5.14. Let us apply Propositions 5.7, 5.8 to the case X = L p (R),
1 ≤ p < ∞, A : D(A) = W 1, p (R) → L p (R), A f = f . Then
T (t) is the translations semigroup, T (t) f (x) = f (x + t). Applying
Proposition 5.7 we get for 0 < θ < 1
(L p (R),W 1, p (R))θ, p = D A (θ, p)

= { f ∈ L p (R) : t →t −θ f (· +t)− f Lp (R) ∈ L p∗ (0,∞)}
= W θ, p (R),
which we knew already (Example 1.8), but this is an alternative proof.

Applying Proposition 3.5 and recalling that D(A2 ) = W 2, p (R) we get
for θ = 1/2
(L p (R), W 2, p (R))θ, p = D A (2θ, p)
so that for θ < 1/2
(L p (R), W 2, p (R))θ, p = W 2θ, p (R),
and for θ > 1/2,
(L p (R),W 2, p (R))θ, p = { f ∈ W 1, p (R) : f ∈ D A (2θ − 1, p)} = W 2θ, p (R).

For θ = 1/2 we use Proposition 5.8: we get
(L p (R), W 2, p (R))1/2, p

p
= f ∈ L p (R) : t → t −1 f (· + 2t) − 2 f (· + t)+ f L p (R) ∈ L ∗ (0, ∞)
= B 1p, p (R),
which coincides with W 1, p (R) only for p = 2.

Choosing X = Cb (R), A : D(A) = Cb1 (R) → Cb (R), A f = f , we
argue similarly (recalling Remark 5.9), and we get
(Cb (R), Cb2 (R))θ,∞ = Cb2θ (R), θ = 1/2,
(Cb (R), Cb2 (R))1/2,∞

| f (x + 2t) − 2 f (x + t) + f (x)|
= f ∈ Cb (R) : supt=0, x∈R <∞ .
t
= B∞,∞
1
(R).
Next example is an important extension of Example 1.8 and of its corol-
laries. We use the notation W 0, p (Rn ) := L p (Rn ).
Example 5.15. For 0 ≤ θ1 < θ2 and for 0 < θ < 1 such that (1 − θ)θ1 +
θθ2 is not integer we have
(Cbθ1 (Rn ), Cbθ2 (Rn ))θ,∞ = Cb(1−θ)θ1 +θθ2 (Rn ), (5.15)
and
(W θ1 , p (Rn ), W θ2 , p (Rn ))θ, p = W (1−θ)θ1 +θθ2 , p (Rn ). (5.16)
More generally, for q ∈ [1, +∞),
(W θ1 , p (Rn ), W θ2 , p (Rn ))θ,q = B (1−θ)θ

p,q
1 +θθ2
(Rn ). (5.17)
In addition, if θ1 and θ2 are not integer, for every q, q1 , q2 ∈ [1, +∞) we

have
(B θp,q
1
1
(Rn ), B θp,q
2
2
(Rn ))θ,q = B (1−θ)θ
p,q
1 +θθ2
(Rn ) (5.18)
(the Besov spaces B sp,q (Rn ) are defined after Example 1.8).
Proof. Let Ai , i = 1, . . . , n, be the realization of the partial derivative Di

in X = Cb (Rn ), or in X = BUC(Rn ), or in X = L p (Rn ), 1 ≤ p < ∞,
with its natural domain: if X = Cb (Rn ), D(Ai ) = { f ∈ Cb (Rn ) :
∃Di f ∈ Cb (Rn )}, if X = BUC(Rn ), D(Ai ) = { f ∈ BUC(Rn ) :
∃Di f ∈ BUC(Rn )}, if X = L p (Rn ), D(Ai ) = { f ∈ L p (Rn ) : ∃Di f ∈
L p (Rn )} (in the latter case we mean the weak partial derivative).
Each Ai satisfies (3.1), with
+∞
(R(λ, Ai ) f )(x) = eλ(xi −s) f (x1 , . . . , xi−1 , s, xi+1 , . . . , xn )ds,
xi
for λ > 0, f ∈ X, x ∈ Rn , so that M = 1 for every i, and R(λ,Ai )R(λ,A j ) =

R(λ, A j )R(λ, Ai ) for every i, j.
Each Ai generates the translation semigroup
(Ti (t) f )(x) = f (x + tei ), t ≥ 0, x ∈ Rn ,
in the usual sense if X = BUC(Rn ) or X = L p (Rn ), in the generalized

sense of Example 5.6 if X = Cb (Rn ).
We apply Proposition 3.11, in its generalized version with any number
of operators, that gives for every σ ∈ (0, 1), q ∈ [1, ∞],

n
(X, K 1 )σ,q = D Ai (σ, q). (5.19)
i=1
Let us take X = Cb (Rn ) or X = BUC(Rn ), q = +∞. We have

K 1 = Cb1 (Rn ), K 1 = BUC1 (Rn ) respectively, and Proposition 5.7 yields
D Ai (σ, ∞) = { f ∈ X : s → f (x1 , . . . , xi−1 , s, xi+1 , . . . , xn ) ∈ Cbσ (R)}.
So, from (5.19) we obtain
(Cb (Rn ), Cb1 (Rn ))σ,∞ = (BUC(Rn ), BUC1 (Rn ))σ,∞ = Cbσ (Rn ).
This is an alternative proof of (1.16), which coincides with (5.15) for

θ1 = 0, θ2 = 1.
Now let us take X = L p (Rn ), q ∈ [1, +∞). We have K 1 = W 1, p (Rn ),
and Proposition 5.7 yields
D Ai (σ, q)
+∞
= f ∈ L p (Rn ) : t σ q−1 | f (x + tei ) − f (x)| p dx dt < +∞ ,
0 Rn
which is easily seen to be equivalent to

D Ai (σ, q)

σ q−1
(5.20)
= f ∈ L (R ) : |t|
p n
| f (x + tei )− f (x)| dx dt < +∞ ,
p
R Rn
and gives, through (5.19), a characterization of (L p (Rn ), W 1, p (Rn ))σ,q .

On the other hand, we already know from Example 1.8 and Exercise 6,
Section 1.1.2, that
(L p (Rn ), W 1, p (Rn ))σ,q = B σp,q (Rn )
(we recall that the right hand side is equal to W σ, p (Rn ) if q = p). So, we
have obtained a characterization of Besov spaces,

σ σ q−1
B p,q (R ) = f ∈ L (R ) :
n p n
|t| | f (x + tei ) − f (x)| p dx dt
R R n
< +∞ ∀i = 1, . . . , n
for 0 < σ < 1, p, q ∈ [1, +∞).

Now we apply Theorem 3.10 for those θ such that θm is not integer,
θm = k + σ , k = [θm], 0 < σ < 1. It yields
(X, K m )θ,q = { f ∈ K k : D α f ∈ (X, K 1 )σ,q , |α| = k}. (5.21)
If X = Cb (Rn ) we have K m = Cbm (Rn ), if X = BUC(Rn ) we have
K m = BUCm (Rn ). In both cases, from (5.21) and (1.16) we get
(X, K m )θ,∞ = { f ∈ K k : D α f ∈ Cbσ (Rn ), |α| = k} = Cbθm (Rn ),
which is (5.15) for θ1 = 0, θ2 ∈ N.
Using m − 1 times Proposition 3.16 (the second statement) and the
Reiteration Theorem, we get for m > k ∈ N such that θ(m − k) ∈
/ N,
(Cbk (Rn ), Cbm (Rn ))θ,∞ = (BUCk (Rn ), BUCm (Rn ))θ,∞ = Cbk+θ(m−k) (Rn ),
so that (5.15) holds for θ1 , θ2 ∈ N. Again by reiteration we obtain (5.15)
in the other cases.
Choosing now X = L p (Rn ) we obtain K m = W m, p (Rn ). Equalities
(5.21) and (5.20) give, still for θm = k + σ , k = [θm], 0 < σ < 1,

(L p (Rn ), W m, p (Rn ))θ,q = f ∈ W k, p (Rn ) : D α f ∈ B σp,q (Rn ), |α| = k
= B θm
p,q (R ),
n
θm θm, p
p (R ) = W (Rn))
n
which yields (5.17) (and also (5.16), recalling that Bp,
with θ1 = 0, θ2 = m.
As in the Hölder case, using m − 1 times Proposition 3.16 and the Rei-
teration Theorem, we get for m, k ∈ N, m > k, such that θ(m − k) ∈ / N,
(W k, p (Rn ), W m, p (Rn ))θ,q = B k+θ(m−k)

p,q (Rn ),
so that (5.16) and (5.17) hold for θ1 , θ2 ∈ N; by reiteration we obtain

(5.17) and (5.16) in the other cases.
Having the above characterizations as starting points, we may char-
acterize several real interpolation spaces between L p or Cb spaces and
domains of elliptic operators.
We begin with the simplest example, which is the Laplacian in Rn . We
recall that if X = L p (Rn ), 1 ≤ p ≤ ∞, and if X = Cb (Rn ), X =
BUC(Rn ), then the domain of the generator A of the Gauss-Weierstrass
semigroup (5.3) is D(A) = { f ∈ X : f ∈ X}, and A f = f , where
is in the sense of distributions. Moreover, if X = L p (Rn ) we have
D(A) = W 2, p (Rn ),
for 1 < p < ∞, while D(A) contains properly W 2, p (Rn ) for p = 1 and
p = ∞. For p = ∞ we may be a bit more precise: if X = L ∞ (Rn ), or
X = Cb (Rn ), or X = BUC(Rn ) we have
2, p
D(A) = { f ∈ ∩ p≥1 Wloc (Rn ) ∩ X : f ∈ X}, (5.22)
p
and also in this case f is a L loc function, not just a distribution. See [88].
Example 5.16. Let A be the generator of the Gauss-Weierstrass semig-
roup (5.3) in X = BUC(Rn ), in X = Cb (Rn ) or in X = L ∞ (Rn ). Then,
for every α ∈ (0, 1),
Cbα (Rn ) = D A (α/2, ∞), Cb2+α (Rn ) = D A (α/2 + 1, ∞). (5.23)
If A is the generator of the Gauss-Weierstrass semigroup (5.3) in X =

L p (Rn ), 1 ≤ p < ∞, then for 0 < α < 1
W α, p (Rn ) = D A (α/2, p), W α+2, p (Rn ) = D A (α/2 + 1, p). (5.24)
Proof. Let us prove (5.23).

The embeddings ⊂ are easy consequences of Example 5.15. Indeed, it
implies that C bα (Rn ) = (BUC(Rn ),BUC2 (Rn ))α/2,∞ . Since BUC(Rn ) ⊂
X and BUC2 (Rn ) ⊂ D(A) in the three cases, then
Cbα (Rn ) = (BUC(Rn ),BUC2 (Rn ))α/2,∞ ⊂ (X,D(A))α/2,∞ = D A (α/2,∞).

Similarly, from Example 5.15 we know that

Cb2+α (Rn ) = (BUC(Rn ), BUC4 (Rn ))(α+2)/4,∞ .
Since BUC(Rn ) ⊂ X and BUC4 (Rn ) ⊂ D(A2 ), then
Cb2+α (Rn ) = (BUC(Rn ), BUC4 (Rn ))(α+2)/4,∞
⊂ (X, D(A2 ))(α+2)/4,∞ = D A (α/2 + 1, ∞),

where the last equality follows from Proposition 3.5.
To prove the opposite inclusions we use estimates (5.4), which imply


 C

 T (t)L(X,Cb1 (Rn )) ≤ 1 + 1/2 , t > 0,
 t
(5.25)



 C 1 C 2 C 3
 T (t)L(X,Cb3 (Rn )) ≤ 1 + 1/2 + + 3/2 , t > 0,
t t t
Replacing T (t) by P(t) := T (t)e−t , the semigroup generated by A − I ,
we get 
 P(t) C

 L(X,Cb1 (Rn )) ≤ 1/2 , t > 0,
 t

 C

 P(t)L(X,C 3 (Rn )) ≤ , t > 0.
b t 3/2
Since D(A) = D(A − I ) then D A (α/2, ∞) = D A−I (α/2, ∞). Using
Theorem 5.12, with E = Cb1 (Rn ), m = 1, β = 1/2, we get
D A−I (α/2, ∞) = (X, D(A − I ))α/2,∞ ⊂ (X, Cb1 (Rn ))α,∞ = Cbα (Rn ).
Therefore,
Cbα (Rn ) ⊃ D A (α/2, ∞).
Since D(A2 ) = D((A− I )2 ) then D A (α/2+1, ∞) = D A−I (α/2+1, ∞).
By Proposition 3.5, the latter space is equal to (X, D((A − I )2 ))(α+2)/4,∞ .
Using again Theorem 5.12, with E = Cb3 (Rn ), m = 2, β = 3/4, we get
D A−I (α/2 + 1, ∞) = (X, D((A − I )2 ))(α+2)/4,∞
⊂ (X, Cb3 (Rn ))(α+2)/3,∞ = Cb2+α (Rn ),

the last equality following from Example 5.15. Therefore,
Cb2+α (Rn ) ⊃ D A (α/2 + 1, ∞),
and (5.23) is proved.
For p > 1, (5.24) may be proved without Theorem 5.12, using the char-
acterizations D(A) = W 2, p (Rn ),D(A2 ) = W 4, p (Rn ) and Example 5.15,
recalling that Proposition 3.5 implies D A (α/2+1, p) = (X,D(A2 ))(α+2)/4,p .
For p = 1 the characterizations of D(A) and D(A2 ) as Sobolev spaces
fail, and we cannot use this shortcut. However, the procedure used in the
proof of (5.23) works also when X = L p (Rn ) for any p ≥ 1, thanks to
estimates (5.6) that imply


 C

 T (t)L(L p ,W 1, p ) ≤ 1 + 1/2 , t > 0,
 t
(5.26)



 C1 C2 C3
 T (t)L(L p ,W 3, p (Rn )) ≤ 1 + 1/2 + + 3/2 , t > 0,
t t t
The details are left to the reader.
A nice consequence of example 5.16 are optimal regularity theorems

for the Laplace equation in Hölder and in fractional Sobolev spaces.
Corollary 5.17. (i) (Schauder Theorem) Let u ∈ C b2 (Rn ) be such that

u ∈ Cbα (Rn ) with 0 < α < 1. Then u ∈ Cbα+2 (Rn ), and
uC α+2 (Rn ) ≤ C(u∞ + uCbα (Rn ) ).

b
(ii) Let u ∈ W 2, p (Rn ) be such that u ∈ W α, p (Rn ) with 0 < α < 1,

1 ≤ p < ∞. Then u ∈ W α+2, p (Rn ), and
uW α+2, p (Rn ) ≤ C(u L p + uW α, p (Rn ) ).
Proof. It is sufficient to recall that D A (θ + 1, p) is the domain of the

part of A in D A (θ, p) for every θ ∈ (0, 1) and p ∈ [1, ∞], and to use
the characterizations of Example 5.16 with A = , θ = α/2 and X =
C b (Rn ), p = ∞ in (i), X = L p (Rn ) in (ii).
If the Laplacian is replaced by a general elliptic operator with bounded

variable coefficients, it still generates an analytic semigroup T (t) in
C b (Rn ) and in L p (Rn ) under mild regularity assumptions, see [2, 88, 50],
but there is not a nice representation formula like (5.3) for T (t). Estim-
ates such as (5.25), (5.26) (and their consequences: Example 5.16 and
Corollary 5.17) are still true if the coefficients are smooth enough, but
the proofs are not obvious. See e.g. [44, 9].
5.3. Regularity in elliptic PDE’s by interpolation

Corollary 5.17 is a nice example of maximal regularity results for the
Laplace equation u = f obtained by interpolation. But the proof is a
sort of lucky strike, because we could characterize both Banach spaces E
and domains of the Laplacian in E as real interpolation spaces between
other spaces and domains. Usually, this is not the case. In this section we
describe a procedure that may be applied to more general situations.
We begin with another interpolation proof of the Schauder theorem. It
uses again estimates in Hölder norms of the derivatives of T (t) f , where
T (t) is the Gauss-Weierstrass semigroup (5.3) generated by the Lapla-
cian.
Lemma 5.18. For 0 ≤ θ1 ≤ θ2 and for each ε > 0 there is C > 0 such
that
Ceεt
T (t)L(C θ1 (Rn ),C θ2 (Rn )) ≤ (θ −θ )/2 . (5.27)
b b t 2 1
Proof. Estimates (5.4) imply that for each ε > 0, and k > h ∈ N ∪ {0},
there is C = C(ε, h, k) such that
Ceεt
T (t)L(C h (Rn ),C k (Rn )) ≤ . (5.28)
b b t (k−h)/2
For θ1 or θ2 non integer, (5.27) can be proved either directly, using for-
mula (5.3), or by interpolation using (5.15) and Theorem 1.6. For in-
stance, using the estimates
T (t)L(Cb (Rn ),Cb1 (Rn )) ≤ Ceεt /t, T (t)L(C 1 (Rn ),C 3 (Rn )) ≤ Ceεt /t,
b b
and the characterizations

(Cb (Rn ), Cb1 (Rn ))α,∞ = Cbα (Rn ), (Cb2 (Rn ), Cb3 (Rn ))α,∞ = Cb2+α (Rn ),
we get T (t)L(C α (Rn ),C 2+α (Rn )) ≤ C eεt /t for t > 0. (By the way, this
b b
shows that T (t) is analytic also in Cbα (Rn )).
Theorem 5.19. Let u ∈ Cbα (Rn ) be such that u ∈ Cbα (Rn ), with 0 <
α < 1. Then u ∈ Cb2+α (Rn ), and there is C > 0, independent of u, such
that
uC 2+α (Rn ) ≤ C(u∞ + uCbα (Rn ) ).
b
Proof. Fix any λ > 0 and set λu − u = f . Let T (t) be the Gauss-
Weierstrass semigroup defined in (5.3). Then for each x ∈ Rn
+∞
u(x) = e−λt (T (t) f )(x)dt. (5.29)
0
Since T (t) f is smooth for t > 0, and using estimates (5.27) with θ1 = α,
θ2 = 2 and ε = λ/2, we can differentiate with respect to x twice, and
obtain u ∈ Cb2 (Rn ),
∞
Di j u(x) = e−λt (Di j T (t) f )(x)dt, i, j = 1, . . . n, x ∈ Rn . (5.30)
0
Using estimates (5.27) gives a further information: taking θ2 ∈ (2, 2 + α)
we get Di j T (t) f C θ2 (Rn ) ≤ C exp(εt)t −1+(α−θ2 )/2 f Cbα , so that u ∈
b
Cb2+θ2 (Rn ). However, taking θ2 = 2 + α we get Di j T (t) f Cbα (Rn ) ≤
C exp(εt)t −1 f Cbα and it is not obvious that Di j u ∈ Cbα (Rn ) because of
the singularity at t = 0. So, we use the following trick: we fix ξ > 0 and
split u = a(ξ ) + b(ξ ), where
ξ ∞
a(ξ ) = e−λt T (t) f dt, b(ξ ) = e−λt T (t) f dt.
0 ξ
Then a(ξ ) ∈ Cbα+θ (Rn )

for each θ ∈ (0, 2), with
ξ
a(ξ )C α+θ ≤C t −θ/2 dt f Cbα = C ξ 1−θ/2 f Cbα ,
b
0
and b(ξ ) ∈ Cb2+α+θ (Rn ), with

+∞
b(ξ )C 2+α+θ ≤ C t −1−θ/2 dt f Cbα = C ξ −θ/2 f Cbα .
b
t
Therefore, for each ξ > 0,
K (ξ, u, Cbα+θ (Rn ), Cb2+α+θ (Rn ))
≤ a(ξ )C α+θ + ξ b(ξ )C 2+α+θ ≤ Cξ 1−θ/2 f Cbα ,

b b
so that, using the definition of (X, Y )γ ,∞ and then Example 5.15,

u ∈ (Cbα+θ (Rn ), Cb2+α+θ (Rn ))1−θ/2,∞ = Cb2+α (Rn ),
We already mentioned that the domain of the Laplacian in Cb (Rn ) is
strictly bigger than Cb2 (Rn ), see formula (5.22). Let us see where the
above proof fails if we take α = 0.
Estimates (5.27) are true also for θ1 = 0, so we still get
K (ξ, u, Cbθ (Rn ), Cb2+θ (Rn )) ≤ Cξ 1−θ/2 f ∞ ,
which implies that u ∈ (Cbθ (Rn ), Cb2+θ (Rn ))1−θ/2,∞ = B∞,∞
2
(Rn ). So,
what we miss in the integer case is the very last step.
Remark 5.20. The proof of Theorem 5.19 could be slightly simplified

starting from formula (5.30) and showing that the second order derivat-
ives Di j u belong to Cbα (Rn ). This follows by the same procedure used in
the proof of the theorem, writing now Di j u = a1 (ξ ) + b1 (ξ ) for every
ξ > 0,
ξ2
a1 (ξ )(x) := (Di j T (t) f )(x) dt, x ∈ Rn ,
0
+∞
b1 (ξ )(x) := (Di j T (t) f )(x) dt, x ∈ Rn .
ξ2
Using estimates (5.27) with = λ/2, θ1 = α, θ2 = 2, 3, we get respect-

ively
ξ2
a1 (ξ )∞ ≤ C t −1+α/2 dt f Cbα = C ξ α f Cbα , (5.31)
0
+∞
b1 (ξ )Cb1 ≤ C t −3/2+α/2 dt f Cbα = C ξ α−1 f Cbα , (5.32)
ξ2
so that K (ξ, Di j u, Cb (Rn ), Cb1 (Rn )) ≤ a1 (ξ )∞ + ξ b1 (ξ )Cb1 ≤

Cξ −α f Cbα which yields Di j u ∈ (Cb (Rn ), Cb1 (Rn ))α,∞ = Cbα (Rn ) by
Example 1.8.
In the last step what we need is just the embedding (Cb (Rn ),
Cb1 (Rn ))α,∞ ⊂ Cbα (Rn ). Looking at its proof suggests a shortcut, namely
for x, y ∈ Rn we can write the above decomposition of Di j u(x)−Di j u(y)
choosing ξ = |x − y| as in the proof of (1.16); using estimates (5.31),
(5.32) we get
|Di j u(x) − Di j u(y)| ≤ 2a1 (|x − y|)∞ + b1 (|x − y|)Cb1 |x − y|
≤ 2C |x − y|α f Cbα + C |x − y|α f Cbα .
Curiously enough, putting together two interpolation arguments one gets

a proof where interpolation could not even be mentioned.
Let us see some extension of Theorem 5.19. First, we may replace

α by α + k, where k is any positive integer, and we get further Hölder
regularity: u, u ∈ C bα+k (Rn ) implies that u ∈ Cbα+k+2 (Rn ).
The same proof works if the Laplace operator is replaced by the
Ornstein-Uhlenbeck operator, because the estimates of Lemma 5.18 hold
for the Ornstein-Uhlenbeck semigroup thanks to (5.10). There is just a
subtle difference, that however does not give big problems. In the case
of the heat semigroup, t → T (t) f is smooth (in fact, analytic) for t > 0
with values in Cbα+θ (Rn ) for each θ > 0, but the same is not true in
the case of the Ornstein-Uhlenbeck semigroup. In fact, it is possible to
show that it is not measurable. Therefore, the integrals that define a(ξ )
and b(ξ ) are not meaningful as C bα+θ and Cb2+α+θ -valued integrals. They
have to be understood pointwise:
ξ
a(ξ )(x) = e−λt (T (t) f )(x)dt,
0
∞
b(ξ )(x) = e−λt (T (t) f )(x)dt, x ∈ Rn .
ξ
Also the Hölder estimates have to be done pointwise; for instance if α +

θ <1
ξ
|a(ξ )(x) − a(ξ )(y)| ≤ e−λt |(T (t) f )(x) − (T (t) f )(y)|dt
0
ξ
≤C t −θ/2 dt f C α |x − y|α+θ .
0
From these examples it is clear that this method may be extended to a

number of situations.
What we need is three Banach spaces Y2 ⊂ Y1 ⊂ X, a smoothing
semigroup, such that T (t)L(X,Yi ) ≤ C exp(ωt)t −γi for t > 0, with
0 ≤ γ1 < 1 < γ2 , and such that the resolvent R(λ, A) of the generator is
represented by formula (5.11) for some λ > ω.
What we get is that the domain of A in X is continuously embedded in
an interpolation space between Y1 and Y2 , precisely in (Y1,Y2)(1−γ1)/(γ2 −γ1),∞ .
Usually the most difficult part is the proof of the estimates T (t)L(X,Yi ) ≤
Ct −γi for t small.
We describe below some situations in which this procedure works.

Estimates (5.27), as well as formula (5.11), are known for semigroups
generated by second order elliptic operators with regular bounded coef-
ficients, and the procedure of theorem 5.19 gives an alternative proof of
Schauder estimates for such operators. However, the method may be ap-
plied to less standard situations.
Example 5.21. Elliptic operators with unbounded regular coefficients
in Rn .
Let A be a second order uniformly elliptic differential operator with reg-

ular, possibly unbounded, coefficients:

n
n
A f (x) = qi j (x)Di j f (x) + bi (x)Di f (x) + v(x) f (x),
i, j=1 i=1
= Tr(Q(x)D 2 f (x)) + B(x), D f (x) + v(x) f (x).
Under several assumptions, it is possible to show that the associated evol-

ution problem

u t (t, x) = Au(t, x), t > 0, x ∈ Rn ,
u(0, x) = u 0 (x), x ∈ Rn ,
with u 0 ∈ Cb (Rn ), has a unique solution u such that u is bounded on

[0, T ]×Rn for each T > 0, and that setting T (t)u 0 = u(t, ·), T (t) enjoys
nice smoothing properties, i.e. it satisfies (5.27) for each 0 ≤ θ1 ≤ θ2 ≤
3, and for suitable ε > 0. Moreover, for λ large and f ∈ Cb (Rn ) the
elliptic problem
λu − Au = f
2, p
has a unique bounded solution u ∈ Cb (Rn ) ∩ Wloc (Rn ) for each p ≥
1, and u is given by the representation formula (5.11). Then the above
procedure gives a Schauder theorem for the operator A.
The hard part of the job is to prove estimates (5.27). They may be ob-
tained either by stochastic methods, see [24], and by deterministic meth-
ods, see [73] and the more recent and sharp [9]. The coefficients may
grow not more than polynomially in [18] and not more than exponen-
tially in [73, 9]. The common assumptions in all these papers are upper
boundedness of v, dissipativity hypotheses on the drift, such as
B(x) − B(y), x − y ≤ ω|x − y|2 ,
and some assumption guaranteeing uniqueness of the bounded solution

to λu − Au = f for bounded f , such as
∃ϕ ∈ C 2 (Rn ) : lim ϕ(x) = +∞, sup Aϕ − λ0 ϕ < +∞

|x|→+∞ x∈Rn
for some λ0 ≥ 0.
In any case, these assumptions cover the Ornstein-Uhlenbeck operator
in the non degenerate case det Q = 0, and more generally operators of
the type u → u+F(x), Du with smooth and Lipschitz continuous F.
Example 5.22. Elliptic operators in infinitely many variables.

The method of Theorem 5.19 was also used to get Schauder estimates
for elliptic operators in separable Hilbert space H , such as
1
(Au)(x) = Tr(Q D 2 u)(x) + Bx, Du(x)
2
where Q ∈ L(H ) is a symmetric positive operator, and B : D(B) → H
is the infinitesimal generator of a strongly continuous semigroup.
The theory of Gaussian measures in Hilbert spaces allows to extend
formulae (5.3) and (5.7) to this setting and to define infinite dimensional
heat and Ornstein-Uhlenbeck semigroups. However, there are additional
difficulties that do not exist in finite dimensions.In the case B = 0,
∞
one needs that the trace of Q (defined as the sum i=1 Qei , ei for each
orthonormal basis {ei : i ∈ N} of H ) is finite. If B is of negative type,
i.e. et B L(H ) ≤ Ce−ωt with C, ω > 0, we can allow also Q = I , but we
need that the operator K t defined in (5.8) has finite trace for each t > 0.
In this second case, estimates (5.27) and the conclusion of Theorem 5.19
hold, with Rn replaced by H . See the papers [22, 23], and the book [32]
as a general reference.
Example 5.23. The degenerate (hypoelliptic) Ornstein-Uhlenbeck oper-
ator.
Let us consider again the Ornstein-Uhlenbeck operator (5.9). We allow
the matrix Q ≥ 0 to be non invertible, but we assume that
det K t > 0, t > 0. (5.33)
This condition is equivalent to the fact that the operator A is hypoelliptic

in the sense of Hörmander ([55]). So, if f ∈ C ∞ (Rn ) and u is a distribu-
tional solution of
Au = f, (5.34)
then u ∈ C ∞ (Rn ). The hypoellipticity condition is also equivalent to
Rank [Q 1/2 , B Q 1/2 , . . . , B n−1 Q 1/2 ] = n, (5.35)
which is well known in control theory and is called Kalman rank condi-
tion. See e.g. [92, Chapter 1] for a proof of the equivalence.
Since A is a degenerate elliptic operator, the usual Hölder spaces are
not the right setting for A. It is more convenient to replace them by
suitable Hölder spaces, adapted to A, defined as follows.
Let k ∈ {0, . . . , n − 1} be the smallest integer such that
Rank [Q 1/2 , B Q 1/2 , . . . , B k Q 1/2 ] = n.

Note that the matrix Q is nonsingular if and only if k = 0. Set V0 =

Range Q 1/2 , Vh = Range Q 1/2 + Range B Q 1/2 +...+ Range B h Q 1/2 .
Of course, Vh ⊂ Vh+1 for every h, and Vk = Rn . Define the orthogonal
projections E h as

 E 0 = projection on V0 ,
(5.36)
 ⊥
E h+1 = projection on (Vh ) in Vh+1 , h = 1, . . . , k − 1.

Then Rn = kh=0 E h (Rn ). By possibly changing coordinates in Rn we
will consider an orthogonal basis {e1 , . . . , en } consisting of generators of
the subspaces E h (Rn ). For every h = 0, . . . , k we define Ih as the set
of indices i such that the vectors ei with i ∈ Ih span E h (Rn ). After the
change of coordinates the second order derivatives which appear in (5.9)
are only the Di j u with i, j ∈ I0 .
The distance d associated to A is defined by

k
d(x, y) = |E h (x − y)|1/(2h+1) ,
h=0
where | · | is the usual Euclidean norm. For α > 0 such that α/(2h +
/ N for h = 0, . . . , k, the space Cdα (Rn ) is the set of all the bounded
1) ∈
functions f : Rn → R such that for every x0 ∈ Rn , 0 ≤ h ≤ k, the
mapping E h (Rn ) → R, x → f (x0 + x) belongs to C α/(2h+1) (E h (Rn )),
with norm bounded by a constant independent of x 0 . It is endowed with
the norm

k
f Cdα (Rn ) = sup f (x0 + ·)C α/(2h+1) (Eh (Rn )) .
h=0 x0 ∈R
n
In particular, for 0 < α < 1 it is the space of the bounded functions f

such that
k −1
α/(2h+1)
[ f ]Cdα (Rn ) = sup | f (x) − f (y)| |E h (x − y)| < ∞.
x,y∈Rn , x = y h=0
Using the representation formula (5.9) it is possible to show that (5.27)

holds provided the usual Hölder spaces are replaced by the spaces Cdθ (Rn).
Using the above arguments we arrive at the following theorem: Let
0 < α < 1, λ > 0. Then for every f ∈ Cdα (Rn ) the problem
λu − Au = f
has a unique distributional solution u in the space of the uniformly con-

tinuous and bounded functions. Moreover u belongs to Cd2+α (Rn ), and
there is C > 0, independent of f , such that
uC 2+α (Rn ) ≤ C f Cdα (Rn ) . (5.37)

d
See [72].
The simplest significant example of this class of operators is the
Kolmogorov operator in R2 ,
Au(x, y) = u x x − xu y ,
in which case k = 1, V0 = R × {0}, V1 = {0} × R, and (with obvious

α,α/3
notation) Cdα (R2 ) = C x y (R2 ), Cd2+α (R2 ) = C x y
2+α,2/3+α/3
(R2 ).
Example 5.24. Elliptic operators in L 1 (Rn ).

Instead of taking as reference space a space of continuous and bounded,
or Hölder continuous functions, we consider now X = L 1 (Rn ). The heat
semigroup T (t) is easily seen to satisfy
C C
DT (t) f L 1 ≤ f L 1 , DT (t) f ∞ ≤ f L 1 , (5.38)
t 1/2 t (n+1)/2
for each t > 0 and f ∈ L 1 (Rn ). The procedure of Theorem 5.19, applied
to the derivatives Di u instead of u, give that if u, u are in L 1 (Rn ) then
Di u ∈ (L 1 (Rn ), L ∞ (Rn ))1/n,∞ = L n/(n−1),∞ (Rn ), (5.39)
where for p ≥ 1, L p,∞ (Rn ) is the Lorentz space defined in Section 1.1.1.
We already said that the domain of the realization of the Laplacian in
L 1 (Rn ) is not contained in W 2,1 (Rn ) if N > 1, i.e. the derivatives of
the functions u in the domain do not belong to W 1,1 (Rn ) in general. One
may ask whether a weaker inclusion holds, i.e. Di u ∈ L n/(n−1) (Rn ) (this
is because, by Sobolev embedding, W 1,1 (Rn ) ⊂ L n/(n−1) (Rn )). But easy
counterexamples show that this is not true, and the most we can get (in
the scale of the Lebesgue and Lorentz spaces) is (5.39).
Estimates (5.38), and consequently embedding (1.24), are true for a
number of semigroups generated by elliptic operators in L 1 spaces, in-
cluding elliptic operators with unbounded Lipschitz continuous coeffi-
cients in Rn . See [77].
Example 5.25. Elliptic operators in fractional Sobolev spaces.

Let us come back to theorem 5.19 for the Laplacian, taking as reference
space X = L p (Rn ) or X = W α, p (Rn ), 1 ≤ p < ∞. It is natural to work
in the interpolation spaces (X, Y )θ, p instead of (X, Y )θ,∞ , but in this
case estimates similar to (5.27),
Ceεt Ceεt
T (t)L(L p (Rn ),W θ, p (Rn )) ≤ , T (t)L(W θ1 , p (R ),W
n θ2 , p (R ))
n ≤ ,
t θ/2 t (θ2 −θ1 )/2
(that are easily seen to be true) are not enough. What we need is
 (θ −θ )/2 −t
 t 2 1 e T (t)L(W θ1 , p (Rn ),W θ2 , p (Rn )) ≤ c(t), t > 0,
(5.40)
 p
c ∈ L ∗ (0, +∞), i = 1, 2.
which is still true if θ1 and θ2 are not integers, but the proof is not immedi-
ate. Once (5.40) is established, we use it to prove that for f ∈ W α, p (Rn ),
the function u given by (5.29) satisfies
ξ → ξ θ/2−1 K (ξ, u, W α+θ, p (Rn ), W 2+α+θ, p (Rn )) ∈ L ∗p (0, +∞),
and hence, by the definition of (X, Y )γ , p and then Example 5.15,
u ∈ (W α+θ, p (Rn ), W 2+α+θ, p (Rn ))1−θ/2, p = W 2+α, p (Rn ),
if α is not integer. This implies that the domain of the Laplacian in the
space W α, p (Rn ) is W 2+α, p (Rn ), if α is not integer. So, we have an altern-
ative proof of Corollary 5.17(ii).
Estimates (5.40), as well as the conclusion, are still true for the
Ornstein-Uhlenbeck semigroup in the nondegenerate case. A proof is
in [75].
5.4. Regularity in evolution equations by interpolation

We follow as far as possible the procedure of Section 5.3 for evolution
problems of the type

 u (t) = Au(t) + f (t), 0 < t < T,

u(0) = 0.
Also for evolution problems, we start with the case where A is the real-
ization of the Laplacian in a space of Hölder continuous functions, and
then we see to which extent it is possible to generalize the procedure.
As in the stationary case, the main ingredients are the representation

formula for u that involves the semigroup T (t) generated by A, i.e. the
variation of constants formula
t t
u(t) = T (t − s) f (s)ds = T (s) f (t − s)ds,
0 0
and the smoothing properties of the semigroup.

The parallel of the Schauder theorem 5.19 is the following result which
was proved for the first time by Kružkhov, Castro and Lopes in [64, 65]
by different methods.
Definition 5.26. The space C 0,α ([0, T ] × Rn ) consists of the continuous
and bounded functions f : [0, T ] × Rn → C which are α-Hölder con-
tinuous with respect to the space variables, uniformly in t. It is endowed
with the norm
f C 0,α = f ∞ + sup [ f (t, ·)]Cbα (Rn ) .

0≤t≤T
Theorem 5.27. Let f ∈ C 0,α ([0, T ] × Rn ). Then problem


 u t (t, x) = u(t, x) + f (t, x), 0 ≤ t ≤ T, x ∈ Rn ,

u(0, x) = 0, x ∈ Rn ,
has a unique classical solution u such that u t and D i j u belong to

C 0,α ([0, T ] × Rn ). There is C > 0 independent of f such that
sup u(t, ·)C 2+α (Rn ) ≤ C sup f (t, ·)Cbα (Rn ) .

b
0≤t≤T 0≤t≤T
Proof. We use the variation of constants formula, that implies easily that
u is continuous and bounded in [0, T ] × Rn .
Let us show that u(t, ·) is bounded with values in Cb2+α (Rn ). Fix t ∈
[0, T ]. For ξ > 0 split again u(t) = a(ξ ) + b(ξ ), where
min(ξ,t)
a(ξ ) = (T (s) f (t − s))(x)ds,
0
t
b(ξ ) = (T (s) f (t − s))(x)ds.
min(ξ,t)
Then a(ξ ) ∈ Cbα+θ (Rn ), and b(ξ ) ∈ Cb2+α+θ (Rn ), for each θ ∈ (0, 2).
Precisely, using (5.27) we get
min(ξ,t)
CeεT
a(ξ )C α+θ (Rn ) ≤ ds sup f (s)Cbα (Rn )
b
0 s θ/2 0≤s≤T
CeεT
≤ ξ 1−θ/2 sup0≤s≤T f (s)Cbα (Rn ) ,
1 − θ/2
t
CeεT
b(ξ )C 2+α+θ (Rn ) ≤ ds sup f (s)Cbα (Rn )
b
min(ξ,t) s 1+θ/2 0≤s≤T
CeεT −θ/2
≤ ξ sup0≤s≤T f (s)Cbα (Rn ) .
θ/2
Therefore,
ξ −(1−θ/2) K (ξ, u(t), Cbα+θ (Rn ), Cb2+α+θ (Rn ))

εT 1 2
≤ Ce + sup f (s)Cbα (Rn ) , ξ > 0,
1 − θ/2 θ 0≤s≤T
so that u(t) belongs to (Cbα+θ (Rn ), Cb2+α+θ (Rn ))1−θ/2,∞ = Cb2+α (Rn ), and
u(t)C 2+α (Rn ) ≤ C sup f (s)Cbα (Rn ) , t ∈ [0, T ].

b
0≤s≤T
To show that u is differentiable with respect to t it is sufficient to examine

the variation of constants formula. For every s ∈ [0, T ] and x ∈ Rn the
function t → (T (t −s) f (s, ·))(x) is continuously differentiable in (s, T ],
and ∂/∂t (T (t − s) f (s, ·))(x) = (T (t − s) f (s, ·))(x). Moreover, by
(5.27) we have
C
|(T (t − s) f (s, ·))(x)| ≤ sup f (σ, ·)C α (Rn ) ,
(t − s)1−α/2 0≤σ ≤T
so that u is differentiable with respect to time and u t = u + f in [0, T ]×

Rn . The statement follows.
Since T (t) maps Cb2+α (Rn ) into itself, and the norm T (t)L(C 2+α (Rn ))
b
is bounded by a constant independent of t, the following corollary holds.
Corollary 5.28. Let f ∈ C 0,α ([0, T ] × Rn ) and u 0 ∈ Cb2+α (Rn ). Then

problem

 u t (t, x) = u(t, x) + f (t, x), 0 ≤ t ≤ T, x ∈ Rn ,

u(0, x) = u 0 (x), x ∈ Rn ,
has a unique classical solution u such that u t and Di j u belong to

C 0,α ([0, T ] × Rn ). There is C > 0 independent of f such that
sup u(t)C 2+α (Rn ) ≤ C(u 0 C 2+α (Rn ) + sup f (t)Cbα (Rn ) ).
b b
0≤t≤T 0≤t≤T
Theorem 5.27 and its corollary are immediately extended to the non de-
generate Ornstein-Uhlenbeck operator, with the same proof. Since the
ingredients are estimates (5.27) and the variation of constants formula for
the solution, it can be extended to a number of other situations, including
Examples 5.21, . . . , 5.23. They can be extended also to fractional So-
bolev spaces, using (X, Y )θ, p spaces such as in example 5.25, see [75].
We do not enter into further details.
Chapter 6
Analytic semigroups and interpolation
Throughout the chapter X is a complex Banach space, A : D(A) ⊂ X →

X is a sectorial operator (i.e., a linear operator satisfying (5.1)), and et A
is the semigroup generated by A.
Let us mention some important properties of analytic semigroups and
of their generators, that will be used in the sequel.
First of all, et A ∈ L(X, D(Am )) for every t > 0 and m ∈ N, and
d /dt m et A = Am et A for t > 0. Moreover there are constants Ck > 0
m
such that

 (a) et A L(X) ≤ C0 eωt , t > 0,
(6.1)

(b) t k (A − ωI )k et A L(X) ≤ Ck eωt , t > 0, k ∈ N,
where ω is the constant in (5.1). Note that for every x ∈ X and 0 < s < t,
the function σ → eσ A x is in C ∞ ([s, t]; D(A)), so that
t t
σA
e x −e x =
tA sA
Ae xdσ = A eσ A xdσ.
s s
The same formula is true also for s = 0, in the sense specified by the
following lemma.
t
Lemma 6.1. For every x ∈ X and t ≥ 0, the integral 0 es A xds belongs
to D(A), and t
A es A xds = et A x − x.
0
If in addition the function s → Aes A x belongs to L 1 (0, t; X), then
t
e x−x =
tA
Aes A xds.
0
Other properties of sectorial operators and analytic semigroups may be

found in [71, Chapter 2].
6.1. Characterization of real interpolation spaces

From now on we denote by M0 , M1 two constants such that et A L(X) ≤
M0 , t Aet A L(X) ≤ M1 , for every t ∈ (0, 1].
Since A is sectorial, the operator A − ωI satisfies (3.1), and we may
use the results of Section 3.1 to characterize the interpolation spaces
(X, D(Am ))θ, p .
Another characterization, which is very useful in abstract parabolic
problems, was found by Butzer and Berens (see e.g. [16]) for m = 1.
Proposition 6.2. For 0 < θ < 1, 1 ≤ p ≤ ∞ we have
(X, D(A))θ, p = {x ∈ X : ϕ(t) := t 1−θ Aet A x ∈ L ∗p (0, 1)},
and the norms · (X,D(A))θ, p and
x → x + ϕ L ∗p (0,1)
are equivalent.
Proof. For every x ∈ (X, D(A))θ, p let a ∈ X, b ∈ D(A) be such that
x = a + b. Then
t 1−θ Aet A x ≤ t 1−θ Aet A a + t 1−θ Aet A b
≤ t −θ M1 a + t 1−θ M0 Ab
≤ max{M0 , M1 }t −θ (a + tb D(A) )
so that
t 1−θ Aet A x ≤ max{M0 , M1 }t −θ K (t, x, X, D(A)),
p
which yields that ϕ ∈ L ∗ (0,1) and ϕ L ∗p (0,1) ≤ max{M0 ,M1 }x(X,D(A))θ, p .
p
Conversely, if ϕ ∈ L ∗ (0, 1) write x as
t
x = (x − e x) + e x = −
tA tA
Aes A x ds + et A x, 0 < t < 1,
0
so that

−θ
t
t −θ
K (t, x, X, D(A)) ≤ t Ae x ds
sA
+ t e x D(A) .
1−θ tA
0
p
Since t 1−θ Aet A x ∈ L ∗ (0, 1),
by Corollary A.13
t the same is true for
t 1−θ v(t), where v is the mean value v(t) = t −1 0 Aes A xds, and
t
1

t → t v(t) L ∗ (0,1) = t → t
1−θ −θ
Ae xds
sA
≤ ϕ L ∗p (0,1) .
p
p θ
0 L ∗ (0,1)
Moreover,
et A x D(A) = et A x + Aet A x ≤ M0 x + t −1+θ ϕ(t).
Therefore
t
−θ −θ
t K (t, x, X, D(A)) ≤ t Aes A xds + t 1−θ M0 x + ϕ(t),
0
p
so that the function t → t −θ K (t, x, X, D(A)) belongs to L ∗ (0, 1), with
norm estimated by C(x + ϕ L ∗p (0,1) ). We recall that it belongs also to
p
L ∗ (1, ∞), with norm not exceeding Cx. Therefore, x ∈ (X, D(A))θ, p ,
Remark 6.3. In the case θ = 1−1/ p with p ∈ (1, +∞), Proposition 6.2
yields
1
x ∈ (X, D(A))1−1/ p, p ⇐⇒ Aet A x p dt < ∞.
0
So, we can make two remarks:

Remark 1. Recalling that d/dt et A x = Aet A x for t > 0, we get
x ∈ (X, D(A))1−1/ p, p ⇐⇒ t → et A x ∈ W 1, p (0, 1; X) ∩ L p (0, 1; D(A)).
As a consequence of Corollary 1.14, we already know that x ∈ (X,

D(A))1−1/ p, p iff x is the trace at t = 0 of a function u ∈ W 1, p (0, 1; X) ∩
L p (0, 1; D(A)). Proposition 6.2 yields that we can take u(t) = et A x.
Remark 2. Assume that the constant ω in (5.1) is ≤ 0, so that the oper-
ator −A is nonnegative in the sense of Definition 4.10, and the powers
(−A)θ are well defined for every θ ∈ (0, 1). If p ∈ (1, +∞) is such that
D((−A)1−1/ p ) ⊂ (X, D(A))1−1/ p, p , then there exists C > 0 such that
1
(−A)1/ p et A x p dt ≤ Cx, x ∈ X.
0
1 1
Indeed, in this case we have 0 (−A)1/ p et A x p dt = 0 Aet A y p dt,
with y = (−A)−(1−1/ p) x ∈ (X, D(A))1−1/ p, p .
However, embeddings such as D((−A)1−1/ p ) ⊂ (X, D(A))1−1/ p, p
are not obvious. An example with equality (not only embedding) is in
Chapter 4: if X is a Hilbert space and A is a self-adjoint operator satisfy-
ing (4.38), by Theorem 4.36 we have D(−A)θ = (X, D(A))θ,2 for every
θ ∈ (0, 1); in particular D(−A)1/2 = (X, D(A))1/2,2 .
Proposition 6.2 has the following generalization.

Proposition 6.4. For 0 < θ < 1, 1 ≤ p ≤ ∞ we have
(X, D(Am ))θ, p = {x ∈ X : ϕm (t) = t m(1−θ) Am et A x ∈ L ∗p (0, 1)},
and the norms · (X,D(Am ))θ, p and
x → x + ϕm L ∗p (0,1)
are equivalent.
Proof. (Sketch) The embedding ⊂ may be proved as in Proposition 6.2,
splitting
t m(1−θ) Am et A x = t m(1−θ) Am et A a + t m(1−θ) et A Am b
if x = a + b with a ∈ X and b ∈ D(Am ). Also the idea of the proof of
the other embedding is similar, but now instead of the kernel Aes A x we
have to use s m−1 Am es A x. For instance, for m = 2 we have
t t
d
s A2 es A x ds = s Aes A x ds
0 0 ds
t
=− Aes A x ds + t Aet A x = x − et A x + t Aet A x.
0
For every t ∈ (0, 1) we split x as
t
x= s A2 es A x ds + et A x − t Aet A x.
0
p
Since t t A e x ∈ L ∗ (0,
1−2θ 2 tA
1) the2 same is true for t 1−2θ v(t), where v
−1 t
is the mean value v(t) = t 0 s A e xds. Therefore
sA
t
t → g(t) = t −2θ s A2 es A xds ∈ L ∗p (0, 1).
0
So,
t −2θ K (t2, x, X, D(A2 ))
t
−2θ
≤t s A e xds + t e x − t Ae x D(A2 )
2 sA 2 tA tA
0
≤ g(t) + t 2−2θ
((M0 + M1 )x + A2 et A x + 2M1 A2 et A/2 x)
p
and t → t −2θ K (t 2 , x, X, D(A2 )) belongs to L ∗ (0, 1), with norm es-
timated by C(ϕ2 L ∗p (0,1) + x). Then t → t −θ K (t, x, X, D(A2 )) ∈
p
L ∗ (0, 1), again with norm less than C(ϕ2 L ∗p (0,1) + x), and the state-
ment follows for m = 2.
If m is arbitrary the procedure is similar, with the kernel s A2 es A x re-
placed by s m−1 Am es A x.
If θm is not integer, Proposition 6.4 may be deduced from Proposi-

tion 3.8 and Proposition 6.2. If θm = n is integer, the considerations of
Remark 5.11 still hold. Analyticity of et A does not add information about
coincidence of D(An ) with any interpolation space.
6.1.1. Exercises
1. Prove that (X, D(A))θ = {x ∈ X : limt→0 t 1−θ Aet A x = 0}, for
each θ ∈ (0, 1).
2. Referring to Exercise 5(ii), Section 1.4.3, prove that Lip(I ;(X,Y )α,∞ )
cannot be replaced in general by C 1 (I ; (X, Y ) α, ∞ ) or by
Lip(I ; (X, Y )α, p ) with p < ∞, considering the function u(t) =
et A x.
6.2. Generation of analytic semigroups by interpolation

We shall use interpolation techniques to prove that certain operators are
sectorial in suitable functional spaces.
Theorem 6.5. Let Y be any interpolation space between X and D(A).

Then the part of A in Y , that is the operator
AY : D(AY ) → Y, D(AY ) = {y ∈ D(A) : Ay ∈ Y }, AY y = Ay
is sectorial in Y .
In particular, for every θ ∈ (0, 1), 1 ≤ p ≤ ∞. the parts of A in
D A (θ, p), in D A (θ), and in [X, D(A)]θ are sectorial operators. Sim-
ilarly, for every k ∈ N the part of A in D A (θ + k, p) is sectorial in
D A (θ + k, p).
Proof. Let λ ∈ Sβ,ω . Since the resolvent operator R(λ, A) commutes

with A on D(A), then R(λ, A)L(D(A)) ≤ M/|λ − ω|. By interpolation
it follows that R(λ, A) ∈ L(Y ) and R(λ, A)L(Y ) ≤ M/|λ − ω|, and the
statement follows.
In the next theorem we use the Stein interpolation Theorem to prove

generation of analytic semigroups in L p spaces.
Theorem 6.6. Let (, µ) be a σ -finite measure space, and let T (t) :
L 2 () + L ∞ () → L 2 ()+L ∞ () be a semigroup such that its restric-
tion to L 2 () is a bounded analytic semigroup in L 2 () and its restric-
tion to L ∞ () is a bounded semigroup in L ∞ (). Then the restriction
of T (t) to L p () is a bounded analytic semigroup in L p (), for every
p ∈ (2, ∞).
Proof. Let θ0 ∈ (0, π/2) and M > 0 be such that T (t) has an analytic
extension to the sector 2 = {z ∈ C : z = 0, |arg z| < θ0 }, and
T (z)L(L 2 ) ≤ M for every z ∈ 2 , T (t)L(L ∞ ) ≤ M for every t > 0.
Let S be the strip {z ∈ C : Re z ∈ [0, 1]}. For every r > 0 and
θ ∈ (−θ0 , θ0 ) define a function h : S → 2 by
h(z) = reiθ(1−z) , z ∈ S,
and define a family of operators z ∈ L(L 2 ) by
z = T (h(z)), z ∈ S.
Then z → z is continuous and bounded in S, holomorphic in the in-

terior of S, with values in L(L 2 ). Consequently, if a, b are simple func-
tions, the function

z → (z a)(x)b(x)µ(dx)

is continuous and bounded in S, holomorphic in the interior of S. If

Re z = 1, h(z) = reθ Im z is a positive real number, so that z ∈ L(L ∞ ).
Moreover
sup it L(L 2 ) ≤ M, sup 1+it L(L ∞ ) ≤ M.

t∈R t∈R
By the Stein interpolation Theorem 2.15, applied with p0 = q0 = 2,

p1 = q1 = +∞, for every s ∈ (0, 1) the operator s has an extension in
L(L p ) with p = 2/(1 − s), and s L(L p ) ≤ M. Note that since s runs
in (0, 1), then p runs in (2, ∞). s is nothing but T (reiθ(1−s) ). Since r
and θ are arbitrary, T (z) ∈ L(L p ) and T (z)L(L p ) ≤ M for every z = 0
with |arg z| < θ0 (1 − s) = 2θ0 / p.
Let us show that z → T (z) is holomorphic in the sector p = {z =
0 : |arg z| < 2θ0 / p} with values in L(L p ) (for the moment we know
only that it is bounded).

Let f ∈ L p , g ∈ L p , and let f n ∈ L p ∩ L 2 , gn ∈ L p ∩ L 2 , be such that

f n → f in L p , gn → g in L p . Then the functions z → T (z) f n , gn
are holomorphic in 2 ⊃ p , and converge to T (z) f, g uniformly in
p because T (z)L(L p ) is bounded in p . Therefore z → T (z) f, g
is holomorphic in p . This implies that z → T (z) is holomorphic with
values in L(L p ).
Theorem 6.6 may be easily generalized as follows: if a semigroup T (t)

is analytic in L p0 and bounded in L p1 then it is analytic in L p for every p
in the interval with endpoints p0 and p1 . But the most common situation
is p0 = 2, p1 = ∞. For instance, if

n
Au = Di (ai j (x)D j u)(x), x ∈ Rn ,
i, j=1
1
where the coefficients ai j are in Hloc (Rn ) and

n
ai j (x)ξi ξ j ≥ 0, x, ξ ∈ Rn ,
i, j=1
then the realization of A in L 2 (Rn ) generates an analytic bounded semig-

roup in L 2 (Rn ), whose restriction to L ∞ (Rn ) is a bounded semigroup in
L ∞ (Rn ). The proof may be found in the book of Davies [33, Ex. 3.2.11].
6.3. Regularity in abstract parabolic equations

In this section X is a complex Banach space, A : D(A) → X is a sec-
torial operator, and T > 0 is fixed.
Let f : [0, T ] → X, and u 0 ∈ X. Cauchy problems of the type

u (t) = Au(t) + f (t), 0 < t < T,
(6.2)
u(0) = u 0
are called abstract parabolic problems, since the most known examples
of sectorial operators are the realizations of elliptic differential operators
in the usual functional Banach spaces (L p spaces, spaces of continuous
or Hölder continuous functions in Rn or in open sets of Rn , etc.).
We shall see several regularity results for (6.2) involving interpolation
spaces or proved by interpolation techniques. The starting point of most
of our analysis is the variation of constants formula for the solution
t
u(t) = e u 0 +
tA
e(t−s)A f (s)ds, 0 ≤ t ≤ T. (6.3)
0
In fact it is not difficult to see that if (6.2) has a solution u ∈ L 1 ((0, T );

D(A)) ∩ W 1,1 ((0, T ); X) then u is given by (6.3).
If (6.2) has a solution u ∈ L p ((0, T ); D(A)) ∩ W 1, p ((0, T ); X) for
some p ≥ 1, then of course f ∈ L p ((0, T ); X) and moreover u 0 ∈
D A (1 − 1/ p, p) if 1 < p < ∞, by Corollary 1.14. Similarly, if (6.2) has
a solution u ∈ C 1 ([0, T ]; X) ∩ C([0, T ]; D(A)) then f ∈ C([0, T ]; X)
and u 0 ∈ D(A), Au 0 + f (0) = u (0) ∈ D(A), since it is the limit of the
incremental ratio at t = 0.
However, the converse statements are not true in general, even for very
good initial data u 0 (e.g. u 0 = 0): the range of the function v defined by
t
v(t) := e(t−s)A f (s)ds, 0 ≤ t ≤ T, (6.4)
0
is not contained in D(A) in general, and v may fail to be differentiable,

even weakly, with values in X.
Some of its regularity properties are proved in the next two proposi-
tions. We set
Mk = sup t k Ak et A L(X) , k ∈ N ∪ {0}, (6.5)

0<t≤T +1
and, for α ∈ (0, 1),
Mk,α = sup t k−α Ak et A L(D A (α,∞),X) , k ∈ N. (6.6)

0<t≤T +1
Proposition 6.7. Let f ∈ L p (0, T ; X), with 1 ≤ p < ∞. Then, for

every α ∈ (0, 1), v belongs to L p ((0, T ); D A (α, p)) ∩ W α, p ((0, T ); X).
Consequently, it belongs to W α−θ, p ((0, T ); D A (θ, p)) for every θ < α <
1 and to C([0, T ]; D A (α, p)) for every α ∈ (0, 1 − 1/ p).
There is C independent of f such that
vW α, p ((0,T );X) + v L p ((0,T );D A (α, p)) ≤ C f L p (0,T ;X) . (6.7)
Note that v does not necessarily belong to W 1−θ, p ((0, T ); D A (θ, p)) for
any θ. Indeed, even the embedding W 1, p ((0, T ); X) ∩ L p ((0, T ); D(A))
⊂ W 1−θ, p ((0, T ); D A (θ, p)) does not hold, in general. From this point
of view the results for p = ∞ are much nicer than for p < ∞, as the
next proposition shows.
Proposition 6.8. Let f ∈ L ∞ (0,T ;X). Then v ∈ C 1−α ([0,T ]; D A (α,1)),

and there is C independent of f such that
vC 1−α ([0,T ];D A (α,1)) ≤ C f L ∞ (0,T ;X) . (6.8)
In particular, v ∈ C α ([0, T ]; X) ∩ C([0, T ]; D A (α, 1)) for every α ∈

(0, 1).
Proof. Due to estimate (6.5), with k = 0, v satisfies

t
v(t) ≤ M0 f (s)ds ≤ M0 T f ∞ , 0 ≤ t ≤ T. (6.9)
0
Since et A L(X) and tet A L(X,D(A)) are bounded in (0, T ], by interpola-
tion there is K 0,α > 0 such that et A L(X,D A (α,1)) ≤ K 0,α t −α for 0 < t ≤
T . Similarly, since t Aet A L(X) and t 2 Aet A L(X,D(A)) are bounded in
(0, T ], by interpolation there is K 1,α > 0 such that Aet A L(X,D A (α,1)) ≤
K 1,α t −α−1 for 0 < t ≤ T .
Therefore, s → e(t−s)A L(X,D A (α,1)) belongs to L 1 (0, t) for every t ∈
(0, T ]. Then v(t) belongs to D A (α, 1) for every α ∈ (0, 1), and
v(t) D A (α,1) ≤ K 0,α (1 − α)−1 T 1−α f L ∞ (0,T ;X) , (6.10)
Moreover, for 0 ≤ s ≤ t ≤ T ,
t

(t−σ )A
s
(s−σ )A

v(t) − v(s) = e −e f (σ )dσ + e(t−σ )A f (σ )dσ
0
s t−σ t s
= dσ Aeτ A f (σ )dτ + e(t−σ )A f (σ )dσ,
0 s−σ s
which implies
s t−σ
v(t) − v(s) D A (α,1) ≤ K 1,α dσ τ −1−α dτ f ∞
0 s−σ
t
+ K 0,α (t − σ )−α dσ f ∞ (6.11)
s

K 1,α K 0,α
≤ + (t − s)1−α f ∞ ,
α(1 − α) 1 − α
so that v is (1 − α)-Hölder continuous with values in D A (α, 1). Estimate

(6.8) follows now from (6.10) and (6.11).
Note that the estimates for v(t) blow up as α → 1, both in Proposi-

tion 6.7 and in Proposition 6.8. So, they are not useful to see whether v
is differentiable with values in X or has values in D(A). To this purpose,
next lemma is useful because it cuts half of the job.
Lemma 6.9. Let u be the function defined by (6.3). The following state-
ments hold.
(a) Let f ∈ L p ((0, T ); X) with 1 ≤ p ≤ ∞. If u ∈ L p ((0, T ); D(A))

then u ∈ W 1, p ((0, T ); X) and u = Au + f a.e. in (0, T ).
(b) Let f ∈ C([0, T ]; X). If u ∈ C([0, T ]; D(A)) then u ∈ C 1 ([0, T ]; X)
and u = Au + f in [0, T ].
t
Proof. First we show that for f ∈ L 1 ((0, T ); X), the integral 0 u(s)ds
belongs to D(A), and
t t
u(t) = u 0 + A u(s)ds + f (s)ds, 0 ≤ t ≤ T. (6.12)
0 0
Indeed, for every t ∈ [0, T ] we have

t t t s
u(s)ds = e u 0 ds +
sA
ds e(s−σ )A f (σ )dσ
0 0
t t 0 t
0
= es A u 0 ds + dσ e(s−σ )A f (σ )ds.
0 0 σ
t t−σ
By Lemma 6.1, the integral σ e(s−σ )A f (σ )ds = 0 eτ A f (σ )dτ is in
D(A), and
t
A e(s−σ )A f (σ )ds = (e(t−σ )A − I ) f (σ ) ∈ L 1 (0, t).
σ
t
Therefore, using again Lemma 6.1, the integral 0 u(s)ds belongs to
D(A), and
t t
A u(s)ds = et A u 0 − u 0 + (e(t−σ )A − 1) f (σ )dσ, 0 ≤ t ≤ T,
0 0
so that (6.12) holds.

If u ∈ L 1 ((0, T ); D(A)), since A is a closed operator, formula (6.12)
may be rewritten as
t t
u(t) = u 0 + Au(s)ds + f (s)ds, 0 ≤ t ≤ T,
0 0
and both statements (a) and (b) follow.

Now we are ready to prove better regularity results for (6.2), including
maximal regularity results.
Maximal (or optimal) regularity is an important and well studied item in
the theory of evolution equations. If E is a Banach space of functions
defined in (0, T ), we say that there is maximal regularity in E if for each
f ∈ E the mild solution u to (6.2) is such that u and Au (exist and)
belong to E, provided of course that the initial datum u 0 satisfies the due
necessary conditions. In Section 3.2.2 and in Section 4.2.1 we already
saw examples of maximal regularity, as a consequence of the theory of
sums of operators.
We prove here the simplest maximal regularity results, with E =

α
C ([ 0, T ]; X), E = C ([ 0, T ]; X) ∩ B ([ 0, T ]; D A (α, ∞)), E =
C([ 0, T ]; D A (α)), following the proof of [85] that uses only estimates
(6.1) and the characterization of D A (α, ∞) given by Proposition 6.2.
Other maximal regularity theorems whose proofs rely only on estimates
(6.1) and on Proposition 6.2 are in [71, 35].
Theorem 6.10. Let 0 < α < 1, f ∈ C α ([0, T ]; X), u 0 ∈ X. Then the

mild solution u of (6.2) belongs to C α ([ε, T ]; D(A)) ∩ C 1+α ([ε, T ]; X)
for every ε ∈ (0, T ), and
(i) if u 0 ∈ D(A) and Au 0 + f (0) ∈ D(A), then u ∈ C 1 ([0, T ]; X) ∩

C([0, T ]; D(A)), and there is C such that
uC 1 ([0,T ];X)+uC([0,T ];D(A)) ≤ C( f C α ([0,T ];X)+u 0 D(A) ); (6.13)
(ii) if u 0 ∈ D(A) and Au 0+f (0) ∈ D A (α,∞), then u , Au ∈ C α ([0,T ]; X),

u ∈ B([0, T ]; D A (α, ∞)), and there is C such that
uC 1+α ([0,T ];X) + AuC α ([0,T ];X) + u B([0,T ];D A (α,∞))
(6.14)
≤ C( f C α ([0,T ];X) + u 0 D(A) + Au 0 + f (0) D A (α,∞) ).
Proof. Let us split u as u = u 1 + u 2 , where

 t

 e(t−s)A ( f (s) − f (t))ds, 0 ≤ t ≤ T,
 u 1 (t) =
0
t (6.15)


 u 2 (t) = e u 0 +
t A
e(t−s)A f (t)ds, 0 ≤ t ≤ T.
0
By Lemma 6.1, u 2 (t) ∈ D(A) for each t ∈ (0, T ]. Since Ae(t−s)A L(X) ≤
M1 (t − s)−1 (estimate (6.5)), and f is Hölder continuous, then the func-
tion s → Ae(t−s)A ( f (s) − f (t)) is in L 1 ((0, t); X). This implies that
u 1 (t) ∈ D(A) for t ∈ [0, T ]. Moreover we have
 t


 (i) Au 1 (t) = Ae(t−s)A ( f (s) − f (t))ds, 0 ≤ t ≤ T,
0 (6.16)



(ii) Au 2 (t) = Aet A u 0 + (et A − 1) f (t), 0 < t ≤ T.
If u 0 ∈ D(A), then (6.16)(ii) holds also for t = 0.

Let us show that Au 1 is Hölder continuous in [0,T ]. For 0 ≤ s ≤ t ≤ T

s

Au 1 (t) − Au 1 (s) = A e(t−σ )A − e(s−σ )A ( f (σ ) − f (s))dσ
0
+ (et A − e(t−s)A )( f (s) − f (t))
t
+ Ae(t−σ )A ( f (σ ) − f (t))dσ,
s
t−σ (6.17)
so that, since A(e(t−σ )A − e(s−σ )A ) = s−σ A2 eτ A dτ ,
s t−σ
Au 1 (t) − Au 1 (s) ≤ M2 (s − σ )α τ −2 dτ dσ [ f ]C α
0 s−σ
+ 2M0 (t −s)α [ f ]C α
t
+ M1 (t −σ )α−1 dσ [ f ]C α
s
s t−σ (6.18)
≤ M2 dσ τ α−2 dτ [ f ]C α
0 s−σ
+ (2M0 + M1 α −1 )(t − s)α [ f ]C α

M2 M1
≤ + 2M0 + (t − s)α [ f ]C α .
α(1 − α) α
Therefore, Au 1 is α-Hölder continuous in [0, T ]. Moreover, Au 2 is obvi-

ously α-Hölder continuous in [ε, T ], for each ε ∈ (0, T ). Summing up,
Au ∈ C α ([ε, T ]; X). Since u ∈ C α ([ε, T ]; X) by Proposition 6.8 and
by the analyticity of t → et A u 0 in (0, ∞), then u ∈ C α ([ε, T ]; D(A)).
Since t → w(t) := u(t + ε) is the mild solution to w = Aw + f (· + ε),
w(0) = u(ε) in the interval [0, T − ε], by Lemma 6.9 w ∈ C 1 ([0, T −
ε]; X) and w (t) = Aw(t) + f (t + ε) for 0 ≤ t ≤ T − ε, so that
u ∈ C 1+α ([ε, T ]; X) for each ε ∈ (0, T ), and u (t) = Au(t) + f (t) for
0 < t ≤ T.
To prove statements (i) and (ii) we have to study the behavior of u near
t = 0.
If u 0 ∈ D(A) we have
Au 2 (t) = et A (Au 0+ f (0)) + et A ( f (t) − f (0)) − f (t), 0 ≤ t ≤ T, (6.19)
so that if Au 0 + f (0) ∈ D(A) then Au 2 is continuous also at t = 0, and

statement (i) follows arguing as before with ε = 0.
In the case where Ax + f (0) ∈ D A (α, ∞), from (6.19) we get, for
0 ≤ s ≤ t ≤ T,
Au 2 (t)− Au 2 (s) ≤ (et A − es A )(Au 0 + f (0))
+ (et A − es A )( f (s) − f (0))
+ (et A − 1)( f (t) − f (s))

t
≤ Aeσ A L(D A (α,∞),X) dσ Au 0 + f (0) D A (α,∞)
s
t

+ s A e dσ
α
σA
[ f ]C α +(M0 +1)(t −s)α [ f ]C α
s L(X)
M1,α
≤ Au 0 + f (0) D A (α,∞) (t − s)α
α

M1
+ + M0 + 1 (t − s)α [ f ]C α ,
α
(6.20)
so that also Au 2 is Hölder continuous, and the estimate
uC 1+α ([0,T ];X) + AuC α ([0,T ];X)
≤ C( f C α ([0,T ];X) + u 0 D(A) + Au 0 + f (0) D A (α,∞) )
follows easily.
The last statement, that u is bounded with values in D A (α, ∞), fol-
lows from the embedding of Exercise 5(i), Section 1.4.3, that implies
also that the condition Au 0 + f (0) ∈ D A (α, ∞) is necessary for u ∈
C 1+α ([0, T ]; X) ∩ C α ([0, T ]; D(A)).
Theorem 6.11. Let 0 < α < 1, u 0 ∈ X, f ∈ C([0, T ]; X) ∩ B([0, T ];

D A (α,∞)), and let u be the mild solution of (6.2).
Then u ∈ C 1 ((0, T ]; X) ∩ C((0, T ]; D(A)), u (t) = Au(t) + f (t) for
0 < t ≤ T , and u ∈ B([ε, T ]; D A (α + 1, ∞)) for every ε ∈ (0, T ).
Moreover, the following statements hold.
(i) if u 0 ∈ D(A), Au 0 ∈ D(A), then u ∈ C 1 ([0, T ]; X)∩C([0, T ];D(A));

(ii) if u 0 ∈ D A (α + 1, ∞), then u and Au belong to C([0, T ]; X) ∩
B([0, T ]; D A (α, ∞)), Au belongs to C α ([0, T ]; X), and there is C
such that
u B([0,T ];D A (α,∞)) + Au B([0,T ];D A (α,∞)) + AuC α ([0,T ];X)
(6.21)
≤ C( f B([0,T ];D A (α,∞)) + u 0 D A (α+1,∞) ).
(iii) if in addition f ∈ C([0, T ]; D A (α)) and u 0 ∈ D A (α + 1), then u ∈

C 1 ([0, T ]; D A (α)) ∩ C([0, T ]; D A (α + 1)), and u ∈ h α ([0, T ]; X).
Proof. Let us consider the function v defined by (6.4). We are going to

show that it solves
v (t) = Av(t) + f (t), 0 < t ≤ T, v(0) = 0, (6.22)
and moreover that v , Av ∈ B([0, T ]; D A (α, ∞)), Av ∈ C α ([0, T ]; X),

and there is C such that
v B([0,T ];D A (α,∞)) + Av B([0,T ];D A (α,∞)) + AvC α ([0,T ];X)
(6.23)
≤ C f B([0,T ];D A (α,∞)) .
Since Ae(t−s)A L(D A (α,∞),X) ≤ M1,α (t − s)1−α for 0 ≤ s < t ≤ T

(estimates (6.6)), then s → Ae(t−s)A f (s) is in L 1 ((0, t); D(A), so that
v(t) belongs to D(A), and
t
Av(t) ≤ M1,α (t − s)α−1 ds f B(D A (α,∞))
0
(6.24)
α
T M1,α
= f B(D A (α,∞)) .
α
Moreover, for 0 < ξ ≤ 1,

t

ξ 1−α
Ae ξA
Av(t) = ξ 1−α A e2 (t+ξ −s)A
f (s)ds

0
t
≤ M2,α ξ 1−α (t + ξ − s)α−2 ds f B([0,T ];D A (α,∞))
0
M2,α
≤ f B([0,T ];D A (α,∞)) ,
1−α
(6.25)
that Av is bounded with values in D A (α, ∞). Let us show that Av is

Hölder continuous with values in X: for 0 ≤ s ≤ t ≤ T we have
s

(t−σ )A

Av(t)− Av(s) ≤ A e −e (s−σ )A
f (σ )dσ
0 t

+A e
(t−σ )A
f (σ )dσ
s
s t−σ
≤ M2,α dσ τ α−2 dτ f B([0,T ];D A (α,∞)) (6.26)
0 s−σ
t
+ M1,α (t − σ )α−1 dσ f B([0,T ];D A (α,∞))
s

M2,α M1,α
≤ + (t −s)α f B([0,T ];D A (α,∞)) ,
α(1−α) α
so that Av is α-Hölder continuous in [0, T ]. Estimate (6.23) follows
now from (6.24), (6.25), (6.26). Moreover Lemma 6.9 implies that v ∈
C 1 ([0, T ]; X) and that it solves (6.22).
Let us consider now the function t → et A u 0 .
If u 0 ∈ D(A) and Au 0 ∈ D(A) then it belongs to C 1 ([0, T ]; X) ∩
C([0, T ]; D(A)). If u 0 ∈ D A (α + 1, ∞), it belongs to B([0, T ]; D A (α +
1, ∞)) ∩ C α ([0, T ]; D(A)). Summing up, statements (i) and (ii) follow.
Let us prove statement (iii). Let f n ∈ C([0, T ]; D(A)) and xn ∈ D(A2 ),
n ∈ N, be such that
lim f n = f in C([0, T ]; D A (α, ∞)), lim xn = u 0 in D A (α, ∞).

n→∞ n→∞
By statement (ii), the solution u n of problem

u n (t) = Au n (t) + f n (t), 0 ≤ t ≤ T, u n (0) = xn ,
belongs to B([0, T ]; D A (α + ε + 1, ∞)) ∩ C([0, T ]; D(A)), and u n be-

longs to C α+ε ([0, T ]; X) for every ε ∈ (0, 1 − α). By Exercise 4(i), Sec-
tion 1.4.3, u n ∈ C([0, T ]; D A (α + 1)); moreover u n ∈ h α ([0, T ]; X). By
estimate (6.21), u n goes to u and u n goes to u in B([0, T ]; D A (α+1, ∞))
and in C α ([0, T ]; X), respectively, as n → ∞. Since C([0, T ]; D A (α +
1)) and h α ([0, T ]; X) are closed in B([0, T ]; D A (α + 1, ∞)) and in
C α ([0, T ]; X), respectively, then u ∈ C([0, T ]; D A (α + 1)) and u ∈
h α ([0, T ]; X).
Theorems 6.10(ii) and 6.11(ii)-(iii) are examples of maximal regular-
ity results. In fact, they are the most important maximal regularity results
in spaces of continuous functions. Maximal L p regularity has been con-
sidered in Section 4.2.2. However, the most precise result about maximal
L p regularity, due to Weis [91], has not been obtained using interpolation
or powers of operators. We quote it for completeness.
Theorem 6.12. Let X be a UMD space, and let A : D(A) → X be

a sectorial operator. Let 1 < p < ∞. The following statements are
equivalent:
(i) for each f ∈ L p ((0, T ); X), the mild solution v to (6.22) belongs to
W 1, p ((0, T ); X) ∩ L p ((0, T ); D(A));
(ii) there are ω ∈ R, θ ≥ π/2 such that family of operators {(λ −
ω)R(λ, A) : λ ∈ Sθ,ω } is R-bounded.
We recall that a subset L of L(X) is R-bounded (Rademacher bounded) if

there is C > 0 such that for all n ∈ N, L 1 , . . . , L n ∈ L and x1 , . . . , xn ∈
X we have

n n
r T x ≤ C r x ,
k k k 2 k k 2
k=1 L ((0,1);X) k=1 L ((0,1);X)
where rk (t) := sign(sin 2k πt) are the Rademacher functions in (0, 1).
Some remarks are in order.
(i) It has been shown by different authors with different methods (e.g.,
[7, 19, 86]) that if X is any Banach space and A is any sectorial
operator, then maximal L p regularity for (6.22) for some p ∈ (1, ∞)
implies maximal L p regularity for every p ∈ (1, ∞);
(ii) if X = H is a Hilbert space, then every sectorial operator satisfies
condition (ii) of Theorem 6.12;
(iii) condition (ii) of Theorem 6.12 is equivalent to R-boundedness of the
set {e−ωz e z A : z ∈ S0,δ } for some δ > 0, and to R-boundedness of
the set {e−ωt et A , te−ωt Aet A : t > 0}.
A detailed account of the theory of maximal L p regularity is in Chapter 2

of the lecture notes [31]. If X = H is a Hilbert space, maximal L 2 regu-
larity for (6.22) may be proved directly by Fourier transform arguments,
without the theory of powers of operators or Theorem 6.12. This has been
done several years ago in [34].
6.3.1. Exercises
1. Show that C([0, T ], D(A)) is dense in C([0, T ], D A (α)). (This has
been used in the proof of Theorem 6.11(iii)).
2. ([86]) Let A be the generator of a semigroup in a Banach space X.
Assume that for some p > 1 there is maximal L p regularity for (6.2)
(i.e., for every u 0 ∈ D A (1 − 1/ p, p) and f ∈ L p ((0, T ); X) the
mild solution of (6.2) belongs to L p ((0, T ); D(A)). Prove that A

is sectorial. (Hint: for x ∈ X and Re λ large consider the function
u(t) := eλt R(λ, A)x and from maximal L p regularity deduce that
AR(λ, A)x ≤ cx. )
Show that if there is maximal regularity in L p ((0, T ); X) for 1 ≤
p ≤ ∞ or in C([0, T ]; X) for problem (6.22), then A is sectorial.
3. Let f ∈ L ∞ ((0, T ); X). Using Proposition 6.4, show that the func-
tion v defined by (6.4) is bounded with values in (X, D(A2 ))1/2,∞ .
6.4. An application: space – time Hölder regularity

in parabolic PDE’s
Consider the problem

u t (t, x) = u(t, x) + f (t, x), 0 ≤ t ≤ T, x ∈ Rn ,
(6.27)
u(0, x) = u 0 (x), x ∈ Rn ,
where f and u 0 are continuous and bounded functions. We read it as

an abstract Cauchy problem of the type (6.2) in the space X = C b (Rn ),
setting u(t) = u(t, ·) and f (t) = f (t, ·). A is the realization of the
Laplace operator in X, and it is the generator of the Gauss-Weierstrass
analytic semigroup (5.3). Although the domain of A contains properly
Cb2 (Rn ), from Example 5.16 we know that for 0 < θ < 1, θ = 1/2,
D A (θ, ∞) = Cb2θ (Rn ), D A (θ + 1, ∞) = Cb2θ+2 (Rn ).
For every f : [0, T ] × Rn → C set f (t) = f (t, ·), 0 ≤ t ≤ T . The

following statements are easy to be checked.
(i) f : [0, T ] → X is continuous iff f is continuous, bounded, and for

every t0 ∈ [0, T ] limt→t0 supx∈Rn | f (t, x) − f (t0 , x)| = 0;
(ii) if 0 < α < 1, then f ∈ C α ([0, T ]; X) if and only if f is continu-
ous and bounded, and moreover sups=t, x∈Rn | f (t, x) − f (s, x)|/|t −
s|α < ∞;
(iii) if 0 < α < 1, α = 1/2, f ∈ B([0, T ]; D A (α, ∞)) iff f (t, ·) ∈
Cb2α (Rn ) for every t, and sup0≤t≤T f (t, ·)C 2α (Rn ) < ∞.
b
So, we may apply Theorems 6.10 and 6.11.

Theorem 6.11(ii) gives an alternative proof of Theorem 5.27. The-
orem 6.10 gives maximal Hölder regularity with respect to time, and pre-
cisely
Theorem 6.13. Let 0 < α < 1, α = 1/2, and let f : [0, T ]×Rn → C be
uniformly continuous, bounded, and such that supt∈[0,T ] f (t, ·)C 2α (Rn )<
b
∞. Let u 0 ∈ Cb2α+2 (Rn ). Then problem (6.27) has a unique solution u
such that u, u t , u are uniformly continuous, bounded, and
supt∈[0,T ] u t (t, ·)C 2α (Rn ) < ∞,
b
supt∈[0,T ] Di j u(t, ·)C 2α (Rn ) < ∞,

b
supx∈Rn u(·, x)C α ([0,T ]) < ∞.

Putting together Theorems 5.27 and 6.13 we get the Ladyzhenskaja –
Solonnikov – Ural’čeva Theorem, see [66] for the original (completely
different) proof. For simplicity we consider only the case 0 < α < 1/2. It
is convenient to adopt the usual notation: we denote by C α,2α ([0, T ]×Rn )
the space of the bounded functions f such that
| f (t, x) − f (s, y)|
sup < ∞,
t =s, x = y |t − s|α + |x − y|2α
and we denote by C 1+α,2+2α ([0, T ] × Rn ) the space of the bounded func-
tions f with bounded f t , Di j f , such that f t , Di j f are in C α,2α ([0, T ] ×
Rn ). It is possible to see that this implies that the space derivatives Di f
are (1/2 + α)-Hölder continuous with respect to t, with Hölder constant
independent of x.
Theorem 6.14 (Ladyzhenskaja–Solonnikov–Ural’čeva). Let 0 < α <
1/2 and let f ∈ C α,2α ([0, T ] × Rn ), u 0 ∈ Cb2α+2 (Rn ). Then problem
(6.27) has a unique solution u ∈ C 1+α,2+2α ([0, T ] × Rn ), and there is
C > 0, independent of f and u 0 , such that
uC 1+α,2+2α ([0,T ]×Rn ) ≤ C( f C α,2α ([0,T ]×Rn ) + u 0 C 2α+2 (Rn ) ).
b
Proof. Almost all follows from putting together the results of Theorems
5.27 and 6.13. It remains to show that the space derivatives Di j u are
time α-Hölder continuous. To this aim we use the fact that Cb2 (Rn ) be-
longs to the class J1−α between Cb2α (Rn ) and Cb2+2α (Rn ) (Exercise 2,
Section 1.4.3). Moreover, since u t (t, ·)C 2α (Rn ) is bounded in [0, T ],
b
then t → u(t, ·) is Lipschitz continuous with values in Cb2α (Rn ), with
Lipschitz constant equal to sup0≤σ ≤T u t (σ, ·)C 2α (Rn ) . So, for 0 ≤ s ≤
b
t ≤ T we have
u(t, ·) − u(s, ·)Cb2 ≤ Cu(t, ·) − u(s, ·)Cα 2α u(t, ·) − u(s, ·)C1−α
2+2α
b b
≤ C((t − s) sup0≤σ ≤T u t (σ, ·)C 2α )α (2 supσ ∈[0,T ] u(σ, ·)C 2+2α )1−α
b b
≤ C (t − s)α
The proof of Theorem 6.14 works without changes for elliptic operators
A with coefficients in Cb2α (Rn ). Indeed, such operators generate ana-
lytic semigroups in Cb (Rn ) and we have D A (α, ∞) = Cb2α (Rn ), D A (α +
1, ∞) = Cb2α+2 (Rn ).
This procedure may be extended to 2m-order uniformly elliptic oper-
ators with regular and bounded coefficients in Rn or in a regular open
subset of Rn with suitable boundary conditions. See [70].
Appendix A
The Bochner integral
A.1. Integrals over measurable real sets

We recall here the few elements of Bochner integral theory that are used
in these notes. Extended treatments, with proofs, may be found in the
books [11, 39, 40, 94]. A nice treatment of the spaces W s, p (a, b; X) is in
the appendix of [35].
X is any real or complex Banach space. We consider the usual
Lebesgue measure in R, and we denote by M the σ -algebra consist-
ing of all Lebesgue measurable subsets of R. If A ⊂ R, 11 A denotes the
characteristic function of the set A.
Definition A.1. A function f : R → X is said to be simple if there
are n ∈ N, x1 , . . . , xn ∈ X, A1 , .., An ∈ M, with meas Ai < ∞ and
Ai ∩ A j = ∅ for i = j, such that
n
f = xi 11 Ai .
i=1
If I ∈ M, a function f : I → X is said to be Bochner measurable if

there is a sequence of simple functions { f n } such that
lim f n (t) = f (t) for almost all t ∈ I.
n→∞
It is easy to
see that every continuous function is measurable.
n
If f = i=1 xi 11 Ai is a simple function we set

n
f (t)dt = xi measAi . (A.1)
R i=1
Definition A.2. Let f : R → X. f is said to be Bochner integrable

if there is a sequence of simple functions { f n } converging to f almost
everywhere, such that

lim f n (t) − f (t)dt = 0.
n→∞ R

Then n → R f n (t)dt is a Cauchy sequence in X. We set

f (t)dt = lim f n (t)dt. (A.2)
R n→∞ R

Arguing as in the case X = R, one sees that R f (t)dt is independent of
the choice of the sequence { f n }. If f is defined on a measurable set, the
above definition can be extended as follows.
Definition A.3. If I ∈ M and f : I → X, f is said to be integrable
over I if the extension f˜ defined by

= f (t), if t ∈ I
f˜(t)
= 0, if t ∈
/I
is integrable. In this case we set

f (t)dt = f˜(t)dt. (A.3)
I R
If I = (a, b), with −∞ ≤ a ≤ b ≤ +∞, we set as usual

b a b
f (t)dt = f (t)dt; f (t)dt = − f (t)dt .
(a,b) a b a
A simple criterion for establishing whether a function is integrable is

stated in the following proposition.
Proposition A.4. Let I ∈ M, and let f : I → X. Then f is integrable
if and only if f is measurable and t → f (t) is Lebesgue integrable on
I . Moreover,

f (t)dt ≤ f (t)dt. (A.4)

I I
From the definition it follows easily that if Y is another Banach space

and A ∈ L(X, Y ), then for every integrable f : I → X the function
A f : I → Y is integrable, and

A f (t)dt = A f (t)dt.
I I

In particular, if ϕ ∈ X then for every integrable f : I → X the function
t → f (t), ϕ is integrable, and

f (t)dt, ϕ = f (t), ϕdt.
I I
It follows that for every couple of integrable functions f, g it holds

(λ f (t) + µg(t))dt = λ f (t) + µ g(t)dt, ∀ λ, µ ∈ C .
I I I
Another important commutativity property is the following one.
Proposition A.5. Let X, Y be Banach spaces, and let A : D(A) ⊂ X →
Y be a closed operator. Let I ∈ M, let f : I → X be an integrable
function such that f (t) ∈ D(A)
for almost all t ∈ I , and A f : I → Y is
integrable. Then the integral I f (t)dt belongs to D(A), and

A f (t)dy = A f (t)dt.
I I
A.2. L p and Sobolev spaces

On the set of all measurable functions on I we define the equivalence
relation
f ∼ g ⇐⇒ f (t) = g(t) for almost all t ∈ I. (A.5)
Definition A.6. L 1 (I ; X) is the set of all equivalence classes of integ-
rable functions f : I → X, with respect to the equivalence relation
(A.5).
Since no confusion will arise, in the sequel we shall identify the equi-
valence class [ f ] ∈ L 1 (I ; X) with the function f itself. We define a
norm on L 1 (I, X) by setting

f L 1 (I ;X) = f (t)dt. (A.6)
I
We define now the spaces L (I ; X) for p > 1.
p
Definition A.7. Let p ∈ (1, +∞], and I ∈ M. L p (I ; X) is the set of all

equivalence classes of measurable functions f : I → X,with respect to
the equivalence relation (A.5), such that t → f (t) belongs to L p (I ).
L p (I ; X) is endowed with the norm
1/ p
f L p (I ;X) = f (t) dt
p
, if p < ∞ (A.7)
I
f L ∞ (I ;X) = ess sup { f (t) : t ∈ I } (A.8)

In the following, if there is no danger of confusion, we shall write f p
instead of f L p (I ;X) .
An extension of a well known property for X = R to general X is in
the next proposition.
Proposition A.8. Let 1 ≤ p < ∞, and let I , J be two real intervals.

Then
L p (I ; L p (J ; X)) ⊂ L p (I × J, X)
if we identify any function f ∈ L p (I ;L p (J ;X)) with the function (t,s) →
f (t)(s), t ∈ I , s ∈ J .
To introduce the Sobolev space W 1, p (a, b; X) we need a lemma.
Lemma A.9. Let p ∈ [1, ∞). Then the operator
L 0 : D(L 0 ) = C 1 ([a, b]; X) → L p (a, b; X), L 0 f = f
is preclosed in L p (a,b;X), that is the closure of its graph is the graph of
a closed operator.
Definition A.10. Let L : D(L) ⊂ L p (a,b;X) be the closure of the oper-
ator L 0 defined in Lemma (A.9). We set
W 1, p (a, b; X) = D(L)
and we endow it with the graph norm. For every f ∈ W 1, p (a, b; X), L f
is said to be the strong derivative of f , and we denote it by f .
In other words, f ∈ W 1, p (a, b; X) if and only if there is a sequence
{ f n } ⊂ C 1 ([a, b]; X) such that f n → f in L p (a, b; X) and f n → g in
L p (a, b; X), and in this case g = f . Moreover we have
f W 1, p (a,b;X) = f L p (a,b;X) + f L p (a,b;X) ∀ f ∈ W 1, p (a, b; X).
Since L is a closed operator, then W 1, p (a, b; X) is a Banach space.

Let f ∈ W 1, p (a, b; X), and let { f n } ⊂ C 1 ([a, b]; X) be such that
f n → f and f n → f in L p (a, b; X). From the equality
t
f n (t) − f n (s) = f n (σ )dσ (A.9)
s
we get, integrating with respect to s in (a, b) and letting n → ∞,

b t
1
f (t) = f (s)ds + (σ − a) f (σ ) dσ , a.e. in (a, b).
b−a a a
Therefore, W 1, p (a, b; X) is continuously embedded in C([a, b]; X). Let-

ting n → ∞ in (A.9) we get also
t
f (t) − f (s) = f (σ ) dσ.
s
Sometimes it is easier to deal with weak (or distributional) derivatives,

defined as follows.
Definition A.11. Let f ∈ L p (a, b; X). A function g ∈ L 1 (a, b; X) is

said to be the weak derivative of f in (a, b) if
b b

f (t)ϕ (t) dt = − g(t)ϕ(t) dt, ∀ϕ ∈ C0∞ (a, b).
a a
It can be shown that weak and strong derivatives do coincide. More pre-
cisely, the following proposition holds.
Proposition A.12. Let f ∈ W 1, p (a, b; X). Then f is weakly differenti-

able, and f is the weak derivative of f .
Conversely, if f ∈ L p (a,b;X) admits a weak derivative g ∈ L p (a,b;X),
then f ∈ W 1, p (a, b; X), and g = f .
For 0 < α < 1 and 1 ≤ p < ∞, the fractional Sobolev space

W α, p (a, b; X) is defined as the subspace of L p (a, b; X) consisting of
the functions f such that

b b
f (t) − f (s)
[ f ]W α, p := ds dt < ∞ .
a a (t − s)1+αp
A.3. Weighted L p spaces

Let I be an interval contained in (0, +∞). For 1 ≤ p ≤ ∞ we denote
p
by L ∗ (I ) the space of the real or complex valued L p functions in I with
respect to the measure dt/t, endowed with its natural norm
+∞ 1/ p
p dt
f p
L ∗ (I ) = | f (t)| , if p < ∞,
0 t
f L ∞
∗ (I )
= ess supt∈I | f (t)|.
p
Dealing with L ∗ spaces, the Hardy-Young inequalities are often more
useful than the Hölder inequality. They hold for every positive measur-
able function ϕ : (0, a) → R, 0 < a ≤ ∞, and every α > 0, p ≥ 1. See
[54, p. 245-246].
 a t p a

 −αp ds dt 1 ds

 (i) t ϕ(s) ≤ p s −αp ϕ(s) p ,
 0 0 s t α 0 s
a a p a (A.10)



 ds dt 1 ds
 (ii) t αp ϕ(s) ≤ p s αp ϕ(s) p
0 t s t α 0 s
The measure m(dt) = dt/t is the Haar measure of the multiplicative

group R+ . So, it is invariant under multiplication: m(A) = m(λA), for
every measurable A ⊂ R+ and λ > 0, and ϕ L ∗p (0,∞) = ϕ(λ·) L ∗p (0,∞) .
p
For every α = 0 the space L ∗ (0, ∞) is invariant under the change
p
of variable t → t α , in the sense that ϕ ∈ L ∗ (0, ∞) iff t → ϕ(t α ) ∈
p
L ∗ (0, ∞), and
ϕ L ∗p (0,∞) = |α|1/ p t → ϕ(t α ) L ∗p (0,∞) .
(This is obviously true also for p = ∞, with the usual convention 1/∞ =
0). In particular, for α = −1 we get an isometry:
ϕ L ∗p (0,∞) = t → ϕ(t −1 ) L ∗p (0,∞) .
Moreover, the change of variable t → t −1 is an isometry also between

p p
L ∗ (1, ∞) and L ∗ (0, 1).
p
If X is any Banach space and 1 ≤ p ≤ ∞ the space L ∗ (I ; X) is the set
of all Bochner measurable functions f : I → X, such that t → f (t) X
p
is in L ∗ (I ). It is endowed with the norm
f L ∗p (I ;X) = t → f (t) X L ∗p (I ) .
In Chapters 1 and 3 we have used the following consequence of inequality

(A.10)(i).
Corollary A.13. Let u be a function such that t → u θ (t) = t θ u(t) be-

p
longs to L ∗ (0, a; X), with 0 < a ≤ ∞, 0 < θ < 1 and 1 ≤ p ≤ ∞.
Then also the mean value

1 t
v(t) = u(s)ds, t > 0 (A.11)
t 0
has the same property, and setting vθ (t) = t θ v(t) we have

1
vθ L ∗p (0,a;X) ≤ u θ L ∗p (0,a;X) (A.12)
1−θ
References
[1] S. AGMON: Lectures on elliptic boundary value problems, Van

Nostrand Math. Studies, Princeton, NJ (1965).
[2] S. AGMON: On the eigenfunctions and the eigenvalues of gen-
eral elliptic boundary value problems, Comm. Pure Appl. Math. 15
(1962), 119–147.
[3] S. AGMON , A. D OUGLIS , L. N IRENBERG: Estimates near the
boundary for solutions of elliptic partial differential equations sat-
isfying general boundary conditions, Comm. Pure Appl. Math. 12
(1959), 623–727.
[4] A. A LMEIDA , P. H ÄST Ö, Interpolation in variable exponent
spaces, Rev. Mat. Complut. 27 (2014), 657–676.
[5] N. BADR, Real interpolation of Sobolev spaces associated to a
weight. Potential Anal. 31 (2009), 345–374.
[6] B. B EAUZAMY: Espaces d’interpolation réels: topologie et géo-
metrie, Springer Verlag, Berlin (1978).
[7] A. B ENEDEK , A.P. C ALDER ÓN , R. PANZONE: Convolution oper-
ators on Banach space valued functions, Proc. Nat. Acad. Sci. USA
48 (1962), 356–365.
[8] J. B ERGH , J. L ÖFSTR ÖM: Interpolation Spaces. An Introduction,
Springer Verlag, Berlin (1976).
[9] M. B ERTOLDI , L. L ORENZI: Estimates for the derivatives for
parabolic problems with unbounded coefficients, Trans. Amer.
Math. Soc. 357 (2005), 2627–2664.
[10] J. B OURGAIN: Some remarks on Banach spaces in which the mar-
tingale differences are unconditional, Arkiv Mat. 21 (1983), 163–
168.
[11] H. B R ÉZIS: Operateurs maximaux monotones et semi-groupes de
contractions dans les espaces de Hilbert, North-Holland Mathemat-
ics Studies 5, Amsterdam (1973).
[12] H. B R ÉZIS: Analyse Fonctionnelle, Masson, Paris (1983).
[13] Y U . B RUDNYI , N. K RUGLJAK: Interpolation functors and inter-

polation spaces, North-Holland, Amsterdam (1991).
[14] D. L. B URKHOLDER: A geometrical characterization of Banach
spaces in which martingale difference sequences are unconditional,
Ann. Probability 9 (1981), 997–1011.
[15] D. L. B URKHOLDER: A geometric condition that implies the exist-
ence of certain singular integrals of Banach space valued functions,
in: “Conference on Harmonic Analysis in Honour of Antoni Zyg-
mund, Chicago 1981”, W. Beckner, A.P. Calderón, R. Fefferman,
P.W. Jones Eds., 270–286, Belmont, Cal. 1983, Wadsworth.
[16] P. L. B UTZER , H. B ERENS: Semigroups of Operators and Approx-
imation, Springer Verlag, Berlin (1967).
[17] F. C OBOS , T. K ÜHN , T. S CHONBEK: One-sided compactness res-
ults for Aronszajn- Gagliardo functors, J. Funct. Anal. 106 (1992),
274–313.
[18] A. P. C ALDER ÓN: Intermediate spaces and interpolation, the com-
plex method, Studia Math. 24 (1964), 113–190.
[19] P. C ANNARSA , V. V ESPRI: On maximal L p regularity for the ab-
stract Cauchy problem, Boll. Un. Mat. Ital. B (6) 5 (1986), 165–175.
[20] S. C AMPANATO: Equazioni ellittiche del I I o ordine e spazi L2,λ ,
Ann. Mat. Pura Appl. 69 (1965), 321–381.
[21] S. C AMPANATO , G. S TAMPACCHIA: Sulle maggiorazioni in L p
nella teoria delle equazioni ellittiche, Boll. U.M.I. 20 (1965), 393–
399.
[22] P. C ANNARSA , G. DA P RATO: Infinite dimensional elliptic equa-
tions with Hölder continuous coefficients, Adv. in Diff. Eqns. 1
(1996), 425–452.
[23] P. C ANNARSA , G. DA P RATO: Schauder estimates for Kolmogorov
equations in Hilbert spaces, in: “Progress in elliptic and parabolic
partial differential equations”, A. Alvino, P. Buonocore, V. Ferone,
E. Giarrusso, S. Matarasso, R. Toscano and G. Trombetti (Editors),
Research Notes in Mathematics 350, Pitman, 100–111, (1996).
[24] S. C ERRAI: Second Order PDE’s in Finite and Infinite Dimension,
Lecture Notes in Math. 1762, Springer-Verlag, Berlin (2001)
[25] R. R. C OIFMAN , G. W EISS: Transference methods in analysis,
Conference Board of the Mathematical Sciences, Regional Confer-
ence Series in Mathematics 31, AMS, Providence (1977).
[26] M. C WIKEL: On (L p0 (A0 ), L p1 (A1 ))θ,q , Proc. Amer. Math. Soc. 44
(1974), 286–292.
[27] M. C WIKEL: Real and complex interpolation and extrapolation of
compact operators, Duke Math. J. 65 (1992), 333–343.
[28] M. C WIKEL , M. M ASTYŁO: Interpolation spaces between the

Lipschitz class and the space of continuous functions, Proc. Amer.
Math. Soc. 124 (1996), 1103–1109.
[29] M. C WIKEL , S. JANSON: Complex interpolation of compact oper-
ators: an update, Proc. Estonian Acad. Sci. Phys. Math. 55 (2006),
164–169.
[30] G. DA P RATO , P. G RISVARD: Sommes d’opérateurs linéaires et
équations différentielles opérationelles, J. Maths. Pures Appliquées
54 (1975), 305–387.
[31] G. DA P RATO , P.C. K UNSTMANN , I. L ASIECKA , A. L UNARDI ,
R. S CHNAUBELT, L. W EIS: Functional analytic methods for evolu-
tion equations. Edited by M. Iannelli, R. Nagel and S. Piazzera. Lec-
ture Notes in Mathematics 1855, Springer-Verlag, Berlin (2004).
[32] G. DA P RATO , J. Z ABCZYK: Second Order Partial Differential
Equations in Hilbert Spaces, London Math. Soc. Lecture Notes
Series 293, Cambridge Univ. Press, Cambridge (2002).
[33] E. B. DAVIES: Heat kernels and spectral theory, Cambridge Univ.
Press, Cambridge (1990).
[34] L. D E S IMON: Un’applicazione della teoria degli integrali sin-
golari allo studio delle equazioni differenziali lineari astratte del
primo ordine (Italian) Rend. Sem. Mat. Univ. Padova 34 (1964),
205–223.
[35] G. D I B LASIO: Linear Parabolic Evolution Equations in L p -
Spaces, Ann. Mat. Pura Appl. 138 (1984), 55–104.
[36] L. D IENING , P. H ARJULEHTO , P. H ÄST Ö , M. R Ů ŽI ČKA, Le-
besgue and Sobolev Spaces with Variable Exponents, Lecture Notes
in Mathematics, 2017, Springer-Verlag, Berlin (2011).
[37] G. D ORE: H ∞ functional calculus in real interpolation spaces, Stu-
dia Math. 137 (1999), 161–167; H ∞ functional calculus in real in-
terpolation spaces, II, Studia Math. 145 (2001), 75–83.
[38] G. D ORE , A. V ENNI: On the closedness of the sum of two closed
operators, Math. Z. 186 (1987), 189–201.
[39] N. D UNFORD , J. S CHWARTZ: Linear Operators, Wiley-Intersci-
ence, New York (1958).
[40] J. D IESTEL , J.J. U HL: Vector Measures, Mathematical Surveys and
Monographs 15, Amer. Math. Soc., Providence (1977).
[41] K.-J. E NGEL , R. NAGEL: One-parameter semigroups for lin-
ear evolution equations, Graduate Texts in Mathematics, vol. 194,
Springer-Verlag, New York (2000).
[42] K.-J. E NGEL , R. NAGEL, A short course on operator semigroups,
Universitext. Springer-Verlag, New York (2006).
[43] G. B. F OLLAND: Subelliptic estimates and function spaces on nil-

potent Lie groups, Ark. Mat. 13 (1975), 161–207.
[44] A. F RIEDMAN: Partial Differential Equations, Holt, Rinehart &
Winston, New York (1969).
[45] B. G AVEAU: Principe de moindre action, propagation de la chaleur
et estimèes sous elliptiques sur certains groupes nilpotents, Acta
Math. 139 (1977), 95–153.
[46] N. E. G RETSKY, J.J. U HL: Bounded linear operators on Banach
function spaces of vector-valued functions, Trans. Amer. Math. Soc.
167 (1972), 263–277.
[47] P. G RISVARD: Commutativité de deux functeurs d’interpolation et
applications, J. Maths. Pures Appl. 45 (1966), 143–290.
[48] P. G RISVARD: Équations differentielles abstraites, Ann. Scient. Éc.
Norm. Sup., 4e série 2 (1969), 311-395.
[49] P. G RISVARD: Caractérisation de quelques espaces d’interpo-
lation, Arch. Rat. Mech. Anal. 25 (1967), 40–63.
[50] D. G UIDETTI: On interpolation with boundary conditions, Math.
Z. 207 (1991), 439–460.
[51] M. H AASE, The functional calculus for sectorial operators, Oper-
ator Theory: Advances and Applications, 169. Birkhäuser Verlag,
Basel (2006).
[52] M. H AASE, Functional analysis. An elementary introduction,
Graduate Studies in Mathematics, 156. American Mathematical So-
ciety, Providence, RI, (2014).
[53] R. H ANKS: Interpolation by the real method between B M O, L α
(0 < α < ∞ and H α (0α < ∞), Indiana Math. J. 26 (1977),
569–645.
[54] G. H. H ARDY, J.E. L ITTLEWOOD , G. P ÒLYA: Inequalities, Cam-
bridge Univ. Press, Cambridge (1934).
[55] L. H ÖRMANDER: Hypoelliptic differential equations of second or-
der, Acta Math. 119 (1967), 147–171.
[56] R. H OWE: On the role of the Heisenberg group in harmonic ana-
lysis, Bull. Amer. Math. Soc. 32 (1980), 821–843.
[57] S. G. K RANTZ: Lipschitz spaces on stratified Lie groups, Trans.
Amer. Math. Soc. 269 (1982), 29–66.
[58] F. J OHN , L. N IRENBERG: On functions of bounded mean oscilla-
tion, Comm. Pure Appl. Math. 14 (1961), 415–426.
[59] T. K ATO: Fractional powers of dissipative operators, Proc. Math.
Soc. Japan 13 (1961), 246–274.
[60] T. K ATO: Fractional powers of dissipative operators, II, Proc.
Math. Soc. Japan 14 (1962), 242–248.
[61] T. K ATO: remarks on pseudo-resolvents and infinitesimal generat-

ors of semigroups, Proc. Japan Acad. 35 (1959), 467–468.
[62] H. K EMPKA , J. V YB ÍRAL, Lorentz spaces with variable exponents,
Math. Nachr. 287 (2014), 938–954.
[63] S. G. K REIN , Y U . P ETUNIN , E.M. S EMENOV: Interpolation of
linear operators, Amer. Math. Soc., Providence (1982).
[64] S. N. K RU ŽHKOV, A. C ASTRO , M. L OPES: Schauder type estim-
ates and theorems on the existence of the solution of fundamental
problem for linear and nonlinear parabolic equations, Dokl. Akad.
Nauk SSSR 20 (1975), 277–280 (Russian). English transl.: Soviet
Math. Dokl. 16 (1975), 60–64.
[65] S. N. K RU ŽHKOV, A. C ASTRO , M. L OPES: Mayoraciones de
Schauder y teorema de existencia de las soluciones del problema
de Cauchy para ecuaciones parabolicas lineales y no lineales, (I)
Ciencias Matemáticas 1 (1980), 55–76; (II) Ciencias Matemáticas
3 (1982), 37–56.
[66] O. A. L ADYZHENSKAJA , V. A. S OLONNIKOV, N. N. U RAL’ ČEVA:
Linear and quasilinear equations of parabolic type, Nauka, Mos-
kow 1967 (Russian). English transl.: Transl. Math. Monographs,
AMS, Providence (1968).
[67] Q. L AI: The Sharp Maximal Function on Spaces of Generalized
Homogeneous Type, J. Funct. Anal. 150 (1997), 75–100.
[68] J. L. L IONS: Théoremes de traces et d’interpolation, (I),. . .,(V); (I),
(II) Ann. Sc. Norm. Sup. Pisa 13 (1959), 389–403; 14 (1960), 317–
331; (III) J. Maths. Pures Appl. 42 (1963), 196–203; (IV) Math.
Annalen 151 (1963), 42–56; (V) Anais de Acad. Brasileira de Cien-
cias 35 (1963), 1–110.
[69] J. L. L IONS , J. P EETRE: Sur une classe d’espaces d’interpolation,
Publ. I.H.E.S. 19 (1964), 5–68.
[70] A. L UNARDI , E. S INESTRARI , W. VON WAHL A semigroup ap-
proach to the time dependent parabolic initial boundary value prob-
lem, Diff. Int. Eqns. 5 (1992), 1275–1306.
[71] A. L UNARDI: Analytic semigroups and optimal regularity in para-
bolic problems, Birkhäuser, Basel (1995).
[72] A. L UNARDI: Schauder estimates for a class of degenerate el-
liptic and parabolic operators with unbounded coefficients, Ann.
Sc. Norm. Sup. Pisa 24 (1997), 133–164.
[73] A. L UNARDI: Schauder theorems for linear elliptic and parabolic
problems with unbounded coefficients in Rn , Studia Math. 128 (2)
(1998), 171–198.
[74] A. L UNARDI: Regularity for a class of sums of noncommuting op-
erators, in: Topics in Nonlinear Analysis, The Herbert Amann An-
niversary volume, J. Escher, G. Simonett Eds., Birkhäuser Verlag,

Basel (1999), 517–533.
[75] A. L UNARDI: Optimal W s, p regularity in parabolic problems, Atti
Acc. Naz. Lincei, Rendiconti, Serie IX, 10 (1999), 25–34.
[76] A. L UNARDI: On generators of noncommuting semigroups: sums,
interpolation, regularity, in: “Evolution Equations, Semigroups and
Functional Analysis”, in Memory of Brunello Terreni, A. Lorenzi,
B. Ruf Eds., Birkhäuser Verlag, Basel (2002), 263–277.
[77] A. L UNARDI , G. M ETAFUNE: On the domains of elliptic operators
in L 1 , Diff. Int. Eqns. 17 (2004), no. 1-2, 73–97.
[78] B. S. M ITJAGIN , E.M. S EMENOV: C k is not an interpolation space
between C and C n , 0 < k < n, Dokl. Akad. Nauk SSSR 228
(1976). English transl.: Soviet Math. Dokl. 17 (1976).
[79] E. P RIOLA: On a class of Markov type semigroups in spaces of uni-
formly continuous and bounded functions, Studia Math. 136 (1999),
271–295.
[80] F. R IESZ , B. S Z . NAGY: Functional Analysis, Dover, New York
(1990).
[81] K. S AKA: Besov spaces and Sobolev spaces on a nilpotent Lie
group, Tohoku Math. J. 31 (1979), 383–437.
[82] R. S EELEY: Norms and domains of the complex powers A zB , Amer.
J. Math. 93 (1971), 299–309.
[83] W. S ICKEL , L. S KRZYPCZAK , J. V YB ÍRAL, Complex interpola-
tion of weighted Besov and Lizorkin-Triebel spaces, Acta Math. Sin.
(Engl. Ser.) 30 (2014), 1297–1323.
[84] W. S ICKEL , L. S KRZYPCZAK , J. V YB ÍRAL, Complex interpola-
tion of weighted Besov and Lizorkin-Triebel spaces (long version),
unpublished manuscript, http://arxiv.org/abs/1212.1614.
[85] E. S INESTRARI: On the abstract Cauchy problem in spaces of con-
tinuous functions, J. Math. Anal. Appl. 107 (1985), 16–66.
[86] P. E. S OBOLEVSKII: Coerciveness inequalities for abstract para-
bolic equations (Russian), Dokl. Akad. Nauk SSSR 157 (1964), 52–
55. English translation: Soviet Math. Dokl. 5 (1964), 894-897.
[87] P. E. S OBOLEVSKII: Fractional powers of coercively positive sums
of operators (Russian) Dokl. Akad. Nauk SSSR 225 (1975), 1271–
1274. English translation: Soviet Math. Dokl. 16 (1975), 1638–
1641.
[88] H. B. S TEWART: Generation of analytic semigroups by strongly
elliptic operators, Trans. Amer. Math. Soc. 199 (1974), 141–162.
[89] L. TARTAR: An introduction to Sobolev spaces and interpola-
tion spaces, Lecture Notes of the Unione Matematica Italiana 3,
Springer, Berlin; UMI, Bologna (2007).
[90] H. T RIEBEL: Interpolation Theory, Function Spaces, Differential

Operators, North-Holland, Amsterdam (1978).
[91] L. W EIS: Operator-valued Fourier multiplier theorems and max-
imal L p -regularity, Math. Ann. 319 (2001), 735–758.
[92] J. Z ABZCYK: Mathematical Control Theory: An Introduction,
Birkhäuser Verlag, Boston (1992).
[93] F. Z IMMERMANN: On vector valued Fourier multiplier theorems,
Studia Math. 93 (1989), 201–222.
[94] K. YOSIDA: Functional Analysis, Springer Verlag, Berlin (1965).
Index
h θ (Rn ), 18 Hardy-Young inequalities, 185

C 0,α ([0, T ] × Rn ), 157 Hausdorff–Young theorem, 58
C0∞ (Rn ), 58 Hilbert transform, 108
Cbθ (), 12 Hille–Yosida Theorem, 129
C0 (Rn ), 18 Hölder spaces, 7, 37
H s, p (Rn ), 57
Jθ (X, Y ), 31 infinitesimal generator, 129
K (t, x), 1 intermediate space, ix
K (t, x, X, Y ), 1 interpolation couple, ix
K θ (X, Y ), 31 interpolation space, ix
L p -multiplier, 111 little-Hölder continuous functions,
V ( p, θ, Y, X), 20 18
W θ, p (), 19 Lorentz spaces, 15, 38–40
BMO(), 40
BUC(Rn ), 11 m-accretive operators, 116
BUC1 (Rn ), 11 Marcinkiewicz interpolation the-
Lip(Rn ), 11 orem, 39
R-boundedness, 176 maximal regularity, 83, 113, 170
ζ -convexity, 108
nonnegative operators, 95
accretive operators, 116
analytic semigroups, 129 positive operator, 85
Balakrishnan formula, 91 ray of minimal growth, 63

Besov spaces, 12 rearrangements, 14
Bochner integrable function, 181 reflection method, 13
Bochner measurable function, 181 Riesz–Thorin theorem, 45, 58
Dirichlet problem, 41 sectorial operators, 130

semigroup, 129
Fourier transform, 58, 111 simple function, 46, 181
fractional Sobolev spaces, 8, 38 spectral decomposition, 124
195
Stampacchia interpolation theorem, traces of W 1, p functions, 23

40 truncated Hilbert transform, 108
Stein interpolation theorem, 59,
165 UMD spaces, 108
strong derivative, 184 Yosida approximations, 120
strongly continuous semigroup,
129
LECTURE NOTES
This series publishes polished notes dealing with topics of current re-
search and originating from lectures and seminars held at the Scuola Nor-
male Superiore in Pisa.
Published volumes
1. M. T OSI , P. V IGNOLO, Statistical Mechanics and the Physics of Flu-
ids, 2005 (second edition). ISBN 978-88-7642-144-0
2. M. G IAQUINTA , L. M ARTINAZZI , An Introduction to the Regularity
Theory for Elliptic Systems, Harmonic Maps and Minimal Graphs,
2005. ISBN 978-88-7642-168-8
3. G. D ELLA S ALA , A. S ARACCO , A. S IMIONIUC , G. T OMASSINI ,
Lectures on Complex Analysis and Analytic Geometry, 2006.
ISBN 978-88-7642-199-8
4. M. P OLINI , M. T OSI , Many-Body Physics in Condensed Matter Sys-
tems, 2006. ISBN 978-88-7642-192-0
P. A ZZURRI, Problemi di Meccanica, 2007. ISBN 978-88-7642-223-2
5. R. BARBIERI, Lectures on the ElectroWeak Interactions, 2007. ISBN
978-88-7642-311-6
6. G. DA P RATO, Introduction to Stochastic Analysis and Malliavin Cal-
culus, 2007. ISBN 978-88-7642-313-0
P. A ZZURRI, Problemi di meccanica, 2008 (second edition). ISBN 978-
88-7642-317-8
A. C. G. M ENNUCCI , S. K. M ITTER , Probabilità e informazione,
2008 (second edition). ISBN 978-88-7642-324-6
culus, 2008 (second edition). ISBN 978-88-7642-337-6
8. U. Z ANNIER, Lecture Notes on Diophantine Analysis, 2009.
ISBN 978-88-7642-341-3
9. A. L UNARDI, Interpolation Theory, 2009 (second edition).
ISBN 978-88-7642-342-0
198 Lecture notes
10. L. A MBROSIO , G. DA P RATO , A. M ENNUCCI, Introduction to Meas-

ure Theory and Integration, 2012.
ISBN 978-88-7642-385-7, e-ISBN: 978-88-7642-386-4
11. M. G IAQUINTA , L. M ARTINAZZI , An Introduction to the Regularity
Theory for Elliptic Systems, Harmonic Maps and Minimal Graphs,
2012 (second edition). ISBN 978-88-7642-442-7, e-ISBN: 978-88-7642-443-4
G. P RADISI, Lezioni di metodi matematici della fisica, 2012.
ISBN: 978-88-7642-441-0
12. G. B ELLETTINI, Lecture Notes on Mean Curvature Flow, Barriers
and Singular Perturbations, 2013.
ISBN 978-88-7642-428-1, e-ISBN: 978-88-7642-429-8
culus, 2014. ISBN 978-88-7642-497-7, e-ISBN: 978-88-7642-499-1
14. R. S COGNAMILLO , U. Z ANNIER, Introductory Notes on Valuation
Rings and Function Fields in One Variable, 2014. ISBN 978-88-7642-500-4,
e-ISBN: 978-88-7642-501-1
15. S. D IPIERRO , M. M EDINA , E. VALDINOCI, Fractional Elliptic Prob-
lems with Critical Growth in the Whole of Rn , 2017.
ISBN 978-88-7642-600-1, e-ISBN: 978-88-7642-601-8
G. P RADISI, Lezioni di metodi matematici della fisica, 2012 (reprint
2018) ISBN: 978-88-7642-441-0
16. A. L UNARDI, Interpolation Theory, 2018 (third edition).
ISBN 978-88-7642-639-1, e-ISBN: 978-88-7642-638-4
Volumes published earlier

G. DA P RATO, Introduction to Differential Stochastic Equations, 1995
(second edition 1998). ISBN 978-88-7642-259-1
L. A MBROSIO, Corso introduttivo alla Teoria Geometrica della Misura
ed alle Superfici Minime, 1996 (reprint 2000).
E. V ESENTINI, Introduction to Continuous Semigroups, 1996 (second
edition 2002). ISBN 978-88-7642-258-4
C. P ETRONIO, A Theorem of Eliashberg and Thurston on Foliations and
Contact Structures, 1997. ISBN 978-88-7642-286-7
Quantum cohomology at the Mittag-Leffler Institute, a cura di Paolo Aluf-
fi, 1998. ISBN 978-88-7642-257-7
G. B INI , C. DE C ONCINI , M. P OLITO , C. P ROCESI, On the Work of
Givental Relative to Mirror Symmetry, 1998. ISBN 978-88-7642-240-9
H. P HAM, Imperfections de Marchés et Méthodes d’Evaluation et Couver-
ture d’Options, 1998. ISBN 978-88-7642-291-1
H. C LEMENS, Introduction to Hodge Theory, 1998. ISBN 978-88-7642-268-3
Seminari di Geometria Algebrica 1998-1999, 1999.
A. L UNARDI, Interpolation Theory, 1999. ISBN 978-88-7642-296-6
199 Lecture notes
R. S COGNAMILLO, Rappresentazioni dei gruppi finiti e loro caratteri,

1999.
S. RODRIGUEZ, Symmetry in Physics, 1999. ISBN 978-88-7642-254-6
F. S TROCCHI, Symmetry Breaking in Classical Systems, 1999 (2000).
ISBN 978-88-7642-262-1
L. A MBROSIO , P. T ILLI, Selected Topics on “Analysis in Metric Spaces”,
2000. ISBN 978-88-7642-265-2
A. C. G. M ENNUCCI , S. K. M ITTER, Probabilità ed Informazione, 2000.
S. V. B ULANOV, Lectures on Nonlinear Physics, 2000 (2001).
ISBN 978-88-7642-267-6
Lectures on Analysis in Metric Spaces, a cura di Luigi Ambrosio e Fran-
cesco Serra Cassano, 2000 (2001). ISBN 978-88-7642-255-3
L. C IOTTI, Lectures Notes on Stellar Dynamics, 2000 (2001).
ISBN 978-88-7642-266-9
S. RODRIGUEZ, The Scattering of Light by Matter, 2001.
ISBN 978-88-7642-298-0
G. DA P RATO, An Introduction to Infinite Dimensional Analysis, 2001.
ISBN 978-88-7642-309-3
S. S UCCI, An Introduction to Computational Physics: – Part I: Grid
Methods, 2002. ISBN 978-88-7642-263-8
D. B UCUR , G. B UTTAZZO, Variational Methods in Some Shape Optim-
ization Problems, 2002. ISBN 978-88-7642-297-3
A. M INGUZZI , M. T OSI, Introduction to the Theory of Many-Body Sys-
tems, 2002.
S. S UCCI, An Introduction to Computational Physics: – Part II: Particle
Methods, 2003. ISBN 978-88-7642-264-5
A. M INGUZZI , S. S UCCI , F. T OSCHI , M. T OSI , P. V IGNOLO, Numer-
ical Methods for Atomic Quantum Gases, 2004. ISBN 978-88-7642-130-0

A.Lunardi - Interpolation Theory

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A.Lunardi - Interpolation Theory

Uploaded by

Copyright:

Available Formats

16

3 Interpolation and domains of operators 63

3.2.2 The sum of two commuting operators . . . . . . 75

4 Powers of positive operators 85

5 Interpolation and semigroups 129

6 Analytic semigroups and interpolation 161

A The Bochner integral 181

Let X, Y be two real or complex Banach spaces. By X = Y we mean

v X∩Y := max{v X , vY }.

Also the sum X + Y := {x + y : x ∈ X, y ∈ Y } is a linear subspace of

v X+Y := inf x X + yY .

As easily seen, X + Y is isometric to the quotient space (X × Y )/D,

X + Y  → X + Y whose restriction to X belongs to L(X) and whose

Lemma 0.1. Let (X, Y ) be an interpolation couple, and let E be an in-

T L(E) ≤ C max{T L(X) , T L(Y ) }.

Proof. The space B of the linear operators T : X + Y  → X + Y such

and taking the infimum over all the decompositions of w we get

Tn w − T w X+Y ≤ Tn − T w X+Y

so that limn→∞ Tn w = T w in X + Y . On the other hand, if w ∈ E then

The general interpolation theory is not devoted to characterize all the

Interpolation theory has a wide range of applications. We shall em-

BUC(Rn ) the space of the uniformly continuous and bounded func-

W θ, p () the fractional Sobolev spaces in an open set , page 18

Let (X, Y ) be a real or complex interpolation couple.

1.1. The K -method

If there is no danger of confusion, we shall write K (t, x) instead of

(X, Y )θ := {x ∈ X + Y : lim+ t −θ K (t, x) = lim t −θ K (t, x) = 0}. (1.3)

Such spaces are called real interpolation spaces.

Note that K (t, x, X, Y ) = t K (t −1 , x, Y, X) for each t > 0. By the

(X, Y )θ, p = (Y, X)1−θ, p , 0 < θ < 1, 1 ≤ p ≤ ∞, (1.4)

(a) Let X = Y . Then X + Y = X, and K (t, x) ≤ min{t, 1}x. There-

Proposition 1.3. For 0 < θ < 1, 1 ≤ p1 ≤ p2 ≤ ∞ we have

X ∩Y ⊂ (X, Y )θ, p1 ⊂ (X, Y )θ, p2 ⊂ (X, Y )θ ⊂ (X, Y )θ,∞ ⊂ X + Y. (1.6)

Proof. As a first step, we show that (X, Y )θ,∞ is contained in X ∩ Y and

K (t, x) = inf a X + tbY ≤ t θ xθ,∞ , t > 0,

so that for every n ∈ N (taking t = 1/n) there are an ∈ X, bn ∈ Y such

x X+Y = K (1, x) ≤ xθ,∞ , ∀x ∈ (X, Y )θ,∞ ,

so that (X, Y )θ,∞ is continuously embedded in X + Y .

xθ,∞ ≤ (min{θ, 1 − θ} p)1/ p xθ, p . (1.8)

Moreover letting t → ∞ we get limt→∞ t −θ K (t, x) = 0. To prove that

0 = lim t −(1−θ) K (t, x, Y, X)

= (xθ, p1 ) p1 / p2 (xθ,∞ )1− p1 / p2 ,

and the statement follows since (X, Y )θ2 ,∞ ⊂ X + Y = X because

Let p = ∞. Then for every t > 0 and n, m ≥ n ε

t −θ K (t, xn − x) ≤ ε + t −θ max{t, 1}xm − x X+Y .

Letting m → +∞, we find t −θ K (t, xn − x) ≤ ε for every t > 0. This

Due again to (1.12), for every δ ∈ (0, 1) we get, for n, m ≥ n ε ,

Theorem 1.6. Let (X 1 , Y1 ), (X 2 , Y2 ) be interpolation couples. If T ∈

Proof. If T is the null operator, there is nothing to prove. If T  = 0, either

so that, taking the infimum over all a, b, we get

and (1.13) follows. From (1.14) it follows also that

limt→0 t −θ K (t, x, X 1 , Y1 ) = limt→∞ t −θ K (t, x, X 1 , Y1 ) = 0 ⇒

⇒ limt→0 t −θ K (t, T x, X 2 , Y2 ) = limt→∞ t −θ K (t, T x, X 2 , Y2 ) = 0,

that is, T maps (X 1 , Y1 )θ into (X 2 , Y2 )θ .

In the papers [17, 27] it was shown that if in addition T : X 1  → X 2 , or

Corollary 1.7. Let (X, Y ) be an interpolation couple. For 0 < θ < 1,

Proof. Set K = R or K = C, according to the fact that X, Y are real

functions, endowed with the norm

It is endowed with the norm · L p + [ · ]W θ, p .

(Cb (Rn ), Cb1 (Rn ))θ,∞ = Cbθ (Rn ), (1.16)

(L p (Rn ), W 1, p (Rn ))θ, p = W θ, p (Rn ), (1.17)

f ∞ ≤ K (1, f, Cb (Rn ), Cb1 (Rn )) ≤ f θ,∞ .

Moreover for x  = y and again for every decomposition f = a + b, with

X + Y → X + Y whose restriction to X belongs to L(X) and whose

Proof. The space B of the linear operators T : X + Y → X + Y such

Proof. If T is the null operator, there is nothing to prove. If T = 0, either

limt→0 t −θ K (t, x, X 1 , Y1 ) = limt→∞ t −θ K (t, x, X 1 , Y1 ) = 0 ⇒

⇒ limt→0 t −θ K (t, T x, X 2 , Y2 ) = limt→∞ t −θ K (t, T x, X 2 , Y2 ) = 0,

In the papers [17, 27] it was shown that if in addition T : X 1 → X 2 , or

Moreover for x = y and again for every decomposition f = a + b, with

(x, 0) of the functions (x, t) → v(x, t) ∈ W 1, p (Rn+1

Therefore, v belongs to V ( p, 1p ,W 1, p (Rn ),L p (Rn )) iff (x, t) → v(t)(x)