ERGODIC PROPERTIES OF CONVOLUTION OPERATORS

JORGE GALINDO AND ENRIQUE JORDÁ

Abstract. Let G be a locally compact group and µ be a measure on

G. In this paper we find conditions for the convolution operators λp pµq,
arXiv:2004.07622v1 [math.FA] 16 Apr 2020

defined on Lp pGq and given by convolution by µ, to be mean ergodic

and uniformly mean ergodic. The ergodic properties of the operators
λp pµq are related to the ergodic properties of the measure µ as well.

1. Introduction
Von Neumann’s mean ergodic theorem proves that the sequence pTrns qn
of averages of the first n powers of a unitary operator T (its Cesàro means)
always converges to the projection onto the subspace of fixed vectors of
T . After the appearance of von Neumann’s mean ergodic theorem, and its
Birkhoff’s contemporary, the pointwise ergodic theorem, the convergence
of averages of measure preserving transforms in diverse settings and senses
rapidly became an important part of Ergodic Theory, see [17, 32] for more
details. In this development, an operator on a Banach space whose sequence
of Cesàro means converges in the strong operator topology came to be known
as mean ergodic. When convergence holds in the operator norm, the coined
term was uniformly mean ergodic.
Our approach to ergodicity can be traced back to [45] where Yosida used
weak clustering of orbits to characterize mean ergodicity of operators on
Banach spaces. Our contemporary antecedent is to be found in the work of
Bonet and Domański [7, 8] that launched a systematic research on ergodicity
from an operator theoretic point of view. Composition and multiplication
operators in spaces of holomorphic functions have since been studied [4,
5, 7, 8] with a recent attention to composition operators acting on spaces
of smooth real functions [20, 30]. In this paper, we address the ergodic
properties of those operators on the Banach space Lp pGq of p-integrable
functions on a locally compact group G that are given by convolution by a
fixed measure on G. We will refer to these operators simply as convolution
operators.
When G is sigma-compact and µ is a probability measure, the operator
given by convolution with µ is a Markov operator, having the Haar measure
mG as sigma-finite invariant measure; the Markov chain generated by µ with
state space G is a random walk on G. If G is compact or Abelian, conditions
for ergodicity and mixing of the random walk are well known, in terms of

Date: April 17, 2020.

2010 Mathematics Subject Classification. Primary 43A05; Secondary 43 A15, 43A20,
46H99, 47A35.
Key words and phrases. mean ergodic operator, uniformly mean ergodic operator, con-
volution operator, locally compact group, amenable group.
1
2 JORGE GALINDO AND ENRIQUE JORDÁ

the algebraic properties of µ (the Itô-Kawada and Choquet-Deny theorems).

See Remark 5.5 for some more information on this approach.
Among the most recent papers dealing with convolution operators from a
point of view related to ours, we can mention the monograph by Derighetti
[13], with special emphasis on their restriction and extension to and from
closed subgroups, and the papers by Neufang, Salmi, Skalski and Spronk
[38], with focus on the wider frame of quantum groups, and by Mustafayev
[36, 37], where ergodicity for multipliers on Banach algebras and convolution
operators on locally compact Abelian groups are studied.
1.1. Outline and summary of results. Our approach leads to charac-
terizations of mean ergodicity and uniform mean ergodicity for convolution
operators defined by several classes of measures. These characterizations can
also be used to compare the convergence of averages of convolution powers of
measures in the vague topology (mean ergodicity of measures) with mean er-
godicity of the corresponding convolution operators. As it turns out, while
mean ergodicity of measures often implies ergodicity of their convolution
operators, the converse does not hold even for weak versions of ergodicity.
We next outline our results, showcasing the case of Abelian groups. Our
methods however are not intrinsically commutative and most of our results
are stated for more general classes of groups. We also have also tried to
determine the limits of what can be expected beyond our results through
a series of examples, both commutative and noncommutative, that appear,
mostly, in Section 7.
Theorem A (Theorem 4.3, Theorem 5.1, Theorem 5.4). Let G be a locally
compact group and µ P M pGq. Consider the following conditions on the
convolution operator λp pµq (see Section 2 for undefined notation):
(a) λp pµq is mean ergodic.
(b) λp pµq is power bounded.
(c) }µ} ď 1.
These conditions can be related as follows.
(1) If G is Abelian and p “ 2, then (a) and (b) are equivalent.
(2) If G is amenable and µ is positive, then all three conditions are
equivalent, for any 1 ă p ă 8.
(3) If G is compact, µ positive and p “ 1, then all three conditions are
equivalent as well.
The last statement of Theorem A actually contains all what needs to be
known about the case p “ 1, for λ1 pµq cannot be mean ergodic unless either
µ is a probability measure and the support of µ is contained in a compact
subgroup of G, see Theorem 5.4, or }µ} ă 1. In this last case we have even
that pλpµn qqn is norm convergent to 0.
Theorem B (Theorem 4.6, Theorem 6.1, Theorem 6.3). Let G be a locally
compact Abelian group and let µ P M pGq. Consider the following conditions
on the convolution operator λp pµq:
(a) λp pµq is uniformly mean ergodic.
(b) kλp pµqk ď 1 and 1 is not an accumulation point of the spectrum of
λp pµq.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 3

Conditions (a) and (b) are then equivalent when any of the following condi-
tions hold:
(1) p “ 2.
(2) µ is positive and 1 ă p ă 8.
(3) G is compact, µ positive and p “ 1 or p “ 8. In either case λp pµq
is uniformly mean ergodic if and only if λ8 pµq is mean ergodic.
In Theorem B, commutativity is only needed to make sure the operator
λ2 pµq is normal. Hence the theorem is also true under this weaker assump-
tion.
Many of our results on ergodicity or mean ergodicity of λp pµq depend on
conditions on the ambient group G. To make them depend, as it is naturally
expected, on the subgroup Hµ generated by the support of µ, we have had
to relate the ergodic behaviour of λp pµq as an operator on Lp pGq with its
behaviour as an operator on Lp pHµ q. Since this is a technically intrincate
issue, we have decided to deal with it in an Appendix at the end of the
paper.

2. Preliminaries
In this section we gather the basic definitions and basic facts around our
two main subjects: ergodicity of operators and convolution operators.

2.1. Ergodic Operators.

Notation. If T P LpXq is a bounded linear operator on a Banach space X,
Trns will denote the Cesàro means:

1 ÿ k
n
Trns “ T .
n k“1

Definition 2.1. We say that a bounded linear operator T P LpXq is:

(1) Weakly mean Ergodic if there is P P LpXq such that limnÑ8 Trns “
P in the weak operator topology.
(2) Mean Ergodic if there is P P LpXq such that limnÑ8 Trns “ P in
the strong operator topology.
(3) Uniformly Mean Ergodic if there is P P LpXq such that limnÑ8 Trns “
P in the operator norm.
Notation. If T P LpXq is a bounded linear operator on a Banach space X
for which kerpT ´ Iq is a complemented subspace we will denote by PT the
projection operator onto kerpT ´ Iq.
The following two basic facts on mean ergodicity of operators can be
found in or deduced from Section 8.4 of [18].
Theorem 2.2. Let T be a bounded linear operator on a Banach space X
such that kT n k{n converges to 0 in the strong operator topology. Then
(1) T is mean ergodic if and only if X “ kerpT ´ Iq ‘ Ran I ´ T .
(2) If T is mean ergodic, then limn Trns “ PT and PT T “ PT “ T PT .
4 JORGE GALINDO AND ENRIQUE JORDÁ

We next state a general version of Yosida’s mean ergodic Theorem. We

include a proof for the sake of completeness. It does not include new ideas
but those of Yosida. See [45] for the original proof and [2, Theorem 2.4],
[18, Theorem 8.22] or [39, Theorem 1.3] for similar statements.
Theorem 2.3. Let X be a Banach space and let τ be a locally convex topol-
ogy on X compatible with σpX, X ˚ q. Then there is a bounded linear operator
P on X such that pTrns xqn is τ -convergent to P pxq for each x P X if and
only the following two conditions hold for each x P X:
n
(1) τ ´ lim T npxq “ 0 and
(2) tTrns pxq : n P Nu is relatively weakly compact.
Proof. The necessity of condition (1) comes from
T n pxq n´1
“ Trns pxq ´ Trn´1s pxq, for any x P X.
n n
Since τ is stronger than the weak topology, Condition (2) is obviously also
necessary.
We now check the sufficiency of (1) and (2). We first observe that Trns pI ´
T q “ n1 pT ´T n`1 q, hence Trns pI ´T qpxq is τ -convergent to 0 for every x P X.
Since, by the Banach Stenihaus theorem, p}Trns }qn is a bounded sequence,
we get
(1) τ ´ lim Trns pvq “ 0 for every v P RanpI ´ T q.
n
Let now x P X. As tTrns pxq : n P Nu is relatively weakly compact, we
get from Eberlein’s theorem an increasing sequence pnk q of natural numbers
and yx P X such that pTrnk s pxqq is weakly convergent to yx . Then, by the
preceding paragraph, and using that T Trnk s “ Trnk s T ,
(2) 0 “ σpX, X ˚ q ´ lim Trnk s ppI ´ T qpxqq “ yx ´ T pyx q.
k
On the other hand
˜n ¸
1 ÿ k
˚ n
x ´ yx “ σpX, X q ´ lim px ´ T pxqq
k nk
n“1

1 ´ÿ
nk
` ˘¯
“ σpX, X ˚ q ´ lim pI ´ T q pI ` T ` ¨ ¨ ¨ ` T n´1 qpxq .
k nk n“1
From this and RanpI ´ T q being a vector subspace we deduce that x ´ yx
is in the norm closure Ran I ´ T . Applying (1) once more we see that
τ ´ limn Trns px ´ yx q “ 0. From this and (2) we conclude that, putting
P x “ yx ,
lim Trns x “ P x, for every x P X.
n
Since kyx k ď supk kTrnk s kkxk, it is clear that P P LpXq. 
Corollary 2.4. (Mean ergodic theorem) If X is reflexive, the sequence
Tn
pTrns qn is bounded (in other words, T is Cesàro bounded) and lim “ 0 in
n n
the strong operator topology, then T is mean ergodic.
Proposition 2.5. If T is Cesàro bounded then rpT q ď 1.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 5

Proof. If rpT q ą 1 then there is M ą 1 such that, eventually, kT n k ě M n ,

n n
which makes the sequence p Tn qn unbounded. Since Tn “ Trns ´ n´1 n Trn´1s ,
this goes against Cesàro-boundedness of T . 
Corollary 2.6. Let X be a Banach space. If T is weakly mean ergodic, then
rpT q ď 1.
And, next, we collect some basic facts on uniform mean ergodicity. (1) is
an immediate consequence of the spectral radius formula, (2) comes again
Tn
from Trns ´ n´1
n Trn´1s “ n and (3) is the classical Yosida Kakutani theorem
[46, Theorem 4, Corollary (i)].
Theorem 2.7. Let T be a bounded linear operator on a Banach space X:
(1) If rpT q ă 1 then T is uniformly mean ergodic (further, T n converges
to 0 in the norm operator topology).
kT n k
(2) If T is uniformly mean ergodic, then lim “ 0.
nÑ8 n
(3) If T is power bounded, compact and mean ergodic then it is uniformly
mean ergodic.
Uniform mean ergodicity can be characterized in a functional analytic
way.
Theorem 2.8 (Dunford, Theorem 8 of [16], Lin, Theorem of [33]). Let T be
a bounded linear operator on the Banach space X. The following assertions
are equivalent:
(1) T is uniformly mean ergodic.
n
(2) RanpI ´ T q2 is closed and limn }Tn } “ 0.
(3) Either 1 P ̺pT q or 1 is a pole of order 1 of the resolvent mapping
n
Rpz, T q and limn }Tn } “ 0.
n
(4) RanpI ´ T q is closed and limn }Tn } “ 0.
n
(5) RanpI ´T q is closed, X “ RanpI ´T q‘kerpI ´T q and limn }Tn } “ 0.
If in addition we have kerpI ´ T q “ t0u then all the satements are equiv-
alent to
n
(6) 1 R σpT q and limn }Tn } “ 0.
The equivalence of the first 3 conditions and that they imply (4) and
(5) were proved by Dunford [16, Theorem 8]. Dunford also proved that (4)
implies (1) if T is assumed to be mean ergodic. Lin proved in [33] that (4)
implies (1) with no extra assumptions, and then also (5), which certainly
implies (4). Condition (6) is simply a particular case of the theorem which
is specially relevant in our work. The decomposition in direct sum of (5)
when there are not fixed points is equivalent to I ´ T being an injective
operator and RanpI ´ T q “ X, i.e. to I ´ T being an isomorphism.
We finish this subsection stating a classic result of Lotz, in the form that
we need it
Theorem 2.9 (Theorem 5 of [34]). Let pΩ, Σ, µq be a positive measure space.
Then every bounded linear operator T on L8 pΩq satisfying limn kT n k{n “ 0
which is mean ergodic is also uniformly mean ergodic.
6 JORGE GALINDO AND ENRIQUE JORDÁ

2.2. Convolution operators. Our convolution operators will be defined

on Banach spaces of functions on a locally compact group. For any such
G, we will the denote by mG its Haar measure, i.e., its (essentially unique)
left invariant measure. The Banach space of (equivalence classes of) Haar
p-integrable functions will be simply denoted as Lp pGq.
The measure mG needs not be right invariant. Its behaviour under right
transaltions is gauged by the modular function ∆G . ∆G is a homomorphism
of G into the multiplicative group of positive real numbers such that
ż ż
´1
f pxy q dmG pxq “ ∆G pyq f pxq dmG pxq.

The Banach space of bounded regular measures on G will be denoted by

M pGq. Hence M pGq “ C0 pGq˚ . The weak* topology σpM pGq, C0 pGqq will
be referred to as the vague topology.
We will always regard L1 pGq as a subalgebra (an ideal, actually) of M pGq
through the embedding f ÞÑ f ¨ mG .
Definitions 2.10. Let µ, µ1 , µ2 P M pGq be bounded regular measures. We
consider:
(1) The convolution of measures:
ż ż
xµ1 ˚ µ2 , f y “ f pxyq dµ1 pxq dµ2 pyq, for every f P C00 pGq.

(2) The left convolution operator 1 ď p ď 8:

λp pµq : Lp pGq Ñ Lp pGq
given by
ż
λp pµqpf qpsq “ pµ ˚ f qpsq :“ f px´1 sq dµpxq, f P Lp pGq, s P G.

(3) The right convolution operator for 1 ď p ď 8:

ρp pµq : Lp pGq Ñ Lp pGq
given by
ż
ρp pµqpf qpsq “ ∆G pxqf psxq dµpxq, f P Lp pGq, s P G.

The ergodic behaviour of the operators λp pµq and ρp pµq is the same when
1 ă p ă 8. This is due to the following easily verifiable fact.
Fact 2.1. Let 1 ă p ă 8 The operators ρp pµq and λp pµq are intertwined by
1{p
the linear isometry Up : Lp pGq Ñ Lp pGq, given by Up pf qpsq “ ∆G ps´1 qf ps´1 q,
i.e. λp pµqUp “ Up ρp pµq.
Proofs for items (1) and (2) of the following theorem can be found in
Theorem 20.12 of [27], for a proof of item (3), see e.g., page 47 of [21].
Theorem 2.11. Let G be a locally compact group and let µ P M pGq. Then:
(1) µ ˚ f P Lp pGq for every f P Lp pGq and }µ ˚ f }p ď }µ} ¨ }f }p. As a
consequence,
(2) λp pµq P LpLp pGqq and }λp pµq} ď }µ}.
(3) kλ1 pµqk “ kµk.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 7

M pGq is a ˚-Banach algebra under convolution with involution defined

A E
by x µ˚ , h y “ µ, h̃ for every h P C0 pGq, where h̃psq “ hps´1 q, for every
s P G. If µ “ µ˚ , the measure is called Hermitian. Since λ2 pµq˚ “ λ2 pµ˚ q,
the operator λ2 pµq is self-adjoint precisely when µ is Hermitian. We will say
that the measure µ is operator-normal when µ ˚ µ˚ “ µ˚ ˚ µ. The operator
λ2 pµq will be normal precisely when µ is operator-normal, in particular when
G is commutative, [27, Theorem 20.23].
1
We will often deal with xLp pGq, Lp pGqy dualities. For p ‰ 2 and 1/p+1/p1 “
1 , this duality is given by:
ż
1
xf, gy “ f pxqgpxq dmG pxq f P Lp pGq, g P Lp pGq.

When p “ 2, the duality is derived from the inner product in L2 pGq and is
therefore given by
ż
xf, gy “ f pxqgpxq dmG pxq f, g P L2 pGq.

2.3. Abelian groups: Fourier-Stieltjes transforms. If G is Abelian,

the Fourier-Stieltjes transform establishes a unitary equivalence between
convolution operators and multiplication operators. We use this subsection
to recall this fact and some of its consequences.
Let G be a locally compact Abelian and let T denote the multiplicative
group of complex numbers of modulus 1. By a character of G we under-
stand a continuous homomorphism of G into T. The set G p of all continuous
characters of G with the topology of uniform convergence on compact sets
acquires the structure of a locally compact Abelian group under pointwise
multiplication. ş
If µ P M pGq, the function µ p Ñ C given by µ
p: G ppχq “ χptq dµptq is a
bounded uniformly continuous function known as the Fourier-Stieltjes trans-
form of µ. The Riemann-Lebesgue theorem shows that, fp P C0 pGq p for
1
every f P L pGq (recall that we simply write f for the measure f ¨ dmG ).
The Fourier-Stieltjes transform restricted to L1 pGq is usually known as the
Fourier transform.
p Ñ
The symbol Mµp will denote the multiplication operator Mµp pf q : L2 pGq
2
L pGq p given by Mµp pf qpχq “ µ
ppχq ¨ f pχq.
Theorem 2.12. Let G be a locally compact Abelian group and let µ P M pGq.
(1) The convolution operator λ2 pµq P LpL2 pGqq is unitarily equivalent to
p
the multiplication operator Mµp P LpL2 pGqq. Hence, }λ2 pµq} “ }p
µ }8 .
p { 2 p
(2) Let Kµ :“ tf : f P kerpI ´ λ2 pµqqu and let Pλ pµq P LpL pGqq be the
projection operator onto Kµ . Then P { λ pµq “ M1A , where Aµ “
µ
p´1 pt1uq.
µ
2.4. Amenable groups. Spectrum of convolution operators. Ame-
nability is a far-reaching property of topological groups that is characterized
by the existence of invariant means on their space of uniformly continuous
bounded functions. All compact groups and all locally compact Abelian
groups are amenable. Free groups and locally compact groups containing
8 JORGE GALINDO AND ENRIQUE JORDÁ

them, such as semisimple Lie groups, are among the most prominent nona-
menable groups. We will refrain from defining what an amenable group is.
For our purposes, we only need to recall the following properties.
Theorem 2.13. Let G be a locally compact group, let µ be a positive measure
on G and let 1 ď p ă 8. Then:
(1) (Theorem 3.2.2 of [21], for instance) If G is amenable, then kµk P
σpλp pµqq.
(2) ([6, Théorème 5]) Hµ is amenable if and only if rpλp pµqq “ kµk.
Proposition 2.14 below is well known when G is Abelian. It is also well-
known that σpλp pµqq Ď σpλ1 pµqq “ σpµq for every locally compact group,
not necessarily amenable, and every µ P M pGq. The inclusion σpλ2 pµqq Ď
σpλp pµqq for amenable groups is not, as far as we know, explicitly stated in
the literature.
Proposition 2.14. Let G be an amenable locally compact and let µ P M pGq.
Then, for any 1 ď p ď q ď 2 or 2 ď q ď p, σpλq pµqq Ď σpλp pµqq.
Proof. Let z P C, z R σpλp pµqq. Then there is an operator S P LpLp pGqq,
such that pλp pµq ´ zIq ˝ S “ S ˝ pλp pµq ´ zIq “ I. Since pλp pµq ´ zIdq
commutes with right translations, so will do its inverse S. In the terminology
of [24] this means that S P Convp pGq. By [24, Theorem C] (proved in
Corollary of [26], see also Section 8.3 of [13]) we have then that S P LpLq pGqq,
what means that z R σpλq pGqq. 
2.5. Fixed points of convolution operators. The following theorem is
part of Corollary 6.6 of [38]. For µ ě 0, it can be deduced from the results
of Derriennic [14, Théorème 8] (also obtained by Mukherjea [35, Theorem 2]
in the second countable case) and Dérriennic and Lin, [15, Proposition 2.1].
Theorem 2.15. Let G be a locally compact group and let µ P M pGq be a
measure with kµk ď 1 such that Hµ is not compact. If f P Lp pGq, 1 ď p ă 8,
and µ ˚ f “ f (a.e.), then f “ 0 (a.e.)
The impact of Theorem 2.15 in the ergodic behaviour of convolution op-
erators is reflected in the following two consequences
Proposition 2.16. Let G be a locally compact group and let µ P M pGq be
a measure with kµk ď 1 such that Hµ is not compact. Then pλp pµrns qf qn
converges to 0 for each f P Lp pµq.
Proof. Since Lp pµq is reflexive and }λp pµq} ď 1, we get from Corollary 2.4
that pλp pµqq is mean ergodic, and pλp pµrns qqn converges in the strong opera-
tor topology to the projection P on the fixed points of λp pµq. Theorem 2.15
yields P “ 0.

Proposition 2.17. Let G be a locally compact group and let µ P M pGq be a
measure with kµk ď 1 such that Hµ is not compact. The following assertions
are equivalent for 1 ď p ă 8:
(1) λp pµq is uniformly mean ergodic.
(2) 1 R σpλp pµqq.
Proof. Follows from Theorem 2.15 and (6) of Theorem 2.8. 
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 9

2.6. Vague ergodicity. Using convolution as multiplication, it also makes

sense to consider the ergodic behaviour of a measure without reference to
the operator it induces. The limitting process is then studied under the
σpM pGq, C0 pGqq topology which is usually called the vague topology. Notice
that by the Alaoglu-Bourbaki theorem, this topology agrees, on bounded
subsets of M pGq, with the σpM pGq, C00 pGqq-topology. Here C00 pGq stands
for the space of continuous functions on G with compact support.
Definition 2.18. Let G be a locally compact group and let µ P M pGq. For
2 n
n P N, we define µrns “ µ`µ `¨¨¨`µ
n . We say that µ is a vague-ergodic
measure if there is a measure µc P M pGq such that, in the vague topology,
lim µrns “ µc .
nÑ8

Probability measures are always vague-ergodic. We recall here this impor-

tant classical result that can be deduced from the Mean Ergodic Theorem.
See Theorem 3.4 below, for the proof of a more general version.
Theorem 2.19 (Theorem 3.0 of [22]). If G is a second countable locally
compact group, then every probability measure µ is vague-ergodic and if µc “
limnÑ8 µrns , then µc is a convolution idempotent measure.

3. General results
In this section we develop tools that are not directly related with ergod-
icity but have a strong impact in our work. Some hints on that impact are
also included in this section.
We first establish the continuity properties of the regular representation
on M pGq. These properties, will permeate most sections of the paper.
Our second tool will be a well-known result on the vague convergence of
powers of probability measures. This has a clear impact on the existence of
fixed points for the corresponding convolution operator.
3.1. Continuity properties of regular representations on M pGq. We
next establish the continuity properties of the left regular representation on
M pGq. This is necessary to connect vague-ergodicity of a measure with the
mean ergodicity of the corresponding convolution operator.
Proposition 3.1. Let G be a locally compact group and consider the map-
ping λp : M pGq Ñ LpLp pGqq.
(1) λp is vague-WOT continuous on norm bounded subsets of M pGq, for
every 1 ă p ă 8.
(2) If Hµ is compact, then λp is vague-SOT sequentially continuous for
every 1 ď p ă 8.
Proof. We start with (1). Let pµα qα be a (norm) bounded net that converges
vaguely to 0 . Since pλp pµα qqα is also norm bounded, and C00 pGq is norm
dense in Lp pGq, we only need to show that pλp pµα qf qα is weakly convergent
to 0 in Lp pGq for each f P C00 pGq. As pλp pµα qf qα is a bounded net in Lp pGq,
and then relatively weakly compact, and σpLp pGq, C00 pGqq is a Hausdorff
topology, it will again suffice to prove that limα x λp pµα qf, h y “ 0 for every
h P C00 pGq.
10 JORGE GALINDO AND ENRIQUE JORDÁ

So, let f, h P C00 pGq. We first observe that, by Fubini’s theorem,

@ ` D˘
x λp pµα qf, h y “ µα , h ˚ fˇ ,
1
where the first bracket refers to the pLp pGq, Lp pGqq, the second to the
pM pGq, C00 pGqq-duality and, for any u : G Ñ C, ǔptq “ upt´1 q.
Once this is clear, it immediately follows from the vague convergence of
pµα qα and the inclusion C00 pGq ˚ Lp pGq Ď C0 pGq, that
@ ` D˘
lim x λp pµα qpf q, h y “ lim µα , h ˚ fˇ “ 0.
α α
For Statement (2) we start with a sequence pµn qn vaguely convergent to
0. As before, we only need to check that kλp pµn qf kp converges to 0 for every
f P C00 pGq. Let K :“ supppf q
Now,
ż ˇA Eˇp
λp pµn qf p “ ˇˇ µn , fqx ˇˇ dx,

p

q “ f pt´1 xq.
where fx denotes the right translate of f by x, so that ˇA fx ptq Eˇ
ˇ ˇp
As above, vague convergence of pµqn implies that limn ˇ µn , fqx ˇ “ 0 for
every x P G. Taking into account
ˇA Eˇp
ˇ q ˇ
ˇ fx , µn ˇ ď kf kp8 ¨ supkµn kpM pGq 1Hµ K pxq,
n
we can apply Lebesgue dominated convergence to conclude limn kλp pµn qf kp “
0, as we wanted to prove. 
Vague-WOT continuity of λ1 and Vague-SOT convergence of λp (1 ă
p ă 8) on bounded sets, both fail when G is not compact, as the following
example shows.
Example 3.2. If G is an infinite discrete group, the mapping λ1 : M pGq Ñ
Lpℓ1 pGqq is not vague-WOT continuous on bounded sets, and, for 1 ă p ă
8, λp : M pGq Ñ Lpℓp pGqq is not vague-SOT sequentially continuous.
Proof. We prove that λp fails to be vague-SOT sequentially continuous for
1 ď p ă 8,. Since ℓ1 pGq has the Schur property, this implies both state-
ments. Let psn qn be a sequence in G with infinitely many different terms.
The sequence pδsn qn clearly converges vaguely to 0. Take any f P ℓ1 pGq,
f ‰ 0. Then, for any 1 ď p ă 8,
kλp pδsn qf kp “ kf kp .
Hence, pλp pδsn qf qn cannot converge to 0 in ℓp pGq. 
Theorem 3.1 establishes an immediate relation between vague-ergodicity
and mean ergodicity of convolution operators.
Proposition 3.3. Let G be a locally compact group and µ P M pGq. The
following statements are equivalent:
(1) µ is vague-ergodic.
n
(2) µ is Cesàro bounded and limn µn “ 0, vaguely.
(3) µ is Cesàro bounded and λp pµq is weakly mean ergodic for all 1 ă
p ă 8.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 11

(4) µ is Cesàro bounded and λp pµq is weakly mean ergodic for some
1 ă p ă 8.
If Hµ is compact we can replace (3) and (4) by
(3)1 µ is Cesàro bounded and λp pµq is mean ergodic for all 1 ď p ă 8.
(4)1 µ is Cesàro bounded and λp pµq is mean ergodic for some 1 ď p ă 8.
Proof. We prove first the necessity of the conditions. If µ is vague-ergodic
it is clear that pkµrns kqn is bounded. The equality,
µn n´1
“ µrns ´ µrn´1s ,
n n
n
shows that µn converges vaguely to 0. Therefore (1) implies (2)
We assume now that (2) holds and fix 1 ă p ă 8. We deduce from
Proposition 3.1 that pλp pµqrns qn is a bounded sequence in LpLp pGqq and
that lim λp pµqn {n “ 0 in the weak operator topology. Theorem 2.3 then
implies that λp pµq is weakly mean ergodic.
Statement (3) certainly implies (4). Assume now (4), i.e., that pµrns qn is
norm bounded and pλp pµqrns qn converges in the weak operator topology to
some P P LpLp pGqq. Since the sequence pµrns qn is norm bounded, Theorem
3.1 (i) implies that for any accumulation point µ of pµrns qn , λp pµq “ P . It
follows that pµrns qn has only precisely one accumulation point, it is therefore
convergent.
The proof of the case when G is compact is completely analogous but
using Proposition 3.1 (2). 
Convolution operators λp pµq can be mean ergodic and yet µ fail to be
vague ergodic, even if G is compact and Abelian, see Remarks 4.4. For an
example with µ ě 0, see Example 7.5.
The foregoing theorem and Corollary 3.3 lead to the following general-
ization of Theorem 2.19 that goes beyond probability measures and second
countable groups.
Theorem 3.4. Let G be a locally compact group and let µ P M pGq. If
p}µn }qn is a bounded sequence, then µ is vague-ergodic.
As a further corollary we get the following version of Theorem 2.15.
Corollary 3.5. Let G be a locally compact group and let µ P M pGq with
}µ} ď 1. Then pµrns qn is vague convergent to 0 if and only if Hµ is not
compact.
Proof. By Proposition 3.4 µ is vague ergodic, i.e. there is µc P M pGq such
that pµrns qn is convergent to µc in the vague topology. Also by Proposition
3.3 pλ2 pµrns qqn is convergent to λ2 pµc q, which must be the projection on the
fixed points of λ2 pµc q. If Hµ is compact then the characteristic function 1Hµ
is a fixed point of λ2 pµq, and then λ2 pµc q is a non null projection, and hence
µc ‰ 0. Conversely, if Hµ is not compact then λ2 pµc q “ 0 by Theorem 2.15,
and thus certainly also µc “ 0. 
Remark 3.6. The condition kµk ď 1 in Corollary 3.5 is imposed by its
dependence on Theorem 2.15. The simple proof of Corollary 3.5 will remain
valid under any condition on µ that keeps pkµn kqn bounded and makes sure
12 JORGE GALINDO AND ENRIQUE JORDÁ

that µc “ 0. Since µc is an idempotent measure, this latter condition is freely

obtained in groups that do not admit nontrivial idempotent measures. Any
locally compact Abelian group with no nontrivial compact subgroup satisfies
this property, [43, Theorem 3.3.2].

4. Operator-normal measures
In this section we restrict our study to measures that give rise to convolu-
tion operators that are normal, i.e., to operator-normal measures according
to our definitions in Section 2. This will automatically involve all measures
when G is Abelian.
Normal operators on Hilbert spaces satisfy the identity rpT q “ kT k, and
that greatly simplifies the analysis of mean ergodicity. The following is an
easy consequence of Theorems 2.2 and 2.8.
Theorem 4.1. Let H be a Hilbert space and let T P LpHq be a normal
operator. Then
(1) The operator T is mean ergodic if and only if it is weakly mean
ergodic if and only if }T } ď 1.
(2) The operator T is uniformly mean ergodic if and only if }T } ď 1 and
1 is not an accumulation point of σpT q.
Proof. Since }T } “ rpT q, corollaries 2.4 and 2.6 imply that T is (weakly)
mean ergodic if and only if kT k ď 1.
If T is uniformly mean ergodic it follows from (1) implies (3) on Theorem
2.8 that 1 cannot be an accumulation point of σpT q.
For the converse we only have to recall that for a normal operator T and
λ P C, RanpT ´ λIq is closed if and only if λ is not an accumulation point
of σpT q, see, e.g., [10, Proposition 4.5 of Chapter XI]. Hence T is uniformly
mean ergodic if 1 is not an accumulation point of σpT q, by (4) implies (1)
of Theorem 2.8. 
Corollary 4.2. Let H be a Hilbert space and let T P LpHq be a normal
operator. Then T is uniformly mean ergodic whenever T 2 is.
Proof. If T is normal also T 2 is. Assume that T is not uniformly mean
ergodic. Then 1 is an accumulation point of σpT q, and therefore also is an
accumulation point of σpT 2 q “ σpT q2 . Therefore T 2 is not uniformly mean
ergodic. 
We will see in Example 7.1 that the converse of Corollary 4.2 is not true.
4.1. Mean ergodicity of normal convolution operators. We can now
completely characterize the mean ergodicity of λ2 pµq when µ is operator-
normal. This provides a complete characterization of the mean ergodicity
of λ2 pµq when G es abelian and µ P M pGq.
Theorem 4.3. Let G be a locally compact group G and let µ be an operator-
normal measure on G. Then the following assertions are equivalent:
(1) The operator λ2 pµq is mean ergodic.
(2) The operator λ2 pµq is weakly mean ergodic.
(3) rpλ2 pµqq “ kλ2 pµqk ď 1.
(4) The operator λ2 pµq is power bounded.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 13

(5) The operator λ2 pµq is Cesàro bounded.

Proof. The equivalence of Statements (1) to (3) follows at once from Theo-
rem 4.1. It is obvious that (3) implies (4) and (4) implies (5). Proposition
2.5 and the normality of λ2 pµq show that Statement (5) implies Statement
(3). 
Remarks 4.4. The assumptions of Theorem 4.3 cannot completely removed
and its conclusions cannot be easily strengthened.
(1) Example 6.24 of [44] shows that every nondiscrete locally compact
Abelian group contains a measure µ P M pGq with kµn k ě 2n and
rpλ2 pµqq “ kλ2 pµqk “ kp µk8 ă 1. The operator λ2 pµq is mean
ergodic but the measure µ is not Cesàro bounded, let alone vague-
ergodic. After Theorem 3.3, one deduces that, at least for operator-
normal measures, vague-ergodicity is strictly stronger than weak
mean ergodicity of the convolution operator.
(2) Both the normality condition µ˚ ˚ µ “ µ ˚ µ˚ and the restriction to
the Hilbert case p “ 2 can be removed if µ ě 0 and Hµ is amenable,
see Theorem 5.1 infra. Example 7.5 shows that Theorem 4.3 is no
longer true when µ is not operator-normal and Hµ is not amenable,
even if µ is positive.
(3) If we keep the condition µ˚ ˚ µ “ µ ˚ µ˚ but consider p ‰ 2, the
result also fails, as witnessed by the following example.
Example 4.5. Let p ą 2. On G “ Z3 , the cyclic group of order 3, there
is a measure µ P M pGq such that kλp pµqk ą 1 but λp pµq is uniformly mean
ergodic.
Proof. If G “ te, x, x2 u, let µ “ pδx ´ δx2 q. An ? elementary computation
yields that, for any p. rpλp pµqq “ kλ2 pµqk “ 3. Computing kλp pµqf kp
with f “ 6´1{p δe ` 6´1{p δx ´ p3{2q´1{p δx2 , one sees that, for every p,
1´ ¯
1 p
kλp pµqkp ě 1 ` 4p .
3
?
? then that for every p 1ą 2, kλp pµqk ą 3. Pick now t with
It is easy to check
kλp pµqk ą t ą 3. The measure µt “ t µ is then the desired measure with
kλp pµt qk ą 1 but rpλp pµt qq ă 1. 
4.2. Uniform mean ergodicity of normal convolution operators.
Turning to uniform mean ergodicity the direct consequence of Theorem 4.1
is the following.
Theorem 4.6. Let G be a locally compact group G and let µ be an operator-
normal measure on G. Then the following assertions are equivalent:
(1) The operator λ2 pµq is uniformly mean ergodic.
(2) kλ2 pµqk “ rpλ2 pµqq ď 1 and 1 is not an accumulation point of
σpλ2 pµqq.
We next extend the preceding Theorem to the case p ‰ 2. This will follow
from Proposition 4.9 which ensures that if uniform mean ergodicity of λq pµq
is assumed for some q, then condition (3) in Theorem 2.8 can be relaxed
as in Theorem 4.1 to characterize uniform mean ergodicity for arbitrary
14 JORGE GALINDO AND ENRIQUE JORDÁ

p. We first need a known result on extension of vector-valued holomorphic

functions.
Definition 4.7. Let X be a Banach space. A subset H Ď X ˚ is said to be
separating in X when x˚ pxq “ 0 for all x˚ P H implies x “ 0.
Proposition 4.8. [19, Theorem 1],[9, Corollary 10, Remark 11] Let X be
a Banach space, H a separating subspace of X ˚ , Ω Ď C a domain, a P Ω.
Let f : Ωztau :Ñ X be a holomorphic function such that x˚ ˝ f admits
holomorphic extension to Ω for each x˚ P H. Then f admits a (unique)
holomorphic extension to Ω .
Proposition 4.9. Let G be a locally compact group and let µ P M pGq. As-
sume there is q ą 1 such that λq pµq is uniformly mean ergodic. Then, for
each 1 ď p ď 8, λp pµq is uniformly mean ergodic if and only if lim }λp pµn q}{n “
0 and 1 is not an accumulation point of σpλp pµqq.
Proof. The necessity follows from (1) implies (3) in Theorem 2.8. We
prove the converse. Let r ą 0 such that pBp1, rqzt1uq X σpλp pµqq “ ∅
and pBp1, rqzt1uq X σpλq pµqq “ ∅. The resolvent mapping restricted to
Bp1, rqzt1u
Rp¨, λp pµqq : Bp1, rqzt1u Ñ LpLp pGqq
is then a holomorphic function. From uniform mean ergodicity in Lq pGq
and Theorem 2.8 we get that, for each f, g P C00 pGq, the function
z ÞÑ x pz ´ 1qRpz, λq pµqpf qq, g y ,
z P Bp1, rqzt1u, is holomorphic and admits a holomorphic extension to 1.
For f, g P C00 pGq define If,g P LpLq pGqq˚ by If,g pT q “ x T pf q, g y. Observe
that tIf,g : f, g P C00 pGqu is a separating subset of LpLp pGqq˚ . We conclude
from Rpz, λq pµqpf q “ Rpz, λp pµqpf qq for each f P C00 pGq that the function
Bp1, rqzt1u Ñ C, z ÞÑ x pz ´ 1qRpz, λp pµqqpf q, g y admits a holomorphic ex-
tension to 1 for each f, g P C00 pGq. The conclusion follows from Proposition
4.8 and (3) implies (1) in Theorem 2.8. 
Theorem 4.10. Let G be a locally compact group and let µ be an operator-
normal measure. Assume also that Hµ is amenable. The following assertions
are equivalent for 1 ď p ď 8.
(1) λp pµq is uniformly mean ergodic.
(2) 1 is not an accumulation point of σpλp pµqq, and limn kλp pµn qk{n “ 0.
Proof. From limn kλp pµn qk{n “ 0 and the spectral radius formula it follows
that rpλp pµqq ď 1. Combining Theorem 4.6 with Proposition 2.14 we get
that (2) implies that λ2 pµq is uniformly mean ergodic. Proposition 4.9 then
proves that Statement (2) implies Statement (1). The other direction comes
directly from Theorem 2.8. 
4.3. Abelian groups. When G is Abelian, Corollaries 4.3 and 4.6 can be
rephrased, using Theorem 2.12, in terms of Fourier-Stieltjes transforms.
Corollary 4.11. Let G be a locally compact Abelian group G and let µ P
M pGq. Then:
(1) The operator λ2 pµq is mean ergodic if and only if kp
µk8 ď 1.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 15

(2) The operator λ2 pµq is uniformly mean ergodic if and only if kp

µk8 ď 1
! )
and 1 R µ p µ
ppχq : χ P G, ppχq ‰ 1 .

There is an obvious relation between mean ergodicity of λ2 pµq and how

p Recall, Section 2.3, that Aµ “ µ
Aµ embeds topologically in G. p´1 pt1uq and
hence that Aµ is always a closed set. We now clarify this.
Corollary 4.12. Let G be a locally compact abelian group and let µ P M pGq.
If λ2 pµq is uniformly mean ergodic, then Aµ is a closed and open subset of
p
G.
Proof. If λ2 pµq is uniformly mean ergodic, we deduce from (2) of Corollary
4.11 that there is δ ą 0 such that:
|p p with χ R Aµ .
µpχq ´ 1| ą δ for all χ P G
We see then that µ̂´1 p1 ´ δ, 1 ` δq “ Aµ and therefore that Aµ is open. 
Remark 4.13. It is easy to find measures with Aµ open that do not produce
uniformly mean ergodic operators λ2 pµq. Take for instance G “ T and
µ “ δs for any s “ e2πix with x R Q. Identifying T p with the discrete group
Z we have that µ p : Z Ñ C is defined by µ ppkq “ sk for k P Z. The range
p is well-known to be dense in T, whence we see that µ does not satisfy
of µ
condition (2) of Corollary 4.11. T p being discrete, Aµ is sure open.
The situation in this respect is quite different if µ P L1 pTq or, more
p P c0 pZq, as we next see.
generally, if µ
Definition 4.14. If G is a locally compact Abelian group, we denote by
p
p P C0 pGq.
M0 pGq the set of measures µ P M pGq such that µ
Since the Riemann-Lebesgue theorem proves that L1 pGq Ă M0 pGq, the
following Corollary applies in particular to all λ2 pf q with f P L1 pGq.
Corollary 4.15. Let G be an locally compact abelian group and let µ P
M0 pGq. Then λ2 pµq is uniformly mean ergodic if and only if kp
µk8 ď 1 and
p
Aµ is a closed and open subset of G.
Proof. After Theorem 4.11 and Corollary 4.12, we only have to prove the
sufficiency part.
Suppose therefore that Aµ is an open (and closed) subset of Gp but there is
p with χn R Aµ for every n P N such that 1 “ limn µ
a sequence pχn qn in G ppχn q.
p we deduce that there is a compact subset K of G
p P C0 pGq,
Since µ p such
that χn P K for every n. There is then χ0 P G p and a subnet pχα qα of pχn qn
such that limα χα “ χ0 . Since limn µ ppχn q “ 1, we have that χ0 P Aµ ; but
Aµ being open, this would imply that the net pχα qα is eventually in Aµ . 
p is discrete, we thus have:
If G is compact, then G
Corollary 4.16. Let G be a compact Abelian group and let µ P M0 pGq. The
following assertions are then equivalent:
(1) λ2 pµq is power bounded.
(2) λ2 pµq is mean ergodic.
(3) }p
µ}8 ď 1.
16 JORGE GALINDO AND ENRIQUE JORDÁ

(4) λ2 pµq is uniformly mean ergodic.

Theorem 4.10 suggests that, for a given µ P M pGq, the uniform mean
ergodicity of λp pµq may depend on p. To confirm this we will need the
fllowing result due to S. Igari.
Theorem 4.17 (Particular case of Theorem 1 of [28]). Let G be a nondis-
crete locally compact Abelian group, let 1 ď p ă 2 and let Φ : r´1, 1s Ñ C.
If Φ does not extend to an entire function, there are µ P M pGq with µ p Ď
ppGq
p
r´1, 1s, and h P L pGq such that pΦ ˝ µpq ¨ p
h is not the Fourier transform of
p
any function in L pGq.
Proposition 4.18. For any nondiscrete locally compact Abelian group G
and for every 1 ď p ă 2 there is a measure µ P M pGq such that λ2 pµq is
uniformly mean ergodic but λp pµq is not.
Proof. Define Φ : r´1, 1s Ñ C by Φptq “ 1{pt ´ zq where z P Tzr´1, 1s. Since
Φ cannot be extended to an entire function we can find µz P M pGq and
hz P Lp pGq with the properties of Theorem 4.17.
Suppose that z R σpλp pµz qq. In that case, there would be an operator
T P LpLp pGqq such that µz ˚ T f ´ zT f “ f for every f P Lp pGq. Taking
Fourier-Stieltjes transforms we see that, for every f P Lp pGq,
Txf px
µz ´ zq “ fp.
The preceding equality applied to f “ hz implies then that
Ty xz ,
xz q ¨ h
hz “ pΦ ˝ µ
which goes against the choice of µz and hz from Theorem 4.17. We conclude
that z P σpλp pµz qq.
Once we have found µz P M pGq with µ p Ď r´1, 1s and z P σpλp pµz qq
xz pGq
we consider µ “ zµz . Then

σpλ2 pµqq “ zx p
µz pGq
is contained in a diameter of the unit circle not passing through 1. According
to Corollary 4.6, the operator λ2 pµq is then uniformly ergodic. On the other
hand, 1 P σpλp pµqq. Since isolated points of the spectrum are necessarily
in the range of the Fourier-Stieltjes transform (see [47, Lemma 2.2]) and
1Rµ p we deduce from Corollary 4.6 that λp pµq is not uniformly mean
ppGq,
ergodic. 

5. Positive measures: mean ergodicity

This section is devoted to make clear that the ergodic behaviour of the
operators λp pµq is simpler when µ is positive.

5.1. Positive measures supported in an amenable subgroup: re-

flexive case. In this section we analyze mean ergodicity of convolution
operators induced by positive measures whose support is contained in an
amenable subgroup. The set of techniques at reach for this case is much
richer and leads to conclusive results.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 17

Theorem 5.1. Let G be a locally compact group and let µ P M pGq be a

positive measure with Hµ amenable. If 1 ă p ă 8, the following assertions
are equivalent:
(1) λp pµq is power bounded.
(2) λp pµq is mean ergodic.
(3) λp pµq is weakly mean ergodic.
(4) λp pµq is Cesàro bounded.
(5) }µ} ď 1.
(6) µ is vague-ergodic.
Proof. By Corollary A.4 we can restrict ourselves to the case when G itself
is amenable. The Mean Ergodic Theorem (Corollary 2.4) proves that (1)
implies (2). This one obviously implies (3) and, by Banach-Steinhaus, (3)
implies (4). Since kµk “ rpλp pµqq, by Theorem 2.13, Corollary 2.5 shows
that (4) implies (5). Since rpλp pµqq “ kµk implies that }λp pµq} “ }µ},
Statement (5) certainly implies Statement (1). Hence, statements (1)–(5)
are all equivalent.
Finally, (5) implies (6), by Theorem 3.4, and (6) implies (3) by Theorem
3.3. 
Remark 5.2. Theorem 5.1 does not hold if Hµ is not amenable, see Example
7.5.
5.2. Positive measures: mean ergodicity in L1 pGq. Mean ergodicity
of λ1 pµq is a much more restrictive condition, as we now see. Here we are
not assuming conditions a priori on Hµ . First of all, we observe that we can
reduce our study to probability measures.
Proposition 5.3. Let G be a locally compact group and let µ P M pGq be
positive.
(1) If }µ} ă 1 then pλ1 pµn qqn is norm convergent to 0, and then λ1 pµq
is uniformly mean ergodic.
(2) If }µ} ą 1 then λ1 pµq is not mean ergodic.
Proof. Observe that, for each n P N, kλ1 pµn qk “ }µn } “ µpGqn . Thus
(1) is immediate and (2) follows from the unboundedness of the sequence
p}λpµn q}{nqn when µpGq ą 1. 
Theorem 5.4. Let µ be a probability measure on G. Then λ1 pµq is mean
ergodic if and only if Hµ is compact.
Proof. If Hµ is compact, we only have to apply (2) implies (3)1 of Theorem
3.3.
Assume now that Hµ is not compact. By Corollaries 2.4 and 2.15, λ2 pµq
is a mean ergodic operator without nonzero fixed points. As a consequence,
kλ2 pµqrns f k2 must converge to 0 for every f P L2 pGq.
Let f P C00 pGq, f ě 0. By the above pxλ1 pµqrns f, gyqn converges to 0 for
each g P C00 pGq, i.e. pλ1 pµqrns f qn converges to 0 in the σpL1 pGq, C00 pGqq-
topology. Since we are assuming that λ1 pµq is mean ergodic, λ1 pµqrns f must
converge in L1 pGq. But, as the σpL1 , C00 pGqq-topology is Hausdorff and is
weaker than the norm topology, we conclude that, in fact,

(3) lim λ1 pµqrns f “ 0.
1
n
18 JORGE GALINDO AND ENRIQUE JORDÁ

This actually holds for every for every f P L1 pGq but we will only need that
fact for some f P C00 pGq, f ě 0, f ‰ 0.
A simple application of Fubini’s theorem shows that, for every f P L1 pGq,
ż ż
pµ ˚ f qpxq dmG pxq “ f pxq dmG pxq

Then, for any f P C00 pGq, f ě 0,

f “ µrns ˚ f “ λ1 pµqrns f ,
1 1 1
If we let n go to infinity and apply (3) we reach a contradiction as soon as
f ‰ 0. 
Remark 5.5. As can be remarked in the proof of Theorem 5.4, the reason
for the failure of λ1 pµq to be mean ergodic lies in its action against posi-
tive functions. A way to ş avoid these
( functions is to consider the subspace
1 1
L0 pGq “ f P L pGq : f dmG “ 0 . Since this subspace ˇis invariant un-
der the action of λ1 , one can consider the operator λ01 pµqˇ 1 and study
L0 pGq

its mean ergodicity. This was done by Rosenblatt [42] who defined a mea-
sure µ P M pGq to be ergodic by convolutions if λ01 pµqrns converges to 0. In
that same paper, Rosenblatt proves that a locally compact group contains a
probability measure that is ergodic by convolutions if and only if the group
is σ-compact and amenable. When G is compact, the Itô-Kawada Theorem
proves that a probability measure µ P M pGq is ergodic by convolutions if
and only if Hµ “ G, and the Choquet-Deny theorem implies that the same
assertion is true when G is Abelian [42, Theorems 1.4 and 1.5], see also [11]
and [29] and the references therein for more on this property. It follows from
the facts just collected that every adapted (that is, with Hµ “ G) probabil-
ity measure supported in a noncompact abelian group satisfies that λ01 pµq
is mean ergodic, while λ1 pµq is not. On the other hand, if G is compact
and Abelian, and µ P M pGq is a probability measure which is not adapted,
then λ1 pµq is mean ergodic but λ01 pµq is not. Neither concept is therefore
stronger than the other.
We can now complete the panorama of Proposition 3.3, Theorem 5.1 and
Theorem 5.4.
Theorem 5.6. Let G be a locally compact group and let µ P M pGq be a
positive measure with Hµ compact. For 1 ď p ă 8, the following assertions
are equivalent:
(1) µ is vague-ergodic.
(2) µ is Cesàro bounded.
(3) }µ} ď 1.
(4) λp pµq is mean ergodic.

Proof. That Statement (1) implies Statement (2) is a simple consequence of

the Banach-Steinhaus theorem. Next, (2) is the same as Cesàro boundedness
of λ1 pµq (by (3) of Theorem 2.11) and the latter implies, by Proposition 2.5,
that rpλ1 pµqq ď 1. Since, by Proposition 2.13, this implies kµk ď 1, we see
that Statement (2) implies Statement (3). (3) implies (4) by Theorem 5.1
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 19

if p ą 1 and by Theorem 5.4 for p “ 1. Finally (4) implies (1) follows from
Proposition 3.3. 

6. Positive measures: uniform mean ergodicity

If µ is positive, the equality kλp pµqk “ kµk makes Theorem 4.10 into the
following somewhat cleaner characterization.
Theorem 6.1. Let G be a locally compact group and let µ be a positive
operator-normal measure with Hµ amenable. The following assertions are
equivalent for 1 ď p ď 8.
(1) λp pµq is uniformly mean ergodic.
(2) 1 is not an accumulation point of σpλp pµqq, and kµk ď 1.
Taking Theorem 5.4 into account, Theorem 6.1 yields the following Corol-
lary. It applies to every probability measure in an Abelian group, although
in that case a simpler proof using duality theory can be applied to every
measure of norm 1.
Corollary 6.2. Let G be a locally compact group and let µ be an operator
normal probability measure such that Hµ is amenable and not compact. Then
1 is an accumulation point of σpλ1 pµqq p“ σpµqq.
Theorem 6.1 leads to a complete characterization in the non reflexive case
that is valid for, at least, all Abelian G.
Theorem 6.3. Let G be a locally compact group and let µ P M pGq be normal
and positive. The following are equivalent
(1) λ1 pµq is uniformly mean ergodic.
(2) λ8 pµq is uniformly mean ergodic.
(3) Either }µ} ă 1, or kµk “ 1, Hµ is compact and 1 is not an accumu-
lation point of σpµq.
(4) λ8 pµq is mean ergodic.
Proof. Statement (2) is equivalent to Statement (4) by Theorem 2.9. The re-
maining equivalences follow from Theorem 4.10 and Theorem 5.4, observing
that λ1 pµq˚ “ λ8 pµ˚ q, Hµ “ Hµ˚ and σpµq “ σpµ˚ q and taking into ac-
count that, for a bounded linear operator on a Banach space X, the uniform
ergodicity of T is equivalent to that of T ˚ . 
Remark 6.4. When G is compact, Theorem 5.6 of [3] can be used to find
examples of positive measures with λ2 pµq uniformly mean ergodic for which
λ1 pµq is not. The measures obtained in [3, Theorem 5.6] are positive, belong
to M0 pGq and have the property that σpλ1 pµqq is the whole unit disk (see
Lemma 4.1 of [3] for this). For any such µ, λ2 pµq is uniformly mean ergodic,
Corollary 4.16, yet λ1 pµq is not, Theorem 6.1. Note that λ1 pµq is mean
ergodic by Theorem 5.4.
6.1. Positive measures with noncompact support. When the support
of a probability measure µ is not contained in a compact subgroup of G,
λp pµq does not have fixed points and its dynamic behaviour is much simpler.
Theorem 6.5. Let G be a locally compact group and let µ P M pGq be a
probability measure with Hµ noncompact. If 1 ă p ă 8 , the following
assertions are equivalent.
20 JORGE GALINDO AND ENRIQUE JORDÁ

(1) λp pµq is uniformly mean ergodic.

(2) rpλp pµqq ă 1.
(3) pλp pµn qqn is norm convergent to 0.
(4) Hµ is not amenable.
Proof. According to Corollary A.4, we can assume along this proof that
G “ Hµ . By Theorem 2.13, this immediately shows that Statements (2)
and (4) are equivalent .
As for the remaining equivalences, only that (1) implies (4) needs proof.
Assume to that end that λp pµq is uniformly mean ergodic. Taking into
account that, by Proposition 2.17, 1 R σpλp pµqq Theorem 2.13, applies then
to show that (4) holds. 
For convolution operators of norm at most one induced by measures µ “
f dmG for some positive f P L1 pGq, uniform mean ergodicity can be neatly
characterized for all p, 1 ď p ă 8.
Theorem 6.6. Let G be a locally compact group and let f ě 0, f P L1 pGq,
with kf k ď 1. Then λp pf q, 1 ď p ă 8, is uniformly mean ergodic if and only
if either supppf q is contained in a compact subgroup of G or rpλp pf qq ă 1.
Proof. After Theorems 6.5 and 5.4, it only remains to prove that λp pf q is
uniformly mean ergodic when the support of f is contained in a compact
subgroup of G.
H
The operator λp f pf q obtained by regarding λp pf q as an operator on
p
L pHf q, is mean ergodic. This follows from Theorem 5.4 when p “ 1 and
from the Mean Ergodic Theorem, Corollary 2.4 in the reflexive case p ą 1.
Since convolution with a function in L1 pHf q defines a compact operator
on Lp pHf q (see, e.g., [12, Exercise 10.4.2]) and, for power bounded compact
operators, mean ergodicity is equivalent to uniform mean ergodicity, (3) of
H
Theorem 2.7, we have that λp f pf q is a uniformly mean ergodic operator.
Corollary A.4 then shows that λG p pf q is uniformly mean ergodic as well. 

The simple example in Remark 4.13 shows that when µ R L1 pGq, Hµ

compact does not necessarily imply that λp pµq is a uniformly mean ergodic
operator.
The characterization of Theorem 6.5 leaves room for a convolution op-
erator λp pµq, with µ positive and Hµ noncompact, to be uniformly mean
ergodic and yet rpλp pµqq “ 1. We see below that this cannot happen when
Hµ is amenable.
Corollary 6.7. Let G be a locally compact group and let µ P M pGq be a
positive measure with Hµ amenable and noncompact. If 1 ă p ă 8, the
following assertions are equivalent:
(1) λp pµq is uniformly mean ergodic.
(2) kµk ă 1.
(3) λp pµn q is norm convergent to 0.
Proof. Only that (1) implies (2) requires proof. The equality }λp pµq} “ }µ}
(Theorem 2.13) permits us to proceed as in the proof of Proposition 5.3 to
get kµk ă 1. 
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 21

Remark 6.8. The above Corollary can be used to prove the converse of
Corollary 4.2 for positive measures on amenable groups that do not contain
nontrivial compact subgroups (as, e.g., G “ R or G “ Z). For Abelian
G, more is true. In that case, it is not difficult to prove (relying on the
positive-definiteness of µp) that ´1 is an accumulation point of µ p if and
ppGq
only if 1 is. It follows that for a probability measure µ P M pGq, λ2 pµq is
uniformly mean ergodic if and only if λ2 pµ2 q is. Positivity of µ is essential
here, see Remark 7.2.

7. Tracing limits
7.1. Counterexamples for nonpositive measures. When G is abelian
and µ positive, Theorem 5.1 and Theorem 5.4 in Section 5 characterize the
mean ergodicity of λp pµq, for 1 ď p ă 8. Also under the same assumptions,
Theorem 6.1 and Theorem 6.3 in Section 6 give a complete characterization
of the uniform mean ergodicity of λp pµq, 1 ď p ď 8. In this section we give
examples showing that these characterizations are not longer true when µ
is not required to be positive.
Since our counterexamples will be Abelian, we recall from subsection 2.3
that, for Abelian G, λ2 pµq is unitarily equivalent to the multiplication op-
erator by µ p As a consequence, the spectrum of λ2 pµq is exactly
p on L2 pGq.
p
ppGq.
µ
The example below shows that Theorem 6.5 fails if µ is not assumed to
be positive, even in the Hilbert case, i.e., that λ2 pµq can be uniformly mean
ergodic even if }µ} “ 1.
Example 7.1. A measure µ P M pZq whose support generates Z, rpλ2 pµqq “
1 and yet λ2 pµq is uniformly mean ergodic. Moreover pλ2 pµn qqn is not norm
convergent.
Proof. Take µ “ p1{2qpδ1 ´ δ2 q. By Corollary 4.6, λ2 pµq will be uniformly
mean ergodic if and only if its spectrum is contained in the unit disc and
does not contain 1 as an accumulation point.
In this case, for every 0 ď t ă 2π,
1 ` it ˘ 1 ` ˘
ppeit q “
µ e ´ e2it “ eit 1 ´ eit .
2 2
For µ it it
ppe q “ 1 one needs that |1´ e | “ 2 and this only happens when t “ π,
but in that case µ ppeit q “ ´1. Hence rpλ2 pµqq “ 1 but 1 R σpλ2 pµqq.
Since µ p pe q “ p´1qn for each n P N and kλ2 pµn qk “ rpλ2 pµn qq “
n it

rpλ2 pµqq “ 1, pλ2 pµn qqn cannot be norm convergent to 0. Observe that
n

λ2 pµq does not have non null fixed points by Theorem 2.15. We conclude
that pλ2 pµn qqn cannot converge in norm. 
Remark 7.2. In the above example 1 P σpλ2 pµ2 qq, hence λ2 pµ2 q is not
uniformly mean ergodic. This shows that, unlike the positive case, the
converse to Corollary 4.2 is not true when µ is not positive.
The following example of a measure µ with kµk ą 1, Hµ “ Z and λp pµq
uniformly mean ergodic for every p, reveals that positivity of µ is not a dis-
posable condition in Proposition 5.3, Theorem 5.6, Theorem 6.3 or Corollary
6.7.
22 JORGE GALINDO AND ENRIQUE JORDÁ

1
? 7.3. Let µ “ 3 pδ1 `nδ0 ´ δ´1 q P M pZqp“ l1 pZqq. For each 1 ă
Example
t ă 3{ 7, ktµk ą 1 and }λp ptµq q} is convergent to 0 and, hence, λp ptµq is
uniformly mean ergodic, for 1 ď p ď 8.
2
Proof. We? only have nto observe that kµ k “ 7{9. As a consequence, as long
as t ă 3{ 7, limn ktµ k “ 0. 

7.2. Uniformly mean ergodic convolution operators induced by

positive measures with large support. We have seen in Theorem 4.3
that for an operator-normal measure µ P M pGq, λ2 pµq is mean ergodic if
and only if kλ2 pµqk ď 1. The same equivalence is shown to hold when µ
is positive and supported in an amenable subgroup of G, Theorem 5.1, this
time for λp pµq and 1 ă p ă 8. In this latter case mean ergodicity of λp pµq
is, in addition, equivalent to vague-ergodicity of µ.
Here we show that, when the measure is positive but not operator-normal
and Hµ is not amenable, none of these equivalences remains true. Our
examples will consist on convolution operators defined by finitely supported
measures in discrete groups. We first collect some facts on kλ2 pµqk for
1 ř
µ “ |S| sPS δs with S Ď G.
Recall that elements of the free group F pXq on the set of generators X
can be described uniquely as words of the form w “ xǫ11 ¨ ¨ ¨ xǫ22 ¨ ¨ ¨ ¨ ¨ xǫnn with
xi P X, ǫi “ ˘1 and ǫi “ ǫj whenever xi “ xj . The length of w is then
the minimum number of terms in such an expression of w. When all the
exponents are positive, then w belongs to the semigroup generated by X.
Theorem 7.4. ř Let G be a discrete group and let S Ď G with |S| “ n.
Consider µ “ sPS δs P M pGq. Then:
(1) (Particular case of [40, Theorem 18.3]) rpλ2 pµqq “ n if and only if
x S y is amenable. If S contains the identity, kλ2 pµqk “ n if and only
if x S y is amenable. ?
(2) ([1, Theorem IV.K]) If S generates a free group: kλ2 pµqk “ 2 n ´ 1.
(3) (Haagerup inequality, [23]) If G “ F pXq is a free group and S con-
sists of words of length n on X, then: kλ2 pµqkLpL2 pGqq ď pn ` 1qkµkl2 .
(4) (strong Haagerup inequality, [31]) If G is a free group and S consists
of words of length n?that are in the semigroup generated by X, then
kλ2 pµqkLpL2 pGqq ď e n ` 1kµkℓ2 .
This is the promised counterexample to Theorem 5.1 for positive measures
which are not supported in amenable subgroups. This example also shows
that the hypothesis of normality of the measure cannot be dropped to get the
necessity of }λ2 pµq} ď 1 when λ2 pµq is uniformly mean ergodic in Theorem
4.3, even if µ is positive.
Example 7.5. Let G be the free group on three generators G “ F px1 , x2 , x3 q.
There is a finitely supported positive measure ν P M pGq such that:
(1) kνk ą kλ2 pνqk ą 1.
(2) λ2 pνq is uniformly mean ergodic.
(3) ν is not vague-ergodic.

Proof. Let µ “ pδx1 ` δx2 ` δx3 q and define, and for ?1 ără ?1 , ν “ rµ.
8 3
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 23

Observe to begin with that, for each n P N, the measures µn is precisely

the characteristic function of the set of all words of length n of the free
semigroup generated by tx1 , x?2 , x3 u.
Since, clearly, kν n kℓ2 “ p 3rqn we deduce from the strong Haagerup
inequality ((4) of Theorem 7.4) that
´? ¯n ?
kλ2 pν n qk ď 3r e n ` 1.
The spectral radius formula together with the choice of r yields
?
rpλ2 pνqq ď 3r ă 1
Therefore, by (1) of Theorem 2.7, λ2 pνq is uniformly mean ergodic (pλ2 pν n qqn
is even convergent to 0)
On the other hand, (2) of Theorem 7.4 shows that
?
kλ2 pνkq “ 8r ą 1.
?
Finally, Since kνk “ 3r ą 8r ą 1 we get

kν n k p3rqn
lim “ lim “ 8,
n n n n
and we see that µ cannot be vague-ergodic. 

8. Open questions
We remark that all our examples of mean ergodic operators are power
bounded. It is natural to conjecture a positive answer to the following
problem.
Problem 1. Let G be a locally compact group. Let µ P M pGq and
1 ă p ă 8. Is it true that λp pµq is power bounded whenever it is mean
ergodic? What if µ ě 0? What if Hµ is amenable?
Examples 7.3 and 7.5 both introduce big measures with pλ2 pµn qq conver-
gent in norm to 0, it seems also natural to ask if an example in the spirit
of 7.1 can be obtained for positive measures (necessarily on non-amenable
groups).
Problem 2. Let G be a free group (or any other nonamenable locally
compact group). Is there any positive measure µ P M pGq such that µpGq ą
1, λ2 pµq is mean ergodic and rpλ2 pµqq “ 1? If the answer is positive, could
µ be taken in such a way that also λ2 pµq is uniformly mean ergodic?
Our last question refers to possible generalizations of Theorem 5.1 to
nonpostive measures or nonamenable groups.
Problem 3. Is there a locally compact group G supporting a measure
which is vague ergodic but λp pµq is not mean ergodic? A positive answer
would provide a convolution operator which is weakly mean ergodic but not
mean ergodic.

Appendix. Convolution operators λp pµq as operators on Lp pHµ q:

reducing to the support
If H is a closed subgroup of G with H Ď Hµ , convolving by µ can be seen
both as an operator on Lp pHq and as an operator on Lp pGq. Our aim is to
show that the ergodic behaviour of both operators is, as expected, the same.
24 JORGE GALINDO AND ENRIQUE JORDÁ

The proofs of these results is rather technical and rely on several involved
results of abstract harmonic analysis. We have therefore preferred to defer
their proof to this Appendix.
These facts are best described when convolution operators by measures
are seen in the wider frame of algebras of p-pseudomeasures. The alge-
bra P Mp pGq of p-pseudomeasures is defined as the weak-operator closure
of tρp pf q : f P L1 pGqu in LpLp pGqq. P Mp pGq is a Banach subalgebra of
LpLp pGqq that contains ρp pµq for every µ P M pGq. It is easy to see that
operators in P Mp pGq commute with left translations.
As a Banach space, the algebra P Mp pGq can be seen as the dual space of a
function algebra Ap pGq known as the Figà-Talamanca Herz algebra. We will
not need to provide a precise defintion of this algebra here. It suffices to say
1
that for any f P Lp pGq and g P Lp pGq, with 1{p ` 1{p1 “ 1, the convolution
f¯ ˚ ǧ P Ap pGq, where f¯psq “ f psq and fˇpsq “ f ps´1 q, and that, for each
T P P Mp pGq, xT, f¯˚ ǧy “ xT f, gy, where the first bracket corresponds to the
1
pP Mp pGq, Ap pGqq-duality and the second to the pLp pGq, Lp pGqq-duality.
In our setting, it would have been more natural to introduce the algebra
of pseudomeasures as the weak-operator closure of tλp pf q : f P L1 pGqu, as
it is often done in the literature. This would have produced a different but
linearly isometric algebra. By technical reasons related to Theorem A.3 we
find it preferable to use the right-handed version here.
Since in this section we are going to see the operator ρp pµq, µ P M pHq,
both as an operator on Lp pGq and as an operator on Lp pHq, where H is a
subgroup of G, it will be convenient to use the notation ρG H
p pµq and ρp pµq
G H
(or λp pµq and λp pµq for the left-handed versions) to distinguish both cases.
The basic tool to explore the relation between operators on Lp pGq and
Lp pHq is the Mackey-Bruhat integration formula described in the next lemma.
Lemma A.1 (Mackey-Bruhat integration formula. Remark 8.2.3 of [41]).
Let G be a locally compact group and let H be a closed subgroup of G. There
is a quasi-invariant measure mG{H on the space of left cosets G{H and a
continuous strictly positive function q : G Ñ R such that:
qpxhq ∆H phq
“ , and
qpxq ∆G phq
ż ż ˆż ˙
f pxhq
f pxq dmG pxq “ dmH phq dmG{H pxq,
9
G{H H qpxhq
where x9 denotes the right coset xH. If H is a normal subgroup of G, qpxq “ 1
for every x P G.
Lemma A.2. Let G be a locally compact group, let H be a closed subgroup
of G and let 1 ă p ă 8. Then,
(1) On bounded subsets of P Mp pGq the σpP Mp pGq, Ap pGqq- and weak
operator topologies coincide.
(2) Restriction defines a linear surjective mapping RH : Ap pGq Ñ Ap pHq
such that for each h P Ap pHq and ε ą 0 there is g P Ap pGq with
khk ď kgk ď khk ` ε and RH pgq “ h.
(3) The adjoint pRH q˚ : P Mp pHq Ñ P Mp pGq is a multiplicative linear
isometry.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 25

(4) If Q P P Mp pHq, f P Lp pGq, x P G and h P H,

pRH q˚ pQqf pxhq “ Qfx phqq 1{p pxhq.
(5) pRH q˚ pρH G
p pµqq “ ρp pµq.

Proof. The first item follows, e.g., from Theorem 6 of [13]. Items (2) and (3)
can be deduced from Theorems A and 1 of [25] and also from [13, Proposition
7.3.5 and Theorem 7.8.4].
Item (4) follows from Theorem 7.8.4 of [13] after noting that the op-
erator f pxhq ÞÑ Qfx phqq 1{p pxhq, f P Lp pGq, coincides with the operator
i : P Mp pHq Ñ P Mp pGq introduced in Definition 7.2.7 loc. cit. Since opera-
tors in P Mp pHq commute with left translations, the formula in (4) defines
this operator unambiguously.
Item (5) follows after applying (4) to ρH p
p pµq and any f P L pGq, x P G:

ppRH q˚ pρH p
p pµqqqf pxhq “ ρH pµq fx phqq
1{p
pxhq
ż
f pxhuq
“ ∆H puq1{p qpxhq1{p dµpuq
qpxhuq1{p
ż
(4) “ ∆G puq1{p f pxhuq dµpuq

“ ρG
p pµqf pxhq.

Theorem A.3. Let G be a locally compact group, H a closed subgroup of
G and µ P M pGq with supppµq Ď H. Let as well 1 ă p ă 8.Then ρpH pµq is
(uniformly, weakly) mean ergodic if and only if ρpG pµq is (uniformly, weakly)
mean ergodic.
Proof. 1. Uniform mean ergodicity.
Since pRH q˚ : P Mp pHq Ñ P Mp pGq is a multiplicative linear isometry and
pRH q˚ pρH G H
p pµqq “ ρp pµq, Lemma A.2, it is clear that ρp pµq is uniformly
mean ergodic if and only if ρG p pµq is uniformly mean ergodic.
2. Weak mean ergodicity.
Lemma A.2 and the Banach-Steinhaus theorem imply that Condition (2) of
Theorem 2.3 holds for ρpH pµq if and only if it holds or ρpG pµq. It also follows
that as soon as either ρpH pµq or ρpG pµq is weakly mean ergodic then λpH pµqrns ,
ρpG pµqrns , n1 ρG pµn q and n1 ρH pµn q will all be bounded in the operator norm.
Suppose now that n1 ρpH pµqn converges to 0 in the weak operator topology
1
and let f P Lp pGq and g P Lp pGq, then
1@ p D 1@ p n D
ρG pµqn f, g “ ρG pµ q, f¯ ˚ ǧ
n n
1@ p n D
“ ρH pµ q, RH pf¯ ˚ ǧq .
n
So, since weak operator topology and σpP Mp pGq, Ap pGqq coincide on bounded
sets of P Mp pGq ((1) of Lemma A.2), n1 ρpG pµn q converges to 0 in the weak
operator topology.
If, conversely, n1 ρpG pµqn converges to 0 in the weak operator topology and
1
f P Lp pHq, g P Lp pHq we can consider u P Ap pGq with RH puq “ f¯ ˚ ǧ and,
26 JORGE GALINDO AND ENRIQUE JORDÁ

noting again that R˚ pρH G

p pµqq “ ρp pµq,
1@ p D 1@ p n D
ρH pµqn f, g “ ρH pµ q, f¯ ˚ ǧ
n n
1@ p n D
“ ρH pµ q, RH puq
n
1@ p n D
“ ρ pµ q, u
n G
and we see that n1 ρpH pµqn converges to 0 in the weak operator topology.
As both conditions of Theorem 2.3 hold precisely for ρG p pµq when they
hold for ρH p pµq, we conclude that ρG pµq is weakly mean ergodic if and only
p
if ρH
p pµq is.
3. Mean ergodicity.
We are now going to use the norm topology version of Theorem 2.3. As
above, Condition (2) will be satisfied for ρG p pµq if and only if it is satisfied
for ρHp pµq.
n
Let µn “ µn . We will show that ρG p pµn q converges to 0 in the strong
operator topology if and only if so does ρH p pµn q. Our approach will follow
closely Chapter 7 of [13].
So, let us first assume that ρH p pµn q converges to 0 in the strong operator
topology. Using the Mackey-Bruhat formula in Lemma A.1, for f P Lp pGq,
ż ż
|ρG pµn qf pshq|p
(5) kρG pµn qf kpLp pGq “ dmH phq dmG{H psq.
9
G{H H qpshq
f pshq
Putting fs phq “ qpshq 1{p and applying the properties of the function q to the

inner integral in (5)

ż ż ˇˇż ˇp
ˇ
|pρG pµn qf qpshq|p ˇ ∆G puq1{p f pshuq ˇ
dmH phq “ ˇ dµn puqˇ dmH phq
H qpshq ˇ
H H qpshq1{p ˇ
ˇ
ż ˇż ˇp
∆G puq1{p f pshuq∆H puq1{p ˇ
ˇ ˇ
“ ˇ dµ n puqˇ dmH phq
Hˇ H ˇ
1{p
qpshuq ∆G puq 1{p

ż ˇˇż ˇp
ˇ
ˇ ∆H puq1{p f pshuq ˇ
“ ˇ dµ n puqˇ dmH phq
ˇ
H H qpshuq 1{p ˇ
“ kpρH pµn qfs qkp P .
L pHq

Since ρH pµn q converges to 0 in the strong operator topology, the sequence

kρH pµn qfs kpLp pHq converges to 0 for every s9 P G{H. By Lebesgue’s dom-
inated convergence theorem, applied to the integral in (5), the sequence
kρG pµn qf kLp pGq will converge to 0 as long as we can see that the functions
kρH pµn qfs kpLp pHq are dominated by some integrable function. But, for each
s9 P G{H, kρH pµn qfs kpLp pHq ď kρH pµn qkp ¨ kfs kpLp pHq , and the Bruhat-Mackey
formula implies that s9 ÞÑ kfs kpLp pHq is integrable with, precisely,
ż
kfs kpLp pHq dmG{H psq
9 “ kf kpLp pGq .
G{H

We conclude so that ρG
p pµn qf converges to 0 in norm.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 27

p p 1
Assume now that ρGp pµn q converges to 0 and let f P L pHq, g P L pHq
with kgk p1 ď 1. If we follow the proof of Theorem 7.3.2 of [13], we can
L pHq
1
find two functions vf P Lp pGq, vg P Lp pGq such that (this is the top formula
of page 115 loc. cit.)
kvg k p1 ď kgk p1 ď 1 and
L pGq L pHq
ˇ@ H Dˇ ˇ@ G Dˇ
ˇ ρp pµn qf, g ˇ ď ˇ ρp pµn qvf , vg ˇ .

It follows then that limnÑ8 kpρH

p pµn qf kLp pHq “ 0.

The equivalence between ρp pµq and λp pµq stated in Fact 2.1 leads to the
following Corollary.
Corollary A.4. Let G be a locally compact group, H a closed subgroup of
G and µ P M pGq with supppµq Ď H. Let as well 1 ă p ă 8. Then:
(1) }λ`pH pµq} ˘“ }λpG`pµq}. Hence,
˘
(2) r λH G
p pµq “ r λp pµq .
(3) λpH pµq is (uniformly, weakly) mean ergodic if and only if λpG pµq is.
Acknowledgements. We have profitted from discussions around these top-
ics with several colleagues. We would like in particular to thank Antoine
Derighetti for always being ready to answer our queries on how to relate
λH G
p pµq and λp pµq, to Michael Lin for helping us to gain perspective on the
probabilistic side of convolution operators, to Nico Spronk and Matthias Ne-
ufang for making us aware of [35] and sharing with us some details of [38],
which permitted us avoid our own, unnecessary, proof of Theorem 2.15, and
to Przemyslaw Ohrysko for pointing us towards the paper [28] and hence
setting us on the path to Proposition 4.18.

References
[1] Charles A. Akemann and Phillip A. Ostrand. Computing norms in group C ˚ -algebras.
Amer. J. Math., 98(4):1015–1047, 1976.
[2] Angela Albanese, José Bonet, Werner Ricker, Mean ergodic operators in Fréchet
spaces Annal. Acad. Sci. Fenn. Math., 34, 401–436, 2009.
[3] Gavin Brown, Costas Karanikas and John H. Williamson, The asymmetry of
M0 pGqMath. Proc. Cambridge Philos. Soc., 91:407–433, (1982).
[4] Marı́a J. Beltrán-Meneu, M. Carmen Gómez-Collado, Enrique Jordá, and David Jor-
net. Mean ergodic composition operators on Banach spaces of holomorphic functions.
J. Funct. Anal., 270 (12):4369–4385, 2016.
[5] Marı́a J. Beltrán-Meneu, M. Carmen Gómez-Collado, Enrique Jordá, and David Jor-
net. Mean ergodicity of weighted composition operators on spaces of holomorphic
functions. J. Math. Anal. Appl., 444(2):1640–1651, 2016.
[6] Christian Berg and Jens Peter Reus Christensen. Sur la norme des opérateurs de
convolution. Invent. Math., 23:173–178, 1974.
[7] José Bonet and Pawel Domański. A note on mean ergodic composition operators on
spaces of holomorphic functions. Rev. R. Acad. Cienc. Exactas Fı́s. Nat. Ser. A Math.
RACSAM, 105(2):389–396, 2011.
[8] José Bonet and Pawel Domański. Power bounded composition operators on spaces of
analytic functions. Collect. Math., 62(1):69–83, 2011.
[9] José Bonet, Leonhard Frerick, and Enrique Jordá. Extension of vector-valued holo-
morphic and harmonic functions. Studia Math., 183(3):225–248, 2007.
28 JORGE GALINDO AND ENRIQUE JORDÁ

[10] John B. Conway. A course in functional analysis, volume 96 of Graduate Texts in

Mathematics. Springer-Verlag, New York, second edition, 1990.
[11] Christophe Cuny, Weak mixing of random walks on groups, J. Theoret. Probab. 16
(2003), 923–933.
[12] H. Garth Dales, Pietro Aiena, Jörg Eschmeier, Kjeld Laursen, and George A. Willis.
Introduction to Banach algebras, operators, and harmonic analysis, volume 57 of
London Mathematical Society Student Texts. Cambridge University Press, Cambridge,
2003.
[13] Antoine Derighetti,Convolution operators on groups,Lecture Note s of the Unione
Matematica Italiana, 11, Springer, Heidelberg; UMI, Bologna, 2011.
[14] Yves Derriennic, Lois ”zèro ou deux” pour les processus de Markov. Applications aux
marches alèatoires. Ann. Inst. H. Poincar Sect. B (N.S.) 12 (1976), no. 2, 111–129
[15] Yves Derriennic and Michael Lin, Convergence of Iterates of Averages of Certain
Operator Representations and of Convolution Powers J. Funct. Anal. 85 (1989), 86–
102.
[16] Nelson Dunford. Spectral theory. Bull. Amer. Math. Soc., 49:637–651, 1943.
[17] Nelson Dunford and Jacob T. Schwartz. Linear operators. Part I. Wiley Classics
Library. John Wiley & Sons, Inc., New York, 1988. General theory, With the as-
sistance of William G. Bade and Robert G. Bartle, Reprint of the 1958 original, A
Wiley-Interscience Publication.
[18] Tanja Eisner, Bálint Farkas, Markus Haase, and Rainer Nagel. Operator theoretic
aspects of ergodic theory, volume 272 of Graduate Texts in Mathematics. Springer,
Cham, 2015.
[19] Karl Grosse-Erdmann. A weak criterion for vector-valued holomorphy. Math. Proc.
Cambridge Philos. Soc., 136(2):399–411, 2004.
[20] Carmen Fernández, Antonio Galbis, Enrique Jordá, Dynamics and spectra of compo-
sition operators on the Schwartz space. J. Funct. Anal. 274 (2018), no. 12, 3503–3530.
[21] Frederick P. Greenleaf. Invariant means on topological groups and their applications.
Van Nostrand Mathematical Studies, No. 16. Van Nostrand Reinhold Co., New York-
Toronto, Ont.-London, 1969.
[22] Ulf Grenander. Probabilities on algebraic structures. Second edition. Almqvist & Wik-
sell, Stockholm; John Wiley & Sons Inc., New York-London, 1968.
[23] Uffe Haagerup. An example of a nonnuclear C ˚ -algebra, which has the metric ap-
proximation property. Invent. Math., 50(3):279–293, 1978/79.
[24] Carl Herz. The theory of p-spaces with an application to convolution operators. Trans
Amer. Math. Soc. 154 (1971), 69–82.
[25] Carl Herz. Harmonic synthesis for subgroups. Ann. Inst. Fourier (Grenoble),
23(3):91–123, 1973.
[26] Carl Herz and Nestor Rivière.Estimates for translation-invariant operators on spaces
with mixed norms. Studia Math. 44 (1972), 511–515.
[27] Edwin Hewitt and Kenneth A. Ross. Abstract harmonic analysis. Vol. I, Springer-
Verlag, Berlin, second edition, 1979.
[28] Satoru Igari.Functions of Lp-multipliers.Tôhoku Math. J. (2) 21 (1969), 304–320.
[29] Wojciech Jaworski, Probability measures on [SIN] groups and some related ideals in
group algebras. Monatsh. Math. 155 (2008), 135–144.
[30] Thomas Kalmes, Power bounded weighted composition operators on function spaces
defined by local properties. J. Math. Anal. Appl. 471 (2019), no. 1-2, 211–238.
[31] Todd Kemp and Roland Speicher. Strong Haagerup inequalities for free R-diagonal
elements. J. Funct. Anal., 251(1):141–173, 2007.
[32] Ulrich Krengel. Ergodic theorems, volume 6 of De Gruyter Studies in Mathematics.
Walter de Gruyter & Co., Berlin, 1985. With a supplement by Antoine Brunel.
[33] Michael Lin. On the uniform ergodic theorem. Proc. Amer. Math. Soc., 43:337–340,
1974.
[34] Heinrich P. Lotz. Uniform convergence of operators on L8 and similar spaces.Math
Z., 190 (1985), 207–220.
 ERGODIC PROPERTIES OF CONVOLUTION OPERATORS 29

[35] Arunava Mukherjea. Limit theorems for probability measures on non-compact groups
and semi-groups. Z. Wahrscheinlichkeitstheorie und Verw. Gebiete 33 (1975/76), no.
4, 273–284.
[36] Heybetkulu Mustafayev, Convergence of iterates of convolution operators in Lp
spaces.Bull. Sci. Math. 152 (2019), 61–92.
[37] Heybetkulu Mustafayev, Mean ergodic theorems for multipliers on Banach algebras,
J. Fourier Anal. Appl. 25 (2019), no. 2, 393.–426.
[38] Matthias Neufang, Pekka Salmi, Adam Skalski, Nico Spronk, Fixed points and limits
of convolution powers of contractive quantum measures, Preprint, 2019. Available at
https://arxiv.org/abs/1907.07337.
[39] Karl Petersen. Ergodic theory, volume 2 of Cambridge Studies in Advanced Mathe-
matics. Cambridge University Press, Cambridge, 1989.
[40] Jean-Paul Pier. Amenable locally compact groups. Pure and Applied Mathematics
(New York). John Wiley & Sons, Inc., New York, 1984. A Wiley-Interscience Publi-
cation.
[41] Hans Reiter and Jan D. Stegemann,Classical Harmonic Analysis and Locally Compact
Groups.Clarendon Press (Oxford), 2000.
[42] Joseph Rosenblatt. Ergodic and mixing random walks on locally compact groups.
Math. Ann., 257(1):31–42, 1981.
[43] Walter Rudin.Fourier analysis on groups of John Wiley and Sons, 1962.
[44] Bertram M. Schreiber, Measures with bounded convolution powers. Trans. Amer.
Math. Soc. 151 (1970), 405–431.
[45] Kôsaku Yosida. Mean ergodic theorem in Banach spaces. Proc. Imp. Acad., 14(8):292–
294, 1938.
[46] K. Yosida, S Kakutani Operator-Theoretical Treatment of Markoff’s Process and
Mean Ergodic Theorem. Annals of Mathematics 42, (1):188–228, 1941.
[47] Misha Zafran. On the spectra of multipliers. Pacific J. Math. 47 (1973), 609–626.

Jorge Galindo, Instituto Universitario de Matemáticas y Aplicaciones (IMAC),

Universidad Jaume I, E-12071, Castellón, Spain.
E-mail: [email protected]

Enrique Jordá, EPS Alcoy, Instituto Universitario de Matemática Pura

y Aplicada IUMPA, Universitat Politècnica de València, Plaza Ferrándiz y
Carbonell s/n E-03801 Alcoy, Spain,
E-mail: [email protected]