j.

differential geometry
99 (2015) 191-213

ON MANIFOLDS SUPPORTING DISTRIBUTIONALLY

UNIQUELY ERGODIC DIFFEOMORPHISMS

Artur Avila, Bassam Fayad & Alejandro Kocsard

Abstract
A smooth diffeomorphism is said to be distributionally uniquely
ergodic (DUE for short) when it is uniquely ergodic and its unique
invariant probability measure is the only invariant distribution (up
to multiplication by a constant). Ergodic translations on tori are
classical examples of DUE diffeomorphisms. In this article we con-
struct DUE diffeomorphisms supported on closed manifolds dif-
ferent from tori, providing some counterexamples to a conjecture
proposed by Forni in [For08].

1. Introduction
When we study the dynamics of a homeomorphism f : M → M (for
the time being we can suppose M is a just compact metric space), we
can consider the induced linear automorphism f  on C 0 (M, C) given by
f  ψ := ψ ◦ f, ∀ψ ∈ C 0 (M, C).
If we endow C 0 (M, C) with the C 0 -uniform topology, f  turns out
to be a continuous linear operator and hence, its adjoint f acts on
the topological dual space (C 0 (M, C)) which coincides, by Riesz rep-
resentation theorem, with M(M ), the space of complex finite measures
on M .
To a certain extent we can say that ergodic theory consists in under-
standing the relation between the “non-linear” dynamics of f and the
linear one of f : M(M ) → M(M ). The fixed points of f , the so-called
f -invariant measures, play a key role in this theory.
When M is a closed smooth manifold and f : M → M is a C r -
diffeomorphism, every linear subspace C k (M, C) ⊂ C 0 (M, C) (where
0 ≤ k ≤ r ≤ ∞) is f  -invariant. Moreover, when C k (M, C) is equipped
with the C k -uniform topology, f  : C k (M, C) → C k (M, C) turns out to
be a continuous isomorphism and hence, its adjoint f acts on Dk (M ),
i.e. the space of distributions up to order k. Of course, the fixed points
of f are called invariant distributions.

Received 11/7/2012.

191
192 A. AVILA, B. FAYAD & A. KOCSARD

As usual, we say f is uniquely ergodic when it exhibits a unique f -

invariant probability measure. On the other hand, when f is C ∞ and
there is only one (up to multiplication by a constant) f -invariant distri-
bution, we shall say that f is distributionally uniquely ergodic, or DUE
for short. Ergodic translations on tori are the archetypical examples of
DUE diffeomorphisms. Recently, the first and third authors showed in
[AK11] that every smooth circle diffeomorphism with irrational rota-
tion number is also DUE.
In 2008, Forni conjectured in [For08] that tori are the only closed
manifolds supporting DUE diffeomorphisms.
In this paper we construct some new DUE systems, providing some
counterexamples to Forni’s conjecture. In fact, our main purpose con-
sists in showing the following

Theorem A. Let P be either

(a) a compact nilmanifold, i.e. P = N/Γ with N a nilpotent connected
and simply connected Lie group and Γ < N a uniform lattice; or
(b) a homogeneous space of compact type, i.e. P = G/H where G is
compact Lie group and H < G a closed subgroup.
Then there exist DUE diffeomorphisms on M := T × P .

It is interesting to remark that so far the most powerful techniques

to study invariant distributions (for dynamical systems which are not
hyperbolic) come from harmonic analysis. However, in general it is very
hard to apply these techniques to dynamical systems which do not ex-
hibit certain “homogeneity” (e.g. they preserve a smooth Riemannian
structure, or are induced by translations on homogeneous spaces).
On the other hand, it is well known that any DUE diffeomorphism
preserving a Riemannian structure is topologically conjugate to an er-
godic torus translation, and after some works of Flaminio and Forni
[FF03, FF07] it was expected that there were no DUE homogeneous
systems supported on homogeneous spaces different from tori. In fact, we
recently learned that Flaminio, Forni, and F. Rodrı́guez-Hertz [RH12]
have shown indeed the validity of Forni’s conjecture for homogeneous
systems. So the main difficulty to prove our result consists in overcoming
this apparent obstruction to apply harmonic analysis tools.

Acknowledgments. Since the larger part of this work was done while
A. Kocsard was visiting the Institut de Mathématique de Jussieu, in
Paris, he would like to thank P. Le Calvez for the hospitality. He is
also grateful to L. Flaminio for some insightful discussions. A.K. was
partially supported by CAPES and CNPq (Brazil).
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 193

2. Preliminaries and notations

2.1. Manifolds, functional spaces, and topology. All through this
paper, M will denote a compact orientable smooth manifold without
boundary. Given any r ∈ N0 ∪ {∞}, we write Diff r (M ) for the group
of C r -diffeomorphisms. The subgroup of C r -diffeomorphisms which are
isotopic to the identity shall be denoted by Diff r0 (M ).
If N denotes any other smooth manifold, we write C r (M, N ) for the
space of C r -maps from M to N . For the sake of simplicity, we shall just
write C r (M ) instead of C r (M, C).
Let us recall that, when r is finite, the uniform C r -topology turns
C r (M ) into a Banach space and C r (M, N ) turns out to be a Banach
manifold.
The space C ∞ (M ) will be endowed with its usual Fréchet topology
which can be defined as the projective limit of the family of Banach
spaces (C r (M ))r∈N . In this case, C ∞ (M, N ) is endowed with a Fréchet
manifold structure.
Of course, for any r ∈ N0 ∪ {∞}, we assume Diff r (M ) equipped with
the C r -uniform topology inherited from its inclusion in C r (M, M ).
Finally, if X is an arbitrary topological space and z ∈ X, we shall
write cc(X, z) to denote the connected component of X containing z.

2.2. Some arithmetical notations. Given any natural number q ∈

N, we write qN := {qn : n ∈ N}.
Whenever we write a single rational number in the form p/q, we
always assume the integers p and q are coprime, i.e. 1 is the great-
est common divisor of p and q. On the other hand, writing a vector
with rational coordinates (p1 /q, . . . , pn /q) ∈ Qn , we shall simply assume
gcd(p1 , . . . , pn , q) = 1.
Given any x ∈ R, we write x to denote the largest integer not
greater than x. Analogously, x denotes the smallest integer greater
than or equal to x.
For any α ∈ Rd we write

αTd := dist(α, Zd ).

Notice that, since α + nTd = αTd for every n ∈ Zd , we can naturally

consider  · Td as defined on Td , too.
We say α = (α1 , . . . , αd ) ∈ Rd is irrational when for every
(n1 , . . . , nd ) ∈ Zd

 d 
 
(1)  ni αi  = 0 =⇒ ni = 0, for i = 1, . . . , d.
 
i=1 Td
194 A. AVILA, B. FAYAD & A. KOCSARD

An irrational vector α is said to be Diophantine if there exist con-

stants C, τ > 0 satisfying
 d 
  C
 α q  ≥ ,
 i i  maxi |qi |τ
i=1 Td

for every (q1 , . . . , qd ) ∈ Zd \{0}. On the other hand, an irrational element

of Rd which is not Diophantine is called Liouville.

2.3. Lie groups.

2.3.1. Generalities. In this work we shall only deal with real con-
nected Lie groups. As usual, if G denotes an arbitrary Lie group, its
identity element is denoted by 1G (or just 0 when G is abelian), its Lie
algebra by g, and we write exp : g → G for the exponential map.
A smooth manifold is called a homogeneous space when it can be
written as G/H, where G denotes a (real, connected) Lie group and H <
G a closed subgroup. We say H is cocompact when G/H is compact,
and we say that G/H is of compact type when G is compact itself.
Clearly, the group G acts naturally (on the left) on G/H and it is well
known that in such a case there exists at most one G-invariant Borel
probability measure on G/H. When such a measure does exist, we will
call it the Haar measure of G/H. A discrete cocompact subgroup will
be called a uniform lattice. Let us recall that the existence of the Haar
measure on G/H is guaranteed whenever either G is compact, or H is
a uniform lattice.
Making some abuse of notation, we will use the brackets [·, ·] to denote
the Lie brackets on g, as well as the commutator operator in G, i.e.
[g, h] := ghg −1 h−1 , for any g, h ∈ G.
More generally, if A, B ⊂ G we define [A, B] := [a, b] : a ∈ A, b ∈ B,
i.e. the subgroup of G generated by commutators of the set subsets A
and B, respectively. And analogously, if h, k ⊂ g, we define [h, k] :=
spanR {[v, w] : v ∈ h, w ∈ k}.
The centers of G and g are defined by Z(G) := {g ∈ G : [g, h] =
1G , ∀h ∈ G} and Z(g) := {v ∈ g : [v, w] = 0, ∀w ∈ g}, respectively.

2.3.2. Tori. The d-dimensional torus will be denoted by Td and will be

identified with Rd /Zd . The canonical quotient projection will be denoted
by π : Rd → Td . For simplicity, we shall simply write T for the 1-torus,
i.e. the circle.
The symbol Lebd will be used to denote the Lebesgue measure on Rd ,
as well as the Haar measure on Td . Once again, for the sake of simplicity,
we just write Leb, and also dx, instead of Leb1 .
For each α ∈ Td , let Rα : Td → Td be the rigid translation Rα : x →
x + α.
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 195

2.3.3. Homogeneous skew-products. Given an arbitrary Lie group

G and any closed subgroup H < G, for any α ∈ T and any γ ∈ C r (T, G),
we define the homogeneous skew-product Hα,γ ∈ Diff r (T × G/H) by

Hα,γ : (t, gH) → (t + α, γ(t)gH), ∀(t, gH) ∈ T × G/H.

The space of C r homogeneous skew-products on T × G/H shall be

denoted by SWr (T × G/H). At this point, it is important to notice that
SWr (T × G/H) is a subgroup and it is closed in Diff r (T × G/H).
Moreover, if G/H admits a Haar measure, and we denote it by ν, then
we clearly have SWr (T × G/H) ⊂ Diff rμ (T × G/H), where μ := Leb ⊗ ν.

2.3.4. Nilmanifolds and Mal’cev theory. Given an arbitrary Lie

group G, the central descending series of G can be recursively defined
by G0 := G and
Gn := [G, Gn−1 ], ∀n ≥ 1.
We say G is nilpotent when Gk = {1G }, for some k ∈ N. The degree of
nilpotency of G is defined as the maximal natural number n such that
Gn = {1G }.
From now on, N shall denote a connected, simply connected nilpotent
Lie group admitting a uniform lattice Γ < N . As usual, the compact
homogeneous space N/Γ is called a (compact) nilmanifold.
It is important to recall that in such a case the exponential map
exp : n → N is a real-analytic diffeomorphism (see Theorem 1.2.1 in
[CG90]). Hence, in this case N as well as n can be identified with the
universal cover of N/Γ.
After Mal’cev [Mal49], a basis {v1 , v2 , . . . , vd } of the Lie algebra n is
called a Mal’cev basis whenever n(i) := spanR {v1 , . . . , vi } is an ideal in
n, for each i ∈ {1, . . . , d}. Moreover, such a basis is said to be strongly
based on Γ when

(2) Γ = exp(Zv1 ) exp(Zv2 ) . . . exp(Zvd ).

In [Mal49] (see also [CG90]) it is proved that there always exists a

Mal’cev basis strongly based on Γ when N and Γ are as above.
Since each n(i) is an ideal in n, N(i) := exp(n(i) ) ⊂ N turns out to
be a (closed) normal subgroup of N , and the quotient N (i) := N/N(i) a
nilpotent connected and simply connected Lie group, itself.
On the other hand, as a consequence of (2) we have
 
Γ(i) := exp Zv1 ⊕ Zv2 ⊕ · · · ⊕ Zvi } ⊂ N(i)

is a discrete subgroup of Γ. Hence, Γ(i) := Γ/Γ(i) can be naturally

identified with a uniform lattice of N (i) .
196 A. AVILA, B. FAYAD & A. KOCSARD

2.4. Distributions and distributional unique ergodicity. Given

any k ∈ N0 , the space of distribution on M up to order k is defined as
the topological dual space of C k (M ) and will be denoted by Dk (M ).
When k = 0, by Riesz representation theorem D0 (M ) can be identified
with the space of finite complex measures on M and so it will also be
denoted by M(M ).
On the other hand, as it is usually done, the topological dual space
of C ∞ (M ) will be denoted simply by D  (M ) and its elements are just
called distributions.
Since all the inclusions C k+1 (M ) → C k (M ) and C ∞ (M ) → C k (M )
are continuous, making some abuse of notation we can consider the
following chain of inclusions (modulo restrictions):
M(M ) = D0 (M ) ⊂ D1 (M ) ⊂ D2 (M ) ⊂ . . . ⊂ D  (M ).
Moreover, since we are assuming M is compact, it is well known that

D  (M ) = Dk (M ).
k≥0

Now, as already mentioned in §1, any f ∈ Diff k (M ) acts linearly on

C k (M ) by pull-back, and the adjoint of this action is the linear operator
f : Dk (M ) → Dk (M ) given by


f T, ψ := T, f  ψ = T, ψ ◦ f , ∀ T ∈ Dk (M ), ∀ ψ ∈ C k (M ).

The space of f -invariant distributions up to order k is defined by
Dk (f ) := {T ∈ Dk (M ) : f T = T }.

Of course, when f is C ∞ we write D  (f ) := k≥0 Dk (f ).
Given any measure μ ∈ M(M ) and r ≥ 0, we define
Diff rμ (M ) := {f ∈ Diff r (M ) : f μ = μ}
and 
Cμr (M ) := φ ∈ C r (M ) : φ dμ = 0 .
M
As usual, we say that f is uniquely ergodic when M(f ) := D0 (f ) is
one-dimensional. We will say that f is distributionally uniquely ergodic
(or just DUE for short) when f is C ∞ and D  (f ) has dimension one.
2.4.1. Coboundaries and distributions. Given any f ∈ Diff r (M ),
any ψ : M → C, and n ∈ Z, the Birkhoff sum is defined by
⎧n−1
⎪
⎨ i=0 ψ ◦ f
i if n ≥ 1;
S ψ = Sf ψ := 0
n n
if n = 0;
⎪
⎩ −n −i
− i=1 ψ ◦ f if n < 0.
We say that ψ is a C  -coboundary for f (with 0 ≤ ≤ r) whenever
there exists u ∈ C  (M ) solving the following cohomological equation:
u ◦ f − u = ψ.
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 197

Observe that in this case it holds that S n ψ = u ◦ f n − u, for any n ∈ Z.

The space of C  -coboundaries will be denoted by B(f, C  (M )). Fol-
lowing Katok [Kat01], we say f is cohomologically C  -stable whenever
B(f, C  (M )) is closed in C  (M ).
Finally, notice that as a straightforward consequence of Hahn-Banach
theorem we get
Proposition 2.1. Given any f ∈ Diff k (M ), with k ∈ N0 ∪ {∞}, it
holds 
clk (B(f, C k (M ))) = ker T,
T ∈Dk (f )

where clk (·) denotes the closure in C k (M ).

2.4.2. Unique ergodicity vs. DUE. There are many well-known ex-
amples of uniquely ergodic systems which are not DUE. Maybe the
simplest one is given by the parabolic map
T2  (x, y) → (x + α, y + x),
with α ∈ T \ (Q/Z) (see [Kat01] for details). Horocycle flows on con-
stant negatively curved closed surfaces and minimal homogeneous flows
on closed nilmanifolds different from tori are more elaborate examples
[FF03, FF07].
On the other hand, a classical result due to Kronecker affirms that
a translation Rα : Td → Td is uniquely ergodic (Lebd is the only Rα -
invariant probability measure) if and only if α = (α1 , . . . , αd ) is irra-
tional.
Moreover, we have the following result, which belongs to the folklore:
Proposition 2.2. Rα is DUE, and it is cohomologically C ∞ -stable
if and only if α is a Diophantine vector.
Proof. See Proposition 2.3 in [AK11] for a proof. q.e.d.
Recently, the first and third authors extended this result in [AK11]
showing that any minimal circle diffeomorphism is also DUE (see [NT12]
for a much simpler proof of non-existence of invariant distributions up
to order 1).
As already mentioned in §1, the main aim of this paper consists in
constructing DUE diffeomorphisms which are not topologically conju-
gate to ergodic translations on tori.

3. Proof of Theorem A: general strategy

As already mentioned in §2.3, in both cases considered in Theorem A
the homogeneous space P admits a Haar measure that will be denoted
by νP . Then the Haar measure of M , which is a homogeneous space
itself, is given by μ := Leb1 ⊗ νP .
198 A. AVILA, B. FAYAD & A. KOCSARD

Let us recall that any homogeneous skew-product on M = T × P

(see §2.3.3) preserves the measure μ. In other words, SWr (T × P ) ⊂
Diff rμ (M ).

3.1. The Anosov-Katok space. Let us consider the horizontal T-

action T : T × M → M given by
 
Tα (t, p) = T α, (t, p) = (t + α, p), ∀(t, p) ∈ M = T × P, ∀α ∈ T.
Then we define the Anosov-Katok space
 
(3) AK∞ (T ) := cl∞ H ◦ Tα ◦ H −1 : α ∈ T, H ∈ SW∞ (T × P ) .
Observe that each Tα ∈ SW∞ (T × P ) and hence, AK∞ (T ) ⊂ SW∞ (T ×
P ).
To prove Theorem A we will show
Theorem 3.1. Generic diffeomorphisms in AK∞ (T ) are DUE. More
precisely, the set
 
DUE(T ) := f ∈ AK∞ (T ) : dim D  (f ) = 1
contains a dense Gδ -subset of AK∞ (T ).
Let us now describe the general strategy to prove Theorem 3.1:
A family (Vn )n≥1 will be called a filtration of Cμ∞ (M ) whenever it
satisfies:
• for every n ≥ 1, Vn ⊂ Cμ∞ (M ) is a closed linear subspace;
Vn ⊂ Vn+1 , for every n ≥ 1;
• 
• n≥1 Vn is dense in Cμ∞ (M ).
Sections 4 and 5 are dedicated to the proof of the following lemma,
in the nilpotent and the compact cases, respectively:
Lemma 3.2. If P is as in Theorem A, then there exists a filtration
of Cμ∞ (M ), called (Vk ), satisfying the following condition:
For every n ∈ N and every q0 ∈ N, there exists q̄ ∈ N and a homoge-
neous skew-product H0,γ ∈ SW∞ (T × P ) such that:
(i) H0,γ ◦ T1/q0 = T1/q0 ◦ H0,γ ;
 −1
(ii) Vn ⊂ B H0,γ ◦ Tp̄/q̄ ◦ H0,γ , C ∞ (M )), for every p̄ ∈ Z coprime with
q̄.
To prove Lemma 3.2, we shall need the following elementary
Lemma 3.3. Let M be an arbitrary manifold, f : M → M a periodic
C r -map (i.e. there exists q ∈ N such that f q = idM ), and φ ∈ C k (M ),
with 0 ≤ k ≤ r ≤ ∞. Then, φ ∈ B(f, C k (M )) iff

q−1
(4) Sfq φ(x) = φ(f j (x)) = 0, ∀x ∈ M.
j=0
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 199

Proof of Lemma 3.3. If φ ∈ B(f, C k (M )), then there exists u ∈ C k (M )

such that φ = u ◦ f − u. Hence, Sfq φ(x) = u(f q (x)) − u(x) = 0, for every
x ∈ M.
Reciprocally, let us suppose (4) holds. Then, using a formula we
learned from [MOP77], we write

1 j
q
v(x) := − Sf φ(x), ∀x ∈ M.
q
j=1

It clearly holds that v ∈ C k (M ), and


q 
1  j 
v(f (x)) − v(x) = − Sf φ(f (x)) − Sfj φ(x)
q
j=1
 
1 q
= − Sf φ(f (x)) − qφ(x) = φ(x),
q

for every x ∈ M . Thus, φ ∈ B(f, C k (M )). q.e.d.

Now, assuming Lemma 3.2, we can prove Theorem 3.1, and hence-
forth, Theorem A, too:

Proof of Theorem 3.1. Let (φm )m∈N be a dense sequence in Cμ∞ (M ) and
define

1
Am := f ∈ AK∞ (T ) : ∃u ∈ C ∞ (M ), u ◦ f − u − φm C m < .
m

Each set Am is clearly open in AK∞ (T ), and by Proposition 2.1, it holds

that

DUE(T ) = Am .
m≥1

Thus, we have to show that each Am is dense in AK∞ (T ). To do that,

consider a fixed set Am , any rational number p0 /q0 , and an arbitrary
homogeneous skew-product H ∈ SW∞ (T × P ).
Let (Vn )n≥1 be a filtration satisfying Lemma 3.2. Then there exists
n ∈ N and φ ∈ Vn such that
 
(5) φ ◦ H −1 − φm  m < 1 .
C m
Now, invoking Lemma 3.2, from n and q0 , we obtain a natural number
q̄ and a homogeneous skew-product H0,γ satisfying (i) and (ii). Then,
200 A. AVILA, B. FAYAD & A. KOCSARD

for each ∈ N, let us define

p̂ := q̄p0 + 1,
q̂ := q̄q0 ,
p̂
p := ,
gcd(p̂ , q̂ )
q̂
q := .
gcd(p̂ , q̂ )
Notice that, for each , p and q are coprime, q is a multiple of q̄, and
p p0
q → q0 , as → +∞.
Then observe that, for every ∈ N and any (t, x) ∈ T × P it holds:
q
SH −1 φ(t, x)
0,γ Tp /q H0,γ

 
 −1
jp  jp 
q
−1
= φ t+ ,γ t + γ(t) x
q q
j=0
 
 −1
j  j
q
(6) −1
= φ t + ,γ t + γ(t) x
q q
j=0

  
q /q̄−1 q̄−1
s r  s r

= φ t + + ,γ t + + γ(t)−1 x
r=0 s=0
q̄ q q̄ q
= 0,
where the last equality is a consequence of condition (ii) of Lemma 3.2
−1
and Lemma 3.3. Thus, we conclude that φ ∈ B(H0,γ Tp /q H0,γ , C ∞ (M )).
Henceforth,
 
−1 −1
(7) φ ◦ H −1 ∈ B HH0,γ Tp /q H0,γ H , C ∞ (M ) ,

for every ∈ N.
On the other hand, Tp /q → Tp0 /q0 in Diff ∞ (M ), as → ∞. Hence,
from (i) of Lemma 3.2, we get
−1 −1 C∞
(8) HH0,γ Tp /q H0,γ H −−→ HTp0 /q0 H −1 , as → ∞.
Now, putting together (5), (7), and (8), we conclude that HTp0 /q0
H −1∈ cl∞ (Am ), as desired. q.e.d.

3.2. Real-analytic DUE diffeomorphisms. Before starting with the

proof of Lemma 3.2, it is interesting to remark that using the techniques
we applied in §3.1 it is possible to prove the existence of real-analytic
DUE diffeomorphisms on M = T × P .
In fact, for the time being let us suppose G is an arbitrary Lie group
admitting a complexification of G that will be denoted by GC .
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 201

Then, for each Δ > 0, let us define CΔ ω (T, G) as the set of real-

analytic functions γ : T → G that admit a holomorphic extension from

the complex band AΔ := {z ∈ C : |Im z| ≤ Δ}/Z to GC . Let us consider
CΔω (T G) endowed with the distance function d given by
Δ

dΔ (γ0 , γ1 ) := sup dGC (γ0 (z), γ1 (z)), ∀γ0 , γ1 ∈ CΔ

ω
(T, G),
z∈AΔ

where dGC denotes a left invariant distance on GC .

Then, taking into account that, for any Δ > 0, CΔ ω (T, G) is dense
∞
in C (T, G), repeating the same argument used in the proof of Theo-
rem 3.1, we can easily show that the set

DUEΔ (T ) := (α, γ) ∈ T×CΔ
ω ω
(T, G) :

−1
H0,γ ◦ Tα ◦ H0,γ ∈ SWω (T × P ) is DUE

is generic in T × CΔ
ω (T, G), and in particular, non-empty.

4. The nilpotent case

All along this section, let us assume P is a compact nilmanifold equal
to N/Γ, where N is a (connected) simply connected nilpotent Lie group
and Γ < N is a uniform lattice.
Observe that any complex function on P can be lifted to its univer-
sal covering, which can be identified with N itself, getting a Γ-invariant
complex function with respect to the Γ-action on N given by right trans-
lations. So, we can naturally identify C ∞ (P ) with
CΓ∞ (N ) := {φ ∈ C r (N ) : φ(xg) = φ(x), ∀x ∈ N, ∀g ∈ Γ} .
Moreover, since the exponential map exp : n → N is a real-analytic
diffeomorphism, we can identify C ∞ (N ) with C ∞ (n), and henceforth,
C ∞ (P ) with a closed linear subspace of C ∞ (N ) = C ∞ (n).
Let V = {v1 , . . . , vd } be a Mal’cev basis of n strongly based on Γ (see
§2.3.4 for details). Fixing this basis, we identify C ∞ (n) with C ∞ (Rd ),
simply writing
 d 

φ xi vi = φ(x1 , x2 , . . . , xd ), ∀(x1 , . . . , xd ) ∈ Rd .
i=1

Thus, making some abuse of notation, we shall assume that

(9) C ∞ (P ) ⊂ C ∞ (N ) = C ∞ (n) = C ∞ (Rd ).
Now let us analyze some of the equivariant conditions a function
φ ∈ C ∞ (Rd ) must satisfy to belong to C ∞ (P ). First, since v1 ∈ Z(n)
(and exp(Zv1 ) ∈ Γ), we conclude that, if φ ∈ C ∞ (P ) ⊂ C ∞ (Rd ), then it
202 A. AVILA, B. FAYAD & A. KOCSARD

is Z-periodic in its first variable. Hence, we can consider the Fourier-like

development
 (1)
(10) φ(x1 , x2 , . . . xd ) = φ̂k (x2 , . . . , xd )e2πikx1 ,
k∈Z

(1) (1)
where each φ̂k ∈ C ∞ (Rd−1 ). Here, the 0th Fourier-function φ̂0 has a
particularly nice interpretation: it can be naturally considered as defined
on the nilpotent Lie group N (1) := N/N(1) , or more precisely, on the
compact nilmanifold N (1) /Γ(1) (see §2.3.4 for these notations).
On the other hand, observe that the basis {v2 +n(1) , v3 +n(1) , . . . , vd +
n(1) } is a Mal’cev one for n/n(1) strongly based on the lattice Γ(i) =
Γ/Γ(i) .
(1)
That means we can repeat our previous argument to prove that φ̂0 is
Z-periodic on its first variable, and hence, we can consider the Fourier-
like development
(1)
 (2)
φ̂0 (x2 , . . . , xd ) = φ̂k (x3 , . . . , xd )e2πikx2 .
k∈Z

(2)
Once again, the Fourier-coefficient function φ̂0 can be considered as
an element of C ∞ (N (2) /Γ(2) ) and the set {v3 + n(2) , . . . , vd + n(d) } is a
Mal’cev basis strongly based on Γ(2) .
By induction, we get a family of Fourier-like coefficients
(j)
φ̂k ∈ C ∞ (Rd−j ), ∀j ∈ {1, . . . , d}, ∀k ∈ Z,
(j)
where each φ̂0 ∈ C ∞ (N (i) /Γ(i) ) ⊂ C ∞ (Rd−j ) and satisfies
(j)
 (j+1)
φ̂0 (xj+1 , . . . xd ) = φ̂k (xj+2 , . . . , xd )e2πikxj+1 .
k∈Z

Now we proceed to define the pseudo-polynomials on P : we shall say

that φ ∈ C ∞ (P ) is a pseudo-polynomial (with respect to V) of degree
less than or equal to n ∈ N iff
(j)
φ̂k ≡ 0, for every j ∈ {1, . . . , d} and |k| > n.
The linear space of pseudo-polynomials on P will be denoted by Pol(P )
and we shall write Poln (P ) for the subspace of pseudo-polynomials with
degree at most n.
Notice that M = T × P is a compact nilmanifold itself; hence we
can talk about pseudo-polynomials on M . In this case, we shall add
the vector v0 := ∂t (which generated the Lie algebra of R) to the basis
V, and hence any function φ ∈ C ∞ (M ) will be written in coordinates
(t, x1 , . . . , xd ), being φ Z-periodic on its first coordinate, too. So we can
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 203

also consider the Fourier-like development

 (0)
φ(t, x1 , . . . , xd ) = φ̂k (x1 , . . . , xd )e2πikt ,
k∈Z

(0)
with each φ̂k ∈ C ∞ (P ). Of course, by analogy with our definition of
pseudo-polynomials on P , we define
(0) (0)
Poln (M ) := {φ ∈ C ∞ (M ) : φ̂0 ∈ Poln (P ), φ̂k ≡ 0, ∀ |k| > n}.
Combining an inductive argument on the dimension of M with clas-
sical Fourier theory, one can easily show
Proposition 4.1. The linear space

Pol(M ) = Poln (M )
n≥0

is dense in C ∞ (M ). In particular, this implies that the family (Vn ) given

by
Vn := Poln (M ) ∩ Cμ∞ (M ), ∀n ≥ 1
is a filtration of Cμ∞ (M ).

Now we will prove Lemma 3.2, assuming the filtration (Vn ) is given
by Proposition 4.1:

Proof of Lemma 3.2 in the nilpotent case. Let us write d := dim N . We

will recursively define, for k ∈ {0, 1, . . . , d}, two sequences γk ∈ C ∞ (T, N )
and (q̄k ) ⊂ N satisfying the following condition: there exists a constant
Ck ∈ R such that for every p ∈ Z coprime with q̄k and every φ ∈ Vn ,
(k)
(11) q
SH −1 φ(t, x) = Ck φ̂0 (γk (t)−1 xN (k) ), ∀(t, x) ∈ T × N,
k Tp/q̄k Hk

(k)
where Hk = H0,γk ⊂ SW∞ (T × P ) and considering φ̂0 as a complex
function on N (k) = N/N(k) (see §2.3.4 for notations).
At this point it is important to notice that, since φ ∈ Vn ⊂ Cμ∞ (M ),
then
(k)
φ̂0 dν = 0, for every 0 ≤ k ≤ d.
P
(d)
Thus, in particular, the complex number φ0 is equal to zero, and so
condition (11) for k = d means that the Birkhoff sum vanishes. By
Lemma 3.3, this is equivalent to φ ∈ B(Hd Tp/q̄d Hd−1 , C ∞ ).
Now let us start with the case k = 0. Observe that without loss of
generality we can assume n < q0 . Let us define γ0 ≡ 1N (so H0 = idM )
and q̄0 := q0 . Hence, for every φ ∈ Vn , and every p ∈ Z coprime with q̄0 ,
204 A. AVILA, B. FAYAD & A. KOCSARD

we have
0 −1
q̄  
q̄0 p
SH −1 φ(t, x) = φ t + j ,x
0 Tp/q̄0 H0 q̄0
j=0
(12)
0 −1
q̄  (0) 2πi(t+ q̄j ) (0)
= φ̂ (x)e 0 = q̄0 φ0 (x),
j=0 ||≤n

for every (t, x) ∈ R × Rd . So condition (11) is verified for k = 0.

Now, suppose we have already defined γk−1 ∈ C ∞ (T, N ) (and then,
Hk−1 = H0,γk−1 ∈ SW∞ (T × G/H)) and q̄k−1 ∈ N, with 1 ≤ k ≤ d; then
let us construct γk and q̄k .
To do this, we start considering an auxiliary C ∞ -function ρ : [0, 1] →
[0, 1] satisfying the following conditions:
(i) ρ(t) = 0, for every t in a neighborhood of 0;
(ii) ρ(t) = 1, for every t in a neighborhood of 1;
(iii) ρ (t) ≥ 0, for all t;
(iv) ρ(1 − t) = 1 − ρ(t), for all t.
Then we use ρ to define a new auxiliary function η : [0, 1] → R as
follows:
   2q0 t  
ρ 2q0 t − 2q0 t + q0 , if t ∈ 0, 12 ;
η(t) :=    
ρ 2q0 (1 − t) − 2q0 (1 − t) + 2q0 (1−t)
q0 , if t ∈ 12 , 1 .

In this way, η turns out to be smooth and it vanishes in some neigh-

borhoods of 0 and 1, so we can consider it as an element of C ∞ (T, R).
Observe that for any m ∈ Z \ {0} with |m| < q0 and any t ∈ T, it holds
that
2q
 0 −1

e2πimη(t+/2q0 )
=0
(13)
0 −1 
q 
= e2πim(η(t)+/q0 ) + e−2πim(η(t)+/q0 ) = 0.
=0

Then we define
q̄k := 2q̄k−1 q0 ,
and γk ∈ C ∞ (T, N ) by

γk (t) := γk−1 (t) exp(η(q̄k−1 t)vk ), ∀t ∈ T.

Assuming the inductive hypothesis, let us prove condition (11) holds

for k, too. Let p ∈ Z be any number coprime with q̄k , φ ∈ Vn arbitrary,
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 205

and (t, x) be any point in T × N . Then we have:

(14)
  
k −1
j
q̄
q̄k j −1
SH −1 φ(t, x) = φ t + , γk t + γk (t) x
k Tp/q̄k Hk q̄k q̄k
j=0

 
2q0 −1 q̄k−1 −1
  j
  j
 
−1
= φ t+ + , γk t + + γk (t) x
q̄k q̄k−1 q̄k q̄k−1
=0 j=0
2q
 0 −1      
(k−1) −1 (k−1)
= Ck−1 φ̂0 exp η q̄k−1 t + vk γk (t) xN
2q0
=0
2q
 0 −1    
(k−1)
= Ck−1 φ̂0 x̃k + η q̄k−1 t + , x̃k+1 , . . . , x̃d
2q0
=0
2q
 0 −1 
= Ck−1 φ̂(k)
m (x̃k+1 , . . . , x̃d )e
2πim(x̃k +η(q̄k−1 t+/2q0 ))

=0 |m|≤n

 2q
 0 −1

= Ck−1 φ̂(k)
m (x̃k+1 , . . . , x̃d )e
2πimx̃k
e2πimη(t+/2q0 )
|m|≤n =0
 
(k) (k)
= q0 Ck−1 φ̂0 (x̃k+1 , . . . , x̃d ) = q0 Ck−1 φ̂0 γk (t)−1 xN (k) ,
where the sixth equality is a consequence of (13) and where (x̃k , x̃k+1 ,
. . . , x̃d ) denotes the “coordinates” of the point γk+1 (t)−1 xN (k−1) in the
Lie algebra n(k−1) , i.e. they satisfy the following equation:
x̃k vk + x̃k+1 vk+1 + . . . + x̃d vd + n(k−1) = exp−1 (γ (t)−1 xN (k−1) ).
N (k−1) k
In this way, (14) shows condition (11) holds for k, finishing the proof
of the lemma. q.e.d.

5. The compact case

We start this section with a geometric construction which is com-
pletely independent of the homogeneous structure of the supporting
manifold. So, for the time being, let us suppose M is an arbitrary smooth
connected closed manifold and μ any Borel probability measure on M .

5.1. Equidistributed loops. Given a finite dimensional subspace E ⊂

Cμ∞ (M ), a smooth loop γ ∈ C ∞ (T, M ) is said to be E-equidistributed
when there exists m ∈ N such that
  
m−1
j

(15) φ γ t+ = 0, ∀t ∈ T, ∀φ ∈ E.
m
j=0

The number m will be called the period of the loop.

206 A. AVILA, B. FAYAD & A. KOCSARD

Theorem 5.1. For every finite dimensional subspace E ⊂ Cμ∞ (M ),

there exists a smooth E-equidistributed loop θ ∈ C ∞ (T, M ).
Proof. Let N := dim E and (φ1 )N i=1 be a basis of E, and define
Φ : M → RN by Φ(x) := (φ1 (x), . . . , φN (x)), for every x ∈ M . For
each m ∈ M , let us write

m
(m)
Φ (x1 , . . . , xm ) := Φ(xj ) ∈ RN , ∀(x1 , . . . , xm ) ∈ M m .
j=1

Let us consider the sets Y (m) , Z (m) ⊂ M m given by

 
(16) Y (m) := x̄ ∈ M m : DΦm (x̄) : Tx̄ M m → RN is surjective ,
 
(17) Z (m) := x̄ ∈ M m : Φm (x̄) = 0 ∈ RN ,
and then define X (m) := Y (m) ∩ Z (m) .
We divide the rest of the proof into several lemmas:
Lemma 5.2. For every n ≥ N , the set Y (n) is non-empty.
Proof of Lemma 5.2. First observe that, given any n ≥ 1,

n
(n)
(18) DΦ (x1 ,...,xn ) (v1 , . . . , vn ) = DΦxj (vj ),
j=1

for every (x1 , . . . , xn ) ∈

Mnand every (v1 , . . . , vn ) ∈ T(x1 ,...,xn ) M n . This
lets us affirm that for any k, n ∈ N and any “forget-some-coordinate”
projection pr : M n+k → M n , it holds that
(19) pr−1 (Y (n) ) ⊂ Y (n+k) .
That means it is enough to show Y (N ) is non-empty. Reasoning by
contradiction, suppose this is not the case. By (18), this implies the set
{DΦx (v) : x ∈ M, v ∈ Tx M }
is contained in a proper linear sub-space of RN , and therefore, we can
find a non-identically zero linear functional L : RN → R such that D(L◦
Φ)x = 0, for every x ∈ M . Of course, since we are assuming M is
connected, this implies L◦Φ : M → RN is a constant function. Since the
coordinate functions of Φ (i.e. functions φ1 , . . . , φN ) have zero integral
with respect to μ, we conclude that L ◦ Φ ≡ 0, contradicting the linear
independence of the set (φi )N (N ) = ∅, and by (19), we get
i=1 . So, Y
Y (n) = ∅, for every n ≥ N . q.e.d.
Lemma 5.3. There exists m ∈ N such that X (m) is non-empty.
Proof of Lemma 5.3. Consider the set
  n
CΦ := λj Φ(xj ) ∈ RN : xj ∈ M, λj > 0, ∀j ∈ {1, . . . , n} .
n≥1 j=1
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 207

Observe CΦ is a convex cone in RN . We claim CΦ = RN . In fact, if

this would not be the case, then there should exist a non-null linear
functional L : RN → R such that L(y) ≥ 0 for every y ∈ CΦ ⊂ RN and,
in particular, L(Φ(x)) ≥ 0, for every x ∈ M . But since the coordinate
functions of Φ belong to Cμ∞ (M ), it holds that

L(Φ(x)) dμ = 0.
M
Hence, L ◦ Φ should be identically equal to zero, contradicting the linear
independence of the coordinate functions (φi )N i=1 . Thus, CΦ = R .
N

By Lemma 5.2, Y (N ) is non-empty, so we can consider an arbitrary

point (z1 , . . . , zN ) ∈ Y ) . By our previous assertion about CΦ , there
(N

exist n ∈ N, positive numbers λ1 , . . . , λn , and points x1 , . . . , xn ∈ M

such that
 n  N
(20) λj Φ(xj ) = −ΦN (z1 , . . . , zN ) = − Φ(zj ).
j=1 j=1

Now, since ΦN (Y (N ) ) is open in RN , we can assume (up to an ar-

bitrary small perturbation of the points z1 , . . . , zN ) that each λj ∈ Q,
and hence we can find p1 , . . . , pn , q ∈ N such that λj = pj /q, for each

1 ≤ j ≤ n. Now, we define m := qN + 1≤j≤n pj and we claim X (m) = ∅.
In fact, if we define (w1 , . . . , wm ) ∈ M m by

z j/q , if 1 ≤ j ≤ qN,
wj := 
xk , if qN < j ≤ qN + k=1 p ,

from (20) we easily conclude (w1 , . . . , wm ) ∈ X (m) . q.e.d.

Now, for each n ≥ 2, let us consider the diffeomorphism σn : M n →
Mn given by
σn (x1 , x2 , . . . , xn ) := (x2 , . . . , xn , x1 ), ∀(x1 , . . . , xn ) ∈ M n .
And we shall prove our last
Lemma 5.4. There exist m ≥ 1 and z̄ ∈ X (m) such that σm (z̄) ∈
cc(X (m) , z̄).
Proof of Lemma 5.4. Let m0 be a natural number such that X (m0 ) is
non-empty, and let z̄ = (x1 , . . . , xm0 ) be any point in X (m0 ) . For each
(q) (q) (q)
q ∈ N, let us consider the point z̄ (q) = (z1 , z2 , . . . , zqm0 ) ∈ X (qm0 )
given by
(q)
zj := x j/q , for 1 ≤ j ≤ qm0 .
We claim that σqm0 (z̄ (q) ) ∈ cc(X (qm0 ) , z̄ (q) ), provided q is sufficiently
large. To prove this, we shall construct a continuous curve ρ : [0, 1] →
X (qm0 ) ⊂ M qm0 , with ρ(0) = z̄ (q) and ρ(1) = σqm0 (z̄ (qm0 ) ).
208 A. AVILA, B. FAYAD & A. KOCSARD

For each 1 ≤ j ≤ qm0 , we write ρj : [0, 1] → M for the j th -coordinate

function of ρ, i.e. ρ(t) = (ρ1 (t), ρ2 (t), . . . , ρqm0 (t)) ∈ M qm0 .
We start defining ρ on the interval [0, 1/2]. To do that, first let us
consider continuous paths α = (α1 , . . . , αm0 ) : [0, 1] → M m0 such that

αi (0) = xi , αi (1) = xi+1 , for 1 ≤ i < m0 ,

αm0 (0) = xm0 , αm0 (1) = x1 .

Now, we choose a (small) neighborhood U of z̄ in M m0 such that

U ⊂ Y (m0 ) and U ∩ X (m0 ) is connected. Since Φ(m0 ) is a submersion on
Y (m0 ) , we can find a continuous path β : [0, 1] → U ⊂ M m0 satisfying
β(0) = z̄ and

(m0 )
  Φ(m0 ) α(t))
(21) Φ β(t) = − , ∀t ∈ [0, 1],
q−1

provided q is sufficiently large. Notice that, since α(1) ∈ X (m0 ) , then

β(1) also belongs to X (m0 ) itself.
Then we define each coordinate function of the path ρ on [0, 1/2] by

β j/q (2t), if 1 ≤ j < m0 q, and j ∈ qZ,
ρj (t) :=
αj/q (2t), if j ∈ {q, 2q, . . . , m0 q},

and every t ∈ [0, 1/2]. Notice that, as a consequence of (21), ρ(t) ∈

X (qm0 ) , for every t ∈ [0, 1/2].
In order to define path ρ on [1/2, 1], let us consider a continuous path
γ : [0, 1] → X (m0 ) joining β(1) to z̄. Such a path γ does exist because
both points β(1) and z̄ belong to U ∩ X (m0 ) , which is a connected open
set of the smooth manifold X (m0 ) , and hence it is arc-wise connected.
Finally, we define ρ on [1/2, 1] by
⎧
⎪
⎨γ j/q (2t − 1), if j ∈ qZ,
ρj (t) := xj+1 , if j ∈ qZ and 1 ≤ j < qm0 ,
⎪
⎩
x1 , if j = qm0 .

In this way, ρ is clearly a continuous path contained in X (qm0 ) and joins

ρ(0) = z̄ to ρ(1) = σqm0 (z̄), as desired. q.e.d.

Finally, let m ≥ 1 and z̄ ∈ X (m) as in Lemma 5.4. Since X (m) ⊂

Mm is a σm -invariant embedded submanifold, and σm is an m-periodic
diffeomorphism, we can find a smooth loop θ̃ ∈ C ∞ (T, X (m) ) satisfying
 
1
(22) θ̃ t + = σm (θ̃(t)), ∀t ∈ T.
m
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 209

Then, for any t ∈ T, if we write θ(t) = (θ1 (t), . . . , θm (t)) ∈ M m , it holds

that
  
  j
m m
(m)
R  0 = Φ (θ1 (t), . . . , θm (t)) =
N
Φ(θj (t)) = Φ θ1 t + ,
m
j=1 j=1

where the last equality is a consequence of (22).

Thus, θ = θ1 is a smooth E-equidistributed loop, as desired. q.e.d.

5.2. The filtration in the compact case. In this section we con-

struct the filtration of Cμ∞ (T × P ) in order to prove Lemma 3.2.
To do this, we return to our homogeneous setting, assuming G is a
compact (connected) Lie group, H < G a closed subgroup, P = G/H,
and M = T × P . For the sake of simplicity of the exposition, we start
assuming H = {1G }. The general case will be gotten easily from this
particular one.
If νG denotes the Haar (probability) measure on G, there are two
unitary representations of G on L20 (G, νG ) := {φ ∈ L2 (G, νG ) : φνG =
0} given by
(Lg φ)(x) := φ(g −1 x),
(23)
(Rg φ)(x) := φ(xg), ∀g, x ∈ G, ∀φ ∈ L20 (G, νG ).
By the classical Peter-Weyl theorem, we know left action L decom-
poses in a direct sum of finite-dimensional irreducible sub-representations,
i.e. there exists a family
 (En )n≥1 of finite-dimension subspaces of
L20 (G, νG ) such that n E n is dense in L20 (G, νG ) and each En is L-
invariant, with no proper L-invariant subspace contained in En . More-
over, these spaces satisfy En ⊂ Cν∞G (G) = C ∞ (G) ∩ L20 (G, νG ), for every
n ≥ 1, and they are also R-invariant (for instance, see §3.3 in [Sep07]
for details).
In particular, this implies that, if γ : T → G an En -equidistributed
with period m, then
  
m−1
j
 m−1  
j

(24) φ γ t+ x = (Rx φ) γ t + = 0,
m m
j=0 j=0

for every t ∈ T, every x ∈ G, and any φ ∈ En .

Now, for each φ ∈ C ∞ (M ) and every k ∈ Z, we define φ̂k ∈ C ∞ (G)
by

(25) φ̂k (x) := φ(t, x)e−2πikt dt, ∀x ∈ G,

T
and
⎧ ⎫
⎨ ⎬
(26) Vn := φ ∈ Cμ∞ (M ) : φ̂0 ∈ Ej , φ̂k ≡ 0, ∀ |k| > n .
⎩ ⎭
j≤n
210 A. AVILA, B. FAYAD & A. KOCSARD

By the Peter-Weyl theorem and classical Fourier series arguments,

we have
Lemma 5.5. The family (Vn ) given by (26) is a filtration for Cμ∞ (M ).
Proof of Lemma 3.2 in the compact case. Let us consider the filtration
(Vj )j≥1 given by (26). As we did in the nilpotent case, without loss of
generality we can assume n < q0 . 

Let γ̃ ∈ C ∞ (T, G) be a E
j≤k j -equidistributed loop in G, and
let m̃ be its period. Then let us define γ : T → G by γ(t) := γ̃(q0 t)
and write q̄ := q0 m̃. Notice H0,γ ∈ SW∞ (T × G) clearly commutes with
T1/q0 . Let us show that condition (ii) of Lemma 3.2 also holds.
To do that, let p be any integer coprime with q̄ and let us consider an
arbitrary φ ∈ Vn . Once again we consider the Fourier-like development
of φ: 
φ(t, x) = φ̂ (x)e2πit , ∀(t, x) ∈ T × G,
||≤n

where each φ̂ ∈ C ∞ (G) is given by (25) and φ̂0 ∈ j≤n Ej .
Then we have
q̄
SH −1 φ(t, x)
0,γ p/q̄ H0,γ
T
q̄−1  
 j  j
= φ t + ,γ t + x
q̄ q̄
j=0
 
 j  j
q̄−1
= φ t + , γ̃ q0 t + x
q m̃
j=0
  
  j
q̄−1
j
= φ̂ γ̃ q0 t + x e2πi(t+ q̄ )
m̃
j=0 ||≤n
  
k
(27) 0 −1 m̃−1
q   2πi(t+ qj + m̃
k
)
= φ̂ γ̃ q0 t + x e 0
m̃
j=0 k=0 ||≤n
  
k
  q0 −1
k  2πi j
m̃−1
= e2πit φ̂ γ̃ q0 t + x e2πi m̃ e q0
m̃
||≤n k=0 j=0
  
 k
m̃−1
= q0 φ̂0 γ̃ q0 t + x
m̃
k=0
  
 k
m̃−1
= q0 Rx φ̂0 γ̃ q0 t + =0
m̃
k=0
for every t ∈ T and every x ∈ G, and where
 the last equality is a
consequence (24) and invariance by Rx of j≤n Ej .
 MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 211

−1
Thus, by Lemma 3.3, it follows from (27) that Vn ⊂ B(H0,γ Tp/q H0,γ ),
as desired. q.e.d.

5.3. The case H = {1G }. Now, let us suppose H < G is a proper

closed subgroup. Since G and H are both compact, they admit unique
Haar probability measures, which will be denoted by νG and νH , respec-
tively. The Haar measure on G/H will be denoted simply by ν.
We will write πH : G → G/H for the canonical projection and we can
define the linear operator ΠH : C ∞ (G) → C ∞ (G/H) by

ΠH ψ(gH) := ψ(gx) dνH (x), ∀ψ ∈ C ∞ (G).

H
Let us remark that ΠH is continuous, closed, and surjective (in fact,
the pull-back by πH is a section of ΠH ) and satisfies ΠH (Cν∞G (G)) =
Cν∞ (G/H). In particular, the family (ΠH (Ej ))j≥1 , where spaces Ej are
defined as in §5.2, turns out to be a filtration of Cν∞ (G/H), where each
ΠH (Ej ) has finite dimension.
Then we have the following
Lemma 5.6. If γ ∈ C ∞ (T, G) is an Ek -equidistributed loop (with
k ∈ N arbitrary), then πH ◦ γ is a ΠH (Ek )-equidistributed loop on G/H.
Proof. Let m denote the period of γ and let φ ∈ Ek be arbitrary.
Then we have
   m−1   
 j  j
m−1
ΠH (φ) πH ◦ γ t + = φ γ t+ y dνH (y)
m m
j=0 j=0 H
 m−1  
  j
= φ γ t+ y dνH (y)
H m
j=0
 m−1   
 j
= (Ry φ) γ t + dνH (y) = 0,
H m
j=0
where the last equality is a consequence of the R-invariance of space Ek .
q.e.d.
Now, using Lemma 5.6 we can easily extend our proof of §5.2 to the
case where H is a proper subgroup. In fact, given any φ ∈ C ∞ (T×G/H)
and any k ∈ Z, once again we can define φ̂k ∈ C ∞ (G/H) by

φ̂k (gH) := φ(t, gH)e2πikt dt, ∀gH ∈ G/H,

T
and so (re)define the filtration (Vn )n∈N of Cμ∞ (T × G/H) analogously
to (26): ⎧ ⎫
⎨ ⎬
Vn := φ ∈ Cμ∞ (T × G/H) : φ̂0 ∈ ΠH (Ej ), φ̂k ≡ 0, ∀ |k| > n ,
⎩ ⎭
j≤n

for every n ∈ N.
212 A. AVILA, B. FAYAD & A. KOCSARD

Then, invoking Lemma 5.6 and the above filtration, mutatis mutandis
we can extend the proof of Lemma 3.2 we did in §5.2 in the case H =
{1G } to the general one.

References
[AK11] A. Avila & A. Kocsard, Cohomological equations and invariant distribu-
tions for minimal circle diffeomorphisms, Duke Math. J. 158 (2011), no. 3,
501–536, MR 2805066, Zbl 1225.37052.
[CG90] L. Corwin & F. Greenleaf, Representations of nilpotent Lie groups and
their applications. Part I, Cambridge Studies in Advanced Mathematics,
vol. 18, Cambridge University Press, Cambridge, 1990, Basic theory and
examples, MR 1070979, Zbl 0704.22007.
[FF03] L. Flaminio & G. Forni, Invariant distributions and time averages for horo-
cycle flows, Duke Math. J. 119 (2003), no. 3, 465–526, MR 2003124, Zbl
1044.37017.
[FF07] , On the cohomological equation for nilflows, Journal of Modern
Dynamics 1 (2007), no. 1, 37–60, MR 2261071, Zbl 1114.37004.
[For08] G. Forni, On the Greenfield-Wallach and Katok conjectures in dimension
three, Contemporary Mathematics 469 (2008), 197–213, MR 2478471, Zbl
1156.37003.
[Kat01] A. Katok, Cocycles, cohomology and combinatorial constructions in ergodic
theory, Smooth ergodic theory and its applications (Seattle, WA, 1999),
Proc. Sympos. Pure Math., vol. 69, American Mathematical Society, Prov-
idence, RI, 2001, In collaboration with E. A. Robinson, Jr., pp. 107–173,
MR 1858535, Zbl 0994.37003.
[Mal49] A. I. Mal’cev, On a class of homogeneous spaces, Izvestiya Akad. Nauk.
SSSR. Ser. Mat. 13 (1949), 9–32, MR 0028842, Zbl 0034.01701.
[MOP77] J. Moulin-Ollagnier & D. Pinchon, Systèmes dynamiques topologiques. I.
Étude des limites de cobords, Bull. Soc. Math. France 105 (1977), no. 4,
405–414, MR 0498967, Zbl 0407.54034.
[NT12] A. Navas & M. Triestino, On the invariant distributions of C 2 circle dif-
feomorphisms of irrational rotation number, to appear in Mathematische
Zeitschrift, 2012, MR 3054331, Zbl 1275.37013.
[RH12] Federico Rodrı́guez-Hertz, personal communication, 2012.
[Sep07] Mark R. Sepanski, Compact Lie groups, Graduate Texts in Mathematics,
vol. 235, Springer, New York, 2007, MR 2279709, Zbl 1246.22001.

Institut de Mathématiques de Jussieu

175 rue du Chevaleret
75013. Paris, France
E-mail address: artur@math.jussieu.fr

Institut de Mathématiques de Jussieu

175 rue du Chevaleret
75013. Paris, France
E-mail address: bassam@math.jussieu.fr
MANIFOLDS SUPPORTING DUE DIFFEOMORPHISMS 213

Instituto de Matemática e Estatı́stica

Univesidade Federal Fluminense
Rua Mário Santos Braga S/N
24.020-140. Niterói, Brazil
E-mail address: akocsard@id.uff.br