lnventiones math.

15, 1-23 (1972)

9 by Springer-Verlag 1972

Ergodicity of Anosov Actions

CHARLES PUGI4* (Berkeley) a n d MICHAEL SHUB** (Waltham)

1. Introduction
In this paper we generalize some ergodicity results of Anosov and
Sinai [1, 2] to group actions more general than Z and R. At the same
time we provide what we consider to be a more natural proof of the
central theorem in [1] concerning the absolute continuity of certain
foliations- see (2.1).
Definition [5]. Let G be a Lie group acting differentiably on M,
A: G--,Diff(M) where M is a compact smooth manifold. We assume
that the orbits of G define a differentiable foliation o~, which is the case
for instance if the G action is locally free (every isotropy group is discrete),
The action is called Anosov if there exists an Anosov e l e m e n t - a n
element g~G such that A ( g ) = f is hyperbolic at ~ [5] and
(1) the G action is locally free, or
(2) G is connected and g is central in G.
We recall that A(g) = f is hyperbolic at ~ means that T f : T M ~ T M
leaves invariant a splitting
E"@ TJ~ G E s= T M
contracting E S more sharply than T~, expanding E" more sharply
than T~. ( T ~ is the bundle of planes tangent to the leaves of ~ )
For example, if {q~t} is an Anosov flow on M then t i-~ q~t defines an
R-action on M and gives the foliation of M by the trajectories. Any
~ot, t + 0 is an Anosov element. Similarly, i f f is an Anosov diffeomorphism
of M then n w-~f" defines a Z-action on M which is Anosov. The leaves
of the orbit foliation are the points of M. Further examples are given
in [3, 5].
In [5] it was proven that Anosov actions are structurally stable,
generalizing another part of the work of Anosov on flows and diffeo-
morphisms.
Definition. The action A: G---, Diff(M) is ergodic iff it is measure
preserving and all invariant functions are constant. Precisely, we require
* University of California at Berkeley, supported by NSF GP-14519 and the Sloan
Foundation.
** Brandeis University, partially supported by NSF GP-9606 and GP-23117.
I lnventionesmath.,Vol.15
2 C. Pugh and M. Shub:

(1) For each g~G, A(g) is measure preserving (respecting some

fixed Lebesgue measure on M).
(2) If f : M - - , R is integrable and, for all g~G, f o A ( g ) = f almost
everywhere on M then f equals a constant, almost everywhere.
Our main theorem is:
( l . l ) T h e o r e m . Suppose A: G--~Diff2(M) is a measure preserving
Anosov action with an Anosov element in the centralizer of G. Then A
is ergodic.
In particular, if G is abelian and A a measure preserving C 2 Anosov
action then A is ergodic.
Theorem (1.1) may be used in conjunction with [6] to give informa-
tion about the ergodic elements of an Anosov action. We give one
example:
(1.2) Theorem. Suppose A: Rk--~Diff2(M) is a measure preserving
Anosov action. Then for every g6 R k off a countable family of hyperplanes
in R k, A(g) is an ergodic diffeomorphism. We recall that a hyperplane is a
translate of a hyperplane through zero.
The idea of the proof is as follows. Let f be the Anosov element.
Then f is hyperbolic at the orbit foliation and so, from [5], we deduce
a stable manifold theory for f By uniqueness and commutativity with f,
the stable and unstable manifolds are A-invariant. We prove that any
strong stable manifold foliation is absolutely continuous, and so is the
center unstable foliation. Then we deduce ergodicity of A as Anosov
and Sinai did, via Birkhoff's Theorem [2]. The center unstable case is
harder than the strong stable, and it would be tempting to try avoiding
it by using [8]. This would require measurability of the center unstable
foliation in the sense of Sinai [8]. But measurability seems no easier to
prove than absolute continuity, nor is it a consequence of being a
foliation in the sense of Anosov [1, p. 18]. See w for an example of this.

2. Pre-Foliations
It is frequently useful and natural to deal with a localized version
of a f o l i a t i o n - w e call it a pre-foliation. It amounts to the continuous
assignment of a disc through each point of a manifold.
Indeed, let M be a compact smooth Riemann manifold and let D k
be the k-disc. The set of all C r, r > 0 , embeddings Dk-+M carrying 0
onto some p ~ M forms a metric space
Emb'(D k, 0; M, p).
The C r distance between two embeddings is defined in the usual w a y -
either via the Riemann metric or a fixed embedding of M into a Euclidean
 Ergodicity of Anosov Actions 3

space. It is easy to see that Emb~(D k, M) is a C ~ fiber bundle over M,

~(h)= h(0) being the projection.
Definition. A pre-foliation of M by C r k-discs is a map p ~-, @ such
that @ is a C r k-disc in M containing p and depending continuously
on p in the following sense: M can be covered by charts, U, in which
p w* ~p is given by
~p=e(p)(D k) pc U
and a: U ~ Embr(D k, U) is a continuous section. If, in addition, these
sections a can all be chosen so that the maps (p, x)w+a(p)(x) are of
class C *, 1 =<s =<r, then the pre-foliation is said to be of class C ~.
Example 1. If ,~- is a C ~ k-foliation of M, r > l, let
~(6)= {xe~: d~(x, p)<=6}
where d~ is the distance in the leaf measured respecting the Riemann
structure in T ~ inherited from TM. Then, for small 6 > 0,
p~(~)
gives a C ~ pre-foliation of M by C ~ k-discs.
Example 2. Let N be a C ~ sub-bundle of k-planes in TM. Then, for
small ,5 > 0,
p ~ expt,(Np (6))

gives a C ~ pre-foliation of M by C ~ k-discs.

Example 3. Let ~/g'" be the unstable manifold foliation of M for a Cr
Anosov diffeomorphisms. For small 6 > 0

p ~ W~(6)=the 6-1ocal unstable manifold through p

gives a pre-foliation of M by C' k-discs. In general this pre-foliation is
not of class C 1 [1, w
On the same note, let us emphasize that for us, a "foliation of M by
C r k-leaves" need not be a C r foliation. The leaves are C r and they vary
locally continuously in the C r sense (this, for r = l , implies that the
union of their tangent planes gives a continuous k-sub-bundle of TM)
but their assembly is not necessarily CL Similarly for pre-foliations.
N o w we shall explain the idea of Poincar6 map along a pre-foliation.
This is the usual "notion of translation in the transversal" for foliations.
Let ff be a pre-foliation of M by C ~ k-discs, r > 1, let q ~ I n t ff~, (Or=the
if-disc through p, and let Dp, Dq be two smooth (m-k)-discs embedded
transverse to ~, at p, q. (See Fig. 1.)

TpDp@ Tp %= TpM, TqDqOTq~v= TqM.

1"
4 C. Pugh and M. Shub:

Fig. 1. The Poincar6 map

Then there is defined a surjection Hp, q: De, q---,Re, q where De, q is a

neighborhood of p in D e
De, q ) Re, q

De Dq
He, q(P)=q ttp, q(Y)e~vC~Dq.
Since ~, depends continuously on y e Dp in the C r sense, r > 1, and ~p
transversally intersects Dq at q, there is uniquely defined a new point of
transversal intersection, Hp, q(y), depending continuously on y near p.
The range of Hp, q, Rp, q, is not in general a neighborhood of q in Dq,
nor is He. q in general a local homeomorphism. On the other hand,
He, q is C s when f~ is of class C "~ and lip, q depends continuously on
p, q, D e, Dq in the C ~ sense. Thus, if f~ is C 1 and q is near p then Hp.q is
a local diffeomorphism.
Next we explain the idea of absolutely continuous foliations. Recall
that a bijection between measure spaces h: U-~ V is absolutely continuous
if it is measurable and is a bijection between the zero sets of U and V.
Definition. A pre-foliation of M by C r k-discs is absolutely continuous
if each of its Poincar6 maps He, q: De,~--,Re, q is absolutely continuous.
Definition. If, in addition, the Radon Nikodym derivative, J, is
continuous and positive, J: De, q--~R,
]ADq(S)= ~ Jdpop S~Rp, q
H~;,~ (S)
then the pre-foliation is said to be measurewise C 1.
The measures Poq, PD~ are the smooth ones induced by the Riemann
structure on TM. Joint continuity in p, q, De, Dq, y is required. Variation
of D e,/)q is done in Emb 1(D"-R, M). J is called the (generalized) Jacobian
of H. Existence of such a J implies, of course, absolute continuity.
(2.1)Theorem. Strong unstable and strong stable foliations are
measurewise C ~. (In particular absolutely continuous.) Precisely: Suppose
 Ergodicity of Anosov Actions 5

f is a C ~ diffeomorphism of M, s>=2, T f leaves E " O E p~= T M invariant

and
sup [ITpP~fII~/< i n f m ( T ~ f ) O<j<r<s, r>l.
p~M p~M

Then there is a unique f-invariant foliation of M by C ~ leaves tangent

to E ~, the strong unstable foliation, ~r ~. It is measurewise C t. Similarly
for strong stable Jbliations.
Remarks. m ( Tp f ) is the co-norm (or minimum norm) of Tp f l E e = T~f ;
that is, m ( T p f ) = l l T ~ p f - l l l -x. Our condition on T f means that all
vectors of E" are expanded more sharply than any vectors in E p~. The
existence of a unique f-invariant foliation of M with C ~ leaves tangent
to E" is proved in [5]. In general, there is no reason to believe E ps can
also be integrated. Notice that HTpSfJI may be > 1 which is why we
write p s - t o indicate pseudo-stable. A more or less explicit formula
for the Jacobian J is developed in the proof of (2.1) given in w3. The
inequality in the hypothesis of (2.1) can be weakened to

i n f m ( T ~ f ) {1TpP~J'l[-J> 1 O<j<r
p~M

but the proof of (2.1) becomes technically harder. If sup [[TW*fH<I,

P
notice that the hypothesis of (2.1) amounts to assuming T " f is an
expansion.
Finally, we wish to point out that our proofs differ substantially
from Anosov's [1] only in that they avoid using continuous differential
forms, dealing directly with the Poincar6 maps instead. In the same
way, they differ from those in [8] in that no emphasis is laid on measure
theoretic generality.

3. Proof that ~t/" Is Measurewise C ~

Although E u, E "s need not be smooth (this would imply measure-
wise C ~ at once) they are H61der.
(3.1) Lemma [c.f. 1]. E u and E ps are O-H61der continuous for some
0>0.
Proof Let E u, E ~'s be smooth approximations to E u, E re and let
~ = {P6L(/~Ps,/~]): tlPll__<l}. Then ~ = ~ ) ~ : , is a smooth disc bundle
over M and T f - ~ acts on ~ in the natural way

F: P - * ( C x + K~P)o(A~+ BxP) -1
for
T~ f _I = ( Ax B~) respecting/~w (~/~,,.
Cx K~
6 C. P u g h a n d M. S h u b :

F is a fiber contraction: it preserves fibers of 9 , covers f - l . M - * M ,

and the Lipschitz constant of F I ~ is < k < 1. In fact k is approximately
/(2 when 2 = inf m (T~ f), p = sup [IT f S f [[, and/~",/~P~ are very near E u, E p~.
The bundle E p~, represented as the graphs of linear maps EPS-~E ",
is an F-invariant section of 9 . But the Invariant Section Theorem
[6.1 of 4] says that the unique F-invariant section of ~ is 0-H61der
continuous if F is C ~ and k L ( f ) ~ 1. Since f is at least C 2, this proves
that, for some 0 > 0, E ps is 0-H61der. Similarly for E ".
Following Anosov we write ~ to denote uniform convergence.
(3.2) Lemma [ 1, p. 136]. Suppose h: D k--, R k is a topological embedding
and (g,) is a sequence of C 1 embeddings D k~-, R k such that

g._~ h J (g,)_~ J

where J(g,) is the Jacobian of g,. Then h is absolutely continuous and has
Jacobian J.
Proof [1, p. 136]. We must show

rues (h A) = ~ J dp A c D k, measurable
A

when dp is Lebesgue measure on D k. Since h is continuous, it suffices

to prove this equality for A = a n arbitrary closed subdisc of D ~. Let
e > 0 be given and choose two other discs A', A" such that A' is interior
to A and A is interior to A". They can be chosen so near A that

J d lt < e,/2
A"-A

because J is continuous. Since g, is a C 1 embedding, mes (g, S) = ~ J(g,) dl t

S
for any measurable S c D k, and since h is a topological embedding

g.A' ~ h A ~ g , A "
for large n. Thus
J(g,) dp < ~ J(g,) dp < ~ J(g,) dp
A' A A"

mes(g,A')< mes(hA) < m e s ( g . A " )

and so l i n e s ( h A ) - ~ J ( g . ) d p l < e for large n. Since ~ J ( g , ) d p ~ ~Sdt~,

A A A
we have show n Imes (h A ) - S J dPl~ e proving the lemma.
A
 Ergodicity of Anosov Actions 7

To state precisely the next lemma, we speak of angles between sub-

spaces of TM. The Riemann structure on T M lets us define
~ (Ap, Bp)= max { ~z (a, Bp): ae A p - O } u { ~ (b, Ap) : be B p - O }
where A v, Bp are linear subspaces of Te M. This amounts to the Haus-
dorff metric on the Grassmanian. The angle between two subbundles
A, B is the supremum of ,~ (Ae, Bp).
(3.3) Lemma. Suppose T M = N @ E P S = E U ~ E ps and N is smooth.
Let fr be the smooth pre-Joliation p~-~fqe(8)=expv(Np(8)). Let fl be
given, O < fl < ~z/2. For small 6 > 0 , each P oincarO map Gp, q: D p, q --r R e, o
along f#(8) is a smooth immersion 0~ E(TDe,(EU)• and ~(TDq,(E")•
Proof The condition on D e, D u is that they be uniformly transverse
to E". Since Gp, q is smooth and its derivative is a continuous function of
p, q, it suffices to prove that ~.Ge, q is a bijection T y D p ~ , , Dq for
y'=Gp, q(y). Since Gp, q=Gy, y, near y, it suffices to verify bijectivity at
y = p . Clearly when y = p = q , this is true. But since the derivative of Gp, q
depends continuously on p, q, D e, Dq and since M and { A p c T e M :
(A v, (EU)• [3} are compact, bijectivity on the diagonal p = q propa-
gates to some h-neighborhood of the diagonal.
Proof of (2.1). Let N be a smooth approximation to E u. Choose fl
so that 0 < f l < n / 2 and ~(EW,(E")• ~z(E w, Nl)<fl. Then choose
h > 0 according to (3.3). Let
~: N. = expy(N,.(6)) yeM
be the resulting smooth pre-foliation. Let f#" be the pre-foliation gotten
from iteration by f " f~"" N" = f " N--y"

Let f#"(c) be the restriction of f~" to radius e.

fr %."0:) = {xe ~,": d~~ y)<~}.
By [5], N"(c)_~ 1r TN"(t:)~E". Thus f acts on pre-foliations
in a natural way and ~/r is the attractive fixed point of this action.
Consider qe Wp" and discs D e, Dq transverse to E". We must study
the Poincar6 map Hp, q: De, q ~ R v , q for the foliation ~#r.. Because ~/" is
a f o l i a t i o n - n o t just a pre-foliation-Hp, o is a homeomorphism and
Re, q is a neighborhood of q in Do.
The relation between lip, o and H y - . p , y - . q is expressed by commu-
tativity of
f-nDp, q n • 1 7 6 q

Is.
Dp, q Hp, q +Rp,q
8 C. Pugh and M. Shub:

since ~//'" is f-invariant. Since f is a diffeomorphism existence of a

continuous positive Jacobian for Hp, q is equivalent to the question for
HI-,p,j--, q. Furthermore, as n--,oo, T ( f -" Op) and T(J .... Dq)~ E p~
[5]. Thus it is no loss of generality to assume

qeW~,(e,/2) r177 r177 (.)

for all n>0. Furthermore, we may shrink Dp SO that D p = D p , q and

Rp.q=range Hp, q is interior to Dq, for existence of J(Hp,,) is a local
question.
Since ~"(c)~/r the Poincar6 map Gp,q of Op to Dq along ~"(e.)
is defined in a unique single valued continuous manner on the domain
Dp, n = 0, 1, 2 . . . . . Thus it is clear that

g,~h

where g, = G"p.o., lOp, Qn =-f~p (g,)(5 Dq, and h = He, q. We show that

g. is an embedding, (a)

J(g,) ~ J = unif lim det (f-"[Ty Dp) (b)

.~ det(j . - n ]Thy Dq)

Then, by (3.2), J is the Jacobian of h=Hp, q. Since the limit in (b) is

uniform, J is continuous, and by symmetry positive. Thus, proof of
(a), (b) demonstrates (2.1).
The proof of (a) is topological and thanks are due to R. PalMs. By
(3.3), (*), the choice of 6, and the naturality of Poincar6 maps, g, is at
least immersion wherever defined. Moreover, both g, and h are defined
on a slightly larger disc/)p, say

gn : L) p ---~D q , fl : L) p --+ D q

and ~ , ~ h . Since ~,, h are locally injective, the theory of mapping

degrees [7] isapplicable. Let Y be a compact neighborhood of Rp, q = h Dp
interior to hDp. For any yeY, degree (h, D~^,y)ff 1 since h is a homeo-
morphism. For large n, ~, ]•b e is very near h l(~Dp and so

~, I0/3p--- h I0/3~, in D q - Y .

Thus, for large n, degree (~,, Dp, y)= 1 for all y e Y, and thus ~:, embeds
g2 ~ Y. The latter contains Dp, for large n, since ~,, ~ h and h- 1 y contains
Dp in its interior. This proves (a).
 Ergodicity of Anosov Actions 9

To prove (b) we express g, in terms of the Poincar6 map along f~,

acted on by f " - t h i s is the straightforward thing to do. Consider
gn : Dp ~ Dq as
g . = j~ ' n o t.rp.,
I-'0
q. o f n

where p , = f - " p , q , = f - " Q,. (Recall that Q. was the point f~p"(e)~Dq.)
Thus q,e ~, and so the Poincar6 map along .a2, Gp.,q.,
. 0 is well defined
on f -" Dp. Moreover
q,e %.(%), %--+0

as n - ~ . For for. is approximately tangent to E" and is thus sharply

expanded by .['" (see Fig. 2).

D~
fn "'Y"~ D~I~..~
.~,-,--.~ i~ ~" _ .y

Fig. 2. The effectoff ~

Using the Chain Rule,

J,~(g,) = det (7"./'"IT r ,~,~,( f -" Dq)) det ( T G ~ ITy ,,, ( f -" D p))
. d e t ( T f -"l~.Dp)

for any yeDp. Since T(f-"DI,)~EV~', T ( f - " D , ) ~ E I'~, and q,~,,,(%)

with %--,0, the middle factor tends uniformly to 1. (b) is therefore
equivalent to
uniflim det(Tf-"lTyDp) - u n i f l i m det(Tf-"l~'DP) (b')
.... det(Tf-"lT~,,,Dq) ,~ det(Tf-"lThrDq)"

Although (b') could be proved directly, we first establish the special

case (as does Anosov in [1]) y = p , TpDp=E~ ~, TqDq=EPL We prove

lim d e t ( T ~ f - " ) exists uniformly. (c)

,~ ~ det (T~Wf-")

TPsf -" denotes T[" "IE p~. By the Chain Rule (c) is equivalent to the
uniform convergence of
f l det ( T f ~ r . / ' - 1)
k=O det(TF~q J--~i
10 C. P u g h a n d M. S h u b :

and this, in turn, is equivalent to the uniform convergence of

]det(Tf~r f -l)_det(Tje~% f - l ) ] .
k=0

Since E pS is 0-H61der with 0 > 0 by (3.1), and f i s C z, TpSf -~ is 0-H61der

and so
idet (Tf_~ p f - 1 ) _ det (Tf-~q f-1)1 <=C d ( f -k p, f -k q)O

for some constant C. Since q~WUp, d ,~j , ' - k p , j~ - k q)=

,<j~ k d(p,q) where

2=infm(T~f)>l. Thus 2 - ~ and our series converges uniformly

by comparison with C ~ ( 2 - ~ k d(p, q). This proves (c).
Now we show how (c) implies (b'). Let ~ps be the projection of TM
onto E ps along E u. Thus rcpS kills E ~ and leaves E ps fixed. Since T f leaves
EU@Eps invariant, T f - " commutes with ~PL Thus

U ' - " I L Dp = (~vs I T~-.y i f - " Dv))-' o T"S f -" o (~ps IZ, D,)
for y6 Dp. Taking determinants gives

det ( T f -"1 "F,,.Dp) = det ( TvVSJ'-") det (rr,PSl~. Dr,)

det (rcP"I Tj- ,> . l - " Dr,)
As n ~ , T ( f - " D p ) ~ E p~ and so the denominator in the preceding
fraction tends uniformly to 1. The same holds when y is replaced by a
point of Dq. Thus, we are reduced to proving

uniflim d e t ( ~ f ~ f - " ) det(nP~] TyDp)

.~ det(T~P~,f -")det(nPSlT~.~,Dq)
(b")
= unif lim det (~P~f-") det (nPsl T.,,Dp)
.~ ~ det (ThUf -") det (rrP~ ] Thy Dq)"

Since g. ~ h and D o is C ~, (b") is equivalent to

uniflim d e t ( ~ " s f - " ) = u n i f l i m det(~r,~'f ") (b'")

.~ det (Tp~, f - " ) .~o~ det (ThP,~f-") '

By (c) - applied to y, hy instead of p, q - t h e second limit exists and is

uniform. To prove that the first exists and equals the second it suffices
to show that
unif lim det(Th~'~f-")
9 - 1. (d)
.~o det(T~V~,f -")
(d) is equivalent to
n--I
uniflim y' Idet(Tfe~h,,f-~)-det(Tfe~,.,,f-')l=O (d')
n~m k=0
 Ergodicity of Anosov Actions 11

by the Chain Rule, as before. Again, this sum is <

n--1
C ~ d ( f - k h y , f - k g , y)~
k=O

for some constant C, since E pS is 0-H61der. Let /~=sup IbTJ'~['ll and

2 = i n f m ( T " f ) . By hypothesis, p < 2 and 2 > 1. Choose
max(if, 1 ) < / ~ < ~ . < 2 .
Since J ' - " hy~W)-%,(~,,),
" J - " (g,y)6ffj ,,,.(e,,) and ff is approximately
tangent to E",
c, < 2 " for large n.
Thus, d ( J - " ( h y ) , f - " ( g , y ) ) < e , < i t " for large n. On the other hand,
d ( f - k ( h y ) , f - k ( g , y ) ) = d ( f " k ( f - " h y ) , f " - k ( f - " g , y ) ) , and for large k,
f - k Dq ..... f - " Dq are nearly tangent to E p~, so that
d ( f - k ( h y ) , f k(g,y))<C'l~"-k ). "
for some constant C'. Thus
n--1 n--1

Zo z~
= C"(t~~+... + t~"~ 2 ,o = C" i~~ \~L-_IT] 2 -"~

which tends to zero as n - ~ ~ . This proves (d'), hence (d), (b"), (b'), (b'),
and ( b ) - completing the p r o o f of (2.1).

4. Measurewise Smoothness of Center Unstable Foliations

The main t h e o r e m of this section, (4.2), is an analogue of (2.1).
Recall that a diffeomorphism f of M is normally hyperbolic at a folia-
tion o~- of M iff 7~]' leaves invariant a splitting T M = E U | 1 7 4 ~,
expanding E" m o r e sharply than U = T ~ , contracting E ~ m o r e sharply
than U, and leaving o~--invariant. The following theorem was proved
in [5].
(4.1) Theorem. / f ,~- is C 1 and f is normally hyperbolic at ,~ then
there are unique f-invariant Jbliations of M, ~lUc" and ~tU~, tangent to
U" = E" 9 U and U ~= U 0 EL Each of their leaves is a union of ,~-Ieaves
and W;"= U w,q",w; ~= U Wq.
qE,~p qE,~p
Here we shall prove
(4.2) Theorem. If f is normally hyperbolic' at ,~,, o~ is C 1, and f is C 2
then ~IU~", ~.cs are measurewise C 1
12 C. Pugh and M. Shub:

Proof We shall utilize a notion generalizing "pre-foliation by discs"

to "pre-foliation by submanifolds'. However, we shall not make the
precise general definition of this, but confine ourselves to the case

U
yE,~p
where N is a smooth subbundle of TM approximating E". In w3, we called
%: %.-- expy(Nv(6))
the pre-foliation by u-discs. Now we are considering the union of all
these u-discs as y ranges over the leaf ~ , . This gives the immersed
manifold ~ , , nearly tangent to E cu. Then let

U
We know that ~ n ~_ ~ cu and T ~ " ~ ECU by [5].
Let D r, Dq be s-discs transversal to E "u through p, q with qE Wf~u.
We must investigate the Poincar6 map Hp, q along ~/g"u. As in w we
may assume
q~ W~(r,/2), p' ~o~p(e/2), D p = d o m a i n Hp,~, diam(Dp)<e]2
without loss of generality. Consider the Poincar6 maps H, = H~,~ along
the Jg" leaves through Dp. As in w3, we must prove that
H, is an embedding, H, ~ H = Hp,q, (A)
J(H,)~J>O. (B)
The proof of (A) is the same as (a) in w because ,Zg" ~ / / ' " " and Hp,, is
a homeomorphism.
Call D = ~ ~,~(c). This D is a smooth disc transverse to E". It is
y~Dp
smoothly fibered by the leaves of ~. For each y~Dp, ~ m ~ , " for all
n > 0 . Thus, the Poincar6 map along the leaves of ~r y~D., would
be smooth if the image disc, Dq lay in D.
For each y~Dp, let y. be the unique point of~ such that

and let y , be the unique point of.~(c) such that

HyeW~(c).
Clearly y, ~
--* y , and p, is the point we called p'.
Choose smooth discs S(y~), X(y,) at y,, y, in D, transverse to E".
We may assume them chosen so that
S,(y,) ~ 2(y.), TX(y,) ~ TS(y.).
 Ergodicity of Anosov Actions 13

Then we may factor H. as h.o ~,,y. where F,,.y~ Dp-~X(y.) is the Poincar6
map along .~- in D and h.: 2(y.)--~Dq is the Poincar6 map along the
leaves of.;r through Dp (see Fig. 3). Note that this factorization depends
on y.

Fig. 3. Factorizing the Poincar6 map/4.

Since 2(y.)~ 20'.) and T L ' ( y . ) ~ T2(y.) and .~- is C ~,

det (~, ~,~,) ~ det (Tv ~. y.)> 0.

Thus (B) will follows from

uniflim J,,.(h.)= f i det(Tf-llTf kY*f-ky:(Y*)) (B')

.~o~ " ~=0 det(Tf-llTf-knyf-kDq)
when H = Hp, q.
As in w3, let/~ = sup t]rcsf ]l, 2 = inf m(r"f), and choose max(l,/~) <
p < 2 < 2 . Then, as in w
f -k H.y~].'Z~,, ff.i.-k)
J'-k H ye WjU_kny(eJ.-k)

for O<k<n and large n, because fr is nearly tangent to E u and is

thus expanded by 2k under fk. We also claim that

d(f -k y,, f -k y.) <=2 -k e

(,)
d(f-kH, y,f-~Hy)<2-kr,

for O<_k<_n and n large. The proof is by induction on k. Since y . , Hy,

/4. y, y. form a twisted trapezoid of small ( < e) diameter whose nearly
14 C. Pugh and M. Shub:

I0 q

Fig. 4. Our twisted trapezoid

parallel opposite edges in Wy'~,fr have length < c, the other edges - being
in Y and Dq must also have length < e (see Fig. 4). This proves (*) for k = 0.
Suppose (*) is valid for k - 1 <n. Let 7 = s u p [ITCf-l[I. Then
d '-k '--k d -k+l -k+l /] - k + l
(j Y., J Y.) =<)' ( f Y., f Y.) < 7 e
by the induction assumption. Thus, f - k y , , f - k H y ' f kH.y ' f - k y .
forms a twisted trapezoid of small (< i' e) diameter whose nearly parallel
opposite edges in fCfrf,.., ff~-k have length -<e2 -k. Its other edges,
being in ~ and f-KDq, must have length <r]2-k; for o~, f-kDq and
ff.-k are essentially perpendicular to each other. This proves (.) for k.
(See Fig. 5.) Note that we used k<n to assure f4"-k is defined and more
or less tangent to E".

g
Fig. 5. General twisted trapezoid

l_l_g h_Lspan(l,g) g.(g,w)-~O

~ l / g and h/g~O.

Now we shall prove (B'). By the Chain Rule

J~, (h.) - det (Ts - ,,.v.Hj~ y., s--n. ~,)det (TJ' "[ ~, X (y.))
det (TJ'-"[ Tn.y Dq)

where H ~ . . . . . . :f-"Z(y.)-~.f-"O. is the Poincar6 map along the

leaves of ~,~6 t'h"rough f-"Dp. Since d(f-"y,,,f-"H,,y)~O and
TZ(y,)__~ TZ(y,), the first term of the numerator tends uniformly to 1.
 Ergodicity of Anosov Actions 15

Thus (B') is equivalent to

uniflim d e t ( T f - " l ~,X(y,)) - unif lim det ( T f -"1 ~,. X (y,)) 9 (B")
.~ + det (TJ '--n I Tu.~, Dq) .++ det(Tf -"lTm, Dq)

As in w3, we can easily demonstrate

I ] det(Tf ~y.f-1) (c)

k=O det(Tj~" k H ) ' f -1)

converges uniformly. For TSf ~ is 0-H61der, 0 >0, and

d(f-ky,,f-kHy)<i, -k.

From (C), it follows that the right hand side of (B") exists. E S is an
exponential attractor, under 7"./"-1, for any plane in TM complementary
to E"". In fact
.~ ( : r f -k z (y,), E ~) __<(t~/'~)k
.f. (r f -k X (y,), E ~)<=(~/ )~)k (**)
(7"["-k Dq, E ~)<=(1~/,~)k

for k<=n and k large, since T X ( y , ) ~ TX(y,) and TX(y,) is complemen-

tary to E c". Since d e t ( T f liP ) is a smooth function of the plane P

Idet (Tf-' IT+.-~,.,f - ~ X(y.))-det(TjL.,,. f-')1 < C(lu/2) k

***)
Idet (7"./ -~ ITf ~H,, f k Do) _ det (T]-~ u,, f - ~)1< C (t~/~.)k

for some constant C. By the Chain Rule, the r.h.s, of (B") converges
uniformly iff
~1 det( Tj-~[ Tf =k),,f -k X(y,))
k=oll det(Tf -llTf knyf -kDq)
does. Convergence of this infinite product follows from comparison
with (C) via (***). Similarly, convergence of the 1.h. s. of (B") to the same
limit is assured if
rt--I
0 = u n i f l i m • Idet(Tf -1 I~- ~,.~f-k~(y,))
"~ k=O (D: y,)
- d e t (T/-11Z+_~, f -k Z(y,))l
n-I
0 = unif lim Z Idet (Tf-11Tf-~u,,~ f -k Dq)
,~o~ k=0 (D: H,y)
- det (TJ'-ll Tf-~uy f - k Dq)l"
16 C. Pugh and M. Shub:

Express the k-th term in (D: y.) as

Idet (Tf-~l Ts ~,,. f -k S ( y . ) ) _ d e t ( T f -tjTj. ~,,.f -k S(y,))[
<=]det(Tf-llTf_ky. f - k S ( y . ) ) - d e t (T~ kr. f-1)1
+ Idet (T~- ~y. f - 1 ) - det (Tj~ y, f-1)1
+ idet (T~_ ~y*f - l ) _ d e t ( T f - l l T y _ ~ y *f - k _r (y,))l
=I+II+III.
By H61der continuity of TSf -~,
II < C' d ( f -k y,, f -k y,)O = C' d ( f ' - k f -" y,, f " - k f -" y,)O
C' lll(n-k)O d ( f -"y,, f -.y.)O < C,[llln-ka~-ng,]O
for some constant C'. Thus, the sum in (D: y.), is
n--1 K n-1
__<y + Z (I+II+IH)
k=0 k=0 k=K+l
K ~, n-1
Z -[-26 (z-ll.l)k-}-fta~-nO21~(n-k)O
k=O k=K+I k=0

for any K, O < K < n - 1 . We used (***) to estimate I, III. This gives a
n--I
bound for the lim sup ~ in (D: y,), which can be made arbitrarily small
n~oO k=0
by taking K large, fixing K, and then letting n tend to ~ . Thus (D:y,)
is proved. The proof of (D: H, y) is the same. This completes the proof
of (D), (B"), (B'), (B) and hence of (4.2).

5. Ergodicity
We now proceed to prove (1.1)-ergodicity of an Anosov action
,4: G-~ Diff 2(M) with Anosov element f in the centralizer of the Lie
group G.
The foliation ~ of M by the components of the A-orbits is C z. (In
fact, we only need ~ C 1; it is f which must be C2.) We shall adopt the
usual, confusing notation that g~G is also considered as the diffeo-
morphism A(g). This is all right if A is the only action considered.
Let
7 = s u p l[TSf[I q=infm(TCf) p = s u p IITCf II 2=infm(TUf)
and choose
? < y < tl<min(1, r/) max(l, p ) < p < 2 < 2 .
Since f is normally hyperbolic at ,~,, we get the f-invariant foliations
~/r ~/f S. They are also G-invariant because of their exponential charac-
 Ergodicity of Anosov Actions 17

terization [5]
W~= { x e M : d ( f -"x, f -" p) 2"-~O as n - * Qo}
W1;~={xeM:d (J ""x
. , J " p)~ -"-~ 0 a s n - * o o }
For gcG commutes with f and so
d ( f -" g x, f -"g p) 2"=d(g f -"x, g f -"p) 2"< L ( g ) d ( f -"x, f -" p) 2"-,O
iffxc Wp".(As usual, L(g) is the Lipschitz constant ofg.) Thus, g Wp = W~p.
Similarly, g Wb~= W~p.
Since the ]:invariant foliations ~r ~t/~c~are defined by
w;s=Uw;
qE,~p q~ff'p
it is clear that g w .pc " = W~p,
~" g w ~ . .-. . .W~p.
.
By (2.1), (4.2) the foliations ~/r #~,s, ~U"", ~r are absolutely con-
tinuous, in fact measurewise C ~. This will let us use the following
Fubini-type lemmas,
(5.1) Lemma. Let ,~ be an absolutely continuous Joliation of M. A set
Z ~ M has measure zero iff almost all leaves of ~ meet Z inessentially.
If the essential maximum of a function cb: M---~R on almost every ,~-leaf
is <=c then the essential maximum oJ q) is <=c.
(5.2) Lemma. / f ~1, ,~'-2 a r e absolutely continuous, complementary
Joliations of M and q): M--*R is a function that is essentially constant
on almost every leaf of ~1 and ,~2 then qb is essentially constant.
Remarks. By "almost all ~ - l e a v e s " we mean all ~ leaves not lying
in a set composed of whole ~-leaves and having measure zero. An
intersection is essential if it has positive or infinite measure, inessential
if it has zero leaf-measure. The essential maximum of a function
eb: M ~ R is inf{sup q~I(M-Z): rues Z = 0 } , and the essential minimum
is sup{infq~[(m-Z): rues Z = 0 } . Since a countable number of zero sets
forms a zero set, inf{ } and sup { } can be replaced by rain { } and max { }.
Proof of(5.1). For completeness, we reproduce part o f [ l , pp. 156-157].
It is obviously no loss of generality to restrict our attention to a neigh-
borhood U of pc M, where the components of the leaves of ~ are discs,
,~v, and where there is a smooth foliation ff by discs complementary
to ~. Thus, there is a local product structure
~z" D k • D " - k - ~ U
sending horizontal discs to ~-leaves, vertical discs to ~-leaves, and
being smooth on Dk• 0, 0 x Dm-k. The measure on the ,~'-leaves and
fq-leaves is the Riemann measure induced by the Riemann structure
on TM. The measures on D k, D"-k are the pull-backs via
Dk~--~Dk• ~+o~ U D " - k . - - ~ O •
2 InventJones math.,Vol. 15
18 C. Pugh and M. Shub:

and the measure o n D k • D m-k is the product measure. Thus,

is absolutely continuous, in fact measure preserving.

Let Z be the set of ~V-leaves intersecting Z essentially. We must
show rues Z = 0 iff mes ,~ = O.
mes Z - O
l [smoothness of ~q]
mes(ffxnZ)=0 fora.e, x~vv v
l [x x D'-k , ~x is absolutely continuous
rc

because ~ is absolutely continuous]

mes(x X om-knT~-lZ)=O for a.e. x ~ D k
l [Fubini Theorem for a product]
rues (~- 1 Z ) = 0

l [Same]
mes(Dkxyc~rc-lZ)=O fora.e, y e O " - k
I [Dk x y " , o~yv is absolutely continuous,
in fact smooth, because fq is smooth]
m e s ( ~ V n Z ) = 0 for a.e. y e ~ ,
l [absolute continuity of ~ ]
mes(,~vnZ)=0 for a.e. Y ~ x ( V x ~ v v)
I [mes ( 4 v n Z) = 0 ~ mes (~v n Z) = 0]
mes(~n2)=0 for a.e. Yefqx (Vxe,~ t')
l [obvious]
mes(2 n fix)=0 forall x ~ v
~ [ 2 , is composed of whole ~V-leaves]
[~ is smooth]
rues(Z)=0-- >mes(Znfqx)=0 fora.e, x ~ v
[fr is smooth]
Thus, mes Z = 0 iff mes Z = 0, proving the first half of (5.1).
 Ergodicity of Anosov Actions 19

Now suppose q~: M ~ R has essential maximum < c on almost all

~ - l e a v e s - t h a t is, for each .N-leaf ~p, there is a set Z e C~p such that
sup q)](@-Zp)<C, and for all ~pp not lying in a zero set of ~-leaves,
~, mes Zp=O. Then Z = ~ u [_)Zp is a zero set by the first half of (5.1),
P
and sup 4J](M - Z) < c, completing the proof of (5.1).
Proof of (5.2). For any ceR, let m ~ = ~ - l ( ( - ~ , c ] ) and let m ~ be
the set of o~a-leaves essentially contained in M ~. Then Z=M~AM~=
(M c - M ~)w (M ~ - M ~) has measure zero. Almost every ~-2 leaf meets Z
inessentially by (5.1). Therefore, almost every YZ-leaf meets M ~ essen-
tially iff it meets M r essentially.
Let e > 0 be small enough so that 2e-local product structure for
~ l , ~ 2 holds for all p~M:

XlGo~pl(~:) X2e~z(g) ~o~xx~(Zg)n ~ z ( z g , ) is a unique point.

Let Mp(e) be this product neighborhood of p in M. For small r,>0, we

also have
~1~2,/~'/~;--,, ~,~2(g/2) c Mp (~)

for all xl ~ ~e 1frO, x2 ~ ~ z (e)(see Fig. 6).

x2

/~,%J
Fig. 6. Local product structure

Let p be a point of M and suppose O,~p2 (g,) meets M c essentially. By

absolute continuity of ~ every other ~pZ(2e) meets M Cessentially for
qe@l(e). Thus, m o s t ~p2 meet M c essentially for qe.~l(e,), and on most
~p2, ~ is essentially constant. Therefore, the essential maximum of
on most ~p2, q~0~l(~:), is <c. By (5.1), the essential maximum of 4~ on
Me(e) is also <c.
On the other hand, suppose @2(~:) meets M c inessentially, By the
absolute continuity of ~1, every other ~qZ(r,/2) meets M C inessentially
for q e ~pl (e,). Thus, most o~2 (e,/2) meet M c inessentially for q e ~ 1 (e) and
2*
20 C. P u g h a n d M. Shub:

on most ~ z , 4) is essentially constant. Therefore, the essential minimum

of 9 on most ~q2, q ~ l ( ~ . ) is >c. By (5.1), the essential minimum of ~b
on Mr(e ) is also >c.
Consequently, for every c~ R,
ess max(dPlMp(~))<=c or ess min(q~lMp(e))>c.
Hence ~b is essentially constant on Mp(e), and so is essentially constant
on each component of M.
Suppose cb: M - ~ R is an A-invariant integrable function. Ergodicity
of A means q~ must be constant almost everywhere. Let lnv(g)=all
integrable g-invariant functions M - * R for g~ G. We are trying to show
Inv(g) is the set of constant functions.
g~G
According to [2, p. 144], we may define a projection I~: L1(M) ~ Inv(g)
by n
1 l k=-
I~ ~o(x)=,~lim 2 n + ~ q~ (gk x) g ~ G.

That is, the limit exists almost everywhere, is integrable, and q~~ I~ q~
is a continuous linear map onto the fixed points of I~, Inv(g). Moreover,
the limits n
I~ q~(x)= lim 1 ~=oCp(gkx) g~G
,~-+~ [ n [ + l k=
exist almost everywhere and I~ ~o(x)= lg ~o(x) for almost all x. That is,
I + = I~- = lg as maps L1(M) ~ Inv (g).
Since the continuous functions are dense in LI(M), their Ir
are dense in Inv(g). Therefore, it is useful to prove
I f r is continuous then Ij. q9 is essentially constant along ~r and ~/r (*)
For any x, y~ W~ and any continuous q~: M - ~ R it is clear that either
both I/~o(x), I f ~o(y) are defined, or neither, and if defined they are
equal. Since I/q~ is defined almost everywhere I f ~p is defined and
constant on almost all ~"-leaves. Since ~q~" is absolutely continuous
and lj7 ~o= II q~ almost everywhere, ly q~ is essentially constant on almost
every ~W" leaf by (5.1). Similarly for ~#F~,proving (.).
By density cb is the limit, almost everywhere o f / r q9with ~0continuous.
Therefore, on almost every ~/r leaf and ~/U~ leaf, cb is the pointwise limit,
almost everywhere on the leaf, of essentially constant functions. Hence ~b
is essentially constant along ~/r ~W~ and ~ : say 9 is essentially constant
on all ~r leaves, ~ leaves, and ~-leaves, not essentially intersecting Z,
mes Z = 0.
The foliations ~lWp", ~r W~" are both (!) smooth. ~ is smooth
on M so it is certainly smooth on W~"; ~r WpTM is smooth because W~'
 Ergodicity of Anosov Actions 21

is s m o o t h and all the other Wq", q e ~ , are gotten from W~ as g W/; = Wq"
for g in the identity c o m p o n e n t of G.
By absolute continuity of ~/Cc" and (5.1), almost every W~" meets Z
inessentially; by (5.1) on such a W~", almost every ~ , Wq" in W~'" meet
Z c~ W~'" inessentiatly. Therefore, by (5.2) on W~ u, 4) is essentially con-
stant on Wr"u. Thus 4) is essentially constant along ~//-c,.
By (5.2) on M and the absolute continuity o f f " " , ~-~, 4) is essentially
constant on M.
6. A Pathological Foliation
Here we give an example to show that there are foliations by s m o o t h
discs which are not measurable in the sense of Sinai [8]. It seems to us
that verification of a foliation's measurability is generally no easier than
verification of its measurewise smoothness. A conversation with N. Kopell
was helpful in cooking up our example.
Let I = [0, 1] and h: I x I-~I be continuous with
(i) h, = h (t, 9): I -~ I is a h o m e o m o r p h i s m , 0-< t_< 1.
(ii) h, = identity for t =<!3, ht=hl for t=> 2.
(iii) hi is not absolutely continuous.
(iv) h, I U is a C ~ embedding for some open dense U c I, 0 < t < 1 .
(v) dh]dt is continuous.
It is easy to construct such an h - w e do it at the end of this section.
Consider the foliation ,~- of I x I whose leaves are the graphs

fl(y)={t,h,y):tel} yel.
By (v), the foliation has a continuous tangent bundle. Since dh,/dt is
smooth on the dense strips {(t, hty): tel, y e U} there is no curve every-
where tangent to leaves but not contained in a leaf. Thus, we have a
foliation in the sense of A n o s o v [1, p. 18],
Let/~ be the usual measure on R 2. Let da, be the s m o o t h induced
R i e m a n n measure on the leaf/3. Let d/~, be the quotient measure on the
space of leaves, N'. If B is a collection of (whole) leaves, then ~t~(B)=
~( U/3)- Suppose that ~ were measurable in the sense of Sinai. Then
limB
there would be a measurable function K : I • I--* R such that
(1) K is positive almost everywhere on I • I.
(2) K is integrable on every leaf/3 not belonging to a set ~ of leaves
h a v i n g / ~ ( ~ ) = 0 and, f o r / 3 ~ . ~ Kd~a= 1.
P
(3) /~(A,/3)~ f ~ Kdaa is an integrable function of /3~.~ if / 3 r
Anp
and if A is measurable in I • I.
22 C, P u g h a n d M. S h u b :

(4) Ix(AmB)=~lx(A, fl)dl~ for any measurable set A c I • and

B
any measurable B c ~ .
Let N be the set where K is not defined or is not positive, p ( N ) = 0 .
By (4) with B = I • 1
/~(N) = ~ p (N, fl) d / ~
B

and so, for a set of leaves ~1 such that/~ ( ~ ) = 0,

flr ~ / J ( N , fl)=O.
Let Z be a zero set of I such that h 1 Z has positive linear measure
and let Bz = U fl(y). Then /x(Bz)>O because [2, l] x hl(Z)~B z. Also
yeZ
/~(Bz,) > 0 for Z'= {ye Z: fl(y)r ~ u ~ }
B~,=y~Z'
U/~(y)=Bz-(~u~O.
N o w let A = [ 0 , 3]
1 • I, B = Bz,. Then A c~B = [0, 89 • Z' so p (Am B) = 0.
Since each f l ~ B z, lies outside ~1, Klfl is almost everywhere positive
on ft. In particular, KIAc~fl is almost everywhere positive, fl~Bz,.
That is
/t(A, fl)> 0 for all flcBz,.
Since/t~ (Bz,) > 0, this proves that

II(A, fl) dlx~>O

Bz,

contradicting (4) for this A and B.

N o w we construct the h o m o t o p y h used to find the foliation. Let U
be an open dense subset of I with measure 89and let
x
u(x)= j" [1 -zu(s)] ds
0

where Zv is the characteristic function of U. This map u: I - - , [0,89

collapses U onto a countable set C c [0, 89 u](I - U) preserves measure,
and u-l(uU)=U. Let g: [0, 89 89 be a h o m e o m o r p h i s m with
g(0)=0, g())= 89 that is not absolutely continuous. Find an open set
V c I , pV= 89 and a collapsing map v: I---, [0,89 with v V = g C , v l ( l - V)
measure preserving, and v-~(v V)= V. Then define hi: I-+I so that
1~-, I

1. 1
[0, 89 - ~ [0, 89
 Ergodicity of Anosov Actions 23

commutes and hi carries /,/-I(c) onto v - l ( g c ) diffeomorphically for all

c~ C. Finally, put
h t (y) = [1 - qo(t)] y + (p (t) h 1(y)

for t, y e I and tp a C oo function R - ~ [ O , I ] with tp=O for t< 89 q~=l

for t > 2. Clearly h, is a h o m e o m o r p h i s m for all t e l and h, is smooth
in t. That is, (i)-(v) are verified.
Post Script. It seems likely that these method apply to metric
transitivity questions for Anosov Actions if such questions make any
sense.
References
1. Anosov, D.V.: Geodesic flows on closed Riemann manifolds with negative curvature.
Proc. of the Steklov Inst. of Math., No. 90 (1967); English translation A. M. S., Providence,
R.I., 1969.
2. - Sinai, Ja. G.: Some smooth ergodic systems. Upseki Mat. Nauk 22, No. 5, 107-172
(1967)=Russian Math. Surveys 22, No. 5, 103-167 (1967).
3. Hirsch, M.W.: Foliations and non-compact transformation groups (to appear).
4. - Pugh, C.C.: Stable manifolds and hyperbolic sets. Proc. of Symposia in Pure Math.,
Vol. XIV, A. M. S., Providence, R. I., 1970.
5. Shub, M.: Invariant manifolds (to appear).
6. Pugh, C.C., Shub, M.: Ergodic elements of ergodic actions. Compositio Mathematica,
Dec., 1970.
7. Schwartz, J.T.: Differential geometry and topology. New York: Gordon and Breach
1968.
8. Sinai, Ja. G.: Dynamical systems with countably multiple Lebesque spectrum II. Izv.
Akad. Nauk SSR, Set. Mat. 30, 15-68 (1966)=A.M.S. Trans. 68, Ser. 2, 34-88.

Charles Pugh Michael Shub

Mathematics Department Mathematics Department
University of California Brandeis University
Berkeley, Calif. 94720 Waltham, Mass.
USA USA

(Received December 10, 1970)