ENTROPY VERSUS HOMOLOGY FOR CERTAIN

DIFFEOMORPHISMS

RUFUS BOWEN
(Received 12 June 1973)

LET M be a compact connected manifold and f: M --t M a diffeomorphism of M. A closed

f-invariant subset A of M is hyperbolic if the tangent bundle restricted to A can be written
as the Whitney sum of continuous Dfinvariant subbundles T,, M = E” + Es with constants
c and 0 < i. < 1 so that

/Df”(o)l] < ci.“]lxil for u E E” and n > 0

and
11Df-“(u)(] < ci.“((uj( for K E E” and n > 0.

One says that f satisfies Axiom A [I41 if (a) the nonwandering set Q(f) = {x E M:
u n U f”( Or) # C$for every neighborhood U of x} is hyperbolic, and (b) the periodic points
“>O

off are dense in n(f). In this case Smale’s spectral decomposition theorem [14] says that
Q(J) = 52, u . . . u R, where the R, are pairwise disjoint f-invariant closed sets with flni
topologically transitive. Now define

H’*(Qi) = {X E M: d(f”(x), Q) -+ 0 as n+co}

lV”(Q,) = {x E M: d(f’-“(x), QJ ---*0 as n + co}

and write Ri > Rj if W“(Q,) n W(Q,) # C#J.We say thatfhas no cycles if 2 can be enlarged
to make a total ordering, i.e. if there is a permutation CTE S, with a(i) 2 o(i) whenever
Ri > Rj. In this case we renumber the R,‘s so !2t < R, < ... < 52, is the total ordering.

We are interested in the eigenvalues of the mapf, on homology. Our main result is the
following (s(A) denotes the spectral radius of a matrix A):

THEOREM. Let f be a d@omorphism on M satisfying Axiom A with no cycles and

dim Szcf) = 0. Let f* : H,(M) + H,(M) be the induced map on real homology. Then
log s(f*) < h(f) where hCf) is the entropy off.

This theorem was conjectured by M. Shub [without the restriction dim QCf) = 0).
The author knows no diffeomorphism with h(f) < logscf,). We thank M. Shub and
E. Spanier for conversations helpful to this paper.
61
62. ENTROPY VERSL’S H05tOLOGY FOR CERT.AIS DlFFEOItORPtfISMS

$1. VARIOUS SILPOTEST hI_APS

Throughout the paper f will always satisfy Axiom A and have no cycles; homology
will have real coefficients. We review the filtrations of [lj]. One can find compact sets
Ml c M, c . . . c hi, = :Cf with

and M\M, u IV(Q) = M\ltl, u n f-k(M\Ms_l) for all 1 < s < nr. We shall assume
k>O

the M, are smooth submanifolds with boundary and dim Al, = dim A4 (this can be obtained
by enlarging the M,‘s slightly if necessary).
Now fix an s. Let X = M-r u kV”(Q,) and fi”(X, hi,_,) be Cech cohomology with
realcoefficients.Theinclusionk: (A’, AI,_,) c (M,, iCf,_r)inducesamapk*: H’(M,, _V_,) =
fik(Ms, MS_,) + B”(X, MS_,).
LEMMA 1. Suppose V is a subspace of Hk(Ms, MS_ ,) with f*(Y) = V. Then k* / V is
one-to-one.
Proof: By the continuity property of Cech theory [S, p. 2611, one has
fi”(X, M-t) = lim Hk(j”“(Ms) u MS-,, .Cfs_,)
m
since

(the limit is the direct limit induced by inclusion maps f”‘(MJ u M3_, cfm(Ms) u M,_,
for m’ > m). If k*(u) = 0, with L‘# 0, then i, *u = 0 for some inclusion i, : cf,(itC,) u
iv,-,, M,_,) --+(MS, M_,). We have a commutative diagram of maps:

where g = f” but with different range. Then (f*)“u = (J‘“)*v = g* i,*~’ = 0. SO (f+)” is not
one-to-one on V; this contradictsf*( V) = V as V is finite dimensional.
Now let qs = dim E,” where x E R,. This is the same for all s E R, since f preserves
E”, flfl, is topologically transitive and E” continuous. For E > 0 small let M’,“(Q,) =
U w,“(x) (see [91h
rsR,
LEMMA 2. Zfdim R, = 0, then

fik(X, X\ W,“(Q,)) = 0 for k > q.

 RUFUS BOWEN 63

Proof: The inclusion, (I+‘:(!&), 2 @‘,U(!&))4 (X, X\IV,Y(Q,)) is a relative homeomorph-

ism, so [5. p. 2661

Ak( x, x\ IV,“(Q)) 1: Ak( IV,Y(Q,), s It’,“@-&)).

These latter groups will be 0 for rl- > q by a general theorem [lo, p. 1521 provided we show
dim IV,“(Q) = q.

For x E R,, Z, = WC’(x) n R, is zero dimensional since R, is. By canonical coordinates

[Id7 P. 7811,PI c: = ,g ~,"W is a neighborhood of x in IV,‘(Q) and a disjoint union.
x
Recall [S], [9] that there is a continuous 4: R, + Emb(DP, M) (D4 is the q-disk) so that
WC”(y) = 4,,(D). The map

lc/.Z,x Dq-+Ux

given by I&, t) = 4”(t) is continuous and a bijection. So

dim U, = dim(Z, x 04) = q,

LEMMA 3. If dim R, = 0, then f*: Hk(Ms, Ms_1)-) Hk(M,; M,_,) and f*: Hk(hfs,
Ms_I)~Hk(Ms, M,_,)arenilpotentfir k > qs.

Proof. Since we are using real coefficients, these maps are duals and we need only
consider the cohomology. First we show that f *: Ak(X, A4 s-l)-+ B'(X, MS_,) is nilpotent.
Since W = X\ W,U(Q,) is a compact set disjoint from Q, and X c R, u U f -"(int M,_,),
">O
we get f "( W) c MS_ 1 for some m > 0. Then

commutes. So (f*)” = (f”)* on fi”(X, M,_,) factors through kk(X, W) and so is zero by
Lemma 2.

If f*: P(M,, M,_,)4P(M,, MS_,) were not nilpotent, then there would be a
subspace V # 0 with f * V = V. The diagram
(I’)”
Hk(M,, MS-,) - Hk(M,, MS-,)

-
(I’)”
fi’(X,
Ik*

MS_,)

commutes. Then

(f*)“k* V = k*(p)” V = k* V # 0

by Lemma 1; but (f*)m = 0 on fi”(X, MS- ,), a contradiction.

6-l ENTROPY VERSUS HOMOLOGY FOR CERT.AlN DIFFEOYORPHlSMS

LEMMA 4. Assume ,ll is orientable and n = dim .Cf. T/renf, on H,(,tf,, .LfS_l) is similar
to (f-l)* on H”-‘(.W lLfx_l. :tl‘, ,_tl,) iff preserves orientation and to -(j-l)* lff recerses
orientation.
Proof. Duality [16, p. 2961 gives an isomorphism

2: H,(int _\i,, int :tIs_r) -+ H”-‘(&I\ :Lf,_,, LII\MS)

(here H”-’ 2 I!?‘-~ as ~\ii\M~_r and .CI\M, are compact submanifolds). We likewise get
an isomorphism
zr: H,(intf(M),, intf(M~_,)) + H”-k(f(;ti\M,_,),f(iti\:t~~)).
Consider the diagram

Here i and j are inclusions; the upper and lower triangles commute because of the naturality
of induced maps. The right square commutes (ZI..+= j*r,) because duality is natural with
respect to inclusion. Iffis orientation preserving, then the back square commutes (rrf* =
(f-‘)*a) because duality is an invariant of oriented manifolds; iffreverses orientation, then
this commutes up to a minus sign. The diagram chase now implies that the left square
commutes up to sign
Remark. A particular case of this lemma was given in Gibbons [6].
THEOREM 1. Assume thatf satis$es Axiom A and the no cl4eproperty and that dim R, = 0.
Then
f* : H,(M,, M_,) --t H,(M,, :W_I)
is nilpotent for k # qS = dim E,“(x E Q,).
Proof: Assume M is orientable. As M is connected,feither preserves or reverses orien-
tation. For k > q the statement is just Lemma 3. For k < q it is enough by Lemma 4 to see
that (f-r)* on H”-k(lV\M,_,, M’\hl,) is nilpotent. Now f-’ satisfies Axiom A and the
no cycle property, Q(f-‘) = Q(f) and

is a filtration forf-‘. Now E,“,/-, = E:, / for s E R,. Lemma 3 applied tof-’ gives us that
(I-‘)* is nilpotent on H’(M\M_,, M‘\ill,) for I > r = dim E,“.,_, = dim E:,/. Since
T,M = El,/ + Ex”,, we get r + q = n. Hence I = n - k = r + q - k > r for k < q.
 RUFUS BOWEN 65

Consider the case where M is not orientable. Let fi be the orientable double cover and
rr: lcf -) M. Thenflifts to a diffeomorphismfon iif. Nowf’satisfies Axiom A,

and a-‘&I, c ..‘c 7r-r Mm._, c &’ is a filtration for J? One might have f] n-‘R, not
topologically transitive but that doesn’t matter. The proof in the orientable case still gives
us thatf, on Hk(n-‘,VS, i;-r MS_,) is nilpotent for k # dim E,” = dim _$‘(x) = q where
x E rr_rR,. For real coefficients the map

“* : Hk(7r-l MS, 11-l MS_,) -+ H,(ICI,, hi,_*)

is surjective ([4]; each singular chain f for M is covered by two ft, fr for #; the map f -+
+[fi +fJ gives a map on chains commuting with d and inducing a right inverse for X*
on homology). It follows that& is nilpotent on H,(MS, M-,) for k # q.

$2. THE NONZERO EIGENVALLJES

Let us recall the structure of a zero-dimensional R,. For an n x n matrix .4 of O’s and
I’swedefinex,c n{l,...,n>by

CA = {(Xi)~Yrxj : Axixi** = 1 V’iEZ}.

If{l,..., n> is given the discrete topology and CA a topology as a subspace of the product,
then CA is a compact metrizable space. There is a shift homeomorphism 0 on x,, defined by

G(xJ~.~ = (~;)~~z where xi = xi+,

&&+xA 1s ca IIe d a subshiji offinite typeif c on CA is topologically transitive. It turns out

that the zero-dimensional basic setsf/R, are topologically conjugate to subshifts of finite
type [2, pages 36-371. One should note that A is not uniquely determined by f/R,. For
B = [Bj] we write 1BI = [/Bij( 1.

THEOREM 2. Let f sarisfy Axiom A and the no cycle property and assume dim R, = 0.
T/ren there is a matrix B with entries 0, + 1 which has the same nonzero eigenvaiues as f, :
H&M,, M-,) -+ HJM,, M_,> andsuch that CJ: xlB, --+ ciB, a subshift qfjinite type conjd-
gate tofIR,.

Proof Let CA be a subshift of finite type conjugate tofl R, and 11:CA -+ R, be a homeo-
morphism with ha = fh. If A is n x n, let C, = {(xJieZ E IA : x0 = k) for 1 < k < n.
Given any E > 0, A and 11may be chosen so that diam h(C,) < E for all I < k < n (n + co
as E + 0); see [2, pages 36-371. Since R, is totally disconnected, E”]R, can be oriented.

Let A(X) = & 1 according to whether DfX : E,” -+ E”/(Xj preserves or reverses orientation.
As A is continuous in s E R,, choose E > 0 so that A is constant on sets of diameter less
than E. Define B = [Bij] by
Bi, = A(h(C,))Aij.

TOP Vol. 13 No. I-E

66 ENTROPY VERSUS HOMOLOGY FOR CERTAIS DlFFEOVORPHISXlS

For x E Q, define Am(l) = i: 1 according to whether Of;“: E,” -t E;imC,)preserves or reverses

orientation. Let
Iv” = {(ro. r,, . , r,): r = (T~)~5z E x.,1-.
For (rO, r,, , r,) the number

A(r, , . . >r,J = &O(r)) = ~r,,,,~,I .,: Br,. ,,rm

depends on (r,, , . , r,) but not the rest of r. By induction one sees that the i, jth entry in
B” equals
1 {A(rO , , r,): (r,, , , r,) E IV,,,, r. = i. r,,, = j].
In particular
TrB” = x{A(r,, , r,): (rO, , r,) E W’,,,, rO = r,>.
Let Per,,, = {X E R, : f”x = I), and h-‘(s) = (r,(s)),, z. Since h conjugates d and f] R,,
we see that r,(x) = r,Jx) when .YE Per,,, In fact I -+ (rO(z), , r,)(x)) defines a bijection
Per, -+ {(rO, , r,) E W, : rO = r,}.
Since A(r,,(x), . . , r,(x)) = A,(X) we get
TrB” = ~{A,(x): .YE Per,).
Now by [14, Q. 761 the index of the fixed QOiIlt .r off” equals (- l)4 A,,,(X). So L(f;“) =
(-I)‘~{AJx): .r~Per,J where f,: H,(h(,, M,_I)+ H,(M,, M5_,).By Theorem 1,
f, is nilpotent in dimensions k # q. So L(f*“) = (- 1)’ Tr f,:.
We get TrB” = Tr(&f4*)m
for all m > 1. From this it follows that f,*and B have the same nonzero eigenvalues,

counting multiplicity (these are determined by the poles of the function exp 1 ‘L TrB”
i (,,I 11
t141,
[31).
Remark. This proof is closely related to the various calculations of the zeta function
U41, 131,1171,[71,hl.
COROLLARY. IfE”IR, can be oriented so that Df 1E” is orientation presewing, then
one can assume B = A. If E”j!2, can be oriented so that Df] E” is orientation rerersing,
one can assume B = -A.

$3. ENTROPY

The topology entropy [I], [2] h(a) of 0: I,, - 1, is equal to log s(A) where s(A) is the
largest absolute value of any eigenvalue of A.
LEMMA 5. For any matrix B,

s(lBl) 3 s(B).
Proof. Consider the norm IJBII, = sup 1BijI. By induction one sees 11 B”jjm < I/ 1 Bl “llm.
Hence by [I l] (and the equivalence of 11 I]D to the operator norm)

s(] BI) = li; i log II I Bj”ll, 3 s(B).

 RUFUS BOWEN 67

THEOREM. Let f satisfy Axiom A and the no cycle property with dim Q(f) = 0. Then
h(f) >4log scf:) where f, is the induced map on real homology.
Proof: Now h(a) = h(fjR,) <h(f); so by Lemma 5 and Theorems 1 and 2

h(f) Z log s(A) 2 log s(f*,r)

where f*.3: H,(M,, ;Li,_,) + H,(M,, M,_l).
Using the exact sequence of homology, by induction on s we get thatf, : H,(rC/,) + H,(MJ
satisfies h(j) 2 log s&J. For M, = M we get our result.
COROLLARY (M. Shub [13]). For f a Morse-Smale dtfiomorphism, all eigenvalues off,
are on the unit circle.
Proof. We have Q(f) finite. So h(f) =0 and s(f*) < 1. Also (sf*-‘) d 1 as
Q(f -‘) = Q(f). This implies the result.

REFERENCES
1. R. L. ADLER, A. G. KONHEIMand M. H. MCANDREW: Topological entropy, Trans. Am. mnth. Sot.
114 (1965) 309-319.
2. R. BOWEN:Topological entropy and Axiom A, Proc. Symp. Pure Math 14 (1970). 23-41.
3. R. BOWENand 0. LANFORD:Zeta functions of restrictions of the shift transformation, Proc. Symp. Pure
Marh. 14 (1970), 43-49.
4. B. ECKMANN:Coverings and Betti numbers, Bull. Am. math. Sot. 55 (1949). 95-109.
5. S. EILENBERG and N. STEENROD:Foundations of Algebraic Topology.
6. J. GIBW&S: A note on one-dimensionai attracting sets in the three-sphere, Proc. Am. math. Sot. 31
(1972). 620-622.
7. J.GUCKENHEIMER: Axiom A and no cycles imply c,(j) rational, Bull. Am. mar/t. Sot. 76 (1970), 592-594.
8. M. HIRSCHand C. PUGH: Stable manifolds and hyperbolic sets, Proc. Symp. Pure Math. 14 (1970).
133-163.
9. M. HIRSCH,J. PALIS.C. PIJGHand M. SHUB: Neighborhoods of hyperbolic sets, Invent. Moth. 9 (1970).
121-134.
10. W. HIJREWICZand H. WALLMAN:Dimension Theory.
11. L. LOOMIS:Introduction to Abstract Harmonic Analysis, p. 75.
12. A. MANNING:Axiom A diffeomorphisms have rational zeta functions, Bull. Lond. math. Sot. 3 (19711,
2 15-220.
13. M. SHUB: Morse-kale Diffeomorphisms are Unipofenf on Homology. Proceedings of the Symposium on
Dynamical Systems, Salvador. Academic Press (I 973).
14. S. SMALE: Differentiable dynamical systems, Bull. Am. moth. Sot. 73 (1967). 747-817.
IS. S. SMALE:The R-stability theorem, Proc. Symp. Pure Math. 14 (1970), 289-297.
16. E. SPANIER:Algebraic Topology.
17. R. WILLIAMS:Zeta function in global analysis, Proc. Symp. Pure Math. 14 (1970), 335-339.