Contemporary Mathematics

Rational Homotopy Theory:

A Brief Introduction
Kathryn Hess
Abstract. These notes contain a brief introduction to rational homotopy
theory: its model category foundations, the Sullivan model and interactions
with the theory of local commutative rings.
Introduction
This overview of rational homotopy theory consists of an extended version of
lecture notes from a minicourse based primarily on the encyclopedic text [18] of
Felix, Halperin and Thomas. With only three hours to devote to such a broad
and rich subject, it was dicult to choose among the numerous possible topics to
present. Based on the subjects covered in the rst week of this summer school,
I decided that the goal of this course should be to establish carefully the founda-
tions of rational homotopy theory, then to treat more supercially one of its most
important tools, the Sullivan model. Finally, I provided a brief summary of the ex-
tremely fruitful interactions between rational homotopy theory and local algebra,
in the spirit of the summer school theme Interactions between Homotopy Theory
and Algebra. I hoped to motivate the students to delve more deeply into the
subject themselves, while providing them with a solid enough background to do so
with relative ease.
As these lecture notes do not constitute a history of rational homotopy theory,
I have chosen to refer the reader to [18], instead of to the original papers, for the
proofs of almost all of the results cited, at least in Sections 1 and 2. The reader
interested in proper attributions will nd them in [18] or [24].
The author would like to thank Luchezar Avramov and Srikanth Iyengar, as
well as the anonymous referee, for their helpful comments on an earlier version of
this article.
Basic notation and terminology. We assume in this chapter that the reader
is familiar with the elements of the theories of simplicial sets and of model categories.
As references we recommend [13] or [25] or the chapter of these lecture notes by
Paul Goerss [20].
2000 Mathematics Subject Classication. 55P60.
Work completed during a visit to the Institut Mittag-Leer (Djursholm, Sweden).
c 0000 (copyright holder)
1
2 KATHRYN HESS
In this chapter, sSet and Top are the categories of simplicial sets and of topo-
logical spaces, respectively. Furthermore, | | : sSet Top denotes the geometric
realization functor, while S

: Top sSet denotes its right adjoint, the singular

simplices functor.
If K is a simplicial set, then C

(K) and C

(K) denote its normalized chain

and cochain complexes, respectively. If X is a topological space, then the singular
chains and cochains on X are S

(X) := C

_
S

(X)
_
and S

(X) := C

_
S

(X)
_
.
A morphism of (co)chain complexes inducing an isomorphism in (co)homology
is called a quasi-isomorphim and denoted

.
A graded vector space is said to be of nite type if it is nite dimensional in
each degree. A CW-complex is of nite type if it has nite number of cells in each
dimension.
Given a category C and two objects A and B in C, we write C(A, B) for the
class of morphisms with source A and target B.
1. Foundations
For the sake of simplicity, we work throughout these notes only with simply
connected spaces. Many of the results presented hold for connected, nilpotent
spaces as well.
1.1. Rationalization and rational homotopy type. Let

H

denote re-
duced homology.
Definition 1.1. A simply connected space X is rational if the following, equiv-
alent conditions are satised.
(1)

X is a Q-vector space.
(2)

H

(X; Z) is a Q-vector space.

(3)

H

(X; Z) is a Q-vector space.

To prove the equivalence of these conditions, one begins by observing that
H

(K(Q, 1); F
p
)

= H

(pt.; F
p
) for all primes p. An inductive Serre spectral sequence
argument then shows that H

(K(Q, n); F
p
)

= H

(pt.; F
p
) for all primes p and all
n 1. The equivalence of conditions (1) and (2) for an arbitrary X then follows
from an inductive argument on the Postnikov tower of X. On the other hand, the
equivalence between conditions (2) and (3) can be easily veried by a Serre spectral
sequence argument.
Example 1.2. For any n 2, let
n,k
denote the homotopy class of the inclusion
of S
n
as the k
th
summand S
n
k
of
_
k1
S
n
k
. The rational n-sphere is dened to be
the complex
(S
n
)
0
:=
_

k1
S
n
k
_
_
_

k2
D
n+1
k
_
,
where D
n+1
k+1
is attached to S
n
k
S
n
k+1
via a representative S
n
S
n
k
S
n
k+1
of

n,k
(k + 1)
n,k+1
. The rational n-disk is then
(D
n+1
)
0
:= (S
n
)
0
I/(S
n
)
0
{0}.
Let
X(r) :=
_

1kr
S
n
k
_
_
_

2kr1
D
n+1
k
_
.
RATIONAL HOMOTOPY THEORY 3
It is clear that for all r, S
n
r
is a strong deformation retract of X(r), which implies
that H
k
X(R) = 0 if k = 0, n. Furthermore, the homomorphism induced in reduced
homology by the inclusion X(r) X(r + 1) is multiplication by r + 1. Since
homology commutes with direct limits and (S
n
)
0
= lim

X(r),

_
(S
n
)
0
; Q) =
_
Q : k = n
0 : else.
Definition 1.3. A pair of spaces (X, A) is a relative CW
0
-complex if X =

n1
X(n), where
(1) X(1) = A;
(2) for all n 1, there is a pushout

Jn
(S
n
)
0
incl.

X(n)

Jn
(D
n+1
)
0

X(n + 1);
(3) X has the weak topology, i.e., U X is open in X if and only if U X(n)
is open in X(n) for all n.
The pairs
_
(S
n
)
0
, S
n
_
and
_
(D
n+1
)
0
, D
n+1
_
are the fundamental examples of
relative CW
0
-complexes.
Note that if A is a rational space and (X, A) is a relative CW
0
-complex, then
X is rational as well.
Definition 1.4. Let X be a simply connected space. A continuous map :
X Y is a rationalization of X if Y is simply connected and rational and

Q :

X Q

Y Q

Y
is an isomorphism.
Observe that a map : X Y of simply connected spaces is a rationalization
if and only if H

(; Q) is an isomorphism.
The inclusions of S
n
into (S
n
)
0
and of D
n+1
into (D
n+1
)
0
are rationalizations.
The rationalization of an arbitrary simply connected space, as constructed in the
next theorem, generalizes these fundamental examples.
Theorem 1.5. Let X be a simply connected space. There exists a relative
CW-complex (X
0
, X) with no zero-cells and no one-cells such that the inclusion
j : X X
0
is a rationalization. Furthermore, if Y is a simply connected, rational
space, then any continuous map f : X Y can be extended over X
0
, i.e., there is
a continuous map g : X
0
Y , which is unique up to homotopy, such that
X
f

B
B
B
B
B
B
B
B
Y
X
0
g

}
}
}
}
}
}
}
}
commutes.
4 KATHRYN HESS
Proof. We provide only a brief sketch of the proof. We can restrict to the
case where X is a 1-reduced CW-complex. The rationalization X
0
can then be
constructed as a CW
0
-complex with rational n-cells in bijection with the n-cells of
X, for all n. The attaching maps of X
0
are obtained by rationalizing the attaching
maps of X. The complete proof can be found in [18, Theorem 9.7].
Continuing in the same vein, one can show that such a cellular rationalization
is unique up to homotopy equivalence, relative to X.
Given a continous map : X Y between simply connected spaces, we let

0
: X
0
Y
0
denote the induced map between their rationalizations, the existence
and uniqueness (up to homotopy) of which are guaranteed by Theorem 1.5.
Definition 1.6. The rational homotopy type of a simply connected space X
is the weak homotopy type of its rationalization X
0
.
Definition 1.7. A continuous map : X Y between simply connected
spaces is a rational homotopy equivalence if the following, equivalent conditions are
satised.
(1)

() Q is an isomorphism.
(2) H

(; Q) is an isomorphism.
(3) H

(; Q) is an isomorphism.
(4)
0
: X
0
Y
0
is a weak homotopy equivalence.
To simplify computations, it is common in rational homotopy theory to restrict
to the class of spaces dened by the following proposition, the proof of which is in
[18, Theorem 9.11].
Proposition 1.8. For any simply connected space X, there is a CW-complex
Z and a rational homotopy equivalence : Z X such that
(1) H

(X; Q) is of nite type if and only if Z is of nite type; and

(2) if dim
Q
H

(X; Q) < , then H

(X; Q) = H
N
(X; Q) if and only if Z is
a nite CW-complex of dimension at most N.
Definition 1.9. A simply connected space X is of nite rational type if con-
dition (1) of Proposition 1.8 is satised.
We can now nally specify clearly the subject presented in these notes.
Rational homotopy theory is the study of rational ho-
motopy types of spaces and of the properties of spaces
and maps that are invariant under rational homotopy
equivalence.
For further information on rationalization, the reader is refered to Section 9 of
[18].
1.2. The passage to commutative cochain algebras. We show in this
section that the category of rational homotopy types of simply connected, nite-type
spaces and of homotopy classes of maps between their representatives is equivalent
to an appropriately dened homotopy category of commutative dierential graded
algebras over Q.
RATIONAL HOMOTOPY THEORY 5
The algebraic category and its homotopy structure. We begin by a rather careful
introduction to the algebraic category in which the Sullivan model of a topological
space lives.
A commutative dierential graded algebra (CDGA) over Q is a commutative
monoid in the category of non-negatively graded, rational cochain complexes (cf.,
[20]). In other words, a CDGA is a cochain complex (A

, d) over Q, endowed with

cochain maps
: Q (A

, d),
called the unit, and
: (A

, d)
Q
(A

, d) (A

, d) : a b a b,
called the product, such that
(1) is graded commutative, i.e., if a A
p
and b A
q
, then a b = (1)
pq
b a;
(2) is associative; and
(3) ( Id
A
) = Id
A
= (Id
A
).
Observe that since is a morphism of chain complexes, the dierential on a CDGA
satises the Leibniz rule, i.e., if a A
p
and b A
q
, then
d(a b) = da b + (1)
p
a db.
Let r N. A CDGA A is r-connected if A
0
= Qand A
k
= 0 for all 0 < k < r+1.
A morphism of CDGAs f : (A

, d, , ) (

A

,

d, , ) is a cochain map such
that f = (f f) and f = . The category of CDGAs over Q and their
morphims is denoted CDGA
Q
.
To simplify notation, we frequently write either A or (A, d) to denote (A

, d, , ).
Furthermore, henceforth in these notes, the notation means
Q
.
In rational homotopy theory, CDGAs with free underlying commutative, graded
algebra play an essential role. Given a non-negatively graded vector space V =

i0
V
i
, let V denote the free, commutative, graded algebra generated by V , i.e.,
V = S[V
even
] E[V
odd
],
the tensor product of the symmetric algebra on the vectors of even degree and of
the exterior algebra on the vectors of odd degree. Given a basis {v
j
| j J} of V ,
we often write (v
j
)
jJ
for V . We also write
n
V to denote the set of elements
of V of wordlength n.
A homomorphism of commutative, graded algebras : V A is determined
by its restriction to V , as is any derivation of commutative, graded algebras :
V A. In particular, the dierential d of a CDGA (V, d) is determined by its
restriction to V .
More generally, the following class of CDGA morphisms is particularly impor-
tant in rational homotopy theory.
Definition 1.10. A relative Sullivan algebra consists of an inclusion of CDGAs
(A, d) (AV, d) such that V has a basis {v

| J}, where J is a well-ordered

set, such that dv

AV
<
for all J, where V

is the span of {v

| < }.
A relative Sullivan algebra is minimal if
< = deg v

< deg v

.
A (minimal) relative Sullivan algebra (V, d) extending (A, d) = (Q, 0) is called a
(minimal) Sullivan algebra.
6 KATHRYN HESS
Remark 1.11. If V
0
= 0 = V
1
, then (V, d) is always a Sullivan algebra and
is minimal if and only if dV
2
V , the subspace of words of length at least two.
Example 1.12. The CDGA ((x, y, z), d), where x, y and z are all of degree 1
and dx = yz, dy = zx and dz = yx, is an example of a CDGA with free underlying
graded algebra that is not a Sullivan algebra.
Recall from Example 1.7 and Example 3.4(1) in [20] that the category Ch

(Q)
of non-negatively graded cochain complexes over Q admits a cobrantly generated
model category structure in which
(1) weak equivalences are quasi-isomorphisms;
(2) brations are degreewise surjections; and
(3) cobrations are degreewise injections, in positive degrees.
The set of generating acyclic cobrations is
J = {j
n
: 0 D(n) | n 1}
where
D(n)
k
=
_
Q : k = n 1, n
0 : else
and d : D(n)
n1
D(n)
n
is the identity ma The set I of generating cobrations
consists of the inclusions
i
n
: S(n) D(n) for n 1 and i
0
: 0 S(0),
where
S(n)
k
=
_
Q : k = n
0 : else.
Consider the pair of adjoint functors
: Ch

(Q) CDGA
Q
: U,
where is the free commutative cochain algebra functor satisfying (C, d) =
(C,

d), where

d is the derivation extending d, and U is the forgetful functor. It
is not dicult, as indicated in Example 3.7 in [20], to show that this adjoint pair
satises the hypotheses of Theorem 3.6 of [20].
There is thus a cobrantly generated model structure on CDGA
Q
, with gen-
erating cobrations and acyclic cobrations
I = {i
n
| n 0} and J = {j
n
| n 1},
where (0) := Q. Let I cell denote the smallest class of morphisms in CDGA
Q
that contains I and that is closed under coproducts, cobase change and sequential
colimits. It is easy to see that Icell is exactly the class of relative Sullivan algebras.
In this model structure on CDGA
Q
, weak equivalences are quasi-isomorphisms,
and brations are degreewise surjections. Cobrations are retracts of relative Sul-
livan algebras. All CDGAs are brant, and the Sullivan algebras are the cobrant
CDGAs.
The next result, which is the Lifting Axiom of model categories in the specic
case of CDGA
Q
, is an important tool in rational homotopy theory.
RATIONAL HOMOTOPY THEORY 7
Proposition 1.13 (The Lifting Lemma). Let
(A, d)
i

(B, d)
p

(A V, D)
g

(C, d)
be a commuting diagram in CDGA
Q
, where i is a relative Sullivan algebra and p is
a surjection. If i or p is a quasi-isomorphism, then g lifts through p to an extension
of f, i.e., there exists a CDGA map h : (AV, D) (B, d) such that hi = f and
ph = g. Furthermore, any two lifts are homotopic relative to (A, d).
We can describe homotopy of CDGA morphisms with source a Sullivan algebra
in terms of the following path objets. Let I denote the CDGA D(1), where the
generators of degrees 0 and 1 are called t and y, respectively. Let
0
: I Q and

1
: I Q denote the augmentations specied by
0
(t) = 0 and
1
(t) = 1.
Proposition 1.14. Let (V, d) be a Sullivan algebra. Two CDGA morphisms
f, g : (V, d) (A, d) are homotopic if and only if there is a CDGA morphism
H : (V, d) (A, d) I such that (Id
A

0
)H = f and (Id
A

1
)H = g.
A careful, degree-by-degree version of the proof of the Small Object Argument
(Theorem 3.5 in [20]) establishes the following useful result.
Proposition 1.15. Any morphism f : (A, d) (B, d) in CDGA
Q
can be
factored as
(A U, D)
p

M
M
M
M
M
M
M
M
M
M
(A, d)
i

q
q
q
q
q
q
q
q
q
q
f

M
M
M
M
M
M
M
M
M
M
(B, d)
(A V, D)
q

q
q
q
q
q
q
q
q
q
q
where i and j are relative Sullivan algebras, and p and q are surjections.
If we are willing to sacrice surjectivity of q, we can obtain minimality of j,
again via a degree-by-degree construction.
Proposition 1.16. Any morphism f : (A, d) (B, d) in CDGA
Q
can be
factored as
(A, d)
f

M
M
M
M
M
M
M
M
M
M
(B, d)
(A W, D)

q
q
q
q
q
q
q
q
q
q
where is a relative Sullivan algebra. In particular, if H
0
A = Q = H
0
B, H
1
f is
injective and H

B is of nite type, then W is of nite type and W = W

2
.
Definition 1.17. The quasi-isomorphism : (A W, D)

(B, d) is a
relative Sullivan minimal model of f : A B. A Sullivan minimal model of the
CDGA (B, d) is a relative Sullivan minimal model : (W, d)

(B, d) of the unit
map : (Q, 0) (B, d).
8 KATHRYN HESS
It is very convenient to know that (relative) Sullivan minimal models are unique
up to isomorphism, which is an immediate consequence of the following proposition.
Proposition 1.18. Suppose that
(A, d)
i

(B, d)
j

(A V, D)

(B W, D)
is a commuting diagram in CDGA
Q
, where i and j are minimal relative Sullivan
algebras, f is an isomorphism and

f is a quasi-isomorphism. Then

f is also an
isomorphism.
The proof of this proposition reduces to showing that if a CDGA endomorphism
of a minimal relative Sullivan algebra (B W, D) that xes B is homotopic to
the identity, then it is equal to the identity.
The functors. We now explain the passage from topology to algebra, starting
with the relationship between simplicial sets and CDGAs.
Definition 1.19. The algebra of polynomial dierential forms, denoted A

, is
the simplicial CDGA given by
A

n
=
_
(t
0
, ..., t
n
; y
0
, ..., y
n
)/J
n
, d
_
,
where deg t
i
= 0 and dt
i
= y
i
for all i and J
n
is the ideal generated by {1

n
i=0
t
i
,

n
j=0
y
j
}. The faces and degeneracies are specied by

i
: A

n
A

n1
: t
k

_
_
_
t
k
: k < i
0 : k = i
t
k1
: k > i
and
s
i
: A

n
A

n+1
: t
k

_
_
_
t
k
: k < i
t
k
+t
k+1
: k = i
t
k+1
: k > i.
The terminology used in this denition is justied by the following observa-
tion. Let
DR
(
n
) be the cochain algebra of smooth forms on
n
, the standard
topological n-simplex. Then

DR
(
n
) = C

(
n
)
A
0
n
A

n
,
where the cochain algebra morphisms induced by the topological face inclusions
and degeneracy mapsagree with
i
and s
i
.
Definition 1.20. Let A

: sSet CDGA
Q
be the functor specied by
A

(K) = sSet(K, A

), with product and dierential dened objectwise. For any

topological space X, let A
PL
(X) := A

_
S

(X)
_
, which we call the CDGA of
piecewise-linear de Rham forms on X.
Since A

(K) is a commutative algebra for every simplicial set K, while C

(K; Q)
usually is not, we cannot expect to be able to dene a natural quasi-isomorphism
of cochain algebras directly from the former to the latter. However, as explained
below, there is a cochain map between them that is close to being an algebra ma
RATIONAL HOMOTOPY THEORY 9
Given f A
n
(K), i.e., f : K A
n

, and x K
n
, write
f(x) =

f(x)dt
1
dt
n
,
so that

f(x) Q[t
1
, ..., t
n
], the ring of polynomials in t
1
, ..., t
n
with coecients in
Q. Dene a graded linear map
_
: A

(K) C

(K; Q) by
_
_
f
_
(x) =
_

f(x)dt
1
dt
n
.
Theorem 1.21 (The Polynomial Stokes-De Rham Theorem). The map
_
is a
map of cochain algebras, inducing an isomorphism of algebras in cohomology.
The proof of this theorem, which can be found in [10], Theorem 2.2 and Corol-
lary 3.4, proceeds by methods of acyclic models.
Theorem 1.21 can in fact be strengthened: as proved in Proposition 3.3 in [10],
the cochain map
_
is actually a strongly homotopy multiplicative map, in the sense
of, e.g., Gugenheim and Munkholm [21].
To compare the homotopy theory of CDGAs and of simplicial sets via the
functor A

, we need A

to be a member of an adjoint pair. We construct its

adjoint as follows.
Definition 1.22. Let K

: CDGA
Q
sSet be the functor specied by
K

(A) = CDGA
Q
(A, A

), with faces and degeneracies dened objectwise.

Let sCDGA
Q
denote the category of simplicial CDGAs. Given any CGDA A
and any simplicial set K, their cartesian product is naturally a simplicial CDGA.
Furthermore, there are natural isomorphisms
CDGA
Q
_
A, sSet(K, A

)
_

= sCDGA
Q
(A K, A

= sSet
_
K, CDGA
Q
(A, A

)
_
,
i.e.,
CDGA
Q
_
A, A

(K)
_

= sSet
_
K, K

(A)
_
.
We therefore have an adjoint pair
A

: sSet CDGA
op
Q
: K

,
which Bouseld and Gugenheim proved to be a Quillen pair in Section 8 of [10].
Definition 1.23. The composite functor
CDGA
Q
K

K
K
K
K
K
K
K
K
K
K
<>

Top
sSet
||

w
w
w
w
w
w
w
w
is called spatial realization.
Let (V, d) be any Sullivan algebra. Let : Id A

be the unit of the

adjoint pair above, and let : S

| | Id be the counit of the adjoint pair

(S

, | |). Consider the commuting diagram

Q

A
PL
_
< (V, d) >
_
APL(
K(V,d)
)

(V, d)

(V,d)

_
K

(V, d)
_
.
10 KATHRYN HESS
Since (V, d) is a Sullivan algebra and A
PL
(
K(V,d)
) is a surjective quasi-isomorphism,
the Lifting Lemma (Proposition 1.13) can be applied to this diagram, establishing
the existence of a CDGA morphism m
(V,d)
: (V, d) A
PL
_
< (V, d) >
_
,
unique up to homotopy, lifting
(V,d)
.
Theorem 1.24. If (V, d) is a simply connected Sullivan algebra of nite type,
then
(1) m
(V,d)
: (V, d) A
PL
_
< (V, d) >
_
is a quasi-isomorphism; and
(2) < (V, d) > is a simply connected, rational space of nite type, such that
there is an isomorphism of graded, rational vector spaces

_
< (V, d) >
_

= hom
Q
(V, Q).
We refer the reader to Section 17 of [18] for the details of the proof of this
extremely important theorem.
To complete the picture, we need to specify the relationship between spatial
realization and homotopy of morphisms.
Theorem 1.25. Let (V, d) and(W, d) be simply connected Sullivan algebras
of nite type.
(1) Let f : (V, d) (W, d) be a CDGA morphism. Then
(V, d)
m
(V,d)

(W, d)
m
(WV,d)

A
PL
_
< (V, d) >
_ APL
_
<f>
_

A
PL
_
< (W, d) >
_
commutes up to homotopy.
(2) Two CDGA morphisms f, g : (V, d) (W, d) are homotopic if and
only if < f > and < g > are homotopic.
(3) Let : X Y be a continuous map between simply connected CW-
complexes of nite type. If there is a homotopy-commutative diagram of
CDGAs
(V, d)

(W, d)

A
PL
(Y )
APL()

A
PL
(X),
then there is a homotopy-commutative diagram of topological spaces
X

< (W, d) >

<f>

< (V, d) >

in which

() Q and

() Q are isomorphisms.
Again, we refer the reader to Section 17 of [18] for the proof of this theorem.
Corollary 1.26. (1) Rational homotopy types of simply connected spaces
of nite rational type are in bijective correspondence with isomorphism
classes of minimal Sullivan algebras.
RATIONAL HOMOTOPY THEORY 11
(2) Homotopy classes of continuous maps between simply connected, nite-
type rational spaces are in bijective correspondence with homotopy classes
of morphisms between simply connected, nite-type Sullivan algebras.
We are now ready to introduce one of the very most important tools in rational
homotopy theory.
Definition 1.27. The Sullivan minimal model of a simply connected topologi-
cal space of nite rational type is the unique (up to isomorphism) Sullivan minimal
model of its algebra of piecewise-linear de Rham forms
: (V, d)

A
PL
(X).
As a consequence of Theorems 1.24 and 1.25, if : (V, d)

A
PL
(X) is a
Sullivan minimal model, then there is an isomorphism of graded, rational vector
spaces
hom
Q
(V, Q)

(X) Q.
In other words, given a Sullivan minimal model of a space, we can read o the
nontorsion part of its homotopy groups from the generators of the model.
2. Sullivan models
Since the CDGA A
PL
(X) is huge and has a complicated product, rational
homotopy theorists prefer to carry out computations with the Sullivan minimal
model, which is free as an algebra, with only nitely many generators in each
dimension if X is of nite rational type. In this section, we provide a brief overview
of the power of the Sullivan model. We begin by providing a few explicit examples
of Sullivan minimal models. We then explore the relationship between topological
brations and the Sullivan model. In particular, we explain the slogan the Sullivan
model of ber is the cober of the Sullivan model and illustrate its application.
A classical and essential numerical homotopy invariant, Lusternik-Schnirelmann
category, is our next subject: its elementary properties, how to calculate it using
the Sullivan model and its additivity. Finally, we present the beautiful and striking
rational dichotomy of nite CW-complexes, the proof of which depends crucially
on Lusternik-Schnirelmann category.
2.1. Examples and elementary construction. As a warmup and an aid to
developing the readers intuition, we calculate a few explicit examples of Sullivan
models. Here, a subscript on a generator always indicates its degree.
Spheres. The Sullivan model of an odd sphere S
2n+1
is
: ((x
2n+1
), 0) A
PL
(S
2n+1
),
where (x) is any representative of the unique cohomology generator of degree
2n + 1. Since is obviously a quasi-isomorphism of CDGAs, the nontorsion and
positive-degree part of

S
2n+1
is concentrated in degree 2n+1, where it is of rank
1.
On the other hand, the Sullivan model of an even sphere S
2n
is
: ((y
2n
, z
4n1
), d) A
PL
(S
2n
),
where dz = y
2
and (y) represents the unique cohomology generator of degree 2n.
Since the square of (y) must be a boundary, there is an acceptable choice of (z).
Again, is clearly a quasi-isomorphism of CDGAs, implying that the nontorsion
12 KATHRYN HESS
part of

S
2n
is concentrated in degrees 2n and 4n 1 and that it is of rank 1 in
each of those degrees.
Complex projective spaces. From the long exact sequences in homotopy of the
brations
S
1
S
2n+1
CP
n
and
S
1
S

CP

,
and the computation above of

S
2n+1
Q, we conclude that

CP
n
Q = Q u
2
Q x
2n+1
and

CP

Q = Q u
2
.
Consequently, the Sullivan model of CP
n
is of the form
:
_
(u
2
, x
2n+1
), d
_
A
PL
(CP
n
),
where dx = u
n+1
, (u) represents the algebra generator of H

(CP
n
; Q), which is a
truncated polynomial algebra on a generator of degree 2, and (x) kills its (n+1)
st
power. The value of dx is nonzero since H

(CP
n
; Q) is zero in odd degrees.
The Sullivan model for CP

is even easier to specify since there can be no

nontrivial dierential. It is
: ((u), 0) A
PL
(CP

),
where (u) represents the algebra generator of H

(CP

; Q), which is a polynomial

algebra on a generator of degree 2
Products. Let (B, ) and (C, ) be monoidal categories. Recall that a functor
F : (B, ) (C, ) is lax monoidal if for all B, B

Ob B, there is a natural
morphism F(B) F(B

) F(B B

) that is appropriately compatible with the

associativity and unit isomorphisms in (B, ) and (C, ).
It is easy to see that A
PL
is a lax monoidal functor, via the natural quasi-
isomorphism
X,Y
, dened to be the composite
A
PL
(X) A
PL
(Y )
APL(p1)APL(p2)
A
PL
(X Y ) A
PL
(X Y )

A
PL
(X Y ),
where p
i
is projection onto the i
th
component, and is the product on A
PL
(XY ).
Given Sullivan models : (V, d) A
PL
(X) and

: (V

, d

) A
PL
(X

),
the Sullivan model of the product space X X

is given by
(V, d) (V

, d

A
PL
(X) A
PL
(X

X,X

A
PL
(X X

).
Formal spaces. A space is formal if its rational homotopy is a formal conse-
quence of its rational cohomology, in the sense of the following denition.
Definition 2.1. A rational CDGA A is formal if there is a zigzag of quasi-
isomorphisms of CDGAs
A

(A).
A space X is formal if A
PL
(X) is a formal CDGA.
From the previous examples, we see that spheres and complex projective spaces
are formal. Furthermore, products of formal spaces are clearly formal. We can also
show that wedges of formal spaces are formal, as follows.
RATIONAL HOMOTOPY THEORY 13
Observe that for any family of pointed spaces {X
j
| j J},
A
PL
(

jJ
X
j
) = A

_
S

jJ
X
j
)
_
= sSet
_
S

jJ
X
j
), A

_
= sSet
_

jJ
S

(X
j
), A

_
=

jJ
sSet
_
S

(X
j
), A

_
=

jJ
A
PL
(X
j
).
Since a product of formal CDGAs is clearly formal, we obtain that a wedge of
formal spaces is formal, too.
Further examples of formal spaces can be found in geometry. Given a com-
pact, connected Lie group G, let K denote the connected component of its neutral
element e, in the subgroup of elements xed by a given involution. The quotient
G/K, which is a symmetric space, is then a formal space, as proved in [11]. Further-
more, Deligne, Griths, Morgan and Sullivan showed in [12] that compact Kahler
manifolds are also formal.
It is easy to construct an example of a nonformal CDGA. Let A = ((u, v, w), d),
where |u| = |v| = 3 and |w| = 5 and where dw = uv. Then
H
n
(A) =
_

_
Q : n = 0, 11
QQ : n = 3, 8
0 : else,
where the classes in degree 8 are represented by uw and vw and the class in degree
11 by uvw. If : A H

(A) is a CDGA map, then (w) = 0 for degree reasons,

which implies that (uw) = 0 = (vw), since is an algebra ma Consequently,
cannot be a quasi-isomorphism.
2.2. Models of ber squares. The Sullivan model is especially well adapted
to studying brations. In particular, as expressed more precisely in the next theo-
rem, the Sullivan model of a ber is the cober of the model.
Theorem 2.2. Let p : E B be a Serre bration such that B is simply
connected and E is path connected. Let F denote the ber of p. Suppose that B or
F is of nite rational type.
(1) Given a Sullivan model : (V, d) A
PL
(B), let
(V, d)
APL(p)

N
N
N
N
N
N
N
N
N
N
N
A
PL
(E)
(V W, D)

o
o
o
o
o
o
o
o
o
o
o
be a factorization of A
PL
(p) as a relative Sullivan algebra, followed
by a quasi-isomorphism. Let (W, D) = Q
(V,d)
(V W, D), and
let

: (W, D)

A
PL
(F) denote the induced ma Then

is a quasi-
isomorphism, i.e., there is a commuting diagram in CDGA
Q
(V, d)

(V W, D)

(W, D)

A
PL
(B)
APL(p)

A
PL
(E)
APL(j)

A
PL
(F)
14 KATHRYN HESS
where is the quotient map and j is the inclusion ma
(2) Given a Sullivan model : (V, d)

A
PL
(B) and a Sullivan minimal
model

: (W, d)

A
PL
(F), there is a relative Sullivan algebra
: (V, d) (V W, D)
such that (W, d)

= Q
(V,d)
(V W, D) and a quasi-isomorphism of
cochain algebras

: (V W, D)

A
PL
(E)
such that the diagram in (1) commutes, i.e., E has a Sullivan model that
is a twisted extension of a Sullivan model of the base by a Sullivan model
of the ber.
We refer the reader to Proposition 15.5 in [18] for the proof of the theorem
above. We remark only that the minimality of (W, d) in (2) is absolutely essential.
Example 2.3. Let S
n

PS
n
p

S
n
be the based path-space bration,
where n is odd. Let : (u, 0)

A
PL
(S
n
) be the Sullivan model of S
n
, and
consider the relative Sullivan algebra (u, 0)

((u, v), d), where |v| = n 1
and dv = u. The cochain algebra ((u, v), d) is clearly acyclic, as is A
PL
(PS
n
),
which implies that A
PL
(p) extends over ((u, v), d) to a quasi-isomorphism of
cochain algebras

: ((u, v), d)

A
PL
(PS
n
). By Theorem 2.2, the induced
cochain algebra map

: (v, 0)

A
PL
(S
n
) is a quasi-isomorphism, which
implies that H

(S
n
; Q)

= Q[v], when n is odd.

More generally, consider the based path-space bration X

PX
p

X,
where X is a simply connected space. Suppose that : (V, d)

A
PL
(X) is a
Sullivan model of X.
Using notation that is standard in rational homotopy theory, let V be the
graded Q-vector space that is the suspension of V , i.e., V
n
= V
n+1
. Let S be the
derivation of (V V ) specied by S(v) = v and S( v) = 0 for all v V . Dene
_
(V V ), D
_
by D v = S(dv), which implies that D v V
+
V , i.e., each
summand of D v contains at least one factor from V . The CDGA
_
(V V ), D
_
is
easily seen to be acyclic.
We can now dene a morphism of CDGAs

:
_
(V V ), D
_
A
PL
(PX)
by setting

(v) = (v) for all v V and

( v) = 0 for all v V and extending

multiplicatively. Since both the source and the target of are acyclic,

is a quasi-
isomorphism. We have therefore constructed a commutative diagram of CDGAs
(V, d)

_
(V V ), D
_

A
PL
(X)
p

A
PL
(PX)
to which we can apply Theorem 2.2. The induced morphism of CDGAs

: (V , 0)

A
PL
(X)
is therefore a quasi-isomorphism, so that H

(X; Q)

= V .
RATIONAL HOMOTOPY THEORY 15
Theorem 2.2 is a consequence of the following, more general result concerning
ber squares, for which the slogan is the model of the pullback is the pushout of
the models.
Theorem 2.4. Let p : E B be a Serre bration, where E is path connected
and B is simply connected, with ber F. Let f : X B be a continuous map, where
X is simply connected. Suppose that B or F is of nite rational type. Consider the
pullback
E
B
X

f

E
p

X
f

B.
Given a commuting diagram of CDGAs
(U, d)

(V, d)

(V W, D)

A
PL
(X) A
PL
(B)
APL(f)

APL(p)

A
PL
(E),
consider the pushout diagram in CDGA
Q
(V, d)

(U, d)

(V W, D)

(U W,

D).
Then the induced map of cochain algebras
(U W,

D)
A
PL
(E
B
X)
is a quasi-isomorphism of CDGAs.
We again refer the reader to [18] for the proof of this theorem, in the guise of
their Proposition 15.8.
Example 2.5. Let X be a simply connected space of nite rational type, with
Sullivan model : (V, d)

A
PL
(X). Consider the free-loop pullback square
LX
e

X
I
p

X X,
where is the diagonal map and where p() =
_
(0), (1)
_
for all paths : I X.
The free loop space LX is thus { X
I
| (0) = (1)}, which is homeomorphic to
X
S
1
, the space of unbased loops on X.
16 KATHRYN HESS
It is easy to check that
(V, d) (V, d)
b

(V, d)

A
PL
(X X)
APL()

A
PL
(X)
commutes, where m is the multiplication map on (V, d) and is the composite
(V, d) (V, d)

A
PL
(X) A
PL
(X)
X,X
A
PL
(X X).
Furthermore
(V, d) (V, d)
b

((V

V ), D)

A
PL
(X X)
APL(p)

A
PL
(X
I
)
commutes as well, where is a relative Sullivan algebra (V

and V

are two copies

of V ), is an appropriate extension of and D is specied as follows.
Let S : (V

V ) (V

V ) be the derivation of degree 1

specied by S(v

) = v = S(v

) and S( v) = 0. Then
D( v) := v

n1
(SD)
n
n!
(v

).
Applying Theorem 2.4 we obtain as a Sullivan model of LX
_
(V V ), D)

= (V, d)
(V,d)
2
((V

V ), D),
i.e., D( v) = S(dv), where S(v) = v and S( v) = 0.
Sullivan and Vigue used this model to prove that if H

(X; Q) requires at least

two algebra generators, then the rational Betti numbers of H

(LX; Q) grow expo-

nentially, which implies in turn that X admits an innite number of distinct closed
geodesics, when X is a closed Riemannian manifold [33].
It is well known (cf., e.g., [28]) that H

(LX; Z)

= HH

_
C

(X; Z)
_
, the Hoch-
schild homology of the cochains on X. The Sullivan model of the free loop space
constructed above is thus also a tool for understanding Hochschild homology.
2.3. Lusternik-Schnirelmann category. One of the most spectacular suc-
cesses of the Sullivan minimal model has been in its application to studying and
exploiting the numerical homotopy invariant known as Lusternik-Schnirelmann (L.-
S.) category.
Definition 2.6. A categorical covering of a space X is an open cover of X
such that each member of the cover is contractible in X. The L.-S. category of a
topological space X, denoted cat X, is equal to n if the cardinality of the smallest
categorical covering of X is n + 1.
L.-S. category is in general extremely dicult to compute. It is trivial, however,
to prove that the L.-S. category of a contractible space is 0 and that cat S
n
= 1 for
all n. Similarly, the L.-S. category of any suspension is 1. More generally, a space
X is a co-H-space if and only if cat X 1.
RATIONAL HOMOTOPY THEORY 17
The proof of this last equivalence is most easily formulated in terms of an
equivalent denition of L.-S. category, which requires the following construction,
due to Ganea.
Definition 2.7. Let p : E X be a bration over a based topological space
(X, x
0
). Let j : F E denote the inclusion of the ber of p over x
0
, with mapping
cone C
j
. Let p : C
j
X denote the induced continuous map, which can be
factored naturally as a homotopy equivalence followed by a bration:
C
j

X.
There is then a commutative diagram
E

p

@
@
@
@
@
@
@
@
E

.}
}
}
}
}
}
}
}
X
called the ber-cober construction on p.
Let p : PX X denote the (based) path bration over X. Iterating the
ber-cober construction repeatedly leads to a commutative diagram
P
0
X = PX
p

P
1
X
p1
.r
r
r
r
r
r
r
r
r
r
r

P
2
X
p2
.j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j

P
n
X
pn
.eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee


X,
in which P
n
X is the n
th
Ganea space for X and p
n
is the n
th
Ganea bration.
Proposition 2.8. If X is a normal space, then cat X n if and only if the
bration p
n
admits a section.
For a proof of this proposition, we refer the reader to Proposition 27.8 in [18].
Another equivalent denition of the L.-S. category of a based space (X, x
0
) is
expressed in terms of the fat wedge on X
T
n
X := {(x
1
, ..., x
n
) X
n
| i such that x
i
= x
0
}.
Proposition 2.9. If X is a path-connected CW-complex, then the following
conditions are equivalent.
(1) cat X n.
(2) The iterated diagonal
(n)
: X

X
n+1
factors up to homotopy through
the fat wedge T
(n+1)
X, i.e., there is a map : X

T
n+1
X such that
the diagram
X

G
G
G
G
G
G
G
G
G

(n)

X
n+1
T
n+1
X
i

t
t
t
t
t
t
t
t
t
commutes up to homotopy.
We refer the reader to Proposition 27.4 in [18] for the proof of this equivalence.
Applying Proposition 2.9, we obtain the following useful upper bound on L.-S.
category (cf., Proposition 27.5 in [18]).
18 KATHRYN HESS
Corollary 2.10. If X is an (r 1)-connected CW-complex of dimension d,
where r 1, then cat X d/r.
On the other hand, a lower bound on cat X is given by the cuplength c(X) of
H

(X; Q), i.e, the greatest integer n such that there exist a
1
, ..., a
n
H

(X; Q)
satisfying a
1
a
n
= 0. We leave it as an easy exercise to prove that c(X) cat X
for all path-connected, normal spaces X.
Example 2.11. Observe that c(CP
n
) = n, so that cat CP
n
n. On the other
hand, CP
n
is 1-connected and of dimension 2n, implying that cat CP
n
2n/2 = n.
Thus, cat CP
n
= n.
Within the realm of rational homotopy theory, it makes sense to consider the
following invariant derived from L.-S. category.
Definition 2.12. The rational category of a simply connected space X, de-
noted cat
0
X, is dened by
cat
0
X := min{cat Y | X and Y have the same rational homotopy type}.
As proved in [18] (Proposition 28.1), if X is a simply connected CW-complex,
then cat
0
X = cat X
0
. Furthermore, it is obvious that cat
0
X cat X for all X.
As we show below, this inequality is sharp, i.e., there are spaces X for which
cat
0
X = cat X. On the other hand, the inequality can certainly be strict, as the
case of a mod p Moore space easily illustrates.
The next theorem has turned out to be a crucial tool in proving numerous
signicant results in rational homotopy theory, such as many of the dichotomy
theorems (cf., Section 2.4).
Theorem 2.13 (The Mapping Theorem). Let f : X Y be a continuous map
between simply connected spaces. If

f Q is injective, then cat

0
X cat
0
Y .
The original proof of the Mapping Theorem relied on Sullivan models. There is
now a purely topological and relatively simple proof, which is given in [18] (Theorem
28.6).
As a rst application of the Mapping Theorem, we mention the amusing and
useful corollary below, which follows immediately from the fact that the natural
map from the (n+1)
st
Postnikov ber to the n
th
Postnikov ber of a space induces
an injection on homotopy groups.
Corollary 2.14. Let X be a connected CW-complex. Let X(n) denote the
n
th
Postnikov ber of X, for all n 1. Then
cat
0
X(n + 1) cat
0
X(n) cat
0
X(2) cat
0
X.
One great advantage of rational category, as opposed to the usual L.-S. category,
is that it can explicitly calculated in terms of the Sullivan model, as stated in the
next theorem.
Theorem 2.15. Let : (V, d)

A
PL
(X) be the Sullivan minimal model
of a simply connected space X of nite rational type. Let (V/
>n
V,

d) denote the
CDGA obtained by taking the quotient of (V, d) by the ideal of words of length
RATIONAL HOMOTOPY THEORY 19
greater than n, and let
(2.1) (V, d)
q

N
N
N
N
N
N
N
N
N
N
N
(V/
>n
V,

d)
((V W), d)
p

m
m
m
m
m
m
m
m
m
m
m
m
be a factorization of the quotient map q as a relative Sullivan algebra, followed by
a surjective quasi-isomorphism. Then cat
0
X n if and only if i admits a CDGA
retraction : ((V W), d)

(V, d), i.e., i = Id
(V,d)
.
The fat wedge formulation of the denition of L.-S. category is crucial in the
proof of this theorem, for which we refer the reader to Propositions 29.3 and 29.4
in [18].
Example 2.16. Since H
n
(S
n
; Q) = Q, the rationalization S
n
0
of the n-sphere
is not contractible and therefore cat
0
S
n
= cat S
n
0
> 0. On the other hand,
cat
0
S
n
cat S
n
= 1, whence cat
0
S
n
= 1, providing the promised example of
equality between rational category and L.-S. category of a space.
This calculation can also be carried out easily using the Sullivan model (cf.,
Section 2.1). If n is odd, then the Sullivan model is ((x), 0), where x is of degree
n. Observe that (x)/
>1
(x) is isomorphic to (x), since x is of odd degree. Since
the quotient map q is itself the identity map in this case, it follows trivially that
cat
0
S
n
1.
If n is even, the relevant Sullivan model is ((y, z), d), where deg y = n, deg z =
2n 1 and dz = y
2
. An easy calculation shows that
((y, z)/
>1
(y, z),

d) = (QQ y Q z, 0).
It is not too dicult to show that the quotient map q factors as
((y, z), d)
q

Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
Q
(QQ y Q z, , 0)
((y, z) W, D)
p

j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
where DW (y, z)
+
W, i.e., the dierential of any generator of W, if nonzero,
is a sum of words, all of which contain at least one letter from W. We can therefore
dene a CDGA retraction by setting (w) = 0 for all w W, implying that
cat
0
S
n
1.
Though Theorem 2.15 does simplify the calculation of rational category by
making it purely algebraic, the computations involved are still dicult, which led
Halperin and Lemaire to propose the following, apparently weaker numerical in-
variant of rational homotopy [23].
Definition 2.17. Let X be a simply connected space of nite rational type,
with Sullivan model : (V, d)

A
PL
(X). If the map i in diagram (2.1) of
Theorem 2.15 admits a retraction as morphisms in the category of (V, d)-modules,
then Mcat
0
X n.
As it turned out, however, the apparent weakness of Mcat
0
was only an illusion.
20 KATHRYN HESS
Theorem 2.18. Mcat
0
X = cat
0
X for all simply connected spaces X of nite
rational type.
For the proof of this theorem, which requires a deep understanding of the
factorization of the quotient map, we refer the reader to Theorem 29.9 in [18].
Theorem 2.18 implies that to show that cat
0
X n, it suces to nd a (V, d)-
module retraction of i in diagram (2.1), which has proven to be a very eective
simplication. We next outline briey one application of this simplication, to the
study of the additivity of L.-S. category.
It is not dicult to show that cat(XY ) cat X+cat Y for all normal spaces
X and Y . At the end of the 1960s Ganea observed that in the only known examples
for which cat(X Y ) = cat X + cat Y , the spaces X and Y had homology torsion
at distinct primes. He conjectured therefore that cat(X S
n
) = cat X + 1 for all
spaces X and all n 1, since S
n
has no homology torsion whatsoever.
In fact, as stated precisely below, if we forget torsion completely and work
rationally, then L.-S. category is indeed additive.
Theorem 2.19. If X and Y are simply connected topological spaces of nite
rational type, then cat
0
(X Y ) = cat
0
X + cat
0
Y.
The proof of this theorem depends in an essential way on Theorem 2.18. We
refer the reader to Sections 29(h) and 30(a) in [18] for further details.
As an epilogue to this story of Sullivan minimal models and L.-S. category, we
mention that in 1997 Iwase applied classical homotopy-theoretic methods to the
construction of a counter-example to Ganeas conjecture [26]. In particular, he
built a 2-cell complex X such that cat(X S
n
) = 2 = cat X.
2.4. Dichotomy. There is a beautiful dichotomy governing nite CW-com-
plexes in rational homotopy theory, expressed as follows: the rational homotopy
groups of a nite CW-complex either are of nite total dimension as graded rational
vector space or grow exponentially. We rst examine the former case, that of elliptic
spaces, then the latter case, that of hyperbolic spaces.
Definition 2.20. A simply connected topological space X is rationally elliptic
if
dimH

(X; Q) < and dim

(X) Q < .
The formal dimension fdimX of a rationally elliptic space X is dened by
fdimX := max{k | H
k
(V, d) = 0}.
The even exponents of a rationally elliptic space X are the postive integers a
1
, ..., a
q
such that there is a basis (y
j
)
1jq
of
even
X Q with deg y
j
= 2a
j
. Similarly,
the odd exponents of X are the positive integers b
1
, ..., b
p
such that there is a basis
(x
i
)
1ip
of
odd
X Q with deg x
i
= 2b
i
1.
A minimal Sullivan algebra (V, d) is elliptic if the associated rational space
< (V, d) > is elliptic. The formal dimension of an elliptic minimal Sullivan algebra
is the formal dimension of the associated space.
Example 2.21. Spheres, complex projective spaces, products of elliptic spaces,
and homogeneous spaces are examples of elliptic spaces.
The following special case of elliptic spaces is important for understanding
general elliptic spaces.
RATIONAL HOMOTOPY THEORY 21
Definition 2.22. A minimal Sullivan algebra (V, d) is pure if dimV < ,
d|
V
even = 0 and d(V
odd
) V
even
. A space is pure if its Sullivan model is pure.
Note that a pure space X is elliptic if and only if dimH

(X; Q) < .
A pure Sullivan algebra (V, d) admits a dierential ltration
F
k
(V, d) = V
even

k
V
odd
.
In particular, dF
k
(V, d) F
k1
(V, d). Write
H
k
(V, d) =
ker
_
d : F
k
(V, d) F
k1
(V, d)
_
Im
_
d : F
k+1
(V, d) F
k
(V, d)
_ .
The following list of the most important properties of pure Sullivan algebras
summarizes Propositions 32.1 and 32.2 in [18].
Proposition 2.23. Let (V, d) be a pure, minimal Sullivan algebra..
(1) dimH

(V, d) < dimH

0
(V, d) < .
(2) If dimH

(V, d) < , then H

n
(V, d) is a 1-dimensional subspace of
H
r
(V, d), where n is the formal dimension of (V, d) and r = max{k |
H
k
(V, d) = 0}.
(3) If dimH

(V, d) < , then r = dimV

odd
dimV
even
. Thus,
dimV
odd
= dimV
even
H

(V, d) = H
0
(V, d).
(4) If dimH

(V, d) < and (V, d) is a Sullivan minimal model of X, then

fdimX =

i
(2b
i
1)

j
(2a
j
1),
where the even and odd exponents of X are a
1
, .., a
q
and b
1
, ..., b
p
, respec-
tively.
We present next a tool for determining whether spaces are elliptic, based on
the notion of pure spaces.
Definition 2.24. Let (V, d) be a minimal Sullivan algebra such that dimV <
. Filter (V, d) by
F
p
(V, d) =

k+lp
(V
even

k
V
odd
)
l
.
The induced spectral sequence is called the odd spectral sequence and converges to
H

(V, d).
Observe that the E
0
-term of the odd spectral sequence is the associated graded
of (V, d

), where d

(V
even
) = 0, d

(V
odd
) V
even
and (d d

)(V
odd
)
V
even

+
V
odd
. We call (V, d

) the associated pure Sullivan algebra of (V, d).

Proposition 2.25. Under the hypotheses of the denition above,
dimH

(V, d) < dimH

(V, d

) < .
Thus, (V, d) is elliptic if and only if (V, d

) is elliptic.
22 KATHRYN HESS
Proof. As the odd spectral sequence converges from H

(V, d

) to H

(V, d),
one implication is clear. An algebraic version of the Mapping Theorem (Theorem
2.13) plays an essential role in the rest of the proof. We refer the reader to Propo-
sition 32.4 in [18] for the complete proof.
Example 2.26. Consider (V, d) = ((a
2
, x
3
, u
3
, b
4
, v
5
, w
7
), d), where the sub-
script of a generator equals its degree and d is specied by da = 0, dx = 0, du = a
2
,
db = ax, dv = ab ux and dw = b
2
vx. Its associated pure Sullivan algebra is
((a, x, u, b, v, w), d

), where d

a = 0, d

x = 0, d

u = a
2
, d

b = 0, d

v = ab and
d

w = b
2
. A straightforward calculation shows that
H

(V, d

) = Q a Q b y/(y
3
) z/(z
3
),
where y is represented by bu av and z is represented by aw bv. In particular
dimH

(V, d

) < , which implies that (V, d) is elliptic.

The next theorem describes the amazing numerology of elliptic spaces and
Sullivan algebras, which imposes formidable constraints on their form.
Theorem 2.27. Let (V, d) be an elliptic, minimal Sullivan algebra of formal
dimension n and with even and odd exponents a
1
, ..., a
q
; b
1
, ..., b
p
. Then:
(1)

p
i=1
(2b
i
1)

q
j=1
(2a
j
1) = n;
(2)

q
j=1
2a
j
n;
(3)

p
i=1
(2b
1
1) 2n 1; and
(4) dimV
even
dimV
odd
.
As a consequence of this theorem, we know, for example, that if (V, d) is
an elliptic, minimal Sullivan algebra of formal dimension n, then V = V
2n1
,
dimV
>n
1 and dimV n.
Proof. One rst proves by induction on dimV that the formal dimensions of
(V, d) and of its associated pure Sullivan algebra are the same, reducing the proof
of the theorem to the pure case. For further details, we refer the reader to Theorem
32.6 in [18].
Definition 2.28. The Euler-Poincare characteristic of a graded vector space
W of nite type is the integer

W
=

i
(1)
i
dimW
i
= dimW
even
dimW
odd
.
It is easy to show that
W
=
H

(W,d)
, for any choice of dierential d on W.
Proposition 2.29. If (V, d) is an elliptic, minimal Sullivan algebra, then

V
0 and
V
0. Furthermore, the following statements are equivalent.
(1)
V
> 0.
(2) H

(V, d) = H
even
(V, d).
(3) H

(V, d) = (y
1
, ..., y
q
)/(u
1
, ..., u
p
), where (u
1
, ..., u
p
) is a regular se-
quence.
(4) (V, d) is isomorphic to a pure, minimal Sullivan algebra.
(5)
V
= 0.
RATIONAL HOMOTOPY THEORY 23
Proof. The proof of this proposition reduces essentially to Poincare series
calculations. We refer the reader to Proposition 32.10 in [18] for details of the
calculations.
Definition 2.30. Let X be an elliptic space with minimal Sullivan model
(V, d). The homotopy Euler characteristic

(X) of X is dened to be
V
.
Propostion 2.29 implies that

(X) 0 for all elliptic spaces X and that

(X) = 0 if and only if X is a pure elliptic space.

Example 2.31 (Application to free torus actions (Example 3 in section 32(e)
of [18])). Let T denote the r-torus, i.e., the product of r copies of S
1
. Suppose that
T acts smoothly and freely on a simply connected, compact, smooth manifold M.
There exists then a smooth principal bundle M M/T and thus a classifying
map M/T BT with homotopy ber M.
If M is elliptic, then M/T is also elliptic, since M/T is compact and BT =
(CP

)
r
. Furthermore,
0

(M/T) =

(M) +

(BT) =

(M) +r,
implying that r

(M).
Now we go to the other extreme.
Definition 2.32. A simply connected space X with the homotopy type of a
nite CW-complex is rationally hyperbolic if dim

(X) Q = .
The following theorem, which justies the terminology hyperbolic, is Theo-
rem 33.2 in [18]. Its proof depends strongly on the Mapping Theorem (Theorem
2.13).
Theorem 2.33. If X is a rationally hyperbolic space, then there exist C > 1
and N Z such that
n

i=0
dim
i
(X) Q C
n
for all n N.
In other words, the rational homotopy groups of X grow exponentially. More-
over, as stated more precisely in the next theorem (Theorem 33.3 in [18]), there
are no long gaps in the rational homotopy groups of X.
Theorem 2.34. If X is a rationally hyperbolic space of formal dimension n,
then for all k 1, there exists i (k, k +n) such that
i
X Q = 0. Furthermore,
for k 0,
k+n1

i=k+1
dim
i
(X) Q
dim
k
X Q
dimH

(X; Q)
.
Consequently, if X has formal dimension n, then
X is rationally elliptic
j
(X) Q = 0 j [2n, 3n 2],
a simple and lovely test of ellipticity.
24 KATHRYN HESS
3. Commutative algebra and rational homotopy theory
In the late 1970s two algebraists, Luchezar Avramov, then at the University
of Soa, and Jan-Erik Roos of the University of Stockholm, discovered and began
to exploit a deep connection between local algebra and rational homotopy theory.
In 1981 they established contact with the rational homotopy theorists, initiating
a powerful synergy that led to a multitude of important results in both elds. In
this section, of a more expository nature than the preceding sections, we describe
certain of the most important results of this collaboration. For further details we
refer the reader to Section 4 of [24].
Roos interest in rational homotopy theory was inspired by work of Jean-Michel
Lemaire on Serres question concerning rationality of Poincare series of the rational
homology of loop spaces and by the work of local algebraists on the analogous
question of Kaplansky and Serre for local rings. More precisely, Lemaire had studied
the Poincare series

n0
dim
Q
H
n
(E; Q) z
n
for E a nite, simply connected CW-complex, while local algebraists were interested
in the series

n0
dim
F
Ext
n
R
(F, F) z
n
for R a local ring with residue eld F. (Here and throughout this section, all local
rings are assumed to be commutative and noetherian.) In both cases, the goal was
to determine whether the series always represented a rational function.
Roos established a research programto study the homological properties of local
rings, in particular those whose maximal ideal m satised m
3
= 0, the rst nontrivial
case for Poincare series calculations. He realized that in order to study local rings,
it was useful, or even necessary, to work in the larger category of (co)chain algebras.
By 1976 he had proved the equivalence of Serres problem for CW-complexes E such
that dimE = 4 and of the Kaplansky-Serre problem for local rings (R, m) such that
m
3
= 0 [31].
Avramov ascribes his original interest in rational homotopy theory to Levins
result from 1965 that if R is a local ring with residue eld F, then Tor
R
(F, F) is
a graded, divided powers Hopf algebra [27]. Since the dual of a graded, divided
powers Hopf algebra is the universal enveloping algebra of a uniquely dened graded
Lie algebra (char F = 0 [30], char F > 2 [1], char F = 2 [32]), to any local ring R is
associated a graded Lie algebra

(R), the homotopy Lie algebra of R. Based on

results in characteristic 0 due to Gulliksen in the late 1960s, Avramov proved that
if R S is a homomorphism of local rings such that S is R-at, then there is an
exact sequence of groups
...
n
R
n
(S R F)
n
(S)
n
(R)

n

n+1
(S R F)
n+1
S ....
The existence of such a long exact sequence of homotopy groups conrmed Av-
ramovs intuition that rational homotopy invariants were analogous to homology
invariants of local rings in arbitrary characteristic.
In 1980 David Anick constructed a nite, simply connected CW-complex E of
dimension 4 such that the Poincare series of the homology of E was not rational
[2], [3]. Anicks construction interested rational homotopy theorists because of its
relation to the dichotomy between elliptic and hyperbolic spaces; see Section 2.4.
RATIONAL HOMOTOPY THEORY 25
Local algebraists were interested because of Rooss result, which allowed the tran-
scription of Anicks space into a local ring (R, m) with m
3
= 0 and with irrational
Poincare series. Shortly after Anicks result became known, Roos and his student
Clas L ofwall discovered other examples of local rings with irrational Poincare series
that they obtained by completely dierent methods [29].
The converging interests of rational homotopy theorists and of local algebraists
led to direct contact between the two groups in 1981. Inspired by the work of Roos
and his colleagues, Felix and Thomas began to work on calculating the radius of
convergence of the Poincare series of a loop space [19], establishing the following
beautiful characterization: a simply connected space E of nite L. -S. category is
rationally elliptic if and only if the radius of convergence of the Poincare series of
E is 1. If E is rationally hyperbolic, then the radius of convergence is strictly less
than 1. Moreover, they found a relatively easily computable upper bound for the
radius of convergence in the case of a hyperbolic, formal space. They also showed
that if A is a noetherian, connected graded commutative algebra over a eld F of
characteristic zero and
A
denotes the radius of convergence of
P
A
(z) =

no
dim
F
Tor
A
n
(F, F) z
n
,
then either
A
= + and A is a polynomial algebra; or
A
= 1, A is a complete
intersection, and the coecients of P
A
(z) grow polynomially; or
A
< 1, A is not a
complete intersection, and the coecients of P
A
(z) grow exponentially. Avramov
later generalized this result to any characteristic [6].
The written version of Avramovs Luminy talk on the close links between local
algebra and rational homotopy theory provides an excellent and thorough introduc-
tion to the subject [7]. His article contains the rst dictionary between rational
homotopy theory and local algebra, explaining how to translate notions and tech-
niques from one eld to the other. Given a theorem in one eld, applying the
dictionary leads to a statement in the other eld that stands a reasonable chance
of being true, though the method of proof may be completely dierent.
Avramov and Halperin wrote another thorough introduction to the subject in
the proceedings of the Stockholm conference of 1983 [8]. It begins at a more ele-
mentary level than the survey article of Avramov in the proceedings of the Luminy
conference, leading the reader from rst principles of dierential graded homolog-
ical algebra to notions of homotopy ber and loop space and on to the homotopy
Lie algebra.
In his introductory article [7], Avramov emphasized the importance of minimal
models in local ring theory. If K
R
is the Koszul complex of a local ring, then there
is a minimal, commutative cochain algebra (V, d) over the residue eld of R that
is quasi-isomorphic to K
R
. Avramov called (V, d) the minimal model of R. He
established its relevance by observing that in degrees greater than 1, the graded
Lie algebra derived from (V, d
2
) is isomorphic to the homotopy Lie algebra of R.
In [17] Felix, Halperin, and Thomas continued the in-depth study of the ho-
motopy Lie algebra of a rationally hyperbolic space begun by Felix and Halperin in
[14]. They showed, for example, that if E is rationally hyperbolic, then its ratio-
nal homotopy Lie algebra is not solvable. Moreover they proposed as conjectures
translations of their theorems into local algebra, where, for a local ring (R, m) with
residue eld F, L.-S. category is replaced by dim
F
(m/m
2
) depth R and innite
dimensional rational homotopy is replaced by R not being a complete intersection.
26 KATHRYN HESS
Recall that
depth R = inf{j | Ext
j
R
(F, R) = 0}.
In [17] Felix, Halperin and Thomas also mentioned a very important conjecture
due to Avramov and Felix, stating that the homotopy Lie algebra of a rationally
hyperbolic space should contain a free Lie algebra on at least two generators. This
conjecture has motivated much interesting work in the study of the homotopy Lie
algebra and has not as yet (2006) been proved.
Using minimal model techniques, Halperin and Bgvad proved two conjectures
due to Roos, which are translations of each other [9]. They showed that
(1) if R is a local ring such that the Yoneda algebra Ext

R
(F, F) is noetherian,
then R is a complete intersection; and
(2) if E is a simply connected, nite CW-complex such that the Pontryagin
algebra H

(E; Q) is noetherian, then E is rationally elliptic.

Their proof is based on a slightly weakened form of the Mapping Theorem that
holds over a eld of any characteristic, as well as on ideas from the article of Felix,
Halperin and Thomas of the previous year [17].
In the spring of 1985 Halperin applied minimal model techniques to answering
on old question concerning the deviations of a local ring [22]. The jth deviation,
e
j
(R), of a noetherian, local, commutative ring R with residue eld F is dim
F

j
(R).
Assmus had shown in 1959 that R is a complete intersection if and only if e
j
(R) = 0
for all j > 2 [4], raising the question of whether any deviation could vanish if R
were not a complete intersection. Halperin succeeded in answering this question,
showing that if R is not a complete intersection, then e
j
= 0 for all j.
The Five Author paper [15] represents a great leap forward in understanding
the structure of the homotopy Lie algebra of a space or of a local ring. The principal
innovation of the Five Author paper consists in exploiting the radical of the homo-
topy Lie algebra, i.e., the sum of all of its solvable ideals, which rational homotopy
theorists had begun to study in 1983. The radical itself is in general not solvable.
Recall, as mentioned above, that the rational homotopy Lie algebra of a rationally
hyperbolic space is not solvable.
Expressed in topological terms, the main theorem of the Five Author paper
states that if E is a simply connected CW-complex of nite type and cat(E) =
m < , then the radical of the homotopy Lie algebra, Rad(E) is nite dimensional
and dimRad(E)
even
m. This is a consequence of two further theorems, both
of which are of great interest themselves. The rst concerns the relations among
the rational L.-S. category of a space and the depth and global dimension of its
homotopy Lie algebra. Recall that the gobal dimension of a local ring R with
residue eld F satises
gl. dim.(R) = sup{j | Ext
j
R
(F, F) = 0}.
The precise statement of this theorem in topological terms is then that if L is the
homotopy Lie algebra of a simply-connected CW-complex of nite type E, then
either
depth UL < cat
0
(E) < gl. dim. UL
or
depth UL = cat
0
(E) = gl. dim. UL.
RATIONAL HOMOTOPY THEORY 27
The second theorem states that under the same hypotheses, if depth UL < , then
Rad(E) is nite dimensional and satises dimRad(E)
even
depth UL. Moreover,
if dimRad(E)
even
= depth UL, then Rad(E) = L.
Rational homotopy theorists have exploited extensively the results of [15] in
developing a deep understanding of the homotopy Lie algebra of rationally hyper-
bolic spaces. The methods the ve authors devised to prove their results have
turned out to be extremely important as well. For example, since their goal was
to relate cat
0
(E) to depth(L), they needed to construct a model of the quotient
cochain algebra (V/
>n
V,

d), where (V, d) is the Sullivan minimal model of E.
Their method for doing so, based on perturbation of a model for (V/
>n
V, d
2
),
proved to be useful in a number of other circumstances, such as in the proof of
Theorem 2.18.
References
[1] M. Andre, Cohomologie des alg`ebres dierentielles o` u op`ere une alg`ebre de Lie, T ohoku Math.
J. 14 (1962), 263311.
[2] D. J. Anick, A counter-example to a conjecture of Serre, Annals of Math. 115 (1982), 133.
[3] D. J. Anick, Comment: A counter-example to a conjecture of Serre, Annals of Math. 116
(1982), 661.
[4] E. Assmus, On the homology of local rings, Ill. J. Math .3 (1959), 187199
[5] L. Avramov, Homology of local at extensions and complete intersection defects, Math. Ann.
228 (1977), 2737.
[6] L. Avramov, Local rings of nite simplicial dimension, Bull. Amer. Math. Soc. 10 (1984)
289291.
[7] L. Avramov, Local algebra and rational homotopy, Homotopie Algebrique et Alg`ebre Locale,
Asterisque, vol. 113114, 1984, pp. 1643.
[8] L. Avramov and S. Halperin, Through the looking glass: a dictionary between rational homo-
topy theory and local algebra, Algebra, Algebraic Topology and their Interactions, Springer
Lecture Notes in Mathematics, vol. 1183, 1986, pp. 127.
[9] R. Bgvad and S. Halperin, On a conjecture of Roos, Algebra, Algebraic Topology and their
Interactions, Springer Lecture Notes in Mathematics, vol. 1183, 1986, 120126
[10] A. K. Bouseld and V. K. A. M. Gugenheim, On PL De Rham Theory and Rational Homotopy
Type, Memoirs of the A. M. S., vol. 179, 1976.
[11] E. Cartan, Sur les nombres de Betti des espaces de groupes close, C.R. Acad. Sci. Paris 187
(1928), 196197.
[12] P. Deligne, P. Griths, J. Morgan and D. Sullivan, Real homotopy theory of K ahler manifolds
Inventiones 29 (1975), 245274.
[13] W. Dwyer and J. Spalinski, Homotopy theories and model categories, Handbook of Algebraic
Topology (I. M. James, ed.), North-Holland, 1995, pp. 73126.
[14] Y. Felix and S. Halperin, Rational L.-S. category and its applications, Trans. A. M. S. 273
(1982), 137
[15] Y. Felix, S. Halperin, C. Jacobsson, C. L ofwall, and J.-C. Thomas, The radical of the homo-
topy Lie algebra, Amer. J. Math. 110 (1988), 301322.
[16] Y. Felix, S. Halperin and J.-M. Lemaire, Rational category and conelength of Poincare com-
plexes, Topology 37 (1998), 743-748.
[17] Y. Felix, S. Halperin and J.-C. Thomas, The homotopy Lie algebra for nite complexes, Publ.
Math. I. H. E. S. 56 (1982), 387410.
[18] Y. Felix, S. Halperin and J.-C. Thomas, Rational Homotopy Theory, Graduate Texts in
Mathematics, vol. 205, Springer-Verlag, 2001.
[19] Y. Felix and J.-C. Thomas, The radius of convergence of Poincare series of loop spaces,
Invent. Math. 68 (1982), 257274.
[20] P. Goerss, Model categories and simplicial methods, Lecture notes from the summer school
on Interactions between Homotopy Theory and Algebra, Contemporary Math., 2006.
[21] V.K.A.M Gugenheim and H. J. Munkholm, On the extended functoriality of Tor and Cotor,
J. Pure Appl. Algebra 4 (1974) , 929.
28 KATHRYN HESS
[22] S. Halperin, The nonvanishing of deviations of a local ring, Comment. Math. Helv. 62 (1987),
646653.
[23] S. Halperin and J.-M. Lemaire, Notions of category in dierential algebra, Algebraic Topol-
ogy: Rational Homotopy, Springer Lecture Notes in Mathematics, vol. 1318, pp. 138153.
[24] K. Hess, A history of rational homotopy theory, History of Topology, Elsevier Science B.V.,
1999, pp. 757796.
[25] M. Hovey, Model Categories, Mathematical Surveys and Monographs, vol. 63, American
Mathematical Society, 1999.
[26] N. Iwase, Ganeas conjecture on Lusternik-Schnirelmann category, Bull. London Math. Soc.
30 (1998), 623634.
[27] G. Levin, Homology of local rings, Ph. D. Thesis, Univ. Chicago, 1965.
[28] J.-L. Loday, Cyclic Homology (Second Edition), Grundlehren der Mathematischen Wis-
senschaften, vol. 30, Springer-Verlag, 1998.
[29] C. L ofwall and J.-E. Roos, Cohomologie des alg`ebres de Lie graduees et series de Poincare-
Betti non rationnelles, C. R. Acad. Sci. Paris 290 (1980) , 733736
[30] J. Milnor and J. Moore, On the structure of Hopf algebras, Annals of Math. 81 (1965),
211264.
[31] J.-E. Roos, Relations between the Poincare-Betti series of loop spaces and local rings,
Seminaire dalg`ebre Paul Dubreil, Springer Lecture Notes in Mathematics, vol. 740, 1979,
pp. 285322
[32] G. Sj odin, Hopf algebras and derivations, J. Algebra 64 (1980), 218229.
[33] D. Sullivan and M. Vigue-Poirrier, The homology theory of the closed geodesic problem, J.
Dierential Geometry 11 (1976), 633644.
EPFL SB IGAT, B atiment BCH, CH-1015 Lausanne, Switzerland
E-mail address: kathryn.hess@epfl.ch