RIGOROUS RESULTS IN TOPOLOGICAL σ-MODEL

Maxim Kontsevich
Max-Planck-Institut für Mathematik, Bonn
and University of California, Berkeley

We give here a review of certain recent develoments in the mathematical formu-

lation of the two-dimensional topological σ-model introduced several years ago by
E. Witten (see [W1,W2]). Our review is very far from being complete and covers
only a small portion of the existing bibliography.
First of all, recall that the partition function of the two-dimensional σ-model is
given by the following Feynman integral:
Z  Z 
′ 2
exp − |φ | + . . . Dφ
φ:Σ−→V Σ

Here Σ is a closed surface of genus g endowed with a Riemannian metric and

V is a Riemannian manifold. Dots in the action denote fermionic and topological
terms.
The classical theory is conformally invariant, i.e. it depends only on the confor-
mal/complex class of metric on Σ. Physicists beleive that the quantum theory is
defined and it is conformally invariant for some special class of metrics on V close
to the class of Einstein metrics. Notice that the standard regularization scheme
of the Feynman integral by a discretization of the surface Σ fails because of the
conformal invariance of the action functional.
Suppose that we have a metric on V giving a conformal field theory. Then, as
usual, we have an infinite-dimensional pre-Hilbert space H of fields with a base
{Oα }. Correlation functions of primary fields in the conformal field theory

hOα1 (p1 ) . . . Oαn (pn )ig

can be considered as real-analytic functions (more precisely, sections of some line

bundles) on the space Mg,n of equivalence classes of surfaces Σ with conformal
structures and n pairwise distinct points p1 , . . . , pn on Σ. This space is a non-
compact orbifold of dimension 6g − 6 + 2n and it is called the moduli space of
smooth complex curves with punctures. Locally it looks like C3g−3+n /Γ where Γ is
a finite group acting linearly in a complex vector space. The space Mg,n is defined
for all g, n ≥ 0 except four cases: g = 0, n ≤ 2 or g = 1, n = 0.
If we include the supersymmetry into the action functional then we get cor-
relators hOα1 . . . Oαn ig for certain fields {Oαi } which are not functions on Mg,n
but differential forms on it. BRST-closed fields give closed forms, and the space
HBRST of BRST-cohomology classes give cohomology classes on Mg,n . Physical

Typeset by AMS-TEX
1
2 MAXIM KONTSEVICH

arguments show that these closed differential forms have smooth enough prolonga-
tion to the Deligne-Mumford compactification Mg,n of Mg,n . By definition, this
compactification (intoduced in [DM]) is the moduli space of curves with marked
points (C; p1 , . . . , pn ) such that
(1) C is a compact connected complex curve,
(2) singularities of C are only simple self-crossing points,
(3) the Euler characteristic of the smooth part of C is equal to 2 − 2g,
(4) marked points pi are pairwise distinct and non-singular,
(5) the automorphism group of (C; p1 , . . . , pn ) is finite.
Such curves are called stable marked curves.
The last condition in the definition of stable curves can be reformulated as
(1) there exists a complete hyperbolic metric on C smooth \ {p1 , . . . , pn }, or,
equivalently,
(2) genus zero components of C smooth \ {p1 , . . . , pn } have at least 3 punctures
and genus 1 components have at least 1 puncture.
In string theory one wants to compute the following string correlation function:
X Z
2g−2
hOα1 . . . Oαn i := λ hOα1 . . . Oαn ig ,
g≥0 Mg,n

where λ is the string coupling constant.

In the topological sector of super-symmetric conformal string theory we integrate
the top-degree component of closed differential forms over the fundamental class of
Mg,n . Mathematically the arising structure was formalized in [KM]. Here is the
list of axioms.
Axioms of topological σ-model.
DATA:
(1) H : a finite-dimensional Z2 -graded (=super) vector space over C,
(2) ( , ) : H ⊗ H − → C : a non-degenerate even scalar product on H,
(3) ( , . . . , )g : H⊗n −→ H∗ (Mg,n , C) : correlators for g, n ≥ 0 , 2 − 2g − n < 0.
AXIOMS:
(1) Sn -invariance,
(2) the splitting axiom and the genus reduction axiom.
The first axiom means that the correlators are invariant under the evident ac-
tion of the permutation group Sn acting simultaneously both on H⊗n and on
H∗ (Mg,n , C).
The splitting axiom is the following:
For each g1 , n1 , g2 , n2 such that 2 − 2gi − ni < 0 and ni > 0 there is a map

Mg1 ,n1 × Mg2 ,n2 −

→ Mg,n

where g = g1 + g2 and n = n1 + n2 − 2. Namely, we glue curves (C; p1 , . . . , pn1 )

and (C ′ ; p′1 , . . . , p′n2 ) into one singular curve
 
[
C C ′ ; p1 , . . . , pn1 −1 , p′2 , . . . , p′n2 
pn =p′1
 RIGOROUS RESULTS IN TOPOLOGICAL σ-MODEL 3

In this way we identify Mg1 ,n1 × Mg2 ,n2 with a sub-orbifold in Mg,n of complex
codimension one (divisor).
The spitting axiom says that

(φi1 . . . φin )g |M =
g1 ,n1 ×Mg2 ,n2

X
⊗ χj φin1 +1 . . . φin



= φi1 . . . φin1 χj g1 g2
j

where {χj } and {χj } are dual bases of H with respect to the scalar product ( , ) .
We demand also an analogous property for the restriction of cohomology classes
on Mg,n to Mg−1,n+2 via the map

(C; p1 , . . . , pn+2 ) 7→ (C/(p1 = p2 ); p3 , . . . , pn+2 )

glueing first two marked points into one singular point.

In fact, all this is only a first approximation to the complete story, because one
wants to consider so called gravitational descendants φi,d where φi ∈ H are fields
and d ≥ 0 is an integer. Naively one can define correlators between gravitational
descendants as Z n
d
Y
(φi1 . . . φin )g (c1 (Tpi C)) i
Mg,n i=1

In reality, gravitational descendants are more complicated objects, but one can
prove using certain recursion relations proven/postulated by physicists that all cor-
relators between gravitational descendendants can be computed using only our
generalized correlators with values in cohomology groups. The string partition
function (after E. Witten, [W2]) is the generating function for these numbers and
it depends on infinitely many variables. E. Witten conjectured in a bit vague form
that this function is equal to some τ -function for an integrable hierarhy.
Gromov-Witten invariants.
Let V be a closed symplectic manifold with sufficiently large cohomology class
[ω] ∈ H2 (V, R) of the symplectic form ω. The exact meaning of words “sufficiently
large” will become clear soon.
One expects after E. Witten ([W1,W2], based on ideas of M. Gromov, [G]) that
there is an associated topological σ-model. The space of observables in this theory
is the sum of cohomology groups of V :
M
H := Hk (V ; C)
k

with the natural Z2 -grading. The scalar product on H is the usual Poincaré pairing:
Z
(φ, ψ) := φ∧ψ
V

Correlators are
(φi1 . . . φin )g :=
X  Z  Z
= exp − ω × (φi1 ⊗ · · · ⊗ φin ) ∧ P.d. (Ig,h,β )
β∈H2 (V ;Z) β Vn
4 MAXIM KONTSEVICH

Here P.d. denotes the Poinaré duality isomorphism between homology and coho-
mology groups of oriented manifolds and
Z
Ig,h,β ∈ HD Mg,n × V n ; Q , D := (1 − g)(dimV − 6) + 2n + 2


c1 (TV )
β

is called the Gromov-Witten invariant of symplectic manifold V .

Geometric meaning of Gromov-Witten invariants.
→ T V, J 2 = −1
To define Ig,h,β let us pick an almost-complex structure J : T V −
on V compatible with the symplectic structure in the obvious way:

ω(v, Jv) > 0 for v 6= 0, ω(u, v) = ω(Ju, Jv), u, v, ∈ Tx V .

Then consider the space Yg,n,β of equivalence classes of holomorphic maps

φ:C −
→V

from complex curves C of genus g with n marked points p1 , . . . , pn to V representing

homology class β ∈ H2 (V ; Z):
φ∗ ([C]) = β
The space Yg,n,β maps to Mg,n × V n ⊂ Mg,n × V n :

(φ; C; p1 , . . . , pn ) 7→ (C; p1 , . . . , pn ) × (φ(p1 ), . . . , φ(pn ))

It is easy to see using the Index Theorem that the dimension of Yg,n,β at eadch
point is greater than or equal to D. Let us assume that the dimension of Yg,n,β is
exactly D. Then this space is naturally oriented.
Then we compactify space Yg,n,β and define Ig,h,β as the image of its fundamental
class. There are several technical problems here: what to do if the dimension of the
space of maps is greater than the expected one, how to prove the independence of
invariants on the choice of J and how to check axioms of topological σ-model?
At the moment there are two different approachs to these problems.
The first solution by Y. Ruan and G. Tian (see [RT]) was formulated in somewhat
different terms. The main idea is to perturb generically Cauchy-Riemann equations
and use Gromov’s compactness theorems. R This approach works only in the case of
manifolds with c1 ≥ 0 (in the sense that C c1 (TV ) ≥ 0 for holomorphic curves C in
V ) and essentially only in the case of genus equal to zero. Also, with this approach it
is very hard to compute Gromov-Witten invariants effectively for complex projective
algebraic varieties.
The second approach (M.K., see [K2]) was designed to work for all genera and
for all manifolds. The main idea was to introduce a new compactification of the
space of holomorphic maps.
Definition. Stable map (φ; C; p1 , . . . , pn ) is a holomorphic map φ from a marked
connected compact curve C to V such that
(1) singularities of C are ordinary double points,
(2) marked points pi are pairwise distinct non-singular points of C,
(3) the group of automorphisms of (C; p1 , . . . , pn ) commuting with φ is finite.
 RIGOROUS RESULTS IN TOPOLOGICAL σ-MODEL 5

The last condition is equivalent to the statement that each contracted (= mapped
to a point of V ) component of C is stable in the Deligne-Mumford sense.
We define Mg,n (V ; β) as the moduli space of stable maps from curves with
arithmetic genus g and degree β. We consider it not just as a topological space but
as a stack (in a sense, we don’t want to forget about finite automorphism groups).
Notice that Mg,n (V ; β) is defined in the case 2 − 2g − n ≥ 0 if β 6= 0.
Basic properties of Mg,n (V ; β) are collected in the following
Theorem.
(1) for compact V the stack Mg,n (V ; β) is compact,
(2) at aT vicinity of each point the space Mg,n (V ; β) can be represented as
X1 X2 where X1 , X2 are two sub-orbifolds in an auxiliary smooth orb-
ifold Y such that dim X1 + dim X2 − dim Y = D,
(3) if V is homogeneous (for example,projective space or a flag variety) then
M0,n (V, β) is smooth stack (= it is an orbifold).

First two properties after some work produce certain “virtual fundamental class”
which is an element of HD (Mg,n (V ; β), Q). The third property means that in the
case of flag varieties and genus zero one can use naive definitions of numbers of
curves.
Recently we checked all desired properties for the second approach. The only
drawback is the lack of the control on the integrality of GW-classes.
The general problem with Gromov-Witten invariants is the convergence question.
Why the series Z
X
exp(− ω) . . .
β∈H2 (V ;Z) β

converges?
In all examples it is the case if one multiplies β ∈ H2 (V, R) by a sufficiently large
positive real number. In general, one expects that the σ-model is defined in an
open domain in the space of all symplectic structures. There is a way to avoid the
problem of convergence: to work over appropriate algebras of formal power series
instead of C (so called Novikov’s rings).
Associativity equation.
Genus zero part of Gromov-Witten invariants can be encoded into an analytic
function (prepotential, or the genus zero partition function) on the Hilbert space
H considered as a complex supermanifold:
∞
1
X Z
F0 (γ) := (γ, . . . , γ)0
n=3
n!
Mg,n

In all examples this series converges in a non-empty open domain U of H. The third
derivative of F0 at each point γ of U can be considered as a symmetric 3-tensor:

∂ 3 F0 |γ ∈ S 3 H∗ = S 3 T ∗ U|γ

The scalar product ( , ) on H can be used to raise one index. Thus we can consider
∂ 3 F0 |γ as a bilinear operation on the tangent space to U. When restricted to points
γ in H2 (V, C) ⊂ H this operation is called usually the quantum product on H.
6 MAXIM KONTSEVICH

The technique of stable maps gives a modification of the prepotential by a polyno-

mial of degree 2. This modificataion has the same third derivative but in examples
it gives shorter formulas. Namely, we redefine F0 as
 Z X ∞ n
1
X Z Y
F0 (γ) := exp − ω p∗i (γ) .
β n=0
n! i=1
β∈H2 (V ;Z)
Mg,n (V ;β)

The prepotential satisfies so called Witten-Dijkgraaf-Verlinde-Verlinde equation

(see [DVV] and [W2]). This very remarkable equation means that the product on
T U is associative. The quantum product is automatically commutative and has a
unit equal to 1 ∈ C = H0 (V, C) ⊂ H.
In our formalism the associativity follows immediately from the splitting axiom
and from certain linear relation among components of the compactification divisor
of M0,n . Denote by DS for S ⊂ {1, . . . , n}, 2 ≤ #S ≤ n − 2, the divisor in M0,n
which is the closure of the moduli of stable curves (C; p1 , . . . , pn ) consisting of two
irreducible components C1 , C2 such that pi ∈ C1 for i ∈ S and pi ∈ C2 for i ∈ / S.
Lemma. We have the following identity in H2 (M0,n , Z)
X X
[DS ] = [DS ] .
S:1,2∈S S:1,3∈S
3,4∈S
/ 2,4∈S
/

Both sides in the equality above are pullbacks under the forgetful map
M0,n − → M0,4 of points D{1,2} , D{1,3} ∈ M0,4 ≃ P1 . It is clear that any two
points on P1 are rationally equivalent as divisors. 
Every cohomology class of M0,n can be obtained as an intersection of several
divisors DS . It follows that whole genus zero part of correlators is determined by top
degree classes, i.e. the Taylor coefficients of F0 . In fact, the associativity equation
for F0 is equivalent to the possibility to reconstruct all lower degree classes:
Theorem (M.K.+Yu. Manin, R. Dijkgraaf+E. Getzler). Genus zero part of
the axioms of the topological sigma-model is equivalent to the associativity equation
on the formal power series F0 .
This theorem was proved via a complicated induction using results of S. Keel
(see the reference in [KM]) about cohomology algebras of spaces M0,n . Recently
E. Getzler obtained another proof using Hodge theory. There are two immediate
corollaries of this theorem:
(1) coupling with gravitational descendants in genus zero is completely deter-
mined by the series F0 ,
(2) there exists an algebraic procedure which made from two solutions of WDVV
equations F0 and F0′ in vector spaces H and H′ respectively a new solution
F0 “×”F0′ in the vector space H ⊗ H′ .
Cohomology-valued correlators in the tensor product of two toplogical field the-
ories are defined as products of correlators in the cohomology rings. The operation
of the tensor product on the level of functions F0 can not be completely elementary.
For example, the prepotential for V = CP 1 is very simple:
t20 t1
+ et1
F0 (t0 , t1 ) =
2
Nevertheless, the prepotential for V × V is quite non-trivial (see [DFI]).
 RIGOROUS RESULTS IN TOPOLOGICAL σ-MODEL 7

Applications of the associativity equation.

The simplest interesting case is V = CP 2 . The Hilbert space of the topological
theory is 3-dimensional:

H = C3 = H0 (V, C) ⊕ H2 (V, C) ⊕ H4 (V, C)

We introduce three coordinates t0 , t1 , t2 in H according to the decomposition above.

The prepotential has the form
∞
t20 t2 + t0 t21 X t3d−1 dt1
F0 (t0 , t1 , t2 ) = + nd 2 e
2 (3d − 1)!
d=1

Here the first summand comes from the contribution of the constant maps and
coefficients nd denote the number of rational curves of degree d passing through
generic 3d − 1 points on CP 2 . After the substitution of this general formula into
the associativity equation we obtain the following relation among numbers nd :

(3d − 4)!a2 b2 (3d − 4)!a3 b

X  
for d ≥ 2 nd = na nb −
(3a − 2)!(3b − 2)! (3a − 1)!(3b − 3)!
a,b≥1: a+b=d

Thus, it is enough to know only one number n1 which is the number of lines passing
through two points, i.e. n1 = 1. By recursion we obtain immediately

n2 = 1, n3 = 12, n4 = 620, n5 = 87304, n6 = 26312976, . . .

First three values are classical in algebraic geometry. Number n4 = 620 was com-
puted in 1873 by H. Zeuthen. Numbers n≥5 are all new.
In general, one can associate with a solution of the WDVV equation associated
with a symplectic manifold an isomonodromy deformation of a linear system of
differential equations with regular singularities on CP 1 in at most dim H+1 points.
In all computed examples the monodromy of this auxiliary equation takes values
in the group of integer matrices GL(N, Z). There is a physical explanation for
it proposed by S. Cecotti and C. Vafa (see [CV]) based on general properties of
massive N = 2 models. Mathematicians expect that any linear differential euation
with an integral monodromy and algebraic coefficients is equivalent to a Picard-
Fuchs equation. This means that one can expect a kind of mirror symmetry for
generic symplectic manifolds, not necessarily Calabi-Yau ones.
In the case V = CP 2 this isomonodromy deformation can be transformed into
the Painlevé VI equation. Using this fact one can find the asymptotic of numbers
nd :
nd
∼ const1 · d−7/2 econst2 d (1 + o(1))
(3d − 1)!
The reason for this is a kind of quasi-homogeneity of prepotential following from
the Riemann-Roch formula (= the Index theorem) for dimensions of spaces of maps.
In physics it correponds to the renormalization group flow. Let us introduce ho-
mogeneous coordinates in Z-graded vector space H. We define vector field L0 on
U by formula
X ti ∂ X ∂
L0 := (deg ti − 2) − 2(Coef f icienti c1 (TV ))
∂ti ∂ti
deg ti 6=2 deg ti =2
8 MAXIM KONTSEVICH

Then we have

L0 (F0 ) = (dimR V − 6)F0 + quadratic terms

Using the quantum product on the tangent space to H one can construct powers
of the vector field L0 :

Ln := L0 × L0 × · · · × L0 (n + 1 times)

for n ≥ −1. The remarkable fact is that vector fields Ln on H form an action of
the Witt algebra (= the Lie algebra of polynomial vector fields on the line):

[Ln , Lm ] = (m − n)Ln+m

Theorem (B. Dubrovin, [D1,D2]). If L0 is diagonalizable in an open set then

the associativity equation is a completely integrable system. Solutions of this system
depend on a finite number of parameters.
This theorem is applicable to many varieties with c1 > 0, including projective
spaces, Grassmanians etc.
Example from the Mirror symmetry.
Let V be a quintic 3-dimensional hypersurface in CP 4 . It is a Calabi-Yau man-
ifold, i.e. a complex manifold with c1 = 0. The prepotential F0 is essentially
encoded in a function of one variable:
∞
5 3 X phys
F (t) := t + Nd exp (dt)
6
d=1

Here again the first coefficient comes from constant maps and other coefficients are

1 1
Ndphys := Nd + Nd/2 + 3 Nd/3 + . . . ∈ Q
23 3

Numbers Nd ∈ Z are (conjecturally) numbers of rational curves of degree d on V .

The formula for Ndphys arised from the consideration of multiple ramified coverings
of rational curves in V as maps from CP 1 to V . The associativity equation gives
no restriction on the sequence Nd because all rational curves are isolated and do
not intersect each other.
The famous mirror formula of P. Candelas et al. (see [COGP] and [Y]) can be
described as follows. Let following four functions
∞
X (5n)! n
ψ0 (z) = z
n=0
(n!)5

∞ 5n
!
X (5n)! X 1
ψ1 (z) = log z · ψ0 (z) + 5 zn
n=1
(n!)5 k
k=n+1

ψ2 (z), ψ3 (z) = . . .
 RIGOROUS RESULTS IN TOPOLOGICAL σ-MODEL 9

form a basis of the space of solutions of the differential equation

 4 !
d d d d d
z − 5z(5z + 1)(5z + 2)(5z + 3)(5z + 4) ψ(z) = 0 .
dz dz dz dz dz

Then  
ψ1 ψ1 ψ2 + ψ0 ψ3
F = .
ψ0 ψ02
The mirror conjecture for quintics for a long time was not formulated well math-
ematically because it is still unknown whether the number of rational curves in
each degree is finite (conjecture of H. Clemens) and how to count contributions of
singular curves.
Calculations show, at least for first few degrees, that the definition with stable
maps give automatically the “physical number” of rational curves. We used the
following approach to the computational definition of Ndphys .
Denote by Xd the space of stable maps of degree d of genus zero curves without
marked points to CP 4 :
Xd := M0,0 (CP 4 , d[CP 1 ])
It is a compact complex orbifold of complex dimension 5d + 1. Let us introduce
a vector bundle Ed of rank 5d + 1 on Xd (in the orbifold sense too). The fiber of
Ed at the point of Xd representing a map φ : C − → CP 4 is defined as the space of
∗
global sections Γ(C, φ O(5)).
Our definition of the number of curves is
Z
Ndphys := c5d+1 (Ed )
Xd

The explanation of this formula is the following. The quintic V is given by an

equation of degree 5:
Q(x1 : x2 : x3 : x4 : x5 ) = 0
We can consider polynomial Q as a global section of the line bundle O(5) on
CP 4 . It induces in the evident way a global section Qd of the vector bundle Ed .
Zeroes of Qd coincide with the moduli of stable maps to V . If we perturb generically
the section Qd it will have finitely many zeroes. The number of these zeroes counted
with signs coming from the natural orientation is equal to the integral of the Euler
class of the bundle Ed .
We calculated using computer first four values of Ndphys , and these numbers
agreed with the mirror formula. Moreover, we obtained a closed formula for the
function F (t) thus reducing the Mirror conjecture to an explicit identity. We will
describe the general scheme of computations in the next section.
Toric methods.
Almost all examples of Calabi-Yau manifolds constructed up to now are ob-
tained by a resolution of singularities of complete intersections in toric varieties.
V. Batyrev and L. Borisov in [BB] proposed a general form of the mirror symmetry
for this class of manifolds. The combinatorial counterpart of the mirror symmetry
in the toric picture is the usual duality of convex cones in real vector spaces (a
generalization of the Legendre transform).
10 MAXIM KONTSEVICH

We will give here a sketch of a procedure giving a closed formula for Gromov-
Witten invariants of arbitrary complete intersections in toric varieties, not only
Calabi-Yau ones. We will illusrtate our method in the case of quintics in CP 4 .
The complex algebraic torus T := (C× )5 acts diagonally on C5 , hence on
CP 4 , on the line bundle O(5), on variety Xd and on the vector bundle Ed . In
T-equivariant case one can use a classical formula of R. Bott (see [B]) reducing the
computation of the integral of any characteristic class of the bundle to the integral
over the submanifold(s) of fixed points:
Eulerequiv (Ed )
Z X Z
Euler(Ed ) = equiv (N α )
Xd α Euler Xd
components Xd
of (Xd )T

Here Eulerequiv denotes the equivariant Euler class of a bundle, index α numerates
connected components of the set (Xd )T of fixed points and NXdα denotes the nornal
bundle to the corresponding component. The right hand side of this identity takes
values apriori in the field of rational functions on the vector space C5 which is the
Lie algebra of T. In fact, the r.h.s. is a constant.
Each point of (Xd )T is representing a map φ from a tree of projective lines
(CP 1 )k to CP 4 with the image of each irreducible component (CP 1 )k invariant
under the action of T. There are two types of components:
(1) contracted ones, φ((CP 1 )k ) is one of 5 fixed points {x1 , . . . , x5 } = (CP 4 )T ,
(2) non-contracted ones, φ((CP 1 )k ) is one of 10 lines passing through two fixed
points and the map φ in some homogeneous T-invariant coordinates looks
like
(z1 : z2 ) 7→ (z1d : z2d : 0 : 0 : 0)
We associate a new tree Γ with such a map by declaring connected components
of φ−1 ({x1 , . . . , x5 }) to be the vertices and non-contracted components to be the
edges. The vertices of Γ are labeled by indices 1, . . . , 5 and the edges are labeled
by degrees d = 1, 2, . . . .
One can see that the connected components of (Xd )T are in one-to-one corre-
spondence with the labeled trees. After a long calculation (using intersection theory
on M0,n ) one can find the contribution of each component into the Bott formula.
This contribution is a product of local weights. Thus, the whole generating func-
tion is equal to the sum over tree diagrams for some abstract “Lagrangian”. The
total sum is equal to the critical value of the action functional. Still the formula is
pretty complicated. Using some tricks we reduced the problem to the compuation
of the critical value of a new functional which is quadratic in all variables except
of a finite number of them.
Now the problem of counting of rational curves on quintics is reduced essentially
to the inversion of an integral operator with the kernel expressed via a generalized
hypergeometric function. We reproduce here from [K2] the resulting functional:
S = S(t, λi ; φij,d , µi ) =
Y 5
Y
d3 (λi − λj )2 (aλi + bλj − dλk )  
1X a+b=d:a,b≥1 k=1 ξi − ξj
exp −td − φij,d φji,d +
2 λi − λj
Y
i,j;d (aλi + bλj )
i6=j a+b=5d:a,b≥1
 RIGOROUS RESULTS IN TOPOLOGICAL σ-MODEL 11

1 X φij,d φij ′ ,d′

+ νi d d′
−
2 λi −λj + λi −λj′
i,j,d,j ′ ,d′
j,j ′ 6=i

X (λi − λj )2 X ξ3 5
X λi − λj
− νi ξi φij,d + νi φij,d + νi i .
d d2 i=1
6
i,j,d i,j,d
i6=j i6=j

Here indices i, j ran from 1 to 5 and d is a positive integer. Variables νi are

defined as
5λi
νi := Q
j:j6=i (λi − λj )

The critical value of S as a function in variables φij,d , µi is independent of λi and

equal to F (t) − 5t3 /6, the generating functions for numbers of rational curves on
quintics.
Applications of the Bott formula.
(1) The heuristic formula of P. Aspinwall and D. Morrison (see [AM]) for the
contribution of multiple coverings of rational curves in 3-dimensional Calabi-
Yau varieties was checked by Yu. Manin in [M] using summation over trees.
(2) One can get a formula for the genus-zero partition function for the complete
intersections in toric varieties (Batyrev-Borisov’s series of mirrors).
(3) One can count (in principle) numbers of higher genus curves in CP n , and,
more generally, in toric varieties.
The problem in the second and the third applications is that for non-homogeneous
toric variety V or if g ≥ 1 spaces Mg,n (V ; β) are singular. Still one can try to apply
the Bott formula because subspaces of fixed points Mg,n (V ; β)T are always smooth.
In the case of projective space and genus zero curves the class in the equivariant
K-theory of the normal bundle to Mg,n (V ; β)T can be decomposed into the sum of
certain standard piece and of the bundle with the fiber equal to H0 (C, φ∗ TV ). In
general situation it is not a vector bundle. Nevertheless, the formal difference

H0 (C, φ∗ TV ) − H1 (C, φ∗ TV )

in the equivariant K-theory is well-defined and we can use it formally. This pro-
cedure gives again some explicit formulas. We hope that in a future it will be
possible to obtain KP-hierarchy from representations of generating functions via
infinite-dimensional determinants.
For the calculations with higher genus curves in the “universal” toric formula
one has to compute certain integral over the moduli spaces of stable curves:

F (x∗ , y∗ , z∗ , λ) =

g
Y
(xαj + hj )
X λ2g−2 X Z j=1
= n
g,n n! α1 ,...,αg Mg,n Y
2−2g−n<0 β1 ,...,βn (yβi + zβi c1 (Tpi C))
i=1
12 MAXIM KONTSEVICH

where formal variables hj are defined via equality

g
Y g
X
(1 + hj ) = ck (Hg ) ∈ H∗ (Mg,n , Q)
j=1 k=0

Here Hg is the g-dimensional vector bundle over Mg,n with the fiber over C
equal to H1 (C, OC ). One can hope in analogy with the pure gravity case (Witten’s
conjecture, see [W2] and [K1]) that this universal function F is equal to the free
energy of some matrix model.

References
[AM] P. S. Aspinwall, D. R. Morrison, Topological field theory and rational curves, Commun.
Math. Phys. 151 (1993), 245–262.
[BB] V. Batyrev and L. Borisov, Dual Cones and Mirror Symmetry for Generalized Calabi-
Yau Manifolds, preprint alg-geom/9402002.
[B] R. Bott, A residue formula for holomorphic vector fields, Jour. Diff. Geom. 1 (1967),
311-330.
[COGP] P. Candelas, X. de la Ossa, P. S. Green, L. Parkes, A pair of Calabi-Yau manifolds as
an exactly soluble superconformal theory, in [Y], pp. 31–95.
[CV] S. Cecotti and C. Vafa, On Classification of N = 2 Supersymmetric Theories, Harvard
and SISSA preprints HUTP-92/A064 and SISSA-203/92/EP.
[DM] P. Deligne, D. Mumford, The irreducibility of the space of curves of given genus, Publ.
Math. I. H. E. S. 36 (1969), 75-110.
[DFI] P. Di Francesco, C. Itzykson, Quantum intersection rings, Saclay preprint SPhT 94/111
and hep-th/9412175.
[DVV] R. Dijkgraaf, H. Verlinde and E. Verlinde, Topological strings in d < 1, Nucl. Phys.
B352 (1991), 59–86.
[D1] B. Dubrovin, Integrable systems in topological field theory, Nucl. Phys. B379 (1992),
627–685.
[D2] B. Dubrovin, Geometry of 2d topological field theories, preprint SISSA-89/94/FM and
hep-th/9407018.
[G] M. Gromov, Pseudo holomorphic curves in symplectic manifolds, Invent. Math. 82
(1985), 307–347.
[K1] M. Kontsevich, Intersection theory on the moduli space of curves and the matrix Airy
function, Commun. Math. Phys. 147 (1992), 1–23.
[K2] M. Kontsevich, Enumeration of rational curves via torus actions, preprint hep-th/9405035.
[KM] M. Kontsevich, Yu. Manin, Gromov-Witten classes, quantum cohomology and enumer-
ative geometry, Comm. Math. Phys. 164:3 (1994), 525–562.
[M] Yu. Manin, Generating functions in algebraic geometry and sums over trees, alg-geom/9407005.
[RT] Y. Ruan, G. Tian, Mathematical theory of quantum cohomology, preprint (1993).
[W1] E. Witten, Topological sigma models, Commun. Math. Phys. 118 (1988), 411-449.
[W2] E. Witten, Two-dimensional gravity and intersection theory on moduli space, Surveys
in Diff. Geom. 1 (1991), 243–310.
[Y] S. T. Yau, ed., Essays on Mirror Manifolds, International Press Co. , Hong Kong, 1992.