arXiv:hep-th/9906046v3 11 Jun 1999 hep-th/9906046 IASSNS-HEP–99/55 Local Mirror Symmetry at Higher Genus Albrecht Klemm ∗ and Eric Zaslow ∗∗ ∗ School of Natural Sciences, IAS, Olden Lane, Princeton, NJ 08540, USA ∗∗ Department of Mathematics, Northwestern University, Evanston, IL 60208, USA Abstract We discuss local mirror symmetry for higher-genus curves. Specifically, we consider the topological string partition function of higher-genus curves contained in a Fano surface within a Calabi-Yau. Our main example is the local P 2 case. The Kodaira-Spencer theory of gravity, tailored to this local geometry, can be solved to compute this partition function. Then, using the results of Gopakumar and Vafa [1] and the local mirror map [2], the partition function can be rewritten in terms of expansion coefficients, which are found to be integers. We verify, through localization calculations in the A-model, many of these Gromov-Witten predictions. The integrality is a mystery, mathematically speaking. The asymptotic growth (with degree) of the invariants is analyzed. Some suggestions are made towards an enumerative interpretation, following the BPS-state description of Gopakumar and Vafa. ∗ e-mail: [email protected]∗∗ e-mail: [email protected]
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
arX
iv:h
ep-t
h/99
0604
6v3
11
Jun
1999
hep-th/9906046
IASSNS-HEP–99/55
Local Mirror Symmetry at Higher Genus
Albrecht Klemm∗ and Eric Zaslow∗∗
∗School of Natural Sciences, IAS, Olden Lane, Princeton, NJ 08540, USA
∗∗Department of Mathematics, Northwestern University, Evanston, IL 60208, USA
Abstract
We discuss local mirror symmetry for higher-genus curves. Specifically, we consider
the topological string partition function of higher-genus curves contained in a Fano surface
within a Calabi-Yau. Our main example is the local P2 case. The Kodaira-Spencer
theory of gravity, tailored to this local geometry, can be solved to compute this partition
function. Then, using the results of Gopakumar and Vafa [1] and the local mirror map [2],
the partition function can be rewritten in terms of expansion coefficients, which are found
to be integers. We verify, through localization calculations in the A-model, many of these
Gromov-Witten predictions. The integrality is a mystery, mathematically speaking. The
asymptotic growth (with degree) of the invariants is analyzed. Some suggestions are made
towards an enumerative interpretation, following the BPS-state description of Gopakumar
The moduli space, Mg,0(β;P), of stable maps from genus g curves into some toric
variety P with image in β ∈ H2(P;Z) can have components of various dimensions. To do
intersection theory on it, we need a fundamental cycle of some “appropriate” dimension –
the virtual fundamental class.
This procedure was developed in [10], and the equivariant localization for spaces with
perfect deformation theories was developed in [11]. In that paper, the authors treat the
case of interest to us: the restrictions to the fixed point loci (under the torus action) of the
equivariant fundamental cycle on the space of holomorphic maps of genus g. The authors
give explicit formulas for the weights.
In our examples, we are interested in calculating Chern classes of Uβ, the bundle over
moduli space whose fiber at a point (C, f) ∈ Mg,0(β;P) is equal to H1(C, f∗KP). It is
not difficult to compute the equivariant Chern classes at the fixed loci.
Using these ingredients, we can perform localization calculations at higher genus. The
Atiyah-Bott fixed point formula – true for virtual classes by [11] – reduces the calculation to
integrals over these fixed loci. The different components of the fixed loci are represented
by different graphs. We sum over the graphs using a computer algorithm. As all the
higher-genus behavior must be centered at the fixed points (there are no fixed higher-
genus curves), the fixed loci are all products of moduli spaces of curves, and the graphs
are decorated by genus data at each vertex as well as degree data at each edge, as in [8].
The integrals over the moduli spaces are performed using the algorithm of Faber [12].
In the following subsections, we summarize this procedure.
2.1. Localization Formulas
As discussed in [2], a Fano surface within a complete-intersection Calabi-Yau con-
tributes to the Gromov-Witten invariants, producing an effective “number” of rational
curves. In fact, the curves are not isolated and there is a whole moduli space of maps
into the Fano surface, arising as a non-discrete zero locus of a section of a bundle over the
moduli space of maps in the Calabi-Yau. Just as spaces of multiple coverings of isolated
3
curves contribute to the Gromov-Witten invariants by the excess intersection formula,4
the contribution from the surface B can be calculated as well. It is
Kgβ =
∫
Mg,0(β;B)
c(Uβ), (2.1)
where β ∈ H2(B;Z) and Uβ is the bundle whose fiber over (C, f) ∈ Mg,0(β;B) is the
vector space H1(C, f∗KB).5
We assume that B has a toric description so that we may use the torus action to define
an action on M0,0(β;B) (moving the image curves) and on Uβ, which inherits the natural
action on the canonical bundle. We evaluate the Gromov-Witten invariants by localizing
to the fixed points using the Atiyah-Bott formula
∫
M
φ =∑
P
∫
P
(
i∗Pφe(νP )
)
,
where the sum is over fixed point sets P, iP is the embedding into M, and e(NP/M ) is the
Euler class of the normal bundle of P in M.
Graber and Pandharipande proved that this formula holds mutatis mutandis, with the
replacement of the integral of a class by evaluation over virtual fundamental cycles (in the
equivariant Chow ring) of equivariant classes of deformation complexes. The equivariant
Chern classes and bundles involved are then computed in the standard way, e.g. by looking
at the weights of sections and taking alternating products over terms in a complex. In our
case, following Kontsevich, the fixed moduli can be labeled by graphs. The fixed locus MΓ
corresponding a graph Γ is a product of moduli spaces:
MΓ =∏
v
Mg(v),val(v).
Here v are the vertices of the graph and g(v) and val(v) are the genus and valence of the
vertex. (At a vertex, which is mapped to a torus fixed point, the choice of a val(v)-marked
genus g(v) curve represents the only moduli.)
4 The contribution, from a component of the zero locus Y of some section, to the Chern class
c(E) of some vector bundle E over M is∫
Y
c(E)c(NY/M )
, where NY/M represents the normal bundle
of Y in M. The equation here follows from the reasoning between eqs. (5.4) and (5.5) in [2].5 Let ev be the evaluation map from Mg,1(β;B) to B and let π be the map Mg,1(β;B) →
Mg,0(β;B) which forgets the marked point. Then Uβ = R1π∗ev∗KB.
4
Graber and Pandharipande [11] found the weights of the inverse of the Euler class of
the normal bundle to be:
1
e(Nvir)=∏
e
(−1)ded2dee
(de!)2(λi(e) − λj(e))2de
∏
a+b=dea,b≥0
k 6=i(e),j(e)
1adeλi(e) + b
deλj(e) − λk
×∏
v
∏
j 6=i(v)
(λi(v) − λj)val(v)−1
×
∏
v
(
∑
F
w−1F
)val(v)−3∏
F∋v
w−1F
if g(v) = 0
∏
v
∏
j 6=i(v)
Pg(v)(λi(v) − λj , E∗)∏
F∋v
1
wF − eFif g(v) ≥ 1
(2.2)
where the case with g(v) = 0 is from [13]. Here i(v) is the fixed-point image of the vertex
v; edges e represent invariant P1’s connecting the fixed point labeled i(e) to j(e); and
flags F are pairs (v, e) of vertices and edges attached to them. ωF = (λi(F ) − λj(F ))/de,
where i(F ) = i(e) and j(F ) = j(e) for the associated edge e ∈ F. eF is the Chern
class of the line bundle over MΓ whose fiber is the point e(F ) ∩ Cv, a “gravitational
descendant” in physical terms – also known as a κ class. We have defined the polynomial
Pg(λ,E∗) =
∑gr=0 λ
rcg−r(E∗), where E is the Hodge bundle with fibersH0(KC) on moduli
space.
The bundle Uβ , whose top Chern class φ we are calculating, has fibers H1(C, f∗KB)
over a point (C, f). Let us now fix B = P2 and put β = d since H2(P2) is one-dimensional.
Since the invariant maps are known and the maps from the genus g(v) > 0 pieces are
constants, we explicitly calculate the weights of the numerator in the Atiyah-Bott formula
to be:
i∗(φ) =∏
v
Λval(v)−1i(v) Pg(v)(Λi(v), E
∗)∏
e
[
3de−1∏
m=1
Λi(e) +m
de(λi(e) − λj(e))
]
, (2.3)
where Λi = λ1 + λ2 + λ3 − 3λi.
2.2. Faber’s algorithm
We therefore have explicit formulas for the class to integrate along MΓ. The integrals
involve the κ classes and the Mumford classes (from the Hodge bundle). The integrals can
be performed by the recursive algorithm developed by Faber.
5
A rough sketch of the idea of Faber’s algorithm is as follows [12]. Witten’s origi-
nal recursion relations from topological gravity [14], proved by Kontsevich [15], suffice to
determine integrals of powers of the ψi’s, which are the first Chern classes of the line
bundles whose fiber is the cotangent space of the corresponding marked point, i = 1, ..., n
(the gravitational descendents). However, the integrals over MΓ involve the λ classes as
well – the Chern classes of the Hodge bundle. These classes can be represented by cycles
involving the boundary components of moduli space as well as the ψi’s.
So a proper intersection theory including boundary classes is needed. The boundary
components are images under maps from Mg−1,n+2 (the map identifying the last two
points; this map is of degree two since swapping the points gives the same boundary
curve) and Mh,s+1 × Mg−h,(n−s)+1 (the map identifying the two last points, which is
degree one except when both n = 0 and h = g/2). Understanding the pull-back of the
boundary classes under these maps completes the reduction of intersection calculations
to lower genera. For example, the ψi’s pull back to ψi’s of corresponding points in the
new moduli spaces. One must also note that since the boundary divisors can intersect
(transversely, in fact), boundary classes must also be pulled back to moduli spaces to
which they do not correspond. We refer the reader to [12] for details.
What remains, then, is to express the Chern classes of the Hodge bundle E in terms of
the boundary divisors. This is precisely the content of Mumford’s Grothendieck-Riemann-
Roch (G-R-R) calculation. Briefly (too briefly!), there is the forgetful map ρ : Mg,1 →
Mg,0 (we now set C ≡ Mg,1), with respect to which we wish to push forward the relative
canonical sheaf ωC,M. Since ρ∗ωC,M = E, the G-R-R formula tells us the Chern character
classes of E (we need that R1ρ∗ωC,M = O). The formula calls for integration of ch(ωC,M)
with the Todd class of the relative tangent sheaf. The former involves the descendents,
while the relative tangent sheaf is dual to the relative cotangent sheaf, which differs from
ωC,M only at the singular locus (on smooth parts, sections of the canonical bundle are just
one-forms). The Todd expansion is responsible for the appearance of Bernoulli numbers,
while the difference from the canonical sheaf introduces the divisor classes.
Therefore, integrals of descendent classes and λ classes can be performed by using
Mumford’s formula (see section 1 of [16]) to convert λ classes to descendents and divisor
classes, then on the boundaries of M pulling these clases back to the moduli spaces which
cover the boundaries. As these spaces have lower genera, the integrals can be performed
recursively. Faber has written a computer algorithm in Maple for this procedure, which
he generously lent us for this calculation.
6
2.3. Summing over graphs
The contribution to the free energy F (g) from genus g curves in the class β is given
from the fixed point formula as
Kgβ =
∑
Γ
1
|AΓ|
∫
MΓ
i∗(φ)
e(Nvir), (2.4)
with e(Nvir)−1 and i∗(φ) as described in (2.2)(2.3). The formal expansion of the integrand
yields cohomology elements of Mg,n, the moduli space of pointed curves at the vertices.
The integral over MΓ splits into integrals over these moduli spaces, against which the co-
homology elements of the right degree have to be integrated. In addition, the contribution
of a given graph must be divided by the order of the automorphism group AΓ, as we are
performing intersection theory in the orbifold sense. AΓ contains the automorphism group
of Γ as a marked graph and a Zdefactor for each edge which maps with degree de.
Hence it remains to construct the graphs Γ which label the fixed point loci of
Mg,k(d,P2) under the induced torus action, then to carry out the summation and in-
tegration. A graph is labeled by a set of vertices and edges, and can be viewed as a
degenerate domain curve with additional data specifying the map to P2. The vertices v
represent the irreducible components Cv. They are mapped to fixed points in P2 under
the torus action. The index of this fixed point i(v) and the arithmetic genus g(v) of the
component Cv are additional data of the graph. The edges e represent projective lines
which are mapped invariantly, with degree de, to projective lines in the toric space con-
necting the pair of fixed points. A graph without the additional data g(v), de, i(v) will be
referred to as undecorated.
The following combinatorial conditions specify a Γg,d,k graph representing a fixed-
point locus of genus g, degree d maps with k marked points:6
(1) if e ∈ Edge(Γ) connects the vertices (u, v) then i(u) 6= i(v)
(2) 1 − #(Vertices) + #(Edges) +∑
v∈Vert(Γ) g(v) = g
(3)∑
e∈Edge(Γ) de = d
(4) 1, . . . , k = ∪v∈Vert(Γ)S(v).
To construct all Γg,d,0 graphs we start by generating a list of all possible undecorated
graphs that can be decorated to yield Γg,d,0 graphs. Hence the number of edges is restricted
6 Condition four is relevant only for maps with k marked points, which we do not consider
(k = 0). S(v) is then the subset of marked points on the component Cv.
7
by (3) while the number of vertices is restricted by (2). The undecorated graphs with nv
vertices can be represented by an nv×nv symmetric incidence matrix, m. Entries mi,j = 1
or 0 represent a link or no link between vi and vj , so with a fixed number of edges ne there
will be(
(nv+1)nv2
ne
)
possibilities. A main problem is to identify among the graphs those
which have different topology. To obtain invariant data, independent of the ordering of the
vertices and depending only on the topology of the graph, we use a so called depth-first-
search algorithm. This starts with a vertex vk and descends level-by-level along all occuring
branches collecting the data of the encountered vertices, namely the valence (and in later
applications the additional information g(v) and i(v) as well as edge data dα for the map),
into a list Tk graded by the level, until all vertices have been encountered. A lexicographic
ordering can be imposed on the combined lists of all vertices TΓ = T1, . . . , Tnv and
TΓi= TΓk
only if Γi is isomorphic to Γk. These invariants will be used for a.) generating
distinct undecorated graphs, b.) decorating them without redundancy, and c.) finding the
automorphism group.7 Fig. 1 shows as an example the generation of genus 2, degree 3
graphs.
We have implemented the above generation of graphs8 of (2.4) in a completely au-
tomated computer algorithm that uses Faber’s algorithm to perform the integrals over
Mg,n. We exhibit the results of these localization calculations in the instanton pieces
F(g)inst =
∑
d>0Kgdq
d of the partition functions F (g) listed below:
7 A useful computer program [17] was used to check the number of undecorated graphs.8 The number of graphs grows very quickly with the genus. E.g., to calculate the d = 5 term
(for g = 0, . . . , 5) in (2.5), one sums over six times 17321, 733101, 2295313, 5353719,
111011442, 203452570 fully-decorated graphs, where the number in braces indicates the num-
ber of graphs with just g(v) and de decorations. For example, in the calculation of (g, d) = (4, 5)
the maximal dimension of a vertex moduli space dim(Mg,n) = 14 occurs in the “star graphs” –
irreducible genus 4 curves with five legs. At this dimension, 22462 different 2-d gravity integrands
must be evaluated.
8
i j
k
i j
i j i
i kj
i j i
i kj
iijj
iiii
kkkk
ijij
1263 2 2 2 2
2 2 2 2 2 2
226 2 2
2
1 1 1 1
3 32
2 2
2 2
n =1
n =2
n =3
g=1g=0 g=2
2
j i jii j i k
i j k j
i j k i
i j
i j
2
2
Fig. 1: The first part of the figure shows a scheme of all undecorated graphs that can be
decorated to g = 2, d = 3 graphs. In the middle we show the 24 different ways to decorate
them by de (small numbers, if different from one) and g(v) (number of circles), with the
order of the automorphism AΓ indicated by the bigger numbers. In the last part of the
figure we show the decoration possibilities by the i(v) indices, which lead to 63 graphs. The
indices (i, j, k) are finally summed over all permutations of (1, 2, 3).
9
F (0) = −t3
18+ 3 q −
45 q2
8+
244 q3
9−
12333 q4
64+
211878 q5
125. . .
F (1) = −t
12+q
4−
3 q2
8−
23 q3
3+
3437 q4
16−
43107 q5
10. . .
F (2) =χ
5720+
q
80+
3 q3
20−
514 q4
5+
43497 q5
8. . .
F (3) = −χ
145120+
q
2016+
q2
336+q3
56+
1480 q4
63−
1385717 q5
336. . .
F (4) =χ
87091200+
q
57600+
q2
1920+
7 q3
1600−
2491 q4
900+
3865234 q5
1920. . .
F (5) = −χ
2554675200+
q
1774080+
q2
14080+
61 q3
49280+
4471 q4
22176−
65308319 q5
98560. . .
(2.5)
2.4. Organizing the partition functions
In order to make enumerative predictions from the partition functions, we need the
analogue of the multiple-cover formula (1/k3). In other words, we need to know how a
“fundamental object,” a holomorphic curve or D-brane (BPS state) of given charge and
spin content, contributes to the partition function.
The functional form of F (g) was derived in [1]:
F (λ) =∞∑
g=0
F (g)λ2g−2 =∑
di,g≥0,k>0
ngdi
1
k(2 sin
kλ
2)2g−2exp[−2πk
∑
i
diti]. (2.6)
Here the di define the homology class (charge) of the BPS state, and ngdi are the
number of BPS states of charge di and left-handed spin content described by g.9
The table below was generated by extracting the integers from the partition functions
as indicated. Results for degrees higher than five come from the B-model calculation,
which we review in the next section.
9 BPS states are killed by the right-handed supersymmetry generators (half), where so(4) ∼=su(2)L ⊗ su(2)R. BPS supermultiplets, formed from a spin (j1, j2) Fock “vacuum,” have states
with left-handed content [( 12) ⊕ 2(0)] ⊗ j1, while the 2j2 + 1 right-handed states only contribute
an overall factor (with sign). Defining I = ( 12) ⊕ 2(0), g labels the representation I⊗g.