nLab basic ideas of moduli stacks of curves and Gromov-Witten theory

This is a sub-entry of Gromov-Witten invariants. See there for further background and context.

This entry is supposed to provide an exposition of some basic ideas underlying Gromov-Witten theory.

raw material: notes taken verbatim in some seminar – needs to be polished


Intro Part I: basics of moduli stacks of curves

question: what is a moduli problem?

  • some kind of object;

  • a notion of “familiy”/“deformation” of these objects

  • some equivalence relation \sim (possibly trivial) on the set of such families S(B)S(B) over BB

from this we get a functor (a presheaf)

F:C opSet F : C^{op} \to Set
BS(B)/ B \mapsto S(B)/_\sim

terminology an object MM is called a fine moduli space if the corresponding represented functor

Y M:C opSet Y_M : C^{op} \to Set
xHom(x,M) x \mapsto Hom(x,M)

is isomorphic to FF, i.e. if it represents FF.

examples in the homotopy category of topological spaces we have

{complexlinebundlesoverB}/isomHom Ho(Top)(B,P ) \left\{ complex line bundles over B \right\}/isom \leftrightarrow Hom_{Ho(Top)}(B, \mathbb{C}P^\infty)

so P \mathbb{C}P^\infty is the classifying space B U ( 1 ) B U(1) for complex line bundle. Similarly for higher rank vector bundles and Grassmannians.

The analogue in the algebraic category is

{linebundlesLoverBwithgeneratingsectionss 0,...,s nΓ(L)}/isomHom(B, n). \left\{ line bundles L over B with generating sections s_0,...,s_n \in \Gamma(L) \right\} / isom \leftrightarrow Hom(B, \mathbb{P}^n).

And also similarly for higher rank vector bundles and Grassmannians.

Despite these examples, in a lot of cases the functors are not representable. We’ll see some of these examples below.

Why are fine moduli spaces desireable?

They allow us to study a single family which tells us universal things about all families. Even if you do not care about families or deformations, moduli spaces can help, because perhaps they can tell you something about trivial families, i.e. the objects that you are studying to begin with.

example studying the cohomology rings of B O ( n ) B O(n) =Gr n( )=Gr_n(\mathbb{R}^\infty) or B U ( n ) B U(n) =Gr n( )=Gr_n(\mathbb{C}^\infty), which are the classifying space and for higher rank real and complex vector bundles gives universal relations (or, rather, the lack thereof!) among Chern classes, etc.

Let’s look at elliptic curves (we’ll work over \mathbb{C}).

the functor of families here is

F:Sch/ opSet F : Sch/\mathbb{C}^{op} \to Set
B{E Bflatfamiliesofellipticcurves} B \mapsto \left\{ \array{ E \\ \downarrow \\ B } flat families of elliptic curves \right\}

Fact: there is no fine moduli space of elliptic curves representing this functor

why not?

there is something called the j-invariant which classifies elliptic curves up to isomorphism


E:y 2=x 3+ax+b E : y^2 = x^3 + a x + b

be an [elliptic curve]] given by parameters a,ba,b. Then its j-invariant is the number

j(E)=2 83 3a 34a 3+27b 2 j(E) = \frac{2^8 3^3 a^3}{4 a^3 + 27 b^2}

so we might guess that the “jj-line” 𝔸 1\mathbb{A}^1 is a fine moduli space for elliptic curves,

i.e. that there is a “universal family” of elliptic curves C𝔸 1C \to \mathbb{A}^1 such that the fiber over j𝔸 1j \in \mathbb{A}^1 is the elliptic cuvre with that jj-invariant.

so that for any family EBE \to B we’d have a pullback

E C B 𝔸 1 \array{ E &\to& C \\ \downarrow && \downarrow \\ B &\to& \mathbb{A}^1 }

where B𝔸 1B \to \mathbb{A}^1 sends a point in BB to the j-invariant of the elliptic curve in the fiber over BB.

Now consider the family

χ=(y 2=x(x1)(x+λ))𝔸 1×𝔸 λ 1× 2 \chi = (y^2 = x(x-1)(x+\lambda)) \subset \mathbb{A}^1 \times \mathbb{A}_\lambda^1 \times \mathbb{P}^2
χ C 𝔸 λ 1 𝔸 1 \array{ \chi &\to& C \\ \downarrow && \downarrow \\ \mathbb{A}^1_\lambda &\to& \mathbb{A}^1 }
j(χ λ)=s 8(λ 2λ+1) 2λ 2(λ1) 2 j(\chi_\lambda) = s^8 \frac{(\lambda^2 - \lambda + 1)^2}{\lambda^2(\lambda-1)^2}

now send 𝔸 λ 1𝔸 λ 1\mathbb{A}^1_\lambda \to \mathbb{A}^1_{\lambda}

by dually sending λ1λ\lambda \mapsto 1-\lambda

check that j(χ λ)=j(χ 1λ)j(\chi_\lambda) = j(\chi_{1-\lambda})

this induces in the fibers a map

ϕ:χχ\phi : \chi \to \chi

ϕ:(x,y,λ)(1x,±iy,1λ) \phi : (x,y,\lambda) \mapsto (1-x,\pm i y , 1-\lambda)

this maps further to (x,y,λ)(x,-y,\lambda). But this would have to be the identity for 𝔸 1\mathbb{A}^1 to be a fine moduli space, which it is not. So 𝔸 1\mathbb{A}^1 at least is not a fine moduli space, even though it might look like one.

so that gives some computational insight that something goes wrong

This argument does not yet prove that there exists no moduli space of elliptic curves. It merely proves that the “j-line” 𝔸 1\mathbb{A}^1 can not be the moduli space of elliptic curves. However, the basic argument can be adapted, if one so desires, to in fact prove that there is no moduli space of elliptic curves.

In fact, below we will see (in an exercise), that the jj-line is a coarse moduli space, as explained below. Moreover, if a coarse moduli space exists, then it is unique up to canonical isomorphism. Since fine moduli spaces, if they exist, are also coarse moduli spaces, and the jj-line is a coarse but not a fine moduli space by the above argument, it follows that no fine moduli space exists.

Abstract argument Yet another, more abstract way, to see that no fine moduli space can exist it to realize that since elliptic curves have nontivial automorphisms, it is possible to construct families of elliptic curves that are locally trivial families (of the form U×EUU \times E \to U for a fixed elliptic curve UU) but which are glued together from these local pieces using nontrivial automorphisms such that the resulting family VXV \to X is not globally trivial, i.e. not globally of the form EXXE \to X \to X.

With a bit of care this alone can be used to show that a fine moduli space cannot exist. This is often summarized by a slogan of the form

Slogan : nontrivial automorphisms of objects prevent the family-assignment of these objects to be representable by a fine moduli space.

However, one has to be careful with interpreting this slogan correctly. Taken naively, the slogan alone would also seem to imply that, since vector spaces have nontrivial automorphisms, no classifying space for “families of vector spaces”, i.e. for vector bundles does exist, while of course this does exist (recall also the example further above). But if one interprets the slogan carefully, it does yield a true statement. For more on that see the discusson at moduli space.

How to “fix” these problems.

  1. add extra structure to the objects under consideration (e.g. add marked points) to make the automorphism groups trivial.

  2. instead of looking for representing topological spaces, look for representing groupoids / stacks.

  3. use coarse moduli spaces MSch/M \in Sch/\mathbb{C} with Ψ M:Fh M\Psi_M : F \to h_M such that

    a) F(Spec())h M(Spec)=hom(Spec,M)F(Spec(\mathbb{C})) \to h_M(Spec \mathbb{C}) = hom(Spec \mathbb{C}, M) is a bijection

    b) given MM' and Ψ M:Fh M\Psi_{M'} : F \to h_{M'} then there exists unique MMM \to M' such that F Ψ M h M Ψ M h M\array{ F && \stackrel{\Psi_{M'}}{\to}&& h_{M'} \\ & {}_{\Psi_M}\searrow && \nearrow \\ && h_M}

So a coarse moduli space is one that at least has the right underlying set of points as the right moduli stack has: as long as we don’t look at families but just at single things, it does give the right classification.

exercise show that the jj-line 𝔸 1\mathbb{A}^1 is, while not a fine moduli space, a coarse moduli space.

exercise Show that if a coarse moduli space exists, then it is unique up to canonical isomorphism.

fact there exists a coarse moduli space M g,nM_{g,n} of Riemann surfaces of genus gg with nn marked points and a fine moduli stack? M g,nM_{g,n} such that for all g,ng,n we have

M g,nM_{g,n} is a smooth Deligne-Mumford stack - aka orbifold

except for (g,n)=(g,n) = (0,0),(0,1),(0,2),(1,0)(0,0), (0,1), (0,2), (1,0) (these are the cases where the automorphism group is infinite, so in these cases we don’t get a Deligne-Mumford stack)


  1. M g,nM_{g,n} is not proper, meaning: not compact, so we don’t have, for example, Poincare duality on M g,nM_{g,n} and no integration theory. Proper, or even better projective, schemes or stacks are just a lot easier to deal with.

  2. Sometimes one wants to study singular curves or families with degeneracies.

Both “issues” can be “resolved” via Deligne-Mumford compactification.

M¯ g,n \bar M_{g,n}

which parameterizes at-most-nodal curves, that are connected, of arithmetic genus gg, with nn smooth marked points, and the group of automorphisms is finite.

M¯ g,n\bar M_{g,n} is a smooth proper Deligne-Mumford stack.

smooth here means smoothness as for orbifolds.

Deligne and Mumford were able to prove many theorems about the ordinary moduli space of curves by studying instead the compactification. For example they were able to prove that M g,nM_{g,n} is irreducible.

Intro Part II: basics of Gromov-Witten theory

Gromov was looking for invariants of symplectic manifolds: his idea was to use JJ-holomorphic curves in compact symplectic manifolds to get symplectic invariants

Edward Witten and other physicists studied worldsheets of strings in some spacetime manifold (e.g a Calabi-Yau 3-fold)

we want to consider now genus gg Riemann surfaces with nn marked points mapping into some space XX

For a fixed

βH 2(X,), \beta \in H_2(X, \mathbb{Z}),
M g,n(X,β) M_{g,n}(X,\beta)

is the space that parameterizes maps

ΣfX \Sigma \stackrel{f}{\to} X

where Σ\Sigma is smooth, and has nn marked points, and such that for [Σ][\Sigma] the fundamental homology class? of Σ\Sigma we have

f *[Σ]=β. f_*[\Sigma] = \beta \,.

this is a smooth Deligne-Mumford stack. (Again we must exclude the cases of small (g,n)(g,n). Excluding these cases, the automorphisms of the surfaces and the automorphisms of the maps are automatically finite.)

similarly, write M¯ g,n(X,β)\bar M_{g,n}(X,\beta) for the same setup but with Σ\Sigma from M¯ g,n\bar M_{g,n} as above in part 1 (the DM compactified moduli stack). EXCEPT here we do not require that Σ\Sigma (with its nn marked points) has finite automorphism group; we require instead that the MAP has finite automorphism group, which means…….(fill in)

This is a proper (compact) Deligne-Mumford stack - but not smooth (not even in the sense of smooth stacks!). This is very important. This is what makes the theory difficult/nontrivial/interesting.

we have

M¯ g,n(pt,0)=M¯ g,n \bar M_{g,n}(pt,0) = \bar M_{g,n}

what do string theorists want to do?

we have evaluation maps

M¯ g,n(X,β)ev iX \bar M_{g,n}(X,\beta) \stackrel{ev_i}{\to} X

where ii labels a marked point, and we want morphisms

H (X) nev 1 *ev 2 *H *(M¯ g,n(X,β)) [M¯ g,n(x,β)] virtual H^\bullet(X)^{\otimes n} \stackrel{ev^*_1 \wedge ev^*_2 \wedge \cdots}{\to} H^*(\bar M_{g,n}(X,\beta)) \stackrel{\int_{[\bar M_{g,n}(x,\beta)]^{virtual}}}{\to} \mathbb{C}

There should be a virtual fundamental class [M¯ g,n(x,β)] virtual[\bar M_{g,n}(x,\beta)]^{virtual} that makes the maps above into the correlation functions of a quantum field theory (the integral would be the path integral of the worldsheet sigma-model) with state space H *(X)H^*(X). We’ll explain some of what this means below.

This virtual fundamental class in algebraic geometry was constructed by Li-Tian and Behrend-Fantechi; in symplectic geometry it was done by Li-Tian.

Why do we want to use a “virtual” fundamental class? Because \bar M_{g,n}(x,\beta) is not smooth, the actual fundamental class may not behave very well. NEEDS ELUCIDATION – it would be nice if someone could give a more complete explanation of why we need a virtual fundamental class

the mathematical structure of GW-theory was elucidated Ruan and then by Maxim Kontsevich and Manin in 1994.

M¯ g,n(X,β) ϕ M¯ g,n \array{ \bar M_{g,n}(X,\beta) \\ \downarrow^\phi \\ \bar M_{g,n} }
I g,n,β:H *(X) H *(M¯ g,n(X,β))ϕ *H *(M¯ g,n) I_{g,n,\beta} : H^*(X)^{\otimes} \to H^*(\bar M_{g,n}(X,\beta)) \stackrel{\phi_*}{\to} H^*(\bar M_{g,n})

where the map ϕ *\phi_* is some kind of “pushforward” or “integration along the fibers”; this map uses the virtual fundamental class

there is also a map

α:M¯ g 1,n 1+1×M¯ g 2,n 2+1M¯ g,n \alpha : \bar M_{g_1,n_1+1} \times \bar M_{g_2,n_2+1} \to \bar M_{g,n}

with g=g 1+g 2g = g_1 + g_2 and n=n 1+n 2n = n_1 + n_2 obtained by gluing the last marked point of the first curve to the first marked point of the second curve.

so consider now the combination of these two maps

H (X) I g,n,β H (M¯ g,n(X,β))ϕ *H *(M¯ g,n) Δ α H (X) n 1H (X)H (X)H (X) n 2 H (M¯ g 1,n 1+1×M¯ g 2,n 2+1) H (X) n 1+1H (X) n 2+1 β 1+β 2=βI g 1,n 1+1,β 1I g 2,n 2+1,β 2 H (M¯ g 1,n 1+1)H (M¯ g 2,n 2+1)) \array{ H^\bullet(X)^{\otimes} &\stackrel{I_{g,n,\beta} }{\to}& H^\bullet(\bar M_{g,n}(X,\beta)) \stackrel{\phi_*}{\to} H^*(\bar M_{g,n}) \\ \downarrow^{- \otimes \Delta \otimes -} && \downarrow^\alpha \\ H^\bullet(X)^{\otimes n_1} \otimes H^\bullet(X) \otimes H^\bullet(X) \otimes H^\bullet(X)^{\otimes n_2} && H^{\bullet}(\bar M_{g_1,n_1+1} \times \bar M_{g_2,n_2+1}) \\ \downarrow^\simeq && \downarrow^{\simeq} \\ H^\bullet(X)^{\otimes n_1 + 1} \otimes H^\bullet(X)^{\otimes n_2 + 1} &\stackrel{\sum_{\beta_1+\beta_2 = \beta}I_{g_1,n_1+1,\beta_1} \otimes I_{g_2,n_2+1,\beta_2}}{\to}& H^{\bullet}(\bar M_{g_1,n_1+1}) \otimes H^\bullet (\bar M_{g_2,n_2+1})) }

where the bottom right iso uses the Künneth formula

here Δ a\Delta_a is a homogeneous basis of H (X)H^\bullet(X)

g ab= XΔ aΔ b g_{a b} = \int_X \Delta_a \wedge \Delta_b
Δ=g abΔ aΔ b \Delta = \sum g^{a b} \Delta_a \wedge \Delta_b


g ab=(g ab) 1 g^{a b} = (g_{a b})^{-1}

so this diagram above says that this satisfies the sewing laws that defines a quantum field theory.

There are various other axioms that Gromov-Witten theory must satisfy, but the sewing law above is the most important. The hard part of all of this is constructing the virtual fundamental class, and then proving that this class indeed makes the sewing law and the other axioms (which I have omitted) satisfied.

The above discussion does not yet reveal much of the rich structure of Gromov-Witten invariants. But GW invariants indeed have a very rich and beautiful mathematical structure. Indeed, if we look at just the easiest part of the theory, namely g=0, we are lead to quantum cohomology and Frobenius manifolds.

There is also the mirror symmetry conjecture, which roughly posits that the GW invariants can be found via calculations that a priori seem completely unrelated.


  • M. Kontsevich, Yu. Manin, Gromov-Witten classes, quantum cohomology, and enumerative geometry, Comm. Math. Phys. 164 (1994), no. 3, 525–562 (euclid).

  • Yuri Manin, Frobenius manifolds, quantum cohomology and moduli spaces, Amer. Math. Soc., Providence, RI, 1999.

  • W. Fulton, R. Pandharipande, Notes on stable maps and quantum cohomology, in: Algebraic Geometry- Santa uz 1995 ed. Kollar, Lazersfeld, Morrison. Proc. Symp. Pure Math. 62, 45–96 (1997)

  • J Robbin, D A Salamon, A construction of the Deligne-Mumford orbifold, J. Eur. Math. Soc. (JEMS) 8 (2006), no. 4, 611–699 (arxiv; pdf at JEMS); corrigendum J. Eur. Math. Soc. (JEMS) 9 (2007), no. 4, 901–905 (pdf at JEMS).

  • J Robbin, Y Ruan, D A Salamon, The moduli space of regular stable maps, Math. Z. 259 (2008), no. 3, 525–574 (doi).

  • Martin A. Guest, From quantum cohomology to integrable systems, Oxford Graduate Texts in Mathematics, 15. Oxford University Press, Oxford, 2008. xxx+305 pp.

  • Joachim Kock, Israel Vainsencher, An invitation to quantum cohomology. Kontsevich’s formula for rational plane curves, Progress in Mathematics, 249. Birkhäuser Boston, Inc., Boston, MA, 2007. xiv+159 pp.

  • Sheldon Katz, Enumerative geometry and string theory, Student Math. Library 32. IAS/Park City AMS & IAS 2006. xiv+206 pp.

  • Eleny-Nicoleta Ionel, Thomas H. Parker, Relative Gromov-Witten invariants, Ann. of Math. (2) 157 (2003), no. 1, 45–96 (doi).

Last revised on March 4, 2024 at 23:37:47. See the history of this page for a list of all contributions to it.