How to describe a momentpolytope using a line bundle
Timothy Goldberg
Friday, February 15, 2008Lie Groups SeminarCornell University
1
AbstractGiven a Hamiltonian group action, proving that a momentimage is a convex polytope is sometimes much easier thanfiguring out exactly what the polytope is. But in certaincases, where the manifold is Kahler and admits a compati-ble holomorphic positive line bundle L, the sections of L cansometimes lead to descriptions of these moment polytopes.I will discuss several results about this, proved by M. Brion,V. Guillemin and R. Sjamaar, L. O’Shea and R. Sjamaar, P.Schutzdeller, and myself.
2
Starring:
(M,ω) compact, connected Kahler manifold
L holomorphic, Hermitian line bundle over M
G compact, connected Lie group
GC complexification of G
Assume:
ω = 12πi
curv(∇).
G,GC act holomorphically on (M,L) by bundleautomorphisms preserving ∇.
3
Where’s the moment map?
∀ξ ∈ g, have two operators on C∞(M,L): ∇(ξM),L(ξ)
Fact: (B. Kostant, 1970)L(ξ) −∇(ξM) is multiplication by an iR-valued function.
Define φ : g → C∞(M) by
φ(ξ) :=1
2πi(L(ξ) −∇(ξM)) .
Can show φ is G-equivariant and d (φ(ξ)) = ω (ξM, ·).
Defines a moment map!
Φ : M → g∗, Φ(x)ξ = φ(ξ)(x)
4
Also starring:
X ⊂M closed, irreducible, analytic subvariety
T ⊂ G maximal torus
t∗+ ⊂ t∗ closed positive Weyl chamber
Λ ⊂ t∗ weight lattice
Λ+ ⊂ t∗+ dominant weight set
ΛQ ⊂ t∗ rational points, ΛQ := Λ⊗Q
B ⊂ GC Borel subgroup corresponding to t∗+
N = [B, B] maximal unipotent subgroup
Example:
G = SU(n), GC = GL(n; C),T = {diagonal elements of G},B = {upper triangular elements of GC},N = {elements of B with 1’s on the diagonal}
5
Kodaira Embedding Theorem: (M,L) is ample.
⇒ ∃ embedding M ↪→ some CPN(holomorphic but not usually Kahler)
So we are really working with algebraic things,not just analytic.
6
The torus case
Moment map for T y M:
ΦT : MΦ // g∗ // // t∗ .
Theorem: (Atiyah, 1982)If X is preserved by TC, then ΦT(X) is the convex hull ofΦT(X
T).
7
T y Γ(M,L), so
Γ(M,L) =⊕λ∈Λ
Γ(M,L)λ, weight space decomposition.
(Note that dimC Γ(M,L) < ∞.)
Alternative Statement of Atiyah’s Theorem:Let j : X ↪→ M be inclusion. If X is preserved by TC, thenΦT(X) is the convex hull of
{λ ∈ Λ | ∃s ∈ Γ(M,L)λ, j∗s 6= 0} .
8
The general case
Definition: The highest weight polytope C(X) of X con-tains λ ∈ ΛQ such that:
• ∃r > 0 with rλ ∈ Λ+, and
• ∃s ∈ Γ(M,Lr)rλ, N-invariant with j∗s 6= 0.
I.e. rλ is a highest weight for the G-representation spaceΓ(M,Lr), and ∃ an eigensection for rλ not vanishingidentically on X.
Put ∆(X) := Φ(X) ∩ t∗+.
Theorem: (M. Brion, 1986)If X is GC-invariant, then C(X) = ∆(X) ∩ΛQ andC(X) = ∆(X).
9
Theorem: (V. Guillemin and R. Sjamaar, 2006)Even if X is only B-invariant, then Brion’s result still holds.
10
What does Γ(M,L) have to do with Φ?
Magic Formula:
Suppose h ⊂ gC = complex Lie subalgebra,s ∈ Γ(M,L) transforms under h by a character χ : h → C.
Then ∀ξ ∈ h,
L(ξM)‖s‖2 = (2Reχ(ξ) + 4πφ(Im ξ)) ‖s‖2.
(For (M,Lr), just replace φ with rφ.)
Example:Let h = gC, χ = 0. Then ∀ξ ∈ g,
L ((iξ)M) ‖s‖2 = 4πφ(ξ)‖s‖2.
11
SemistabilityDefinition: x ∈M is:
• algebraically semistable if ∃ r > 0 and s ∈ Γ(M,Lr)Gwith s(x) 6= 0.
• analytically semistable if 0 ∈ Φ(GC · x
).
Theorem:algebraic semistability ⇐⇒ analytic semistability
(⇒) follows from Magic Formula.(⇐) is hard. (V. Guillemin and S. Sternberg in 1982, F.Kirwan in 1984, L. Ness in 1984, R. Sjamaar in 1995.)
From this we can prove:
0 ∈ C(X) ⇐⇒ 0 ∈ Φ(X).
12
The shifting trickGiven λ0 ∈ Λ+, want to construct (M ′, ω ′), L ′, X ′, andΦ ′ : M ′ → g∗ such that
• 0 ∈ ∆(X) ⇐⇒ λ0 ∈ ∆(X ′) and
• 0 ∈ C(X) ⇐⇒ λ0 ∈ C(X ′).
Choose λ0 ∈ Λ+.
Gλ0= stabilizer of λ0 w.r.t. coadjoint action G y g∗
Pλ0= parabolic subgroup of GC corresponding to λ0
Mλ0:= GC/Pλ0
(flag variety)≈ G/Gλ0
(homogenous space)≈ G · λ0 (coadjoint orbit)
L−λ0:= (GC × C−λ0
) /Pλ0
Symplecticω ′ comes from G ·λ0, andΦλ0: Mλ0
→ g∗ is thenegative of the inclusion
Mλ0≈ G · λ0 ↪→ g∗.
13
Borel-Weil Theorem:Γ(Mλ0
, L−λ0) ∼= V(λ0)
∗ := dual of irreducible representationof G with highest weight λ0.
Put
M ′ = M×Mλ0, L ′ = L� L−λ0
, X ′ = X×Mλ0,
ω ′ = ω+ωλ0, Φ ′ = Φ+Φλ0
,
and let G y M ′ diagonally.
Γ(M ′, L ′)G ∼= (Γ(M,L)⊗ Γ(Mλ0, L−λ0
))G
∼= (Γ(M,L)⊗ V(λ0)∗)G
∼= Hom (V(λ0), Γ(M,L))G
∼= Γ(M,L)Nλ0.
Φ(x) = λ0 ⇐⇒Φ ′(x, λ) = Φ(x) +Φλ0
(λ0) = Φ(x) − λ0 = 0.
14
Real resultsSuppose
• (τ, β) y (M,L) and σ y GC are antiholomorphicinvolutions,
• β preserves ∇,
• σ preserves G, T , and B.
σ y G σ y g, g∗
Compatibility with Hamiltonian action:
• Distibution: ∀g ∈ G, x ∈M, τ(g · x) = σ(g) · τ(x).
• Anti-equivariance: ∀x ∈M, Φ (τ(x)) = −σ (Φ(x)).
σ, τ ! complex conjugationMτ, Gσ ! “real parts” of M, G.
15
Note:
x ∈Mτ ⇒ Φ(x) = Φ (τ(x)) = −σ (Φ(x))⇒ σ (Φ(x)) = −Φ(x)
SoΦ (Mτ) ⊂ g∗−1.
Theorem: (L. O’Shea and R. Sjamaar, 2000)Suppose X is preserved by GC and τ, and Xτ contains asmooth point. Then
∆(Xτ) = ∆(X) ∩ g∗−1.
16
Theorem: (G., 2007) O’Shea and Sjamaar’s theorem holdseven if X is only preserved by B and τ.
Mantra:The real part of the moment polytope is (ought to be) themoment polytope of the real part.
17
Strategy of proofs:
1. Show ∆(Xτ) ∩ΛQ = C(Xτ) ∩ g∗−1.
(Involves variant on the shifting trick.)
2. Show C(Xτ) ∩ g∗−1 = C(X) ∩ g∗−1.
(Xτ contains a Lagrangian submanifold,⇒ Xτ Zariski dense in X.)
3. Use previous results about C(X) and ∆(X).
Shifting trick variant:Need involution α y Mλ0
≈ G · λ0.Since this λ0 is in g∗−1, can define α := −σ y G · λ0.
Proposition: (G · λ0)α = Gσ · λ0.
18