Aim of the talk
• Explain “localization formulae".
A “localization formula" is a “compressed”
formula for sums over a large set
(integral points inside a convex polytope) or for some special integrals
on manifolds.
These special integrals are related to action of a symmetry group,
like rotation group on the manifold
or gauge transformations.
Method
• Use of equivariant cohomology as modified de Rham complex (Witten, Berline-Vergne, ∼ 1982) using the symmetry group.
Localization is
a generalization of the fundamental theorem
of calculus.
Inverse problem
• “Decompress" the short formulae
to obtain back the wanted information:
Witten formula for integrals over reduced spaces ∼ 1992.
Compute it by residue methods.
•
Relate discrete and continuous pictures.
Classical mechanics versus quantum mechanics.
Volume versus number of integral points
in convex polytopes.
Some Applications
Conjecture of Guillemin-Sternberg on Quantization of compact classical spaces
Meinrenken-Sjamaar, ∼ 1999,
and Generalization by Paradan ∼ 2003.
Conjecture on Geometric quantization
and transversally elliptic operators
Beautiful mathematical formulae 7→ algorithmic computations
First some examples of what we can compute related to convex polytopes.
Then : geometry,
equivariant cohomology, elliptic operators.
• But ideas occurred in the reverse order:
We: many collaborators
Welleda Baldoni, Matthias Beck, Nicole Berline, Michel
Brion, Charles Cochet, Michel Duflo, Shrawan Kumar,
Jesus de Loera, Paul-Emile Paradan, Andras Szenes.
A beautiful formula.
0 1 3 4 5 6 10000
10000
X
i=0
q
i= 1
1 − q + q
100001 − q
−1. - depends only of the end points.
N
X
i=0
1 = (N + 1).
Left hand side exponentially larger than the right hand side .
(see Barvinok’s talk at this ICM)
Atiyah-Bott fixed point formula.
A
X
i=−A
q
i= q
−A1 − q + q
A1 − q
−1Simple case of Atiyah-Bott fixed point formula.
x
p− p+
Inverse problem: Decompression
Given a short expression for a sum,
compute an individual term of the sum.
Understand its geometric meaning, if it comes from geometry.
In the preceding example,
knowing left hand side, want right hand side q
−A1 − q + q
A1 − q
−1= X
i
c
iq
ithat is, find the coefficients of q
i.
• Answer c
i:= 1 , if −A ≤ i ≤ A :
geometric meaning from Atiyah-Bott theorem:
non zero if and only if
The continuous version: Integrals Continuous version of sums; P
Bi=A
q
i=
1−qqB−1+
1−qqAZ
BA
e
xtdx
the fundamental theorem of calculus
gives a compressed expression for the integral
Z
BA
e
xtdx = e
Btt + e
At−t e
xt=
dxd(e
xt/t) :
apply fundamental theorem of calculus.
Relation between continuous and discrete versions Local Euler Maclaurin formula (Berline-Vergne 2005) P ⊂ R
dconvex polytope with rational vertices,
F face of P
D
Flocal differential operators of infinite order with constant coefficients, such that
h a polynomial function:
X
a∈P∩Zd
h(a) = Z
P
h + X
F;dim(F)<d
Z
F
D
Fh
D
Fcomputed in polynomial time (d, degree h) fixed.
N
X 1 = N + 1
+ 1
.
Work in progress: implement this formula Example: sums of polynomial functions over points in Z
2inside polygons in R
2;
with parameters: Sum of x
481x
482on T ∩ Z
2dilated by the factor N .
N = 11
5; result with 604 digits (expected number of digits).
Proof: prove on basic examples
Cone decompositions in convex geometry
B
X
i=A
q
i= q
A1 − q + q
B1 − q
−1= − q
A−11 − q
−1− q
B+11 − q
χ([A, B ]) = χ([A, ∞[) + χ(] − ∞, B ]) − χ( R )
= χ( R ) − χ(] − ∞, A[) − χ(]B, ∞[).
A B
=
A B−
A BP
∞n=−∞
q
n= 0
Cone decompositions
A
C
B
=
ΓA A
C B
+
ΓB A
C
B
+
ΓC A
C
B
−
ΓBC A C B
−
ΓAC A
C B
−
ΓAB A
C B
+
ΓABC A
C B
Use of Brion theorem and of Barvinok algorithm
Brion: Prove formulae for cones and exponentials.
Barvinok: Signed decomposition of cones;
7→ Theoretical result: P
P∩Zd
h
Polynomial time in the data of P , h when (d, order) fixed.
Cones:=Basic examples.
The geometric analogue will be
tubular neighborhoods of a critical set.
Move on Geometry: Stationary phase M a compact manifold of dimension n , f smooth function on M ,
dm smooth density.
F (t) :=
Z
M
e
itf(m)dm
finite number of non-degenerate critical points:
When t → ∞
F (t) ∼ X
p∈C
e
itf(p)∞
X
k≥0
a
p,kt
−n2 +k. Asymptotically, the integral “localizes"
at a finite number of critical points p .
Example
M = {x
2+ y
2+ z
2= A
2}
x
−A A x
f := x two critical points projecting on A and −A .
Duistermaat-Heckman :Exact stationary phase
F (t) :=
Z
M
e
itxdm = Z
A−A
e
itxdx = e
iAtit + e
−iAt−it Here F (t) is exactly equal
to the first term of the local expression.
Duistermaat-Heckman theorem
M compact symplectic manifold, dimension n , with circle action, f Hamiltonian vector field
dm Liouville measure.
Z
M
e
itfdm = X
p∈C
e
itf(p)a
p,0t
−n2.
Reason: equivariant cohomology.
Definitions
K compact Lie group acting on manifold M . k Lie algebra of K . φ ∈ k ,
vector field V φ : For φ ∈ k , V
xφ := d
dǫ exp(−ǫφ) · x|
ǫ=0.
0
Equivariant complex
Let A(M ) be the algebra of differential forms on M .
d the exterior derivative.
ι(V ) the contraction by a vector field V . Equivariant form: α : k → A(M )
α commutes with action of K .
The equivariant de Rham operator D (Witten (82), Berline-Vergne (82))
D(α)(φ) := d(α(φ)) − ι(V φ)α(φ).
Then D
2= 0 .
Equivariant cohomology
Equivariant cohomology algebra
H
∞( k , M ) = Ker(D)/Im(D ).
If α(φ) polynomial in φ , obtain H
pol( k , M ) : = topological equivariant cohomology (Henri Cartan, Atiyah-Bott)
But need to integrate e
iα(φ)for M non compact:
First application: Fourier transform of the measure on x
2+ y
2− z
2= 1 .
(Harish-Chandra character formulae for
non compact reductive groups)
Equivariant integration R
M
α(φ) well defined if M compact oriented.
( integration top degree term α(φ)[dim M ] ).
This is a function on k .
may be defined as a generalized function on k , when M is not compact:
F (φ) a test function on k ; R
k
α(φ)F (φ)dφ :
differential form on M .
If this differential form is integrable on M , then R
M
α is defined h
Z
M
α, F dφi :=
Z
M
Z
k
α(φ)F (φ)dφ.
Hamiltonian spaces
Examples of equivariantly closed forms arise in Hamiltonian geometry.
M symplectic manifold, symplectic form Ω . Hamiltonian action of K on M :
Moment map µ : M → k
∗for every φ ∈ k ,
d(hφ, µi) = ι(V φ) · Ω.
Noether’s theorem: hφ, µi constant on the orbit of V φ .
Atiyah-Guillemin-Sternberg theorem:
K abelian, M compact.
The image of M by µ is a convex polytope.
(Kirwan polytope if K connected compact)
Example M := P
2( C ) .
|z
1|
2+ |z
2|
2+ |z
3|
2= 1
[z
1, z
2, z
3] ∼ [e
iθz
1, e
iθz
2, e
iθz
3] Action (e
iθ1z
1, e
iθ2z
2, z
3) .
Moment map (|z
1|
2, |z
2|
2) : Image
ij
Homogeneous polynomials of degree n : z
1iz
2jz
3n−(i+j)Basis: integral points in t
1≥ 0, t
2≥ 0, t
1+ t
2≤ n .
Equivariant symplectic form M Hamiltonian manifold.
Ω(φ) := hφ, µi + Ω is a closed equivariant form.
1
(2iπ )
dimM/2Z
M
e
iΩ(φ)Fourier transform of the pushforward
of the Liouville measure under the moment map.
Basic Example 1: R
2with action of S
1Coordinates (x, y ) .
Ω := dx ∧ dy.
V φ := φ(y∂
x− x∂
y).
µ := x
2+ y
22
Equivariant symplectic form is
Ω(φ) = φ( x
2+ y
22 ) + dx ∧ dy.
The moment map µ(x, y) =
x2+y2 2Image of dxdy under the moment map µ :
0
More generally R
2nwith diagonal action Then image of R
2n: Cone,
and image of measure a multivariate spline function.
Multivariate spline functions.
Example:
(Z
1, Z
2, Z
3) ∈ C
3⊕ C
3⊕ C
3e
iθ1Z
1, e
i(θ1+θ2)Z
2, e
iθ2Z
3Moment map (a = |Z
1|
2+ |Z
2|
2, b = |Z
2|
2+ |Z
3|
2) : G(a, b) = b
5(7a
2− 7ab + 2b
2)
14 · 5!
b
G(b, a)
G(a, b)
Basic Example 2: The cotangent bundle T
∗S
1Free action of S
1M := T
∗S
1= S
1× R . Coordinates (θ, t) .
Ω := dt ∧ dθ.
V φ = −φ∂
θ.
µ = t.
The moment map of µ[t, θ] = t
Image of the measure dt ∧ dθ under µ :
0
1 2iπ
Z
T∗S1
e
iΩ(φ)= δ
0(φ).
Fourier transform of the pushforward
of the Liouville measure under the moment map.
Example 3: The sphere
x
−A A x
Ω := dy ∧ dz
x , V φ := φ(y∂
z− z∂
y).
µ := x.
Relation with Duistermaat-Heckman
Ω(φ) = φ x + Ω.
1 2iπ
Z
M
e
iΩ(φ)= Z
M
e
iφ xdm.
1 2iπ
Z
M
e
iΩ(φ)= ( e
iAφiφ + e
−iAφ−iφ ).
Push-forward of Liouville measure K abelian:
Duistermaat-Heckman proved exact stationary phase formula by showing that if
M
red(ξ ) = µ
−1(ξ )/K symplectic reduction, then
Z
M×k
e
−i<ξ,φ>e
iΩ(φ)dφ = vol(M
red(ξ )).
Reduced spaces for R
2and T
∗S
1are points, so vol(M
red(ξ )) = 1 or 0 ,
if ξ belong to µ(M ) or not.
G(a, b) symplectic volume of a Kahler manifold of complex dimension 7 .
Kahler cone of dimension 2 :
Two generalizations
Abelian and "non abelian" localization formulae.
Abelian localization
(Berline-Vergne, Witten, Atiyah-Bott ∼ 1982) S
1acting on compact M
isolated fixed points.
α(φ) closed equivariant form:
(2π)
−dim2MZ
M
α(φ) = X
p∈{fixed points}
i
∗pα(φ)
p det
TpML
p(φ) .
Will explain the reason later.
Witten "non abelian" localization formula ∼ 1992 K compact group, M Hamiltonian:
µ : M → k
∗.
M
red= µ
−1(0)/K symplectic reduction with symplectic form Ω
red.
α(φ) closed equivariant form with polynomial coefficients gives a cohomology class α
redon M
redThen
Z
M×k
e
iΩ(φ)α(φ)dφ = c ∗ Z
µ−1(0)/K
e
iΩredα
redIf α = 1 , Duistermaat theorem.
Equivariant cohomology with generalized coefficients α(φ) distribution with value forms on M :
Z
k
α(φ)F (φ)dφ is a differential form on N .
The operator D is well defined: (Duflo-Kumar-Vergne (1993));
Parallel theory to equivariant index theorem
for Atiyah-Singer transversally elliptic operators
Equivariant index: a distribution on the group.
Localization or 1 = 0
Basic observation ( Paradan ∼ 2000) K compact Lie group acting on M : κ a K -invariant vector field
tangent to the orbits of K and never vanishing:
Theorem
1 = 0
in equivariant cohomology with generalized coefficients.
That is 1 = DB with
B defined with generalized coefficients.
Example
Example C
∗with coordinates [r, θ ] . 1 = D (iY
+(φ)dθ)
0
Indeed:
D (iY
+(φ)dθ)
= iY
+(φ)d(dθ)) − iφY
+(φ)ι(∂
θ)dθ = 1
Applications
M manifold with action of K : κ a K -invariant vector field tangent to the orbits of K :
C = ∪
FC
Fthe set of zeroes of κ ,
divided in connected components;
1 = 0 on M − C :
Reduces the computation on a neighborhood of the
zeroes of κ .
Abelian localization
For an action of S
1, use κ = V φ :
Zeroes of κ : Fixed points of the action.
Integration of closed equivariant form reduced to the basic case R
2(Tangent space to fixed points is a sum of R
2).
Abelian localization formula
Witten non abelian localization M → k
∗Hamiltonian manifold.
k ∼ k
∗κ
m= d
dǫ exp(ǫµ(m)) ∗ m|
ǫ=0Zeroes of κ : Critical set of |µ|
2.
µ = 0
one connected component of the zeroes of κ . Witten idea: use Kirwan vector field:
7→ non abelian localization formula;
reduces essentially double integration on M × k
to integration on T
∗S
1× R
by considering N neighborhood of µ = 0
Tubular neighborhoods of the critical set of kµk
2x2
−A 0 A x
−A A
p− p+
Computations
Interest: Use both localization formulae:
abelian+non abelian; 7→
to obtain computations of intersection pairings on some reduced spaces
(as toric varieties, moduli spaces of flat bundles);
Residue computation theoretical: Jeffrey-Kirwan,
Brion-Vergne, Szenes-Vergne, de Concini-Procesi;
Efficient algorithms for Computations
Volume V (a, b) of the reduced fiber µ : C
9→ R
2V (a, b) = res
x1=0res
x2=0e
ax1e
bx2dx
1dx
2(x
1)
3(x
2)
3(x
1+ x
2)
3. That is integration on
|x
1| = ǫ
1, |x
2| = ǫ
2with ǫ
1< ǫ
2. Reverse the order if a ≤ b .
Efficient (exact) computation of number of points in Network polytopes (Baldoni-de Loera-Vergne)
of Kostant partition functions (Beck-Baldoni-Cochet-Vergne)
Multiplicities for tensor products for representations of
compact groups (Cochet).
Equivariant integration on non compact spaces K compact Lie group acting on M :
κ a K -invariant vector field tangent to the orbits of K .
assume C = zeroes of κ compact
: then as 1 = 0 on M − C , there exists a canonical class P
κ∈ H
−∞( k , M )
with compact support (neighborhood of C ) and equal to 1 (without support condition).
Equivariant integration defined on H
∞( k , M ) by Z
M
α(φ)P
κ(φ).
The result is well defined as a generalized function:
(depending possibly on κ if M is non compact).
The equivariant volume and its quantized version Basic example: T
∗S
1Equivariant volume:
1 2iπ
Z
T∗S1
e
iΩ(φ)= Z
R
e
itφdt = δ
0(φ).
Quantized version of T
∗S
1= L
2(S
1) .
L
2(S
1) = ⊕
∞n=−∞C e
inθ.
Tr
L2(S1)(e
iφ) =
∞
X
n=−∞
e
inφ= δ
1(e
iφ).
Basic example R
2Equivariant volume:
1 2iπ
Z
R2
e
iΩ(φ)=
Z
∞0
e
irφdr == 1
−iφ . Quantization of R
2:
Stone Von-Neumann theorem (1930) :
F ock( C ) = {f ∈ O( C );
Z
C
|f (z )|
2e
−|z|2dxdy < ∞}.
F ock = ⊕
∞n=0C z
nTr
F ock(e
iφ) =
∞
X e
inφ= 1
1 − e
iφ.
Formula for the character of the quantized space M symplectic manifold : classical space.
H (M ) Hilbert space with a representation Still mysterious
( Kirillov for homogeneous Hamiltonian spaces:
The orbit method )
I reinterpreted Kirillov’s formula for the character as an equivariant cohomology formula.
Works for more general cases.
In general ?? I conjecture
an equivariant cohomology formula in some cases
Quantization and transversally elliptic operators Let M be a Hamiltonian manifold.
Symplectic form ω integral.
K compact acting on M in a Hamiltonian way.
µ : M → k
∗moment map.
κ Kirwan Vector field:
assume zeroes of κ compact.
Theorem: Paradan-Berline-Vergne
There exists a (virtual) Hilbert space H (M ) (the index of a transversally elliptic operator), and a representation of K in H (M )
Tr
H(M)(e
iφ) = Z
M
e
iΩ(φ)T odd(φ, M )P
κ(φ)
T odd(φ, M ) is an equivariant cohomology
Quantization and reduction
When M is compact, H (M ) constructed as the index of an elliptic operator:
Guillemin-Sternberg conjecture.
Theorem: Meinrenken-Sjamaar:
The multiplicities of the representations of K in H (M ) are related to the moment map
( dim H (M )
K= dim H (M
red) ).
Qualitative statement:
The irreducible representations of K
are classified by integral orbits of K in k
∗(orbits of highest weight).
Any representation arising in H (M ) is associated to an orbit
contained in µ(M ) ⊂ k
∗.
Proper moment map.
M non compact: H (M ) computed as
the index of a transversally elliptic operator if zeroes of Kirwan vector field is compact.
Conjecture
The conjecture of Guillemin-Sternberg should be true;
That is H (M ) reflects well the geometry of M .
Levels of energy
True for M = T
∗K ,
M = V symplectic space.
0
The spectrum of iφ in H (M ) is
composed of the discrete values of the energy.
Discrete series
True for discrete series (Paradan);
λ