Quantization

P. Deligne

School of Mathematics Institute for Advanced Study Princeton, NJ 08540 e-mail: deligne@math.ias.edu The present notes do not pretend to any originality. We have tried to present di erent aspects of what quantization can mean, including the case of odd variables. Complex polarizations are not considered.

and \small". For this, we need some notion of size 1 on M . For instance M could carry a Riemannian metric, with curvature of size at most 1 and injectivity radius of size at least 1. The symplectic form should be of size 1, and } is a small number. The 2-form which really matters is !=}. Our sign convention for the Poisson bracket corresponding to ! is the following: the Hamiltonian vector eld X (f ) de ned by a function f is given by X (f )g = ff; gg, and df = iX (f )!. If M = R2 with coordinates p, q and if ! = dp^dq, then fp; qg = 1. A quantization of M consists in a complex Hilbert space H and in a rule to attach to functions f on M operators f ^ acting on H. The rule f 7! f ^ should be C -linear, for f real, f ^ should be hermitian and, in a sense I will not try to make precise, f 7! f ^ should almost be an homomorphism: 1^ should be the identity and for slowly varying functions f and g, (1.1) (fg)^

1. Let (M; !) be a symplectic manifold of dimension 2d. We will use vague words \big"

f ^ g^
1

mod O(}):

If (1:1) did hold exactly, the f ^ would commute the one with the other. The symplectic structure controls the failure of commutativity (1.2) (1:2)0

f ^ ; g^]
(X (f )g)^

i}ff; gg^ i=} f ^ ; g^]

mod O(}2 ):

Formula (1.2) can be rewritten mod O(}):

The symplectic vector eld X (f ), if it is well behaved at in nity, exponentiates to an automorphism exp(X (f )) of (M; !), with Taylor's formula reading exp(X (f )) g = exp(derivation X (f ))(g): Formula (1:2)0 should hold in its integrated form (1.3) (exp(X (f )) g)^ exp(i=}f ^ )g^ exp( i=}f ^ ) mod O(}): We don't care about the exact rule f 7! f ^. It matters mainly mod O(}). In the deformation quantization point of view, (1:1) is the rst term in an expansion ( -product) (1:1)0

f ^ g^ = (fg)^ + c1(f; g)^} + c2(f; g)^}2 +

;

with ci(f; g) given by a C -bilinear di erential operator in f and g. The formula (1.2) becomes c1(f; g) c1(g; f ) = iff; gg. If one disposes of quantizations H(}),with } tending to 0, (1; 1)0 can be an asymptotic expansion in }. if one rede nes f 7! f ^ to be f ^ + d1(f )^ } + with di (f ) a linear di erential operator in f , (1:1)0 changes to an equivalent -product. Assumption (1:1) means that H almost localizes on M : an element h of H can be P decomposed as a sum hi of elements localized in small regions Ui of M : if x 2 Ui , f ^ hi is about f (x):hi . Because of the non commutativity (1:2), localization does not make sense for regions smaller than expressed by the uncertainty principle p q 2 }. If we were only to assume (1:1) and (1:2), a multiple Hn of a quantization H of M would again be a quantization. A \ nite multiplicity" assumption is that if f is a C 1
2

function with compact support on M , the operator f ^ is of trace class. If this holds, I would expect that for some integer m (rather: one for each connected component of M ), one has Z ^ Tr(f ) m f (x)dx where, in Darboux local coordinates, dx is the product of the dpi dqi =2 sure). We want m = 1: (1.4) Tr(f )
^

}

(Liouville mea-

Z

f (x)dx

The basic example of almost localization, in the sense meant above, is when M is a cotangent bundle T V and H the space of half densities on V . If f is peaked at x, say in a gaussian fashion, and if g is a real function, with dg slowly varying, then feig=} is localized around (x; dgx) in M . If f is slowly varying, then f:eig=} is localized around the section x 7! dgx of M ! V , above the support of f . The operator f ^ has a natural de nition when f is a ne linear on the bers of T V over V . If f is the pull back of a function on V , f ^ is multiplication by f . If f is linear on each cotangent space, hence identi ed with a vector eld F on V , f ^ = i}LF . For those f , (1.2) holds exactly. To de ne f ^ for more complicated functions requires auxiliary choices. On T V , we have a canonical 1-form : in local coordinates qi on V , giving local P coordinates (pi ; qi) on T V , is pi dqi. The trivial unitary line bundle L, with the connection i =}: rf = df i =}f , is a prequantization line bundle: its curvature is i!=}, for ! the symplectic form d . A real function g on V de nes the lagrangian section L(g): x 7! (dg)x of T V . The pull back to L(g) of eig=} is an horizontal section of L. For any symplectic manifold M , let M or simply be the ber space over M whose ber m at m 2 M is the space of lagrangian (= maximal isotropic) subspace of the tangent space Tm of M at m. To take into account the Maslov index story, it is best to take prequantization line bundles as living on . They should verify (a) (b) below. (a) The curvature 2-form R is the pull back from M of i!=}:

2. Example: cotangent bundles.

r2 = i!=}:
3

In particular, on each ber m of ! M , (L; r) is at. One has 1 ( m ) = Z and (b) the monodromy of (Lr) on m is multiplication by i.
Remarks: (i) In the same way that an hamiltonian vector eld X (f ) acts on a prequantization line bundle, with the action depending on f , and not only on X (f ), it also acts on (L; r) as above: X (f ) is symplectic, hence lifts to , the question is local on , and locally on (L; r) is the inverse image of a prequantization line bundle on M . If we exponentiate, we see that exp(X (f )) acts by an automorphism of (M; !; L; r). The in nitesimal action on sections of L is given by

(2.1)

@X (f ) = rX (f ) if=}

(ii) To make sense of (b), we should make precise our sign convention identifying 1 ( m ) with Z. For the symplectic vector space R2, form dp^dq, the generator is given by the path

7 ! line spanned by (q; p) = (cos ; sin )

(0

):

(iii) Let V be a symplectic vector space, and 0 be the space of lagrangian linear subspaces of V . If S1 and S2 in 0 intersect transversally, one de nes as follows a preferred homotopy _ class of path from S1 to S2. Identify S2 to S1 by s2 7! !(s2 ; s1 ). Choose a basis ei of S1 and let e01 be the dual basis of S2. The path is

7 ! subspace spanned by the cos ei + sin e0i

(0

=2):

It depends only on the quadratic form for which the basis (ei ) is orthonormal. The choice of this quadratic form running over a contractible set, the homotopy class of the path does not depend on the choice. If V is of dimension 2d, the preferred path from S1 to S2, followed by the preferred path from S2 to S1, is d times the preferred generator of 1( 0 ). We now come back to the case where M is a cotangent bundle T V . Let 0 be the subbundle, with 0 m consisting of those lagrangian subspaces L of Tm intersecting m transversally the vertical subspace of Tm (tangent to the ber of T V ! V ). If we pull back the prequantization line bundle on T V to 0, it extends uniquely to L on satisfying the conditions (a) (b).
4

Let L be a lagrangian subvariety of M . Let s be the section of over L: x 7! tangent space of L at x. The section s is with values in 0 if and only if L is locally of the form L(g). As L is lagrangian, s L is at and it makes sense ot speak of a slowly varying section of s L on L. Generalizing the f:eig=} considered previously, to a slowly varying half density u on L, with values in s L, corresponds u] in H, with (2.2)

k u]k2

Z

S

hu; ui :

It is localized near Supp(u) S M = T V . A function f de nes an automorphism exp(X (f ))of (M; L; r) (Remark (ii) above), and an automorphism exp( i=}f ^) of H. Those automorphisms should (almost) preserve the construction u 7! u]: (2.3) exp(X (f )) u] exp(i=}f ^) u]:

Inner products are given as follows. Fix (Si; ui ) (i = 1; 2) as above, and assume that S1 and S2 intersect transversally. Let x be an intersection point. The lagrangian subspaces s1 (x) and s2(x) of TxM intersect transversally, and the preferred class of path from s1 (x) to s2 (x) (Remark (iii) above) gives an isomorphism : Ls2(x) ! Ls1(x). The symplectic structure puts s1(x) and s2 (x) in duality, identifying the line of half densities of S1 at x with the dual of the similar line for S2. The inner product hu1(x); u2(x)i is hence just a number. The stationary phase gives (2.4)

h u1]; u2]i

X

x2S1 \S2

(2 i})d=2 hu1(x); u2(x)i

for M of dimension 2d. Our convention for inner products ha; bi is: antilinear in a, linear in b. In the at case, one can de ne a quantization as the data of f 7! f ^, just for f a linear function, with (1.2) holding exactly. The purely imaginary linear functions form a Lie algebra, for i}ff; gg, and we want a unitary representation f 7! f ^ of this Lie aglebra, with i^ = i:Id. It is more convenient to ask for a unitary representation of the corresponding
5

3. The at case is when M is an a ne space, with a translation invariant symplectic form.

Lie group. Formula (1.4) becomes a request for irreducibility. Weyl's quantization f 7! f ^, for a general f , is given by (3.1) exp(`)^ = exp(`^) for ` a purely imaginary linear function, a generalf ^ being deduced from (3.1) by considering f as a superposition of exp(`), by Fourier transform. For Weyl's quantization, formula (1.4) holds exactly. Understood in those terms, quantization in the at case is unique, up to isomorphisms. If M is the cotangent bundle to an a ne space V , 2. gives its Schrodinger model as L2(V ): if V0 is the vector space of translations of V , one has M = V0_ V ; for f (the pull back of) a function on V , f ^ is multiplication by f . For ` a linear form on V0_ , identi ed with a vector v 2 V0, `^ is i}@v , and exp(i`)^ is (x) 7! (x + }v). This unicity assertion depends on the assumption that the at symplectic variety M is of nite dimension.
Remark. In the at case, the data on M of a prequantization line bundle de nes a quantization H, as de ned above, up to unique isomorphism. This is made plausible by the fact that the group of automorphisms of

(a ne symplectic space M; L ; r) and of (a ne symplectic space M; H; f 7! f ^) are the same extension of the a ne symplectic group (Sp n translations) by U 1 . The relation between L and the quantization H can be xed as follows. Let H1 be the C 1-vectors in H, for the action of the Heisenberg group. In the Schroedinger model: the Schwartz functions on V . Let H 1 H be the dual of H1. Then, a linear lagrangian subspace L of M , given with u on L of the form (translation invariant half density). ( at section s L ), de ne u] in H 1, and the inner product formula (2.4) holds exactly, to give the inner product of continuous superpositions of u]. Formula (2.3) holds exactly, for f quadratic and u as above. In particular, if a linear form ` vanishes on L, one has `^ u] = 0.
6

4. We now go over to the superworld. A super Hilbert space is a mod 2 graded complex

vector space H , with an even sesquilinear form (u; v), antilinear in u and linear in v, for which (v; u) = ( 1)p(u)p(v)(u; v) ; with a positivity and a completeness condition. As positivity condition, we take (u; u) > 0 for u even; u 6= 0 (4.1) i(u; u) > 0 for u odd; u 6= 0: This positivity condition is stable by tensor product, if ( ; ) for H 0 H 00 is de ned by (u0 u00 ; v0 v00 ) = ( 1)p(u00 )p(v0)(u0 ; v0 )(u00 ; v00 ); in accordance with the sign rule. The adjoint T y of an operator T is de ned by (4.2) (Tu; v) = ( 1)p(T )p(u)(u; T y v): The physicist don't like (4.1), and prefer to work with the ordinary mod 2 graded Hilbert space with the inner product h ; i de ned by (u; v) = hu; vi for u; v even (4.3) (u; v) = i hu; vi for u; v odd: The adjoint T y is related to the corresponding ordinary adjoint T by T y = T for T even (4.4) T y = iT for T odd: The de nition (4.2) has the unsettling e ect that the eigenvalues of an odd self adjoint operator are in i1=2 :R. However, the eigen spaces being neither even nor odd, there is not much wish to consider them. In a quantization of a supermanifold M , to functions f on M should correspond operators f ^, acting on a super Hilbert space H, with f ^ of the same parity as f . If f is real, one should have f ^ = f ^y . Because of (3.4), if M is a supermanifold, the physicists will declare to be \real" the real even f , and the i 1=2 for real. With this rule, if 1 and 2 are odd and \real", i 1 2 is \real" (as well as real, being even). Example: on R1;1, with coordinates (x; ), if we put = i 1=2 , the vector eld @ + i @x is a multiple of a real vector eld: @ + i @x = i1=2 (@ + @x):
7

5. Let now M be a super a ne space, with a translation invariant symplectic form. In

imitation of 3., we de ne a quantization as the data of a super Hilbert space H, and of ` 7! `^ attaching to a linear function on M an operator on H, with ` and `^ of the same parity. The quantization ` 7! `^ should be linear, real: `^ = (`^)y for ` real, map 1 to Id, and obey (5.1)

`^; m^] = i}ff; gg^:

One should also require some irreducibility. The condition (5.1) requires that for real linear odd functions `, the Poisson bracket f`; `g (a constant) be negative: (5.2)

f`; `g 0:

Indeed, `^; `^] = 2`^2 = `^y `^ must be i}f`; `g. The sign in (5.2) is due to the choices of sign in (4.1) and (5.1), the latter repeating (1.2). As in 3., one extends ` 7! `^ to polynomial functions by requiring that for a product of even or odd linear functions, one has (`1 : : : `n)^ = 1=n!
X

`^(1) : : : `^(n)

( 2 Sn)

with given by the sign rule. In particular, is + if the `i are all even, and "( ) if they are all odd.

6. Let W be a real vector space with a positive de ned symmetric bilinear form B. Let
f`; mg = B 1 (`; m)

M be the a ne space de ned by W : it is of dimension (0; dim W ), and linear forms on W are odd functions on M . De ne a Poisson bracket by
(6.1)

for `; m 2 W _. If ei is an orthonormal basis of W , with dual basis ei, and if ei is veiwed as an odd function on M , the Poisson bracket (6.1) is given by the super symplectic form (6.2)

!= 1 2

X

dei dei :

8

1 One W , let Q be the quadratic form 2 B 1 (`; `). If H is a mod 2 graded module over the (mod 2 graded) Cli ord algebra C (W ; Q), with the module structure written c`(`) for ` 2 W , (5.1) holds for

(6.3)

`^ = (i})1=2 c`(`):
p

p For dim W = 1, one can take H = (C even) (C odd), with c`(e1) = 1=0 2 1=0 2 . For the obvious Hilbert space structure of H, and the corresponding super Hilbert space structure, one has then c`(e1 ) = c`(e1) and for any ` in V ,

(6.4)

`^ = (`^)y :

A general W can be written as an orthogonal direct sum of lines, and the corresponding tensor product of super Hilbert spaces is a Cli ord module for which (6.4) holds, if `^ is de ned by (6.3). If H1 is a graded submodule with the induced Hilbert space structure, (6.4) continues to hold for H1. This proves the existence of a super Hilbert space H with a Cli ord module structure, irreducible as a graded module, for which (6.4) holds when `^ is de ned by (6.3). We now assume W of even dimension. They are the two isomorphy classes of H as above, exchanged by the parity change H 7! H L, for L a (0; 1)-dimensional super Hilbert space. Let (e1 ; : : : ; en) be an orthonormal basis (for B 1 ), and e1; : : : ; en] be the corresponding density on M . Another orthonormal basis would give the same, or the opposite density. The choice of the density e1; : : : ; en] picks out one of the two isomorphy classes of H: the one for which the following analogue of (1.4) holds. For any f on M , the supertrace Tr(f ^) is given by (6.5) Tr(f ^) = }n=2
Z

e1; : : : ; en ]f

Justi cation: writing H as a tensor product, one reduces to the case n = 2. For n = 2, 1 1 2 c`(e1 )c`(e2 ) has square 1=4 and eigenvalues i=2. It follows that (e1e2 )^ = 2 (e^e^ e^e^) = e^e^ = i}c`(e1 )c`(e2 ) has eigenvalues }=2, hence supertrace }. 2 1 1 2
9

even and M odd, as in 6. The tensor product of quantizations of M + and M gives one for M . If M + is the cotangent bundle of an a ne space V , and if M is obtained as in 6. from an even dimensional quadratic vector space W , this tensor product can be realized as the space of L2-density on V with values in a Cli ord module for W _, and an analogue of 1.4, 6.5 holds.

7. A super symplectic at space M as in 5. can be decomposed as M + M , with M +

8. Let V be a manifold and W be an orthogonal vector bundle on V , with a connection r

respecting the structural symmetric bilinear form B. We take M to be the ber product over V of T V and W . Over any basis B, a section of M over V is the data of an even section p of T V and of an odd section of W . A section e of W _ de nes an odd function 7! e( ) on M . A local coordinate system (qi ) on V and a local basis (e ) of W give a local coordinate system (qi ; pi ; ) on M . We assume that W is even dimensional, oriented and Spin. It gives rise to a bundle of super Hilbert spaces HV on V , reproducing point by point what we got in 6., and the connection r on W gives one on HV . Let H be the super Hilbert space of half densities on V with values in HV . To any function f on M , of degree 1 on the bers of M=V , when pi is viewed as of degree 1 and as of degree 1=2, one attaches an operator f ^ as follows. (a) For (the pull back of) a function f on V , f ^ is multiplication by f . (b) For the odd function de ned by a section e of W _,

e^ = (i})1=2 c`(e)
as in 6.3. This is extended as in 5. to de ne f ^ for any function f on W . (c) If f is a function on T V , linear on each ber of T V=V and identi ed with a vector eld F , f ^ = i}rF : In (c), if x in H is the product of a section h of HV by a half density v, rF (x) := rF (h):v + h:LF (v).
10

The super vector space of operators f ^, for f of degree 1 on the bers of M=V , is stable by bracket. By (1.2), the bracket of operators corresponds to a Poisson bracket, which we now compute. For functions on W , it is as in 6.: functions on V commute with functions on W and for `, m odd linear functions, identi ed with sections of W _, one has

f`; mg = B 1(`; m):
For f a function on T V , linear on the bers and corresponding to a vector eld F on V , ff; g is rF on functions on W . It remains to compute f ^; g^] as being of the form i}ff; gg^ when f and g are linear functions on T V , corresponding to vector elds F and G. We have

f ^; g^] = i}rF ; i}rG] = ( i})2 r F;G] + R(F; G) = i} ff; gg^ V i}R(F; G) T
for f ; gT V the Poisson bracket on T V and F the curvature 2-form of W , with values in SO(W ) (which acts on HV ). Let S be the matrix of S 2 SO(W ), relatively to a local basis e of W : S is (x ) 7! (S x ). Let e be the dual basis, and move the indices up and down using B. The action of S on HV is then the Cli ord multiplication by S e e (sum on and ) in the Cli ord algebra. This follows from

S e e ; e ] = S (e e ; e ] e ; e ]e ) (superbrackets) 1 = 2 S (e b ; b ; e ) = S b e = S e = S (e )
for the Lie algbra action. If we apply this to the curvature, and if are the odd functions on M corresponding to the e , this gives f ^; g^] = i}ff; gg^ for

ff; gg = ff; ggT

V

R(F; G)

;

This Poisson bracket corresponds to the following symplectic 2-form . Let T V be the canonical 1-form pdq on T V , as well as its pull back to M . Let W be the 1-form
11

on W whose inverse image by a section is 1 ( ; r ). If (e ) is a local basis of W , and 2 if the connection is given by the endomorphism valued 1-form :

r(x e ) = (dx +
one has With those notations,
W=1 2

x )e ;
):

(d

+

=d T V +d V:
X X

If (e ) is an orthonormal basis, one has

d

1 V =2

(d

+

)2 +

;

R

:

12