• Tidak ada hasil yang ditemukan

Representations and Line Bundles

We have seen that the splitting principle allows us to regard the Chern classes of a vector bundle E as (symmetric) polynomials in the first Chern classes of the line bundles of a flag bundle associated to E. Given an m-dimensional vector space V, the spaceF `(V) of all complete flags on V can be identified with GL(V)/P, whereP is the group of upper triangular matrices. (This follows because GL(V) is transitive on the flags and P, the stabilizer of the standard flag 0 ⊂ he1i ⊂ he1, e2i ⊂ · · · ⊂ he1, . . . , emi=V, is isomorphic to the stabilizer of any given flag.) We can associate any one-dimensional representation χ : P → C to an equivariant line bundle over the flag manifold F `(V) as follows:

L(χ) = GL(V)×C/((gp, z)∼(g, χ(p)z)) (8.14) for g ∈ GL(V), p ∈ P, and z ∈ C. The projection of L(χ) onto F `(V) is just (g, z)→π (gP). Under the action of GL(V) given by h(gp, z) = (hgp, z), the following diagram commutes:

L(χ) −−−−→GL(V) L(χ)

π

 y

 yπ F `(V) −−−−→GL(V) F `(V),

since (hgp, z) = (hg, χ(p)z). The line bundleL(χ) is thus equivariant with respect to the bundle projection.

Conversely, suppose L is an equivariant line bundle over F `(V). Then P acts on the fiber over eP, so this fiber is a one-dimensional representation χ of P. Let us show that the line bundle L(χ) corresponding to this representation is isomorphic to L. Lety∈Llie in the fiber overeP. Then we claim that the mapr :L(χ)→Lgiven by r(g, z) = z(g·y) is an isomorphism. We have r(gp, z) = z(gp·y) = z(gχ(p)y) = r(g, χ(p)z), so this map is well-defined. Since G acts transitively on the fibers of L, and multiplication by z is a surjective map on any given fiber, r is surjective. For injectivity, suppose that z1(g1 ·y) = z2(g2 ·y). If z1 6= 0, then y = z11z2g11g2 ·y, so g−11 g2 ∈ P. This means that g2 = g1p for some p ∈ P and z−11 z2χ(p) = 1, so z2 = z1χ(p)−1. Thus, as elements of L(χ), (g2, z2) = (g1p, χ(p)−1z1) = (g1, z1), so r is indeed injective. The correspondence between line bundles and one-dimensional representations of P is therefore a bijection.

We can identify the characters χwith line bundles on a flag manifoldF `(V) more explicitly. Consider the tautological filtration[41]

0 =U0 ⊂U1 ⊂U2 ⊂ · · · ⊂Um =F `(V)×V (8.15) of vector bundles overF `(V), whereF `(V)×V is the product bundle, and Uk is the k-dimensional bundle over F `(V) whose fiber over a flag V1 ⊂ · · · ⊂ Vm is Vk. It follows from the splitting principle that the cohomology ringH(F `(V)) is generated by the first Chern classes of the line bundles Li =Ui/Ui−1, setting xi =c1(L1). The identity matrix fixes the standard flag {e1, . . . , em}. Therefore, over eP, the fiber of Li is Vi/Vi−1, where Vi = he1, . . . , eii. If v = Pi

k=1αiei ∈ Vi and p ∈ P, then p·v =w+piiei, wherew∈Vi1 andpiiis the ith diagonal entry ofp. We have shown

the following.

Theorem 8.4.1 If Li is the line bundle over a flag manifold defined as above, then the character χ associated to Li is the map taking p to pii.

Let us adapt this machinery to the problem at hand. Recall that we have two complex vector spaces Aand B of dimensionsdAand dB, respectively, together with a map φ: Grk(A)→GrkdB(A⊗B) given byφ(V) =V ⊗B. We wish to compute the action of the induced map φ :H(GrkdB(A⊗B))→H(Grk(A)).

LetQAandQABbe the quotient bundles of Theorem 8.2.2 over the Grassmannians Grk(A) and GrkdB(A⊗B) respectively. The Chern classes of these bundles are the classes of the special Schubert varieties in the cohomology rings. By the splitting principle, the associated flag bundles Fl(A) and Fl(A ⊗ B) have pullbacks which split as a direct sum of line bundles Li of the respective tautological filtrations. The cohomology of the Grassmannians embeds in the cohomology of these pullbacks, so we may determine φ by its action on the Chern classes of the pullback bundle of Fl(A⊗B).

It follows from the definition of pullback bundles that the bundle φ(L(χi)) is the set of triples (gPA, φ(g), z) ∈ GL(A)/PA ×GL(A⊗B)×C with the identification (gPA, φ(g ·p), z) ∼ (gPA, φ(g), χ(φ(p))z). This means that φ(L(χi)) = L(φi)).

And the pullback of the map induced by φ on the characters of the group PAB is readily computed: for a matrix X ∈ PA, and the character χi taking a matrix to its ith diagonal entry, we have φi)(X) = χi(φ(X)) = χi(X ⊗I) = χdi/dBe(X). So φi) = χdi/dBe. Now we can calculate the action ofφ on the Chern classes:

Theorem 8.4.2 φ(xi) = c1(L(χdi/dBe)).

Proof

φ(xi) = φ(c1(L(χi))) (8.16)

= c1(L(χi))) (8.17)

= c1(L(φi))) (8.18)

= c1(L(χdi/dBe)). (8.19)

2

Chapter 9

Determining the Inequalities

In this chapter we use our knowledge of howφbehaves to explicitly derive inequalities relating the spectra ofρAB and ofρA. We work out some examples in low dimensions.

We also restate how to obtain the inequalities in the language of representation theory.

We discuss recent progress in symplectic geometry that shows that the inequalities derived using our method are sufficient. Finally, we prove that ifdB12d2A, then the inequalities simplify greatly.

9.1 Putting It All Together

Let ρA = TrBρAB, and let λ, µ, and ˜λ denote the spectra of ρAB, −ρAB, and ρA, respectively. Theorem 6.2.2 can be interpreted cohomologically as saying that if

φπ)∪σ˜ν 6= 0, (9.1)

where σπ ∈ H(Gr(kdB, dAdB)) and ˜σν ∈ H(Gr(k, da)) are Schubert classes, then the spectra µ and ˜λ must satisfy the inequalities

Xν(i)˜λi+X

π(i)µi ≤0. (9.2)

Now φπ) is an integer combination of Schubert classes, φπ) = X

i

niσ˜πi. (9.3)

For each of these classes, ˜σπi ∪σ˜ν 6= 0 iff ν contains the complement of πi in the k×(n−k) rectangle. But if we consider the case whereν is in fact the complement ofπi, then we see that the Inequalities 9.2 are the strongest in this case; for any other ν0 ⊃ν, the inequalities determined by ν0 are implied by the inequalities determined byν. So it is sufficient to consider complements of each Schubert class ˜σπi contained in φπ), in order to obtain the inequalities relating −ρAB and ρA. Now if µ is the spectrum of −ρAB, then the spectrum λ of ρAB is given by λi =−µdA−i+1 (since the ordering of the eigenvalues is reversed). Given binary strings π,πˆ ∈ dAkdB

satisfying ˆ

π(i) =π(dAdB−i+ 1), so that ˆπ is simply the string π in reverse, the Schubert cell Sˆπ corresponds to the complementary partition to that of Sπ. This means that we obtain inequalities

Xν(i)˜λi ≤X

π(i)λi (9.4)

whenever φˆπ) contains σˆν (where ˆν is the complementary partition to ν) as a summand. It then follows that Inequalities 9.4 are obtained wheneverφπ) contains σν as a summand.

Theorem 8.2.2 says that the lth Chern class cl(Q) of the universal quotient bun- dle Q over the Grassmannian Gr(k, n) is equal to the special Schubert class σl ∈ H(Gr(k, n)). And the splitting principle allows us to conclude that

cl(Q) =el(x1, . . . , xnk), (9.5) where xi =c1(Li) is the first Chern class of theith split component of f(Q), andel is the lth elementary symmetric polynomial. Because the special Schubert classesσl generate the cohomology ring, we therefore have a surjective ring homomorphism

ψ˜: Λn−k →H(Gr(k, n)) el(x1, . . . xn−k) 7→σl.

We may compose the map ˜ψ with the involution ω : Λnk → Λnk, ω(ek) = hk, to

obtain a map

ψ : Λn−k →H(Gr(k, n)) hl(x1, . . . xn−k) 7→σl.

Now, by the Pieri rule, it follows that for any partition λ, ψ(sλ(x1, . . . , xn−k)) =σλ. Thus, we may determine how φ acts on H(Gr(kdB, dAdB)) by determining how the map xi 7→xdi/dBe acts on Schur functions.