for Pσ∨ ={Pn
i=1λimσ,i |0≤λi <1}. In particular, if σ is smooth, then S( ˜χ(Uσ,L)) = χmσ
Qn
i=1(1−χmσ,i).
Proof. Completeness of X forces to ˜χ(X,L) be in Z[M], and from Lemma 7.2.1 shows that
S( ˜χ(X,L)) = ˜χ(X,L)) =X
p≥0
(−1)p X
γ∈[l]p
S( ˜χ(Uσγ,L)).
Lemma 7.2.4 (a) says that it suffices to consider the terms for dimσγ =n. On the other hand, in general, we use Σmax, but completeness of X also makes us possible to use the open cover U ={Uσ}σ∈Σ(n) for the ˇCech complex. This is because we have no cone of dimension less than nisolated in NR'Rn. Thus, the zero ˇCech complex only consists ofn-dimensional cones, i.e., Cˇ0(U,L) =L
σ∈Σ(n)H0(Uσ,L). Hence, we only have the case for p= 0, so that part (a) is done.
Part (b) is just an immediate consequence of Lemma 7.2.4 (b) and (c).
Now, we finally have the Brion’s equalities. Rather, Theorem 7.2.6 will be used in the next section.
Corollary 7.2.7 (Brion’s Equalities). Let P ⊆ MR ' Rn be a full dimensional lattice polytope, and for each vertex v∈P, let Cv = Cone(P∩M −v)⊆MR. Then:
(a) P
m∈P∩Mχm =P
v:vertexχv· S P
m∈Cv∩Mχm .
(b) If P is simple and v ∈ P is a vertex, let mv,1, . . . , mv,n ∈ M be the minimal generators of Cv. Then
X
m∈P∩M
χm = X
v:vertex
χv·P
m∈Pv∩Mχm Qn
i=1(1−χmv,i) .
Proof. Consider the toric variety XP and the line bundle LXP(DP). Note that DP = P
F facesaFDF is Cartier, where aF is from (3.2.1). Moreover, this divisor is basepoint free Cartier, so is nef. Thus, applying Demazure vanishing (theorem 6.2.3) shows that
˜
χ(XP,L) =P
m∈P∩Mχm. For vertices v ofP,σv∨ =Cv and (7.2.3) implies that
˜
χ(Uσv,L) =χv· X
m∈σ∨v∩M
χm =χv· X
m∈Cv∩M
χm
Using Theorem 7.2.6 concludes our proof.
Theorem 7.3.1(Equivariant Riemann-Roch). For a smooth complete toric varietyX= XΣ and a line bundle L =OX(D) of a torus-invariant divisor D on X, we have
χT(L) = Z
Xeq
chT(L)TdT(X).
In this theorem, χT(L),R
Xeq, chT(L), and TdT(X) are equivariant versions of the corresponding objects appearing in Theorem 7.0.1.
The Equivariant Euler Characteristic. Let T = TN is the torus associated to M.
Since we will deal with power series, we need the completion of (ΛT)Q=HT•({pt},Q) = L∞
k=0HTk({pt},Q), say
Λ =b HbT•({pt},Q) =
∞
Y
k=0
HTk({pt},Q).
From (6.3.1), using symmetric algebra withs: SymQ(M)'(ΛT)Q, ifm∈M, thens(m) is a degree 2 element of Λ so that the exponentialb
es(m)= 1 +s(m) +1
2s(m)2+ 1
6s(m)3+· · · ∈Λ.b
For a complete toric variety X = XΣ and the line bundle L = OX(D) of a torus- invariant divisorDon X, theequivariant Euler characteristic of L is defined by
χT(L) = X
m∈M n
X
i=0
(−1)idimHi(X,L)mes(m)∈Λ.b (7.3.1) The Nonequivariant Limit.Leti∗pt:Λb !Qbe the map that sends elements of posi- tive degree to zero, andi∗X :HT•(X,Q)!H•(X,Q).i∗pt andi∗X are our “nonequivariant limit” for passing equivariant version of HRR to ordinary HRR we desired. In particular, i∗pt takes the equivariant Euler characteristic to the ordinary Euler characteristic
i∗pt(χT(F)) =χ(F) =
n
X
i=0
(−1)idimHi(X,F). (7.3.2) Equivariant Chern Characters and Todd Classes. In this time we use another completion of the equivariant cohomologyHT•(X,Q). The completion is
HbT•(X,Q) =
∞
Y
k=0
HTk(X,Q).
The nonequivariant limit mapi∗X :HT•(X,Q) !H•(X,Q) extends to a ring homomor- phism
i∗X :HbT•(X,Q)!H•(X,Q) since elements of degree >2dimX map to zero.
By Proposition 6.3.1, a torus-invariant divisor D = P
ρaρDρ gives the equivariant cohomology class [D]T =P
ρaρ[Dρ]T ∈HT2(X,Q). Then we define theequivariant Chern
character chT(L) ofL =OX(D) to be
chT(L) =e[D]T = 1 + [D]T + 1
2![D]2T +· · · ∈HbT•(X,Q).
On the other hand, we have a nice form of Todd class of X from Theorem 7.1.4, it is desirable to define theequivariant Todd class ofX to be
TdT(X) =
r
Y
i=1
[Dρ]T
1−e−[Dρ]T.
Considering this with Bernoulli numbers shows that TdT(X)∈HbT•(X,Q). Sincei∗X[Dρ]T = [Dρ] by Proposition 6.3.1, we have
i∗XchT(L) = ch(L) and i∗XTdT(L) = Td(L). (7.3.3) The Equivariant Integral.Note that the map R
X :H•(X,Q)!Qis the generalized Gysin map of the constant function p :X ! {pt}. It also gives the equivariant Gysin map
p!:HT•(X,Q)!HT•({pt},Q) which maps HTk(X,Q) to HTk−2dimX({pt},Q). We writeR
Xeq instead ofp!, so that Z
Xeq
:HbT•(X,Q)!HbT•({pt},Q) =Λ.b This is called theequivariant integral. Note thatR
Xeq andR
X are compatible with taking the nonequivariant limit (Proposition A.1.3 ). This will make it easy to derive HRR from equivariant Riemann-Roch.
Delta Map as an Isomorphism.The delta map from (6.3.2) induced by the inclusion of the fixed point set XT ⊆X can be extended to the completion
δ :HbT•(X,Q)−! M
σ∈Σ(n)
HbT•({xσ},Q) = M
σ∈Σ(n)
Λ.b
IfSbe the multiplicative set consisting of all finite products of nonzero degree 2 elements of Λ with the empty product, then the localized mapb
δS :HbT•(X,Q)S −! M
σ∈Σ(n)
ΛbS. (7.3.4)
becomes an isomorphism.
Local Decomposition of Equivariant Euler Characteristics. We decompose the equivariant Euler chracteristicχT(L)∈Λ by local terms with respect tob n-dimensional cones, and this decomposition will appear in ΛbS. First, define the ring homomorphism
Φ :Z[M]−!Λ,b χm7−!es(m).
Φ extends a well-defined map between to two localizations: Q[M]S and ΛbS, where S is the multiplicative set consisting of all finite products of nonzero degree 2 elements ofΛ.b In Q[M]S, all forms of the type 1−χm are invertible, and in ΛbS, the image of 1−χm under Φ is s(m)1−es(m)s(m), where 1−es(m)s(m) =−1−12s(m)− · · · is invertible inΛ. Thus, web have well-defined ring homomorphism
Φ :Q[M]S −!ΛbS. Then, by definitions, it easy to see that
Φ( ˜χ(X,L)) =χT(L)
when X=XΣ is complete. Using Σ, we now define the local version of χT(L).
Definition 7.3.2. Let X=XΣ be complete of dimension nand set L =OX(D) for a torus-invariant Cartier divisorD. Given σ∈Σ(n), define
χTσ(L) = Φ(S( ˜χ(Uσ,L))).
Applying Φ to Theorem 7.2.6 directly gives the local decomposition of equivariant Euler characteristic.
Theorem 7.3.3. Let X = XΣ be complete of dimension n and set L = OX(D) for a torus-invariant Cartier divisor D with Cartier data {mσ}σ∈Σ(n). Then:
(a) χT(L) =P
σ∈Σ(n)χTσ(L) in ΛbS.
(b) If σ ∈ Σ(n) is simplicial and mσ,1, . . . , mσ,n ∈ M are the minimal generators of σ∨ ⊆MR, then
χTσ(L) = es(mσ)·P
m∈Pσ∨∩Mes(m) Qn
i=1(1−es(mσ,i)) for Pσ∨ ={Pn
i=1λimσ,i |0≤λi <1}. In particular, if σ is smooth, then χTσ(L) = es(mσ)
Qn
i=1(1−es(mσ,i)).
Fixed Points.The next step is to decompose the equivariant integral. LetX=XΣ, and assume thatXis complete and simplicial. Given σ∈Σ(n),V(σ) ={xσ}is a fixed point of the torus action. Consider the mapp◦iσfromxσ to pt, whereiσ :V(σ) ={xσ},−!X and p:X!{pt}, induces a commutative diagram
HbT•(X,Q)
i∗σ
((
HbT•({pt},Q) =Λb
p∗
OO
∼ //HbT•({xσ},Q)
(7.3.5)
where the isomorphism on the bottom is (p ◦iσ)∗ = i∗σ ◦p∗. From now on, for α ∈ HbT•(X,Q), we can regardi∗σ(α)∈as an element of Λ.b
The following lemma is useful when we prove later propositions.
Lemma 7.3.4. Let X = XΣ be complete and simplicial of dimension n. Also let σ ∈ Σ(n) and ρ∈Σ(1). Then:
(a) If ρ /∈σ(1), then i∗σ([Dρ]T) = 0.
(b) If ρ∈σ(1), theni∗σ([Dρ]T) = −1l s(m), wherel be the smallest positive integer such thatlDρ is Cartier and m∈M satisfies lDρ∩Uσ = div(χm).
Proof. Using Proposition 6.3.1 (c), for any ρ ∈Σ(1), we have jσ∗([Dρ]T) = [Dρ∩Uσ]T, when we factor iσ as {xσ} ,! Uσ
jσ
,! X. [Dρ∩Uσ]T is zero, because Dρ∩Uσ = ∅ if ρ6∈σ(1).
For ρ∈σ(1), using Proposition 6.3.1 (b) and (c) and our hypothesis, we obtain jσ∗([lDρ]T) = [lDρ∩Uσ]T = [div(χm)]T =−s(m)·1∈HbT•(Uσ,Q).
Note that the constant map to a point makes all asΛ-modules, andb s(m) is inΛ. Thus,b considering s(m) as a scalar implies thati∗σ([lDρ]T) =−s(m)·1.
We also have the equivariant Gysin mapiσ! :HbT•({pt},Q)!HbT•(X,Q) from Propo- sition A.1.1. Here is an important property of this map.
Proposition 7.3.5. Let X =XΣ be complete and simplicial. If σ ∈Σ(n), then iσ!(1) = mult(σ) Y
ρ∈σ(1)
[Dρ]T. Proof. Note that Q
ρ∈σ(1)[Dρ]T and iσ!(1) live in HT2n(X,Q), see Proposition A.1.1. We show the following two things:
Z
Xeq
iσ!(1) = 1 and Z
Xeq
mult(σ) Y
ρ∈σ(1)
[Dρ]T = 1. (7.3.6)
i∗σ0(iσ!(1)) = 0 and i∗σ0( Y
ρ∈σ(1)
[Dρ]T) = when σ0∈Σ(n), σ0 6=σ. (7.3.7) For (7.3.6), by Proposition A.1.1 (b) and R
Xeq = p!, p ◦ iσ : {xσ} ! {pt} im- plies that p!(iσ!(1)) = p!◦iσ!(1) = 1. On the other hand, from Proposition A.1.1 (a), R
Xeq
Q
ρ∈σ(1)[Dρ]T is a degree zero element in bλ. Using the commutative diagram in Proposition A.1.3 and Proposition 6.3.1 (d),
i∗pt Z
Xeq
Y
ρ∈σ(1)
[Dρ]T = Z
X
i∗X( Y
ρ∈σ(1)
[Dρ]T) = Z
X
Y
ρ∈σ(1)
i∗X[Dρ]T = Z
X
Y
ρ∈σ(1)
[Dρ].
By Lemma 6.3.4, and the fact theR
X[xσ] = 1, we are done with desired result.
To achieve (7.3.7), for different cones σ6=σ0 in Σ(n), Proposition A.1.2 gives 0 //HbT•({xσ0},Q)'Λb
HbT•({xσ},Q)
OO
iσ! //HbT•(X,Q)
i∗
σ0
OO
Thus we havei∗σ0◦iσ!= 0. Also, Lemma 7.3.4 implies that the second equality in (7.3.7).
(7.3.6) and (7.3.7) imply that iσ!(1) and mult(σ)Q
ρ∈σ(1)[Dρ]T are equal inHT2n(X,Q).
Decomposing the Equivariant Integral.Now we decompose the equivariant integral, for smooth complete toric varietyX =XΣ. This is important because when we do real computation for explicit varieties as examples, applying nonequivariant limit is just the process of extracting some information from this equivariant integral.
From the constant mapp:X!{pt},HbT•(X,Q) is aΛ-module viab p∗in the commu- tative diagram (7.3.5). In addition, we have the Gysin map R
Xeq =p!:HbT•(X,Q) !Λ,b being aΛ-module homomorphism, because of Proposition A.1.1 (c). These maps extendsb to localizations at S, where S is the multiplicative set of all finite products of nonzero degree 2 elements of Λ. Note that we identifiedb Λ byb HbT•({xσ},Q) through (p◦iσ)∗, which is the inverse of the map (p◦iσ)!.
In the next theorem, we will work on the localization HbT•(X,Q)S, which is a ΛbS- module.
Theorem 7.3.6. If X = XΣ is an n-dimensional smooth complete toric variety and α∈HbT•(X,Q), then
Z
Xeq
α= (−1)n X
σ∈Σ(n)
i∗σ(α) Qn
i=1s(mσ,i)
in ΛbS, where mσ,1, . . . , mσ,n are the minimal generators of σ∨ ⊆ MR for σ ∈ Σ(n). In particular, when X is simplicial, we have
Z
Xeq
α= (−1)n X
σ∈Σ(n)
mult(σ∨)i∗σ(α) Qn
i=1s(mσ,i)
Proof. As a first step, we calculate without summation over n-dimensional cones Z
Xeq
α· (−1)ni∗σ(α) Qn
i=1s(mσ,i) =p!
(−1)n(α·i∗σ(α)) Qn
i=1s(mσ,i)
=p!
(−1)niσ!(i∗σ(α)) Qn
i=1s(mσ,i)
by Proposition A.1.1 (c)
=
(−1)n Qn
i=1s(mσ,i)
p!(iσ!(i∗σ(α)))
=
(−1)n Qn
i=1s(mσ,i)
i∗σ(α) ∵identification via (p◦iσ)∗. Next, we will use the map δS :HbT•(X,Q)S ' L
σ∈Σ(n)ΛbS from (7.3.4). Denote the identity in ΛbS with respect to σ ∈ Σ(n) by eσ. Let ui be the minimal generator of ρi ∈σ(1) for 1≤i≤n. Then smoothness ofσ gives the minimal generators mσ,i of σ∨ as the dual basis, mult(σ) = 1, and Cartier divisors Dρi with Dρi ∩Uσ = div(χmσ,i).
Thus,
δ(iσ!(1)) =X
σ0
i∗σ0(iσ!(1))·eσ0
=i∗σ(iσ!(1))·eσ ∵(7.3.7)
=
n
Y
i=1
i∗σ([Dρi]T)·eσ by Proposition 7.3.5
=
n
Y
i=1
(−s(mσ,i))·eσ by Lemma 7.3.4 (b) with l= 1.
Now, consider that injectivity ofδ gives that δ
X
σ∈Σ(n)
(−1)niσ!(1) Qn
i=1s(mσ,i)
= X
σ∈Σ(n)
eσ =δ(1) =⇒ X
σ∈Σ(n)
(−1)niσ!(1) Qn
i=1s(mσ,i) = 1∈HbT•(X,Q)S Hence, using the above, we can deduce the desired result for the smooth case.
For the simplicial case, using the fact that mult(σ)mult(σ∨) = Qn
i=1li, where li is the smallest positive integer such thatmi=lim0i ∈M for the dual basism0i, after simple computation, we are done.
Proof of Equivariant RR.Now, we prove the equivariant Riemann-Roch theorem for smooth complete toric varieties.
Proof of Theorem 7.3.1. Denoteα= chT(L)TdT(X).D is Cartier with the Cartier data {mσ}σ∈Σ(n), satisfying D|Uσ= div(χmσ). Set mi = mσ,i, which are the minimal generatorsni ofσ, and Di =Dρi forρi ∈Σ(1) withDi|Uσ= div(χmi). Then,
i∗σ(chT(L)) =i∗σ(e[D]T) by Definition
=ei∗σ([D]T)
=es(mσ) by the proof of Lemma 7.3.4.
i∗σ(TdT(X)) =i∗σ( Y
ρ∈Σ(1)
[Dρ]T
1−e−[Dρ]T) = Y
ρ∈Σ(1)
i∗σ
[Dρ]T
1−e−[Dρ]T
= Y
ρ∈σ(1)
i∗σ
[Dρ]T
1−e−[Dρ]T
by Lemma 7.3.4 (a)
= Y
ρ∈σ(1)
−s(mi)
1−es(mi) by Lemma 7.3.4 (b)
Thus, we have
χT(L) = X
σ∈Σ(n)
es(mσ) Qn
i=1(1−es(mi)) by Theorem 7.3.3 (b)
= X
σ∈Σ(n)
(−1)n Qn
i=1s(mi) ·es(mσ)· Qn
i=1s(mi) Qn
i=1(1−es(mi))
= X
σ∈Σ(n)
(−1)n Qn
i=1s(mi) ·i∗σ(chT(L))·i∗σ(TdT(X))
= X
σ∈Σ(n)
(−1)n Qn
i=1s(mi) ·i∗σ(α)
∵i∗σ is a ring homomorphisim
= Z
Xeq
α by Theorem 7.3.6.
Passing from Equivariant RR to HRR.Using nonequivariant limits, we conclude our goal.
Theorem 7.3.7. For an invertible sheafL on a smooth complete toric varietyX=XΣ, we have
χ(L) = Z
X
ch(L)Td(X).
Proof. SinceXis smooth,L =OX(D) for some Cartier divisorDonX. We may assume
that this is a torus-invariant divisor since DivTN(X)!Cl(X) is surjective. Then χ(L) =i∗pt(χT(L)) Definition of nonequivariant limit
=i∗pt( Z
Xeq
chT(L)TdT(X)) by equivariant RR
= Z
X
i∗X(chT(L)TdT(X)) by Proposition A.1.3
= Z
X
i∗X(chT(L))i∗X(TdT(X)) sincei∗X is a ring homomorphism
= Z
X
ch(L)Td(X)
In fact, if we modify the the equivariant Todd class of a simplicial toric variety X and use the quotient construction ofX, then we can obtain a simplicial version of HRR.
The Simplicial Case.We introduce the equivariant Riemann-Roch theorem and HRR whenX=XΣ is complete and simplicial. In this case, Todd class cannot be represented by [Dρ] forρ∈Σ(1) as before. Thus it need to be modified.
Let X = XΣ be a complete simplicial toric variety. Consider the group G in the quotient construction ofX, namely
G=
(tρ)∈(C∗)Σ(1) |Y
ρ
thm,uρ ρi= 1 for all m∈M . Each ρ∈Σ(1) gives the character
χρ: (C∗)Σ(1)−!C∗
defined by projection on the ρth factor. Then for any coneσ ∈Σ, let Gσ =
g∈G|χρ(g) = 1 for all ρ6∈σ(1)
=
(tρ)∈(C∗)Σ(1) |tρ= 1 for ρ6∈σ(1), Y
ρ∈σ(1)
thm,uρ ρi= 1 for m∈M . One can show that
Gσ 'Nσ/(Σρ∈σ(1)Zuρ), so that|Gσ|= mult(σ).
Then we set
GΣ= [
σ∈Σ
Gσ ⊆G
and define theequivariant Todd class ofX to be TdT(X) = X
g∈GΣ
Y
ρ∈Σ(1)
[Dρ]T
1−χρ(g)e−[Dρ]T (7.3.8)
whenX =XΣ is complete and simplicial. Here is the result proved in [12] by Brion and Vergne. This result has also been proved by Edididin and Graham, [13], using different methods.
Theorem 7.3.8. Let X=XΣ be a complete simplicial toric variety. If L =OX(D) is the line bundle of a torus-invariant Cartier divisor D onX, then
χT(L) = Z
Xeq
chT(L)TdT(X), where TdT(X) is defined in (7.3.8).
Once we have a simplicial version of equivariant Riemann-Roch, the proof of Theorem 7.3.7 gives HRR in the simplicial case as follows.
Corollary 7.3.9. Let X = XΣ be a complete simplicial toric variety. If L is a line bundle on X, then
χ(X) = Z
X
ch(L)Td(X), where
Td(X) = X
g∈GΣ
Y
ρ∈Σ(1)
[Dρ] 1−χρ(g)e−[Dρ].