singularities of a certain class of toric varieties1576/fulltext.pdf · singularities of a certain...
TRANSCRIPT
Singularities of A Certain Class of Toric
Varieties
A dissertation presented
by
Himadri Mukherjee
to
The Department of Mathematics
In partial fulfillment of the requirements for the degree of
Doctor of Philosophy
in the field of
Mathematics
Northeastern University
Boston, Massachussetts
April, 2008
1
Singularities of A Certain Class of Toric Varieties
by
Himadri Mukherjee
ABSTRACT OF DISSERTATION
Submitted in partial fulfillment of the requirements
for the degree of Doctor of Philosophy in Mathematics
in the Graduate School of Arts and Sciences of
Northeastern University, April 2008
2
Abstract
Hibi considered the class of algebras k[L] = k[xα|α ∈ L] with straightening laws
associated to a finite distributive lattice L in his paper [2]. In that paper he
proves that these algebras are normal and integral domains. This result along
with the work of Sturmfels and Eisenbud [7] on binomial prime ideals implies
that the affine varieties associated to the algebra k[L] are normal toric varieties.
In the present work we will consider the toric variety X(L) = spec(k[L]), we will
give the combinatorial description of the cone σ associated to it. The final result
will be to give a standard monomial basis for the tangent cone Txτ where xτ is a
singular point associated to a torus orbit Oτ for the action of the torus T , where
τ is a face of the cone σ
3
ACKNOWLEDGMENTS
I am deeply grateful to my advisor Venkatramani Lakshmibai
for her lessons, support and patience throughout this time.
Also I would like to thank the Department of Mathematics of
Northeastern University for financial support.
4
To My Family,
baba, ma, eban EbK ipyalŇek,
jŇbn mrenr sŇmana qaĹaey
bńzu eH Aamar
reyqo daĎĹaey
5
Contents
1 Preface 8
2 Introduction 9
2.1 Main Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Main Result 1. . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.2 Main Result 2. . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Sketch of the proof of the main result 1: . . . . . . . . . . . . . . 12
2.3 Sketch of proof of main result 2. . . . . . . . . . . . . . . . . . . . 12
2.4 Generalities on toric varieties . . . . . . . . . . . . . . . . . . . . 14
2.4.2 Resume of combinatorics of affine toric varieties . . . . . . 15
2.4.3 Toric ideals (cf. [17]) . . . . . . . . . . . . . . . . . . . . . 17
2.4.11 Varieties defined by binomials . . . . . . . . . . . . . . . . 20
2.4.15 Orbit decomposition in affine toric varieties . . . . . . . . 22
3 Distributive lattices and Toric Varieties 24
3.1 Generalities on finite distributive lattices . . . . . . . . . . . . . . 24
3.1.10 Generalities on distributive lattices . . . . . . . . . . . . . 26
3.2 The variety X(L) . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6
3.2.5 Some more lattice theory . . . . . . . . . . . . . . . . . . . 33
4 A Basis for Cotangent Space 38
4.1 Cone and dual cone of XL: . . . . . . . . . . . . . . . . . . . . . . 38
4.1.11 Analysis of faces of σ . . . . . . . . . . . . . . . . . . . . . 42
4.2 Generation of the cotangent space at Pτ . . . . . . . . . . . . . . 45
4.3 A basis for the cotangent space at Pτ . . . . . . . . . . . . . . . . 52
4.3.1 Determination of the degree one part of J(τ): . . . . . . . 53
5 Standard Basis for Tangent Cone 56
5.1 Analysis of diamond relations: . . . . . . . . . . . . . . . . . . . . 56
5.2 Tangent cone at Pτ . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2.2 Determination of I(τ): . . . . . . . . . . . . . . . . . . . . 74
7
CHAPTER 1. PREFACE 8
Chapter 1
Preface
Hibi studies a class of algebras with straightening laws in [2], which are defined
by binomial relations as a quotient of full polynomial algebras. These algebras
come up as degenerations of Schubert varieties. My work deals with constructing
a standard basis for the tangent cone to these toric varieties at a singular point.
Chapter 2
Introduction
Let K denote the base field which we assume to be algebraically closed of arbitrary
characteristic. Given a distributive lattice L, let XL denote the Hibi variety -
the affine variety in A#L whose vanishing ideal is generated by the binomials
XτXϕ − Xτ∨ϕXτ∧ϕ in the polynomial algebra K[Xα, α ∈ L]; here, τ ∨ ϕ (resp.
τ ∧ ϕ) denotes the join - the smallest element of L greater than both τ, ϕ (resp.
the meet - the largest element of L smaller than both τ, ϕ). In the sequel, we shall
refer to the quadruple (τ, ϕ, τ ∨ ϕ, τ ∧ ϕ) ((τ, ϕ) being a skew pair) as a diamond.
These varieties were extensively studied by Hibi in [13] where Hibi proves that
XL is a normal variety. On the other hand, Eisenbud-Sturmfels show in [7] that
a binomial prime ideal is toric (here, “toric ideal” is in the sense of [17]). Thus
one obtains that XL is a normal toric variety. We refer to such an XL as a Hibi
toric variety.
For L being the Bruhat poset of Schubert varieties in a minuscule G/P , it is
shown in [9] that XL flatly deforms to G/P (the cone over G/P ), i.e., there exists
a flat family over A1 with G/P as the generic fiber and XL as the special fiber.
More generally for a Schubert variety X(w) in a minuscule G/P , it is shown in
9
10
[9] that XLw flatly deforms to X(w) (here, Lw is the Bruhat poset of Schubert
subvarieties of X(w)). In a subsequent paper (cf. [11]), the authors of loc.cit.,
studied the singularities of XL,L being the Bruhat poset of Schubert varieties
in the Grassmannian; further, in loc.cit., the authors gave a conjecture giving a
necessary and sufficient condition for a point on XL to be smooth, and proved in
loc.cit., the sufficiency part of the conjecture. Subsequently, the necessary part of
the conjecture was proved in [4] by Batyrev et al. The toric varieties XL, L being
the Bruhat poset of Schubert varieties in the Grassmannian play an important
role in the area of mirror symmetry; for more details, see [3, 4]. We refer to such
an XL as a Grassmann-Hibi toric variety.
In this thesis, we first determine an explicit description (cf. §4.1) of the cone σ
associated to a Hibi toric variety and the dual cone σ∨ (note that if Sσ denotes
the semigroup of integral points in σ∨, then K[XL], the K-algebra of regular
functions on XL can be identified with the semigroup algebra K[Sσ]). Using this
we determine the cotangent space at any point on XL very explicitly as described
below.
For each face τ of σ, there exists a distinguished point Pτ in the torus orbit Oτ
corresponding to τ ; namely, identifying a (closed) point in XL with a semigroup
map Sσ → K∗∪{0}, Pτ corresponds to the semigroup map fτ which sends u ∈ Sσ
to 1 or 0 according as u is in τ⊥ or not. We have that fτ induces an algebra map
ψτ : K[XL] → K whose kernel is precisely MPτ , the maximal ideal corresponding
to Pτ . For α ∈ L, we shall denote by Pτ (α), the co-ordinate of Pτ (considered as
a point of A#L) corresponding to α. Set
Dτ = {α ∈ L |Pτ (α) 6= 0}
It turns out that Dτ is an embedded sublattice of L; further, Dτ determines the
2.1. Main Results. 11
local behavior at Pτ . To make this more precise, identifying K[XL] as the quotient
of the polynomial algebra K[Xα, α ∈ L] by the ideal generated by the binomials
XτXϕ −Xτ∨ϕXτ∧ϕ, let xα denote the image of Xα in K[XL]. For α ∈ L, set
Fα =
xα, if α 6∈ Dτ
1− xα, if α ∈ Dτ
Then we have that MPτ is generated by {Fα, α ∈ L}. Fix a maximal chain Γ in
Dτ . Let Λτ (Γ) denote the sublattice of L consisting of all maximal chains of L
containing Γ. Let Eτ denote the set of all α ∈ L such that there exists a β ∈ Dτ
such that (α, β) is a diagonal of a diamond whose other diagonal is contained in
L \ Dτ . Let Yτ (Γ) = Λτ (Γ) \ Eτ . For F ∈ MPτ , let F denote the class of F
in MPτ /M2Pτ
. We define an equivalence relation (cf. §4.2) on L \ Dτ in such a
way that for two elements θ, θ′ in the same equivalence class, we have F θ = F θ′ .
Denoting by Gτ (Γ) the set of equivalence classes [θ], where θ ∈ Yτ (Γ) \ {Γ∪Eτ}and F [θ] is non-zero in MPτ /M2
Pτ, we have
2.1 Main Results.
2.1.1 Main Result 1.
Theorem 1: (cf. Theorem 4.3.12) {F [θ], [θ] ∈ Gτ (Γ)}∪{F γ, γ ∈ Γ} is a basis for
the cotangent space MPτ /M2Pτ
.
Using the above result, we determine Sing XL (cf. Theorem 4.3.15):
Let SL = {τ < σ |#Gτ (Γ) + #Γ > #L}.
Theorem 2: (cf. Theorem 4.3.15) Sing XL = ∪τ∈SL
X(Dτ ).
(Here, X(Dτ ) denotes the Hibi variety associated to the distributive lattice Dτ .)
2.2. Sketch of the proof of the main result 1: 12
2.1.2 Main Result 2.
We first determine the vanishing ideal J(τ) of the tangent cone at a singular point
Pτ on the Hibi toric variety XL (here, Pτ is the distinguished point corresponding
to a face τ of the convex polyhedral cone associate to XL). We then introduce
a monomial order > (on the polynomial algebra containing J(τ)), and determine
in J(τ), the initial ideal of J(τ). Using MaCaulay’s Theorem [6], we obtain a
“standard monomial basis” for TCPτ XL (cf. Theorem 5.2.8).
2.2 Sketch of the proof of the main result 1:
Using our explicit description of the generators for σ, σ∨ (cf. §4.1), we first
determine explicitly the embedded sublattice associated to a face τ of σ. We
then analyze the local expression around Pτ for any f ∈ I(XL), the vanishing
ideal of XL, and show the generation of the degree one part of gr (RL,MPτ )
by {F θ, θ ∈ Yτ (Γ)} (here, RL = K[XL], and MPτ is the maximal ideal in RL
corresponding to Pτ ). We then define the equivalence relation on L \ Dτ . The
linear independence of {F [θ] [θ] ∈ Gτ (Γ)} ∪ {F γ, γ ∈ Γ} in MPτ /M2Pτ
is proved
using the defining equations of XL (as a closed subvariety of A#L), thus proving
Theorem 1. Theorem 2 is then deduced from Theorem 1.
2.3 Sketch of proof of main result 2.
Let σ be the convex polyhedral cone associate to the toric variety XL. Let Sσ be
the semigroup of integral points in σ∨ (the cone dual to σ), then K[XL], the K-
algebra of regular functions on XL can be identified with the semigroup algebra
K[Sσ]). For each face τ of σ, there exists a distinguished point Pτ in the torus
2.3. Sketch of proof of main result 2. 13
orbit Oτ corresponding to τ ; namely, identifying a (closed) point in XL with a
semigroup map Sσ → K∗ ∪ {0}, Pτ corresponds to the semigroup map fτ which
sends u ∈ Sσ to 1 or 0 according as u is in τ⊥ or not. We have that fτ induces an
algebra map ψτ : K[XL] → K whose kernel is precisely MPτ , the maximal ideal
corresponding to Pτ . For α ∈ L, we shall denote by Pτ (α), the co-ordinate of Pτ
(considered as a point of A#L) corresponding to α. Set
Dτ = {α ∈ L |Pτ (α) 6= 0}
It turns out that Dτ is an embedded sublattice of L; further, Dτ determines the
local behavior at Pτ . To make this more precise, identifying K[XL] as the quotient
of the polynomial algebra K[Xα, α ∈ L] by the ideal generated by the binomials
XτXϕ −Xτ∨ϕXτ∧ϕ, let xα denote the image of Xα in K[XL]. For α ∈ L, set
Fα =
xα, if α 6∈ Dτ
1− xα, if α ∈ Dτ
Then we have that MPτ is generated by {Fα, α ∈ L}.
Fix a maximal chain Γ in Dτ . Let Λτ (Γ) denote the sublattice of L consisting of
all maximal chains of L containing Γ. This sublattice plays a crucial role in our
discussion as explained below:
For F ∈ Mτ , let F denote the class of F in Mτ/M2τ . Then J(τ) gets identified
with the kernel of the surjective map
π : K[Xθ, θ ∈ L] → gr(R, Mτ ), Xθ 7→ Fθ
For r ∈ N, let π(r) be the restriction of π to the degree r part of the polynomial
algebra K[Xθ, θ ∈ L]. Then one of our main results is that for a diamond D
with (τ, ϕ), the diamond relation fD := XτXϕ − Xτ∨ϕXτ∧ϕ a diamond relation
2.4. Generalities on toric varieties 14
is in the ideal generated by ker π(1) and the diamond relations arising from
diamonds in Λτ (Γ). This result together with the description of the cotangent
space as determined in [15] reduces our discussion to an analysis of the diamond
relations arising from diamonds in the distributive lattice Λτ (Γ); notice one
obvious simplification, namely, by going to Λτ (Γ), we have “shrunk” Dτ to just
one chain (namely Γ) in Dτ . The precise analysis of the diamond relations arising
from diamonds in Λτ (Γ) enables us to determine J(τ) explicitly. The details will
be discussed in chapter 5
The sections are organized as follows: In §2.4, we recall generalities on toric
varieties. In §3.1, we recall some basic results on distributive lattices. In §3.2, we
introduce the Hibi variety X(L), and recall some basic results on X(L). In §4.1,
we determine generators for the cone σ (and the dual cone σ∧); further, for a face
τ of σ, we introduce Dτ and derive some properties of Dτ . In §4.2, we present our
main results on the tangent and cotangent spaces at the distinguished point Pτ
in the orbit corresponding to τ . In §5.1 we study the diamond relations in view
of the local descriptions. tangent cone at point Pτ in §5.2, we describe the ideal
generators and the standard basis for the tangent cone in §5.2.2.
2.4 Generalities on toric varieties
Since our main object of study is a certain affine toric variety, we recall in this
section some basic definitions on affine toric varieties. Let T = (K∗)m be an
m-dimensional torus.
Definition 2.4.1. (cf. [8], [14]) An equivariant affine embedding of a torus T is
an affine variety X ⊆ Al containing T as an open subset and equipped with a
T -action T ×X → X extending the action T ×T → T given by multiplication of
2.4. Generalities on toric varieties 15
the group structure of T . If in addition X is normal, then X is called an affine
toric variety.
2.4.2 Resume of combinatorics of affine toric varieties
Let M be the character group of T . Let X be a toric variety with a torus action
by T ; let R = K[X]. We note the following:
• M-gradation: We have an action of T on R, and hence writing R as a
sum of T weight spaces, we obtain a M -grading for R: R = ⊕{χ∈M}
Rχ, where
Rχ = {f ∈ R|tf = χ(t)f, ∀t ∈ T}.
• The semi group S: Let S = {χ ∈ M |Rχ 6= 0}. Then via the multiplication
in R, S acquires a semi group structure. Thus S is a semi subgroup of M , and R
gets identified with the semi group algebra K[S].
• Finite generation of S: In view of the fact that R is a finitely generated
K-algebra, we obtain that S is a finitely generated semi subgroup of M.
• Generation of M by S: The fact that T and X have the same function field
(since T is a dense open subset of X) implies that M is generated by S.
• Saturation of S: The normality of R (being identified as the semi group
algebra K[S]) implies that S is saturated (recall that a semi subgroup A in M
is saturated, if for χ ∈ M , rχ ∈ A =⇒ χ ∈ A (where r is an integer > 1;
equivalently, A is precisely the lattice points in the cone generated by A, i.e.,
A = θ ∩ A where θ =∑
aixi, ai ∈ R+, xi ∈ A)).
Thus given an affine toric variety X (with torus action by T ), X determines a
finitely generated, saturated semi subgroup S of M which generates M in such a
way that the semi group algebra is the ring of regular functions on X.
Let us also observe that starting with a finitely generated, saturated semi
2.4. Generalities on toric varieties 16
subgroup S of M which generates M , SpecK[S] is an affine toric variety (with
torus action by T ).
This sets up a bijection between {affine toric varieties with torus action by T}and {finitely generated, saturated semi subgroups of M which generate M}.
See [14] for details.
Let θ :=∑
aixi, ai ∈ R+, xi ∈ be the cone generated by S. Then θ is a
rational convex polyhedral cone (i.e., a cone with a (finite) set of lattice points
as generators), and θ is not contained in any hyperplane in MR := M ⊗ R. Let
σ := θ∨ = {v ∈ M∗R|v(f) ≥ 0,∀f ∈ θ}
(here, M∗R is the linear dual of MR) Then σ is a strongly convex rational convex
polyhedral cone (one may take this to be the definition of a strongly convex
rational convex polyhedral cone, namely, a rational convex polyhedral cone τ is
strongly convex if τ∨ is not contained in any hyperplane; equivalently, τ does not
contain any non-zero linear subspace.)
Thus an affine toric variety X (with a torus action by T ) determines a rational
convex polyhedral cone σ in such a way that K[X] is the semigroup algebra K[Sσ],
where Sσ is the semi subgroup in M consisting of the set of lattice points in σ∨.
Conversely, starting with a strongly convex rational convex polyhedral cone σ,
Sσ := σ∨ ∩ M is a finitely generated, saturated semi subgroup of M which
generates M ; and hence determines an affine toric variety X (with a torus action
by T ), namely, X = Spec K[Sσ].
2.4. Generalities on toric varieties 17
2.4.3 Toric ideals (cf. [17])
Let A = {χ1, . . . , χl} be a subset of Zd. Consider the map
πA : Zl+ → Zd, u = (u1, . . . , ul) 7→ u1χ1 + · · ·+ ulχl.
Let K[x] := K[x1, . . . , xl], K[t±1] := K[t1, . . . , td, t−11 , . . . , t−1
d ].
The map πA induces a homomorphism of semigroup algebras
πA : K[x] → K[t±1], xi 7→ tχi .
Definition 2.4.4. (cf. [17]) The kernel of πA is denoted by IA and called the
toric ideal associated to A.
Note that a toric ideal is prime. We shall now show that we have a bijection
between the class of toric ideals and the class of equivariant affine toroidal
embeddings.
Consider the action of T (= (K∗)d) on Al given by
t(a1, · · · , αl) = (tχ1a1, · · · , tχlal)
Then V(IA), the affine variety of the zeroes in K l of IA, is simply the Zariski
closure of the T -orbit through (1, 1, . . . , 1). Let MA be the Z-span of A (inside
X(T ) = Zd), and dA the rank of MA. Let SA be the sub semigroup of Zd generated
by A. Let TA be the dA-dimensional torus with MA as the character group. Then
we have
Proposition 2.4.5. V(IA) is an equivariant affine embedding of TA (of dimension
dA); further, K[V(IA)] is the semigroup algebra K[SA].
Remark 2.4.6. In the above definition, we do not require V(IA) to be normal.
Note that V(IA) is normal if and only if SA is saturated. A variety of the form
2.4. Generalities on toric varieties 18
V(IA), is also sometimes called an affine toric variety. But in this paper, by an
affine variety, we shall mean a normal toric variety as defined in Definition 2.4.1.
Remark 2.4.7. Consider the action of T (= (K∗)d) on K l given by tei = tχiei
(here, ei, 1 ≤ i ≤ l are the standard basis vectors of K l). Then V(IA), the
affine variety of the zeroes in K l of IA is simply the Zariski closure of the T -orbit
through (1, 1, . . . , 1). In particular, V(IA) is an equivariant affine embedding of
T = (K∗)d.
The vanishing ideal of X
Let us return to the affine toric variety X ↪→ Al with a torus action by T (= (K)∗)d
as in Definition 2.4.1. Let Tl = (K∗)l (an l-dimensional torus). Consider T as
a subtorus of Tl. The natural action of Tl on Al induces an action of T on Al,
and the embedding X ↪→ Al is T -equivariant. Hence if we denote the images
of {x1, · · · , xl} in K[X] by {y1, · · · , yl}, then {y1, · · · , yl} are T -weight vectors
with weights {χ1, · · · , χl}, say. Then {χ1, · · · , χl} is a set of generators for the
semi group Sσ (notation being as in §2.4.2, namely, K[X] equals the semi group
algebra K[Sσ]). Denoting, A := {χ1, · · · , χl}, we have that IA of Definition 2.4.4
is precisely the vanishing ideal of X. Thus we obtain
Proposition 2.4.8. I(X), the vanishing ideal of X (for the embedding X ↪→ Al)
is a toric ideal.
Remark 2.4.9. Note that we also have the saturation property for A (since Sσ
is a saturated sub semigroup of X(T ))
Recall the following (see [17]).
2.4. Generalities on toric varieties 19
Proposition 2.4.10. The toric ideal IA is spanned as a K-vector space by the
set of binomials
{xu − xv | u,v ∈ Zn+ with πA(u) = πA(v)}. (*)
(Here, a binomial is a polynomial with at most two terms.)
An example
Let us fix the integers n1, . . . , nd > 1, and let n = Πdi=1ni, m =
∑di=1 ni. Let
el1, . . . , e
lnl
be the unit vectors in Znl , for 1 ≤ l ≤ d. For 1 ≤ ξ1 ≤ n1, . . . , 1 ≤ξd ≤ nd, define
aξ1...ξd= e1
ξ1+ · · ·+ ed
ξd∈ Zn1 ⊕ · · · ⊕ Znd
and let
An1,...,nd= {aξ1...ξd
| 1 ≤ ξ1 ≤ n1, . . . , 1 ≤ ξd ≤ nd}.
The corresponding map
πA : Zn1·····nd+ → Zn1+···+nd
is defined as follows: for 1 ≤ l ≤ d and 1 ≤ il ≤ nl fixed, the (n1+· · ·+nl−1+il)-th
coordinate of πA(u) is given by∑
uξ1...ξl−1ξlξl+1...ξd, the sum being taken over the
elements (ξ1, . . . , ξl−1, ξl, ξl+1, . . . , ξd) of C(n1, . . . , nd) with ξl = il. We call this
subset the l-th slice of C(n1, . . . , nd) defined by il, and denote it by {ξl = il}. The
components (or entries) of an element u ∈ Zn1·····nd are indexed by the elements
(i1, . . . , id) of C(n1, . . . , nd). If (j1, . . . , jd) ∈ {ξl = il}, sometimes we also say that
uj1...jditself belongs to the slice {ξl = il}.
The map πA induces the map
πA : k[x11...1, . . . , xξ1ξ2...ξd, . . . , xn1n2...nd
] → k[t11, . . . , t1n1 , . . . , td1, . . . , tdnd]
2.4. Generalities on toric varieties 20
given by
xξ1...ξd7→ t1ξ1 . . . tdξd
, for 1 ≤ ξ1 ≤ n1, . . . , 1 ≤ ξd ≤ nd.
2.4.11 Varieties defined by binomials
Let l ≥ 1, and let X be an affine variety in Al, not contained in any of the
coordinate hyperplanes {xi = 0}. Further, let X be irreducible, and let its
defining prime ideal I(X) be generated by l binomials
xai11 . . . xail
l − λixbi11 . . . xbil
l , 1 ≤ i ≤ m . (∗)
Consider the natural action of the torus Tl = (K∗)l on Al,
(t1, . . . , tl) · (a1, . . . , al) = (t1a1, . . . , tlal).
Let X(Tl) = Hom(Tl,Gm) be the character group of Tl, and let εi ∈ X(Tl) be the
character
εi(t1, . . . , tl) = ti , 1 ≤ i ≤ l.
For 1 ≤ i ≤ m, let
ϕi =l∑
t=1
(ait − bit)εt .
Set T = ∩mi=1 ker ϕi, and X◦ = {(x1, . . . , xl) ∈ X | xi 6= 0 for all 1 ≤ i ≤ l}.
Proposition 2.4.12. Let notations be as above.
(1) There is a canonical action of T on X.
(2) X◦ is T -stable. Further, the action of T on X◦ is simple and transitive.
(3) T is a subtorus of Tl, and X is an equivariant affine embedding of T .
Proof 2.4.13. (1) We consider the (obvious) action of T on Al. Let (x1, . . . , xl) ∈X, t = (t1, . . . , tl) ∈ T , and (y1, . . . , yl) = t · (x1, . . . , xl) = (t1x1, . . . , tlxl). Using
2.4. Generalities on toric varieties 21
the fact that (x1, . . . , xl) satisfies (∗), we obtain
yai11 . . . yail
l = tai11 . . . tail
l xai11 . . . xail
l = λitbi11 . . . tbil
l xbi11 . . . xbil
l = λiybi11 . . . ybil
l ,
for all 1 ≤ i ≤ m, i.e. (y1, . . . , yl) ∈ X. Hence t · (a1, . . . , al) ∈ X for all t ∈ T .
(2) Let x = (x1, . . . , xl) ∈ X◦, and t = (t1, . . . , tl) ∈ T . Then, clearly
t · (x1, . . . , xl) ∈ X◦. Considering x as a point in Al, the isotropy subgroup
in Tl at x is {id}. Hence the isotropy subgroup in T at x is also {id}. Thus the
action of T on X◦ is simple.
Let (x1, . . . , xl), (x′1, . . . , x′l) ∈ X◦. Set t = (t1, . . . , tl), where ti = xi/x
′i. Then,
clearly t ∈ T . Thus (x1, . . . , xl) = t · (x′1, . . . , x′l). Hence the action of T on X◦ is
simple and transitive.
(3) Now, fixing a point x ∈ X◦, we obtain from (2) that the orbit map t 7→ t · xis in fact an isomorphism of T onto X◦. Also, since X is not contained in any
of the coordinate hyperplanes, the open set Xi = {(x1, . . . , xl) ∈ X | xi 6= 0}is nonempty for all 1 ≤ i ≤ l. The irreducibility of X implies that the sets Xi,
1 ≤ i ≤ l, are open dense in X, and hence their intersection
X◦ =l⋂
i=1
Xi = {(x1, . . . , xl) ∈ X | xi 6= 0 for any i}
is an open dense set in X, and thus X◦ is irreducible. This implies that T is
irreducible (and hence connected). Thus T is a subtorus of Tl. The assertion that
X is an equivariant affine embedding of T follows from (1) and (2).
Remark 2.4.14. One can see that the ideal I(X) is a toric ideal in the sense of
Definition 2.4.4; more precisely, A = {ρi, 1 ≤ i ≤ l}, where ρi = εi
∣∣T
(here K[T ]
is identified with K[t±11 , . . . , t±1
d ], and the character group X(T ) with Zd).
2.4. Generalities on toric varieties 22
2.4.15 Orbit decomposition in affine toric varieties
We shall preserve the notation of §2.4.2. As in §2.4.2, let X be an affine toric
variety with a torus action by T . Let K[X] = K[Sσ]. We shall denote X also by
Xσ.
Let us first recall the definition of faces of a convex polyhedral cone:
Definition 2.4.16. A face τ of σ is a convex polyhedral sub cone of σ of the
form τ = σ ∩ u⊥ for some u ∈ σ∨, and is denoted τ < σ.
We have that Xτ is a principal open subset of Xσ, namely,
Xτ = (Xσ)f
Each face τ determines a (closed) point Pτ in Xσ, namely, it is the point
corresponding to the maximal ideal in K[X](= K[Sσ]) given by the kernel of
eτ : K[Sσ] → K, where for u ∈ Sσ, we have
eτ (u) =
1, if u ∈ τ⊥
0, otherwise
Remark 2.4.17. As a point in Al, Pτ may be identified with the l-tuple with 1
at the i-th place if χi is in τ⊥, and 0 otherwise (here, as in §2.4.3, χi denotes the
weight of the T -weight vector yi - the class of xi in K[Xσ]).
Let Oτ denote the T -orbit in Xσ through Pτ . We have the following orbit
decomposition in Xσ:
Xσ = ∪θ≤σ
Oθ
Oτ = ∪θ≥τ
Oθ
dim τ + dimOτ = dimXσ
2.4. Generalities on toric varieties 23
(here, by dimension of a cone τ , one means the vector space dimension of the
span of τ).
See [8], [14] for details.
Thus τ 7→ Oτ defines an order reversing bijection between {faces of σ} and {T -
orbit closures in Xσ}. In particular, we have the following two extreme cases:
1. If τ is the 0-face, then Pτ = (1, · · · , 1) (as a point in Al), and Oτ = T , and is
contained in Xθ,∀θ < σ. It is a dense open orbit.
2. If τ = σ, then Pτ (as a point in Al) is the l-tuple with 1 at the i-th place if χi
is in τ⊥, and 0 otherwise, and Oτ = {Pτ}, and is the unique closed orbit.
For a face τ , let us denote by Nτ the sublattice of N (the Z-dual of M) generated
by the lattice points of τ . Let N(τ) = N/Nτ , and M(τ), the Z-dual of N(τ). For
a face θ of σ such that θ contains τ as a face, set
θτ := (θ + (Nτ )R)/(Nτ )R
(here, for a lattice L, LR = L⊗R). Then the collection {θτ , σ > θ > τ} gives the
set of faces of the cone στ (⊂ N(τ)R).
Lemma 2.4.18. For a face τ < σ, Oτ gets identified with the toric variety
SpecK[Sστ ]. Further, K[Oτ ] = K[Sσ ∩ τ⊥].
Proof 2.4.19. The first assertion follows from the description of the orbit
decomposition ( and the definition of στ ). For the second assertion, we have
Sστ = σ∨τ ∩M(τ) = σ∨ ∩ (τ⊥ ∩M) = Sσ ∩ τ⊥
(note that we have, M(τ) = τ⊥ ∩M).
Chapter 3
Distributive lattices and Toric
Varieties
3.1 Generalities on finite distributive lattices
In this section we recall elementary properties of Distributive lattice.
A partially ordered set which sometime also called a poset is defined as a set p
with a transitive relation ≥. If x ≥ y, in phrase we call x is greater than y or y
is less than x. If x, y ∈ P and x ≥ y and x 6= y then we call x is strictly greater
than y, and write it as x > y. An element x ∈ P is called maximal if x is greater
than all y ∈ P . In the case when an element x ∈ P is less than all the elements
y ∈ P then we y the minimal element of the poset.
Definition 3.1.1. A finite poset P is called bounded if there exist both maximal
and minimal elements and are unique. For the rest of the text we denote 1 as the
unique maximal element of a bounded poset. and 0 for the minimal element of
the same bounded poset.
24
3.1. Generalities on finite distributive lattices 25
Example Let S be a non empty set with finite elements. Let P(S) be the set of
subsets of S, then the inclusion among the subsets define a partial order. A ≥ B
if A ⊂ B. In this poset there is a minimal element Φ and the set S considered
as a subset of itself is a maximal element which are clearly unique. This gives an
example of a bounded poset.
Definition 3.1.2. A totally ordered subset C of a finite poset P is called a chain,
and the number #C − 1 is called the length of the chain.
A chain in a bounded poset p is called a maximal chain if it contains both the
minimal and the maximal element.
Definition 3.1.3. A bounded poset P is said to be graded (or also ranked) if
all maximal chains have equal length (note that 1 and 0 belong to any maximal
chain).
Example The bounded poset in the above example is also a ranked poset; each
maximal chain has length equal to the number of elements in the set S.
Definition 3.1.4. Let P be a graded poset. The length of a maximal chain in P
is called the rank of P .
Definition 3.1.5. In a graded poset P for elements λ, µ ∈ P with λ ≥ µ, the
poset {τ ∈ P | µ ≤ τ ≤ λ} is called the interval from µ to λ, and denoted by
[µ, λ]. The interval is clearly a graded poset. We define the rank of [µ, λ] by lµ(λ);
if µ = 0, then we denote lµ(λ) by just l(λ).
Theorem 3.1.6. Let P be a graded poset, x, y ∈ P two element x ≥ y then the
interval [y, x] is a graded poset.
Proof 3.1.7. Lets call the interval I, I gets the poset structure from the poset
relation of the larger poset P , its also bounded since the element x is the maximal
3.1. Generalities on finite distributive lattices 26
element where y is the minimal element of the poset. To prove that all the
maximal chains have same length lets assume that M1,M2 are two maximal chains
with different lengths. We give an order relation ≥ on the totally ordered subsets
of the graded poset P given by the inclusion. Under this order any inductively
ordered chain must have a maximal element since the set of all totally ordered
chains in P is finite. Note that this maximal is a maximal chain of the poset P .
So we can extend any totally ordered chain of the lattice P to a maximal chain.
Remark 3.1.8. In the later part of the work, we refer to l(λ) as the level of λ in
L.
Definition 3.1.9. Let P be a graded poset, and λ, µ ∈ P , with λ ≥ µ. The
ordered pair (λ, µ) is called a cover (and we also say that λ covers µ) if lµ(λ) = 1.
3.1.10 Generalities on distributive lattices
Definition 3.1.11. A lattice is a partially ordered set (L,≤) such that, for every
pair of elements x, y ∈ L, there exist elements x ∨ y and x ∧ y, called the join,
respectively the meet of x and y, defined by:
x ∨ y ≥ x, x ∨ y ≥ y, and if z ≥ x and z ≥ y, then z ≥ x ∨ y,
x ∧ y ≤ x, x ∧ y ≤ y, and if z ≤ x and z ≤ y, then z ≤ x ∧ y.
It is easy to check that the operations ∨ and ∧ are commutative and associative.
Definition 3.1.12. An element z ∈ L is called the zero of L, denoted by 0, if
z ≤ x for all x in L. An element z ∈ L is called the one of L, denoted by 1, if
z ≥ x for all x in L.
3.1. Generalities on finite distributive lattices 27
Definition 3.1.13. Given a lattice L, a subset L′ ⊂ L is called a sublattice of L
if x, y ∈ L′ implies x ∧ y ∈ L′, x ∨ y ∈ L′.
Definition 3.1.14. Two lattices L1 and L2 are em isomorphic if there exists a
bijection ϕ : L1 → L2 such that, for all x, y ∈ L1,
ϕ(x ∨ y) = ϕ(x) ∨ ϕ(y) and ϕ(x ∧ y) = ϕ(x) ∧ ϕ(y).
Definition 3.1.15. A lattice is called distributive if the following identities hold:
x ∧ (y ∨ z) = (x ∧ y) ∨ (x ∧ z) (3.1)
x ∨ (y ∧ z) = (x ∨ y) ∧ (x ∨ z). (3.2)
Example of a distributive lattice would be the poset of the powerset of a set S.
Where the operations ∨ and ∧ are given by union and intersection respectively.
Definition 3.1.16. An element z of a lattice L is called join-irreducible (resp.
meet irreducible ) if z = x ∨ y (respectively z = x ∧ y) implies z = x or z = y.
The set of join-irreducible (resp. meet irreducible) elements of L is denoted by
JL (resp. ML), or just by J (resp. M) if the reference to the lattice L is obvious
in the context.
Examples: 1. The lattice of all subsets of the set {1, 2, . . . , n} is denoted by
B(n), and called the Boolean algebra of rank n.
A good class of example of distributive lattice would be the chain product lattices
defined below.
2. Chain product lattice: Given an integer n ≥ 1, C(n) C(n) will denote
the chain {1 < · · · < n},C(n1, . . . , nd) and for n1, . . . , nd > 1, C(n1, . . . , nd) will
denote the chain product lattice C(n1) × · · · × C(nd) consisting of all d-tuples
3.1. Generalities on finite distributive lattices 28
(i1, . . . , id), with 1 ≤ i1 ≤ n1, . . . , 1 ≤ id ≤ nd. For (i1, . . . , id), (j1, . . . , jd) in
C(n1, . . . , nd), we define
(i1, . . . , id) ≤ (j1, . . . , jd) ⇐⇒ i1 ≤ j1, . . . , id ≤ jd .
We have
(i1, . . . , id) ∨ (j1, . . . , jd) = (max{i1, j1}, . . . , max{id, jd})
(i1, . . . , id) ∧ (j1, . . . , jd) = (min{i1, j1}, . . . , min{id, jd}).C(n1, . . . , nd) is a finite distributive lattice, and its zero and one are (1, . . . , 1),
(n1, . . . , nd) respectively.
Note that there is a total order / on C(n1, . . . , nd) extending <, namely the
lexicographic order, defined by (i1, . . . , id) / (j1, . . . , jd) if and only if there exists
l < d such that i1 = j1, . . . , il = jl, il+1 < jl+1. Also note that two elements
(i1, . . . , id) / (j1, . . . , jd) are non-comparable with respect to ≤ if and only if there
exists 1 < h ≤ d such that ih > jh.
Sometimes we denote the elements of C(n1, n2, . . . , nd) by xi1...id , with 1 ≤ i1 ≤n1, . . . , 1 ≤ id ≤ nd.
In the case x y ∈ L, where L is a lattice, when x is neither larger nor less than y,
we call x and y non comparable.
One has the following embedding of distributive lattices into sublattices of boolean
algebra. (see [1]):
Theorem 3.1.17. Any finite distributive lattice is isomorphic to a sublattice of
a Boolean algebra of finite rank, and hence, in particular, to a sublattice of a
finite chain product lattice.
We will soon concentrate on the maximal chains of a distributive lattice, and in
turn we will build up these maximal chains with covers. As a good understanding
of the covers the following easy corollary to the definition of a cover will be useful.
3.1. Generalities on finite distributive lattices 29
Corollary 3.1.18. Let (τ, λ1), (τ, λ2) be two covers in a distributive lattice L.
Then λ1 ∧ λ2 is covered by both λ1 and λ2
Definition 3.1.19. A sublattice of the form {τ, φ, τ ∨ φ, τ ∧ φ}, with τ, φ ∈ L
non-comparable is called a diamond, and is denoted by D{τ, φ, τ ∨ φ, τ ∧ φ} or
also just D(τ, φ). The pair (τ, φ) (respectively (τ ∨ φ, τ ∧ φ))) is called the skew
(respectively main) diagonal of the diamond D(τ, φ).
A diamond of the form {x∨ y, x, y, x∧ y} where x and y are non comparable and
x∨y covers both x and y, hence x and y both covers x∧y by the above corollary,
will be called a simple diamond.
Definition 3.1.20. A subset I of a poset P is called an ideal of P if for all
x, y ∈ P ,
x ∈ I and y ≤ x imply y ∈ I.
Theorem 3.1.21. (Birkhoff) Let L be a distributive lattice with 0, and P the
poset of its nonzero join-irreducible elements. Then L is isomorphic to the lattice
of finite ideals of P , by means of the lattice isomorphism
α 7→ Iα = {τ ∈ P | τ ≤ α}, α ∈ L.
Since in the later part of the work we would be required to collect the set of all
non comparable pairs in a particular distributive lattice, L we set.
Q(L) = {(τ, ϕ), τ, ϕnoncomparable}
In the sequel, Q(L) will also be denoted by just Q when there is no possible
confusion.
We prove the following lemma to give another criterion for the join irreducibles
in terms of the language of covers.
3.1. Generalities on finite distributive lattices 30
Lemma 3.1.22.
JL = {τ ∈ L | there exists at most one cover of the form (τ, λ)}.
Proof 3.1.23. The proof is clear from the observation that if τ ∈ L covers
more than one element, let us say two of them are α1, α2, then, consider α1 ∨ α2
which is larger than both α1 and α2. But since τ ≥ α1 and τ ≥ α2, we
have τ ≥ α1 ∨ α2 ≥ α1 but since τ was a cover for α1 this means that either
α1 = α1 ∨ α2 which in turn implies that α1 ≥ α2. Once again the repeated
argument considering that τ is a cover of α2 we must have α1 = α2 contrary
to our assumption that these are two different elements. On the other hand if
α1 ∨ α2 = τ then τ being the join of these two is no longer a join irreducible
contradicting our hypothesis. This ends the proof.
For α ∈ L, let Iα be the ideal corresponding to α under the isomorphism in
Theorem 3.1.21.
Lemma 3.1.24. Let (τ, λ) be a cover in L. Then Iτ equals Iλ∪{β} for some
β ∈ JL.
Proof 3.1.25. If τ ∈ JL, then λ is the unique element covered by τ ; it is clear
that in this case that Iτ = Iλ ∪ {τ} since by definition Iτ = {α, α ≤ τ} which is
same as {τ} ∪ {α ≤ λ} = Iτ ∪ {τ}.
Let then τ 6∈ JL. By the previous lemma we know that τ must cover more than
one element. Let λ′ be another element covered by τ . Let φ = λ ∧ λ′. Now
note that φ = τ since τ covers λ and we have τ ≥ φ ≤ λ which is possible only
when τ = φ. Now by induction on the level of τ , we have, Iλ = Iφ ∪ {β}, for
some β ∈ JL (note that (λ, φ), (λ′, φ) are both covers in view of Corollary 3.1.18).
Then Iλ′ ∪{β} is an ideal contained in Iτ (note that if γ in JL is such that γ < β,
then γ is in Iλ, and hence in Iτ ). Hence if θ is the element of L corresponding to
3.2. The variety X(L) 31
the ideal Iλ′ ∪ {β} (under the bijection given by Theorem 3.1.21), then we have,
λ′ < θ ≤ τ . Hence, we obtain that θ = τ (since (τ, λ′) is a cover).
Starting point of induction: Let τ be an element of least length among all θ’s
such that θ covers an element θ′. Then the above reasoning implies clearly that
τ ∈ JL, and therefore, Iτ = Iλ ∪ {τ}.
Note that as a corollary to the above lemma we get that the length of a maximal
chain is less than equal to the number of elements in the join irreducible set. In
fact we have the following proposition.
Proposition 3.1.26. Any maximal chain in (the ranked poset) L has cardinality
equal to #JL.
Proof 3.1.27. which follows from the observation I1 equals JL along with the
lemma 3.1.24
3.2 The variety X(L)
Consider the polynomial algebra K[xα, α ∈ L]; let I(L) be the ideal generated by
{xαxβ − xα∨βxα∧β, α, β ∈ L}. Then one knows (cf.[13]) that K[xα, α ∈ L] /I(L)
is a normal integral domain; in particular, we have that I(L) is a prime ideal.
Let X(L) be the affine variety of the zeroes in Kn of I(L) (where, n = #L).
Since the ideal I(L) is a binomial ideal we can say that X(L) is a affine normal
variety defined by binomials, and hence by §we know that the variety is a Toric
embedding. That is its an affine variety with a dense torus T with an action
on X(L) extending the action of T on T . We also know that since the toric
embedding X(L) is normal it must be a toric variety associated to a normal cone.
3.2. The variety X(L) 32
To discuss the singularity of the variety X(L) we will first calculate the dimension
of X(L). Since a toric variety X contains a dense torus T the dimension of X is
same as the dimension of the torus T . To compute the dimension of the variety
X(L) we will compute the dimension of the torus T .
Before we calculate the dimension we must define a suitable combinatorial set,
which will be essential later. Let
I = {(τ, φ, τ ∨ φ, τ ∧ φ) | (τ, φ) ∈ Q}, where
Q = {(τ, φ) | τ, φ ∈ L non-comparable}.
Let Tl = (K∗)l, π : X(Tl) → X(T ) be the canonical map, given by restriction,
and for χ ∈ X(Tl), denote π(χ) by χ. Let us fix a Z-basis {χτ | τ ∈ L} for X(Tl).
For a diamond D = (τ, φ, τ ∨ φ, τ ∧ φ) ∈ I, let χD = χτ∨φ + χτ∧φ − χτ − χφ.
Lemma 3.2.1. We have
(1) X(T ) ' X(Td)/ ker π.
(2) ker π is generated by {χD | D ∈ I}.
Proof 3.2.2. The canonical map π is, in fact, surjective, since T is a subtorus of
Td. Now (1) follows from this. The assertion (2) follows from the definition of T .
Let X(JL) be the Z-span of {χθ, θ ∈ JL}. As an immediate consequence of the
above Lemma, we have
Corollary 3.2.3. π maps X(JL) isomorphically onto its image.
Proof 3.2.4. The result follows since there cannot be a diamond D contained
completely in JL.
3.2. The variety X(L) 33
3.2.5 Some more lattice theory
For α ∈ L, let Iα be the ideal corresponding to α under the isomorphism in
Theorem 3.1.21. Let us define
ψα =∑
θ∈Iα
χθ.
Taking a total order on L extending the partial order, we see easily that the
transition matrix expressing ψα’s in terms of χθ’s is triangular with diagonal
entries equal to 1. Let α1 ≥ α2 ≥ . . . . . . ≥ αn denote the total order on the
lattice L extending the partial order relation on L. Further if we call δα,β the
indicator function whether xβ occurs in ψα. That is δα,β = 1 if and only if xβ is
a summand of ψα. Then we have the following matrix:
Ψ =
ψα1
ψα2
...
ψαn
= (δα,β)
Since the matrix is upper triangular with diagonal entries equal to 1 we have the
following obvious lemma.
Lemma 3.2.6. {ψα, α ∈ L} is a Z-basis for X(Tn) the torus Tn being embedded
as an open set in the affine space An.
Clearly Since ψα is a basis for the torus Tn and the torus T embedded into it, we
get the following lemma.
If we consider the surjective (restriction) map π : X(Tn) → X(T ), and denoting
π (ψα) by ψα, we obtain
Lemma 3.2.7. The set {ψα | α ∈ L} generates X(T ) as a Z-module.
3.2. The variety X(L) 34
Since we know that rank of T is equal to the rank of the character group X(T ).
Calculating the dimension of the kernel of the surjection map will be sufficient to
calculate the dimension of the torus T .
Lemma 3.2.8. Let α ∈ L. Then ψα is in the Z-span of {χθ, θ ≤ α, θ ∈ JL}. In
particular, {χθ, θ ≤ α, θ ∈ J} generates X(T ) as a Z-module.
Proof 3.2.9. (by induction)
Case 1: α ∈ JL (may suppose α 6= 0, since if α = 0, then the result is clear).
There exists a unique β ∈ L covered by α. We have
ψα = χα + ψβ
By induction, ψβ belongs to the Z-span of {χθ, θ ≤ β, θ ∈ JL}. The result follows
from this.
Case 2: α 6∈ JL.
Hypothesis implies that there exist α1, α2 both of which are covered by α. Then
α1, α2 are non-comparable since if possible α ≥ α1 ≥ α2 will contradict the
hypothesis that α covers α2. We have α1 ∨ α2 = α; let β = α1 ∧ α2. We have
ψα = ψα1+ ψα2
− ψβ
The result follows by induction.
Combining the above Lemma with Corollary 3.2.3, we obtain
Proposition 3.2.10. The set {χτ | τ ∈ JL} is a Z-basis for X(T ).
Now, since dim X(L) = dim T , we obtain
Theorem 3.2.11. The dimension of X(L) is equal to #JL.
3.2. The variety X(L) 35
Combining the above theorem with Lemma 3.1.24 and Corollary 3.1.26, we obtain
Corollary 3.2.12. 1. The dimension of XL is equal to the cardinality of the
set of elements in a maximal chain in L.
2. Fix any chain β1 < · · · < βd, d being #JL. Let γi+1 be the element
of JL corresponding to the cover (βi+1, βi), i ≥ 1; set γ1 = β1. Then
JL = {γ1, · · · , γd}
Definition 3.2.13. For a finite distributive lattice L, we call the cardinality of
JL the dimension of L, and we denote it by dim L.
Lemma 3.2.14. Let P = (Pθ)θ∈L ∈ X(L) be such that Pτ 6= 0 for any τ ∈ JL.
Then Pθ 6= 0 for any θ ∈ L.
Proof 3.2.15. (by induction) Let θ ∈ L. If θ ∈ JL, there is nothing to check.
Let then θ ∈ L \ JL.
Let θ be a minimal element of L \ JL. This implies that every τ ∈ L such that
τ < θ belongs to JL. The fact that θ ∈ L \ JL implies that there are at least two
elements θ1, θ2 of L which are covered by θ. Note that θ1, θ2 are not comparable.
We have θ1 ∨ θ2 = θ. Let µ = θ1 ∧ θ2. We have PθPµ = Pθ1Pθ2 . Now Pθ1 6= 0,
Pθ2 6= 0, since θ1, θ2 ∈ JL. Hence we obtain that Pθ 6= 0.
Let now φ be any element of L \ JL. Assume, by induction, that Pτ 6= 0 for any
τ < φ. Since φ 6∈ JL, there are at least two elements φ1, φ2 of L which are covered
by φ. We have φ1 ∨ φ2 = φ, PφPδ = Pφ1Pφ2 , where δ = φ1 ∧ φ2. Also Pφ1 6= 0,
Pφ2 6= 0 (since φ1, φ2 are both < φ). Hence we obtain Pφ 6= 0.
We would like to list all the toric subvarieties of the toric variety X(L), towards
that we define an important class of distributive sublattices.
3.2. The variety X(L) 36
Definition 3.2.16. A sublattice L′ of L is called an embedded sublattice of L if
τ, φ ∈ L, τ ∨ φ, τ ∧ φ ∈ L′ ⇒ τ, φ ∈ L′.
Given a sublattice L′ of L, let us consider the variety X(L′), and consider
the canonical embedding X(L′) ↪→ A(L′) ↪→ A(L) (here A(L′) = A#L′ ,
A(L) = A#L).
In the following lemma we classify all the toric subvarieties of the variety X(L).
Proposition 3.2.17. X(L′) is a subvariety of X(L) if and only if L′ is an
embedded sublattice of L.
Proof 3.2.18. Under the embedding X(L′) ↪→ A(L), X(L′) can be identified
with
{(xτ )τ∈L ∈ A(L) | xτ = 0 if τ 6∈ L′, xτxφ = xτ∨φxτ∧φ for τ, φ ∈ L′ noncomparable }.
Let η′ be the generic point of X(L′). We have X(L′) ⊂ X(L) if and only if
η′ ∈ X(L).
Assume that η′ ∈ X(L). Let τ , φ be two noncomparable elements of L such that
τ ∨φ, τ ∨φ are both in L′. We have to show that τ, φ ∈ L′. If possible, let τ 6∈ L′.
This implies η′τ = 0. Hence either η′τ∨φ = 0, or η′τ∧φ = 0, since η′ ∈ X(L). But
this is not possible (note that τ ∨ φ, τ ∧ φ are in L′, and hence η′τ∨φ and η′τ∧φ are
both nonzero).
Assume now that L′ is an embedded sublattice. We have to show that η′ ∈ X(L).
Let τ , φ be two non-comparable elements of L. The fact that L′ is a sublattice
implies that if η′τ∨φ or η′τ∧φ is zero, then either η′τ , or η′φ is zero. Also, the fact
that L′ is an embedded sublattice implies that if η′τ or η′φ is zero, then either η′τ∨φ
or η′τ∧φ is zero. Further, when τ, φ, τ ∨ φ, τ ∧ φ ∈ L′,
η′τη′φ = η′τ∨φη
′τ∧φ.
3.2. The variety X(L) 37
Thus η′ satisfies the defining equations of X(L), and hence η′ ∈ X(L).
Chapter 4
A Basis for Cotangent Space
4.1 Cone and dual cone of XL:
In this section we shall determine the cone and the dual cone of XL, L being
a finite distributive lattice. As in the previous sections, denote the poset of
join-irreducibles of L by JL; let d = # JL. Denote JL = {β1, · · · , βd}. Let T
be a d-dimensional torus. Identifying T with (K∗)d, let {fβ, β ∈ JL} denote the
standard Z-basis for X(T ), namely, for u ∈ T, u = (uβ, β ∈ JL), fβ(u) = uβ. Then
K[X(T )] may be identified with K[u±1β1
, · · · , u±1βd
], the ring of Laurent polynomials.
If χ ∈ X(T ), say χ =∑
aβfβ, then as an element of K[X(T )], χ will also be
denoted by∏
uaβ
β .
Denote by I(JL) the poset of ideals of JL. For A ∈ I(JL), set
fA :=∑z∈A
fz
Recall (cf. Theorem 3.1.21) the bijection L → I(JL), θ 7→ Aθ := {τ ∈ JL | τ ≤ θ}.
Lemma 4.1.1. {fAθ, θ ∈ JL} is a Z-basis for X(T ).
38
4.1. Cone and dual cone of XL: 39
Proof 4.1.2. Take a total order on JL extending the partial order. Then the
matrix expressing {fAθ, θ ∈ JL} in terms of the basis {fθ, θ ∈ JL} is easily seen
to be triangular with diagonal entries equal to 1. The result follows from this.
Let A = {fA, A ∈ I(JL)}. Then as a consequence of the above Lemma, we obtain
Corollary 4.1.3. A generates X(T ); in particular, dA = d, dA being as in
Remark 2.4.7.
For A ∈ I(JL), denote by mA the monomial:
mA :=∏τ∈A
uτ
in K[X(T )]. If α is the element of L such that Iα = A (cf. Theorem 3.1.21), then
we shall denote mA also by mα. Consider the surjective algebra map
F : K[Xα, α ∈ L] → K[mA, A ∈ I(JL)] (⊂ K[X(T )]), Xα 7→ mA, A = Iα
Then with notation as in Definition 2.4.4, we have, F = πA, ker F = IA. Further,
Proposition 2.4.7 and Corollary 4.1.3 imply the following
Proposition 4.1.4. V(IA) = SpecK[mA, A ∈ I(JL)] and is of dimension
d(= #JL) (here, V(IA) is as in Remark 2.4.7)
Let us denote V(IA) by Y .
Lemma 4.1.5. Kernel of F is generated by {XαXβ −Xα∨βXα∧β, α, β ∈ L}
Proof 4.1.6. We have (in view of Lemma 3.1.24) that if (α, λ) is a cover, then
Iα equals Iλ∪{β} for some β ∈ JL. A repeated application of this result implies
that if γ ≤ α, and m = l(α)− l(γ) (here, l(β) denotes the level of β (cf. Remark
3.1.8)), then there exist α1 · · · , αm in JL such that Iα \ Iγ = {α1 · · · , αm}. Let
4.1. Cone and dual cone of XL: 40
now β, β′ be two non-comparable elements in L. Let α = β ∨ β′, γ = β ∧ β′. Let
l(β)− l(γ) = r, l(β′)− l(γ) = s. Then there exist β1, · · · , βr, and β′1, · · · , β′s in JL
such that
Iβ \ Iγ = {β1, · · · , βr}
Iβ′ \ Iγ = {β′1, · · · , β′s}
Iα \ Iβ = {β′1, · · · , β′s}
Iα \ Iβ′ = {β1, · · · , βr}Hence we obtain
mβ = mγuβ1 · · · uβr
mβ′ = mγuβ′1 · · ·uβ′s
mα = mβuβ′1 · · ·uβ′s
mα = mβ′uβ1 · · ·uβr
From this it follows that
mαmγ = mβmβ′
Thus for each diamond in L, i.e., a quadruple (β, β′, β ∨ β′, β ∧ β′), we have
XβXβ′ − Xβ∨β′Xβ∧β′ is in the kernel of the surjective map F . Hence F factors
through K[XL]; hence, we obtain closed immersions (of irreducible varieties):
Y ↪→ XL ↪→ A#L
(Y being V(IA)). But dimension considerations imply that Y = XL (note that in
view of Proposition 4.1.4, dimY = d = dimXL (cf. Theorem 3.2.11)), and the
result follows.
As an immediate consequence, we obtain (as seen in the proof of the above
Lemma)
Theorem 4.1.7. We have an isomorphism K[XL] ∼= K[mA, A ∈ I(JL)].
4.1. Cone and dual cone of XL: 41
Denote M := X(T ), the character group pf T . Let N = Z-dual of M , and
{ey, y ∈ JL} be the basis of N dual to {fz, z ∈ JL}. As above, for A ∈ I(JL), let
fA :=∑z∈A
fz
Let V = NR(= N⊗ZR). Let σ ⊂ V be the cone such that XL = Xσ. Let σ∨ ⊂ V ∗
be the cone dual to σ. Let Sσ = σ∨ ∩M , so that K[XL] equals the semi group
algebra K[Sσ].
As an immediate consequence of Theorem 4.1.7, we have
Proposition 4.1.8. The semigroup Sσ is generated by {fA, A ∈ I(JL)}.
Let M(JL) be the set of maximal elements in the poset JL. Let Z(JL) denote
the set of all covers in the poset JL (i.e., (z, z′), z > z′ in the poset JL, and there
is no other element y ∈ JL such that z > y > z′). For a cover (y, y′) ∈ Z(JL),
denote
vy,y′ := ey′ − ey
Proposition 4.1.9. The cone σ is generated by {ez, z ∈ M(JL), vy,y′ , (y, y′) ∈Z(JL)}.
Proof 4.1.10. Let θ be the cone generated by {ez, z ∈ M(JL), vy,y′ , (y, y′) ∈Z(JL)}. Then clearly any u in Sσ is non-negative on the generators of θ, and
hence σ∨ ⊆ θ∨. We shall now show that θ∨ ⊆ σ∨, equivalently, we shall show
that Sθ ⊆ Sσ. Let f ∈ M , say, f =∑
z∈JL
azfz. Then it is clear that f is in Sθ if
and only if
(∗) az ≥ 0, for z maximal in JL, and ax ≥ ay, for x, y ∈ JL, x < y
Claim: Let f ∈ M . Then f has property (*) if and only if f =∑
A∈I(JL)
cAfA,
4.1. Cone and dual cone of XL: 42
cA ∈ Z+.
Note that Claim implies that Sθ ⊆ Sσ, and the required result follows.
Proof of Claim: The implication ⇐ is clear.
The implication ⇒: Let f ∈ Sθ, say, f =∑
z∈JL
azfz. The hypothesis that f
has property (*) implies that az being non-negative for z maximal in JL, ax is
non-negative for all x ∈ JL. Thus f =∑
z∈JL
azfz, az ∈ Z+, z ∈ JL. Further, the
property in (*) that ax ≥ ay if x < y implies that
{x | ax 6= 0}
is an ideal in JL. Call it A; in the sequel we shall denote A also by If . Let
m = min{ax, x ∈ A}. Then either f = mfA in which case the Claim follows, or
f = mfA +f1, where f1 is in Sθ (note that f1 also has property (*)); further, If1 is
a proper subset of A. Thus proceeding we arrive at positive integers m1, · · · ,mr,
elements f1, · · · , fr (in Sθ) with Ai := Ifi, an ideal in JL, and a proper subset of
Ai−1 := Ifi−1such that
f = mfA + m1fA1 + · · ·+ mrfAr
(In fact, at the last step, we have, fr = mrfAr .) The Claim and hence the required
result now follows.
4.1.11 Analysis of faces of σ
We shall concern ourselves just with the closed points in XL. So in the sequel,
by a point in XL, we shall mean a closed point. Let τ be a face of σ. Let Pτ be
the distinguished point of Oτ with the associated maximal ideal being the kernel
4.1. Cone and dual cone of XL: 43
of the map
K[Sσ] → K,
u ∈ Sσ, u 7→
1, if u ∈ τ⊥
0, otherwise
Then for a point P ∈ XL (identified with a point in Al), denoting by P (α), the
α-th co-ordinate of P , we have,
Pτ (α) =
1, if fIα ∈ τ⊥
0, otherwise
With notation as above, let
Dτ = {α ∈ L |Pτ (α) 6= 0}
We have,
Lemma 4.1.12. Dτ is an embedded sublattice of L.
Proof 4.1.13. Let θ, δ be a pair of non-comparable elements in Dτ . The fact that
Pτ (θ), Pτ (δ) are non-zero, together with the diamond relation xθxδ = xθ∨δxθ∧δ
implies that
Pτ (θ∨ δ), Pτ (θ∧ δ) are again non-zero, and are equal to 1 (note that any non-zero
co-ordinate in Pτ is in fact equal to 1). Thus, Dτ is a sublattice of L.
The above reasoning implies that if θ, δ in Dτ form the main diagonal in a diamond
D in L, then Pτ (α), Pτ (β) are non-zero, and are equal to 1, where α, β form the
skew diagonal of D. Hence α, β are in D, and hence D ⊂ Dτ . Thus we obtain
that Dτ is an embedded sublattice of L.
Conversely, we have
4.1. Cone and dual cone of XL: 44
Lemma 4.1.14. Let D be an embedded sublattice in L. Then D determines a
unique face τ of σ such that Dτ equals D.
Proof 4.1.15. Denote P to be the point in A#L with
P (α) =
1, if α ∈ D
0, otherwise
Then P is in L (since D is an embedded sublattice in L, the co-ordinates of P
satisfy all the diamond relations). Set
u =∑
α∈D
fIα , τ = σ ∩ u⊥
Then clearly, P = Pτ and Dτ = D
Thus in view of the two Lemmas above, we have a bijection
{ faces of σ} bij↔{ embedded sublattices of L}
Proposition 4.1.16. Let τ be a face of σ. Then we have Oτ = XDτ .
Proof 4.1.17. Recall (cf. Lemma 2.4.18) that K[Oτ ] = K[Sσ ∩ τ⊥]. From the
description of Pτ , we have that τ⊥ is the span of {fIα , α ∈ Dτ}; hence
(∗) Sσ ∩ τ⊥ = {∑
α∈Dτ
cαfIα , cα ∈ Z+}
Thus we obtain
(∗∗) Sσ ∩ τ⊥ = Sστ
(here, στ is as in Lemma 2.4.18). On the other hand, if η is the cone associated to
the toric variety XDτ , then by Proposition 4.1.8 we have that Sη is the semigroup
generated by {fIα , α ∈ Dτ}. Hence we obtain that η = στ (in view of (*) and
(**)). The required result now follows.
4.2. Generation of the cotangent space at Pτ 45
4.2 Generation of the cotangent space at Pτ
For α ∈ L, let us denote the image of Xα in RL (under the surjective map
K[Xα, α ∈ L] → RL) by xα. Let Rτ = K[X(Dτ )], the co-ordinate ring of
X(Dτ ), and πτ : RL → Rτ be the canonical surjective map induced by the closed
immersion X(Dτ ) ↪→ XL; clearly, kernel of πτ is generated by {xθ, θ ∈ L \ Dτ}.Set
Fα =
xα, if α 6∈ Dτ
1− xα, if α ∈ Dτ
Denoting by Mτ , the maximal ideal in RL corresponding to Pτ , we have (cf.
§4.1.11)
Lemma 4.2.1. The ideal Mτ is generated by {Fα, α ∈ L}.
A set of generators for the cotangent space Mτ/M2τ
For F ∈ Mτ , let F denote the class of F in Mτ/M2τ .
Lemma 4.2.2. Fix a maximal chain Γ in Dτ . For any β ∈ Dτ \ Γ, we have that
in Mτ/M2τ , F β is in the span of {F γ, γ ∈ Γ}
Proof 4.2.3. Fix a β ∈ Dτ \ Γ; denote by lτ (β), the level (cf. Remark 3.1.8) of β
considered as an element of the distributive lattice Dτ . We shall prove the result
by induction on lτ (β). If lτ (β) = 0, then β coincides with the (unique) minimal
element of Dτ , and there is nothing to prove. Let then lτ (β) ≥ 1. Let β′ in Dτ
be such that β covers β′. Then Iβ \ Iβ′ = {θ}, for an unique θ ∈ JDτ (cf. Lemma
3.1.24). Then there exists a unique cover (γ, γ′) in Γ such that Iγ \ Iγ′ = {θ} (cf.
Lemma 3.1.24).
Claim: Fβ − Fβ′ ≡ Fγ − Fγ′(modMτ ).
4.2. Generation of the cotangent space at Pτ 46
First observe that the fact that θ belongs to Iβ, Iγ and does not belong to Iβ′ , Iγ′
implies
(∗) γ′ 6≥ β, β′ 6≥ γ
We now divide the proof of the Claim into the following two cases.
Case 1: γ′ < β.
This implies that γ′ is in fact less than β′, and γ < β (since, Iβ \ Iβ′ = Iγ \ Iγ′ =
{θ}, and θ 6∈ Iγ′); this in turn implies that β′ 6≤ γ (for, otherwise, β′ ≤ γ would
imply γ′ < β′ < γ, not possible, since, (γ, γ′) is a cover). Let D be the diamond
having (γ, β′) as the skew diagonal. Now γ′ being less than both γ and β′, we get
that γ′ ≤ β′ ∧ γ, and in fact equals β′ ∧ γ (note that the fact that γ′ ≤ β′ ∧ γ < γ
together with the fact that (γ, γ′) is a cover in Dτ implies that γ′ = β′ ∧ γ). In a
similar way, we obtain that β = β′ ∨ γ. Now the diamond relation xβ′xγ = xβxγ′
implies the claim in this case (by definition of Fξ’s).
Case 2: γ′ 6≤ β.
This implies that γ 6≤ β.
If γ > β, then clearly
γ = γ′ ∨ β, β′ = γ′ ∧ β
Claim in this case follows as in Case 1.
Let then γ 6≥ β. Then we have that (γ, β), (γ′, β′) are non-comparable pairs (note
that γ′ ≤ β′ if and only if γ′ < β). Denote δ = γ ∧ β, δ′ = γ′ ∧ β′. The facts that
Iδ = Iγ ∩ Iβ, Iδ′ = Iγ′ ∩ Iβ′ , Iγ = Iγ′ ∪ {θ}, Iβ = Iβ′ ∪ {θ}
imply that
Iδ = Iδ′ ∪ {θ}
4.2. Generation of the cotangent space at Pτ 47
Hence we obtain
Iγ′ ∩ Iδ = Iγ′ ∩ Iδ′ = Iδ′ ; Iγ′ ∪ Iδ = Iγ
Iβ′ ∩ Iδ = Iβ′ ∩ Iδ′ = Iδ′ ; Iβ′ ∪ Iδ = Iβ
Thus we obtain that
γ′ ∧ δ = δ′, γ′ ∨ δ = γ
β′ ∧ δ = δ′, β′ ∨ δ = β
Considering the diamonds with skew diagonals (γ′, δ), (β′, δ)respectively, the
diamond relations
xγ′xδ = xγxδ′ , xβ′xδ = xβxδ′
imply that in Mτ/M2τ , we have the following relations
F γ − F γ′ = F δ − F δ′ ; F β − F β′ = F δ − F δ′
Hence we obtain that Fβ − Fβ′ ≡ Fγ − Fγ′(modMτ ) as required.
This completes the proof of the Claim. Note that Claim implies the required
result (by induction on lτ (β)). (Note that when lτ (β) = 1, then β′ is the (unique)
minimal element of Dτ , and hence β′ ∈ Γ. The result in this case follows from
the Claim).
Under the (surjective) map πτ : RL → Rτ , denote πτ (Mτ ) by M ′τ ; then πτ induces
a surjection πτ : Mτ/M2τ → M ′
τ/M′2τ .
Corollary 4.2.4. {πτ (F γ), γ ∈ Γ} is a basis for M ′τ/M
′2τ .
Proof 4.2.5. We have that dimM ′τ/M
′2τ ≥ dimX(Dτ ) = #Γ (cf. Corollary
3.2.12(1)). On the other hand, the above Lemma implies that dimM ′τ/M
′2τ ≤ #Γ,
and the result follows.
Lemma 4.2.6. Let α ∈ L \ Dτ be such that there exists an element β ∈ Dτ and
a diamond D in L such that
4.2. Generation of the cotangent space at Pτ 48
(1) (α, β) is a diagonal (main or skew) in D.
(2) D ∩Dτ = {β}.
Then Fα ∈ M2τ .
Proof 4.2.7. Let us denote the remaining vertices of the diamond by θ, δ. Writing
the diamond relation xαxβ = xθxδ in terms of the Fξ’s, we have,
xαxβ − xθxδ = Fα(1− Fβ)− FθFδ
Hence in RL, we have,
Fα = FαFβ + FθFδ
The required result follows from this.
The set Eτ
Define
Eτ := {α ∈ L, α as in Lemma 4.2.6}
In the sequel, we shall refer to an element α ∈ Eτ as an Eτ -element.
The equivalence relation:
For two distinct elements θ, δ ∈ L \ Dτ , we say θ is equivalent to δ (and denote
it as θ ∼ δ) if there exists a sequence θ = θ1, · · · , θn = δ in L \ Dτ such that
(θi, θi+1) forms one side of a diamond in L whose other side is contained in Dτ .
For θ ∈ L \ Dτ , we shall denote by [θ], the set of all elements of L \ Dτ equivalent
to θ, if there exists such a diamond as above having θ as one vertex; if no such
diamond exists, then [θ] will denote the singleton set {θ}. Clearly, for all θ in a
given equivalence class, Fθ (mod M2τ ) is the same (by consideration of diamond
relations), and we shall denote it by F [θ] or also just F θ. Note also that in view
4.2. Generation of the cotangent space at Pτ 49
of Lemma 4.2.6, F θ = 0 in Mτ/M2τ , if [θ] ∩ Eτ 6= ∅. We shall refer to [θ] as a
Eτ -class or a non Eτ -class according as [θ] ∩ Eτ is non-empty or empty.
The sublattice Λτ (Γ): Fix a chain Γ in Dτ . Let Λτ (Γ) denote the union of all
the maximal chains in L containing Γ. Note that α in L is in Λτ (Γ) if and only
if α is comparable to every γ ∈ Γ.
Lemma 4.2.8. Λτ (Γ) is a sublattice of L.
Proof 4.2.9. Let (θ, δ) be a pair of non-comparable elements in Λτ (Γ). Then for
any γ ∈ Γ, we have (by definition of Λτ (Γ)) that either γ is less than both θ and
δ or greater than both θ and δ; in the former case, γ is less than both θ ∨ δ and
θ ∧ δ, and in the latter case, γ is greater than both θ ∨ δ and θ ∧ δ.
Remark 4.2.10. Λτ (Γ) need not be an embedded sublattice:
Take L to be the distributive lattice consisting of {(i, j), 1 ≤ i, j ≤ 4} with the
partial order (a1, a2) ≥ (b1, b2) ⇔ ar ≥ br, r = 1, 2. Take
Dτ := {(i, j), 2 ≤ i, j ≤ 3}, Γ := {(2, 2), (3, 2), (3, 3)}
Then
Λτ (Γ) = Γ ∪ I1 ∪ I2
where I1 = {(1, 1), (1, 2), (2, 1)}, I2 = {(3, 4), (4, 3), (4, 4)}. Consider x =
(2, 4), y = (4, 2); then x ∨ y = (4, 4), x ∧ y = (2, 2). Now x ∨ y, x ∧ y are in
Λτ (Γ), but x, y are not in Λτ (Γ). Thus Λτ (Γ) is not an embedded sublattice.
Proposition 4.2.11. Let θ0 ∈ L \ Dτ . Then θ0 ∼ µ, for some µ ∈ Λτ (Γ) ∪ Eτ .
Proof 4.2.12. Let us denote L′τ = L \ Dτ . By induction, let us suppose that
the result holds for all θ ∈ L′τ , θ > θ0; we shall see that the proof for the case
4.2. Generation of the cotangent space at Pτ 50
when θ0 is a maximal element in L′τ (the starting point of induction) is included
in the proof for a general θ0.
If θ0 ∈ Λτ (Γ), then there is nothing to prove. Let then θ0 6∈ Λτ (Γ). Fix γ1 minimal
in Γ such that θ0 and γ1 are non-comparable. Let ξ = γ1 ∧ θ0, θ1 = γ1 ∨ θ0. We
divide the proof into the following two cases.
Case 1: γ1 is the minimal element of Γ (note that γ1 is the minimal element of
Dτ also).
We have ξ ∈ Λτ (Γ) (since ξ < γ1); further, ξ 6∈ Dτ (again, since ξ < γ1, the
minimal element of Dτ ).
Subcase 1(a): Let θ1 be in Dτ . Then considering the diamond with
(θ1, γ1), (θ0, ξ) as opposite sides, we have, θ0 ∼ ξ, and the result follows (note
that ξ ∈ Λτ (Γ)).
Subcase 1(b): Let θ1 6∈ Dτ . This implies θ0 ∈ Eτ , and the result follows.
Case 2: γ1 is not the minimal element of Γ.
Let γ0 ∈ Γ be such that γ0 is covered by γ1 in Dτ . Then ξ ≥ γ0 (since, both θ0
and γ1 are > γ0).
Subcase 2(a): Let ξ > γ0. Then the fact that (γ1, γ0) is a cover in Dτ together
with the relation γ0 < ξ < γ1 implies that ξ ∈ Λτ (Γ) \ Γ; in particular, ξ 6∈ Dτ .
Then as in Case 1, we obtain that θ0 ∈ Eτ if θ1 6∈ Dτ , and θ0 ∼ ξ, if θ1 ∈ Dτ ; and
the result follows (again note that ξ ∈ Λτ (Γ)).
Subcase 2(b): Let ξ = γ0. We first note that θ1 6∈ Dτ ; for θ1 ∈ Dτ would imply
that θ0 ∈ Dτ (since Dτ is an embedded sublattice (cf. Lemma 4.1.12)). Hence by
induction hypothesis, we obtain that
(∗) θ1 ∼ η, for some η ∈ Λτ (Γ) ∪ Eτ
Also, considering the diamond with (θ1, θ0), (γ1, γ0) as opposite sides, we have,
4.2. Generation of the cotangent space at Pτ 51
θ0 ∼ θ1. Hence, the result follows in view of (*).
Note that the above proof includes the proof of the starting point of induction,
namely, θ0 is a maximal element in L′τ . To make this more precise, let γ1, ξ, θ1
be as above. Proceeding as above, we obtain (by maximality of θ0) that θ1 ∈ Dτ .
Hence Subcases 1(b) and 2(b) do not exist (since in these cases θ1 6∈ Dτ ). In
Subcases 1(a) and 2(a), we have ξ ∈ Λτ (Γ), and θ0 ∼ ξ. The result now follows
in this case.
Lemma 4.2.13. Let (θ1, δ1), (θ2, δ2) be two skew pairs in Λτ (Γ) such that either
θ1 ∨ δ1 = θ2 ∨ δ2 or θ1 ∧ δ1 = θ2 ∧ δ2, say θ1 ∨ δ1 = θ2 ∨ δ2, and γi := θi ∧ δi, i = 1, 2
are in Γ. Then γ1 = γ2.
Proof 4.2.14. Assume γ1 6= γ2. We have γ1, γ2 are comparable (since Γ is totally
ordered), say γ1 > γ2. Denote α := θ1∨δ1 = θ2∨δ2. The fact that any element of
Λτ (Γ) is comparable to all elements of Γ together with the hypothesis that (θ2, δ2)
is a skew pair implies that either θ2, δ2 are both > γ1 or are both < γ1. If θ2, δ2
are both > γ1, then we would obtain γ2(= θ2 ∧ δ2) ≥ γ1 which is not true. Hence
we obtain that θ2, δ2 are both < γ1; this implies that α(= θ2 ∨ δ2) < γ1 which is
again not true. Hence our assumption is wrong and the result follows.
Let Yτ (Γ) = Λτ (Γ) \ Eτ . Let us write
Yτ (Γ) = Γ∪Zτ (Γ)
Gτ (Γ) = {[θ], θ ∈ Zτ (Γ), [θ] is a non Eτ−class}
Combining the above Proposition with Lemmas 4.2.2,4.2.6, we obtain
Proposition 4.2.15. Mτ/M2τ is generated by {F [θ], [θ] ∈ Gτ (Γ)}∪{F γ, γ ∈ Γ}.
4.3. A basis for the cotangent space at Pτ 52
4.3 A basis for the cotangent space at Pτ
In this section, we shall show that {F [θ], [θ] ∈ Gτ (Γ)}∪{F γ, γ ∈ Γ} is in fact a
basis for Mτ/M2τ . We first recall some basic facts on tangent cones and tangent
spaces.
Let X = Spec R ↪→ Al be an affine variety; let S be the polynomial algebra
K[X1, · · · , Xl]. Let P ∈ X, and let MP be the maximal ideal in K[X]
corresponding to P (we are concerned only with closed points of X). Let A =
OX,P , the stalk at P ; denote the unique maximal ideal in A by M(= MP RMP).
Then Spec gr(A,M), where gr(A,M) = ⊕j∈Z+
M j/M j+1 is the tangent cone to X
at P , and is denoted TCP X. Note that
gr(R, MP ) = gr(A,M)
Tangent cone & tangent space at P : Let I(X) be the vanishing ideal of X
for the embedding X = SpecR ↪→ Al. Expanding F ∈ I(X) in terms of the local
co-ordinates at P , we have the following:
• TP (X), the tangent space to X at P is the zero locus of the linear forms of F ,
for all F ∈ I(X), i.e, the degree one part in the (polynomial) local expression for
F .
• TCP (X), the tangent cone to X at P is the zero locus of the initial forms (i.e.,
form of smallest degree) of F , for all F ∈ I(X).
In the sequel, we shall denote the initial form of F by IN(F ).
We collect below some well-known facts on singularities of X:
Facts: 1. dimTP X ≥ dim X with equality if and only if X is smooth at P .
2. X is smooth at P if and only if gr(R,MP ) is a polynomial algebra.
4.3. A basis for the cotangent space at Pτ 53
4.3.1 Determination of the degree one part of J(τ):
We now take X = XL, P = Pτ ; we shall denote I = I(XL),MP = Mτ . As above,
for F ∈ Mτ , let F denote the class of F in Mτ/M2τ . Let J(τ) be the kernel of the
surjective map
fτ : K[Xθ, θ ∈ L] → gr(R,Mτ ), Xθ 7→ Fθ
For r ∈ N, let f (r) be the restriction of f to the degree r part of the polynomial
algebra K[Xθ, θ ∈ L]. We are interested in
f (1)τ : ⊕
θ∈LKXθ → Mτ/M
2τ
We shall first describe a complement to the kernel of f(1)τ , and then deduce a basis
for Mτ/M2τ (which will turn out to be the set {F [θ], [θ] ∈ Gτ (Γ)}∪{F γ, γ ∈ Γ}).
Let Vτ be the span of {Fβ, β ∈ Dτ} (in Mτ/M2τ ); by Lemmas 4.2.2, we have that
Vτ is spanned by {F γ, γ ∈ Γ}. Let Wτ be the span of {Fθ, θ 6∈ Dτ} (in Mτ/M2τ ).
Lemma 4.3.2. {F γ, γ ∈ Γ} is a basis for Vτ .
Proof 4.3.3. The surjective map πτ : Mτ/M2τ → M ′
τ/M′2τ induces a surjection
Vτ → M ′τ/M
′2τ , and the result follows from Corollary 4.2.4.
Lemma 4.3.4. The sum Vτ + Wτ is direct.
Proof 4.3.5. Let v ∈ Vτ ; by Lemma 4.3.2, we can write v =∑γ∈Γ
aγFγ. Now if
v ∈ Wτ , then under the map
πτ : Mτ/M2τ → M ′
τ/M′2τ
we obtain that πτ (v) = 0 (since, Wτ ⊆ ker πτ ). Hence we obtain that
πτ (∑γ∈Γ
aγFγ) = 0 in M ′τ/M
′2τ ; this in turn implies that aγ = 0,∀γ (cf. Corollary
4.2.4). Hence v = 0 and the required result follows.
4.3. A basis for the cotangent space at Pτ 54
Let us write ⊕θ∈L
KXθ = Aτ ⊕Bτ where
Aτ = ⊕β∈Dτ
KXβ, Bτ = ⊕θ 6∈Dτ
KXθ
Write f(1)τ = g
(1)τ + h
(1)τ , where g
(1)τ (respectively h
(1)τ ) is the restriction of f
(1)τ to
Aτ (respectively Bτ ). Note that we have surjections
g(1)τ : Aτ → Vτ , h(1)
τ : Bτ → Wτ
As a consequence of Lemma 4.3.4, we get the following
Corollary 4.3.6. ker f(1)τ = ker g
(1)τ ⊕ ker h
(1)τ
Proof 4.3.7. The inclusion“⊇” is clear. Let then v ∈ ker f(1)τ ; write v = a + b,
where a ∈ Aτ , b ∈ Bτ . Then denoting a′ = f(1)τ (a), b′ = f
(1)τ (b), we have, 0 = a′+b′.
Also, we have that a′ ∈ Vτ , b′ ∈ Wτ . Hence in view of Lemma 4.3.4, we obtain
a′ = 0 = b′. This implies that a ∈ ker g(1)τ , b ∈ ker h
(1)τ , and the result follows.
Lemma 4.3.8. The span of {Xγ, γ ∈ Γ} is a complement to the kernel of g(1)τ .
Proof 4.3.9. We have, g(1)τ (Xβ) = Fβ. By Lemma 4.3.2, we have that {Fγ, γ ∈ Γ}
is a basis for Vτ . The result now follows.
Let {ξ1, · · · , ξr} be a complete set of representatives for Gτ (Γ).
Lemma 4.3.10. The span of {Xξ1 , · · · , Xξr} is a complement to the kernel of
h(1)τ .
Proof 4.3.11. A typical element F of I(XL) such that IN( F ) is in Bτ is of the
form
F = F1 + F2
where F1 is a linear sum of diamond relations arising from diamonds having
precisely one vertex in Dτ , and F2 is a linear sum of diamond relations arising
4.3. A basis for the cotangent space at Pτ 55
from diamonds having precisely one side in Dτ . Note that in a typical term in F1,
the linear term is of the form aαXα, aα ∈ K where α ∈ Eτ ; similarly, in a typical
term in F2, the linear term is of the form bθδ(Xθ −Xδ), θ, δ 6∈ Dτ , θ ∼ δ, bθδ ∈ K.
Hence the kernel of h(1)τ is generated by {Xα, α ∈ Eτ} ∪ {(Xθ −Xδ), θ ∼ δ}. The
required result now follows in view of Proposition 4.2.11.
Theorem 4.3.12. {F [θ], [θ] ∈ Gτ (Γ)}∪{F γ, γ ∈ Γ} is a basis for Mτ/M2τ .
Proof 4.3.13. In view of Lemmas 4.3.8, 4.3.10 and Corollary 4.3.6, we obtain
that
⊕γ∈Γ
KXγ ⊕ ⊕1≤i≤r
KXξiis a complement to the kernel of the surjective map
f(1)τ : ⊕
θ∈LKXθ → Mτ/M
2τ . The result now follows.
Let TτXL denote the tangent space to XL at Pτ . As an immediate consequence
of the above Theorem, we have the following
Corollary 4.3.14. dimTτXL = #Gτ (Γ) + #Γ.
In view of the above Corollary, we obtain that XL is singular along Oτ if and only
if #Gτ (Γ) + #Γ > #L. Let
SL = {τ < σ |#Gτ (Γ) + #Γ > #L}
(here, σ is the cone associated to XL). We obtain from Proposition 4.1.16 the
following
Theorem 4.3.15. Sing XL = ∪τ∈SL
X(Dτ ).
Chapter 5
Standard Basis for Tangent Cone
5.1 Analysis of diamond relations:
For a diamond D in L, with (θ, δ) as the skew diagonal, let
fD = XθXδ −Xθ∨δXθ∧δ
We shall refer to fD as a diamond relation.
The ideal b(τ): Let b(τ) be the ideal in K[Xα, α ∈ L] generated by ker π(1) and
the diamond relations fD′ for D′ in Λτ (Γ).
Our main goal in this section is to show that given a diamond D in L, IN fD
is in b(τ). Our first step is to collect IN fD, for all diamonds D in L. In our
discussion in this section, we will be repeatedly using the following two lemmas.
Lemma 5.1.1. Let D be a diamond with diagonals (θ1, θ4), (θ2, θ3). Suppose
either
Fθ1 ≡ Fθ2(modM2τ ), Fθ3 ≡ Fθ4(modM2
τ )
or
Fθ1 ≡ Fθ3(modM2τ ), Fθ2 ≡ Fθ4(modM2
τ )
56
5.1. Analysis of diamond relations: 57
Then fD ∈ b(τ)
Proof 5.1.2. Let us prove the case when Fθ1 ≡ Fθ3(modM2τ ), Fθ2 ≡
Fθ4(modM2τ )) (proof of the other case being similar). Let (θ2, θ3) be the skew
diagonal. Hypothesis implies that Xθ1 − Xθ3 , Xθ2 − Xθ4 are in ker π(1). The
assertion follows by writing
fD(= ±(Xθ2Xθ3 −Xθ1Xθ4)) = ±[Xθ3(Xθ2 −Xθ4)−Xθ4(Xθ1 −Xθ3))]
Lemma 5.1.3. Let D1, D2 be two diamonds with diagonals
{(θ1, θ4), (θ2, θ3)}, {(δ1, δ4), (δ2, δ3)} respectively; further, let (θ2, θ3), (δ2, δ3) be
the skew diagonals for D1, D2 respectively. Suppose Fθj≡ Fδj
(modM2τ ).
Then fD1 ≡ fD2(mod b(τ)).
Proof 5.1.4. The assertion follows by writing fD1(= Xθ2Xθ3 −Xθ1Xθ4) as
fD1 = Xθ3(Xθ2 −Xδ2)−Xθ4(Xθ1 −Xδ1)−Xδ1(Xθ4 −Xδ4) + Xδ2(Xθ3 −Xδ3) + fD2
(note that hypothesis implies that Xθj−Xδj
, j = 1, 2, 3, 4 are in ker π(1)).
As mentioned in the beginning of this section, we want to collect IN fD, for all
diamonds D in L. Since we know the description of Mτ/M2τ (and hence ker π(1)),
we shall suppose the following:
(i) In view of Lemma 4.1.12 and Lemma 4.2.4, we may suppose that any diagonal
of D is not contained in Dτ .
(ii) In view of Lemma 4.2.2, we may suppose that any side of D is not contained
in Dτ (note that in this case, in view of that Lemma, IN fD ∈ ker π(1)).
In view of (i), (ii), we may suppose that #{D ∩Dτ} ≤ 1.
(iii) Let #{D∩Dτ} = 1; let β be the vertex of D such that β ∈ Dτ . Let α be the
other vertex on the diagonal of D through β. Let (θ, δ) be the other diagonal of D.
5.1. Analysis of diamond relations: 58
We obtain that α ∈ Eτ (note that θ, δ 6∈ Dτ ), and hence IN fD = Xα(∈ ker π(1))
(cf. Lemma 4.2.6).
Thus we may suppose that D ∩Dτ 6= ∅.
(iv) In view of Lemma 4.2.6, we may suppose that any side of D is not contained
in Eτ (note that in this case, in view of that Lemma, fD ∈ b(τ)). Thus we may
suppose that # D ∩ Eτ ≤ 2.
For the rest of this section, D will denote a diamond in L satisfying
D ∩Dτ = ∅, # D ∩ Eτ ≤ 2
In particular, fD is homogeneous (and hence IN fD = fD).
Let b(τ) be as above (namely, the ideal in K[Xα, α ∈ L] generated by ker π(1)
and the diamond relations fD′ for D′ in Λτ (Γ)). Let us denote Γ by
Γ = {γn+1 > · · · > γ1}
where n is the rank of the distributive lattice Dτ . Denote γ0 := 0 (the “zero
element” of L). For 0 ≤ i ≤ n + 1, let
Λi(Γ) := {θ ∈ L | (θ, γr) comparable, r ≤ i}
We have
Λi(Γ) ⊃ Λi+1(Γ), Λn+1(Γ) = Λτ (Γ), Λ0(Γ) = L
(Note that if θ ∈ L \ Λτ (Γ), then θ ∈ Λi(Γ) where i + 1 is the smallest such that
(θ, γi+1) is skew.)
Lemma 5.1.5. Λi(Γ) is a sublattice of L.
Proof 5.1.6. Let (τ, ϕ) be a skew pair in Λi(Γ). Then for 1 ≤ r ≤ i, we have
that γr is comparable to both τ, ϕ. In fact, γr is either less than both τ, ϕ or
5.1. Analysis of diamond relations: 59
greater than both τ, ϕ; hence, τ ∨ ϕ, τ ∧ ϕ are both comparable to γr, and we
obtain that τ ∨ ϕ, τ ∧ ϕ are in Λi(Γ).
Let D be a diamond in Λi(Γ) for some i, 0 ≤ i ≤ n. Let bi+1(τ) be the ideal
in K[Xα, α ∈ L] generated by ker π(1) and the diamond relations fD′ for D′ in
Λi+1(Γ).
Proposition 5.1.7. fD is in bi+1(τ).
Before proving the Proposition, let us derive the following important consequence
(which is in fact the essence of this section as mentioned above).
Corollary 5.1.8. fD is in b(τ).
Proof 5.1.9. This is immediate from Proposition 5.1.7 by decreasing induction
on i; the starting point being i = n + 1 for which the result is clear
(note that fD ∈ bn+1(τ)(= b(τ))).
Proof of Proposition 5.1.7: Let us denote the main diagonal of D by (θ4 > θ1),
and the skew diagonal by (θ2, θ3). We break our discussion into the two cases: D
contained in Λi(Γ) \ Λi+1(Γ) and D not contained in Λi(Γ) \ Λi+1(Γ).
A. Let D be contained in Λi(Γ) \ Λi+1(Γ). Denote γ := γi+1. We have
(θj, γ), j = 1, 2, 3, 4 are skew. Denote
εj := θj ∨ γ, δj := θj ∧ γ, j = 1, 2, 3, 4
By distributivity of join, meet operations, we have
ε2 ∨ ε3 = ε4, ε2 ∧ ε3 = ε1
δ2 ∨ δ3 = δ4, δ2 ∧ δ3 = δ1
5.1. Analysis of diamond relations: 60
For simplicity, let us refer to the diamond with εj (respectively δj) as vertices
as the ε-diamond (respectively δ-diamond), whenever (ε2, ε3) (resp. (δ2, δ3)) is
skew. Note that εj, δj, j = 1, 2, 3, 4 belong to Λi+1(Γ); for, (θj, γ) being skew,
j = 1, 2, 3, 4, for any γr, r ≤ i, we have that either both θj, γ are > γr or both
< γr. Thus for r ≤ i, εj, γr (respectively δj, γr) are comparable, and in addition
εj, γ (respectively δj, γ) are comparable (note that γ = γi+1).
Also, the fact that θj 6∈ Dτ implies (by consideration of the diamond having (θj, γ)
as the skew diagonal):
(∗) At most one of {εj, δj} is in Dτ , j = 1, 2, 3, 4
(since Dτ is an embedded sublattice of L).
Case 1: Let D ∩Eτ = ∅. This implies (by (*)) that precisely one of {εj, δj} is in
Dτ for j = 1, 2, 3, 4 (by definition of Eτ ). If either one of the pairs (ε1, ε4), (ε2, ε3)
is contained in Dτ , then εj ∈ Dτ for j = 1, 2, 3, 4.
(This is clear if ε2, ε3 are comparable - for then {ε2, ε3} = {ε1, ε4}; if (ε2, ε3) is
skew, then this follows from the fact that Dτ is an embedded sublattice of L).
As a consequence we obtain (by considering the diamond with (θj, γ) as the skew
diagonal)
Fθj≡ Fδj
(modM2τ ), j = 1, 2, 3, 4
If δ2, δ3 are comparable, say δ2 > δ3, then δ2 = δ4,δ3 = δ1. Hence we obtain
Fθ2 ≡ Fθ4(modM2τ ) Fθ3 ≡ Fθ1(modM2
τ )
It follows (in view of Lemma 5.1.1) that fD ∈ b(τ)).
If (δ2, δ3) is skew, then we obtain (in view of Lemma 5.1.3) that fD ≡fD′(mod b(τ)) where D′ is the δ-diamond (⊆ Λi+1). Hence it follows that
fD ∈ bi+1(τ).
5.1. Analysis of diamond relations: 61
Interchanging the roles of the ε’s and δ’s, we obtain that if either one of the pairs
(δ1, δ4), (δ2, δ3) is contained in Dτ , then we obtain the following:
(i) fD ∈ b(τ), if ε2, ε3 are comparable.
(ii) If (ε2, ε3) is skew, then fD ≡ fD′(mod b(τ)) where D′ is the ε-diamond
(⊆ Λi+1), and fD ∈ bi+1(τ).
To discuss the remaining possibilities, we shall denote the diagonals of D by
(θj, θm), (θk, θl). In view of the fact that precisely one of {εh, δh} is in Dτ for
h = j, k, l, m, we are left to discuss the case when the following hold:
(a) precisely one of the pairs (εj, εk), (εl, εm) is contained in Dτ and the other pair
has empty intersection with Dτ .
(b) precisely one of the pairs (δj, δk), (δl, δm) is contained in Dτ and the other pair
has empty intersection with Dτ .
Let {εj, εk} be contained in Dτ . This implies that {δl, δm} is contained in Dτ . We
have
(‡) Fδj≡ Fδk
(modM2τ ), Fεl
≡ Fεm(modM2τ )
(This is clear if (εj, εm) (resp. (δj, δm)) is non-skew in which case εj = εk, εl = εm
(resp. δj = δk, δl = δm); if (εj, εm) (resp. (δj, δm)) is skew, then this follows
by consideration of the ε-diamond (resp. δ-diamond)). Also considering the
diamonds with (θr, γ), r = j, k, l, m as the skew diagonals, we have
(‡‡) Fθh≡ Fδh
(modM2τ ), h = j, k Fθh
≡ Fεh(modM2
τ ), h = l,m
Hence we obtain (from (‡), (‡‡))
Fθj≡ Fθk
(modM2τ ), Fθl
≡ Fθm(modM2τ )
From this it follows (in view of Lemma 5.1.1) that fD is in bi+1(τ).
5.1. Analysis of diamond relations: 62
This completes the discussion in the case D ∩ Eτ = ∅.
Case 2: Let #{D ∩ Eτ} = 1.
As in Case 1, let us denote the diagonals of D by (θj, θm), (θk, θl). Let θj ∈ Eτ
(and hence the other vertices are not in Eτ ). As in Case 1, we may suppose that
none of the pairs (εk, εl), (εj, εm), (δk, δl), (δj, δm) is contained in Dτ . Also, since
θh 6∈ Eτ , h 6= j, we obtain that precisely one of {εh, δh}, h 6= j is in Dτ . Hence,
precisely one of the pairs in {(εk, εm), (εl, εm), (δk, δm), (δl, δm)} is contained in Dτ ,
say, {εk, εm} is contained in Dτ ; this implies
Fεl≡ Fεj
(modM2τ )
(This is clear if εj, εm are comparable - for then εj = εl εk = εm}; if (εj, εm) is
skew, then this follows by the consideration of the ε-diamond and the fact that
εk, εm are in Dτ .) This together with the consideration of the diamonds with
(θl, γ), (θj, γ) as skew diagonals implies
(1) Fθl≡ Fεl
≡ Fεj≡ Fθj
(modM2τ )
Also, we have, δl ∈ Dτ (since precisely one of {εl, δl} is in Dτ and εl 6∈ Dτ ). Hence
we obtain the following:
If δj ∈ Dτ , then
(2) Fθk≡ Fδk
≡ Fδm ≡ Fθm(modM2τ )
(Note that if δj, δm are comparable, then δk = δm, δj = δl; if (δj, δm) is skew then
Fδk≡ Fδm ≡ Fθm(mod M2
τ )). It follows from (1), (2) and Lemma 5.1.1 that fD is
in bi+1(τ).
If δj 6∈ Dτ , then
δk ∈ Eτ , Fθk≡ Fδk
(modM2τ )
5.1. Analysis of diamond relations: 63
(note that δl ∈ Dτ , δm 6∈ Dτ ). We have Xθj, Xδk
are in ker π(1) (since θj, δk ∈ Eτ ),
and Xθk−Xδk
(and therefore Xθk) ∈ ker π(1). Hence it follows that fD ∈ bi+1(τ).
Case 3: Let #{D ∩ Eτ} = 2.
This implies (in view of (iv) above) that a diagonal of D is contained in Eτ .
As in Case 1, let us denote the diagonals of D by (θj, θm), (θk, θl), and let us
suppose that {θj, θm} is contained in Eτ . This together with the hypothesis that
#{D ∩ Eτ} = 2 implies that θk, θl 6∈ Eτ , and hence precisely one of {εk, δk}and one of {εl, δl} is in Dτ . Also, as in Case 1, we may suppose that none of
the pairs (εk, εl), (εj, εm), (δk, δl), (δj, δm) is not contained in Dτ . Hence we obtain
that precisely one of the pairs (εk, δl), (εl, δk) is contained in Dτ and the other has
empty intersection with Dτ ; say {εk, δl} is contained in Dτ . In view of the fact
that at least one of {εj, εm} is not in Dτ , we have the following possibilities:
If both εj, εm are not in Dτ , then (εj, εm) is skew necessarily (since εk ∈ Dτ ), and
we have
εl ∈ Eτ , Fθl≡ Fεl
(modM2τ )
If εm ∈ Dτ , εj 6∈ Dτ , then Fθm ≡ Fδm(modM2τ ). Further, we have
if δj 6∈ Dτ , then δk ∈ Eτ , Fθk≡ Fδk
(modM2τ )
if δj ∈ Dτ , then Fθk≡ Fδk
≡ Fδm ≡ Fθm(modM2τ )
(Note that if δj 6∈ Dτ , then (δj, δm) is skew necessarily (since δm 6∈ Dτ , δl ∈ Dτ ).
If δj ∈ Dτ , then Fδk≡ Fδm ≡ Fθm(modM2
τ ) whether (δj, δm) is skew or not). It
follows that fD ∈ bi+1(τ).
The discussion of the case εj ∈ Dτ , εm 6∈ Dτ is completely analogous (the roles of j
and m are interchanged) in the above discussion and we obtain that fD ∈ bi+1(τ).
This finishes the proof of the Proposition in case A.
5.1. Analysis of diamond relations: 64
B. Let D be not contained in Λi(Γ) \ Λi+1(Γ). As above, denote γ := γi+1.
Let us denote the skew diagonal of D by (θ2, θ3), and the main diagonal by
(θ4 > θ1). The hypothesis implies that at least one of {θ2, θ3} 6∈ Λi+1(Γ) (note
that if both {θ2, θ3} are in Λi+1(Γ), then Lemma 5.1.5 would imply that D is
contained in Λi+1(Γ), and the result follows). We divide the discussion into two
cases:
(1) precisely one of {θ2, θ3} is in Λi+1(Γ)
(2) both θ2, θ3 are not in Λi+1(Γ).
CaseB1: Precisely one of {θ2, θ3} is in Λi+1(Γ), say, θ3 ∈ Λi+1(Γ) (the proof
for the case θ2 ∈ Λi+1(Γ) being similar). This implies that θ3, γ(= γi+1) are
comparable, and the pair (θ2, γ) is skew. We shall prove the case when θ3 < γ
(the proof for the case θ3 > γ being similar). We have θ1 < γ; further
δ3 = θ3, δ1 = θ1, ε1 = ε3 = γ
(by the definition of the δi’s and the εi’s).
We will be using the following four facts:
Fact 1: ε4 = ε2
(This follows, since, ε4 = ε3 ∨ ε2 = γ ∨ ε2 = ε2.)
Fact 2: If δ2, δ3 are comparable, then δ2 < δ3(= θ3) necessarily (since δ2 < θ2,
and (θ2, θ3) is skew), and we have
δ4 = δ3(= θ3), δ2 = δ1(= θ1)
(This follows since δ4 = δ3 ∨ δ2 = δ3. Similarly we have δ2 = δ1).
Fact 3: If θ4, γ are comparable, then γ < θ4 necessarily, and we have
δ4 = γ, θ4 = ε4
5.1. Analysis of diamond relations: 65
(This follows since θ4 < γ would imply that θ2 < γ, which is not true.)
Fact 4: If δ4 6∈ Dτ , then (θ4, γ) is skew necessarily.
(This follows since θ4, γ are comparable would imply (cf. Fact 3) that δ4 ∈ Dτ .)
We now divide the discussion into the following two subcases:
Subcase B1(a): Let δ2 ∈ Dτ .
By considering the diamond with (θ2, γ) as the skew diagonal we obtain that
ε2 6∈ Dτ ; hence
(∗) Fθ2 ≡ Fε2(modM2τ )
Also Fact 2 implies that (δ2, δ3) is skew necessarily (since δ1(= θ1) 6∈ Dτ ).
If δ4 6∈ Dτ , then considering the δ-diamond, and the diamond with (θ4, γ) as skew
diagonal (note that (θ4, γ) is skew (cf. Fact 4)), we obtain
θ3(= δ3), θ4 ∈ Eτ
(note that δ2 ∈ Dτ and therefore ε4(= ε2) 6∈ Dτ )). It follows that fD is in
fD ∈ b(τ) (since Xθ3 , Xθ4 are in ker π(1)).
Let then δ4 ∈ Dτ .
Claim 1: Fθ2 ≡ Fθ4(mod M2τ ).
If θ4, γ are comparable, then θ4 = ε4 (cf. Fact 3). The Claim follows from this in
view of (*) (note that ε4 = ε2 by Fact 1).
If (θ4, γ) is skew, then
(∗∗) Fθ4 ≡ Fε4(modM2τ )
(note that δ4 ∈ Dτ ). Claim 1 follows from (*)and (**) (in view of Fact 1).
Now in view of Fact 2 and the hypothesis that δ4 ∈ Dτ , we have that (δ2, δ3) is
5.1. Analysis of diamond relations: 66
skew; hence by considering the δ-diamond, we have
(∗ ∗ ∗) Fθ1 ≡ Fθ3(mod M2τ )
(note that δ1 = θ1, δ3 = θ3, and δ2, δ4 ∈ Dτ ). It follows from Claim 1, (***), and
Lemma 5.1.1 that fD ∈ b(τ).
Subcase B1(b): Let δ2 6∈ Dτ .
(i) Let ε2(= ε4) ∈ Dτ . Then we get
(†) Fθ2 ≡ Fδ2(modM2τ )
Also, we have (cf. Fact 1)
(††) δ1 = θ1, δ3 = θ3
The hypothesis that ε4 ∈ Dτ implies that (θ4, γ) is skew (cf. Fact 3), and hence
we obtain
(† † †) Fθ4 ≡ Fδ4(modM2τ )
If δ2, δ3 are comparable, then θ3 = δ4, θ1 = δ2 (cf. Fact 2), and we obtain (by
(†), (††), († † †))Fθ2 ≡ Fθ1(modM2
τ ), Fθ4 ≡ Fθ3(modM2τ )
It follows that fD ∈ b(τ) (cf. Lemma 5.1.1).
If (δ2, δ3) is skew, then by considering the δ-diamond, we obtain in view of
(†), (††), († † †), and Lemma 5.1.3 that fD ∈ bi+1(τ).
(ii) Let ε2(= ε4) 6∈ Dτ . This implies (by considering the diamond with (θ2, γ) as
the skew diagonal)
(¦) θ2 ∈ Eτ
5.1. Analysis of diamond relations: 67
(note that by hypothesis we have δ2 6∈ Dτ ).
If θ1 ∈ Eτ , then clearly fD ∈ b(τ).
Let then θ1 6∈ Eτ .
Claim 2: θ4 ∈ Eτ .
If δ2, δ3 are comparable, then (θ4, γ) is skew necessarily (since in this case
δ4(= θ3) 6∈ Dτ (cf. Fact 2)), and θ4 ∈ Eτ (by considering the diamond with
(θ4, γ) as skew diagonal, and also noting that by hypothesis ε4 6∈ Dτ ).
If (δ2, δ3) is skew, then by considering the δ-diamond we obtain that δ4 6∈ Dτ
(since δ1(= θ1) 6∈ Eτ ). Hence we obtain that (θ4, γ) is skew necessarily, and
therefore θ4 ∈ Eτ (note that by hypothesis ε4 6∈ Dτ ).
Thus Claim 2 follows. Claim 2 together with (¦) implies that fD ∈ b(τ).
This completes the discussion in Case B1.
Case B2: θ2, θ3 6∈ Λi+1(Γ).
This implies that at least one of {θ4, θ1} is in Λi+1(Γ). We divide the discussion
into the following two cases:
(1) precisely one of {θ4, θ1} is in Λi+1(Γ)
(2) both θ4, θ1 are in Λi+1(Γ).
Subcase B2(a): Precisely one of {θ4, θ1} is in Λi+1(Γ), say, θ4 ∈ Λi+1(Γ) (the
proof for the case θ1 ∈ Λi+1(Γ) being similar). Thus θ4, γ are comparable, and
the pair (θ1, γ) is skew.
We first gather some facts:
Fact 5: We have θ4 > γ necessarily (since (θi, γ) is skew for i = 2, 3), and hence
θ4 = ε4, δ4 = γ(∈ Dτ ). Hence {ε1, ε4} is not contained in Dτ . This in turn
5.1. Analysis of diamond relations: 68
implies that {ε2, ε3} is not contained in Dτ (this is clear if ε2, ε3 are comparable -
then {ε2, ε3} = {ε1, ε4}; if (ε2, ε3) is skew, this follows from the fact that Dτ is a
sublattice of L).
Fact 6: (δ2, δ3) is skew
(Say, δ2 > δ3. We have
γ = δ4 = δ2 ∨ δ3 = δ2
which is not true, since (θ2, γ) is skew and δ2 = θ2 ∧ γ.)
In particular, we get the δ-diamond (⊆ Λi+1(Γ)).
Fact 7: If δ1 ∈ Dτ , then the δ-diamond is contained in Dτ (since δ4(= γ) ∈ Dτ ,
and Dτ is a embedded sublattice), and hence
Fθj≡ Fεj
(modM2τ ), j = 1, 2, 3, 4
If ε2, ε3 are comparable, say, ε2 > ε3, then ε2 = ε4, ε3 = ε1, and hence
Fθ2 ≡ Fθ4(modM2τ ), Fθ1 ≡ Fθ3(modM2
τ )
It follows that fD ∈ b(τ) (cf. Lemma 5.1.1).
If (ε2, ε3) is skew, then by Lemma 5.1.3, we obtain that fD ≡ fD′(mod b(τ)), D′
being the ε-diamond (⊆ Λi+1(Γ)). For similar considerations, we may suppose
{δ2, δ3} is not contained in Dτ (since Dτ is a sublattice of L).
Hence we shall suppose that δ1 6∈ Dτ , and at least one of {δ2, δ3} is not in Dτ .
Fact 8: In view of Facts 5 & 7, we may suppose that none of the pairs
(ε1, ε4), (ε2, ε3), (δ1, δ4), (δ2, δ3) is contained in Dτ .
We now divide our discussion into the following two subcases:
Subsubcase B2(a1): At least one of {θ1, θ4} is in Eτ . Let us suppose that
θ1 ∈ Eτ (the proof for the case when θ4 ∈ Eτ being similar). Hence (by our
5.1. Analysis of diamond relations: 69
general reduction that a side of D is not contained in Eτ ), we obtain that θ2, θ3
are not in Eτ . This implies that for j = 2, 3 precisely one of {εj, δj} is in Dτ . From
this we get (in view of Fact 8) that precisely one of {ε2, δ3}, {ε3, δ2} is contained
in Dτ and the other has empty intersection with Dτ ; let us suppose that {ε2, δ3}is contained in Dτ , and ε3, δ2 6∈ Dτ (proof of the other case being similar). This
implies
Fθ2 ≡ Fδ2(modM2τ )
Fθ3 ≡ Fε3(modM2τ )
Fδ1 ≡ Fδ2(modM2τ )
(by considering the diamonds with skew diagonals (θ2, γ), (θ3, γ), (δ2, δ3) (cf. Fact
6 and the hypothesis that θ2, θ3 6∈ Λi+1(Γ)) respectively, and also observing that
ε2, δ3, δ4 are in Dτ ).
If ε1 ∈ Dτ , then Fθ1 ≡ Fδ1(modM2τ ), and hence we obtain
Fθ1 ≡ Fθ2(mod M2τ )
Hence it follows fD ∈ b(τ) (since Xθ1−Xθ2 , Xθ1 ∈ ker π(1), we get Xθ2 ∈ ker π(1)).
If ε1 6∈ Dτ , then ε3 ∈ Eτ (by consideration of the ε-diamond; note that (ε2, ε3) is
skew (for,ε2, ε3 comparable would imply {ε2, ε3} = {ε1, ε4} which is not possible
since ε2 ∈ Dτ while ε1, ε4(= θ4) are not in Dτ ). This together with the fact that
Fθ3 ≡ Fε3(modM2τ ) implies that Xθ3 ∈ ker π(1). Hence we obtain that fD ∈ b(τ)
(since by hypothesis, θ1 ∈ Eτ , and hence Xθ1 ∈ ker π(1)).
Subsubcase B2(a2): Both θ1, θ4 are not in Eτ .
This implies that ε1 ∈ Dτ (since δ1 6∈ Dτ (cf. Fact 7) ), and
(1) Fθ1 ≡ Fδ1(modM2τ )
(by considering the diamond with (θ1, γ) as the skew diagonal).
5.1. Analysis of diamond relations: 70
Further, precisely one of {ε2, ε3} is in Dτ ; this is clear if ε2, ε3 are comparable (in
which case {ε2, ε3} = {ε1, ε4}, and ε4(= θ4) 6∈ Dτ , and ε1 ∈ Dτ ). If (ε2, ε3) is skew,
then in the ε-diamond, we have that ε4(= θ4) 6∈ Eτ , and ε1 ∈ Dτ .
Let us then suppose that ε2 ∈ Dτ , ε3 6∈ Dτ (proof of the other case being similar).
Hence (by considering the diamond with (θ2, γ) as the skew diagonal) we obtain
(2) Fθ2 ≡ Fδ2(modM2τ )
Also,
(3) Fε3 ≡ Fθ4(modM2τ )
(If ε2, ε3 are comparable, then the facts that ε1, ε2 ∈ Dτ , ε4(= θ4) 6∈ Dτ imply that
ε1 = ε2, ε3 = θ4, and hence (3) follows. If (ε2, ε3) is skew, then by considering the
ε-diamond and using the facts that ε1, ε2 ∈ Dτ , we obtain that Fε3 ≡ Fε4(modM2τ ),
and (3) follows since ε4 = θ4.)
If δ3 ∈ Dτ , then by considering the δ-diamond (note (δ2, δ3) is skew (cf. Fact
6))) and the diamond with (θ3, γ) as skew diagonal, we obtain (noting that
δ4(= γ) ∈ Dτ )
(4) Fδ1 ≡ Fδ2(modM2τ )
(5) Fθ3 ≡ Fε3(modM2τ )
From (1),(2),(3),(4), (5) we obtain
Fθ1 ≡ Fθ2(modM2τ ), Fθ3 ≡ Fθ4(modM2
τ )
Hence in view of Lemma 5.1.1, we obtain that fD ∈ b(τ).
If δ3 6∈ Dτ , then δ1 ∈ Eτ (note that in the δ-diamond, δ1 6∈ Dτ (cf. Fact 7),
δ2 6∈ Dτ (since ε2 ∈ Dτ ) while δ4 ∈ Dτ ); similarly, consideration of the diamond
5.1. Analysis of diamond relations: 71
with (θ3, γ) as the skew diagonal implies θ3 ∈ Eτ (note that ε3, δ3 6∈ Dτ ). The
facts that δ1, θ3 ∈ Eτ together with (1) imply that Xθ1 , Xθ3 are in ker π(1) and
hence fD ∈ b(τ).
Subcase B2(b): Both θ1, θ4 are in Λi+1(Γ) (and θ2, θ3 are not in Λi+1(Γ))
We have θ1 < γ < θ4 necessarily.
Fact 9: We have θ1 = δ1, θ4 = ε4, ε1 = γ = δ4.
As in B2(a) (cf. Fact 6), we have
Fact 10: (δ2, δ3), (ε2, ε3) are skew.
(The proof of the assertion that (ε2, ε3) is skew is similar to that of Fact 6; namely,
say ε2 > ε3. We have
γ = ε1 = ε2 ∧ ε3 = ε3
which is not true (since (θ3, γ) is skew, and ε3 = θ3 ∨ γ).
In particular, we get the δ-diamond and the ε-diamond (contained in Λi+1(Γ)).
Fact 11: In view of Facts 9 & 10, we have that none of the pairs
(ε1, ε4), (ε2, ε3), (δ1, δ4), (δ2, δ3) is contained in Dτ .
Subsubcase B2(b1): At least one of {θ1, θ4} is in Eτ .
Let us suppose that θ1 ∈ Eτ (the proof for the case when θ4 ∈ Eτ being similar).
Hence (by our general reduction that a side of D is not contained in Eτ ), we
obtain that θ2, θ3 are not in Eτ . This implies that for j = 2, 3 precisely one of
{εj, δj} is in Dτ . From this we get (in view of Fact 11) that precisely one of
{ε2, δ3}, {ε3, δ2} is contained in Dτ and the other has empty intersection with Dτ ;
let us suppose that {ε2, δ3} is contained in Dτ , and ε3, δ2 6∈ Dτ (proof of the other
5.2. Tangent cone at Pτ 72
case being similar). This implies
Fθ2 ≡ Fδ2(modM2τ )
Fδ1 ≡ Fδ2(modM2τ )
Fθ3 ≡ Fε3(modM2τ )
Fε3 ≡ Fε4(modM2τ )
(by considering the diamonds with skew diagonals (θ2, γ), (δ2, δ3), (θ3, γ), (ε2, ε3)
respectively). This combined with the fact that δ1 = θ1, ε4 = θ4 implies
Fθ2 ≡ Fθ1(modM2τ ), Fθ3 ≡ Fθ4(modM2
τ )
Hence we obtain (in view of Lemma 5.1.1) that fD ∈ b(τ).
Subcase B2(b2): Both θ1, θ4 are not in Eτ .
This implies (in view of Facts 9 & 11) that precisely one of {ε2, ε3}, and that
precisely one of {δ2, δ3} is in Dτ (by considering the ε-diamond, δ-diamond
respectively); also, consideration of the diamond with (θj, γ), j = 2, 3 as the
skew diagonal implies that for j = 2, 3 at most one of {εj, δj} is in Dτ . Hence we
obtain that precisely one of {ε2, δ3}, {ε3, δ2} is contained in Dτ and the other has
empty intersection with Dτ . The rest of the proof is as in Subcase B2(b1), and
we obtain that fD ∈ b(τ).
This completes the proof of the case B2(b) and hence that of Proposition 5.1.7.
5.2 Tangent cone at Pτ
In this section, we construct “standard monomial basis” for TCPτ XL, the tangent
cone to XL at Pτ . We first determine J(τ). We then introduce a monomial order
> and determine in J(τ), the initial ideal of J(τ). Using MaCaulay’s Theorem,
we obtain a “standard monomial basis” for TCPτ XL.
5.2. Tangent cone at Pτ 73
Let X = SpecR ↪→ Al be an affine variety. Let P ∈ X, and let MP be the
maximal ideal in K[X] corresponding to P (we are concerned only with closed
points of X). Let A = OX,P , the stalk at P ; denote the unique maximal ideal in
A by M(= MP RMP). Then Spec gr(A,M), where gr(A,M) = ⊕
j∈Z+
M j/M j+1 is
the tangent cone to X at P , and is denoted TCP X. Note that
gr(R, MP ) = gr(A,M)
Remark 5.2.1. Let I(X) be the vanishing ideal of X for the embedding
X = Spec R ↪→ Al. Expanding f ∈ I(X) in terms of the local co-ordinates
at P , we have the following:
• TP (X), the tangent space to X at P is the zero locus of the linear forms of
f, f ∈ I(X).
• TCP (X), the tangent cone to X at P is the zero locus of the initial forms (i.e.,
form of smallest degree) of f, f ∈ I(X).
In the sequel, we shall denote the initial form of f by IF (f). We have a natural
surjective map
Sym(MP /M2P ) → gr(R, MP )
(here, Sym(MP /M2P ) is the symmetric algebra of MP /M2
P ). Thus we have a
natural embedding
TCP (X) ↪→ TP (X)(= Ar)
r being the dimension of TP (X), and TCP (X) is the zero locus of the ideal I in
Sym(MP /M2P ) such that gr(R, MP ) is the quotient Sym(MP /M2
P )/I (note that
I consists of IF (f), f ∈ I(X)).
We collect below some well-known facts on singularities of X:
Facts: 1. dimTP X ≥ dim X with equality if and only if X is smooth at P .
5.2. Tangent cone at Pτ 74
2. X is smooth at P if and only if multP X equals 1.
3. X is smooth at P if and only if the canonical surjective map Sym(MP /M2P ) →
gr(R, MP ) is an isomorphism, i.e., if and only if gr(R, MP ) is a polynomial
algebra.
5.2.2 Determination of I(τ):
We now take X = XL(= SpecK[Sσ]), and P = Pτ , τ being a face of σ (notation
as in §4.1). We shall denote XL by X, K[XL] by R. Let J(τ) be the kernel of
the surjective map
π : K[Xθ, θ ∈ L] → gr(R, Mτ ), Xθ 7→ Fθ
Note that J(τ) consists of the initial forms of elements of I(XL) (I(XL) being
the ideal in K[Xθ, θ ∈ L] generated by the diamond relations). For r ∈ N,
let π(r) be the restriction of π to the degree r part of the polynomial algebra
K[Xθ, θ ∈ L]; then ker π(r) = J(τ)r, the r-th graded piece of the homogeneous
ideal J(τ). By [15], we have that ker π(1) = ker g(1)τ ⊕ker h
(1)τ , ker h
(1)τ is generated
by {Xα, α ∈ Eτ} ∪ {(Xθ − Xδ), θ ∼ δ}; further, by Lemma 4.2.11, for any
θ ∈ L \ Dτ , there exists a µ ∈ Λτ (Γ) ∪ Eτ such that θ ∼ µ. In view of this
and Corollary 5.1.8, we may work with K[Xθ, θ ∈ Λτ (Γ)], and identify J(τ) with
the kernel of the surjective map
fτ : K[Xθ, θ ∈ Λτ (Γ)] → gr(R,Mτ ), Xθ 7→ Fθ
Note that J(τ) consists of the initial forms of elements of I(XΛτ (Γ)) (the ideal in
K[Xθ, θ ∈ Λτ (Γ)] generated by the diamond relations arising from diamonds in
Λτ (Γ)).
5.2. Tangent cone at Pτ 75
We shall next describe ker π(2), deduce a set of generators for ker π(r), r ∈ N, and
then determine a set of generators for J(τ). We first start with an analysis of the
diamond relations in Λτ (Γ).
For a diamond D in Λτ (Γ), with (θ, δ) as the skew diagonal, let
fD = XθXδ −Xθ∨δXθ∧δ
Certain quadratic relations on TCP XL: We have the following three types
of elements of J(τ)2 arising from the diamond relations:
(i) Let D be such that
• D ∩ Γ = ∅.
• One vertex, say α, is in Eτ (Eτ being as in §4.2). Denote the other vertex of
the diagonal through α by ξ.
• The diagonal not containing α has empty intersection with Eτ . Denote this
diagonal by (θ, δ).
We have that fD = XθXδ − XαXξ (in local co-ordinates around Pτ ) and hence
XθXδ −XαXξ belongs to J(τ)2. Now Xα ∈ J(τ)1 (cf. Lemma 4.2.6). Hence we
obtain that XθXδ is in J(τ)2.
(ii) Let D be such that D ∩ {Γ ∪ Eτ} = ∅. Then fD remains homogeneous (in
local co-ordinates around Pτ ) and belongs to J(τ)2.
(iii) Let D1, · · · , Dr be a set of diamonds such that there exists a pair (α, γ) in
Λτ (Γ), α ∈ Eτ , γ ∈ Γ such that (α, γ) is a diagonal of Di; note that (α, γ) is
necessarily the main diagonal of Di (since α ∈ Λτ (Γ)), and the other diagonal
of Di has empty intersection with Γ (since any γ ∈ Γ can not be on the skew
diagonal of any diamond in Λτ (Γ)). Further, for each i, let the other diagonal of
Di have empty intersection with Eτ . Denote this diagonal by (θi, δi); note that
5.2. Tangent cone at Pτ 76
by Lemma 4.2.13, if r > 1, then γ is uniquely determined by α. If r > 1, let us
further suppose that D1, · · · , Dr are all the diamonds in Λτ (Γ) having (α, γ) as
the main diagonal.
(a) If r = 1 (i.e., there exists a unique skew pair (θ, δ) with {θ∨δ, θ∨δ} = {α, γ}),then denoting the corresponding diamond by D, we have IN(fD) = Xα (and
Xα ∈ J(τ)1).
(b) If r > 1, then for a pair (i, j), i 6= j, we have, IF (fDi− fDj
) ≡ XθiXδi
−Xθj
Xδj(modJ(τ)); hence we obtain that Xθi
Xδi−Xθj
Xδj∈ J(τ)2.
In the sequel, we shall refer to a D as in (i) (resp.(ii),(iii)) as a Type I (resp. Type
II,Type III ) diamond. The elements of J(τ) appearing in (i)(resp. (ii),(iii)(a),(b))
above will be referred to as a Type I-element (resp. Type II,III-elements) of J(τ).
Ideal generators for J(τ):
Let a(τ) denote the ideal in the polynomial algebra K[Xθ, θ ∈ Λτ (Γ)] generated
by elements in J(τ) of Type I,II,III (as described above). Clearly a(τ) ⊆ J(τ).
Theorem 5.2.3. The inclusion a(τ) ⊆ J(τ) is an equality.
Proof 5.2.4. Let h ∈ I(XΛτ (Γ)). We are required to show that IF (h) belongs to
a(τ). We shall first write h as a polynomial expression in the fD’s, and then find
IF (h). Now h is a polynomial expression in the fD’s, D being of Type I,II or III
of §5.2.2. Let
h = H1 + H2 + H3
where H1, H2, H3 are polynomial expression in the fD’s D’s being of Type I,II ,
III respectively. Let d be the degree of IF (h). Now for any l, the degree l forms
of H1, H2 are clearly in a(τ) (since for D of Type I,II, fD is in a(τ)). Thus we are
5.2. Tangent cone at Pτ 77
led to analyze the degree l forms in H3 (for any l). Let
(∗) H3 =∑
α∈{Λτ (Γ)∩Eτ}
∑i
Aα,i[Xα(1−Xγ)−XθiXδi
]
where (α, γ), (θi, δi) are the diagonals of diamonds of Type III. Since Xα ∈ a(τ)
for α ∈ Eτ , we are led to analyze the degree d forms in a H3 as above, where we
may suppose that none of the monomials present in Aα,i’s is divisible by any Xα′
for α′ ∈ Eτ . If∑
α∈{Λτ (Γ)∩Eτ}
∑i
Aα,iXα 6= 0
then the result is clear (since, Xα ∈ a(τ), for α ∈ Eτ ). Let then
∑
α∈{Λτ (Γ)∩Eτ}
∑i
Aα,iXα = 0
Hence we obtain
(∗∗)∑
i
Aα,i = 0, ∀α
(in view of our assumption that none of the monomials in Aα,i’s is divisible by
any Xα′ for α′ ∈ Eτ ). Hence we obtain
H3 = −∑
α∈{Λτ (Γ)∩Eτ}
∑i
Aα,i XθiXδi
Now (**) implies that if a certain monomial m occurs with a coefficient s in∑i
Aα,i, then −m also occurs with the same coefficient s in∑i
Aα,i. Hence
denoting by Mα the set of all monomials appearing in the polynomial∑
i Aα,i,
we have,
(∗ ∗ ∗) H3 =∑
α
∑
m∈Mα
m(XθiXδi
−XθjXδj
)
where (θi, δi), (θj, δj) are skew diagonals of diamonds Di, Dj in Λτ (Γ) for which
(α, γ) is the main diagonal (cf. (iii) of §5.2.2). The required result now follows
5.2. Tangent cone at Pτ 78
(note that (***) implies that H3 is in the homogeneous ideal generated by Type
III (b) elements of §5.2.2, and hence any homogeneous component - in particular,
IN H3 is again in that ideal).
A monomial order in K[Xθ, θ ∈ Λτ (Γ)]: Take a total order  on Λτ (Γ)
extending the partial order. Define Xθ Â Xδ if θ ≺ δ; a monomial will be
written as Xθ1 · · ·Xθr where θ1 º · · · º θr. Given two monomials m1,m2, define
m1 Â m2 if either deg m1 > deg m2 or deg m1 = deg m2, and m1 ≥ m2 (under
the lexicographic order induced by Â, i.e., if m1 = Xθ1 · · ·Xθr ,m2 = Xδ1 · · ·Xδr ,
then there exists a t < r such that θi = δi, i < t and θt  δt).
Let in J(τ) be the initial ideal of J(τ); note that
in J(τ) = {in f, f ∈ J(τ)}
where for a polynomial g ∈ K[Xθ, θ ∈ L], in g is the greatest (under Â) monomial
occurring in g. Let
Eτ (Γ) = Λτ (Γ) ∩ Eτ
Let notation be as in §5.2.2. Further, given an equivalence class [θ] we shall
denote the maximal element (under Â) in {δ, δ ∼ θ} by [θ]max
Proposition 5.2.5. in J(τ) is generated by
{Xθ, θ 6= [θ]max},
{Xα, α ∈ Eτ (Γ)},
{XθXδ, (θ, δ) being a diagonal of a diamond of Type I}
{XθXδ, (θ, δ) being the skew diagonal of a diamond of Type II}
{XθXδ, (θ, δ) being the skew diagonal of a diamond of Type III with r > 1,
and (θ, δ) 6= max {([θi]max, [δi]max)} }
5.2. Tangent cone at Pτ 79
(Note that in the last set (θi, δi) are all the skew diagonals such that
θi ∨ δi = θ ∨ δ, θi ∧ δi = θ ∧ δ with the main diagonal as in Type III.)
Proof 5.2.6. Let bτ be the ideal generated by the set of monomials as described
in the Proposition; clearly, we have that bτ ⊆ in J(τ). We have that in J(τ)
consists of initial monomials in the initial forms of elements of I(XΛτ (Γ)) (the
ideal in K[Xθ, θ ∈ Λτ (Γ)] generated by diamond relations arising from diamonds
in Λτ (Γ)). Now a typical element h ∈ I(XΛτ (Γ)) is a polynomial expression in the
fD’s, D being of Type I,II or III of §5.2.2. Let
h = H1 + H2 + H3
where H1, H2, H3 are polynomial expression in the fD’s D’s being of Type I,II ,
III respectively. Proceeding as in the proof of Proposition 5.2.3, we see that in h
is in bτ .
Definition 5.2.7. Call a monomial in K[Xθ, θ ∈ Λτ (Γ)] standard if it is not in
the collection as described in the above Proposition.
As an immediate consequence of MaCaulay’s theorem (cf. [6], Theorem 15.3), we
have the following
Theorem 5.2.8. The standard monomials form a basis for the ring of regular
functions on TCpτ X(L).
Theorem 5.2.9. Let τ be such that Eτ (Γ) is empty. Further, for each θ ∈ L, let
[θ] consist of just {θ}. Then TCpτ X(L) is again a Hibi variety.
Proof 5.2.10. Hypothesis implies that J(τ) is generated by just Type II
elements, namely, the diamond relations in Λτ (Γ). Thus we obtain that
TCpτ X(L) is simply the Hibi variety associated to the distributive lattice Λτ (Γ).
Bibliography
[1] M. Aigner, Combinatorial Theory, Springer-Verlag, New York, 1979.
[2] T. Hibi, Distributive lattices, affine semigroup rings, and algebras with
straightening laws,Commutative Algebra and Combinatorics, Advanced
Studies in Pure Math. 11 (1987) 93-109.
[3] Victor V. Batyrev, Ionut Ciocan-Fontanine, Bumsig Kim, Duco van Straten,
Conifold Transitions and Mirror Symmetry for Calabi-Yau Complete
Intersections in Grassmannians, Nuclear Phys. B 514 (1998), no. 3, 640-666.
[4] Victor V. Batyrev, Ionut Ciocan-Fontanine, Bumsig Kim, Duco van Straten,
Mirror Symmetry and Toric Degenerations of Partial Flag Manifolds, Acta
Math. 184 (2000), no. 1, 1-39.
[5] A. Bjorner, A. Garsia and R. Stanley, An introduction to Cohen-Macaulay
partially ordered sets, I. Rival (editor), D. Reidel Publishing Company, 1982,
583-615.
[6] D. Eisenbud, Commutative algebra with a view toward Algebraic
Geometry, Springer-Verlag, GTM, 150.
[7] D. Eisenbud and B. Sturmfels, Binomial ideals, preprint (1994).
80
5.2. Tangent cone at Pτ 81
[8] W. Fulton, Introduction to Toric Varieties, Annals of Math. Studies 131,
Princeton U. P., Princeton N. J., 1993.
[9] N. Gonciulea and V. Lakshmibai, Degenerations of flag and Schubert
varieties to toric varieties, Transformation Groups, vol 1, no:3 (1996), 215-
248.
[10] N. Gonciulea and V. Lakshmibai, Singular loci of ladder determinantal
varieties and Schubert varieties, J. Alg., 232 (2000), 360-395.
[11] N. Gonciulea and V. Lakshmibai, Schubert varieties, toric varieties and
ladder determinantal varieties , Ann. Inst. Fourier, t.47, 1997, 1013-1064.
[12] R. Hartshorne, Algebraic Geometry, Springer-Verlag, New York, 1977.
[13] T. Hibi, Distributive lattices, affine semigroup rings, and algebras with
straightening laws, Commutative Algebra and Combinatorics, Advanced
Studies in Pure Math. 11 (1987) 93-109.
[14] G. Kempf et al, Toroidal Embeddings, Lecture notes in Mathematics,N0. 339,
Springer-Verlag, 1973.
[15] V. Lakshmibai and H. Mukherjee, Singular loci of Hibi toric varieties,
(submitted to J. Ramanujan Math. Society)
[16] V. Lakshmibai and B. Sandhya, Criterion for smoothness of Schubert
varieties in SL(n)/B, Proc. Indian Acad. Sci. (Math. Sci.), 100, (1990),
45-52.
[17] B. Sturmfels, Grobner bases and convex polytopes, University Lecture series,
A.M.S., vol 8.
5.2. Tangent cone at Pτ 82
[18] D. G. Wagner, Singularities of toric varieties associated with finite
distributive lattices, preprint (1995).