You are on page 1of 50

Algebraic approach to contraction families

arXiv:2302.10867v1 [math.AG] 21 Feb 2023

Takuma Hayashi∗

Abstract
In this paper, we give a purely algebraic approach to the contraction
group scheme predicted by Bernstein–Higson–Subag and constructed by
Barbasch–Higson–Subag. We also compare quotient schemes of contrac-
tion group schemes with other related schemes, equipped with actions of
contraction group schemes.

1 Introduction
The basic idea of contractions after Inönü–Wigner in [27] is to obtain a new
Lie algebra (group) by replacing its structure constant with a parameter t and
sending it to zero. On this course, we obtain a one-parameter family of Lie
algebras (groups), which is called a contraction family.
Around 2016, Bernstein, Higson, and Subag started a project on an alge-
braic formalism for Inönü–Wigner’s contraction families ([4, 5]). In fact, they
predicted the existence of group schemes over the complex and real projective
lines for an algebraic model of the contraction families attached to groups with
involutions in [4]. They also gave examples for some classical symmetric pairs.
After that, Barbasch, Higson, and Subag constructed the group schemes in [3].
Their group schemes generalize the example of a non-reductive group scheme in
[12, Section 5]. They also proved that the proposed group schemes are smooth
([3, Proposition 4.3]), and that the fibers at t = 0, ∞ are the Cartan motion
groups ([3, Proposition 4.2]). In the real setting, they also proved in [3, The-
orem 5.1] the fiber at positive (resp. negative) real points are isomorphic to
the given real group (resp. the real form of the complexfication attached to the
composition of the given conjugate action and the involution).
Their strategy is as follows:
• Choose a faithful representation G → SLn of a complex group G to the
special linear group SLn of degree n.
• Define a new faithful representation
√ ±1 √ ±1
G ×Spec C Spec C [ t ] → SL2n ×Spec C Spec C [ t ] .
∗ Department of Pure and Applied Mathematics, Graduate School of Information Science

and Technology, Osaka University, 1-5 Yamadaoka, Suita, Osaka 565-0871, Japan, hayashi-
t@ist.osaka-u.ac.jp

1
√ √ √ ±1
• See the conjugate action defined by t ↦ − t in SL2n ×Spec C Spec C [ t ]
√ ±1
to descend G ×Spec C Spec C [ t ] to a group scheme Gt over C [t±1 ].

• Take the Zariski closure in SL2n ×Spec C P1C to obtain a group scheme G
over P1C , where P1C is the complex projective line.
• Use the classical topology to compute the fibers at t = 0, ∞ and to prove
that G is smooth over P1C .
• If G is defined over the real numbers, induce the anti-holomorphic involu-
tion on G from that on G to obtain a group scheme over the real projective
line. Its fibers are also computed by unwinding their definitions.
It is still an interesting problem to treat contraction families in an algebraic
way to let us work over general ground rings with 1/2. This allows us to iterate
the one-parameter contraction described above to obtain contraction of multi-
variables (multi-contraction). Once we prove the smoothness over a general
ground ring, we repeat this result to deduce smoothness of multi-contraction
schemes. Another point is to obtain arithmetic structures of the families. In

fact, the lack of q for positive rational numbers in general should lead to a
family of algebraic groups which are not isomorphic to a given rational algebraic
group with an involution.
The purpose of this paper is to give a purely algebraic and direct approach to
their theory: We work over general ground rings k with 1/2. We define the con-
traction algebra A over the polynomial ring k [t] explicitly for a commutative

k-algebra A with an involution θ (see Definition 2.1). In particular, the construc-
tion of A is functorial in (A, θ). As we will occasionally prove, A ⊗k[t] k [ t]
is isomorphic to the extended Rees algebra (see (6)). In other words, A can be
obtained by a certain faithfully flat descent of the extended Rees algebra (see
also [5, Sections 2.1.2 and 2.1.3] for appearance of the extended Rees algebras).
The fiber At0 of A at t = t0 for a unit t0 of k is a twisted k-form of A [x] /(x2 −t0 )
with respect to the quadratic Galois extension k [x] /(x2 − t0 ) ⊃ k by manifest:
Proposition 1.1 (Proposition 2.5 (2) and (3), Lemma 2.4 (3)). Define in-
volutions on k [x] /(x2 − t0 ) and A [x] /(x2 − t0 ) by σk ∶ a + bx ↦ a − bx and
σA ∶ a + bx ↦ θ(a) − θ(b)x respectively.
(1) The canonical homomorphism ϕ ∶ k [x] /(x2 − t0 ) → A [x] /(x2 − t0 ) com-
mutes with the involution, that is, ϕ ○ σk = σA ○ ϕ. Equivalently, σA is
semi-linear over k [x] /(x2 − t0 ) for σk , i.e., we have

σA (ϕ(α)β) = ϕ(σk (α))σA (β)

for all α ∈ k [x] /(x2 − t0 ) and β ∈ A [x] /(x2 − t0 ).


(2) There is an isomorphism of k-algebras from At0 onto the fixed point sub-
algebra of A [x] /(x2 − t0 ) by σA . Moreover, its scalar extension gives rise
to an isomorphism At0 [x] /(x2 − t0 ) ≅ A [x] /(x2 − t0 ).

2
(3) If the equation x2 = t0 admits a solution in k, then we have a k-algebra
isomorphism At0 ≅ A.
This generalizes the description of fibers of G at nonzero real points in [3,
Theorem 5.1].
Our main results are:
Theorem 1.2 (Theorems 2.17, 3.5). Suppose that A is smooth over k. Let
I ⊂ A be the ideal generated by {a ∈ A ∶ θ(a) = −a}.
(1) There is a k-algebra isomorphism A/(t) ≅ SymA/I I/I 2 , where SymA/I I/I 2
is the symmetric algebra over A/I generated by I/I 2 .
(2) The structure homomorphism k [t] → A is smooth.
(3) If A is a Hopf algebra over k, A is naturally equipped with the structure
of a Hopf algebra over k [t].

Part (1) is proved by a direct computation. We prove (2) by showing


(i) A is flat over k [t];
(ii) A ⊗k[t] k [t±1 ] is smooth over k [t±1 ];
(iii) A ⊗k[t] k [t] /(t) is smooth over k [t] /(t);
(iv) A is finitely presented over k [t].
The crucial√ step is (i). This is verified by studying a certain filtration on
A ⊗k[t] k [ t]. It is technical to check (iv). We work locally in the étale topol-
ogy of Spec A and then may and do replace k with a Noetherian ring. Then
√ ±1
the assertion follows from Hilbert’s basis theorem. The Hopf algebra structure
stated in (3) is derived from the base change of that on A to k [ t ].
Let us also note:
Proposition 1.3 (Corollary 3.12, see also Corollary 2.9 and Propositions 3.11,
A.5). Our definition agrees with that in [3].

On the course of its proof, we interpret the Zariski closure in [3] as the
scheme-theoretic closure, which leads us to a direct construction of contraction
families of real groups in the fashion of [3] for complex groups.
In our definition of contraction families, the compatibility of the complex
and real settings is verified in a more general context:
Proposition 1.4 (Propositions 2.6, 3.10). The assignment (A, θ) ↝ A and the
construction of the Hopf algebra structure in Theorem 1.2 (3) commute with flat
base change functors in a canonical way.

3
For simplicity, we only work over the polynomial ring k [t]. One can work
over the projective line like [3] if one wishes by working over k [t−1 ] and gluing
the algebras over k [t±1 ] as pointed out in [5, Remark 3.2.1] (see Remark 2.28
for details). Similar results still hold since the statements are local with respect
to the two principal open subsets of the projective line.
One may also replace k → A with any affine morphism over Z [1/2] since the
statements so far are local in the base. The affine hypothesis for the structure
morphism can not be removed since we do not have θ-stable affine open covering
in general when the base is affine. We still try to discuss the local behavior of the
contraction. Namely, we study the contraction algebra attached to a localization
of an algebra with an involution:
Proposition 1.5 (Variant 2.14). Let (A, θ) be a commutative algebra with
an involution over a commutative ring k with 1/2, and f ∈ A. If θ(f ) = f
(resp. θ(f ) = −f ) then θ extends to an involution of the localization Af . More-
over, the contraction algebra of Af is√ isomorphic to the localization of the con-
traction algebra A of A by f (resp. tf ) if θ(f ) = f (resp. θ(f ) = −f ).

We discuss this because we can glue up the affine contraction schemes locally
to give global contraction schemes if a nice θ-stable affine open covering exists
by chance (cf. Example 2.15).
Our algebraic approach allows us to apply the operation of contraction twice
if we are given mutually commutative two involutions (double contraction). We
prove an analogous result to Proposition 1.3 (see Corollary 2.11).
As an application of removal of the group structure in our construction and
the functoriality of A, we should be able to define the contraction of group
actions on schemes. This is guaranteed by the preservation of tensor products
which holds under mild assumptions:
Proposition 1.6 (Proposition 2.26, Theorem 1.2 (2), Example 2.24). The as-
signment (A, θ) ↝ A respects the tensor product if either k is a field, or A is
smooth over k.
For a digression, we compute the differential structure of the contraction
of affine group schemes and their actions (Proposition 3.8, Corollary 3.9). In
particular, we guarantee that the spectrum G = Spec A of the contraction of a
commutative Hopf algebra A is a group analog of the contraction g of the Lie
algebra of G = Spec A in [5, Section 2.1.3].
As a typical example of group actions on affine schemes in representation
theory, one can think of the contraction families attached to affine symmetric
varieties. That is, let G be a reductive algebraic group over a field F of char-
acteristic not 2 with an involution θ, and K be the fixed point subgroup by
θ. Then X = G/K is an affine variety over F , equipped with a natural involu-
tion (the Matsushima criterion). It gives rise to a contraction family X over
F [t] by taking the spectrum of the contraction algebra of the coordinate ring
of X = G/K. For example, this connects algebraic models of compact and non-
compact symmetric spaces by thinking of the Cartan involution over the field

4
R of real numbers; The manifolds of real points of the fibers at nonzero points
are not Riemannian symmetric since they are disconnected in general. Indeed,
let F s be the separable closure of F . Write H 1 (Γ, K(F s )) and H 1 (Γ, G(F s ))
for the first Galois cohomology of K and G respectively (see [33, Chapter I,
section 5.1 and Chapter II, section 1.1]). According to [7, Chapter II, Caution
of Section 6.8] and [33, Chapter I, section 5.4, Corollary 1], there is a one-to-one
correspondence between the set of G(F )-orbits in the set (G/K)(F ) of F -points
of the algebraic variety G/K and the kernel of the canonical map

H 1 (Γ, K(F s )) → H 1 (Γ, G(F s )).

For explicit computation of the Galois cohomology in the case of F = R, see [1]
for example.
Example 1.7. Put F = R and G = SLn with n ≥ 1. Set θ = ((−)T )−1 ,
where (−)T denotes the transposition of matrices. Then the fiber of X at
t = 1 (resp. t = −1) is identified with SLn / SO(n) (resp. SU(n)/ SO(n)). One
can see that (SU(n)/ SO(n))(R) = SU(n, R)/ SO(n, R). On the other hand,
(SL(n)/ SO(n))(R) has two connected components. To understand them more
concretely, put n = 2. Then (SU(2)/ SO(2))(R) is the complex projective line,
and (SL2 / SO(2))(R) is the disjoint union of the upper and lower half planes.
One can connect more symmetric varieties by thinking of other involutions.
For instance, we can connect the real affine symmetric varieties attached to
commuting Galois actions on a given complex reductive algebraic group with
respect the Galois group of the extension C/R (cf. [3, Theorem 5.2, Example
5.1]). We note that if one is still interested in symmetric spaces of real Lie
groups, a simple solution to the component issue is to take the unit component.
For example, this allows us to only pick up the upper half plane in Example 1.7.
As for a formalism to connect symmetric varieties, it is a natural question
to compare X with the fppf quotient G/K in the sense of [11]:
Theorem 1.8 (Theorem 4.6). The base point x0 = K of X naturally extends
to an F [t]-point x0 of X. Moreover, the G-orbit containing x0 is open in X
and it is isomorphic to G/K.
If F = R, we can again take the unit components of the manifolds of real
points fiberwisely to obtain the same family whose fibers at nonzero points are
symmetric spaces (Corollary 4.9).
More generally, we introduce the quotient scheme of contraction families
attached to a pair of algebraic groups with compatible involutions (the beginning
of Section 4). As another extremal example of this quotient, we discuss the case
of θ-stable parbaolic subgroups Q when G is connected reductive. We can
come up with two other possible families related to G/Q. To explain the first
candidate, let g and q be the Lie algebras of G and Q respectively. Notice that
G/Q can be identified with the G-orbit in the Grassmannian of g containing q
if G is of type (RA) in the sense of [13, Définition 5.1.6], for example, if the
characteristic of F is zero or G is connected semisimple of adjoint type (see

5
[13, Remarques 5.1.7 and Proposition 5.1.3]). As its contraction analog, we
can define the G-orbit of the Grassmannian Gr(g) of g containing q (if it is
represented by a scheme).
The second candidate is to follow the idea of [3]. I.e., we embed Gt /Qt to
Gr(g) and take the scheme-theoretic closure Gt /Qt in the sense of [28, Chapter
2, Exercise 3.17] (see Definitions A.1, A.2). It is naturally equipped with an
action of G (Proposition 4.22). They are related as follows:
Theorem 1.9 (Propositions 4.22, 4.23, Corollary 4.28). Assume the following
conditions:
(i) G is simply connected.
(ii) G and K are of type (RA).
(iii) The equality
Np (k ∩ q; q) = p ∩ q
holds (see (11) for the definition of Np (k ∩ q; q)).

Then:
(1) The closed subscheme Gt /Qt ⊂ Gr(g) is G-invariant.
(2) The G-orbit in Gr(g) attached to q coincides with G/Q. In particular,
G/Q is representable.

(3) The orbit map G/Q ↪ Gr(g) factors through Gt /Qt .


We remark that we can still work nicely without the simply connected as-
sumption, but then the centralizer is slightly different from Q in general. We
resolve this problem by taking the unit component of G in the sense of [6,
Définition 3.1] (see also [6, Cas particulier 3.4, the proof of Théorème 3.10] and
[22, Corollaire (15.6.5)]). We need the condition Np (k ∩ q; q) = p ∩ q for good
behavior of the centralizer subgroup at t = 0. We can also work over an arbitrary
ground ring with 1/2, but we need to assume a residual version of the equality
Np (k ∩ q; q) = p ∩ q. The general statement is given in Theorem 4.26.

Acknowledgments
I am grateful to Eyal Subag for the answer to my question in the online seminar
“Suita Representation Theory Seminar” on adding parameters of contraction.
Corollary 2.11 is due to him. I also thank him for pointing out some typos of a
draft of this paper.
I am grateful to Professor Hisayosi Matumoto for stimulating comments.
Thanks Yoshiki Oshima for discussions in an early stage of this work.
This work was supported by JSPS KAKENHI Grant Number 21J00023.

6
Organization of this paper
In Section 2, we study the general formalism of contraction algebras A and their
spectra X = Spec A. In Section 3, we lift the structures of Hopf algebras and
cogroup coactions on given commutative algebras with involutions to those on
contraction algebras. Their differential structures are also studied. In Section
4, we compare quotient schemes of contraction group schemes and other related
objects. In Appendix A, a short discussion on scheme-theoretic images is given.

Notation
We denote the ring of integers by Z. Let R (resp. C) denote the field of real
(resp. complex) numbers.
Let k be a commutative ring. We refer to the group of units of k as k × .
Write idk for the identity map of k.
Let A be a commutative ring, and M be an A-module. For a nonnegative
integer n, write SymnA M for the nth symmetric product of M as an A-module.
Let SymA M be the symmetric algebra of M over A.
For a commutative ring k and k-modules M and N , write Homk (M, N ) for
the k-module of k-homomorphisms from M to N .
Let k be a commutative ring. For a (small) set Λ, write AΛ k for the affine
Λ-space over k, i.e., AΛ = Spec k [xλ ∶ λ ∈ Λ]. For a finitely generated and
projective k-module V , let Gr(V ) denote the Grassmanian scheme of V , i.e.,
k

the disjoint union of the Grassmanian schemes of all ranks (see [19, Section
(8.6)]). This is a projective k-scheme by [19, Remarks 8.24, Example 13.69, and
the closed immersion (8.8.8) of Remark 8.21]. For a k-module M , we denote
the copresheaf M of abelian groups on the category of commutative k-algebras
by M (R) = M ⊗k R. We remark that M is represented by the affine k-scheme
Spec Symk Homk (M, k) if M is finitely generated and projective as a k-module.
For a scheme S, we denote its structure sheaf by OS . For a morphism
f ∶ X → Y of schemes, write f ♯ ∶ OY → f∗ OX for its structure homomorphism.

2 Contraction algebras
Let k be a Z [1/2]-algebra. Let A be a commutative k-algebra with an involution
θ. Henceforth write

Aθ = {a ∈ A ∶ θ(a) = a}, A−θ = {a ∈ A ∶ θ(a) = −a}.

Since 2 ∈ A× , we have a decomposition A = Aθ ⊕ A−θ . Write X = Spec A.


The symbol t will be the variable for the parameter of contractions.
√ ±1
Definition 2.1. (1) Write A for the k [t]-subalgebra of A [ t ] generated
by Aθ and 1

t
A−θ . We call it the contraction algebra.

7
(2) Set
√ ±1
At = Aθ [t±1 ] ⊕ √ A−θ ⊗k k [t±1 ] ⊂ A [ t ] .
1
t
This is naturally equipped with the structure of a k [t±1 ]-algebra.
(3) For a unit t0 ∈ k × , set At0 ∶= At /(t − t0 ) and Xt0 = Spec At0 .
(4) Write X = Spec A and Xt = Spec At . We call X the contraction scheme.
(5) We denote A0 = A/(t) and X 0 = Spec A0 .
We will apply similar notations to morphisms.
Though X is called a family in [4, 5, 3], we prefer to use the terminology
“scheme” in order to emphasize scheme-theoretic perspectives.
Example 2.2. Put θ = idA . Then we have A = A [t].
Example 2.3 (double contraction). Let η be another involution of A over k
commuting with θ. Let A(θ) be the contraction algebra attached to (A, θ).
Then η naturally extends to an involution of A(θ), which we denote by the
same symbol η. Write A(θ, η) for the contraction algebra attached to (A, η).
We call A(θ, η) the double contraction of A attached to (θ, η). One can define
multi-contractions in a similar way.
Let us note fundamental relations of the algebras appearing above:
Lemma 2.4. (1) We have a canonical isomorphism A ⊗k[t] k [t±1 ] ≅ At of
k [t±1 ]-algebras.
√ ±1
(2) The inclusion map At ↪ A [ t ] extends to an isomorphism

√ ±1 √ ±1
At ⊗k[t±1 ] k [ t ] ≅ A [ t ]

√ ±1
of k [ t ]-algebras.

(3) For t0 ∈ k × , we have a natural isomorphism of k [x] /(x2 − t0 )-algebras

At0 ⊗k k [x] /(x2 − t0 ) ≅ A ⊗k k [x] /(x2 − t0 ).

Proof. The k [t±1 ]-algebra A ⊗k[t] k [t±1 ] can be identified with the k [t±1 ]-
√ ±1
subalgebra of A [ t ] generated by Aθ and √1t A−θ through the canonical iso-

√ ±1 √ ±1
morphism
A [ t ] ⊗k[t] k [t±1 ] ≅ A [ t ]

since the localization map k [t] → k [t±1 ] is flat. Part (1) then follows by defini-
tion of At .

8
Part (2) is verified by the following identification:
√ ±1
A[ t ]

= Aθ [t±1 ] ⊕ √ A−θ ⊗k k [t±1 ] ⊕ √ Aθ [t±1 ] ⊕ A−θ ⊗k k [t±1 ]


1 1
t t
= At ⊕ √ At
1
t
√ ±1
≅ At ⊗k[t±1 ] k [ t ] .

√ ±1
Take its base change by the evaluation map evx ∶ k [ t ] → k [x] /(x2 −t0 ) at

t = x to obtain (3). In fact, evx is well-defined since x is a unit of k [x] /(x2 −t0 )
0 x = 1 in k [x] /(x − t0 ). Since the restriction of evx to k [t ] factors
by x ⋅ t−1 2 ±1

through the evaluation map evt0 to k ⊂ k [x] /(x − t0 ) at t = evx (t) = x2 = t0 ,


2

the base change of the left hand side in (2) is isomorphic to

At ⊗k[t±1 ] k ⊗k k [x] /(x2 − t0 ) ≅ At0 ⊗k k [x] /(x2 − t0 );

√ ±1
k [t±1 ] k[ t ]
evt0 evx

k k [x] /(x2 − t0 ).

This completes the proof.

√ ±1
There are conceptual proofs of (2) and (3) from the perspectives of Galois
theory. Recall that k [t±1 ] ⊂ k [ t ] is a quadratic Galois extension in the sense
√ √ √ ±1
of [24, Definition 1.3.1] for the involution t ↦ − t of k [ t ]. Similarly, for
a unit t0 ∈ k × , k [x] /(x2 − t0 ) is a (possibly split) quadratic Galois extension of
k for a + bx ↦ a − bx.
Proposition 2.5. Let t0 ∈ k × .
√ ±1 √i √
(1) Define an involution on A [ t ] by ∑i ai t ↦ ∑i θ(ai )(− t)i . Then
√ ±1 √ ±1
the canonical homomorphism k [ t ] → A [ t ] commutes with the in-
volution. Moreover, its fixed point subalgebra is At .
(2) Define an involution on A [x] /(x2 − t0 ) by a + bx ↦ θ(a) − θ(b)x. Then the
canonical homomorphism k [x] /(x2 − t0 ) → A [x] /(x2 − t0 ) commutes with
the involution. Moreover, its fixed point subalgebra Aθ ⊕ A−θ x is naturally
isomorphic to At0 as a commutative k-algebra.
(3) If we are given an element α ∈ k such that α2 = t0 then α determines a
k-algebra isomorphism At0 ≅ A.

9
We remark that (2) and (3) generalize the latter half of [3, Theorem 5.1].

√ ±1 √ ±1 √
Proof. Part (1) is straightforward. Take the base change of (1) by the quotient
map k [ t ] → k [ t ] /(t − t0 ), and replace the symbol t with x to deduce
(2) since base changes respect Galois extensions. We check directly as in (1)
that the fixed point subalgebra of A [x] /(x2 −t0 ) coincides with Aθ ⊕A−θ x. Part
(3) is obtained by sending x to α.
Then the isomorphisms of (2) and (3) of Lemma 2.4 follow by generalities
on the Galois descent (see [25, Theorem A.3]).
For a positive integer n ≥ 1, write (A−θ )n for the Aθ -submodule of A spanned
by products of n elements of A−θ . For convention, we set (A−θ )0 ∶= Aθ . Then it
follows by definition that A is expressed as

tn √ ±1
Aθ [t] ⊕ (⊕n≥0 A−θ √ ) ⊕ ⊕n≥2 √ n (A−θ )n ⊂ A [ t ] .
1
(1)
t t
Proposition 2.6. Let k → k ′ be a flat homomorphism of commutative Z [1/2]-
algebras. Put B = k ′ ⊗k A and η = k ′ ⊗k θ. Then there is a canonical isomorphism
k ′ ⊗k A ≅ B of k ′ [t]-algebras.
√ ±1
Proof. Since k ′ is flat over k, A ⊗k k ′ maps injectively into (A ⊗k k ′ ) [ t ].
Therefore the proof will be completed by comparing the generators. Since we
have a canonical splitting A = Aθ ⊕ A−θ , one has canonical isomorphisms

k ′ ⊗k Aθ ≅ B η , k ′ ⊗k A−θ ≅ B −η .

We next see the relation with the definition in [3]. We begin with an ele-
mentary observation from a corresponding result in ring theory. To state it, let
us introduce the following notations: For a ∈ A, write
a+θ(a) a−θ(a)
a+ = 2
, a− = 2
.

For an indexed element aλ ∈ A, we will write aλ± = (aλ )± .


Lemma 2.7. Let {aλ } be a (possibly infinite) generator of A.
(1) As a k-algebra, Aθ is generated by elements of the form aλ+ and aλ− aµ−
(λ, µ ∈ Λ).
(2) As an Aθ -module, A−θ is generated by elements of the form aλ− (λ ∈ Λ).
For a tuple I = (λ1 , λ2 , . . . , λn ) ∈ Λn (n ≥ 0), write

aI = ∏ni=1 aλi , a(I,+) = ∏ni=1 aλi + , a(I,−) = ∏ni=1 aλi −

We also put ∣I∣ = n.

10
Proof. Let a ∈ A. Choose a presentation
n
a = ∑ cI aI = ∑ cI ∏(aλ+ + aλ− ),
I∈Λn I=(λ1 ,λ2 ,...,λn )∈Λn i=1

where cI ∈ k for each tuple I. Expand the right hand side to obtain an expression

a= ∑ cIJ a(I,+) a(J,−) ,


I∈Λn ,J∈Λm

where cIJ ∈ k.
For (1), assume θ(a) = a. Then we have


a + θ(a)
a= = cIJ a(I,+) a(J,−) .
2 I,J
∣J∣ is even

For each J = (λ1 , λ2 , . . . , λ2n ), we have


n
a(J,−) = ∏(aλ2i−1 − aλ2i − ).
i=1

Therefore a is expressed by a polynomial of aλ+ and aλ− aµ− .


For (2), assume θ(a) = −a. Then we have


a − θ(a)
a= = cIJ a(I,+) a(J,−) .
2 I,J
∣J∣ is odd

For each J = (λ1 , λ2 , . . . , λ2n+1 ), write λJ = λ2n+1 and J ○ = (λ1 , λ2 , . . . , λ2n ).


Then we have


a − θ(a)
a= = cIJ a(I,+) a(J ○ ,−) aλJ − .
2 I,J
∣J∣ is odd

Since cIJ a(I,+) a(J ○ ,−) belongs to Aθ for each pair (I, J) of tuples with ∣J∣ odd,
a is expressed as an Aθ -linear combination of elements of the form aλ− .
As an immediate consequence, we obtain:
Proposition 2.8. Let {aλ }λ∈Λ be as in Lemma 2.7. Then A is generated by
the elements of the forms aλ+ and √1t aλ− as a k [t]-algebra (λ ∈ Λ).

Corollary 2.9. Choose a generator {aλ }λ∈Λ of A. Write ι ∶ X ↪ AΛ for the


corresponding closed immersion. Then there is a natural closed immersion

ι̃ ∶ X ↪ A4Λ ⊗k k [t]

satisfying the following properties:

11
(i) The map

√ ±1 √ ±1 ι̃⊗k[t] k[ t ] √ ±1
√ ±1

X ⊗k k [ t ] ≅ X ⊗k[t] k [ t ] ÐÐÐÐÐÐÐ→ A4Λ ⊗k k [ t ]

(see Lemma 2.4 for the first isomorphism) is expressed as



x ↦ ( √ −1 ),
1 ι(x) + ι(θ(x)) t(ι(x) − ι(θ(x)))
2 t (ι(x) − ι(θ(x))) ι(x) + ι(θ(x))
√ ±1
where x runs through R-points of X for k [ t ]-algebras R.

(ii) X exhibits the scheme-theoretic closure of Xt along


ι̃⊗k[t] k[t±1 ]
Xt ≅ X ⊗k[t] k [t±1 ] ÐÐÐÐÐÐ→ A4Λ ⊗k k [t±1 ] ↪ A4Λ ⊗k k [t]

(see Lemma 2.4 (1) for the first isomorphism).


Proof. We discuss the corresponding algebra homomorphisms. Write

f ∶ k [xλ ∶ λ ∈ Λ] → A

for the surjective k-algebra homomorphism defined by xλ ↦ aλ . Define

ft ∶ k [t±1 , xijλ ; 1 ≤ i, j ≤ 2, λ ∈ Λ] → At

by
ft (xiiλ ) = aλ+ (i ∈ {1, 2})
ft (x12λ ) = √t t aλ−
ft (x21λ ) = √1t aλ−

√ ±1
for λ ∈ Λ. In fact, it is clear that each of the right hand sides belongs to At . Its
base change to k [ t ] has the following identification:
√ ±1 √ ±1 √ ±1
k [ t , xijλ ; 1 ≤ i, j ≤ 2, λ ∈ Λ] → At ⊗k[t±1 ] k [ t ] ≅ A [ t ] ; (2)

xiiλ ↦ aλ+ (i ∈ {1, 2})


x12λ ↦ √t t aλ− (3)
x21λ ↦ √1t aλ− .
Consider the diagram

k [t, xijλ ; 1 ≤ i, j ≤ 2, λ ∈ Λ] k [t±1 , xijλ ; 1 ≤ i, j ≤ 2, λ ∈ Λ]


f˜ ft
(4)

A At .

The proof will be completed by showing the following assertions:

12
(I) the left vertical arrow f˜ is the diagram (4) exists;
(II) f˜ is surjective.
In fact, set ι̃ = Spec f˜. Then ι̃ is a closed immersion by (II). Condition (i) holds
by the expression (3). Condition (ii) follows from (II) and
Ker f˜ = Ker ft ∩ k [t, xijλ ; 1 ≤ i, j ≤ 2, λ ∈ Λ] .
Part (I) follows by definition of ft and A. In fact, each variable xijλ maps
to an element of A. Part (II) follows from Proposition 2.8. This completes the
proof.
Corollary 2.10. Let η be an involution on A over k commuting with θ. Write
Aθ,η = Aθ ∩ Aη , Aθ,−η = Aθ ∩ A−η , A−θ,η = A−θ ∩ Aη , A−θ,−η = A−θ ∩ A−η .
We denote the contraction parameters of A(θ) and A(θ, η) by t1 and t2 re-
√ ±1 √ ±1
spectively to distinguish the variables of the first and second contractions. Then
A(θ, η) is identified with the k [t1 , t2 ]-subalgebra of A [ t1 , t2 ] generated by

Aθ,η , Aθ,−η √ , A−θ,η √ , A−θ,−η √ √ .


1 1 1
t2 t1 t1 t2
In particular, one has a canonical isomorphism A(θ, η) ≅ A(η, θ) because of the
symmetry of the description of A(θ, η) in θ and η.
√ ±1 √ ±1 √ ±1
Proof. One may regard A(θ) [ t2 ] as a subalgebra of A [ t1 , t2 ]. There-
√ ±1
fore we may work within A(θ) [ t2 ].
Notice that A(θ) is generated by

Aθ,η , A−θ,η √ ⊂ A(θ)η


1
t1
and
Aθ,−η , A−θ,−η √ ⊂ A(θ)−η .
1
t1
The assertion now follows from Proposition 2.8.
Corollary 2.11 (Subag). Suppose that we are given an involution η on A over
k commuting with θ. Choose a closed immersion i ∶ X ↪ AΛ k as in Corollary 2.9.
Define X(θ) = Spec A(θ) and X(θ, η) = Spec A(θ, η). Let ĩθ ∶ X(θ) ↪ A4Λ k[t1 ]
be the map in Corollary 2.9. Let ĩθ,η ∶ X(θ, η) ↪ A16Λ k[t1 ,t2 ] be the map obtained
by applying ĩθ to Corollary 2.9. Then X(θ, η) exhibits the scheme-theoretic
closure of X(θ, η) ⊗k[t1 ,t2 ] k [t±1
1 , t2 ] in Ak[t±1 ,t±1 ] . The base change of ĩθ,η to
±1 16Λ

√ ±1 √ ±1 1 2

k [ t1 , t2 ] is identified with the map


√ ±1 √ ±1
√ ±1 √ ±1 ; x ↦ (xij ),
X ⊗k k [ t1 , t2 ] ↪ A16Λ
1
k[ t1 , t2 ] 4

13
where
xjj = i(x) + i(θx) + i(ηx) + i(θηx) (1 ≤ j ≤ 4)
x12 = x34√= i(x) + i(θx) + i(ηx) + i(θηx)
x13 = x√24 =√ t2 (i(x) + i(θx) − i(ηx) − i(θηx))
x14 = t1 t2 (i(x) − i(θx) − i(ηx) + i(θηx))
√ −1
x21 = x43 = t1 (i(x) − i(θx) + i(ηx) − i(θηx))
√ −1 √
x23 = t1 t2 (i(x) − i(θx) − i(ηx) + i(θηx))
√ −1
x31 = x42 = t2 (i(x) + i(θx) − i(ηx) − i(θηx))
√ √ −1
x32 = t1 t2 (i(x) − i(θx) − i(ηx) + i(θηx))
√ −1 √ −1
x41 = t1 t2 (i(x) − i(θx) − i(ηx) + i(θηx)).

Corollary 2.12. If A is of finite type over k, then so is A over k [t].


Corollary 2.13. The assignment (A, θ) ↝ A respects surjective maps.
It turns out in scheme theory that contraction of affine k-schemes with invo-
lutions respects closed immersions. For a digression, let us also prove an open
analog of Corollary 2.13:
Variant 2.14. Let f ∈ Aθ ∪ A−θ . Define f ∈ A by


⎪f (f ∈ Aθ )
f = ⎨√

⎪ (f ∈ A−θ ).

tf

(1) Write B = Af for the localization of A by f . Then θ induces an involution


on B, which we will denote by the same symbol θ.
(2) We have a canonical isomorphism Af ≅ B.
Proof. Part (1) is clear. We prove (2). Since the canonical homomorphism
p ∶ A → B respects the involution, it induces a homomorphism p ∶ A → B. We
wish to prove that p(f ) is a unit of B in order to extend p to a map α ∶ Af → B.
To see this, recall that p(f ) is a unit of B by definition of B = Af . The assertion
for f ∈ Aθ then follows since p restricts to p ∶ Aθ → B θ . Assume f ∈ A−θ . In this
case, √1t p(f )−1 belongs to B and

p(f ) √ p(f )−1 = 1.


1
t
This shows that p(f ) is a unit as desired.
√ ±1
We next construct its inverse. We may and do regard Af as a subset of
A [ t ] since localization is flat in general. The canonical homomorphism
√ ±1 √ ±1 √ ±1
f
A → A [ t ] extends to q ∶ B [ t ] → A [ t ] by definition of f . In fact,
√ ±1
f f
f is a unit of A [ t ] by definition of f .
f

14
We next prove that q∣B factors through Af . In particular, we will obtain
β ∶ B → Af . We may restrict ourselves to generators. Each element of B is
expressed as fan with a ∈ A and n ≥ 0 (put f 0 = 1). We may multiply f n to a and
f n if necessary to assume that n is even. In particular, the denominator belongs
to Aθ . Then we can easily see that fan belongs to B θ (resp. B −θ ) if and only if
there exists a positive integer m such that θ(a)f m = af m (resp. θ(a)f m = −af m ).
We may multiply f m if necessary to assume that m is even. In particular, fan
belongs to B θ (resp. B −θ ) if and only if there exists a positive even integer m
such that af m ∈ Aθ (resp. af m ∈ A−θ ). Let us take m in each case. We prove
that q ( fan ) lies in Af by case-by-case study:

af m af m
q( ) = q ( n+m ) = n+m ∈ Af
a
f n f f
a
if f ∈ Aθ and fn
∈ Bθ ;
√ n+m
af m af m t
q ( n ) = q ( n+m ) =
a
∈ Af
f f f n+m

if f ∈ A−θ and a
fn
∈ Bθ ;

af m 1 af m 1
q( √ ) ( √ ) √
a 1
= q = n+m ∈ Af
fn t f n+m t t f

if f ∈ Aθ and a
fn
∈ B −θ ;
√ n+m−1
af m 1 af m t
q ( n √ ) = q ( n+m √ ) =
a 1
∈ Af
f t f t f n+m

if f ∈ A−θ and fan ∈ B −θ (remember that n and m are even).


Finally, we prove that α and β are mutually inverse. To see that β ○ α
coincides with the identity map, we may restrict to A since the localization
map A → Af is an epimorphism of commutative rings. Then we may restrict to
the generator of A in Definition 2.1. In this case, the coincidence is evident by
construction of α and β. Conversely, we compute α ○ β. We may again restrict
to B θ ∪ √1t B −θ . Let fan ∈ B θ ∪ B −θ . We may and do assume n even and take m
as in the former paragraph. Then we have

(α ○ β) ( ) = α( n) = n
a a a
f n f f

15
a
if f ∈ Aθ and fn
∈ Bθ ;

af m
(α ○ β) ( ) (α ( )
a
= ○ β)
fn f n+m
√ n+m
⎛ af m t ⎞
⎝ f ⎠
=α n+m

√ n+m 1 1 n+m
= af m t ( √ )
f t
a
= n
f

if f ∈ A−θ and a
fn
∈ B θ since 1
f
∈ B −θ ;

af m 1
(α ○ β) ( √ ) = (α ○ β) ( n+m √ )
a 1
f n f
t t
af m 1
= α( √ n+m )
t f
af m 1
= √
t f n+m
= n√
a 1
f t

if f ∈ Aθ and a
fn
∈ B −θ ;
√ n+m−1
⎛ af m t ⎞
(α ○ β) ( n √ ) = α
a 1
f t ⎝ f n+m

√ n+m−1 1 1 n+m
= af m t ( √ )
f t

= n√
a 1
f t

if f ∈ A−θ and fan ∈ B −θ . Therefore β ○ α is the identity on the generator. This


completes the proof.

Example 2.15. Assume that k is a Z [1/2, −1]-algebra. Following [19, Corol-
lary 13.33], identify P1k with the moduli scheme of ordered pairs of generators
of k. Namely, for a commutative k-algebra R, the R-point set P1k (R) of P1k is
naturally identified with the set of isomorphism classes of locally free R-modules
L of rank 1 and pairs (a1 , a2 ) of elements of L such that a1 and a2 generate
L as R-modules. We define an involution θ on P1k by (L, a1 , a2 ) ↦ (L, a2 , −a1 ).
Define affine open immersions Spec k [wi ] ↪ P1k (i ∈ {1, 2}) by
√ √ √ √
(k [w1 ] , 1 + −1w1 , −1 + w1 ), (k [w2 ] , w2 + −1, w2 −1 + 1).

16
They are θ-stable. Moreover, the induced involutions on the two affine lines
are given by θ ∶ wi ↦ −wi . The intersection of these affine open subschemes is
Spec k [wi±1 ]. The transition map Spec k [w1±1 ] ≅ Spec k [w2±1 ] of the two affine
open subschemes on their intersection is given by w1 ↦ w2−1 . The elements wi
and wi−1 generate the coordinate rings k [wi±1 ], and they belong to k [wi±1 ] .
−θ

Therefore we can glue up contraction of these affine lines. In fact, one can
identify the contraction algebras of (Ai , θ) with k [t, vi ] by √
wi
t
↦ vi . In view of
Variant 2.14, the gluing isomorphism is given by

k [t, v1 ]tv1 ≅ k [t, v2 ]tv2 ; v1 ↦


1
tv2

1
(use the equality √tw = w2t 1t ).
2
The resulting scheme is the open subscheme of
B ∶= Proj k [t, x, y, z] /(x2 + y 2 − tz 2 )
obtained by removing a projective line at t = 0. In fact, for a commutative k-
algebra R, the R-point set of B is identified with the set of isomorphism classes of
locally free R-modules L of rank 1 and triples (a1 , a2 , a3 ) of generators of L such
that ∑3i=1 ai ⊗ ai = 0 in L ⊗R L. Define two open immersions Spec k [t, vi ] → B
by
√ √
(k [t, v1 ] , 1 + tv12 , −1(tv12 − 1), 2v1 ), (k [t, v2 ] , tv22 + 1, −1(1 − tv22 ), 2v2 ).
These maps satisfy the gluing condition. Moreover, the resulting map is an
isomorphism on t ≠ 0, and it is the open immersion from √ the affine line Spec k [v1 ]
into Proj k [x, y, z] /(x2 + y 2 ) defined by (k [v1 ] , 1, −1, v1 ) in terms of a similar
moduli description. We remark that Proj k [x, y, z] /(x2 + y 2 ) is identified with
the union of two projective lines intersecting at a single k-point by
√ √
x2 + y 2 = (x + −1y)(x − −1y).
√ √
One can catch the other affine line by replacing −1 with − −1.
As an application of our description of A, we prove corresponding results
to [3, Propositions 4.2 and 4.3] in a purely algebraic way. Let I be the ideal of
A generated by A−θ . We note that ⊕n≥0 I n /I n+1 is naturally equipped with the
structure of a graded A/I-algebra.
Proposition 2.16. There exists a natural isomorphism A0 ≅ ⊕n≥0 I n /I n+1 of
k-algebras.
Proof. It follows by the expression (1) that A0 can be identified with
⊕n≥0 (A−θ )n /(A−θ )n+2 .
Since A = Aθ ⊕ A−θ , we have
I n = (A−θ )n ⊕ (A−θ )n+1 (5)
for n ≥ 0. Therefore ⊕n≥0 (A−θ )n /(A−θ )n+2 can be identified with the graded
algebra ⊕n≥0 I n /I n+1 .

17
Theorem 2.17. Suppose that A is smooth over k.
(1) There exists an isomorphism A0 ≅ SymA/I I/I 2 of k-algebras. In particu-
lar, A0 is smooth over k.
(2) The k [t]-algebra A is smooth.
Towards the proof, we give preliminary observations from ring theory and
algebraic geometry.
Lemma 2.18. Let p ∶ B → C be a surjective homomorphism of smooth commu-
tative algebras over a Noetherian ring R. Write J = Ker p. Let B [u, Ju−1 ] be
the extended Rees algebra of (B, J), i.e.,

B [u, Ju−1 ] = B [u] ⊕ ⊕n≥1 J n u−n ⊂ B [u±1 ] .

Then the structure homomorphism R [u] → B [u, Ju−1 ] is smooth.


Proof. It suffices to check the following conditions:
(i) B [u, Ju−1 ] ⊗R[u] R [u±1 ] is smooth over R [u±1 ].

(ii) B [u, Ju−1 ] ⊗R[u] R [u] /(u) is smooth over R [u] /(u).

(iii) B [u, Ju−1 ] is flat over R [u].

(iv) B [u, Ju−1 ] is finitely presented over R [u].

Condition (i) holds since B [u, Ju−1 ] ⊗R[u] R [u±1 ] is identified with B [u±1 ].
For (ii), identify B [u, Ju−1 ]⊗R[u] R [u] /(u) and R [u] /(u) with ⊕n≥0 J n /J n+1
and R respectively. To prove that ⊕n≥0 J n /J n+1 is smooth over R, notice that
the morphism Spec p ∶ Spec C ↪ Spec B of affine schemes is a regular im-
mersion by [23, Théorème (17.12.1)]. [23, Proposition (16.9.8)] then implies
⊕n≥0 J n /J n+1 ≅ SymC J/J 2 . Since J/J 2 is finitely generated and projective as
a C-module from [23, Proposition (16.9.8)], SymC J/J 2 is a smooth C-algebra.
Since C is smooth over R, so is SymC J/J 2 .
To prove (iii), set Fn B [u, Ju−1 ] = B [u] ⊕ ⊕1≤i≤n J i u−i ⊂ B [u, Ju−1 ] for each
nonnegative integer n. It suffices to prove that Fn B [u, Ju−1 ] is flat as an R [u]-
module for every n by the passage to the direct limit.
FFor 0 ≤ m ≤ n, set Fnm B [u, Ju−1 ] = ⊕i≥−m J m ui ⊕ ⊕m+1≤i≤n J i u−i . Then we
have Fnn B [u, Ju−1 ] = ⊕i≥−n J n ui and

Fnm B [u, Ju−1 ] /Fnm+1 B [u, Ju−1 ] ≅ ⊕i≥−m J m /J m+1 ui


≅ J m /J m+1 ⊗C C [u]
C J/J ⊗C C [u]
2
≅ Symm

for 0 ≤ m ≤ n − 1 (recall the first paragraph). Since J/J 2 is a finitely generated


and projective C-module, so is Symm 2 m 2
C J/J . In particular, SymC J/J is flat as

18
an C-module. Therefore Fnm B [u, Ju−1 ] /Fnm+1 B [u, Ju−1 ] is flat as an C [u]-
module. Since C is flat over R, Fnm B [u, Ju−1 ] /Fnm+1 B [u, Ju−1 ] is flat as an
R [u]-module. One can also prove by induction that J m is also a flat R-module
for 0 ≤ m ≤ n. In fact, if m = 0 then the assertion holds since J 0 = B is smooth
over R. For 1 ≤ m ≤ n, consider the short exact sequence

0 → J m → J m−1 → J m−1 /J m → 0.

The induction hypothesis implies that J m−1 is flat as an R-module. We also


proved that J m−1 /J m ≅ Symm−1 J/J 2 is flat over R. Therefore J m is flat as a
R-module, and the induction proceeds. We conclude that Fnn B [u, Ju−1 ] and
C

Fnm B [u, Ju−1 ] /Fnm+1 B [u, Ju−1 ] are flat as R [u]-modules. A similar descend-
ing induction implies Fn0 B [u, Ju−1 ] = Fn B [u, Ju−1 ] is flat as an R [u]-module.
Finally, we prove (iv). It is evident by definition that B [u, Ju−1 ] is generated
by B and Jt−1 as an R [u]-algebra. Since B is smooth over R, B is of finite type
over R. Therefore one can find a finite set {ci } of generators of B over R. Since R
is Noetherian, so is B by Hilbert’s basis theorem. Therefore one can find a finite
set {c′j } of generators of J as a B-module. It is easy to show that {ci } ∪ {c′j u−1 }
generates B [u, Ju−1 ] as an R [u]-algebra. In particular, B [u, Ju−1 ] is of finite
type over R [u]. The assertion then follows from Hilbert’s basis theorem.
We next make an attempt to remove the Noether hypothesis on R. To
clarify the scheme-theoretic considerations below, we interpret and generalize
the former lemma to a scheme-theoretic statement.
Lemma 2.19. Let S be a scheme, and j ∶ Z ↪ Y be a closed immersion of
smooth S-schemes with the ideal sheaf J ⊂ OY . Let OY [u, Ju−1 ] be the extended
Rees algebra of (OY , J), i.e.,

OY [u, Ju−1 ] = OY [u] ⊕ ⊕n≥1 Jn u−n ⊂ OZ [u±1 ] .

Then Spec OY [u, Ju−1 ] is smooth over Spec OS [u].


Since J is a quasi-coherent OY -module, and the collection of quasi-coherent
OY -modules is closed under formation of small colimits and finite limits,

OY [u, Ju−1 ]

is a quasi-coherent OY -algebra. Therefore Spec OY [u, Ju−1 ] makes a sense.


Proof. It is easy to show that Jn respects the flat base changes for n ≥ 0. Namely,
suppose that we are given a Cartesian diagram

j′
Z′ Y′
p′ p
j
Z Y,

19
where p is flat. Let J′ ⊂ OY ′ be the ideal attached to the closed immersion j ′ .
Then we have a canonical isomorphism p∗ Jn ≅ (J′ )n for n ≥ 0. We may therefore
work locally in the étale (flat) topology of Y by definition of the extended Rees
algebra.
Let U be the complementary open subscheme to Z in Y . Then we have
C ′ ∣U ≅ OY [t±1 ] since J∣U = OY ∣U . This implies that Spec OY [u, Ju−1 ] ×Y U is
smooth over Spec OS [u]. We next work on an open neighborhood of Z in Y .
In this case, we may replace Z ↪ Y with the standard closed immersion

Spec OS [u1 , . . . , um ] ↪ Spec OS [u1 , u2 , . . . , un ]

of affine spaces over S for some m ≤ n by [23, Corollaire (17.12.2)]. In this case,
we may replace S with Spec Z by the explicit description of J. The assertion
now follows from Lemma 2.18.
Proof of Theorem 2.17. Recall that A/I is a smooth k-algebra by [24, the proof
of Lemma 3.1]. Then (1) follows from the same line as the third paragraph of
the proof of Lemma 2.18 (recall Proposition 2.16).
√ (2). In virtue of the
We next prove √ faithfully flat descent, we may verify
that A ⊗k[t] k [ t] is smooth over k [ t]. A similar argument to Lemma 2.4
√ √ √ −n
(2) implies
A ⊗k[t] k [ t] ≅ A [ t] ⊕ ⊕n≥1 I n t (6)
(use the equalities (1) and (5)). The assertion now follows from Lemma 2.19.
As we can easily see in examples, the fiber of X at t = 0 is disconnected in
general even if X is connected. This may sometimes cause difficulties in analysis
of geometric structures of X (cf. Section 4). We can resolve this problem in
practice by taking connected components fiberwisely. To be more precise, let x0
be a θ-invariant k-point of X. Regard x0 as a homomorphism A → k and take
its contraction to obtain a k [t]-point x0 of X. If X is smooth, we can apply
[22, Corollaire (15.6.5)] to (X, x0 ) to obtain an open subscheme of X which we
will denote by X ○ since X is smooth and affine over k [t]. Similarly, we define
(X θ )○ . We remark that these notations will not be confusing in this paper since
x0 is clear from the context. In fact, the scheme X ○ will appear in Section 4,
and all of x0 there will arise from the unit of a smooth affine group scheme.
One can literally apply the argument of [12, Section 5.12] to X to obtain:
Proposition 2.20. Assume the following conditions:
(i) X is smooth over k.
(ii) The fibers of X are connected.
(iii) The open subscheme (X θ )○ ⊂ X θ is closed.
Then the open subscheme X ○ ⊂ X attached to x0 is affine.
We end this section with study of preservation of tensor products.

20
Construction 2.21. For commutative k-algebras with involutions (A, θ) and
(B, η), define a canonical homomorphism
√ ±1 √ ±1
iA,B ∶ A ⊗k[t] B → A [ t ] ⊗k[√t±1 ] B [ t ]

by the componentwise embedding:


√ ±1
A ⊗k[t] B → (A ⊗k[t] B) ⊗k[t] k [ t ]
√ ±1 √ ±1
≅ (A ⊗k[t] k [ t ]) ⊗k[√t±1 ] (B ⊗k[t] k [ t ])
√ ±1 √ ±1
≅ A [ t ] ⊗k[√t±1 ] B [ t ] ,

where the first map is given by the unit. We define


√ ±1 √ ±1 √ ±1
iA,B,C ∶ A ⊗k[t] B ⊗k[t] C → A [ t ] ⊗k[√t±1 ] B [ t ] ⊗k[√t±1 ] C [ t ]

for an additional commutative k-algebra with involution (C, ζ) in a similar way.


Condition 2.22. The structure homomorphism k [t] → A is flat.
Example 2.23. Condition 2.22 holds if A is smooth over k by Theorem 2.17

√ ±1
Example 2.24. Condition 2.22 holds if k = F is a field. In fact, regard A as an
F [t]-submodule of A [ t ] to see that A is a torsion-free F [t]-module. Since
F [t] is a PID, A is flat over F [t].
Example 2.25. Suppose that Condition 2.22 is satisfied. Then for a flat ho-
momorphism k → k ′ of commutative rings, the contraction algebra for A ⊗k k ′
satisfies Condition 2.22 by Proposition 2.6.
Proposition 2.26. For i ∈ {1, 2, 3}, let (Ai , θi ) be commutative k-algebras with
involutions. Set
(Aij , θij ) = (Ai ⊗k Aj , θi ⊗k θj )
for (i, j) ∈ {1, 2, 3}2, and

(A123 , θ123 ) = (A1 ⊗k A2 ⊗k A3 , θ1 ⊗k θ2 ⊗k θ3 ).

(1) The composition of the canonical isomorphism


√ ±1 √ ±1 √ ±1
A1 [ t ] ⊗k[√t±1 ] A2 [ t ] ≅ A12 [ t ] (7)

with iA1 ,A2 is surjective onto A12 . We denote the resulting map

A1 ⊗k[t] A2 ↠ A12

by IA1 ,A2 .

21
(2) Th map IA1 ,A2 is natural in both A1 and A2 .
(3) The compositions of iA1 ,A2 ,A3 with the canonical isomorphisms
√ ±1 √ ±1 √ ±1
A1 [ t ] ⊗k[√t±1 ] A2 [ t ] ⊗k[t] A3 [ t ]
√ ±1 √ ±1
≅ A12 [ t ] ⊗k[√t±1 ] A3 [ t ]
√ ±1
≅ A123 [ t ] ,

√ ±1 √ ±1 √ ±1
A1 [ t ] ⊗k[√t±1 ] A2 [ t ] ⊗k[t] A3 [ t ]
√ ±1 √ ±1
≅ A1 [ t ] ⊗k[√t±1 ] A23 [ t ]
√ ±1
≅ A123 [ t ]

coincide with the compositions of the surjective maps


IA1 ,A2 ⊗k[t] A3 IA12 ,A3
A1 ⊗k[t] A2 ⊗k[t] A3 ÐÐÐÐÐÐÐÐ→ A12 ⊗k[t] A3 → A123 ,

A1 ⊗k[t] IA2 ,A3 IA1 ,A23


A1 ⊗k[t] A2 ⊗k[t] A3 ÐÐÐÐÐÐÐÐ→ A1 ⊗k[t] A23 → A123
√ ±1
with the inclusion A123 ⊂ A123 [ t ] respectively.

(4) Let C ∶ A12 ≅ A21 denote the canonical isomorphism. Then the diagram

A1 ⊗k[t] A2 A2 ⊗k[t] A1
IA1 ,A2 IA2 ,A1
C
A12 A21

√ ±1 √ ±1
√ ±1
C⊗k k[ t ]
A12 [ t ] A21 [ t ]

commutes, where the upper horizontal arrow is the canonical isomorphism.


(5) If one of (A1 , θ1 ) and (A2 , θ2 ) satisfies Condition 2.22 then iA1 ,A2 is in-
jective. In particular, IA1 ,A2 is an isomorphism of k [t]-algebras.
(6) If two of (A1 , θ1 ), (A2 , θ2 ), and (A3 , θ3 ) satisfy Condition 2.22 then iA1 ,A2 ,A3
is injective.

22
(7) Put A2 = k and θ2 = idk . Write

r ∶ A1 ⊗k k ≅ A1
√ ±1 √ ±1 √ ±1
l ∶ k ⊗k A3 ≅ A3
µA1 ,k ∶ A1 [ t ] ⊗k[√t±1 ] k [ t ] ≅ A12 [ t ]
√ ±1 √ ±1 √ ±1
µk,A3 ∶ k [ t ] ⊗k[√t±1 ] A3 [ t ] ≅ A23 [ t ]

for the canonical isomorphisms. Then the diagrams

A1 ⊗k[t] k [t]
IA1 ,k r
A12 A1
iA1 ,k

√ ±1 √ ±1 √ ±1 r[ √ ±1
√ ±1
t ]
A1 [ t ] ⊗k[√t±1 ] k [ t ] A12 [ t ] A1 [ t ]
µA1 ,k

k [t] ⊗k[t] A3
Ik,A3 l
A23 A3
ik,A3

√ ±1 √ ±1 √ ±1 √ ±1
√ ±1
l[ t ]
k [ t ] ⊗k[√t±1 ] A3 [ t ] A23 [ t ] A3 [ t ]
µk,A3

commute.
Proof. For (1), it will suffice to compare the generators

Aθ11 ⊗ 1, √ A−θ ⊗ 1, 1 ⊗ Aθ22 , √ A−θ1 ⊗ √ A−θ


1 1
1 1 2
1 2
t t t
of A1 ⊗k[t] A2 and

2 , √ A1 ⊗ Aθ22 , Aθ11 ⊗ √ A−θ


1 −θ1 1
Aθ11 ⊗ Aθ22 , A−θ
1
1
⊗ A−θ 2
2
2

t t
of A12 . Since an involution fixes the unit in general, we have

Aθ11 ⊗ 1 ⊂ Aθ11 ⊗ Aθ22 , 1 ⊗ Aθ22 ⊂ Aθ11 ⊗ Aθ22

√ A−θ ⊗ 1 ⊂ √ A−θ ⊗ Aθ22 , 1 ⊗ Aθ22 ⊂ Aθ11 ⊗ √ A−θ


1 1
1 1
1 2
1 1 2
t t t

√ A−θ ⊗ √ A−θ ⊂ ( √ A−θ ⊗ Aθ22 ) ⋅ (Aθ11 ⊗ √ A−θ


2 ).
1 1
1 2
1 1
1 2
1 2 1
t t t t
These formulas of containment show that the composite map of (7) with iA1 ,A2
factors through A12 . This map is onto A12 from

Aθ11 ⊗ Aθ22 = (Aθ11 ⊗ 1) ⋅ (1 ⊗ Aθ22 )

23
= t ⋅ √ A−θ ⊗ √ A−θ
1 1
A−θ
1
1
⊗ A−θ
2
2
2
2

t t

√ A−θ ⊗ Aθ22 = ( √ A−θ ⊗ 1) ⋅ (1 ⊗ Aθ22 )


1 1
1 1
1 1
t t

Aθ11 ⊗ √ A−θ = (Aθ11 ⊗ 1) ⋅ (1 ⊗ √ A−θ


2 ).
1 2
1 2
2
t t
√ ±1
Part (2) follows by seeing values of the map in A12 [ t ].
Parts (3), (4), and (7) are evident by definitions. Part (6) will follow from
(3) and (5). Therefore the proof will be completed by showing (5). The map
iA1 ,A2 is identified with the composition of the sequence of canonical maps
√ ±1
A1 ⊗k[t] A2 → A1 ⊗k[t] A2 [ t ]
√ ±1 √ ±1
≅ (A1 ⊗k[t] k [ t ]) ⊗k[√t±1 ] A2 [ t ]
√ ±1 √ ±1
≅ A1 [ t ] ⊗k[√t±1 ] A2 [ t ] .

√ ±1
The first map is the base change of the inclusion A2 ↪ A2 [ t ]. The sec-
√ ±1
ond map is the canonical isomorphism using the k [ t ]-algebra structure of
√ ±1
A2 [ t ]. The isomorphism in the third row is obtained by Lemma 2.4. If
(A1 , θ1 ) satisfies Condition 2.22, then the first map is injective. This proves
that iA1 ,A2 is injective if (A1 , θ1 ) satisfies Condition 2.22. One can see that
iA1 ,A2 is injective if (A2 , θ2 ) satisfies Condition 2.22 in a similar way. This
proves (5).
Corollary 2.27. The assignment (A, θ) ↝ A determines a lax symmetric
monoidal functor from the symmetric monoidal category of commutative k-
algebras with involutions to that of k [t]-algebras. Moreover, it restricts to a
symmetric monoidal functor from the symmetric monoidal category of com-
mutative k-algebras with involutions satisfying Condition 2.22 to that of flat
k [t]-algebras.
√ ±1
Proof. We can reduce the first assertion to the fact that − ⊗k k [ t ] is sym-

metric monoidal by seeing the relations in the Laurent polynomial algebras of
tensor products of commutative k-algebras with variable t. For the unitality,
recall Example 2.2 if necessary. For the latter statement, suppose that we are
given pairs (A, θ) and (B, η) of commutative k-algebras with involutions satis-
fying Condition 2.22. Put (C, ζ) = (A ⊗k B, θ ⊗k η). Then we have a natural
isomorphism A ⊗k[t] B ≅ C. Since A and B are flat over k [t], so is C. The
latter assertion now follows from the former one.

24
Remark 2.28 ([5, Remark 3.2.1]). If one would like to work over the projec-
tive line P1k over k, define the corresponding contraction schemes over the two
principal open affine lines in P1k . Let Āt be the k [t±1 ]-algebra obtained from At
by switching the action of t with t−1 . Then the map Āt → At defined by
atn ↦ at−n (a ∈ Aθ )
↦ a t √t (a ∈ A−θ )
−n+1
tn
a√ t

is a k [t±1 ]-algebra isomorphism (cf. [3, the last line of Section 3]). Use this map
and Lemma 2.4 (1) to glue the schemes over the two principal open affine lines
of P1k to obtain a scheme over P1k .
Remark 2.29. If one wishes to work with n-contraction over the projective
n-scheme Pnk , work on principal open affine n-spaces in Pnk , and glue up the
contraction schemes on them in a similar way to Remark 2.28.

3 Hopf structure
Let k be a Z [1/2]-algebra. Let A be a commutative Hopf k-algebra with an
involution. Let
∆ ∶ A → A ⊗k A
ǫ∶A→k
i∶A→A
be the comultiplication, counit, and antipode of A respectively. We will use the
Sweedler notation of [36, Section 1.2].
In this section, we put the structure of a Hopf k [t]-algebra on A. We also
give a group theoretic analog of Corollary 2.9, which is rather a straightforward
generalization of [3, Definition 4.2]. We follow the notations in the former
√ ±1 √ ±1
section, but write G = Spec A.
The k [ t ]-algebra A [ t ] is equipped with the structure of a Hopf
algebra by the base change from A. One can put the structure of a Hopf algebra
over k [t±1 ] on At by the Galois descent (cf. Proposition 2.5, [25, Theorem
√ ±1
A.3]). In fact, one can define the structure homomorphisms by restriction from
A [ t ] to At . We wish to restrict the structure homomorphisms of Hopf
√ ±1
algebras on A [ t ] and At to A. This follows as a formal consequence of
Corollary 2.27 under Condition 2.22. Let us note below how we can construct
the structure homomorphisms.
Lemma 3.1. The counit ǫ is zero on A−θ .
Proof. It is straightforward: We have

ǫ(a) = ǫ(a − θ(a)) = (ǫ(a) − ǫ(a)) = 0


1 1
2 2
for a ∈ A−θ .

25
√ ±1
Construction 3.2 (Counit). The map i ⊗k k [ t ] restricts to a k [t]-algebra

√ ±1
homomorphism
ǫ ∶ A → k [t] ⊂ k [ t ]

by Lemma 3.1. Explicitly, ǫ is computed by

ǫ∣Aθ [t] = ǫ∣Aθ ⊗k k [t]

ǫ∣(⊕ tn
−θ √ )⊕⊕n≥2 √1n (A−θ )n
= 0.
n≥0 A
t t

Construction 3.3 (Antipode). Since i commutes with θ, i respects Aθ and


√ ±1
A−θ . Hence i ⊗k k [ t ] naturally restricts to a k [t]-algebra automorphism i
of A (recall the definition of A).
In the rest of this section, we assume Condition 2.22.
Construction 3.4 (Comultiplication). Put (B, η) = (A ⊗k A, θ ⊗k A). Then we
have a commutative diagram

∆ IA,A
A B ∼ A ⊗k[t] A
iA,A

√ ±1 √ ±1 √ ±1 √ ±1
√ ±1
∆⊗k k[ t ]
A[ t ] B[ t ] ∼
A [ t ] ⊗k[√t±1 ] A [ t ]

by Proposition 2.26. We denote the composite horizontal k [t]-algebra homo-


morphism A → A ⊗k[t] A by the same symbol ∆. In computations, we will
√ ±1 √ ±1
regard ∆(a) as an element of A [ t ] ⊗k[√t±1 ] A [ t ] rather than A ⊗k[t] A
through the counterclockwise sequence of arrows for a ∈ A.
We now record our result as a formal statement:
Theorem 3.5. The k [t]-algebra A is a Hopf algebra for the homomorphisms
(∆, ǫ, i).
Set K = Spec A/I. According to [24, the proof of Lemma 3.1], K ⊂ G can be
identified with the fixed point subgroup scheme by Spec θ.
Let g be the Lie algebra of G (see [16, Chapter II, §4] for the general for-
malism). We denote the differential of θ by the same symbol. Set

k = {x ∈ g ∶ θ(x) = x}

p = {x ∈ g ∶ θ(x) = −x}.
It follows by definition that k is naturally identified with the Lie algebra of K.
Corollary 3.6. If A is smooth over k then Spec A0 is isomorphic to K ⋉ p as
an affine group scheme over k.

26
For a digression, let us compute the differential structures:
Construction 3.7 ([5, 2.1.3]). Write [−, −]g for the Lie bracket of g. Define a
Lie algebra over k [t] as follows:

g = g ⊗k k [t] ,


⎪[x, y]g (x ∈ k, y ∈ g)
[x, y]g = ⎨

⎪ t [x, y]g (x, y ∈ p),

where [x, y]g is the Lie bracket of x and y in g. We also set gt ∶= g ⊗k[t] k [t±1 ].
Proposition 3.8. Suppose that A is finitely presented over k. Then the Lie
algebra of G is isomorphic to g.
Let Derk (A, k) denote the k-derivations of A on k, where k is regarded
as an A-module for ǫ. We use similar notations for other commutative Hopf
algebras. According to [16, Chapter I, §4, 2.2 Proposition and Chapter II, §4,
3.6 Corollary], one can identify Derk (A, k) with the Lie algebra of G. The Lie
bracket of Derk (A, k) is given by

[x, y]g (a) = ∑(x(a(1) )y(a(2) ) − y(a(1) )x(a(2) )).


(a)

Proof. Let ξ ∈ Derk[t] (A, k [t]). Then define a k [t]-module homomorphism


x ∶ A [t] → k [t] by

⎪ξ(a) (a ∈ Aθ )

x(a) = ⎨

⎪ ξ ( √1t a) (a ∈ A−θ ).

It is easy to show that x belongs to Derk[t] (A [t] , k [t]) (use Lemma 3.1).
Conversely, suppose that we are given a k [t]-derivation x of A [t]. Then
define a k [t]-module homomorphism ξ ∶ A → k [t] by

ξ(a) = x(a) (a ∈ Aθ )
ξ ( √t a) = x(a)
1
(a ∈ A−θ )
ξ∣⊕n≥2 √1n (A−θ )n = 0.
t

One can easily prove that ξ belongs to Derk[t] (A, k [t]) (use Lemma 3.1).
These constructions give a k [t]-module isomorphism

Derk[t] (A, k [t]) ≅ Derk[t] (A [t] , k [t]).

Since A is finitely presented over k, the canonical map

k [t] ⊗k Derk (A, k) → Derk[t] (A [t] , k [t])

is an isomorphism (use [23, Corollaire (16.4.22)]). Let

φ ∶ Derk[t] (A, k [t]) → g


27
denote the resulting isomorphism of k [t]-modules.
Finally, we check that the transferred Lie bracket on g from
Derk[t] (A, k [t])
coincides with [−, −]g . We denote the Lie bracket on Derk[t] (A, k [t]) by [−, −].
Let x, y ∈ g and a ∈ A. Then we have
φ−1 ([φ(x), φ(y)])(a)
= φ−1 ([φ(x), φ(y)])(a+ + a− )

= [φ(x), φ(y)] (a+ + √ a− )


1
t

= ∑ φ(x)(a(1)+ )φ(y)(a(2)+ ) + ∑ φ(x) ( √ a(1)− ) φ(y) ( √ a(2)− )


1 t
(a) (a) t t

− ∑ φ(y)(a(1)+ )φ(x)(a(2)+ ) − ∑ φ(y) ( √ a(1)− ) φ(x) ( √ a(2)− )


1 t
(a) (a) t t

+ ∑ φ(x)(a(1)+ )φ(y) ( √ a(2)− ) + ∑ φ(x) ( √ a(1)− ) φ(y)(a(2)+ )


1 1
(a) t (a) t

− ∑ φ(y)(a(1)+ )φ(x) ( √ a(2)− ) − ∑ φ(y) ( √ a(1)− ) φ(x)(a(2)+ )


1 1
(a) t (a) t
by
∆(a + θ(a)) = ∑(a(1) ⊗ a(2) + θ(a(1) ) ⊗ θ(a(2) ))
(a)

= 2 ∑(a(1)+ ⊗ a(2)+ + a(1)− ⊗ a(2)− ),


(a)

∆(a − θ(a)) = ∑(a(1) ⊗ a(2) − θ(a(1) ) ⊗ θ(a(2) ))


(a)

= 2 ∑(a(1)+ ⊗ a(2)− + a(1)− ⊗ a(2)+ ).


(a)

Hence by construction of the bijection


Derk[t] (A, k [t]) ≅ Derk[t] (A [t] , k [t]),
we get
φ−1 ([φ(x), φ(y)])(a) = ∑ x(a(1)+ )y(a(2)+ ) + ∑ x(a(1)− )y(a(2)− t)
(a) (a)

− ∑ y(a(1)+)x(a(2)+ ) − ∑ y(a(1)− )(a(2)− t)


(a) (a)

+ ∑ x(a(1)+ )y(a(2)− ) + ∑ x(a(1)− )y(a(2)+ )


(a) (a)

− ∑ y(a(1)+)x(a(2)− ) − ∑ y(a(1)− )x(a(2)+ )


(a) (a)

28
We proceed further computations by case-by-case study. Observe that if
x ∈ k and a ∈ A−θ then
1
x(a) = x(a − θ(a)) = 0;
2
if x ∈ p and a ∈ Aθ then
1
x(a) = x(a + θ(a)) = 0.
2
Therefore the last formula in the former paragraph is computed as follows:
x∈k
φ−1 ([φ(x), φ(y)])(a) = ∑ x(a(1)+ )y(a(2)+ ) − ∑ y(a(1)+ )x(a(2)+ )
(a) (a)

+ ∑ x(a(1)+ )y(a(2)− ) − ∑ y(a(1)− )x(a(2)+ )


(a) (a)

= ∑ (x(a(1)+ )y(a(2) ) − y(a(1) )x(a(2)+ ))


(a)

= ∑(x(a(1) )y(a(2) ) − y(a(1))x(a(2) ))


(a)

= [x, y]g (a)

(use x ∈ k for the third equality);


x, y ∈ p

φ−1 ([φ(x), φ(y)])(a) = ∑ x(a(1)− )y(a(2)−t) − ∑ y(a(1)− )x(a(2)− t)


(a) (a)

= t ∑(x(a(1) )y(a(2) ) − y(a(1) )x(a(2) ))


(a)

= t [x, y]g (a)

(use x, y ∈ p for the second equality).


This completes the proof.
Corollary 3.9. Let Y = Spec B be an affine k-scheme, equipped with an action
of G, and η be an involution of B. Assume:
(i) The action of G on Y respects the involutions of G and Y ;
(ii) A satisfies Condition 2.22.

Then Y is naturally equipped with an action of G. Moreover, its differential


action is given by


⎪x(b) (x ∈ k, b ∈ B η )
x(b) = ⎨√
⎪ tx(b) (x ∈ p, b ∈ B η )

29


⎪ √ x(b)
1
(x ∈ k, b ∈ B −η )
x ( √ b) = ⎨ t
1

⎪ (x ∈ p, b ∈ B −η )
t ⎩x(b)
under the identification of Proposition 3.8, where for x ∈ g ⊂ g, x is the image
of x in the Lie algebra of G.
We remark that x is uniquely determined by the formulas above through
the Leibniz rule.
Proof. The first assertion follows from Proposition 2.26 in a similar way to
Corollary 2.27. To compute the differential action, write ρ ∶ B → B ⊗k A for the
algebra homomorphism corresponding to the action map G ×Spec k Y → Y . We
will apply the Sweedler notation to ρ and ρ.
For ξ ∈ Derk[t] (A, k [t]), the differential action is given by ξ(−) ∶= (1 ⊗ ξ) ○ ρ.
Take x ∈ g. Then for b ∈ B η , we have

x(b) = ∑(b(1)+ x(a(2)+ ) + b(1)− x(a(2)− ) t)
(b)

by
ρ(b) = ρ(b + θ(b)) = ∑(b(1)+ ⊗ a(2)+ + b(1)− ⊗ a(2)− );
1
2 (b)

for b ∈ B −η , we have

x ( √ b) = ∑ (b(1)+ x(a(2)− ) + √ b(1)− x(a(2)+ ))


1 1
t (b) t

by
ρ(b) = ρ(b − θ(b)) = ∑(b(1)+ ⊗ a(2)− + b(1)− ⊗ a(2)+ ).
1
2 (b)

Hence we have
x(b) = ∑ b(1)+ x(a(2)+ )
(b)

∑(b(1) x(a(2) +) + θ(b(1) )x(a(2)+ ))


1
=
2 (b)

∑(b(1) x(a(2) ) + θ(b(1) )x(θ(a(2) )))


1
=
2 (b)

= (x(b) + x(θ(b)))
1
2
= x(b)

30
for x ∈ k and b ∈ B η ;

x(b) = ∑ b(1)− x(a(2)− ) t
(b)

∑(b(1) x(a(2)− ) − θ(b(1) )x(a(2)− )) t
1
=
2 (b)

= ∑(b(1) x(a(2) ) + θ(b(1) )x(θ(a(2) ))) t
1
2 (b)

(x(b) + x(θ(b)))
t
=
√2
= tx(b)

for x ∈ p and b ∈ B η ;

x ( √ b) = √ ∑ b(1)− x(a(2)+ )
1 1
t t (b)

= √ ∑(b(1) x(a(2)+ ) − θ(b(1) )x(a(2)+ ))


1
2 t (b)

= √ ∑(b(1) x(a(2) ) − θ(b(1) )x(θ(a(2) )))


1
2 t (b)

= √ (x(b) − x(θ(b)))
1
2 t
= √ x(b)
1
t
for x ∈ k and b ∈ B −η ;

x ( √ b) = ∑ b(1)+ x(a(2)− )
1
t (b)

∑(b(1) x(a(2)− ) + θ(b(1) )x(a(2)− ))


1
=
2 (b)

∑(b(1) x(a(2) ) − θ(b(1) )x(θ(a(2) )))


1
=
2 (b)

= (x(b) − x(θ(b)))
1
2
= x(b)

for x ∈ p and b ∈ B −η .

The following assertion is evident by construction:

31
Proposition 3.10. Let k → k ′ be a flat homomorphism of commutative Z [1/2]-
algebras. Write B = k ′ ⊗k A. The canonical map

k ′ ⊗k A ≅ B

of k ′ [t]-algebras in Proposition 2.6 is an isomorphism of Hopf algebras over


k ′ [t]. We remark that this statement makes a sense since B satisfies Condition
2.22 by Example 2.25.
For a positive integer n, let SLn be the special linear group scheme of degree
n over k.
Proposition 3.11. Suppose that we are given a representation ι ∶ G → SLn
(n ≥ 1) which is a closed immersion as a morphism of schemes. Regard SLn as
a closed subscheme of An . Then the morphism ι̃ ∶ G ↪ A4n ⊗k k [t] in Corollary
2 2

2.9 factors through SL2n ⊗k k [t]. Moreover, the resulting map G → SL2n ⊗k k [t]
is a homomorphism of group schemes.
√ ±1
Proof. The map ι̃ ⊗k[t] k [ t ] is expressed as

√ √ √ √ −1
g↦( )( )( )
tIn − tIn ι(g) 0 tIn − tIn
In In 0 ι(θ(g)) In In

under the identification


√ ±1 √ ±1
G ⊗k k [ t ] ≅ G ⊗k[t] k [ t ] ,

√ ±1
where In is the nth unit matrix (cf. [3, Definition 4.1]). Therefore ι̃⊗k[t] k [ t ]
√ ±1
factors through SL2n ⊗k k [ t ]. It descends to a homomorphism

G ⊗k[t] k [t±1 ] → SL2n ⊗k k [t±1 ] .

Since SL2n is closed in An , ι̃ factors through SL2n ⊗k k [t] (recall Corollary


2

2.9). One can verify that the resulting map G ↪ SL2n ⊗k k [t] is a morphism
of affine group schemes over k [t] by a similar argument to Corollary 2.27. In
fact, one can check that the corresponding map of the coordinate rings is a
homomorphism of Hopf algebras over k [t] by comparing the comultiplications
√ ±1 √ ±1
in
A [ t ] ⊗k[√t±1 ] A [ t ] .

This completes the proof.


Corollary 3.12. Suppose k = F ∈ {R, C}. Let G be an affine algebraic group
over F with an involution. Then the attached group scheme G of ours is iso-
morphic to the contraction family constructed in [3, Definition 4.2 and Theorem
5.1].

32
As we mentioned in the introduction, we only work over the polynomial ring
F [t] just for simplicity.
Proof. For existence of a faithful representation of G over F , see [29, Theorem
4.9]. We can find a faithful representation of G to a special linear group by
taking the determinant.
Henceforth fix a faithful representation ι ∶ G ↪ SLn for some positive integer
n. Then the assertion for F = C follows from Proposition 3.11 and [28, Chapter
2, Exercise 3.17 (e)]. Put F = R. We may assume the matrix S in [3, Lemma 5.1]
to be the unit since our faithful representation is defined over the real numbers.
The real form of [3, Theorem 5.1] now coincides with our G by Propositions
3.11 and A.5 (use Proposition 3.11 twice; one each of R, C).

4 Contraction families of quotient varieties


In this section, we study quotient of contraction group schemes. For the tech-
nical reason on existence of quotient schemes, we assume the ground ring k = F
to be a field of characteristic not 2.
Let G be a smooth affine algebraic group over F with an involution θ, and
H be a θ-invariant smooth subgroup of G. Then we have contraction group
schemes G and H. It follows from the naturality of our Hopf structure and
Corollary 2.13 that H is a closed subgroup scheme of G. The fppf quotient
G/H exists since F [t] is a PID ([2, 4.C. Théorème], Example 2.24).
The following assertion is immediate from Theorem 2.17:
Proposition 4.1. Suppose that G and H are smooth over F . Then G/H is a
smooth scheme of finite presentation (equivalently, smooth, quasi-compact, and
quasi-separated) over F [t].
Proof. We have a Cartesian diagram
pr
G ×Spec F [t] H G
a π

π
G G/H

by definition of G/H, where π is the quotient map, a is the action map, and
pr is the projection. In virtue of [17, Théorème 10.1.1] or [2, Théorème 6], we
see that π is faithfully flat and locally of finite presentation. Since the structure
homomorphism H → Spec F [t] is smooth (Theorem 2.17), so is pr. Similarly,
since the morphism H → Spec F [t] is affine, so is pr. In particular, pr is quasi-
compact and (quasi-)separated. Therefore the fppf descent implies that π is
smooth of finite presentation. Recall also that G is smooth over F [t] (Theorem
2.17). It also follows by definition that the structure morphism G → Spec F [t]
is affine, in particular, quasi-compact and (quasi-)separated. The assertion now
follows from [23, Lemma (17.7.5) and Proposition (17.7.7)].

33
In this section, we examine relations of this quotient with other related G-
schemes. Before we begin this, let us renew the notations on Lie algebras from
the former section: For a smooth affine group scheme over a commutative ring
k, we denote its Lie algebra by the corresponding small German letter. Its
adjoint representation will be denoted by Ad. The Lie bracket will be denoted
by [−, −]. We remark that there is no conflict of the notations for the bold
German letters and the subscript (−)t in virtue of Proposition 3.8 and Lemma
2.4. For an involution θ on the given group scheme, we will denote the induced
involution on its Lie algebra by the same symbol θ.
Let us also introduce an additional notation for Section 4.2: For a smooth
affine group scheme L over a commutative ring k and a smooth affine subgroup
scheme M , we denote the normalizer of m in l with respect to the adjoint
representation by NL (m) (see [9, Définition 2.3.3] for a general formalism). We
remark that m is a sub-copresheaf of l by the definition of the Lie algebras
of group schemes as copresheaves on the category of commutative k-algebras.
According to [10, Corollaire 4.11.8] and [13, Rappel 5.3.0], NL (m) is represented
by a closed subgroup scheme of L which is of finite presentation over L. In
particular, NL (m) is of finite presentation over k.

4.1 Contraction families of symmetric varieties


Let K ⊂ G be the fixed point subgroup by θ. Assume that G is (possibly discon-
nected) reductive. Then so is K ([32, the beginning of Section 1]). Therefore
X ∶= G/K is an affine variety by the Matsushima criterion (see [30, 31]). We
denote the attached involution on X to θ by the same symbol. In this section,
we study X and its fiber X 0 at t = 0. Let X θ be the fixed point subvariety of
X by θ. Let p ⊂ g be as in Section 3.

Proposition 4.2. The F -variety X θ is finite étale. In particular, X 0 is iso-


morphic to the disjoint union of copies of p if F is separably closed.
Proof. We may assume that F is algebraically closed. Recall that X θ is a
smooth F -variety by [24, Lemma 3.1]. In the rest, we may identify the closed
points of X and X θ with their F -points.
For each point gK ∈ X(F ), the tangent space of X at gK can identified with
g/ Ad(g)k. If θ(g)K = gK then θ induces an automorphism

g/ Ad(g)k → g/ Ad(θ(g))k = g/ Ad(g) Ad(g −1 θ(g))k = g/ Ad(g)k.

Moreover, the tangent space of X θ at gK coincides with its fixed point subspace.
If an element x ∈ g satisfies θ(x) ∈ Ad(g)k then x belongs to Ad(g)k by

Ad(g)−1 x = θ(Ad(θ(g)−1 )θ(x))


= θ(Ad(θ(g)−1 g) Ad(g)−1 θ(x))
∈ θ(Ad(K(F ))k)
= k.

34
This implies that the tangent space of X θ at gK is trivial. Therefore X θ is a
smooth affine algebraic variety of dimension zero. Equivalently, X θ is a finite
étale F -variety. This completes the proof.
For a digression, let us note how we can compute X θ in a special case:
Corollary 4.3. Suppose F to be separably closed. Assume that G is connected
and simply connected. Let H 1 (θ, G(F )) and H 1 (θ, K(F )) denote the first group
cohomology of G(F ) and K(F ) for θ respectively ([33, Chapter I, section 5.1]).
Then X θ (F ) is bijective to the kernel of the canonical map

H 1 (θ, K(F )) → H 1 (θ, G(F )). (8)

Proof. The set of K(F )-orbits in X θ (F ) is bijective to the kernel of (8) by [7,
Chapter II, Caution of Section 6.8] and [33, Chapter I, section 5.4, Corollary 1].
One deduces from [35, Theorem 8.1] and Proposition 4.2 that K acts trivially
on X θ . Therefore the set of K(F )-orbits in X θ (F ) is exactly (identified with)
X θ (F ). This completes the proof.
Remark 4.4. Since θ acts trivially on K, H 1 (θ, K(F )) is identified with the
set of K(F )-conjugacy classes of the elements of K(F ) of order at most two.
Remark 4.5. The cohomology H 1 (θ, G(F )) for F = C and complexified Car-
tan involutions θ was studied in [1]. We can apply the computations of [1] to
H 1 (θ, K(C)) in this case since θ is still the complexified Cartan involution for
K.
Let x0 = K be the base point of X. This lifts to an F [t]-point x0 of X since
θ(x0 ) = x0 . We denote the centralizer subgroup of G at x0 by ZG (x0 ) (see [9,
Définition 2.3.3])
Theorem 4.6. (1) We have ZG (x0 ) = K.
(2) The G-orbit attached to x0 is an open subscheme of X. Moreover, it is
an isomorphism onto X ○ defined by x0 if G is connected.
Proof. We may assume that F is algebraically closed. Since X is smooth and
affine by Theorem 2.17 (cf. the proof of Proposition 4.1), X is separated and
locally of finite presentation over F [t]. Hence the centralizer ZG (x0 ) is repre-
sented by a closed subgroup of finite presentation of G ([6, Exemples 6.2.4. b)]).
In particular, ZG (x0 ) is of finite presentation over F [t] since G is affine and
smooth over F [t] (Theorem 2.17). The action of G at x0 gives rise to a map
G → X, whose restriction to K coincides with the constant map at x0 . In virtue
of the functoriality of contraction, K fixes x0 . Therefore K is contained in
ZG (x0 ).
We wish to prove that K = ZG (x0 ). Notice that K is flat and locally of
finite presentation over F [t] by K = K ⊗F F [t]. In view of [23, Corollaire
(17.9.5)], it will suffice to see the equality at t = 0 and the locus of t ≠ 0. To
see the action on these loci, we remark that the fppf quotient commutes with

35
any base change. Therefore the action of G on X at t = 0 can be identified
with that of the Cartan motion group K ⋉ p on the normal bundle TX θ X. The
√ ±1
equality now follows from the description of TX θ X in Proposition 4.2. To see
the equality on t ≠ 0, we may take the base change to F [ t ] to identify the
action with the base change of the original action of G on G/K. This proves
(1).
We next prove (2). The orbit map G/K = G/ZG (x0 ) → X is an isomor-
phism onto X ○ at t = 0 by Proposition 4.2. This map is also an isomorphism
on the locus of t ≠ 0 from Lemma 2.4 since the fppf quotient commutes with
arbitrary base changes. Moreover, X = X ○ on this locus if G is connected since
X is geometrically connected in this case (recall Lemma 2.4). Therefore the
orbit map is an open immersion by [23, Corollaire (17.9.5)]. If G is connected
then the orbit map factors through X ○ since X ○ is open in X. Moreover, the
resulting map G/K → X ○ is an isomorphism from [23, Corollaire (17.9.5)]. This
completes the proof.
Corollary 4.7. The quotient scheme G/K is affine if G is connected,
Proof. Combine Theorem 4.6 (2) with Proposition 2.20. We remark that the
condition (iii) of Proposition 2.20 holds since we are assuming k = F is a field.

If one is interested in symmetric spaces of real Lie groups, one can simply
resolve the difference of X and G/K by an analytic analog of [22, Corollaire
(15.6.5)]:
Proposition 4.8. Let p ∶ M → N be a submersion of smooth manifolds, and
s be a section of p. For each y ∈ N , let p−1 (y)○ be the connected component of
p−1 (y) containing s(y). Set

M ○ = ∐ p−1 (y)○ .
y∈N

Then M ○ is an open submanifold of M .


Proof. Take x ∈ M ○ . We wish to find an open neighborhood of x ∈ M contained
in M ○ . Write y = p(x). Since p−1 (y) is a manifold ([18, Chapter 1, Theorem
5.6]), one can choose a continuous map c ∶ [0, 1] → p−1 (y) satisfying c(0) = x
and c(1) = s(y), where [0, 1] ⊂ R is the closed interval from 0 to 1. For each
t ∈ [0, 1], one can take open neighborhoods Ut ∋ c(t) and Vt ∋ p(c(t)) = p(x)
with the following properties by [18, Chapter 1, the proof of Theorem 5.6]:
(i) Vt = p(Ut );
(ii) For each element y ′ ∈ Vt , p−1 (y ′ ) ∩ Ut is connected.
In fact, shrink W in [18, Chapter 1, the proof of Theorem 5.6] to assume that
W is the product of an open subset of Rn containing 0 and an open ball in
Rm−n around 0. We may also replace V1 and U1 by a smaller open subset V1′ ∋ y

36
and p−1 (V1′ ) ∩ U1 respectively with the property that s(V1′ ) ⊂ U1 to assume that
s(V1 ) ⊂ U1 .
Since [0, 1] = ∪t∈[0,1] c−1 (Ut ), one can find a finite subset {0, 1} ⊂ I ⊂ [0, 1]
such that [0, 1] = ∪t∈I c−1 (Ut ). We construct a sequence t0 , t1 , . . . , tn of distinct
elements of I with the following properties:
(i) t0 = 0 and tn = 1;

j=0 (c (Utj ) ∩ c (Uti )) ≠ ∅.


−1 −1
(ii) For 1 ≤ i ≤ n, ∪i−1

(iii) For 0 ≤ i ≤ n − 2, c−1 (Uti ) ∩ c−1 (U1 ) = ∅.


Set t0 = 0. For i ≥ 1, define ti ∈ I as follows: If c−1 (Uti−1 ) ∩ c−1 (U1 ) ≠ ∅ then set
n = i and ti = 1; Suppose otherwise. Write Ii−1 = I ∖ {t0 , t1 , . . . , ti−1 }. Then we
have
[0, 1] = (∪i−1
j=0 c (Utj )) ∪ ∪t∈Ii−1 c (Ut ).
−1 −1

We have elements 0 ∈ c−1 (U0 ) ⊂ ∪i−1


j=0 c (Utj ) and 1 ∈ ∪t∈Ii−1 c (Ut ) since 1 ∈ Ii−1 .
−1 −1

Since [0, 1] is connected, (∪j=0 c (Utj ))∩(∪t∈Ii−1 c (Ut )) is nonempty. One can
i−1 −1 −1

and do choose ti ∈ Ii−1 such that (∪i−1 j=0 c (Utj )) ∩ c (Uti ) ≠ ∅. This procedure
−1 −1

will stop since I is a finite set.


For each 1 ≤ i ≤ n, choose 0 ≤ m(i) ≤ i − 1 such that
c−1 (Utm(i) ) ∩ c−1 (Uti ) ≠ ∅.

Set V = ∩ni=1 p(Utm(i) ∩ Uti ). Then we have V ⊂ ∩ni=0 Vti since m(1) = 0. Observe
that V contains p(x). In fact, for any 1 ≤ i ≤ n, take an element
u ∈ c−1 (Utm(i) ) ∩ c−1 (Uti ).

Then c(u) is contained in Utm(i) ∩ Uti . We thus get


p(x) = p(c(u)) ∈ p(Utm(i) ∩ Uti ).

We also note that V is open since p is an open map ([18, Chapter 1, Theorem
5.6]). We define a descending sequence n(j) by
n(−1) = n, n(j + 1) = m(n(j)).

This sequence eventually stops, i.e., one can find a nonnegative integer j0 such
that n(j0 ) = 0.
We prove that the open neighborhood U0 ∩ p−1 (V ) of x in M is contained in
M . Take any element x′ ∈ U0 ∩p−1 (V ). Write y ′ = p(x′ ). Then for each 1 ≤ i ≤ n,

p−1 (y ′ ) ∩ Utm(i) ∩ Uti ≠ ∅ by definition of V . For each −1 ≤ j ≤ j0 − 1, fix an


element x′n(j) ∈ p−1 (y ′ ) ∩ Utn(j+1) ∩ Utn(j) . For convention, we set x′n(−2) ∶= s(y ′ ).
Then x′n(−2) = s(y ′ ) belongs to U1 = Utn = Utn(−1) . According to our choice of
Ut , one can connect x′n(j−1) and x′n(j) in p−1 (y ′ ) ∩ Utn(j) for each −1 ≤ j ≤ j0 − 1
(recall y ′ ∈ V ⊂ Vtn(j) ). Therefore we reach

x′n(j0 −1) ∈ p−1 (y ′ ) ∩ Utn(j0 ) = p−1 (y ′ ) ∩ Ut0 = p−1 (y ′ ) ∩ U0 .

37
from s(y ′ ) by a path. We can connect x′n(j0 −1) with x′ in p−1 (y ′ ) ∩ U0 since
p−1 (y ′ ) ∩ U0 is connected. Therefore x′ belongs to M ○ . This completes the
proof.
Corollary 4.9. Put F = R. Regard X and G/K as R-schemes by composing
the structure morphisms to Spec R [t] with the canonical morphism

Spec R [t] → Spec R.

Then the open immersion G/K → X induces a diffeomorphism

(G/K)(R)○ ≅ X(R)○

of the manifolds of the fiberwise unit components. Moreover, the fiber at t = t0 ∈


R is Gt0 (R)○ /K(R) ∩ Gt0 (R)○ (resp. p) if t0 ≠ 0 (resp. t0 = 0), where Gt0 (R)○
is the unit component of the Lie group Gt0 (R).

4.2 Contraction families of partial flag schemes


In this section, we study the case where H = Q is a θ-stable parabolic subgroup.
We start with a quite more general setting. I.e., let k be an arbitrary com-
mutative ring with 1/2, and G be a reductive group scheme over k in the sense
of [12, Définition 2.7], i.e., a smooth affine group scheme over k whose geometric
fibers are connected reductive algebraic groups. We say that G is of type (RA)
if for every root α relative to a maximal torus T of each geometric fiber of G,
the positive generator of the ideal {χ(α) ∈ Z ∶ χ ∈ HomZ (X ∗ (T ), Z)} ⊂ Z is a
unit of k, where X ∗ (T ) is the character group of T ([13, Définition 5.1.6]).
Let us give a quick review on parabolic subgroups and partial flag schemes
from SGA 3. A smooth subgroup scheme Q of G is called a parabolic subgroup
if it is so at every geometric fiber in the classical sense ([15, Définition 1.1]). Let
PG denote the total flag scheme, i.e., the moduli scheme of parabolic subgroups
of G ([15, Section 3.2, Théorème 3.3]). Let type G be the finite étale k-scheme of
parabolic types of G (see [14, Section 3] and [15, Section 3.2, Définition 3.4]). We
have a canonical morphism PG → type G ([15, Section 3.2]). One can recognize
from the arguments of [15] that this map exhibits the étale quotient map of PG
by the conjugate action of G. For a parabolic subgroup Q ⊂ G, we call the image
of Q in (type G)(k) the (parabolic) type of Q ([15, Section 3.2, Définition 3.4]).
For each k-point x of type G, the fiber of the morphism PG → type G at x is
called the partial flag scheme of type x, and it will be denoted by PG,x (cf. [15,
Corollaire 3.6]).
Example 4.10 ([15, Corollaire 3.6]). There is a canonical parabolic type de-
noted by ∅. This corresponds to the flag scheme, i.e., the moduli scheme of
Borel subgroups ([13, Corollaire 5.8.3 (i)]).
One can define a morphism

PG → Gr(g) (9)

38
by assigning the Lie algebras of parabolic subgroups (see [10, Corollaire 4.11.8]).
According to [13, Corollaire 5.3.3], (9) is a monomorphism if G is of type (RA).
Since PG and Gr(g) are projective ([15, Corollaire 3.5]), (9) is a closed immersion
in this case ([20, Théorème 5.5.3 (i) and Corollaire 5.4.3 (i)], [23, Corollaire
18.12.6]). For a parabolic type x, we will denote the restriction of (9) to the
closed subscheme PG,x by ιx . For a latter argument, let us record a general
observation on the normalizer of the Lie algebra of a parabolic subgroup here:
Lemma 4.11. Suppose that G is of type (RA). Then for a parabolic subgroup
Q of G, we have NG (q) = Q.
Proof. This follows by the monomorphicity of (9) and [15, Proposition 1.2].
Take an involution of θ. Then Gt is a reductive group scheme over k [t±1 ]
by Lemma 2.4 and [12, the sentence below Définition 2.7]. Let us note a general
remark for finding elements of (type Gt )(k [t±1 ]):
√ ±1
Lemma 4.12. Let ¯be the Galois involution on type G ⊗k k [ t ] with respect
√ ±1
to the quadratic Galois extension k [ t ] /k [t±1 ]. We denote the involution on
type G induced from θ by the same symbol. Then we have a canonical bijection
√ ±1
(type Gt )(k [t±1 ]) ≅ {x ∈ (type G) (k [ t ]) ∶ θ(x̄) = x} .

√ ±1
Proof. The Galois involution on (type G) (k [ t ]) corresponding to the k [t±1 ]-
√ ±1
form Gt of G ⊗k k [ t ] is given by x ↦ θ(x̄). The assertion then follows from
generalities on Galois descent.
√ ±1
Example 4.13. Let x ∈ (type G)(k) such that θ(x) = x. Then x ⊗k k [ t ]
descends to an element of (type Gt )(k [t±1 ]).
Example 4.14. Let Q be a θ-stable parabolic subgroup of G. Then the
parabolic type of Q attaches an element of (type Gt )(k [t±1 ]) by Example 4.13.
This coincides with the type of Qt .
Example 4.15. Assume k = F is a field. Then the minimal parabolic subgroups
of G determine a unique element of (type Gt )(k [t±1 ]) by [7, 20.9 Theorem].
Henceforth we assume that G is of type (RA). Then Gt is also of type (RA)
by Lemma 2.4 and [13, Remarques 5.1.7 c)].
Take x ∈ (type Gt )(k [t±1 ]). Let j0 denote the canonical open immersion
Spec k [t±1 ] ↪ Spec k [t] of affine k-schemes. Compose the closed immersion ιx
with the base change jGr(g) ∶ Gr(gt ) ↪ Gr(g) of j0 to get an affine immersion
PGt ,x ↪ Gr(g).
Definition 4.16. We call the scheme-theoretic closure of PGt ,x in Gr(g) the
partial flag scheme of G of type x, and denote it by PG,x . If x = ∅ then we call
it the flag scheme of G, and refer to it as BG .

39
Remark 4.17. A key idea to prove that the total flag scheme of a reductive
group scheme without the (RA) hypothesis was to pass to the adjoint group
in the sense of [13, Définition 4.3.6] (see [13, the proof of Théorème 5.8.1]).
Therefore passing to the adjoint group of G is another possible definition of
partial flag schemes without the (RA) hypothesis. To be more precise, let G′
be the quotient of G by its center Z (see [13, Corollaire 4.1.7 and Proposition
4.3.5 (ii)]). Then Z is θ-stable in G, and therefore θ descends to an involution of
G′ . We note that Zt is of finite presentation over k [t±1 ] by Lemma 2.4 (2), [13,
Corollaire 4.1.7], and [21, Proposition 2.1 b)]. Similarly, the center of Gt is flat
of finite presentation over k [t±1 ] by[13, Corollaire 4.1.7] and [21, Proposition 2.1
a)]. With these in mind, one can prove in a similar way to Theorem 4.6 that G′t
is canonically isomorphic to the adjoint group of Gt . We identify the total flag
scheme and the scheme of parabolic types of Gt with those of G′t respectively.
Then imbed PGt into Gr(g′t ). Finally, take the schematic closure of PGt ,x in
Gr(g′ ). One can see that similar results to the below hold under this version
of the definition (the statements need minor modifications). We also note that
there is no canonical map Gr(g′ ) → Gr(g) even if G is of type (RA) since g′ is
not a quotient of g in general. For example, think of G = SL2n+1 with n ≥ 1,
the trivial involution, and k = Z [1/2]. Therefore it is difficult to compare these
two definitions in general. On the other hand, the definitions coincide if G is of
type (RA) and Z is smooth by [19, Proposition 8.17 (2)] and Proposition A.6
since g′ ≅ g/z in this case.
Remark 4.18. One can see that PGt ,x is scheme-theoretically dense in PG,x
by [34, Lemma 01RG].
Example 4.19. Let Q be a θ-stable parabolic subgroup Q ⊂ G, and x be the
attached parabolic type of Gt . In this case, we write PG,x = Gt /Qt . We note
that PGt ,x is naturally identified with Gt /Qt by Example 4.14 and [15, Corollaire
3.6].
Example 4.20. Let G be SL2 or SO(2, 1) ≅ PGL2 . Then the flag scheme
of G is isomorphic to B ∶= Proj k [t, x, y, z] /(x2 + y 2 − tz 2 ) through a natural
identification of Proj k [t, x, y, z] with the Grassmannian scheme of rank 2. In
fact, one can deduce from computations in Example 2.15 that jGr(g) ○ ι∅ is an
isomorphism onto

Proj k [t±1 , x, y, z] /(x2 + y 2 − tz 2 ) ≅ Proj k [t, x, y, z] /(x2 + y 2 − tz 2 ) ⊗k[t] k [t±1 ]



(see [19, Remark 13.27])√ if k contains −1; This statement for general k is
verified by adding −1 and applying the Galois descent. Therefore the flag
scheme of G coincides with B by Example A.3 and Propositions A.5, A.6.
As a consequence, one finds that the √ flag scheme of the current G is not
smooth at t = 0. In fact, we may add −1 to the base ring k. Then the fiber
at t = 0 is not smooth at the intersection of the two projective lines (recall the
description of the fiber of B at t = 0 in Example 2.15).
Let us record two basic observations on PG,x .

40
Proposition 4.21. We have a canonical isomorphism PG,x ⊗k[t] k [t±1 ] ≅ PGt ,x .
Proof. Since ιx is a closed immersion, the assertion follows from Proposition
A.5 and Example A.4.
Proposition 4.22. The partial flag scheme PG,x is a G-invariant closed sub-
scheme of Gr(g).
Proof. Let jGt denote the canonical immersion Gt ↪ G. Then we have an affine
immersion
jGt ×Spec k[t] (jGr(g) ○ ιx ). (10)
In particular, (10) is quasi-compact. Indeed, the morphisms jGt and jGr(g)
are affine since they are obtained by base changes of j0 . Since ιx is a closed
immersion, ιx is affine.
Consider the commutative diagram

(10)
Gt ×Spec k[t±1 ] PGt ,x Gt ×Spec k[t] PGt ,x G ×Spec k[t] Gr(g)

jGr(g) ○ιx
PGt ,x Gr(g),

where the vertical arrows are the action maps. In view of the functoriality of
scheme-theoretic images in [34, Lemma 01R9] and the definition of PG,x , the
proof will be completed by showing that the scheme-theoretic image of (10) is
G ×Spec k[t] PG,x .
The affine immersion (10) has a factorization
jGt ×Spec k[t] PGt ,x
Gt ×Spec k[t] PGt ,x ÐÐÐÐÐÐÐÐÐÐ→ G ×Spec k[t] PGt ,x
G×Spec k[t] (jGr(g) ○ιx )
ÐÐÐÐÐÐÐÐÐÐÐÐ→ G ×Spec k[t] Gr(g).

The first map is obtained by the base change of j0 . In virtue of Example A.3
and Propositions A.5, A.6, we may compute the scheme-theoretic image of (10)
as that of G ×Spec k[t] (jGr(g) ○ ιx ). The assertion now follows by Proposition A.5
and the definition of PG,x .
We next turn into another candidate for a contraction analog of partial flag
schemes. Let Q be a θ-stable parabolic subgroup of G.
Proposition 4.23. The partial flag scheme Gt /Qt contains q as a k [t]-point.
Proof. Recall that we have a canonical factorization

Gt /Qt ↪ Gt /Qt ↪ Gr(g)


j i

41
by [28, Chapter 2, Exercise 3.17 (d)]. Consider the commutative diagram
qt

Spec k [t±1 ] Gt /Qt Gt /Qt


j i
Gr(g)

j0 q

Spec k [t] ,

where the left horizontal arrow is given by the base point. Then the section of
Gr(g) attached to q is a closed immersion since Gr(g) is separated over k [t].
Therefore its restriction to Spec k [t±1 ] is quasi-compact. Moreover, the k [t]-
point q of Gr(g) exhibits Spec k [t] as the scheme-theoretic closure of Spec k [t±1 ]
by Example A.3 and Proposition A.6. The dotted arrow now exists by [28,
Chapter 2, Exercise 3.17 (d)]. This shows the assertion.
As we briefly explained in the introduction, we wish to compare the G○ -
orbit attached to q with Gt /Qt . For this, let us introduce some notations: For
a k-module V and a prime ideal P ⊂ k, we write

Vκ(P) ∶= V ⊗k κ(P),

where κ(P) in the right hand side is the residue field of k at P. Suppose that
we are given k-modules l and mi (i ∈ {1, 2, 3}) with m3 ⊂ l and together with a
k-linear map [−, −] ∶ m1 ⊗k m2 → l. We note that in latter applications, l will be
a Lie algebra over k, mi will be k-submodules for i ∈ {1, 2, 3}, and [−, −] will be
the restriction of the Lie bracket of l. Let us set

Nm1 (m2 ; m3 ) ∶= {x ∈ m1 ∶ [x, m2 ] ⊂ m3 }. (11)

We remark that Nm1 (m2 ; m3 ) is identified with the fiber product

m1 ×Homk (m2 ,l) Homk (m2 , m3 ) (12)

of the map m1 → Homk (m2 , l) corresponding to [−, −] and the inclusion map

Homk (m2 , m3 ) ↪ Homk (m2 , l).

Set p = {x ∈ g ∶ θ(x) = −x} as in Section 3.


To state our comparison theorem, we also need additional observations:
Lemma 4.24. The k-submodules k, p, q, p ∩ q, and k ∩ q are direct summands
of g.
Although we should have already known that q is a direct summand of g
on the course of defining the map (9), we show this here for convenience to the
reader. In particular, we do not need the (RA) hypothesis here.

42
Proof. Notice that g/q is finitely generated and projective as a k-module. In fact,
this statement is local in the Zariski topology of Spec k. This therefore follows
from [10, Corollaire 4.11.8] (recall that g is finitely generated and projective).
In particular, q is a direct summand of g.
The assertions for k and p follow from the decomposition g = k ⊕ p. Since q
is θ-stable, it restricts to the decomposition

q = (q ∩ k) ⊕ (q ∩ p). (13)

Therefore the assertions for p ∩ q and k ∩ q are verified by combining that for q
with (13).
Lemma 4.25. (1) The group scheme K ○ is reductive.
(2) If G is simply connected, i.e., the geometric fibers of G are so in the
classical sense, then K ○ = K.
(3) The closed subgroup scheme Q∩K ○ ⊂ K ○ is parabolic. In particular, Q∩K ○
has connected fibers.
(4) We have (Q ∩ K)○ = Q ∩ K ○ .
(5) The group schemes G○ and Q○ are affine.
(6) If G is simply connected then G○ = G and Q○ = Q.
Remark that we do not need the (RA) hypothesis of G here.
Proof. For (1), see [32, the beginning of Section 1] and [8, Proposition 3.1.3].
Part (2) is a consequence of [35, Theorem 8.1]. Part (3) follows from (1), [8,
Proposition 3.1.3], and [26, Propositions 5.2.5, 5.3.1]. For (4), we may work
fiberwisely to assume k = F is a field by definition of the unit component. Then
the assertion follows since Q ∩ K ○ is a connected open subgroup of Q ∩ K. Part
(5) follows from [8, Proposition 3.1.3], (4), and Proposition 2.20. For (6), we
may see on the locus of t ≠ 0 and at t = 0. The equality on t ≠ 0 follows from
Lemma 2.4. The equality at t = 0 for G follows from (2). The equality at t = 0
for Q is verified by (2) and (4).
Theorem 4.26. Assume the following conditions:
(i) G and K ○ are of type (RA).
(ii) the equality

Npκ(P) (kκ(P) ∩ qκ(P) ; qκ(P) ) = pκ(P) ∩ qκ(P) (14)

holds for every prime ideal P of k.


Then:
(1) We have NG○ (q) = Q○ . In particular, the G○ -orbit in Gt /Qt attached to
q ∈ Gt /Qt (k [t]) (recall Proposition 4.22) is isomorphic to G○ /Q○ .

43
(2) The G○ -orbit of (1) is representable.
The statements makes a sense since kκ(P) , pκ(P) , and qκ(P) are naturally
regarded as κ(P)-subspaces of gκ(P) for every prime ideal P ⊂ k by Lemma
4.24.
Remark 4.27. The condition of Theorem 4.26 is stable under formation of
K ○ (k)-conjugations of Q. Moreover, the orbit map G○ /Q○ ↪ Gr(g) only de-
pends on the K ○ (k)-conjugacy classes. On the other hand, different K ○ (k)-
conjugacy classes give different orbits in general. For example, if k = C, then
the two affine lines at t = 0 in Example 4.20 arise from the two closed K ○ (C)-
orbits in P1 (C).
Corollary 4.28. Assume the following conditions:
(i) G is simply connected.
(ii) G and K are of type (RA) (recall that K is reductive under the assumption
(i) by Lemma 4.25 (1) and (2)).
(iii) The equality

Npκ(P) (kκ(P) ∩ qκ(P) ; qκ(P) ) = pκ(P) ∩ qκ(P)

holds for every prime ideal P of k.


Then:
(1) The G-orbit in Gt /Qt attached to q ∈ Gt /Qt (k [t]) is isomorphic to G/Q.
(2) The G-orbit of (1) is representable.
Proof. This is immediate from Theorem 4.26 and Lemma 4.25 (6).
Towards the proof of Theorem 4.26, we need a copresheaf analog of (11):
Let l and mi (i ∈ {1, 2, 3}) be as before. Assume:
(i) m2 is finitely presented as a k-module, and
(ii) m3 is a direct summand of l.
Then we define a copresheaf Nm1 (m2 , m3 ) on the category of commutative k-
algebras by

Nm1 (m2 , m3 )(R)


= {x ∈ m1 ⊗k R ∶ x ∈ Nm1 ⊗k S (m2 ⊗k S, m3 ⊗k S) for every R−algebra S}
= Nm1 ⊗k R (m2 ⊗k R, m3 ⊗k R),

where x is regarded as an element of m1 ⊗k S in the second line through the


identification m1 ⊗k S ≅ (m1 ⊗k R) ⊗R S and the unit m1 ⊗k R → (m1 ⊗k R) ⊗R S.
The last equality follows by S-linear formations.

44
Lemma 4.29. (1) For any commutative flat k-algebra R, we have a canonical
isomorphism

Nm1 (m2 ; m3 ) ⊗k R ≅ Nm1 ⊗k R (m2 ⊗k R, m3 ⊗k R).

(2) Suppose that k = F is a field (of characteristic not 2). Then we have

Nm1 (m2 ; m3 ) = Nm1 (m2 ; m3 ).

We do not need the hypothesis 1/2 ∈ k for the formalism of Nm1 (m2 ; m3 ).
The field F in (2) can be of characteristic 2 if one wants.
Proof. Part (1) follows from (12) through a formal argument: We have

Nm1 (m2 ; m3 ) ⊗k R
≅ (m1 ×Homk (m2 ,m3 ) Homk (m2 , l)) ⊗k R
≅ (m1 ⊗k R) ×Homk (m2 ,m3 )⊗k R (Homk (m2 , l) ⊗k R)
≅ (m1 ⊗k R) ×HomR (m2 ⊗k R,m3 ⊗k R) HomR (m2 ⊗k R, l ⊗k R)
≅ Nm1 ⊗k R (m2 ⊗k R, m3 ⊗k R)

for any commutative flat k-algebra R. In fact, the second isomorphism follows
since flat base changes respect fiber products. The third isomorphism is verified
by the fact that m2 is finitely presented as a k-module. We used the assumption
(ii) on m3 in the last isomorphism to apply (12) to Nm1 ⊗k R (m2 ⊗k R, m3 ⊗k R).
Part (2) is immediate from (1).
Lemma 4.30. (1) The definitions of

k, p, q ∩ k = {x ∈ q ∶ θ(x) = x}, q ∩ p = {x ∈ q ∶ θ(x) = −x}

commute with arbitrary base changes, i.e., for any commutative k-algebra
R, we have a canonical isomorphism

k ⊗k R ≅ {x ∈ g ⊗k R ∶ (θ ⊗k R)(x) = x}

p ⊗k R ≅ {x ∈ g ⊗k R ∶ (θ ⊗k R)(x) = −x}
(q ∩ k) ⊗k R ≅ {x ∈ q ⊗k R ∶ (θ ⊗k R)(x) = x}.
(q ∩ p) ⊗k R ≅ {x ∈ q ⊗k R ∶ (θ ⊗k R)(x) = −x}.

(2) The intersections k ∩ q and p ∩ q commute with arbitrary base changes.


That is, for any k-algebra R, we have canonical isomorphisms

(k ∩ q) ⊗k R ≅ (k ⊗k R) ∩ (q ⊗k R)

(p ∩ q) ⊗k R ≅ (p ⊗k R) ∩ (q ⊗k R).

45
Proof. Since Q is θ-stable, so is q in g. Moreover,

g = k ⊕ p, q = (k ∩ q) ⊕ (p ∩ q)

exhibit the eigenspace decompositions of g and q respectively for the involution


θ. Since we are working over Z [1/2]-algebras, these decompositions commute
with arbitrary base changes. This shows (1). Part (2) is a formal consequence
of (1).
Proof of Theorem 4.26. For (1), we may only see the assertions at t = 0 by
a similar argument to Theorem 4.6 (use Proposition 4.21 and Example 4.19).
Recall that the fiber of Q○ at t = 0 is (Q ∩ K ○) ⋉ p ∩ q by Lemma 4.25. We wish
to prove that NK ○ ⋉p (q0 ) = (Q ∩ K ○ ) ⋉ p ∩ q. Since the Lie algebra of the fiber
of Q○ at t = 0 is q0 , NK ○ ⋉p (q0 ) contains (Q ∩ K ○ ) ⋉ p ∩ q. We wish to prove the
converse containment.
We note that the adjoint representation of K ○ ⋉ p on the Lie algebra k ⋉ p of

K ⋉ p is expressed as

Ad((g, x))(y, z) = (Ad(g)y, Ad(g)([x, y] + z)), (15)

where (g, x) ∈ (K ○ ⋉ p)(R) and (y, z) ∈ (k ⋉ p) ⊗k R with R running through


all commutative k-algebras. Let R be a commutative k-algebra, and (g, x)
be an R-point of K ○ ⋉ p. If (g, x) normalizes ((q ∩ k) ⋉ (p ∩ q)) ⊗k R then
g ∈ (K ○ ∩Q)(R) by (15) and Lemmas 4.25 (2), 4.11. Henceforth we may assume
g ∈ (K ○ ∩ Q)(R). Then (g, x) normalizes ((q ∩ k) ⋉ (p ∩ q)) ⊗k R if and only if x
belongs to Np (k ∩ q; p ∩ q)(R).
The proof of (1) will be completed by showing Np (k ∩ q; p ∩ q) = p ∩ q. It is
evident that the left hand side contains the right hand side. To see the converse,
observe that the left hand side is equal to Np (k ∩ q; q) by Lemma 4.30 (2). A
similar argument to [13, Rappel 5.3.0] shows that Np (k ∩ q; q) is represented by
an affine k-scheme of finite presentation. The right hand side p ∩ q is represented
by a smooth affine k-scheme by Lemma 4.24. We may therefore pass to the fibers
by [23, Corollaire (17.9.5)]. That is, we wish to prove

Np (k ∩ q; q) ⊗k κ(P) = p ∩ q ⊗k κ(P)

for each prime ideal P ⊂ k. Lemmas 4.29 and 4.30 imply

Np (k ∩ q; q) ⊗k κ(P) ≅ Npκ(P) (k ∩ qκ(P) ; qκ(P) )


≅ Npκ(P) ((k ∩ q)κ(P) ; qκ(P) )
≅ Npκ(P) (kκ(P) ∩ qκ(P) ; qκ(P) ).

The assertion now follows from the hypothesis (14). This shows (1).
Part (2) follows from Theorem 2.17 and [17] Théorème 10.1.2.

46
A Scheme theoretic image
Following [28, Chapter 2, Exercise 3.17] and [34, Sections 01R5, 01RA, 01U2],
we collect a few basic facts on scheme-theoretic image which we use in this
paper.
Definition A.1 ([28, Chapter 2, Exercise 3.17]). Let f ∶ X → Y be a quasi-
compact morphism of schemes. Then we write Im f ∶= Spec OY / Ker f ♯ , and call
it the scheme-theoretic image of f .
Definition A.2 ([34, Definition 01RB, Lemmas 01RD, 01RG]). Let i ∶ Y ↪ X
be a quasi-compact (not necessarily open) immersion. Then we call Im i the
scheme-theoretic closure of Y in X. We say Y is scheme-theoretically dense in
X if Im i = X, in which case i is automatically an open immersion.
The following evident fact will be used repeatedly in Section 4.2:
Example A.3 (cf. [34, Example 056A]). Let k be a commutative ring. Then
Spec k [t±1 ] is scheme-theoretically dense in Spec k [t].
Example A.4 ([28, Chapter 2, Proposition 2.24]). Let i ∶ Y ↪ X be a closed
immersion. Then we have a canonical isomorphism Y ≅ Im i. We will identify
them to write Y = Im i.
The scheme-theoretic image commutes with formation of flat base changes:
Proposition A.5 ([34, Lemma 081I]). Let f ∶ X → Y be a quasi-compact
morphism of schemes, and g ∶ Y ′ → Y be a flat morphism of schemes. Then the
scheme-theoretic image of the projection X ×Y Y ′ → Y ′ is canonically isomorphic
to Im f ×Y Y ′ .
Once we find a candidate for the scheme-theoretic image, the following as-
sertion is useful for proving that it is exactly so:

Proposition A.6. Let X → Y → Z be quasi-compact morphisms of schemes.


f g

If Im f = Y then we have Im g = Im(g ○ f ). In particular, if Im f = Y and g is a


closed immersion then we have Im(g ○ f ) = Y .
Proof. We remark that the equality Im f = Y holds if and only if f ♯ is monic.
Assume these equivalent conditions. Then its direct image
g∗ f ♯ ∶ g∗ OY → g∗ f∗ OX = (g ○ f )∗ OX
is monic. Notice also that the structure homomorphism
(g ○ f )♯ ∶ OZ → (g ○ f )∗ OX
can be identified with the composite map
g♯ g∗ f ♯
OZ → g ∗ OY ↪ g ∗ f ∗ OX .
Therefore the kernels of g ♯ and (g ○ f )♯ coincde. The assertion now follows by
definition of the scheme-theoretic image.

47
References
[1] J. Adams and O. Taı̈bi. Galois and Cartan cohomology of real groups.
Duke Math. J., 167(6):1057–1097, 2018.
[2] S. Anantharaman. Schémas en groupes, espaces homogènes et espaces
algébriques sur une base de dimension 1. In Sur les groupes algébriques,
pages 5–79. Bull. Soc. Math. France, Mém. 33. 1973.
[3] D. Barbasch, N. Higson, and E. Subag. Algebraic families of groups and
commuting involutions. Internat. J. Math., 29(4):1850030, 18, 2018.

[4] J. Bernstein, N. Higson, and E. Subag. Contractions of representations and


algebraic families of Harish-Chandra modules. Int. Math. Res. Not. IMRN,
(11):3494–3520, 2020.
[5] J. Bernstein, N. Higson, and E. Subag. Algebraic families of Harish-
Chandra pairs. Int. Math. Res. Not. IMRN, (15):4776–4808, 2020.
[6] J. E. Bertin. Généralités sur les préschémas en groupes. In Schémas en
Groupes (Sém. Géométrie Algébrique, Inst. Hautes Études Sci., 1963/64),
Fasc. 2a, Exposé 6b, page 112. Inst. Hautes Études Sci., Paris, 1965.
[7] A. Borel. Linear algebraic groups, volume 126 of Graduate Texts in Math-
ematics. Springer-Verlag, New York, second edition, 1991.
[8] B. Conrad. Reductive group schemes. In Autour des schémas en groupes.
Vol. I, volume 42/43 of Panor. Synthèses, pages 93–444. Soc. Math. France,
Paris, 2014.
[9] M. Demazure. Structures algébriques, cohomologie des groupes. In Schémas
en Groupes (Sém. Géométrie Algébrique, Inst. Hautes Études Sci., 1963),
Fasc. 1, Exposé 1, page 42. Inst. Hautes Études Sci., Paris, 1963.
[10] M. Demazure. Fibrés tangents, algèbres de Lie. In Schémas en Groupes
(Sém. Géométrie Algébrique, Inst. Hautes Études Sci., 1963), Fasc. 1,
Exposé 2, page 40. Inst. Hautes Études Sci., Paris, 1963.
[11] M. Demazure. Topologies et faisceaux. In Schémas en Groupes (Sém.
Géométrie Algébrique, Inst. Hautes Études Sci., 1963), Fasc. 1, Exposé 4,
page 91. Inst. Hautes Études Sci., Paris, 1963.
[12] M. Demazure. Groupes réductifs—généralités. In Schémas en Groupes
(Sém. Géométrie Algébrique, Inst. Hautes études Sci., 1964), Fasc. 6, Ex-
posé 19, page 34. Inst. Hautes Études Sci., Paris, 1965.
[13] M. Demazure. Groupes réductifs: Déploiements, sous-groupes, groupes-
quotients. In Schémas en Groupes (Sém. Géométrie Algébrique, Inst.
Hautes Études Sci., 1964), Fasc. 6, Exposé 22, page 106. Inst. Hautes
Études Sci., Paris, 1965.

48
[14] M. Demazure. Automorphismes des groupes réductifs. In Schémas en
Groupes (Sém. Géométrie Algébrique, Inst. Hautes Études Sci., 1963/64),
Fasc. 7, Exposé 24, page 87. Inst. Hautes Études Sci., Paris, 1965.
[15] M. Demazure. Sous-groupes paraboliques des groupes réductifs. In
Schémas en Groupes (Sém. Géométrie Algébrique, Inst. Hautes Études Sci.,
1963/64), Fasc. 7, Exposé 26, page 91. Inst. Hautes Études Sci., Paris,
1966.
[16] M. Demazure and P. Gabriel. Introduction to algebraic geometry and al-
gebraic groups, volume 39 of North-Holland Mathematics Studies. North-
Holland Publishing Co., Amsterdam-New York, 1980. Translated from the
French by J. Bell.
[17] P. Gabriel. Construction de préschémas quotient. In Schémas en Groupes
(Sém. Géométrie Algébrique, Inst. Hautes Études Sci., 1963/64), Fasc. 2a,
Exposé 5, page 37. Inst. Hautes Études Sci., Paris, 1963.
[18] L. Godinho and J. Natário. An introduction to Riemannian geometry.
Universitext. Springer, Cham, 2014. With applications to mechanics and
relativity.
[19] U. Görtz and T. Wedhorn. Algebraic geometry I. Schemes. Springer
Studium Mathematik—Master. Springer Spektrum, Wiesbaden, second
edition, [2020] ©2020. With examples and exercises.
[20] A. Grothendieck. Éléments de géométrie algébrique. II. Étude globale
élémentaire de quelques classes de morphismes. Inst. Hautes Études Sci.
Publ. Math., (8):222, 1961.
[21] A. Grothendieck. Groupes de type multiplicatif: Homomorphismes dans un
schéma en groupes. In Schémas en Groupes (Sém. Géométrie Algébrique,
Inst. Hautes Études Sci., 1963/64), pages Fasc. 3, Exposé 9, 37. Inst.
Hautes Études Sci., Paris, 1964.
[22] A. Grothendieck. Éléments de géométrie algébrique. IV. Étude locale des
schémas et des morphismes de schémas. III. Inst. Hautes Études Sci. Publ.
Math., (28):255, 1966.
[23] A. Grothendieck. Éléments de géométrie algébrique. IV. Étude locale des
schémas et des morphismes de schémas IV. Inst. Hautes Études Sci. Publ.
Math., (32):361, 1967.
[24] T. Hayashi. Half-integrality of line bundles on partial flag schemes of clas-
sical Lie groups, 2021. arXiv:2104.13304.
[25] T. Hayashi. SO(3)-homogeneous decomposition of the flag scheme of SL3
over Z [1/2], 2021. arXiv:2111.07905.

49
[26] T. Hayashi, F. Januszewski. Families of twisted D-modules and arithmetic
models of Harish-Chandra modules. arXiv:1808.10709.
[27] E. Inonu and E. P. Wigner. On the contraction of groups and their repre-
sentations. Proc. Nat. Acad. Sci. U.S.A., 39:510–524, 1953.
[28] Q. Liu. Algebraic geometry and arithmetic curves, volume 6 of Oxford
Graduate Texts in Mathematics. Oxford University Press, Oxford, 2002.
Translated from the French by Reinie Erné, Oxford Science Publications.
[29] J. S. Milne. Algebraic groups, volume 170 of Cambridge Studies in Advanced
Mathematics. Cambridge University Press, Cambridge, 2017. The theory
of group schemes of finite type over a field.
[30] E. A. Nisnevič. Affine homogeneous spaces and finite subgroups of arith-
metic groups over function fields. Funkcional. Anal. i Priložen., 11(1):73–
74, 1977.
[31] R. W. Richardson. Affine coset spaces of reductive algebraic groups. Bull.
London Math. Soc., 9(1):38–41, 1977.
[32] R. W. Richardson and T. A. Springer. The Bruhat order on symmetric
varieties. Geom. Dedicata, 35(1-3):389–436, 1990.
[33] J-P. Serre. Galois cohomology. Springer Monographs in Mathematics.
Springer-Verlag, Berlin, english edition, 2002. Translated from the French
by Patrick Ion and revised by the author.
[34] The Stacks Project Authors. Stacks Project. http://stacks.math.
columbia.edu, 2023.
[35] R. Steinberg. Endomorphisms of linear algebraic groups. Memoirs of the
American Mathematical Society, No. 80. American Mathematical Society,
Providence, R.I., 1968.
[36] M. E. Sweedler. Hopf algebras. Mathematics Lecture Note Series. W. A.
Benjamin, Inc., New York, 1969.

50

You might also like