2

7844tp.new.indd 1 7/27/10 10:34 AM

Algebra and Discrete Mathematics ISSN: 1793-5873

Managing Editor: Rüdiger Göbel (University Duisburg-Essen, Germany)

Editorial Board: Elisabeth Bouscaren, Manfred Droste, Katsuya Eda,

Emmanuel Dror Farjoun, Angus MacIntyre, H.Dugald Macpherson,

José Antonio de la Peña, Luigi Salce, Mark Sapir, Lutz Strüngmann,

Simon Thomas

The series ADM focuses on recent developments in all branches of algebra and

topics closely connected. In particular, it emphasizes combinatorics, set theoretical

methods, model theory and interplay between various fields, and their influence on

algebra and more general discrete structures. The publications of this series are of

special interest to researchers, post-doctorals and graduate students. It is the

intention of the editors to support fascinating new activities of research and to

spread the new developments to the entire mathematical community.

Vol. 1: Aspects of Infinite Groups: A Festschrift in Honor of Anthony Gaglione

eds. Benjamin Fine, Gerhard Rosenberger & Dennis Spellman

Vol. 2: Grassmannians of Classical Buildings

by Mark Pankov

LaiFun - Grassmannians of Classical Blds.pmd 6/21/2010, 10:31 AM 2

NE W J E RSE Y • L ONDON • SI NGAP ORE • BE I J I NG • SHANGHAI • HONG KONG • TAI P E I • CHE NNAI

World Scientifc

Mark Pankov

University of Warmia and Mazury, Poland

Algebra and Discrete Mathematics Vol. 2

7844tp.new.indd 2 7/27/10 10:34 AM

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

For photocopying of material in this volume, please pay a copying fee through the Copyright

Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to

photocopy is not required from the publisher.

ISBN-13 978-981-4317-56-6

ISBN-10 981-4317-56-X

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,

electronic or mechanical, including photocopying, recording or any information storage and retrieval

system now known or to be invented, without written permission from the Publisher.

Copyright © 2010 by World Scientific Publishing Co. Pte. Ltd.

Published by

World Scientific Publishing Co. Pte. Ltd.

5 Toh Tuck Link, Singapore 596224

USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601

UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

Printed in Singapore.

GRASSMANNIANS OF CLASSICAL BUILDINGS

Algebra and Discrete Mathematics — Vol. 2

LaiFun - Grassmannians of Classical Blds.pmd 6/21/2010, 10:31 AM 1

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Dedicated

to

my wife Inna

and

daughters Sabina and Varvara

v

This page intentionally left blank This page intentionally left blank

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Preface

Tits buildings or simple buildings are combinatorial constructions success-

fully exploited to study various types of groups (classical, simple algebraic,

Kac–Moody). One of historical backgrounds of this concept is Cartan’s

well-known classiﬁcation of simple Lie groups. We refer to [Abramenko

(1996); Brown (1989); Garrett (1997); Ronan (1989); Scharlau (1995); Tits

(1974)] for various aspects of building theory.

Buildings can be obtained from groups admitting Tits systems. Such

groups form a suﬃciently wide class which contains classical groups, reduc-

tive algebraic groups and others. The formal deﬁnition of a building is pure

combinatorial and does not depend on a group. In [Tits (1974)] a build-

ing is deﬁned as a simplicial complex with a family of subcomplexes called

apartments and satisfying certain axioms. All apartments are isomorphic

to the simplicial complex obtained from a Coxeter system which deﬁnes the

building type.

The vertex set of a building can be labeled by the nodes of the diagram

of the associated Coxeter system. The set of all vertices corresponding

to the same node is called a Grassmannian (more general objects were

investigated in [Pasini (1994)]). This term is motivated by the fact that

every building of type A

n

is isomorphic to the ﬂag complex of an (n + 1)-

dimensional vector space and the Grassmannians of the building can be

identiﬁed with the Grassmannians of this vector space. Every building

Grassmannian has a natural structure of a partial linear space (point-line

geometry); this partial linear space is called the Grassmann space associated

with the Grassmannian.

The aim of this book is to present both classical and more recent re-

sults on Grassmannians of buildings of classical types (A

n

, B

n

= C

n

, D

n

).

These results will be formulated in terms of point-line geometry. A large

vii

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

viii Grassmannians of Classical Buildings

portion of them is a part of the area known as characterizations of geomet-

rical transformations under mild hypotheses. Roughly speaking, we want

to show that some mappings of Grassmannians can be extended to map-

pings of the associated buildings. Other results are related with structural

properties of apartments. Also we show that our methods work for some

geometric constructions non-related with buildings — Grassmannians of

inﬁnity-dimensional vector spaces, the sets of conjugate linear involutions

and Grassmannians of exchange spaces.

The book is self-contained and prospective audience includes researchers

working in algebra, combinatorics and geometry, as well as, graduate and

advanced undergraduate students. The requirement to the reader is know-

ledge of basics of algebra and graph theory.

Acknowledge

The present book contains the main results of my Doctor of Science

(habilitation) thesis presented in Institute of Mathematics NASU and I

thank A. Samoilenko and V. Sharko for supporting my research.

I am grateful to H. Van Maldeghem who read a preliminary version of

the book and made a long list of valuable comments. Also I am grateful to

H. Havlicek and M. Kwiatkowski for useful discussions and remarks.

Finally, I thank J. Kosiorek for drawing the pictures.

Mark Pankov

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Contents

Preface vii

0. Introduction 1

1. Linear Algebra and Projective Geometry 7

1.1 Vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.1.1 Division rings . . . . . . . . . . . . . . . . . . . . 8

1.1.2 Vector spaces over division rings . . . . . . . . . . 10

1.1.3 Dual vector space . . . . . . . . . . . . . . . . . . 14

1.2 Projective spaces . . . . . . . . . . . . . . . . . . . . . . . 17

1.2.1 Linear and partial linear spaces . . . . . . . . . . 17

1.2.2 Projective spaces over division rings . . . . . . . . 19

1.3 Semilinear mappings . . . . . . . . . . . . . . . . . . . . . 20

1.3.1 Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . 20

1.3.2 Mappings of Grassmannians induced by semilinear

mappings . . . . . . . . . . . . . . . . . . . . . . . 21

1.3.3 Contragradient . . . . . . . . . . . . . . . . . . . . 25

1.4 Fundamental Theorem of Projective Geometry . . . . . . 26

1.4.1 Main theorem and corollaries . . . . . . . . . . . . 26

1.4.2 Proof of Theorem 1.4 . . . . . . . . . . . . . . . . 28

1.4.3 Fundamental Theorem for normed spaces . . . . . 32

1.4.4 Proof of Theorem 1.5 . . . . . . . . . . . . . . . . 33

1.5 Reﬂexive forms and polarities . . . . . . . . . . . . . . . . 37

1.5.1 Sesquilinear forms . . . . . . . . . . . . . . . . . . 37

1.5.2 Reﬂexive forms . . . . . . . . . . . . . . . . . . . 38

1.5.3 Polarities . . . . . . . . . . . . . . . . . . . . . . . 40

ix

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

x Grassmannians of Classical Buildings

2. Buildings and Grassmannians 43

2.1 Simplicial complexes . . . . . . . . . . . . . . . . . . . . . 43

2.1.1 Deﬁnition and examples . . . . . . . . . . . . . . 43

2.1.2 Chamber complexes . . . . . . . . . . . . . . . . . 46

2.1.3 Grassmannians and Grassmann spaces . . . . . . 47

2.2 Coxeter systems and Coxeter complexes . . . . . . . . . . 49

2.2.1 Coxeter systems . . . . . . . . . . . . . . . . . . . 49

2.2.2 Coxeter complexes . . . . . . . . . . . . . . . . . . 52

2.2.3 Three examples . . . . . . . . . . . . . . . . . . . 53

2.3 Buildings . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.3.1 Deﬁnition and elementary properties . . . . . . . 55

2.3.2 Buildings and Tits systems . . . . . . . . . . . . . 57

2.3.3 Classical examples . . . . . . . . . . . . . . . . . . 59

2.3.4 Spherical buildings . . . . . . . . . . . . . . . . . 62

2.3.5 Mappings of the chamber sets . . . . . . . . . . . 63

2.4 Mappings of Grassmannians . . . . . . . . . . . . . . . . . 65

2.5 Appendix: Gamma spaces . . . . . . . . . . . . . . . . . . 67

3. Classical Grassmannians 69

3.1 Elementary properties of Grassmann spaces . . . . . . . . 70

3.2 Collineations of Grassmann spaces . . . . . . . . . . . . . 75

3.2.1 Chow’s theorem . . . . . . . . . . . . . . . . . . . 75

3.2.2 Chow’s theorem for linear spaces . . . . . . . . . . 77

3.2.3 Applications of Chow’s theorem . . . . . . . . . . 78

3.2.4 Opposite relation . . . . . . . . . . . . . . . . . . 80

3.3 Apartments . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.3.1 Basic properties . . . . . . . . . . . . . . . . . . . 83

3.3.2 Proof of Theorem 3.8 . . . . . . . . . . . . . . . . 85

3.4 Apartments preserving mappings . . . . . . . . . . . . . . 87

3.4.1 Results . . . . . . . . . . . . . . . . . . . . . . . . 87

3.4.2 Proof of Theorem 3.10: First step . . . . . . . . . 89

3.4.3 Proof of Theorem 3.10: Second step . . . . . . . . 93

3.5 Grassmannians of exchange spaces . . . . . . . . . . . . . 95

3.5.1 Exchange spaces . . . . . . . . . . . . . . . . . . . 95

3.5.2 Grassmannians . . . . . . . . . . . . . . . . . . . . 96

3.6 Matrix geometry and spine spaces . . . . . . . . . . . . . 100

3.7 Geometry of linear involutions . . . . . . . . . . . . . . . 102

3.7.1 Involutions and transvections . . . . . . . . . . . . 102

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Contents xi

3.7.2 Adjacency relation . . . . . . . . . . . . . . . . . . 104

3.7.3 Chow’s theorem for linear involutions . . . . . . . 108

3.7.4 Proof of Theorem 3.15 . . . . . . . . . . . . . . . 110

3.7.5 Automorphisms of the group GL(V ) . . . . . . . . 113

3.8 Grassmannians of inﬁnite-dimensional vector spaces . . . 114

3.8.1 Adjacency relation . . . . . . . . . . . . . . . . . . 114

3.8.2 Proof of Theorem 3.17 . . . . . . . . . . . . . . . 116

3.8.3 Base subsets . . . . . . . . . . . . . . . . . . . . . 118

3.8.4 Proof of Theorem 3.18 . . . . . . . . . . . . . . . 118

4. Polar and Half-Spin Grassmannians 123

4.1 Polar spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 125

4.1.1 Axioms and elementary properties . . . . . . . . . 125

4.1.2 Proof of Theorem 4.1 . . . . . . . . . . . . . . . . 126

4.1.3 Corollaries of Theorem 4.1 . . . . . . . . . . . . . 129

4.1.4 Polar frames . . . . . . . . . . . . . . . . . . . . . 130

4.2 Grassmannians . . . . . . . . . . . . . . . . . . . . . . . . 133

4.2.1 Polar Grassmannians . . . . . . . . . . . . . . . . 133

4.2.2 Two types of polar spaces . . . . . . . . . . . . . 136

4.2.3 Half-spin Grassmannians . . . . . . . . . . . . . . 138

4.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

4.3.1 Polar spaces associated with sesquilinear forms . . 141

4.3.2 Polar spaces associated with quadratic forms . . . 146

4.3.3 Polar spaces of type D

3

. . . . . . . . . . . . . . . 147

4.3.4 Embeddings in projective spaces and classiﬁcation 149

4.4 Polar buildings . . . . . . . . . . . . . . . . . . . . . . . . 150

4.4.1 Buildings of type C

n

. . . . . . . . . . . . . . . . . 150

4.4.2 Buildings of type D

n

. . . . . . . . . . . . . . . . 150

4.5 Elementary properties of Grassmann spaces . . . . . . . . 151

4.5.1 Polar Grassmann spaces . . . . . . . . . . . . . . 151

4.5.2 Half-spin Grassmann spaces . . . . . . . . . . . . 154

4.6 Collineations . . . . . . . . . . . . . . . . . . . . . . . . . 159

4.6.1 Chow’s theorem and its generalizations . . . . . . 159

4.6.2 Weak adjacency on polar Grassmannians . . . . . 161

4.6.3 Proof of Theorem 4.8 for k < n −2 . . . . . . . . 162

4.6.4 Proof of Theorems 4.7 and 4.8 . . . . . . . . . . . 164

4.6.5 Proof of Theorem 4.9 . . . . . . . . . . . . . . . . 165

4.6.6 Remarks . . . . . . . . . . . . . . . . . . . . . . . 172

4.7 Opposite relation . . . . . . . . . . . . . . . . . . . . . . . 174

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

xii Grassmannians of Classical Buildings

4.7.1 Opposite relation on polar Grassmannians . . . . 174

4.7.2 Opposite relation on half-spin Grassmannians . . 175

4.8 Apartments . . . . . . . . . . . . . . . . . . . . . . . . . . 178

4.8.1 Apartments in polar Grassmannians . . . . . . . . 178

4.8.2 Apartments in half-spin Grassmannians . . . . . . 181

4.8.3 Proof of Theorem 4.15 . . . . . . . . . . . . . . . 184

4.9 Apartments preserving mappings . . . . . . . . . . . . . . 186

4.9.1 Apartments preserving bijections . . . . . . . . . 186

4.9.2 Inexact subsets of polar Grassmannians . . . . . . 187

4.9.3 Complement subsets of polar Grassmannians . . . 194

4.9.4 Inexact subsets of half-spin Grassmannians . . . . 199

4.9.5 Proof of Theorem 4.16 . . . . . . . . . . . . . . . 201

4.9.6 Embeddings . . . . . . . . . . . . . . . . . . . . . 202

4.9.7 Proof of Theorems 4.17 and 4.18 . . . . . . . . . . 203

Bibliography 207

Index 211

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Chapter 0

Introduction

This short chapter is an informal description of the main objects of the

book — buildings and their Grassmannians. All precise deﬁnitions will be

given in Chapter 2.

The simplest buildings are so-called Coxeter complexes — a class of

simplicial complexes deﬁned by Coxeter systems. Let W be a group and

S be a set of generators for W such that each element of S is an involu-

tion. The pair (W, S) is a Coxeter system if the group W has the following

presentation

S : (ss

)

m(s,s

)

= 1, (s, s

) ∈ S ×S, m(s, s

) < ∞ ,

where m(s, s

) is the order of ss

**. The associated Coxeter complex Σ(W, S)
**

is the simplicial complex whose simplices can be identiﬁed with special

subsets of type wX with X ⊂ S and w ∈ W. Every Coxeter system

can be uniquely (up to an isomorphism) reconstructed from its diagram.

The diagram associated with (W, S) is the graph whose vertex set is S

and s, s

∈ S are connected by m(s, s

**) − 2 edges. In the case when W is
**

ﬁnite, we get a Dynkin diagram without directions. There is a complete

description of all ﬁnite Coxeter systems.

Similarly, more complicated buildings can be obtained from the Tits

systems. A Tits system (G, B, N, S) is a structure on a group G consisting

of two subgroups B, N which span G and a set of generators S of the

quotient group

W := N/(B ∩ N);

note that the pair (B, N) is called a BN-pair. By one of the basic properties

of the Tits systems, (W, S) is a Coxeter system. The second remarkable

property is the fact that every subgroup containing B, such subgroups are

called special, can be reconstructed from elements of W and B. Moreover,

1

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

2 Grassmannians of Classical Buildings

there is a natural one-to-one correspondence between special subgroups and

subgroups of W generated by subsets of S. The building associated with

the Tits system (G, B, N, S) is the simplicial complex ∆(G, B, N, S) whose

simplices can be identiﬁed with special subsets of type gP, where g ∈ G

and P is a special subgroup. If w ∈ W then for every g, g

∈ N belonging

to w and every special subgroup P we have gP = g

**P; denote this special
**

subset by wP. All such special subsets form a subcomplex Σ isomorphic

to Σ(W, S). The left action of the group G on the building deﬁnes the

family of subcomplexes gΣ, g ∈ G. These subcomplexes are isomorphic to

Σ(W, S) and called apartments.

Following [Tits (1974)], we deﬁne an abstract building as a simplicial

complex ∆ with a family of subcomplexes called apartments and satisfying

the following axioms:

• all apartments are Coxeter complexes,

• for any two simplices of ∆ there is an apartment containing both of

them,

• a technical condition concerning the existence of “nice” isomorphisms

between apartments.

By this deﬁnition, every Coxeter complex is a building with a unique apart-

ment. So, there exists a Coxeter system (W, S) such that all apartments of

∆ are isomorphic to Σ(W, S). The type of the building ∆ is deﬁned by the

diagram of (W, S). We restrict ourselves to so-called thick buildings only;

in such buildings maximal simplices form a suﬃciently wide class. Coxeter

complexes do not satisfy this condition.

Let V be an (n+1)-dimensional vector space and ∆(V ) be the ﬂag com-

plex of V , i.e., the simplicial complex consisting of all ﬂags of V . For every

base B of V the subcomplex Σ

B

which consists of all ﬂags formed by linear

subspaces spanned by subsets of B is called the apartment of ∆(V ) asso-

ciated with the base B. Every Σ

B

is isomorphic to the simplicial complex

of the Coxeter system (W, S), where W is the group of all permutations

on the set {1, . . . , n + 1} and S is formed by all transpositions (i, i + 1);

the associated diagram is A

n

. The simplicial complex ∆(V ) together with

the family of all such apartments is a building of type A

n

. Note that this

building can be obtained from the Tits system of the group GL(V ).

Let ∆ be a building and (W, S) be the associated Coxeter system. The

vertex set of ∆ can be naturally decomposed in |S| disjoint subsets called

Grassmannians. If the building is associated with a Tits system for a certain

group G then the Grassmannians are the orbits of the left action of G on

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Introduction 3

the vertex set. In the general case, this decomposition is related to the fact

that the vertex set of ∆ can be labeled by elements of S and this labeling

is unique up to a permutation on S.

Let G be a Grassmannian of ∆. The intersection of G with an apartment

of ∆ is called an apartment of the Grassmannian G. Two distinct elements

a, b ∈ G are said to be adjacent if there exists a simplex P ∈ ∆ such that

P ∪{a} and P ∪{b} are maximal simplices. In this case, the subset formed

by all c ∈ G such that P ∪ {c} is a maximal simplex will be called the line

joining a and b. So, we get a structure known as a partial linear space or

a point-line geometry, i.e., a set of points together with a family of subsets

called lines and satisfying some simple axioms. This partial linear space is

said to be the Grassmann space associated with G.

The term “Grassmannian” is motivated by the following example. Let V

be an (n+1)-dimensional vector space. The Grassmannians of the building

∆(V ) are the usual Grassmannians G

k

(V ), k ∈ {1, . . . , n}, formed by all

k-dimensional linear subspaces of V . The Grassmann spaces corresponding

to G

1

(V ) and G

n

(V ) are the projective space associated with V and the

dual projective space, respectively. In particular, any two distinct elements

of these Grassmannians are adjacent. The Grassmann space of G

k

(V ),

1 < k < n, is more complicated. It contains non-adjacent elements. Note

that our adjacency relation coincides with the classical adjacency relation

introduced in [Chow (1949)]: two elements of G

k

(V ) are adjacent if and

only if their intersection is (k −1)-dimensional.

Recall that a building is spherical if the associated Coxeter system is

ﬁnite, and it is irreducible if the diagram is connected. Irreducible thick

spherical buildings of rank ≥ 3 were classiﬁed in [Tits (1974)]. There are

precisely the following seven types of such buildings:

A

n

B

n

= C

n

D

n

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

4 Grassmannians of Classical Buildings

E

6

E

7

E

8

F

4

The ﬁrst three types are called classical, the remaining four are known

as exceptional. Every thick building of type A

n

(n ≥ 3) is isomorphic

to the ﬂag complex of a certain (n + 1)-dimensional vector space. All

thick buildings of types C

n

and D

n

can be obtained from polar spaces, see

Chapter 4. Exceptional buildings are related with so-called metasymplectic

and parapolar spaces, see [Cohen (1995)].

In the present book we will consider Grassmannians associated with

buildings of classical types only. Some information concerning Grassman-

nians of exceptional buildings can be found in [Cohen (1995)].

Investigation of building Grassmannians goes back to [Chow (1949);

Dieudonn´e 2 (1951)] (see also Chapter III in [Dieudonn´e (1971)]) and is

continued in [Cameron (1982)]. Currently, there are several research direc-

tions:

• axiomatic characterizations of Grassmann spaces, a survey can be found

in [Cohen (1995)];

• embeddings in projective spaces, hyperplanes and generalized rank, see

[Cooperstein (2003)] for a survey;

• subspaces of Grassmann spaces [Cooperstein, Kasikova and Shult

(2005); Cooperstein (2005, 2007)];

• apartment properties [Blok and Brouwer (1998); Cooperstein and Shult

(1997); Cooperstein, Kasikova and Shult (2005); Pankov 3 (2007)];

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Introduction 5

• characterizations of geometrical transformations of Grassmannians un-

der “mild hypotheses” [Havlicek (1995); Huang (1998, 2000, 2001);

Huang and Kreuzer (1995); Kreuzer (1998); Pankov, Pra˙ zmowski and

˙

Zynel (2006)] and the author’s papers refereed in the book.

We describe all apartments preserving mappings of Grassmannians asso-

ciated with buildings of classical types and collineations (isomorphisms)

of the corresponding Grassmann spaces. Actually, the Fundamental Theo-

rem of Projective Geometry and classical Chow’s theorems [Chow (1949)]

are partial cases of our results. The methods are based on deep structural

properties of apartments (connections between the adjacency relation and

apartments). Roughly speaking, we work in the latter two directions men-

tioned above. Also, we establish similar results for some geometric construc-

tions non-related with buildings, for example, Grassmannians of inﬁnite-

dimensional vector spaces and the sets of conjugate linear involutions. One

of them joins Chow’s theorem with Dieudonn´e–Rickart’s classiﬁcation of

automorphisms of the linear group [Dieudonn´e 1 (1951); Rickart (1950)].

This page intentionally left blank This page intentionally left blank

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Chapter 1

Linear Algebra and Projective

Geometry

In Sections 1.1 and 1.2 we consider vector spaces over division rings and

the associated projective spaces; it must be pointed out that vector spaces

are not assumed to be ﬁnite-dimensional.

The main objects of the chapter are so-called semilinear mappings of

vector spaces and the mappings of Grassmannians induced by them (Section

1.3).

By the Fundamental Theorem of Projective Geometry, every

collineation (isomorphism) of projective spaces is induced by a semilin-

ear isomorphism of the associated vector spaces. In Section 1.4 we prove

a more general result known as Faure–Fr¨ olicher–Havlicek’s version of the

Fundamental Theorem.

The second result of Section 1.4 is Mackey’s theorem concerning isomor-

phisms of the lattices formed by closed linear subspaces of normed vector

spaces (real and complex). This theorem states that all such isomorphisms

are induced by linear and conjugate-linear homeomorphisms of the associ-

ated normed spaces (the second possibility is realized only in the complex

case).

The Fundamental Theorem of Projective Geometry can be reformu-

lated in the following form [Baer (1952)]: every isomorphism of the lattices

formed by linear subspaces of vector spaces is induced by a semilinear iso-

morphism. By this reason, Mackey’s result is interpreted as the Fundamen-

tal Theorem for normed spaces.

In Section 1.5 we recall basic facts on reﬂexive sesquilinear forms.

7

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

8 Grassmannians of Classical Buildings

1.1 Vector spaces

1.1.1 Division rings

Let R be a non-empty set with additive and multiplicative operations +

and satisfying the following conditions:

• (R, +) is an Abelian group (the identity element is denoted by 0),

• (R ¸ ¦0¦, ) is a group (the identity element is denoted by 1).

If the distributive axioms

a(b +c) = ab +ac and (b +c)a = ba +ca ∀ a, b, c ∈ R

hold then we say that (R, +, ) is a division ring. A division ring is called

a ﬁeld if the multiplicative operation is commutative.

Exercise 1.1. Show that 0a = a0 = 0 and −1a = −a for every a ∈ R.

For every division ring R we deﬁne the opposite (or dual ) division ring

R

∗

as follows: R and R

∗

have the same set of elements and the same additive

operation, the multiplicative operation ∗ on R

∗

is deﬁned by the formula

a ∗ b = ba.

It is clear that R

∗∗

coincides with R, and we have R

∗

= R only in the case

when R is a ﬁeld.

The following result is well-known; see, for example, Theorem 1.14 in

[Artin (1957)].

Theorem 1.1 (J. H. M. Wedderburn). Every ﬁnite division ring is a

ﬁeld.

Let R and R

be division rings. We say that a mapping σ : R → R

is

a homomorphism if

σ(a +b) = σ(a) +σ(b) and σ(ab) = σ(a)σ(b)

for all a, b ∈ R. Bijective homomorphisms are called isomorphisms.

Exercise 1.2. Show that the equalities

σ(0) = 0, σ(1) = 1, σ(−1) = −1

hold for every non-zero homomorphism σ : R → R

.

Proposition 1.1. Every non-zero homomorphism σ : R → R

is injective.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 9

Proof. If σ : R → R

**is not injective then σ(a) = 0 for a certain non-zero
**

element a ∈ R and we have

σ(b) = σ(aa

−1

b) = σ(a)σ(a

−1

b) = 0

for every b ∈ R.

An isomorphism of a division ring to itself is said to be an automorphism.

The group formed by all automorphisms of R is denoted by Aut(R). An iso-

morphism of R to the opposite division ring R

∗

is called anti-automorphism

of R (if R is a ﬁeld then every anti-automorphism is an automorphism).

Proposition 1.2. Every non-zero homomorphism of R to itself is identity.

In particular, Aut(R) = ¦1

R

¦.

Proof. Let σ : R →R be a homomorphism. For every n ∈ N we have

n = 1 + + 1

. ¸¸ .

n

,

hence

σ(n) = σ(1) + +σ(1)

. ¸¸ .

n

= 1 + + 1

. ¸¸ .

n

= n.

This implies that σ(m) = m for all m ∈ Z. Since σ preserves division, the

restriction of σ to Q is identity.

For every real number a > 0 there exists b ∈ R such that a = b

2

, and

we have

σ(a) = σ(b)

2

> 0.

This means that σ is order preserving: if a > b then a −b > 0 and

σ(a) −σ(b) = σ(a −b) > 0

which implies that σ(a) > σ(b).

Every a ∈ R ¸ Q corresponds to unique section of rational numbers

¦ q ∈ Q : q < a ¦, ¦ r ∈ Q : a < r ¦.

Since σ is order preserving, it transfers this section in the section corre-

sponding to σ(a). On the other hand, σ sends every section to itself (its

restriction to Q is identity) and we get σ(a) = a.

The group Aut(C) is not trivial. It contains, for example, the complex

conjugate mapping z → ¯ z.

Exercise 1.3. Show that every non-zero continuous homomorphism of C

to itself is identity or coincides with the complex conjugate mapping. Hint:

the restriction of every non-zero homomorphism σ : C →C to Q is identity

and σ(i) = ±i.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

10 Grassmannians of Classical Buildings

There are a lot of other automorphisms of C, but they are non-

continuous; moreover, there exist non-surjective homomorphisms of C to

itself.

Example 1.1 (Finite ﬁelds). Every ﬁnite ﬁeld coincides with a certain

Galois’ ﬁeld GF(p, r) (p is a prime number). If r = 1 then we get Z

p

and

the group of automorphisms is trivial. In the general case, our ﬁeld consists

of p

r

elements; its multiplicative group is a cyclic group of order p

r

− 1;

the group of automorphisms is a cyclic group of order r generated by the

automorphism a → a

p

.

Example 1.2 (The division ring of real quaternions). Consider the

division ring H formed by the real quaternion numbers a + bi + cj + dk.

This is a 4-dimensional vector space over R (the canonical base consists of

1, i, j, k) with the multiplicative operation deﬁned by the following condi-

tions

i

2

= j

2

= k

2

= −1,

ij = k, jk = i, ki = j.

This division ring is non-commutative (for example, ik = i

2

j = −j = −ki).

Every automorphism of H is inner (x → qxq

−1

for a certain q ,= 0). The

conjugate mapping

a +bi +cj +dk → a −bi −cj −dk

is an anti-automorphism.

1.1.2 Vector spaces over division rings

Let R be a division ring and (V, +) be an Abelian group (the identity

element is 0). Let also

R V → V, (a, x) → ax

be a left action of R on V satisfying the following conditions:

(1) 1x = x for all x ∈ V ,

(2) a(bx) = (ab)x for all a, b ∈ R and x ∈ V .

If the distributive axioms

a(x +y) = ax +ay and (a +b)x = ax +bx ∀ a, b ∈ R, x, y ∈ V

hold then we say that V is a left vector space over R. Elements of V and

R are called vectors and scalars, respectively.

Similarly, every right action of R on V satisfying the conditions

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 11

(3) x1 = x for all x ∈ V ,

(4) (xb)a = x(ba) for all a, b ∈ R and x ∈ V

deﬁnes a right vector space over R if the corresponding distributive axioms

hold. Clearly, this action can be considered as a left action, then (4) must

be rewritten as

a(bx) = (ba)x or a(bx) = (a ∗ b)x,

where ∗ is the multiplicative operation of the opposite division ring.

Therefore, every right vector space over R can be presented as a left

vector space over R

∗

, and conversely. In what follows we restrict ourselves

to left vector spaces only.

Let V and V

**be left vector spaces over a division ring R. A mapping
**

l : V → V

is said to be linear if

l(x +y) = l(x) +l(y) and l(ax) = al(x)

for all x, y ∈ V and a ∈ R. We say that the vector spaces V and V

are

isomorphic if there exists a linear bijection of V to V

. Linear bijections

are called linear isomorphisms. The group of all linear automorphisms of

V (linear isomorphisms of V to itself) is denoted by GL(V ).

A subset S ⊂ V is a linear subspace of V if

ax +by ∈ S

for all x, y ∈ S and a, b ∈ R. By this deﬁnition, ¦0¦ and V are linear

subspaces. The intersection of any collection of linear subspaces is a linear

subspace. For every linear mapping l : V → V

**all vectors x ∈ V satisfying
**

l(x) = 0 form a linear subspace of V ; it is called the kernel of l and denoted

by Ker l.

For every subset X ⊂ V the intersection of all linear subspaces contain-

ing X is called the linear subspace spanned by X and denoted by ¸X); this

subspace consists of all linear combinations

a

1

x

1

+ +a

k

x

k

,

where x

1

, . . . , x

k

∈ X and a

1

, . . . , a

k

∈ R. A subset X ⊂ V is said to be

independent if the linear subspace ¸X) is not spanned by a proper subset

of X, in other words, for any distinct non-zero vectors x

1

, . . . , x

k

∈ X the

equality

a

1

x

1

+ +a

k

x

k

= 0

holds only in the case when all scalars a

1

, . . . , a

k

are zero.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

12 Grassmannians of Classical Buildings

An independent subset X ⊂ V is called a base of V if ¸X) coincides

with V . Similarly, we deﬁne bases of linear subspaces. Every independent

subset X ⊂ V is a base of the linear subspace ¸X).

Proposition 1.3. Every independent subset X ⊂ V can be extended to a

base of V ; in particular, bases of V exist.

Proof. Let X be the set of all independent subsets containing X. This set

is non-empty (X ∈ X) and it is partially ordered by the inclusion relation.

If Y is a linearly ordered subset of X (for any Y, Y

∈ Y we have Y ⊂ Y

or Y

**⊂ Y ) then the subset
**

Z :=

_

Y ∈Y

Y

is independent. Indeed, suppose that

a

1

x

1

+ +a

k

x

k

= 0

holds for non-zero vectors x

1

, . . . , x

k

∈ Z and non-zero scalars a

1

, . . . , a

k

;

consider Y

1

, . . . , Y

k

∈ Y containing x

1

, . . . , x

k

, respectively; one of these

subsets contains the others and is not independent, a contradiction. By

Zorn lemma, the set X has maximal elements. Every maximal element of

X is a base of V containing the subset X.

Theorem 1.2. Any two bases of a vector space have the same cardinality

(possible inﬁnite).

Proof. See Section II.2 in [Baer (1952)].

The cardinality of bases of V is called the dimension of V and denoted

by dimV . Since every linear subspace can be considered as a vector space,

the dimension of linear subspaces also is deﬁned.

Every linear isomorphism maps independent subsets to independent

subsets and bases to bases; thus isomorphic vector spaces have the same di-

mension. Every one-to-one correspondence between bases of V and V

can

be uniquely extended to a linear isomorphism between these vector spaces.

Proposition 1.4. For any two linear subspaces S, U ⊂ V there exists a

base of V such that S and U are spanned by subsets of this base.

Proof. Suppose that S ∩ U ,= 0. We take any base X of S ∩ U and

extend it to bases B

S

and B

U

of S and U, respectively. Show that the

subset B

S

∪ B

U

is independent (this is trivial if one of the subspaces is

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 13

contained in the other, and we consider the case when S and U are not

incident).

If the statement fails then there exist distinct vectors

x

1

, . . . , x

k

∈ B

S

, y

1

, . . . , y

m

∈ B

U

such that

k

i=1

a

i

x

i

+

m

j=1

b

j

y

j

= 0

and all scalars a

i

, b

j

are non-zero. Then

k

i=1

a

i

x

i

= −

m

j=1

b

j

y

j

∈ S ∩ U.

Since B

S

and B

U

both are independent, all x

i

and y

j

belong to X = B

S

∩B

U

which contradicts the fact that X is independent.

By Proposition 1.3, the subset B

S

∪ B

U

can be extended to a base of

V . This base is as required.

In the case when S ∩ U = 0, the proof is similar. We leave it as an

exercise for the reader.

For any two subsets X, Y ⊂ V we deﬁne

X +Y := ¦ x +y : x ∈ X, y ∈ Y ¦.

If X, Y are linear subspaces then the sum X +Y coincides with the linear

subspace ¸X, Y ).

Let S be a linear subspace of V . The associated quotient vector space

V/S is formed by all subsets x + S (we have x + S = y + S if x − y ∈ S)

and the vector space operations on V/S are deﬁned as follows

(x +S) + (y +S) := (x +y) +S, a(x +S) := ax +S.

A linear subspace U is called a complement of S if

S ∩ U = 0 and S +U = V ;

in this case, U is isomorphic to V/S (the mapping x → x + S is a linear

isomorphism). Therefore, all complements of S have the same dimension

called the codimension of S and denoted by codimS.

Proposition 1.5. Let S and U be linear subspaces of V . If these subspaces

both are ﬁnite-dimensional then

dim(S +U) = dimS + dimU −dim(S ∩ U).

In the case when at least one of these subspaces is inﬁnite-dimensional, we

have

dim(S +U) = max¦ dimS, dimU ¦.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

14 Grassmannians of Classical Buildings

This statement is a simple consequence of Proposition 1.4 and the fol-

lowing remark.

Remark 1.1. Recall that the cardinality of a set X is denoted by [X[. The

sum of two cardinalities α and β is the cardinality [X ∪ Y [, where X and

Y are disjoint sets of cardinalities α and β (respectively). If α ≤ β and β

is inﬁnite then α +β = β.

For every linear subspace S ⊂ V we have

dimS + codimS = dimV ;

in the inﬁnite-dimensional case, this formula means that

max¦ dimS, codimS ¦ = dimV.

Denote by ((V ) the set of all linear subspaces of V . The Grassmannians of

V can be deﬁned as the orbits of the action of the group GL(V ) on ((V );

in what follows we do not consider two trivial Grassmannians formed by 0

and V , respectively. If dimV = n is ﬁnite then for every k ∈ ¦1, . . . , n−1¦

we write (

k

(V ) for the Grassmannian consisting of all k-dimensional linear

subspaces of V . In the case when dimV = α is inﬁnite, there are the

following three types of Grassmannians:

• β < α and (

β

(V ) consists of all linear subspaces of dimension β (the

codimension of these linear subspaces is α),

• β < α and (

β

(V ) consists of all linear subspaces of codimension β (the

dimension of these linear subspaces is α),

• (

α

(V ) = (

α

(V ) consists of all linear subspaces whose dimension and

codimension is α.

1.1.3 Dual vector space

Let V be a left vector space over a division ring R. Linear mappings of

V to R are called linear functionals. We deﬁne the sum v + w of linear

functionals v and w as follows

(v +w)(x) := v(x) +w(x) ∀ x ∈ V ;

and for any linear functional v and any scalar a ∈ R the linear functional

va is deﬁned by the formula

(va)(x) := v(x)a ∀ x ∈ V.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 15

The set of all linear functionals together with the operations deﬁned above

is a right vector space other R. The associated left vector space over R

∗

will be denoted by V

∗

and called the dual vector space.

Let B = ¦x

i

¦

i∈I

([I[ = dimV ) be a base of V . There is a one-to-

one correspondence between linear functionals and mappings of the base B

to R, since every linear functional can be uniquely reconstructed from its

restriction to B. Denote by B

∗

the subset formed by the linear functionals

x

∗

i

, i ∈ I, satisfying

x

∗

i

(x

j

) = δ

ij

(δ

ij

is Kroneker symbol). This subset is independent.

If dimV = n < ∞ then for every x

∗

∈ V

∗

we have

x

∗

=

n

i=1

x

∗

(x

i

)x

∗

i

.

Thus B

∗

is a base of V

∗

and dimV = dimV

∗

. The base B

∗

will be called

dual to B.

If V is inﬁnite-dimensional then B

∗

is not a base of V

∗

. Indeed, a

linear functional cannot be presented as a linear combination of elements

from B

∗

if it takes non-zero values on an inﬁnite subset of B. In this case,

dimV < dimV

∗

(see Section II.3 in [Baer (1952)] for the details).

Lemma 1.1. The kernel of every non-zero linear functional is a linear

subspace of codimension 1.

Proof. Let v ∈ V

∗

¸ ¦0¦. If the codimension of Ker v is greater than

1 then consider S ∈ (

2

(V ) contained in a complement of Ker v. There

exist linearly independent vectors x, y ∈ S satisfying v(x) = v(y). Then

v(x −y) = 0 which contradicts the fact that S intersects Ker v precisely in

0.

Exercise 1.4. Show that

codim(S ∩ U) ≤ codimU + 1

for every linear subspace U ⊂ V and every linear subspace S ⊂ V of

codimension 1. As a consequence establish that

codim(S

1

∩ ∩ S

k

) ≤ k

for any linear subspaces S

1

, . . . , S

k

⊂ V of codimension 1.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

16 Grassmannians of Classical Buildings

For every subset X ⊂ V the linear subspace

X

0

:= ¦ v ∈ V

∗

: v(x) = 0 ∀ x ∈ X ¦

is known as the annihilator of X. Similarly, for every Y ⊂ V

∗

the linear

subspace

Y

0

:= ¦ y ∈ V : w(y) = 0 ∀ w ∈ Y ¦

is called the annihilator of Y . If S and U are linear subspaces of V or linear

subspaces of V

∗

then S ⊂ U implies that U

0

⊂ S

0

.

Proposition 1.6. If S ⊂ V is a linear subspace of ﬁnite codimension then

dimS

0

= codimS and S

00

= S. Similarly, if U is a ﬁnite-dimensional

linear subspace of V

∗

then codimU

0

= dimU and U

00

= U.

Proof. Suppose that codimS = k < ∞. Consider a base B = ¦x

i

¦

i∈I

of V such that S is spanned by a subset of B. Let B

∗

be the subset of

V

∗

which consists of all linear functionals x

∗

i

, i ∈ I, satisfying x

∗

i

(x

j

) = δ

ij

.

There are precisely k distinct vectors from B which do not belong to S,

denote them by x

i

1

, . . . , x

i

k

. The restriction of v ∈ S

0

to B can take non-

zero values only on these vectors. Therefore, x

∗

i

1

, . . . , x

∗

i

k

form a base of S

0

and dimS

0

= k. Since S coincides with the intersection of the kernels of

x

∗

i

1

, . . . , x

∗

i

k

, we have S

00

= S.

Let ¦v

1

, . . . , v

k

¦ be a base of U. Then U

0

is the intersection of the

kernels of v

1

, . . . , v

k

and, by the second part of Exercise 1.4, its codimension

is equal to m ≤ k. Since dimU

00

= m (it was established above), the

inclusion U ⊂ U

00

implies that U

00

= U and m = k.

Denote by (

ﬁn

(V ) and (

ﬁn

(V ) the sets of all proper linear subspaces of

V with ﬁnite dimension and ﬁnite codimension, respectively. If V is ﬁnite-

dimensional then (

ﬁn

(V ) and (

ﬁn

(V ) both coincide with ((V ). Proposition

1.6 gives the following.

Corollary 1.1. The annihilator mapping S → S

0

is a bijection of (

ﬁn

(V )

to (

ﬁn

(V

∗

). This bijection reverses inclusions:

S ⊂ U ⇐⇒ U

0

⊂ S

0

.

If dimV = n < ∞ then the annihilator mapping transfers (

k

(V ) to

(

n−k

(V

∗

) for every k ∈ ¦1, . . . , n −1¦. In the inﬁnite-dimensional case, it

sends (

k

(V ) to (

k

(V

∗

) for every natural k.

Every vector x ∈ V deﬁnes the linear functional v → v(x) of V

∗

(an

element of V

∗∗

). If V is ﬁnite-dimensional then this correspondence is a

linear isomorphism between V and V

∗∗

. In the ﬁnite-dimensional case, we

will identify the second dual vector space V

∗∗

with V .

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 17

1.2 Projective spaces

In the ﬁrst subsection we introduce the point-line geometry language which

will be exploited throughout the book. Projective spaces over division rings

will be considered in the second subsection.

1.2.1 Linear and partial linear spaces

Let P be a non-empty set and L be a set consisting of proper subsets of P.

Elements of P and L will be called points and lines, respectively. Two or

more points are said to be collinear if there is a line containing them. We

say that the pair Π = (P, L) is a partial linear space if the following axioms

hold:

(1) each line contains at least two points and for every point there is a line

containing it;

(2) for any two distinct collinear points p and q there is precisely one line

containing them, this line will be denoted by p q.

A linear space is a partial linear space, where any two points are collinear.

In what follows we will always suppose that a partial linear space contains

more than one line.

We say that partial linear spaces Π = (P, L) and Π

= (P

, L

) are

isomorphic if there exists a bijection f : P → P

such that f(L) = L

; this

bijection is called a collineaton of Π to Π

. A bijection of P to P

is said

to be a semicollineaton of Π to Π

**if it maps lines to subsets of lines. An
**

injection of P to P

**sending lines to subsets of lines is called an embedding
**

of Π in Π

**if distinct lines go to subsets of distinct lines.
**

Let Π = (P, L) be a partial linear space. For every subset X ⊂ P we

deﬁne

L

X

:= ¦ L ∩ X : L ∈ L, [L ∩ X[ ≥ 2 ¦;

the pair (X, L

X

) is called the restriction of the partial linear space Π to the

subset X.

The collinearity graph of Π is the graph whose vertex set is P and whose

edges are pairs of distinct collinear points. Our partial linear space is called

connected if the collinearity graph is connected. In a connected graph the

distance d(v, w) between two vertices v and w is the smallest number i such

that there is a path of length i connecting v and w; every path between v

and w consisting of d(v, w) edges is said to be a geodesic. If Π is connected

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

18 Grassmannians of Classical Buildings

then we deﬁne the distance between two points of Π as the distance between

the corresponding vertices of the collinearity graph.

We say that S ⊂ P is a subspace of Π if for any two distinct collinear

points p, q ∈ S the line p q is contained in S. By this deﬁnition, every

set of mutually non-collinear points is a subspace. A subspace is said to

be singular if any two points of this subspace are collinear (the empty set

and one-point subspaces are singular). The intersection of any collection of

subspaces is a subspace; moreover, it is a singular subspace if all subspaces

from the collection are singular.

Exercise 1.5. Show that maximal singular subspaces exist and every sin-

gular subspace is contained in a certain maximal singular subspace. Hint:

use Zorn lemma.

The minimal subspace containing a subset X ⊂ P (the intersection of

all subspaces containing X) will be called the subspace spanned by X and

denoted by ¸X). In the general case, the subspace spanned by a clique

of the collinearity graph (a subset consisting of mutually collinear points)

does not need to be singular. We say that a subset X ⊂ P is independent

if the subspace ¸X) is not spanned by a proper subspace of X.

Let S be a subspace of Π (possible S = P). An independent subset

X ⊂ S is said to be a base of S if ¸X) = S. We deﬁne the dimension of

S as the smallest cardinality α such that S has a base of cardinality α + 1

(if the cardinality α is inﬁnite then α + 1 = α). The dimensions of the

empty set and one-point subspaces are equal to −1 and 0 (respectively),

lines are 1-dimensional subspaces. Two-dimensional linear spaces and two-

dimensional singular subspaces of partial linear spaces are called planes.

A triple of mutually collinear points is said to be a triangle if these

points are not collinear (in other words, the points span a plane).

A proper subspace of a partial linear space is called a hyperplane if it

has a non-empty intersection with every line (it is clear that a hyperplane

contains a line or intersects it precisely in a point).

Recall that in a projective plane each line contains at least three points

and any two distinct lines have a non-empty intersection (by the second

axiom of partial linear spaces, this intersection is a point). A projective

space is a linear space where every line contains at least three points and

the following axiom holds: if a, b, c, d are distinct points and the lines ab

and cd have a non-empty intersection then the lines ac and bd have a non-

empty intersection. Projective spaces can be also deﬁned as linear spaces

where all planes are projective.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 19

1.2.2 Projective spaces over division rings

Let V be a left vector space over a division ring and dimV ≥ 3 (the

dimension of V is not assumed to be ﬁnite). For every S ∈ (

2

(V ) the

subset consisting of all 1-dimensional linear subspaces of S, in other words,

(

1

(S) is called a line of (

1

(V ). Denote by L

1

(V ) the set of all such lines

and consider the pair

Π

V

:= ((

1

(V ), L

1

(V )).

For any distinct P

1

, P

2

∈ (

1

(V ) there is the unique line (

1

(P

1

+ P

2

) con-

taining them. Every line (

1

(S) contains at least three points (if vectors x, y

form a base of S then ¸x), ¸y), ¸x+y) are three distinct points on the line).

Thus Π

V

is a linear space. In the case when dimV = 3, any two lines have

a non-empty intersection and Π

V

is a projective plane.

If P

1

, P

2

, P

3

, P

4

are distinct points of Π

V

and P

1

+P

2

has a non-zero in-

tersection with P

3

+P

4

then all P

i

are contained in a certain 3-dimensional

linear subspace; in particular, P

1

+ P

3

and P

2

+ P

4

have a non-zero inter-

section. Therefore, Π

V

is a projective space.

Exercise 1.6. Show that o ⊂ (

1

(V ) is a subspace of Π

V

if and only if

there exists a linear subspace S ⊂ V such that o = (

1

(S).

Let A be a subset of (

1

(V ). For every P ∈ A we choose a non-zero

vector x

P

∈ P and consider the set X := ¦x

P

¦

P∈X

. It follows from Exercise

1.6 that A is an independent subset of Π

V

if and only if X is an independent

subset of V . Hence A is a base of the subspace (

1

(S), S ∈ ((V ) if and

only if X is a base of the linear subspace S. Thus the dimension of (

1

(S)

is equal to dimS − 1; in particular, Π

V

is a projective space of dimension

dimV −1.

Theorem 1.3. If the dimension of a projective space is not less than 3

(the dimension can be inﬁnite) then this projective space is isomorphic to

the projective space associated with a certain vector space.

Proof. See Chapter VII in [Baer (1952)].

Remark 1.2. For projective planes this fails. All projective spaces over

division rings satisfy Desargues’ axiom; but there exist non-Desarguesian

projective planes.

Denote by P

∗

(V ) the Grassmannian consisting of all linear subspaces

of codimension 1 in V (in other words, P

∗

(V ) coincides with (

n−1

(V ) or

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

20 Grassmannians of Classical Buildings

(

1

(V ) if V is n-dimensional or inﬁnite-dimensional, respectively). A subset

A ⊂ P

∗

(V ) is said to be a line if there is a linear subspace S ⊂ V of

codimension 2 such that A consists of all elements of P

∗

(V ) containing S.

Let L

∗

(V ) be the set of all such lines. An easy veriﬁcation shows that

Π

∗

V

:= (P

∗

(V ), L

∗

(V ))

is a linear space; moreover, it is isomorphic to Π

V

∗ by the annihilator

mapping (Corollary 1.1). The projective space Π

∗

V

is called dual to Π

V

.

Exercise 1.7. Show that for every ﬁnite-dimensional subspace o of Π

∗

V

there exists S ∈ (

ﬁn

(V ) such that o consists of all elements of P

∗

(V )

containing S. Hint: use Corollary 1.1.

1.3 Semilinear mappings

Throughout the section we suppose that V and V

**are left vector spaces
**

over division rings R and R

, respectively.

1.3.1 Deﬁnitions

We say that a mapping l : V → V

is semilinear if it is additive:

l(x +y) = l(x) +l(y) ∀ x, y ∈ V,

and there exists a homomorphism σ : R → R

such that

l(ax) = σ(a)l(x) (1.1)

for all a ∈ R and x ∈ V . In the case when R = R

and σ is identity, we

get a linear mapping. As for linear mappings, all vectors x ∈ V satisfying

l(x) = 0 form a linear subspace called the kernel of l and denoted by Ker l.

If l is non-zero (Ker l ,= V ) then there exists the unique homomorphism

σ : R → R

**satisfying (1.1) and the mapping l is said to be σ-linear.
**

Every semilinear mapping between real vector spaces is linear. In the

general case, semilinear mappings form a more wide class.

A semilinear mapping of V to V

**is called a semilinear isomorphism if it
**

is bijective and the associated homomorphism of R to R

is an isomorphism.

If l : V → V

**is a σ-linear isomorphism then the inverse mapping l
**

−1

is a σ

−1

-linear isomorphism of V

**to V . Semilinear isomorphisms map
**

independent subsets to independent subsets and bases to bases. Semilinear

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 21

isomorphisms of V to V

exist if and only if dimV = dimV

and R is

isomorphic to R

.

Exercise 1.8. Show that every semilinear bijection sending independent

subsets to independent subsets is a semilinear isomorphism.

The following example shows that there exist semilinear bijections over

non-surjective homomorphisms of division rings.

Example 1.3. The complexiﬁcation mapping of R

2n

to C

n

:

(x

1

, y

1

, . . . , x

n

, y

n

) → (x

1

+y

1

i, . . . , x

n

+y

n

i)

is a semilinear bijection; the associated homomorphism is the natural em-

bedding of R in C.

The group of all semilinear automorphisms of V (semilinear isomor-

phisms of V to itself) is denoted by ΓL(V ). It contains GL(V ) as a sub-

group. Consider the homomorphism of ΓL(V ) to Aut(R) which sends ev-

ery σ-linear automorphism of V to σ. The kernel of this homomorphism is

GL(V ). Hence GL(V ) is a normal subgroup of ΓL(V ) and the correspond-

ing quotient group is isomorphic to Aut(R).

Example 1.4. The homothetic transformation x → ax, x ∈ V , is a semilin-

ear automorphism of V (the associated automorphism of R is r → ara

−1

).

This transformation is linear only in the case when the scalar a belongs to

the center of the division ring R. The group of all homothetic transforma-

tions H(V ) is isomorphic to the multiplicative group of the division ring;

moreover, it is a normal subgroup of ΓL(V ).

1.3.2 Mappings of Grassmannians induced by semilinear

mappings

Every semilinear mapping l : V → V

**induces the mapping
**

(l)

1

: (

1

(V ) ¸ (

1

(Ker l) → (

1

(V

)

¸x) → ¸l(x)), x ,∈ Ker l.

In the case when l is a semilinear isomorphism, this mapping is a

collineation of Π

V

to Π

V

.

If l : V → V

is a σ-linear mapping then for every non-zero scalar a ∈ R

the mapping al is σ

-semilinear with

σ

(r) = aσ(r)a

−1

∀ r ∈ R,

and (al)

1

coincides with (l)

1

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

22 Grassmannians of Classical Buildings

Proposition 1.7. Let l : V → V

**be a semilinear mapping such that l(V )
**

contains two linearly independent vectors. If s : V → V

is a semilinear

mapping satisfying (l)

1

= (s)

1

then s = al for a certain scalar a ∈ R

.

This statement is a direct consequence of the following lemma.

Lemma 1.2. Let l : V → V

and s : V → V

**be additive mappings such
**

that

s(x) ∈ ¸l(x)) ∀ x ∈ V.

If l(V ) contains two linearly independent vectors then s = al for a certain

scalar a ∈ R

.

Proof. For every vector x ∈ V there exists a scalar a

x

∈ R

(possible

zero) such that

s(x) = a

x

l(x).

Suppose that vectors l(x) and l(y) both are non-zero. Then

a

x

l(x) +a

y

l(y) = s(x +y) = a

x+y

(l(x) +l(y))

and a

x

= a

y

= a

x+y

if l(x), l(y) are linearly independent. If l(y) is a

scalar multiple of l(x) then we take any vector z ∈ V such that l(x) and

l(z) are linearly independent (by our hypothesis, this is possible) and get

a

x

= a

z

= a

y

. Therefore, for all vectors x ∈ V satisfying l(x) ,= 0 the scalar

a

x

is a constant. The equality l(x) = 0 implies that s(x) = 0 and we have

s(x) = al(x) for every scalar a ∈ R

.

Proposition 1.8 ([Zick (1983)]). Let l : V → V

be an additive mapping

which satisﬁes

l(¸x)) ⊂ ¸l(x)) ∀ x ∈ V.

If l(V ) contains two linearly independent vectors then l is semilinear.

Proof. For a non-zero scalar a ∈ R consider the mapping x → l(ax).

Lemma 1.2 implies the existence of a scalar σ(a) ∈ R

such that

l(ax) = σ(a)l(x) ∀ x ∈ V.

We set σ(0) := 0 and get a mapping σ : R → R

**. For any scalars a, b ∈ R
**

and any vector x ∈ V we have

σ(a +b)l(x) = l(ax +bx) = l(ax) +l(bx) = (σ(a) +σ(b))l(x),

σ(ab)l(x) = l(abx) = σ(a)σ(b)l(x)

which means that σ is a homomorphism.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 23

It was noted above that for every semilinear isomorphism l : V → V

the mapping (l)

1

is a collineation of Π

V

to Π

V

. Conversely, we have the

following.

Proposition 1.9. Let l : V → V

**be a semilinear mapping such that (l)
**

1

is a collineation of Π

V

to Π

V

. Then l is a semilinear isomorphism of V

to V

.

Proof. Let S ∈ (

2

(V ) and ¦x, y¦ be a base of S. Then a → ¸x + ay) is

a bijection of the division ring R to (

1

(S) ¸ ¦¸y)¦. The mapping (l)

1

sends

¸x +ay) to

¸l(x) +σ(a)l(y)),

where σ : R → R

**is the homomorphism associated with l. Since the
**

restriction of (l)

1

to the line (

1

(S) is a bijection on the line (

1

(l(S)), the

homomorphism σ is surjective; hence it is an isomorphism of R to R

.

The latter implies that l(V ) is a linear subspace of V

; then l(V ) = V

(otherwise, (l)

1

is not surjective). Since Ker l = 0, the mapping l is a

semilinear isomorphism.

Every semilinear isomorphism of V to V

**induces bijections between the
**

Grassmannians of the same indices; in particular, it deﬁnes a collineation

of Π

∗

V

to Π

∗

V

.

Proposition 1.10. Suppose that ( and (

**are Grassmannians of V and
**

V

**(respectively) with the same index. Let
**

¯

l : ( → (

and ¯ s : ( → (

be the

bijections induced by semilinear isomorphisms l : V → V

and s : V → V

,

respectively. If

¯

l = ¯ s then s = al for a certain scalar a ∈ R

.

Proof. Let P ∈ (

1

(V ). Consider the set A consisting of all elements of

( containing P. Then

P =

S∈X

S

and

l(P) =

S∈

¯

l(X)

S, s(P) =

S∈¯ s(X)

S.

Since

¯

l = ¯ s, we have l(P) = s(P) for every P ∈ (

1

(V ). In other words,

(l)

1

= (s)

1

and Proposition 1.7 gives the claim.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

24 Grassmannians of Classical Buildings

A semilinear injection l : V → V

**is called a semilinear embedding of
**

V in V

**if it maps independent subsets to independent subsets. Semilinear
**

embeddings of V in V

exist only in the case when dimV ≤ dimV

and R

isomorphic to a certain division subring of R

.

Example 1.5. The mapping

(x

1

, . . . , x

n

) → (x

1

+ 0i, . . . , x

n

+ 0i)

is a semilinear embedding of R

n

in C

n

. Similarly, the natural embedding

of Q

n

in R

n

is a semilinear embedding.

Now we consider the mappings of Grassmannians induced by semilinear

embeddings. We restrict ourselves to the case when dimV = n is ﬁnite (the

inﬁnite-dimensional case is similar).

Let l : V → V

**be a semilinear embedding. Then for any linear subspace
**

S ⊂ V we have

dim¸l(S)) = dimS,

and the mapping

(l)

k

: (

k

(V ) → (

k

(V

)

S → ¸l(S))

is deﬁned for every k ∈ ¦1, . . . , n −1¦.

Exercise 1.9. Show that the following assertions are fulﬁlled:

(1) (l)

k

is injective for every k ∈ ¦1, . . . , n −1¦,

(2) (l)

1

is an embedding of Π

V

in Π

V

,

(3) (l)

n−1

is an embedding of Π

∗

V

in Π

∗

V

if dimV

= n,

(4) if s : V → V

**is a semilinear embedding such that (l)
**

k

= (s)

k

for a

certain k ∈ ¦1, . . . , n −1¦ then s is a scalar multiple of l.

Proposition 1.11. Let l : V → V

**be a semilinear embedding such that
**

(l)

k

is bijective for a certain k ∈ ¦1, . . . , n − 1¦. Then l is a semilinear

isomorphism.

Proof. In the case when k = 1, our bijection maps any triple of collinear

points of Π

V

to collinear points of Π

V

and any triple of non-collinear

points to non-collinear points. Hence it is a collineation of Π

V

to Π

V

. By

Proposition 1.9, l is a semilinear isomorphism.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 25

Suppose that k > 1. Let S

∈ (

k−1

(V

). We choose U

i

∈ (

k

(V

),

i = 1, 2, such that

S

= U

1

∩ U

2

.

Since (l)

k

is bijective, there exist U

i

∈ (

k

(V ), i = 1, 2, satisfying

¸l(U

i

)) = U

i

.

Then U

1

∩ U

2

belongs to (

k−1

(V ) (we leave the details for the reader) and

¸l(U

1

∩ U

2

)) = S

.

Thus (l)

k−1

is surjective, and by the statement (1) of Exercise 1.9, it is

bijective. Step by step, we establish that (l)

1

is bijective.

1.3.3 Contragradient

Now suppose that V and V

**have the same ﬁnite dimension n. Recall that
**

in this case V = V

∗∗

and V

= V

∗∗

. For every x ∈ V and x

∗

∈ V

∗

we will

write x

∗

x instead of x

∗

(x).

Let u : V → V

**be a σ-linear isomorphism. The adjoint mapping
**

u

∗

: V

∗

→ V

∗

is deﬁned by the condition

u

∗

(x

∗

) x = σ

−1

(x

∗

u(x)) ∀ x ∈ V, x

∗

∈ V

∗

.

This is a (σ

−1

)-linear isomorphism and u

∗∗

= u. The inverse mapping

ˇ u := (u

∗

)

−1

: V

∗

→ V

∗

is known as the contragradient of u. Since

(u

∗

)

−1

= (u

−1

)

∗

,

the contragradient of the contragradient coincides with u. An easy veriﬁ-

cation shows that

ˇ u(x

∗

) u(x) = σ(x

∗

x) ∀ x ∈ V, x

∗

∈ V

∗

.

Therefore, ˇ u transfers the annihilator of a linear subspace S ⊂ V to the

annihilator of u(S) and we get

u(U

0

)

0

= ˇ u(U)

for every linear subspace U ⊂ V

∗

. So, the mapping

(

k

(V

∗

) → (

k

(V

∗

)

U → u(U

0

)

0

coincides with (ˇ u)

k

for every k ∈ ¦1, . . . , n −1¦.

Exercise 1.10. Show that the contragradient mapping u → ˇ u is an iso-

morphism of ΓL(V ) to ΓL(V

∗

) transferring GL(V ) to GL(V

∗

).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

26 Grassmannians of Classical Buildings

1.4 Fundamental Theorem of Projective Geometry

1.4.1 Main theorem and corollaries

Let V and V

be left vector spaces over division rings R and R

, respectively.

The dimensions of V and V

**are assumed to be not less than 3. Let also
**

l : V → V

**be a semilinear injection. The mapping
**

(l)

1

: (

1

(V ) → (

1

(V

)

P → ¸l(P))

does not need to be injective, see Example 1.3. For any P, P

1

, P

2

∈ (

1

(V )

satisfying P ⊂ P

1

+P

2

(a triple of collinear points of Π

V

) we have

¸l(P)) ⊂ ¸l(P

1

)) +¸l(P

2

)).

If ¸l(P

1

)) = ¸l(P

2

)) then the line (

1

(P

1

+ P

2

) goes to a point. In the case

when ¸l(P

1

)) and ¸l(P

2

)) are distinct, the restriction of (l)

1

to (

1

(P

1

+P

2

)

is a bijection on a subset of the line (

1

(¸l(P

1

+P

2

))).

The following generalized version of the Fundamental Theorem of Pro-

jective Geometry was proved in [Faure and Frolicher (1994)] and [Havlicek

(1994)], independently.

Theorem 1.4. Suppose that a mapping f : (

1

(V ) → (

1

(V

) satisﬁes the

following conditions: for any P, P

1

, P

2

∈ (

1

(V )

P ⊂ P

1

+P

2

=⇒ f(P) ⊂ f(P

1

) +f(P

2

)

and f((

1

(V )) is not contained in a line. Then f is induced by a semilinear

injection of V to V

.

Theorem 1.4 together with Proposition 1.9 give the classical version of

the Fundamental Theorem of Projective Geometry.

Corollary 1.2. Every collineation of Π

V

to Π

V

is induced by a semilinear

isomorphism of V to V

.

Remark 1.3. Denote by PΓL(V ) the group formed by all collineations of

the projective space Π

V

to itself. Consider the homomorphism of ΓL(V )

to PΓL(V ) which sends l to (l)

1

. The Fundamental Theorem of Projective

Geometry (Corollary 1.2) states that this homomorphism is surjective. By

Proposition 1.7, the kernel of the homomorphism is H(V ) (Example 1.4).

Therefore, PΓL(V ) is isomorphic to the quotient group ΓL(V )/H(V ). In

the case when V is ﬁnite-dimensional, the contragradient isomorphism of

ΓL(V ) to ΓL(V

∗

) (Exercise 1.10) sends H(V ) to H(V

∗

); hence it induces

an isomorphism between PΓL(V ) and PΓL(V

∗

).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 27

Corollary 1.3. If V and V

**have the same ﬁnite dimension then every
**

semicollineation of Π

V

to Π

V

is a collineation.

Proof. Every semicollineation of Π

V

to Π

V

satisﬁes the conditions of

Theorem 1.4; hence it is induced by a semilinear mapping l : V → V

. Our

semicollineation maps every triple of collinear points to collinear points and

it is suﬃcient to show that triples of non-collinear points go to non-collinear

points.

If this fails then there exist linearly independent vectors x, y, z ∈ V such

that l(x), l(y), l(z) are linearly dependent. Let B be a base of V containing

x, y, z. Since dimV = dimV

is ﬁnite, l(B) spans a proper subspace of V

which contradicts ¸l(B)) = ¸l(V )) = V

.

Corollary 1.3 will be generalized in Section 3.5.

Problem 1.1. Is there a semicollineation of Π

V

to Π

V

in the case when

dimV > dimV

?

It was noted in the previous section that semilinear isomorphisms of V

to V

induce collineations of Π

∗

V

to Π

∗

V

**. We can show that there exist no
**

other collineations of Π

∗

V

to Π

∗

V

**only in the case when our vector spaces
**

are ﬁnite-dimensional.

Corollary 1.4. If V and V

**are ﬁnite-dimensional then every collineation
**

of Π

∗

V

to Π

∗

V

is induced by a semilinear isomorphism of V to V

.

Proof. Let f be a collineation of Π

∗

V

to Π

∗

V

. It can be considered as a

collineation of Π

V

∗ to Π

V

∗. By the Fundamental Theorem of Projective

Geometry, the latter collineation is induced by a semilinear isomorphism

u : V

∗

→ V

∗

and

f(S) = u(S

0

)

0

for every linear subspace S ⊂ V of codimension 1. By Subsection 1.3.3, we

have f = (ˇ u)

1

in the ﬁnite-dimension case.

Exercise 1.11. Show that every semicollineation of Π

∗

V

to Π

∗

V

is a

collineation if dimV = dimV

< ∞.

Let X be a partially ordered set. The supremum of a subset Y ⊂ X

is the least element of X which is greater than or equal to all elements of

Y . Similarly, the inﬁmum of Y is the greatest element of X which is less

than or equal to every element of Y . The supremum and inﬁmum of Y

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

28 Grassmannians of Classical Buildings

are denoted by sup Y and inf Y , respectively. If sup Y (inf Y ) exists then

it is unique. The set X is called a lattice if every subset consisting of two

elements has the supremum and inﬁmum. The set ((V ) is partially ordered

by the inclusion relation. Since for all S, U ∈ ((V ) we have

sup¦S, U¦ = S +U and inf¦S, U¦ = S ∩ U,

it is a lattice. Every semilinear isomorphism of V to V

induces an order

preserving bijection of ((V ) to ((V

) (an isomorphism between the lat-

tices). The classical Fundamental Theorem of Projective Geometry can be

reformulated in the following form (see, for example, [Baer (1952)]).

Corollary 1.5. Let f : ((V ) → ((V

**) be an order preserving bijection: for
**

all S, U ∈ ((V )

S ⊂ U ⇐⇒ f(S) ⊂ f(U).

Then f is induced by a semilinear isomorphism of V to V

.

Proof. It is not diﬃcult to prove that f transfers (

1

(V ) and (

2

(V ) to

(

1

(V

) and (

2

(V

**), respectively. Hence the restriction of f to (
**

1

(V ) is

a collineation of Π

V

to Π

V

and there exists a semilinear isomorphism

l : V → V

**such that f(P) = l(P) for all P ∈ (
**

1

(V ). Then for every

S ∈ ((V )

f((

1

(S)) = (

1

(f(S))

coincides with

l((

1

(S)) = (

1

(l(S))

and we get f(S) = l(S).

1.4.2 Proof of Theorem 1.4

Let f : (

1

(V ) → (

1

(V

**) be a mapping which satisﬁes the conditions of
**

Theorem 1.4: for any P, P

1

, P

2

∈ (

1

(V )

P ⊂ P

1

+P

2

=⇒ f(P) ⊂ f(P

1

) +f(P

2

)

and f((

1

(V )) is not contained in a line. This means that the restriction of

f to every line is a constant or a bijection to a subset of a line.

Lemma 1.3. Let v

1

, v

2

, v

3

∈ V and w

1

, w

2

, w

3

∈ V

be vectors satisfying

the following conditions:

• f(¸v

i

)) = ¸w

i

) for i = 1, 2, 3,

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 29

• f(¸v

1

+v

2

)) = ¸w

1

+w

2

) and f(¸v

1

+v

3

)) = ¸w

1

+w

3

),

• w

1

, w

2

, w

3

are linearly independent.

Then

f(¸v

2

+v

3

)) = ¸w

2

+w

3

) and f(¸v

1

+v

2

+ v

3

)) = ¸w

1

+w

2

+w

3

).

Proof. It is clear that v

1

, v

2

, v

3

are linearly independent and ¸v

1

+v

2

+v

3

)

is the intersection of

¸v

1

+v

2

) +¸v

3

) and ¸v

1

+v

3

) +¸v

2

).

Then f(¸v

1

+v

2

+v

3

)) is contained in the intersection of

¸w

1

+w

2

) +¸w

3

) and ¸w

1

+ w

3

) +¸w

2

).

The latter intersection coincides with ¸w

1

+w

2

+w

3

) and we get the second

equality. Similarly,

¸v

2

+v

3

) = (¸v

2

) +¸v

3

)) ∩ (¸v

1

) +¸v

1

+v

2

+v

3

))

and f(¸v

2

+v

3

)) is contained

(¸w

2

) +¸w

3

)) ∩ (¸w

1

) +¸w

1

+w

2

+w

3

))

which gives the ﬁrst equality.

Lemma 1.4. Let P, Q ∈ (

1

(V ) and x ∈ P + Q with x ,∈ Q. Then there

exists a unique vector y ∈ Q such that P = ¸x +y).

Proof. Trivial.

By our hypothesis, there exist vectors x

1

, x

2

, x

3

∈ V such that

f(¸x

1

)), f(¸x

2

)), f(¸x

3

))

are not contained in a line. Our ﬁrst step is to establish the existence of

vectors y

1

, y

2

, y

3

∈ V

**satisfying the conditions
**

f(¸x

i

)) = ¸y

i

) and f(¸x

i

+x

j

)) = ¸y

i

+y

j

).

Proof. Let us take any non-zero vector y

1

∈ f(¸x

1

)). Since ¸x

1

) is con-

tained in ¸x

1

+x

2

) +¸x

2

),

y

1

∈ f(¸x

1

)) ⊂ f(¸x

1

+x

2

)) +f(¸x

2

)).

Lemma 1.4 implies the existence of y

2

∈ f(¸x

2

)) such that

f(¸x

1

+x

2

)) = ¸y

1

+y

2

).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

30 Grassmannians of Classical Buildings

If y

2

is zero then f(¸x

1

+x

2

)) coincides with f(¸x

1

)) and the restriction of

f to the line associated with the linear subspace

¸x

1

+x

2

) +¸x

1

) = ¸x

1

) +¸x

2

)

is constant; in this case, we get f(¸x

1

)) = f(¸x

2

)) which is impossible.

Thus y

2

,= 0. Similarly, we establish the existence of a non-zero vector

y

3

∈ f(¸x

3

)) satisfying

f(¸x

1

+x

3

)) = ¸y

1

+y

3

).

By Lemma 1.3, the vectors y

1

, y

2

, y

3

are as required.

Let x ∈ V ¸ ¦0¦. We choose i ∈ ¦1, 2, 3¦ such that f(¸x)) ,= f(¸x

i

)).

The restriction of f to the line associated with the linear subspace

¸x) +¸x

i

) = ¸x +x

i

) +¸x)

is injective and we apply Lemma 1.4 to

y

i

∈ f(¸x

i

)) ⊂ f(¸x

i

+x)) +f(¸x)).

So, there exists a vector l(x) ∈ f(¸x)) such that

f(¸x

i

+x)) = ¸y

i

+l(x)).

Since f(¸x

i

+ x)) ,= f(¸x

i

)), we have l(x) ,= 0. We set l(0) := 0 and assert

that l : V → V

is a semilinear mapping.

Show that the mapping l : V → V

**is well-deﬁned (the deﬁnition of the
**

vector l(x) does not depend on the choice of the number i ∈ ¦1, 2, 3¦).

Proof. Suppose that

f(¸x

1

)) ,= f(¸x)) ,= f(¸x

2

)).

There exists a non-zero vector l(x) ∈ f(¸x)) satisfying

f(¸x

1

+x)) = ¸y

1

+l(x)).

We need to show that

f(¸x

2

+x)) = ¸y

2

+l(x)).

If f(¸x)) is not contained in f(¸x

1

)) + f(¸x

2

)) then the required equality

follows from Lemma 1.3. In the case when

f(¸x)) ⊂ f(¸x

1

)) +f(¸x

2

)),

we consider successively the triples x

1

, x

3

, x and x

3

, x

2

, x.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 31

Our next step is the equality l(v +w) = l(v) +l(w) for all v, w ∈ V (the

mapping l is additive).

Proof. First consider the case when l(v) and l(w) are linearly inde-

pendent. We choose i ∈ ¦1, 2, 3¦ such that f(¸x

i

)) is not contained in

f(¸v)) +f(¸w)) (this is possible, since f(¸x

1

)), f(¸x

2

)), f(¸x

3

)) are not con-

tained in a line). We have

f(¸x

i

+v)) = ¸y

i

+l(v)) and f(¸x

i

+w)) = ¸y

i

+l(w)),

and Lemma 1.3 implies that

f(¸v +w)) = ¸l(v) +l(w)) and f(¸x

i

+v +w)) = ¸y

i

+l(v) +l(w)).

Thus

¸l(v +w)) = ¸l(v) +l(w)) and ¸y

i

+l(v +w)) = ¸y

i

+l(v) +l(w)).

By the ﬁrst equality, l(v + w) = a(l(v) + l(w)) for a certain scalar a ∈ R

;

the second equality guarantees that a = 1.

Now suppose that l(w) is a scalar multiple of l(v). Then

f(¸v)) = f(¸w)) = f(¸v +w)).

We take any vector z ∈ V such that l(w), l(z) are linearly independent.

Then l(w+z) = l(w)+l(z); moreover, l(v+w), l(z) are linearly independent

and

l(v +w +z) = l(v +w) +l(z). (1.2)

Since f(¸w)) ,= f(¸z)), the restriction of f to the line associated with the

linear subspace ¸w) +¸z) is injective and

f(¸v)) = f(¸w)) ,= f(¸w +z)).

This means that l(v), l(w + z) are linearly independent; hence

l(v +w +z) = l(v) +l(w +z) = l(v) +l(w) +l(z)

and (1.2) gives the claim.

For every non-zero vector x ∈ V and any non-zero scalar a ∈ R we have

¸l(ax)) = f(¸ax)) = f(¸x)) = ¸l(x)).

In other words, l(¸x)) ⊂ ¸l(x)) and the mapping l : V → V

satisﬁes the

conditions of Proposition 1.8. Therefore, l is semilinear and f = (l)

1

.

Remark 1.4. This proof was taken from the paper [Faure (2002)] dedicated

to A. Fr¨ olicher.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

32 Grassmannians of Classical Buildings

1.4.3 Fundamental Theorem for normed spaces

Let F be the ﬁeld of real or complex numbers. Let also N be a normed space

over F; this means that N is a vector space (over F) with a real-valued

function x → [[x[[ called a norm and satisfying the following conditions:

• [[x[[ ≥ 0 for all x ∈ N, and [[x[[ = 0 implies that x = 0,

• [[x +y[[ ≤ [[x[[ +[[y[[ for all x, y ∈ N,

• [[ax[[ = [a[ [[x[[ for every vector x ∈ N and every scalar a ∈ F.

Our normed space is, in a natural way, a matric space (hence it is also a

topological space); the metric is deﬁned by

d(x, y) := [[x −y[[ ∀ x, y ∈ N.

Denote by (

cl

(N) the set of all closed linear subspaces of N. Every ﬁnite-

dimensional linear subspace is closed; hence (

cl

(N) coincides with ((N) if

the dimension of N is ﬁnite.

The sum of two closed linear subspaces does not need to be closed. For

linear subspaces S, U ⊂ N we denote by S U the closure of S +U; since

the closure of a linear subspace is a linear subspace, we have SU ∈ (

cl

(N).

The set (

cl

(N) is partially ordered by the inclusion relation; moreover, for

all S, U ∈ (

cl

(V )

sup¦S, U¦ = S U and inf¦S, U¦ = S ∩ U.

So, (

cl

(N) is a lattice. Note that (

cl

(N) is not a sublattice of the lattice

((N) if N is inﬁnite-dimensional.

Example 1.6. The vector space F

n

is normed by

[[x[[ := ([x

1

[

2

+ +[x

n

[

2

)

1/2

, x = (x

1

, . . . , x

n

);

in particular, (F, [ [) is a 1-dimensional normed space.

Example 1.7. The vector space C

F

([0, 1]) consisting of all continuous func-

tions g : [0, 1] → F is normed by

[[g[[ := sup¦ [g(x)[ : x ∈ [0, 1] ¦.

Example 1.8. Let p ∈ N and l

p

(F) be the vector space formed by all

sequences x = ¦x

n

¦

n∈N

⊂ F satisfying

[[x[[

p

:=

_

∞

n=1

[x

n

[

p

_

1/p

< ∞.

Then (l

p

(F), [[ [[

p

) is a normed space.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 33

Let M be other normed vector space over F. We will investigate order

preserving bijections of (

cl

(N) to (

cl

(M) (isomorphisms between the lat-

tices). Recall that a bijection h between two topological spaces is called

a homeomorphism if h and h

−1

both are continuous. Clearly, every linear

homeomorphism of N to M induces an order preserving bijection of (

cl

(N)

to (

cl

(M).

Now suppose that our normed spaces are complex. If a semilinear

mapping l : N → M is continuous then the associated homomorphism

σ : C →C is continuous. By Exercise 1.3, σ is identity or the complex con-

jugate mapping. In what follows a semilinear mapping between complex

vector spaces will be called conjugate-linear if the associated homomor-

phism of C to itself is the complex conjugate mapping. Every conjugate-

linear homeomorphism of N to M induces an order preserving bijection of

(

cl

(N) to (

cl

(M).

Theorem 1.5 (G. W. Mackey, P. A. Fillmore, W. E. Longstaﬀ).

Let f : (

cl

(N) → (

cl

(M) be an order preserving bijection:

S ⊂ U ⇐⇒ f(S) ⊂ f(U)

for all S, U ∈ (

cl

(N). If N and M are real normed spaces then f is induced

by a linear homeomorphism of N to M. In the case when N and M are

inﬁnite-dimensional complex normed spaces, f is induced by a linear or

conjugate-linear homeomorphism of N to M.

Remark 1.5. If N and M are ﬁnite-dimensional complex normed spaces

then every semilinear isomorphism of N to M induces an order preserving

bijection of ((N) = (

cl

(N) to ((M) = (

cl

(M).

Theorem 1.5 was ﬁrst proved in [Mackey (1942)] for real normed spaces.

The complex version of this result was given in [Fillmore and Longstaﬀ

(1984)].

1.4.4 Proof of Theorem 1.5

We need some elementary facts concerning linear and conjugate-linear map-

pings. Let N and M be normed spaces over F = R, C. For a linear mapping

l : N → M (if M = F then l is a linear functional) the following conditions

are equivalent:

• l is continuous,

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

34 Grassmannians of Classical Buildings

• l transfers bounded subsets to bounded subsets (a subset X is bounded

if there exists a real non-negative number a such that [[x[[ ≤ a for all

x ∈ X).

In the complex case, the same holds for conjugate-linear mappings. In

[Rudin (1973)] (Section 1.32) this statement is proved for linear mappings;

for conjugate-linear mappings the proof is similar. If l : N → M is a

continuous linear or conjugate-linear mapping then the number

[[l[[ := sup¦ [[l(x)[[ : x ∈ N, [[x[[ ≤ 1 ¦

is ﬁnite and called the norm of l; it is clear that

[[l(x)[[ ≤ [[l[[ [[x[[ and [[al[[ = [a[ [[l[[

for all x ∈ N and a ∈ F.

Now we start to prove the theorem. Let f : (

cl

(N) → (

cl

(M) be an

order preserving bijection. Then f transfers (

1

(N) and (

2

(N) to (

1

(M)

and (

2

(M), respectively. This means that the restriction of f to (

1

(N) is

a collineation of Π

N

to Π

M

. By the Fundamental Theorem of Projective

Geometry, it is induced by a certain semilinear isomorphism l : N → M.

As in the proof of Corollary 1.5, we establish that

f(S) = l(S) ∀ S ∈ (

cl

(N).

We need to show that l is a homeomorphism of N to M.

Real case. The mapping l is linear. Let v : M → R be a non-zero

continuous linear functional. Then Ker v is a closed linear subspace of

codimension 1 and

S := l

−1

(Ker v)

is a closed linear subspace of codimension 1 in N. For every closed lin-

ear subspace of codimension 1 there exists a continuous linear functional

whose kernel coincides with this linear subspace (a consequence of the Hahn-

Banach Theorem, see Section 3.5 in [Rudin (1973)]). Consider a continuous

linear functional w : N →R such that Ker w = S. We ﬁx z ∈ N satisfying

w(z) = 1. Every x ∈ N can be presented in the form x = y +w(x)z, where

y ∈ S. Then

v(l(x)) = v(l(y)) +v(w(x)l(z)) = w(x)v(l(z)) (1.3)

(since l(y) ∈ Ker v).

Let X be a bounded subset of N. Then w(X) is a bounded subset of R

and (1.3) guarantees that the same holds for v(l(X)). Since v is arbitrary

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 35

taken, the set v(l(X)) is bounded for every continuous linear functional

v : M → R, in other words, l(X) is weakly bounded. In a normed space

every weakly bounded subset is bounded (Section 3.18 in [Rudin (1973)]).

Thus l transfers bounded subsets to bounded subsets; hence it is continuous.

Similarly, we establish that l

−1

is continuous.

Complex case. The normed spaces are assumed to be inﬁnite-

dimensional. Let σ be the automorphism of C associated with l. In this

case, we have

v(l(x)) = σ(w(x))v(l(z))

instead of (1.3). The arguments given above work only in the case when σ

is identity or the complex conjugate mapping. Therefore, we need to show

that l is linear or conjugate-linear. This fact is a direct consequence of

Kakutani–Mackey’s result which will be proved below (Proposition 1.12).

Lemma 1.5. Let σ be an automorphism of C. If for every sequence of

complex numbers z

n

→ 0 the sequence ¦σ(z

n

)¦

n∈N

is bounded then σ is

continuous.

Proof. By the additivity of σ, it is suﬃcient to verify that σ is continuous

in 0. Indeed, if z

n

→ z then (z

n

−z) → 0 and

σ(z

n

−z) → 0 =⇒ σ(z

n

) → σ(z).

Suppose that σ is non-continuous in 0. Then there exist a sequence z

n

→ 0

and a real number a > 0 satisfying

[σ(z

n

)[ > a ∀ n ∈ N.

Clearly, ¦z

n

¦

n∈N

contains a subsequence ¦z

n

¦

n∈N

such that [z

n

[ < 1/n

2

for

every n ∈ N, in other words, nz

n

→ 0. Since the restriction of σ to N is

identity,

[σ(nz

n

)[ = n[σ(z

n

)[ > na ∀ n ∈ N

and the sequence ¦σ(nz

n

)¦

n∈N

is unbounded, a contradiction.

Lemma 1.6. If N is inﬁnite-dimensional then there exist an indepen-

dent subset ¦x

n

¦

n∈N

⊂ N and a sequence of continuous linear functionals

¦v

n

¦

n∈N

of N such that

v

i

(x

j

) = δ

ij

. (1.4)

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

36 Grassmannians of Classical Buildings

Proof. It is not diﬃcult to choose two linearly independent vectors

x

1

, x

2

∈ N and two continuous linear functionals v

1

, v

2

of N satisfying

(1.4). Suppose that (1.4) holds for linearly independent vectors x

1

, . . . , x

n

and continuous linear functionals v

1

, . . . , v

n

. Then there exists a continuous

linear functional v

n+1

such that

• all x

i

belong to Ker v

n+1

,

• v

n+1

(x

n+1

) = 1 for a certain vector x

n+1

which does not belong to

¸x

1

, . . . , x

n

)

(see Section 3.5 in [Rudin (1973)]). We deﬁne

x

n+1

:= x

n+1

−

n

i=1

v

i

(x

n+1

)x

i

.

Then v

n+1

(x

n+1

) = 1 and v

i

(x

n+1

) = 0 for all i ≤ n.

Lemma 1.7. If N is inﬁnite-dimensional then it contains an indepen-

dent subset ¦x

n

¦

n∈N

satisfying the following condition: for every bounded

sequence of scalars ¦a

n

¦

n∈N

there exists a continuous linear functional

v : N → F such that

v(x

n

) = a

n

∀ n ∈ N.

Proof. Let X = ¦x

n

¦

n∈N

and ¦v

n

¦

n∈N

be as in the previous lemma. We

can assume that

[[v

n

[[ = 1/2

n

∀ n ∈ N

(indeed, for every n ∈ N there exists a scalar b

n

such that [[b

n

v

n

[[ = 1/2

n

and we can take x

n

∈ ¸x

n

) satisfying b

n

v

n

(x

n

) = 1).

Let ¦a

n

¦

n∈N

be a bounded sequence of scalars and a = sup [a

i

[. For

every vector

x = v

1

(x)x

1

+ +v

n

(x)x

n

∈ ¸X)

we deﬁne

v(x) := v

1

(x)a

1

+ +v

n

(x)a

n

.

Then

[v(x)[ ≤ [a

1

[ [[v

1

[[ [[x[[ + +[a

n

[ [[v

n

[[ [[x[[

≤ a[[x[[(1/2 + + 1/2

n

) < a[[x[[.

So, v : ¸X) → F is a continuous linear functional; by the Hahn-Banach

Theorem, it can be extended to a continuous linear functional of N.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 37

Proposition 1.12 ([Kakutani and Mackey (1946)]). Suppose that N

and M are inﬁnite-dimensional complex normed spaces. Let l : N → M be a

semilinear isomorphism which sends closed linear subspaces of codimension

1 to closed linear subspaces. Then l is linear or conjugate-linear.

Proof. Let σ be the automorphism of C associated with l. Let also

¦x

n

¦

n∈N

be a subset of N with the property described in the previous

lemma. Show that for every sequence of complex numbers a

n

→ 0 the

sequence ¦σ(a

n

)¦

n∈N

is bounded.

If ¦σ(a

n

)¦

n∈N

is unbounded then ¦a

n

¦

n∈N

contains a subsequence

¦b

n

¦

n∈N

such that

[σ(b

n

)[ ≥ n[[l(x

n

)[[ ∀ n ∈ N. (1.5)

Let v : N → C be a continuous linear functional satisfying v(x

n

) = b

n

for

every n. We take z ∈ N such that v(z) = 1. Then x

n

= y

n

+ b

n

z, where

y

n

∈ Ker v. We have

l(x

n

)/σ(b

n

) = l(y

n

/b

n

) +l(z).

By (1.5),

l(x

n

)/σ(b

n

) → 0 and l(−y

n

/b

n

) → l(z).

Hence l(z) belongs to the closure of l(Ker v). Ker v is a closed linear sub-

space of codimension 1 and the linear subspace l(Ker v) is, by our hypoth-

esis, closed. Thus l(z) belongs to l(Ker v). Since l is bijective, we get

z ∈ Ker v which contradicts v(z) = 1.

By Lemma 1.5, σ is continuous. This means that it is identity or the

complex conjugate mapping.

1.5 Reﬂexive forms and polarities

1.5.1 Sesquilinear forms

Let V be a left vector space over a division ring R and σ : R → R be an

anti-automorphism (an isomorphism of R to the opposite division ring R

∗

).

We say that

Ω : V V → R

is a sesquilinear form over σ or simple a σ-form if

Ω(x +y, z) = Ω(x, z) + Ω(y, z), Ω(z, x +y) = Ω(z, x) + Ω(z, y),

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

38 Grassmannians of Classical Buildings

Ω(ax, by) = aΩ(x, y)σ(b)

for all vectors x, y, z ∈ V and all scalars a, b ∈ R; if σ is identity (this is

possible only in the commutative case) then Ω is a usual bilinear form.

For every y ∈ V the mapping x → Ω(x, y) is a linear functional of V

and we get a σ-linear mapping of V to V

∗

. Conversely, every semilinear

mapping l : V → V

∗

gives a sesquilinear form

(x, y) → l(y) x.

Therefore, there is a one-to-one correspondence between sesquilinear forms

on V and semilinear mappings of V to V

∗

.

Our form Ω is said to be non-degenerate if the associated semilinear

mapping of V to V

∗

is injective (if V is ﬁnite-dimensional then dimV =

dimV

∗

and it is a semilinear isomorphism). This condition is equivalent

to the fact that for every non-zero y ∈ V there exists x ∈ V such that

Ω(x, y) ,= 0.

1.5.2 Reﬂexive forms

Suppose that dimV = n is ﬁnite. A sesquilinear form Ω : V V → R is

called reﬂexive if

Ω(x, y) = 0 =⇒ Ω(y, x) = 0

for every x, y ∈ V . We give a few examples.

Example 1.9. A non-zero sesquilinear form Ω : V V → R is said to be

symmetric or skew-symmetric if

Ω(x, y) = Ω(y, x) ∀ x, y ∈ V

or

Ω(x, y) = −Ω(y, x) ∀ x, y ∈ V,

respectively. In each of these cases, the associated anti-automorphism is

identity (we leave the veriﬁcation to the reader); hence R is commutative.

A sesquilinear form Θ : V V → R is called alternating if

Θ(x, x) = 0 ∀ x ∈ V.

Every alternating form is skew-symmetric. Conversely, if the characteristic

of R is not equal to 2 then every skew-symmetric form on V is alternating.

In the case of characteristic 2, the classes of symmetric and skew-symmetric

forms are coincident. Non-degenerate alternating forms exist only on even-

dimensional vector spaces over ﬁelds.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 39

Example 1.10. Let σ be a non-identity anti-automorphism of R. We say

that a non-zero sesquilinear form Ω : V V → R is σ-Hermitian or skew

σ-Hermitian if

Ω(x, y) = σ(Ω(y, x)) ∀ x, y ∈ V

or

Ω(x, y) = −σ(Ω(y, x)) ∀ x, y ∈ V,

respectively. An easy veriﬁcation shows that Ω is a σ-form and σ

2

= 1

R

in

each of these cases.

It is clear that the forms considered above are reﬂexive. Also note that

for every reﬂexive σ-form Ω : V V → R and every non-zero scalar a ∈ R

the scalar multiple

(x, y) → Ω(x, y)a

is a reﬂexive σ

-form with

σ

(b) = a

−1

σ(b)a ∀ b ∈ R.

There is a complete description of all non-degenerate reﬂexive forms.

Theorem 1.6 (G. Birkhoﬀ, J. von Neumann). If Ω : V V → R is

a non-degenerate reﬂexive form then one of the following possibilities is

realized:

• R is commutative and Ω is symmetric or alternating,

• Ω is a scalar multiple of a Hermitian form.

Proof. See, for example, Section 1.6 in [Dieudonn´e (1971)] or Chapter 7

in [Taylor (1992)].

Let Ω : V V → R be a reﬂexive form. We say that two vectors x, y ∈ V

are orthogonal and write x ⊥ y if Ω(x, y) = 0. The orthogonality relation

is symmetric (by the reﬂexivity). If X and Y are subsets of V then X ⊥ Y

means that every x ∈ X is orthogonal to all y ∈ Y ; in this case, we say that

the subsets X and Y are orthogonal. The linear subspace consisting of all

vectors orthogonal to a subset X ⊂ V is called the orthogonal complement

of X and denoted by X

⊥

.

A non-zero vector x ∈ V satisfying Ω(x, x) = 0 is said to be isotropic.

A linear subspace S is called totally isotropic if the restriction of Ω to S is

total zero, in other words, S is contained in S

⊥

. The Grassmannian formed

by all k-dimensional totally isotropic subspaces will be denoted by (

k

(Ω).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

40 Grassmannians of Classical Buildings

From this moment we suppose that the form Ω is non-degenerated. Let

u : V → V

∗

be the semilinear isomorphism associated with Ω. Then

Ω(x, y) = u(y) x ∀ x, y ∈ V.

Since the orthogonal relation is symmetric, we have

S

⊥

= u(S)

0

= u

−1

(S

0

) (1.6)

for every linear subspace S ⊂ V ; in particular,

dimS

⊥

= codimS.

This implies that dimS

⊥⊥

= dimS; since S ⊂ S

⊥⊥

, we get

S

⊥⊥

= S.

If S is totally isotropic then S ⊂ S

⊥

and dimS ≤ codimS.

The mapping S → S

⊥

is a bijective transformation of ((V ) sending

(

k

(V ) to (

n−k

(V ). This transformation is order reversing:

S ⊂ U ⇐⇒ U

⊥

⊂ S

⊥

.

The square of the transformation is identity.

Remark 1.6. Taking S

0

= U in (1.6) we obtain u(U

0

)

0

= u

−1

(U) which

implies that

ˇ u(U) = u

−1

(U).

Therefore, (ˇ u)

k

= (u

−1

)

k

for all k. This means that ˇ u is a scalar multiple

of u

−1

. Since ˇ u = (u

∗

)

−1

, we have u

∗

= au for non-zero a ∈ R.

1.5.3 Polarities

Let V be as in the previous section. A bijection π : (

1

(V ) → (

n−1

(V ) is

called a polarity if

P ⊂ π(Q) ⇐⇒ Q ⊂ π(P) ∀ P, Q ∈ (

1

(V ). (1.7)

Example 1.11. Let be ⊥ be the orthogonal relation associated with a

non-degenerate reﬂexive form. Then

S ⊂ U

⊥

⇐⇒ U ⊂ S

⊥

for any S, U ∈ ((V ) (since the orthogonal relation is symmetric). This

means that the restriction of the transformation S → S

⊥

to (

1

(V ) is a

polarity.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Linear Algebra and Projective Geometry 41

So, every non-degenerate reﬂexive form deﬁnes a polarity. Conversely,

we have the following.

Proposition 1.13. For every polarity π : (

1

(V ) → (

n−1

(V ) there exists

a non-degenerate reﬂexive form Ω such that the restriction of the transfor-

mation S → S

⊥

(where ⊥ is the orthogonal relation associated with Ω) to

(

1

(V ) coincides with π.

Proof. Show that π is a collineation of Π

V

to Π

∗

V

. Let P

1

, P

2

be distinct

elements of (

1

(V ) and A be the set formed by all (n−1)-dimensional linear

subspaces containing P

1

+P

2

. By (1.7), our polarity transfers

(

1

(π(P

1

) ∩ π(P

2

)) (1.8)

to the set A. Clearly, P ∈ (

1

(V ) belongs to the line (

1

(P

1

+P

2

) if and only

if it is contained in every element of A. By (1.7), the latter is equivalent

to the fact that π(P) contains every element of the set (1.8). Thus

P ⊂ P

1

+ P

2

⇐⇒ π(P

1

) ∩ π(P

2

) ⊂ π(P)

and π is a collineation.

The Fundamental Theorem of Projective Geometry guarantees the exis-

tence of a semilinear isomorphism u : V → V

∗

such that π(P) = u(P)

0

for

every P ∈ (

1

(V ). It follows from (1.7) that the sesquilinear form deﬁned

by u is reﬂexive.

This page intentionally left blank This page intentionally left blank

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Chapter 2

Buildings and Grassmannians

This chapter is a survey of basic facts concerning buildings and related

topics. Buildings of classical groups will be considered as examples. We

introduce the concepts of building Grassmannians and the associated Grass-

mann spaces. Grassmannians of ﬁnite-dimensional vector spaces and Grass-

mannians formed by totally isotropic subspaces of non-degenerate reﬂexive

forms are special cases of this general construction.

In the second part of the chapter, we prove Abramenko–Van

Maldeghem’s result on apartments preserving mappings of the chamber sets

of buildings (maximal simplices of a building are called chambers). Also

we show that the same method cannot be applied to apartments preserving

mappings of building Grassmannians.

2.1 Simplicial complexes

2.1.1 Deﬁnition and examples

Let X be a non-empty set and ∆ be a set consisting of ﬁnite subsets of X

(we do not require that X is ﬁnite). Suppose that the following conditions

hold:

(1) every one-element subset belongs to ∆,

(2) if A ∈ ∆ then every subset of A belongs to ∆.

Then ∆ is called a simplicial complex; elements of X and ∆ are said to be

vertices and simplices, respectively.

Let Y be a subset of X and Σ be a simplicial complex whose vertex set

is Y . If Σ ⊂ ∆ then Σ is called a subcomplex of ∆.

43

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

44 Grassmannians of Classical Buildings

Let ∆ and ∆

be simplicial complexes whose vertex sets are X and X

,

respectively. A mapping f : X → X

is said to be a morphism of ∆ to ∆

if

f(∆) ⊂ ∆

**(the image of every simplex is a simplex). A bijective morphism
**

f : X → X

is an isomorphism if f(∆) = ∆

. Isomorphisms of a simplicial

complex to itself are called automorphisms.

Suppose that X is a set with an incidence relation ∗ (this is a symmetric

and reﬂexive binary relation). A subset consisting of pairwise incident

elements is said to be a ﬂag. The ﬂag complex associated with the relation

∗ is the simplicial complex whose vertex set is X and whose simplices are

ﬁnite ﬂags.

Now we give a few examples.

Example 2.1. Consider the set of all proper subsets of ¦1, . . . , n+1¦ with

the natural incidence relation (two subsets are incident if one of them is

contained in the other). The associated ﬂag complex is denoted by A

n

.

Example 2.2. Let us consider the 2n-element set

J := ¦1, . . . , n, −1, . . . , −n¦.

A subset X ⊂ J is said to be singular if

j ∈ X =⇒ −j ,∈ X.

We write ¸ for the set of all singular subsets. Every maximal singular

subset consists of n elements and for every i ∈ ¦1, . . . , n¦ it contains i

or −i. For every k ∈ ¦1, . . . , n¦ we denote by ¸

k

the set of all singular

subset consisting of k elements (every one-element subset is singular and

we identify ¸

1

with J). The ﬂag complex associated with the natural

incidence relation on ¸ will be denoted by C

n

.

Now deﬁne

X

+

:= ¦1, . . . , n¦, X

−

:= ¦1, . . . , n −1, −n¦

and consider the sets

¸

+

:= ¦ X ∈ ¸

n

: n −[X

+

∩ X[ is even ¦,

¸

−

:= ¸

n

¸ ¸

+

= ¦ Y ∈ ¸

n

: n −[X

+

∩ Y [ is odd ¦.

Then X

+

and X

−

are elements of ¸

+

and ¸

−

, respectively; moreover,

Y ∈ ¸

n

is an element of ¸

−

if and only if n −[X

−

∩ Y [ is even. Also note

the following remarkable property: for X, Y ∈ ¸

n

the number n −[X ∩ Y [

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 45

is odd if and only if one of these subsets belongs to ¸

+

and the other is an

element of ¸

−

. Let

¸

∗

:= ¸ ¸ ¸

n−1

.

We deﬁne an incidence relation ∗ on ¸

∗

as follows: in the case when X ∈ ¸

k

,

k ≤ n − 2, and Y ∈ ¸

∗

, we write X ∗ Y if X and Y are incident in the

usual sense (one of these subsets is contained in the other); if X ∈ ¸

+

and

Y ∈ ¸

−

then X ∗ Y means that X ∩ Y belongs to ¸

n−1

(for example, X

+

and X

−

are incident). The associated ﬂag complex is called the oriﬂamme

complex and denoted by D

n

.

Example 2.3. Let V be a ﬁnite-dimension vector space over a division ring

and ∆(V ) be the ﬂag complex deﬁned by the natural incidence relation on

the set of all proper linear subspaces of V . If Ω is a non-degenerate reﬂexive

form on V then we write ∆(Ω) for the subcomplex of ∆(V ) consisting of

all ﬂags formed by totally isotropic subspaces of Ω.

Now suppose that V is a (2n)-dimensional vector space over a ﬁeld whose

characteristic is not equal to 2 and Ω is a non-degenerate symmetric form

on V . The dimension of maximal totally isotropic subspaces is assumed to

be equal to n. Then every (n − 1)-dimensional totally isotropic subspace

is contained in precisely two maximal totally isotropic subspaces. The

action of the group SO(Ω) (the group formed by all linear automorphisms

of determinant 1 which preserve the form Ω) on the Grassmannian (

n

(Ω) is

not transitive: there are precisely two orbits which will be denoted by (

+

(Ω)

and (

−

(Ω). Note that for two maximal totally isotropic subspaces S and U

the codimension of S ∩ U in S is odd if and only if one of these subspaces

belongs to (

+

(Ω) and the other is an element of (

−

(Ω). Let (

∗

(Ω) be the

set of all totaly isotropic subspaces whose dimension is not equal to n −1.

As in the previous example, we deﬁne the oriﬂamme incidence relation ∗

on (

∗

(Ω). If S ∈ (

k

(Ω), k ≤ n −2, and U ∈ (

∗

(Ω) then S ∗ U means that

S and U are incident in the usual sense; for S ∈ (

+

(Ω) and U ∈ (

−

(Ω) we

write S ∗ U if their intersection is (n −1)-dimensional. The associated ﬂag

complex is called the oriﬂamme complex of Ω and denoted by Orif(Ω).

We will restrict ourselves to simplicial complexes satisfying the following

condition: every simplex is contained in a certain maximal simplex and all

maximal simplices have the same cardinality. The rank of a such simplicial

complex is the number of vertices in maximal simplices. The ﬂag complexes

considered in Examples 2.1, 2.2 and 2.3 satisfy this condition.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

46 Grassmannians of Classical Buildings

2.1.2 Chamber complexes

Let ∆ be a simplicial complex of rank n. Two maximal simplices C and C

**are said to be adjacent if
**

[C ∩ C

[ = n −1.

Let Ch(∆) be the set of all maximal simplices of ∆ and Γ

ch

(∆) be the graph

whose vertex set is Ch(∆) and whose edges are pairs of adjacent elements.

We say that ∆ is a chamber complex if the graph Γ

ch

(∆) is connected. In

this case, maximal simplices of ∆ are said to be chambers; and we deﬁne the

distance d(C, C

) between two chambers C and C

**as the distance between
**

the corresponding vertices in the graph Γ

ch

(∆).

In a chamber complex of rank n, every simplex consisting of n − 1

vertices is called a panel. A chamber complex is said to be thick if every

panel is contained in at least 3 chambers; and it is called thin in the case

when every panel is contained in precisely 2 chambers.

Lemma 2.1. If f and g are isomorphisms between thin chamber complexes

∆ and ∆

**such that the restrictions of f and g to a certain chamber are
**

coincident then f = g.

Proof. Suppose that the restrictions of f and g to a chamber C

0

are

coincident and consider a chamber C adjacent with C

0

. Then f(C) and

g(C) are chambers containing the panel

f(C

0

∩ C) = g(C

0

∩ C)

and distinct from the chamber f(C

0

) = g(C

0

). Thus f(C) = g(C) (since

our chamber complexes are thin) and the restrictions of f and g to C are

coincident. By connectedness, the same holds for every chamber of ∆.

Let ∆ be a simplicial complex whose vertex set is X. We say that ∆ is

labeled by a set S if there exists a mapping α : X → S which satisﬁes the

following conditions:

• the restriction of α to every maximal simplex is bijective,

• for all A, B ∈ ∆ the inclusion A ⊂ B implies that α(A) ⊂ α(B);

in this case, the mapping α is called a labeling of ∆. If α is a labeling of

∆ by S then for any permutation t on the set S the mapping tα also is a

labeling of ∆.

Lemma 2.2. If ∆ is a chamber complex labeled by a set S then any two

labelings of ∆ by S are coincident up to a permutation on S.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 47

Proof. Let α and β be labelings of ∆ by S. Denote by α

and β

their

restrictions to a chamber C

0

. Since α

and β

**are bijective, there exists a
**

permutation t on S such that β

= tα

. Show that β = tα.

We take any chamber C adjacent with C

0

. Then C ¸ C

0

and C

0

¸ C are

one-point sets. Suppose that

C

0

¸ C = ¦x¦ and C ¸ C

0

= ¦y¦.

Then

α(x), α(y) ∈ S ¸ α(C ∩ C

0

).

Since the latter subset consists of unique element, we get α(x) = α(y).

The same holds for the labeling β. The equality β(x) = tα(x) (recall that

x ∈ C

0

) implies that β(y) = tα(y). Therefore, the restrictions of β and tα

to C are coincident. The connectedness of Γ

ch

(∆) gives the claim.

Lemma 2.3. Let ∆ be a labeled chamber complex. If 3 distinct chambers

of ∆ are mutually adjacent then their intersection is a panel.

Proof. Let α be a labeling of ∆. If C

1

and C

2

are adjacent chambers

then

C

1

¸ C

2

= ¦x¦ and C

2

¸ C

1

= ¦y¦

for certain vertices x, y and α(x) = α(y) (see the proof of the previous

lemma). If a chamber C

3

is adjacent with C

1

, C

2

and it does not contain

the panel C

1

∩C

2

then x, y ∈ C

3

; since α(x) = α(y), this contradicts the fact

that the restriction of α to C

3

is bijective. Therefore, C

1

∩ C

2

is contained

in C

3

.

2.1.3 Grassmannians and Grassmann spaces

Let ∆ be a chamber complex of rank n and α be a labeling of ∆ by a set

S. Then [S[ = n. For every s ∈ S we deﬁne the (α, s)-Grassmannian

(

α,s

(∆) := α

−1

(s).

If α

**is another labeling of ∆ by the set S then, by Lemma 2.2, there exists
**

a permutation t on the set S such that α

**= tα and every Grassmannian
**

(

α

,s

(∆) coincides with a certain Grassmannian (

α,s

(∆).

Therefore, the vertex set of ∆ can be decomposed in n disjoint subsets

which will be called Grassmannians. This decomposition does not depend

on a labeling of the complex ∆.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

48 Grassmannians of Classical Buildings

Let ( be a Grassmannian of ∆. Two distinct vertices a, b ∈ ( are said

to be adjacent if there exist adjacent chambers A and B such that a ∈ A

and b ∈ B; this is equivalent to the existence of a panel P such that P ∪¦a¦

and P ∪ ¦b¦ are chambers.

The Grassmann graph Γ

G

associated with the Grassmannian ( is the

graph whose vertex set is ( and whose edges are pairs of adjacent vertices.

Proposition 2.1. The Grassmann graph Γ

G

is connected.

Proof. Let a, b ∈ (. We take chambers A and B such that a ∈ A and

b ∈ B. Every path in Γ

ch

(∆) connecting A and B induces a path in Γ

G

which connects a and b.

Every simplex of ∆ intersects ( in at most one vertex. Consider a panel

P which does not intersect (. The set consisting of all x ∈ ( such that

P ∪ ¦x¦ is a chamber will be called the line of ( associated with (deﬁned

by) the panel P. Clearly, any two distinct elements of this line are adjacent.

Denote by L the set of all such lines. The partial linear space G := ((, L)

is known as the Grassmann space or the shadow space corresponding to

the Grassmannian (. The collinearity relation of G coincides with the

adjacency relation and the associated collinearity graph is the Grassmann

graph Γ

G

.

Let f be an isomorphism of ∆ to a chamber complex ∆

. Then ∆

is

labeled and f transfers Grassmannians to Grassmannians (if α is a labeling

of ∆ then αf

−1

is a labeling of ∆

**); moreover, f induces collineations
**

between the Grassmann spaces.

The term “Grassmannian” is motivated by the following example.

Example 2.4. Let V be an n-dimensional vector space over a division

ring. The ﬂag complex ∆(V ) is labeled by the dimension function and its

rank is equal to n −1. This is a chamber complex (it will be shown later).

The Grassmannians of ∆(V ) coincide with the Grassmannians of the vector

space V . If k = 1, n−1 then any two distinct elements of (

k

(V ) are adjacent

and the associated Grassmann spaces are the projective space Π

V

and the

dual projective space Π

∗

V

. In the case when 1 < k < n − 1, two elements

of (

k

(V ) are adjacent if their intersection is (k − 1)-dimensional (this is

equivalent to the fact that the sum of these linear subspaces is (k + 1)-

dimensional); every line consists of all elements of (

k

(V ) “lying between”

two incident linear subspaces of dimension k −1 and k + 1.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 49

2.2 Coxeter systems and Coxeter complexes

2.2.1 Coxeter systems

Let W be a group generated by a set S. If s, s

∈ S then we write m(s, s

)

for the order of the element ss

**. We suppose that m(s, s) = 1 for every
**

s ∈ S and m(s, s

) ≥ 2 if s ,= s

**. The ﬁrst condition guarantees that every
**

element of S is an involution. We have m(s, s

) = m(s

**, s) and the equality
**

m(s, s

) = 2 holds if and only if the involutions s and s

commute. Denote

by I the set of all pairs (s, s

) ∈ S S such that m(s, s

) is ﬁnite.

The pair (W, S) is called a Coxeter system if the group W has the

presentation

¸ S : (ss

)

m(s,s

)

= 1, (s, s

) ∈ I );

this means that every mapping h of the set S to a group G satisfying

(h(s)h(s

))

m(s,s

)

= 1 ∀ (s, s

) ∈ I

can be extended to a homomorphism of W to G.

The diagram associated with the Coxeter system (W, S) is the graph

whose vertex set is S and any two distinct s, s

∈ S are connected by

m(s, s

) − 2 edges (in the case when m(s, s

**) = ∞, we draw one edge
**

labeled by ∞). Note that two generators are not connected by an edge if

they commute.

Coxeter systems (W, S) and (W

, S

**) are called isomorphic if there exists
**

an isomorphism h : W → W

such that h(S) = S

**. Coxeter systems are
**

isomorphic if and only if they have the same diagram.

A Coxeter system is called irreducible if its diagram is connected.

Remark 2.1. Let ¦S

i

¦

i∈I

be the connected components of the diagram

associated with a Coxeter system (W, S). We deﬁne W

i

:= ¸S

i

) for every

i ∈ I. Then each (W

i

, S

i

) is an irreducible Coxeter system and W is the

“bounded” direct product of all W

i

, Subsection IV.1.9 in [Bourbaki (1968)].

A Coxeter system (W, S) is called ﬁnite if W is ﬁnite (the group W does

not need to be ﬁnite if S is ﬁnite). There are precisely 11 types of ﬁnite

irreducible Coxeter systems (Subsection VI.4.1 in [Bourbaki (1968)]):

A

n

B

n

= C

n

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

50 Grassmannians of Classical Buildings

D

n

E

6

E

7

E

8

F

4

4

G

2

3

H

3

3

H

4

i

l

2

(i), i = 3 or i ≥ 5

Let W be a group generated by a set S. The length l(w) = l

S

(w) of

an element w ∈ W (with respect to the generating set S) is the smallest

number n such that w has an expression w = s

1

. . . s

n

with each s

i

∈ S.

An expression w = s

1

. . . s

n

is called reduced if n = l(w). We say that the

pair (W, S) satisﬁes the exchange condition if for every reduced expression

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 51

w = s

1

. . . s

n

and every s ∈ S satisfying l(sw) ≤ n there exists i ∈ ¦1, . . . , n¦

such that

sw = s

1

. . . ˆ s

i

. . . s

n

(the symbol ˆmeans that the corresponding term is omitted).

Theorem 2.1. Let W be a group and S ⊂ W be a set of involutions which

generates W. Then (W, S) satisﬁes the exchange condition if and only if it

is a Coxeter system.

Proof. See Subsection IV.1.6 in [Bourbaki (1968)].

The following properties of a Coxeter system (W, S) can be drawn from

the exchange condition.

Exercise 2.1. Show that if s

1

, . . . , s

n

, s

1

, . . . , s

n

∈ S, w = s

1

. . . s

n

=

s

1

. . . s

n

and l(w) = n then

¦s

1

, . . . , s

n

¦ = ¦s

1

, . . . , s

n

¦.

Hint: the proof is induction by l(w); apply the exchange condition to s

1

w

and s

1

w.

By Exercise 2.1, for every w ∈ W there exists a subset S

w

⊂ S such

that every reduced expression of w is formed by all elements of S

w

. For

every subset X ⊂ S we deﬁne W

X

:= ¸X).

Exercise 2.2. Show that W

X

consists of all w ∈ W satisfying S

w

⊂ X.

Hint: establish that

S

sw

⊂ ¦s¦ ∪ S

w

**for every s ∈ S and prove that
**

S

ww

⊂ S

w

∪ S

w

induction by l(w).

Using Theorem 2.1 and Exercise 2.2 we establish the following.

Theorem 2.2. Let (W, S) be a Coxeter system. Then for every subset

X ⊂ S the pair (W

X

, X) is a Coxeter system, and

W

X

∩ W

Y

= W

X∩Y

for all subsets X, Y ⊂ S.

Proof. See Subsection IV.1.8 in [Bourbaki (1968)].

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

52 Grassmannians of Classical Buildings

2.2.2 Coxeter complexes

Let (W, S), S = ¦s

1

, . . . , s

n

¦ be a Coxeter system. We deﬁne

W

k

:= ¸S ¸ ¦s

k

¦), k = 1, . . . , n;

and for every subset J = ¦j

1

, . . . , j

m

¦ ⊂ ¦1, . . . , n¦

W

J

:= ¸S ¸ ¦s

j

1

, . . . , s

j

m

¦).

Then, by Theorem 2.2,

W

J

= W

j

1

∩ ∩ W

j

m

.

Exercise 2.3. Show that

wW

i

= w

W

j

=⇒ i = j and ww

−1

∈ W

i

.

Thus wW

i

= W

i

implies that w ∈ W

i

, and we have w = 1 if the latter

equality holds for all i.

The Coxeter complex Σ(W, S) is the simplicial complex whose vertex set

consists of all special subsets wW

k

with w ∈ W and k ∈ ¦1, . . . , n¦; special

subsets X

1

, . . . , X

m

form a simplex if there exists w ∈ W such that

X

1

= wW

j

1

, . . . , X

m

= wW

j

m

.

We identify this simplex with the special subset

X

1

∩ ∩ X

m

= wW

J

,

where J = ¦j

1

, . . . , j

m

¦. Then every maximal simplex

¦wW

1

, . . . , wW

n

¦

will be identiﬁed with the element w ∈ W.

The complex Σ(W, S) is ﬁnite if the Coxeter system is ﬁnite.

Proposition 2.2. The Coxeter complex Σ(W, S) is a thin chamber com-

plex.

Proof. Denote by C

0

the maximal simplex formed by W

1

, . . . , W

n

. Then

wC

0

is the maximal simplex corresponding to w ∈ W. If w = s

i

1

. . . s

i

k

then

C

0

, C

1

= s

i

1

C

0

, C

2

= s

i

1

s

i

2

C

0

, . . . , C

k

= s

i

1

. . . s

i

k

C

0

= wC

0

is a path in Γ

ch

(∆). Therefore, C

0

can be connected with every element of

Ch(∆) and the graph Γ

ch

(∆) is connected.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 53

Let P

k

be the panel formed by

W

1

, . . . ,

ˆ

W

k

, . . . , W

n

(the vertex W

k

is omitted). If a chamber wC

0

contains P

k

then

w ∈

i=k

W

i

= ¦1, s

k

¦

(by Exercise 2.3). Thus P

k

is contained only in the chambers C

0

and s

k

C

0

.

Then wP

k

is contained only in the chambers wC

0

and ws

k

C

0

.

The mapping wW

k

→ s

k

is the canonical labeling of Σ(W, S) by the set

S (by Lemma 2.2, any other labeling of Σ(W, S) by S is the composition of

the canonical labeling with a permutation on S). We write Aut

0

(Σ(W, S))

for the group of all automorphisms of Σ(W, S) preserving the canonical

labeling. In the general case, there exist automorphisms of Σ(W, S) which

do not belong to this group (see Example 2.5).

Exercise 2.4. For every w ∈ W denote by l

w

the automorphism of Σ(W, S)

which sends every special subset X to wX. Show that w → l

w

is an

isomorphism of W to the group Aut

0

(Σ(W, S)). Hint: use Lemma 2.1 to

prove the surjectivity.

The Grassmannians of Σ(W, S) coincide with the orbits of the action

of the group W = Aut

0

(Σ(W, S)) on the vertex set of Σ(W, S). Denote

by (

k

(W, S) the Grassmannian containing W

k

; it consists of all special

subsets wW

k

, w ∈ W. Grassmannians of Coxeter complexes are closely

related with the concept of so-called Coxeter matroids [Borovik, Gelfand

and White (2003)].

2.2.3 Three examples

In this subsection we consider the Coxeter systems of types A

n

, C

n

, D

n

and

show that the associated Coxeter complexes are A

n

, C

n

, D

n

(respectively).

Example 2.5 (Type A

n

, n ≥ 1). Let W = S

n+1

, where S

n+1

is the sym-

metric group consisting of all permutations of the set I := ¦1, . . . , n + 1¦.

Let also S be the set formed by the transpositions

s

i

= (i, i + 1), i = 1, . . . , n.

Then (W, S) is a Coxeter system whose diagram is

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

54 Grassmannians of Classical Buildings

S

n-1

S

n

S

1

S

2

S

3

For every number k ∈ ¦1, . . . , n¦ the subgroup W

k

is the stabilizer of the

subset ¦1, . . . , k¦ and the mapping

wW

k

→ ¦w(1), w(2), . . . , w(k)¦

is well-deﬁned. This is an isomorphism of Σ(W, S) to the complex A

n

(Example 2.1); it transfers the Grassmannian (

k

(W, S) to the set of all k-

element subsets of I. The automorphism of the complex A

n

sending every

subset X ⊂ I to I ¸X induces an automorphism of Σ(W, S) which does not

preserve the canonical labeling.

Example 2.6 (Type B

n

= C

n

, n ≥ 2). We say that a permutation s on

the set J := ¦1, . . . , n, −1, . . . , −n¦ is symplectic if

s(−j) = −s(j) ∀ j ∈ J.

All symplectic permutations form the group denoted by Sp

n

. Elements of

this group preserve ¸ (the set of all singular subsets, see Example 2.2).

The group Sp

n

is generated by the “symplectic transpositions”

s

i

= (i, i + 1)(−i, −(i + 1)), i = 1, . . . , n −1,

and the transposition

s

n

= (n, −n).

Suppose that W = Sp

n

and S is the set of generators considered above.

Then (W, S) is a Coxeter system with the diagram

S

n-1

S

n

S

1

S

2

S

3

As in the previous example, the subgroup W

k

is the stabilizer of the subset

¦1, . . . , k¦ and the mapping

wW

k

→ ¦w(1), w(2), . . . , w(k)¦

is an isomorphism of Σ(W, S) to the complex C

n

(Example 2.2). This

isomorphism sends the Grassmannian (

k

(W, S) to ¸

k

(the set formed by

all singular subsets consisting of k elements).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 55

Example 2.7 (Type D

n

, n ≥ 4). In this example we suppose that W is

the subgroup SO

n

⊂ Sp

n

generated by the set S consisting of s

1

, . . . , s

n−1

(deﬁned in the previous example) and the “symplectic transposition”

s

n

= (n −1, −n)(−(n −1), n).

Then (W, S) is a Coxeter system whose diagram is

S

1

S

2

S

3

S

n-1

S

n-2

S

n

The action of the group W on the set ¸

n

is not transitive. For example,

W does not contain permutations transferring

X

+

:= ¦1, . . . , n¦ to X

−

:= ¦1, . . . , n −1, −n¦

(since every symplectic permutation sending X

+

to X

−

is the composition

of (n, −n) and an element of W

n

). The reader can show that ¸

+

and ¸

−

(Example 2.2) are the orbits of the action of W on ¸

n

and the mapping

wW

k

→ ¦w(1), w(2), . . . , w(k)¦ if k ≤ n −2,

wW

n−1

→ ¦w(1), . . . , w(n −1), w(−n)¦,

wW

n

→ ¦w(1), . . . , w(n −1), w(n)¦

is an isomorphism of Σ(W, S) to the oriﬂamme complex D

n

(Example

2.2). This isomorphism transfers the Grassmannians of Σ(W, S) to the

sets ¸

1

, . . . , ¸

n−2

, ¸

−

, ¸

+

.

2.3 Buildings

2.3.1 Deﬁnition and elementary properties

Let ∆ be a simplicial complex and A be a set of subcomplexes of ∆ satisfying

the following axioms:

(1) every element of A is isomorphic to a Coxeter complex,

(2) for any two simplices of ∆ there is an element of A containing both of

them.

Then ∆ is a chamber complex (since every Coxeter complex is a chamber

complex, Proposition 2.2). The simplicial complex ∆ is a building if the

following additional axiom holds:

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

56 Grassmannians of Classical Buildings

(3) for all Σ, Σ

∈ A such that Σ∩Σ

**contains a chamber of ∆ there exists
**

an isomorphism of Σ to Σ

**preserving pointwise all simplices contained
**

in Σ ∩ Σ

**(by Lemma 2.1, such isomorphism is unique);
**

in this case, elements of A are said to be apartments and A is called a system

of apartments.

The axiom (3) guarantees that any two apartments Σ, Σ

∈ A are iso-

morphic: we take chambers C ∈ Σ and C

∈ Σ

**and consider an apartment
**

Σ

containing them; by the axiom (3), Σ

is isomorphic to both Σ and Σ

.

Similar arguments can be used to prove the following property (we leave

the details for the reader):

(4) if Σ, Σ

∈ A then for any simplices A, B ∈ Σ ∩ Σ

there exists an

isomorphism of Σ to Σ

**preserving A and B pointwise.
**

Remark 2.2. The axiom systems (1), (2), (3) and (1), (2), (4) are equiva-

lent, Section IV.1 in [Brown (1989)]. Moreover, by Section 4.3 in [Garrett

(1997)], the axiom (1) can be drawn from the axioms (2) and (3).

Example 2.8. Every Coxeter complex can be considered as a building with

unique apartment.

Let ∆ be a building and A be a system of apartments for ∆. We write

A

C

for the set of all apartments containing a chamber C. By the axiom

(3), for any two apartments Σ, Σ

∈ A

C

there is a unique isomorphism f

ΣΣ

of Σ to Σ

**whose restriction to C is identity. We have
**

f

Σ

Σ

= f

Σ

Σ

f

Σ

Σ

for all Σ, Σ

, Σ

∈ A

C

(because f

Σ

Σ

f

Σ

Σ

is an isomorphism of Σ

to Σ

preserving C pointvise). Now we ﬁx an apartment Σ ∈ A

C

. Since the

union of all apartments from A

C

coincides with ∆ (by the axiom (2)), all

isomorphisms

f

Σ

Σ

, Σ

∈ A

C

,

ﬁt together to give a surjective morphism of ∆ to Σ. This morphism will

be denoted by ρ

Σ,C

and called the retraction of ∆ on Σ centered at C.

If α is a labeling of Σ (apartments are labeled by the axiom (1)) then

αρ

Σ,C

is a labeling of ∆. Therefore, we have the following.

Proposition 2.3. Every building is a labeled chamber complex.

Proposition 2.4. Let Σ ∈ A and C, C

**be chambers of Σ. Then Σ contains
**

every geodesic of the graph Γ

ch

(∆) connecting C and C

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 57

Proof. Suppose that

C = C

0

, C

1

, . . . , C

k

= C

is a geodesic of Γ

ch

(∆) which is not contained in Σ. We choose an index i

such that C

i−1

∈ Σ and C

i

,∈ Σ. There is a unique chamber C

∈ Σ distinct

from C

i−1

and containing the panel C

i−1

∩ C

i

. Consider the retraction

ρ = ρ

Σ,C

. The chamber ρ(C

i

) ∈ Σ contains the panel

ρ(C

i−1

∩ C

i

) = C

i−1

∩ C

i

,

hence it coincides with C

i−1

or C

**. Since there is an apartment containing
**

both C

and C

i

, we have

ρ(C

i

) ,= ρ(C

) = C

.

So ρ(C

i

) = C

i−1

and

C = C

0

, C

1

, . . . , C

i−1

= ρ(C

i

), ρ(C

i+1

), . . . , ρ(C

k

) = C

is a part in Γ

ch

(∆) (if A, B are adjacent chambers then ρ(A), ρ(B) are

adjacent or coincident). This path contains at most k − 1 edges which

contradicts the assumption that d(C, C

) = k.

In general, the building ∆ can admit diﬀerent systems of apartments;

but the union of any collection of apartment systems is again an apart-

ment system, Section IV.4 in [Brown (1989)]. This implies the existence

of a largest system of apartments. Moreover, there exists a unique (up to

isomorphism) Coxeter system (W, S) such that all apartments of ∆ are iso-

morphic to the Coxeter complex Σ(W, S), Section IV.3 in [Brown (1989)].

2.3.2 Buildings and Tits systems

Suppose that G is a group spanned by proper subgroups B and N; more-

over, B ∩N is a normal subgroup of N. Let S be a set of generators of the

quotient group

W := N/(B ∩ N).

If w ∈ W then for any two elements g

1

and g

2

belonging to the class w we

have g

−1

2

g

1

∈ B ∩ N and g

1

B = g

2

B; thus wB and

C(w) := BwB

are well-deﬁned. We require that the following two technical conditions

hold:

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

58 Grassmannians of Classical Buildings

(1) C(s)C(w) ⊂ C(w) ∪ C(sw) for all s ∈ S and w ∈ W,

(2) sBs

−1

is not contained in B for every s ∈ S.

Then (G, B, N, S) is said to be a Tits system and W is called the Weyl group

of this Tits system; also we say that the subgroup B, N form a BN-pair of

the group G.

The conditions (1) and (2) give the following remarkable properties,

Section V.2 in [Brown (1989)]:

(3) every s ∈ S is an involution and (W, S) is a Coxeter system;

(4) C(w) ∩ C(w

) = ∅ if w ,= w

;

(5) for every subgroup W

= ¸S

) with S

⊂ S

BW

B :=

w∈W

C(w)

is a subgroup of G; in particular, we have BWB = G, and BW

B

coincides with B if W

= ¦1¦;

(6) C(s)C(w) = C(sw) if l(sw) ≥ l(w), and C(s)C(w) = C(w) ∪ C(sw) if

l(sw) ≤ l(w).

Now assume that S = ¦s

1

, . . . , s

n

¦. For every number k ∈ ¦1, . . . , n¦ and

every subset J ⊂ ¦1, . . . , n¦ we deﬁne W

k

and W

J

as in Subsection 2.2.2.

By the property (5), all

P

J

:= BW

J

B, J ⊂ ¦1, . . . , n¦,

are subgroup of G and

P

J

= P

j

1

∩ ∩ P

j

m

if J = ¦j

1

, . . . , j

m

¦.

These subgroups are called special. Every special subgroup contains B.

Conversely, every subgroup of G containing B is special, Section V.2 in

[Brown (1989)].

Exercise 2.5. Show that

gP

i

= g

P

j

=⇒ i = j and gg

−1

∈ P

i

.

Consider the simplicial complex ∆ = ∆(G, B, N, S) whose vertex set

consists of all special subsets gP

k

with g ∈ G and k ∈ ¦1, . . . , n¦; special

subsets X

1

, . . . , X

m

form a simplex if there exists g ∈ G such that

X

1

= gP

j

1

, . . . , X

m

= gP

j

m

.

We identify this simplex with the special subset

X

1

∩ ∩ X

m

= gP

J

,

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 59

where J = ¦j

1

, . . . , j

m

¦. Then maximal simplices will be identiﬁed with

special subsets gB.

Remark 2.3. A subgroup of G is called parabolic if it is conjugate to a spe-

cial subgroup. There is a one-to-one correspondence between simplices of

∆ and parabolic subgroups, given by gP → gPg

−1

(it easy follows from the

fact that there are no two distinct special subgroups which are conjugate).

If w ∈ W then for any two elements g

1

and g

2

of the class w we have

g

1

P

J

= g

2

P

J

for every J ⊂ ¦1, . . . , n¦ and denote this special subset by

wP

J

. Let Σ be the subcomplex of ∆ formed by all special subsets wP

J

with w ∈ W, J ⊂ ¦1, . . . , n¦; it will be called the fundamental apartment

of ∆. The mapping

wW

k

→ wP

k

is an isomorphism of the Coxeter complex Σ(W, S) to the fundamental

apartment Σ. For every g ∈ G denote by l

g

the automorphism of ∆ trans-

ferring every special subset X to gX. The subcomplexes

gΣ := l

g

(Σ), g ∈ G,

are called apartments of ∆. Then ¦gΣ¦

g∈G

is an apartment system and ∆

is a building, Section V.3 in [Brown (1989)].

The mapping gP

k

→ s

k

is the canonical labeling of the building ∆ by

the set S (any other labeling of ∆ by S is the composition of the canonical

labeling with a permutation on S, Lemma 2.2). Denote by Aut

0

(∆) the

group formed by all automorphisms of ∆ preserving the canonical labeling

(there exist automorphisms of ∆ which do not belong to this group, see

Remark 2.4). The mapping g → l

g

is a homomorphism of G to Aut

0

(∆);

it does not need to be injective and surjective (Remark 2.4).

The Grassmannians of ∆ coincide with the orbits of the left action of

the group G on the vertex set of ∆. Denote by (

k

(∆) the Grassmannian

containing P

k

, it consists of all special subsets gP

k

.

2.3.3 Classical examples

In this subsection we consider buildings associated with general linear, sym-

plectic, and orthogonal groups.

Example 2.9 (The general linear group). Let G = GL(V ), where V

is an (n+1)-dimensional vector space over a division ring. We take any base

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

60 Grassmannians of Classical Buildings

X

0

= ¦x

1

, . . . , x

n+1

¦ of this vector space and denote by B the stabilizer of

the maximal ﬂag

¸x

1

) ⊂ ¸x

1

, x

2

) ⊂ ⊂ ¸x

1

, . . . , x

n

)

in the group G. Let also N ⊂ G be the stabilizer of the projective base

¸x

1

), . . . , ¸x

n+1

). Then B ∩ N consists of all linear automorphisms pre-

serving each ¸x

i

). This is a normal subgroup of N and the corresponding

quotient group is isomorphic to the symmetric group S

n+1

. Let S be the

set of generators of S

n+1

considered in Example 2.5. Then (G, B, N, S) is

a Tits system, Section V.5 in [Brown (1989)]. The associated building ∆

can be identiﬁed with the ﬂag complex ∆(V ) (Example 2.3). Indeed, for

every number k ∈ ¦1, . . . , n¦ the subgroup P

k

is the stabilizer of the linear

subspace ¸x

1

, . . . , x

k

) and the mapping

gP

k

→ g(¸x

1

, . . . , x

k

))

is the required isomorphism. For every base X ⊂ V denote by Σ

X

the

subcomplex of ∆(V ) consisting of all ﬂags formed by the linear subspaces

spanned by subsets of X; this subcomplex is isomorphic to the complex

A

n

(Example 2.1). Then the fundamental apartment Σ is identiﬁed with

Σ

X

0

and gΣ = Σ

g(X

0

)

for every g ∈ G. The Grassmannian (

k

(∆) (the

Grassmannian containing P

k

) is (

k

(V ). The associated Grassmann spaces

were described in Example 2.4.

Remark 2.4. By the Fundamental Theorem of Projective Geometry

(Corollary 1.5), every element of Aut

0

(∆(V )) is induced by a semilinear

automorphism of V . For every homothetic transformation the associated

automorphism of ∆(V ) is identity. Hence the homomorphism g → l

g

of

GL(V ) to the group Aut

0

(∆(V )) is not injective. This homomorphism is

surjective only in the case when every automorphism of the associated di-

vision ring is inner, for example, if V is a real vector space or a vector space

over the division ring of real quaternion numbers (indeed, if l is a σ-linear

automorphism of V such that (l)

1

= (s)

1

for a certain linear automorphism

s : V → V then l is a scalar multiple of s which is equivalent to the fact that

the automorphism σ is inner). Also note that semilinear isomorphisms of V

to V

∗

(if they exist) induce automorphisms of ∆(V ) which do not preserve

the canonical labeling.

Example 2.10 (The symplectic group). Let Ω be a non-degenerate al-

ternating form on a (2n)-dimensional vector space V over a ﬁeld. Then

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 61

every 1-dimensional linear subspace is totally isotropic and the dimension

of maximal totally isotropic subspaces is equal to n. A base X ⊂ V is said

to be an Ω-base if for each x ∈ X there is precisely one y ∈ X satisfying

Ω(x, y) ,= 0. Such bases form a suﬃciently wide class.

We ﬁx an Ω-base

X

0

= ¦x

1

, y

1

, . . . , x

n

, y

n

¦

satisfying Ω(x

i

, y

i

) ,= 0 for all i. Suppose that G = Sp(Ω) (the group formed

by all linear automorphisms of V preserving the form Ω). Let B ⊂ G be

the stabilizer of the ﬂag

¸x

1

) ⊂ ¸x

1

, x

2

) ⊂ ⊂ ¸x

1

, . . . , x

n

)

(all linear subspaces in this ﬂag are totally isotropic) and N ⊂ G be the

stabilizer of the projective base

¸x

1

), ¸y

1

), . . . , ¸x

n

), ¸y

n

).

Then B ∩ N is a normal subgroup of N and the corresponding quotient

group is isomorphic to Sp

n

. If S is the set of generators of Sp

n

considered

in Example 2.6 then (G, B, N, S) is a Tits system, Section V.6 in [Brown

(1989)]. As in the previous example, we show that the associated building

can be identiﬁed with the subcomplex ∆(Ω) ⊂ ∆(V ) (Example 2.3). The

fundamental apartment is the intersection of ∆(Ω) with Σ

X

0

, and every

apartment is the subcomplex ∆(Ω) ∩ Σ

X

, where X is a certain Ω-base.

These apartments are isomorphic to the complex C

n

(Example 2.2).

The Grassmannians of this building are (

k

(Ω), k ∈ ¦1, . . . , n¦. Two

elements of (

n

(Ω) are adjacent if their intersection is (n −1)-dimensional;

every line in (

n

(Ω) consists of all maximal totally isotropic subspaces con-

taining a certain element of (

n−1

(Ω). In the case when k < n, elements

S, U ∈ (

k

(Ω) are adjacent if S ⊥ U and S ∩ U belongs to (

k−1

(Ω) (this is

equivalent to the fact that S + U is an element of (

k+1

(Ω)); the lines are

deﬁned as for Grassmannians of ﬁnite-dimensional vector spaces (by pairs

of incident totally isotropic subspaces of dimension k −1 and k + 1).

Example 2.11 (The orthogonal group). Let V be a (2n)-dimensional

vector space over a ﬁeld whose characteristic is not equal to 2. Let also

Ω be a non-degenerate symmetric form deﬁned on V . The dimension of

maximal totally isotropic subspaces is assumed to be equal to n. We say

that a base X ⊂ V is an Ω-base if it consists of isotropic vectors and for

each x ∈ X there is precisely one y ∈ X satisfying Ω(x, y) ,= 0. As in the

previous example, these bases form a suﬃciently wide class.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

62 Grassmannians of Classical Buildings

Consider the oriﬂamme complex Orif(Ω) (Example 2.3). Every Ω-base

X deﬁnes the subcomplex Orif

X

⊂ Orif(Ω) consisting of all “oriﬂamme

ﬂags” formed by the totally isotropic subspaces spanned by subsets of X;

this subcomplex is isomorphic to the oriﬂamme complex D

n

(Example 2.2).

Let us ﬁx an Ω-base X

0

= ¦x

1

, . . . , x

2n

¦ and a maximal simplex (a maximal

oriﬂamme ﬂag) in the subcomplex Orif

X

0

. Suppose that G = SO(Ω) and

denote by B the stabilizer of this maximal simplex in G. Let N ⊂ G

be the stabilizer of the projective base ¸x

1

), . . . , ¸x

2n

). Then B ∩ N is a

normal subgroup of N and the corresponding quotient group is isomorphic

to SO

n

. Let S be the set of generators of the group SO

n

from Example

2.7. Then (G, B, N, S) is a Tits system, Section V.7 in [Brown (1989)].

Standard arguments show that the associated building can be identiﬁed

with the oriﬂamme complex Orif(Ω). Every apartment of this building is

Orif

X

, where X is an Ω-base.

The Grassmannians of this building are (

k

(Ω), k ≤ n − 2, and (

δ

(Ω),

δ ∈ ¦+, −¦. The adjacency relation on (

k

(Ω), k ≤ n−2, is deﬁned as in the

previous example. Two elements of (

δ

(Ω) are adjacent if their intersection

belongs to (

n−2

(Ω); every line consists of all elements of (

δ

(Ω) containing

a certain (n −2)-dimensional totally isotropic subspace.

Example 2.12. Let G be a reductive algebraic group over a ﬁeld and B be

a Borel subgroup of G containing a maximal torus T (we refer [Humphreys

(1975)] for the precise deﬁnitions). Denote by N the normalizer of T in

G. Then B and N form a BN-pair of G. The associated Tits system and

building are described in [Tits (1974)] (Chapter 5).

2.3.4 Spherical buildings

A building ∆ is called spherical if the associated Coxeter system is ﬁnite.

In this case, by Proposition 2.4, the diameter of the graph Γ

ch

(∆) is ﬁnite

and equal to the diameter of the graph Γ

ch

(Σ), where Σ is an apartment

of ∆. We say that two chambers of a spherical building are opposite if the

distance between them is maximal (is equal to the diameter).

The term “spherical” is motivated by the following.

Proposition 2.5. In a spherical building every apartment Σ is the union

of all geodesics connecting two opposite chambers C, C

∈ Σ.

Proof. See Section IV.5 in [Brown (1989)].

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 63

Corollary 2.1. Every spherical building admits a unique system of apart-

ments; in other words, if ∆ is a spherical building and A, A

are apartment

systems for ∆ then A = A

.

Corollary 2.2. Every isomorphism between spherical buildings is apart-

ments preserving (sends apartments to apartments).

Proof. Let ∆, ∆

be spherical buildings and A, A

be their systems of

apartments. If f is an isomorphism of ∆ to ∆

then f(A) is an apartment

system for ∆

. By Corollary 2.1, we have f(A) = A

.

A building is called irreducible if the associated Coxeter system is irre-

ducible. Irreducible thick spherical buildings of rank ≥ 3 were classiﬁed in

[Tits (1974)]. There are three classical types

A

n

, B

n

= C

n

, D

n

,

and four exceptional types

F

4

and E

i

, i = 6, 7, 8

(the building type is the type of the associated Coxter system). In particu-

lar, there exists no thick building of type H

i

, i = 3, 4.

We restrict ourselves to buildings of classical types only. Every thick

building of type A

n

, n ≥ 3, is isomorphic to the ﬂag complex ∆(V ), where

V is an (n +1)-dimensional vector space over a division ring (this fact is a

reformulation of Theorem 1.3). All thick buildings of types C

n

and D

n

can

be obtained from so-called polar spaces (this construction will be considered

in Chapter 4).

2.3.5 Mappings of the chamber sets

Let ∆ and ∆

be buildings. Isomorphisms of ∆ to ∆

**(if they exist) induce
**

isomorphisms between the graphs Γ

ch

(∆) and Γ

ch

(∆

). Conversely, we have

the following.

Theorem 2.3 (J. Tits). If the diagram of the Coxeter system associated

with ∆ does not contain edges labeled by ∞ then every isomorphism of

Γ

ch

(∆) to Γ

ch

(∆

) is induced by an isomorphism of ∆ to ∆

.

Proof. See [Tits (1974)], p.51.

Theorem 2.4 ([Abramenko and Van Maldeghem (2000)]). If ∆ is

a thick spherical building then for any two distinct chambers C

1

, C

2

∈ ∆

the following conditions are equivalent:

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

64 Grassmannians of Classical Buildings

• C

1

and C

2

are adjacent,

• there exists a chamber C ,= C

1

, C

2

such that no chamber is opposite to

a unique member of the set ¦C, C

1

, C

2

¦.

In particular, if ∆

**also is a thick spherical building then every bijection of
**

Ch(∆) to Ch(∆

**) preserving the relation to be opposite (two chambers of ∆
**

are opposite if and only if their images are opposite) is an isomorphism of

Γ

ch

(∆) to Γ

ch

(∆

).

The intersection of an apartment of ∆ with the chamber set Ch(∆) will

be called an apartment in Ch(∆). We say that a mapping

f : Ch(∆) → Ch(∆

)

is apartments preserving if its restriction to every apartment of Ch(∆) is a

bijection on a certain apartment of Ch(∆

**). Every apartments preserving
**

mapping is injective (since for any two chambers there is an apartment

containing them). We want to show that apartments preserving mappings

are adjacency preserving; our proof will be based on the following charac-

terization of the adjacency relation in terms of apartments.

Lemma 2.4. Let ∆ be a thick building. Distinct chambers C, C

∈ ∆ are

adjacent if and only if the intersection of all apartments of Ch(∆) contain-

ing them coincides with ¦C, C

¦.

Proof. If the intersection of all apartments of Ch(∆) containing both

C and C

coincides with ¦C, C

¦ then, by Proposition 2.4, C and C

are

adjacent.

Conversely, suppose that C and C

**are adjacent and denote by A the
**

intersection of all apartments of Ch(∆) containing C, C

. Assume that

A ,= ¦C, C

¦.

First we establish the existence of C

∈ A ¸ ¦C, C

¦ such that

C, C

, C

or C

, C, C

is a geodesic in Γ

ch

(∆). We take any geodesic γ of Γ

ch

(∆) connecting C

with a certain element of A ¸ ¦C, C

**¦; by Proposition 2.4, γ is contained
**

in A. If γ contains C

then the ﬁrst possibility is realized. If C

does not

belong to γ then γ contains a chamber C

(C

,= C

**) adjacent with C; this
**

chamber is not adjacent with C

(otherwise, by Lemma 2.3, C

contains

the panel C ∩ C

which is impossible, since C, C

, C

∈ A and apartments

are thin chamber complexes).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 65

Consider the ﬁrst case (C, C

, C

**is a geodesic); the second is similar.
**

We choose a chamber

ˆ

C containing the panel C

∩C

and distinct from C

and C

**(our building is thick). The chambers C and
**

ˆ

C are not adjacent

(it follows from Lemma 2.3). Hence C, C

,

ˆ

C is a geodesic in Γ

ch

(∆). Then

every apartment containing C and

ˆ

C contains C

; but C

**does not belong
**

to this apartment which contradicts C

∈ A.

Theorem 2.5. Suppose that ∆ and ∆

**are thick buildings of the same rank.
**

Then every apartments preserving mapping of Ch(∆) to Ch(∆

) preserves

the adjacency relation (two chambers of ∆ are adjacent if and only if their

images are adjacent); in particular, every apartments preserving bijection

of Ch(∆) to Ch(∆

) is an isomorphism of Γ

ch

(∆) to Γ

ch

(∆

).

Proof. Let f : Ch(∆) → Ch(∆

**) be an apartments preserving mapping.
**

It was noted above that f is injective. Lemma 2.4 guarantees that f trans-

fers adjacent chambers to adjacent chambers.

Let C and

ˆ

C be chambers of ∆ and / be an apartment of Ch(∆)

containing them. Suppose that the rank of ∆ is equal to n. Then /

contains precisely n distinct chambers adjacent with C; denote them by

C

1

, . . . , C

n

. Their images

f(C

1

), . . . , f(C

n

) ∈ f(/)

are adjacent with f(C). Since the rank of ∆

**also is equal to n, the apart-
**

ment f(/) does not contain other chambers adjacent with f(C). Hence if

f(C) and f(

ˆ

C) are adjacent then we have

ˆ

C = C

i

for a certain i. Therefore,

C and

ˆ

C are adjacent if and only if their images are adjacent.

Remark 2.5. Theorem 2.5 is a modiﬁcation of a result obtained in [Abra-

menko and Van Maldeghem (2008)]. We do not assume that the mapping

is bijective, but require that the buildings are of the same rank.

2.4 Mappings of Grassmannians

The intersections of a building Grassmannian with apartments of the asso-

ciated building are called apartments of this Grassmannian. We say that

a mapping between two building Grassmannians is apartments preserving

if its restriction to every apartment is a bijection to an apartment. Apart-

ments preserving mappings are injective, since for any two elements of a

building Grassmannian there is an apartment containing them.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

66 Grassmannians of Classical Buildings

Example 2.13. Let V be an n-dimensional vector space over a division

ring. Every apartment of the Grassmannian (

k

(V ) consists of all k-

dimensional linear subspaces spanned by subsets of a certain base of V .

In the cases when k = 1, n −1, this is a base of the projective space Π

V

or

the dual projective space Π

∗

V

, respectively.

Let V and V

**be vector spaces of the same ﬁnite dimension n ≥ 3.
**

It is not diﬃcult to prove that every apartments preserving bijection of

(

1

(V ) to (

1

(V

**) (a bijection which sends bases of Π
**

V

to bases of Π

V

) is a

collineation of Π

V

to Π

V

; similarly, every apartments preserving bijection

of (

n−1

(V ) to (

n−1

(V

) is a collineation of Π

∗

V

to Π

∗

V

. In Chapters 3 and

4 we establish such kind results for other Grassmannians associated with

buildings of classical types.

Let ∆ and ∆

**be buildings of the same classical type X
**

n

, X ∈ ¦A, C, D¦.

Let also ( and (

be Grassmannians of ∆ and ∆

, respectively. Denote by

G and G

**the associated Grassmann spaces (Subsection 2.1.3).
**

We suppose that G and G

**both are not projective spaces. In this
**

case, we show that every isomorphism of Γ

G

to Γ

G

(Subsection 2.1.3) is a

collineation of G to G

**. For projective spaces this fails (since any two points
**

of a projective space are collinear and every bijection between projective

spaces gives an isomorphism of their collinearity graphs).

We describe all collineations of G to G

**and show that they can be ex-
**

tended to isomorphisms of ∆ to ∆

**(for projective spaces the latter state-
**

ment is trivial). In almost all cases (except the case when X ∈ ¦C, D¦ and

n = 4), these results easy follow from elementary properties of maximal sin-

gular subspaces of Grassmann spaces. Note that such kind theorems were

ﬁrst proved in [Chow (1949)] for Grassmannians of ﬁnite-dimensional vector

spaces and Grassmannians formed by maximal totally isotropic subspaces

of non-degenerate reﬂexive forms.

One of our main results is the description of all apartments preserving

mappings of ( to (

**; in particular, we establish that every apartments
**

preserving bijection of ( to (

is a collineation of G to G

(hence it can be

extended to an isomorphism of ∆ to ∆

).

The idea used to prove Theorem 2.5 does not work for building Grass-

mannians, since for any a, b ∈ ( the intersection of all apartments of (

containing both a, b coincides with ¦a, b¦.

Our method is based on combinatorial properties of so-called maximal

inexact subsets. We say that a subset of an apartment is inexact if there

exist other apartments containing this subset. Certain Grassmannians have

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Buildings and Grassmannians 67

maximal inexact subsets of precisely one type; but there exist other Grass-

mannians with maximal inexact subsets of two diﬀerent types. We char-

acterize the adjacency relation in terms of maximal inexact subsets. Using

this characterization we show that every apartments preserving mapping of

( to (

**is adjacency preserving (two elements of ( are adjacent if and only
**

if their images are adjacent).

Following [Cooperstein, Kasikova and Shult (2005)] we present results

of “opposite nature” (characterizations of apartments in terms of the ad-

jacency relation) for certain Grassmannians associated with buildings of

classical types.

2.5 Appendix: Gamma spaces

A partial linear space Π = (P, L) is called a gamma space if it satisﬁes the

following axiom:

(Γ) if a point is collinear with two distinct points of a line then it is collinear

with all points of this line.

This axiom implies that for any point p ∈ P and any line L ∈ L the set of

all points on L collinear with p is empty, or consists of a single point, or

coincides with L. In linear spaces the axiom holds trivially.

Now we establish some elementary properties of gamma spaces. It will

be shown later that all Grassmann spaces associated with buildings of clas-

sical types are gamma spaces.

Let Π = (P, L) be a gamma space. Consider a subset X ⊂ P consisting

of mutually collinear points (a clique in the collinearity graph of Π) and

suppose that [X[ ≥ 2. We write [X]

1

for the set formed by all points

belonging to the lines joining points of X (a point q belongs to [X]

1

if there

exist distinct points p, p

∈ X such that q is on the line pp

). It is clear that

X ⊂ [X]

1

.

Exercise 2.6. Show that [X]

1

is a clique of the collinearity graph.

For every natural i ≥ 2 we deﬁne

[X]

i

:= [[X]

i−1

]

1

.

Then [X]

i

is contained in [X]

j

if i ≤ j, and, by Exercise 2.6, each [X]

i

is a

clique of the collinearity graph.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

68 Grassmannians of Classical Buildings

Proposition 2.6. If X is a clique in the collinearity graph of a certain

gamma space and [X[ ≥ 2 then

¸X) =

∞

i=1

[X]

i

is a singular subspace.

Proof. Denote by S the union of all [X]

i

. If p and q are distinct points

of S then p ∈ [X]

i

and q ∈ [X]

j

for certain i, j. It is clear that p and q

both belong to [X]

m

, where m = max¦i, j¦; the line joining these points is

contained in [X]

m+1

. Thus S is a singular subspace. Since ¸X) contains

every [X]

i

, we have X ⊂ S ⊂ ¸X) which gives the claim.

Corollary 2.3. If X is a clique in the collinearity graph of a gamma space

Π = (P, L) and a point p ∈ P is collinear with all points of X then p is

collinear with all points of ¸X).

Using Zorn lemma we can show that every graph has maximal cliques

and every clique is contained in a certain maximal clique.

Proposition 2.7. In a gamma space the class of maximal singular sub-

spaces coincides with the class of maximal cliques of the collinearity graph.

Proof. By Proposition 2.6, every maximal clique X of the collinearity

graph is contained in the singular subspace ¸X). Let S be a maximal

singular subspace containing ¸X) (Exercise 1.5). Since S is a clique of the

collinearity graph, we have X = ¸X) = S.

Every maximal singular subspace U is contained in a certain maximal

clique Y of the collinearity graph. It was established above that Y is a

maximal singular subspace and we get U = Y .

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Chapter 3

Classical Grassmannians

It was mentioned in the previous chapter that every thick building of type

A

n−1

(n ≥ 4) is isomorphic to the ﬂag complex ∆(V ), where V is an n-

dimensional vector space over a division ring. The Grassmannians of ∆(V )

are the usual Grassmannians (

k

(V ). The associated Grassmann spaces will

be denoted by G

k

(V ); their elementary properties will be studied in Section

3.1. Note that G

1

(V ) = Π

V

and G

n−1

(V ) = Π

∗

V

.

Let V

**be other n-dimensional vector space over a division ring. Fol-
**

lowing our programme, we will examine collineations of G

k

(V ) to G

k

(V

)

and apartments preserving mappings of (

k

(V ) to (

k

(V

).

The classical Chow’s theorem will be proved in Section 3.2. It states

that every collineation of G

k

(V ) to G

k

(V

), 1 < k < n − 1, is induced by

a semilinear isomorphism of V to V

or a semilinear isomorphism of V to

V

∗

(the second possibility can be realized only in the case when n = 2k).

Also we characterize the adjacency relation in terms of the relation to be

opposite (as in Theorem 2.4) and show that every bijection of (

k

(V ) to

(

k

(V

**), 1 < k < n −1, preserving the opposite relation is a collineation of
**

G

k

(V ) to G

k

(V

).

If k = 1, n − 1 then apartments of (

k

(V ) are bases of Π

V

and Π

∗

V

,

respectively. In Section 3.3 we characterize apartments of (

k

(V ), 1 < k <

n − 1, in terms of the adjacency relation. This characterization follows

from a more general result concerning apartments in parabolic subspaces of

G

k

(V ). As an application, we describe all subspaces of G

k

(V ) isomorphic

to the Grassmann spaces of ﬁnite-dimensional vector spaces.

Section 3.4 is dedicated to apartments preserving mappings. All apart-

ments preserving bijections of (

k

(V ) to (

k

(V

) are collineations of G

k

(V )

to G

k

(V

). If 1 < k < n − 1 then every apartments preserving mapping

of (

k

(V ) to (

k

(V

) is induced by a semilinear embedding of V in V

or a

69

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

70 Grassmannians of Classical Buildings

semilinear embedding of V in V

∗

(as in Chow’s theorem, the second possi-

bility can be realized only in the case when n = 2k). There is an example

showing that the latter statement fails for k = 1, n −1.

In Sections 3.5–3.7 we demonstrate how our methods work for some

constructions rather similar to Grassmannians of ﬁnite-dimensional vector

spaces — spine spaces, Grassmannians of exchange spaces and the sets of

conjugate linear involutions. In particular, Chow’s theorem and the theo-

rem on apartments preserving bijections are closely related with classical

Dieudonn´e–Rickart’s results on automorphisms of the group GL(V ).

In Section 3.8 we extend some results to the case of inﬁnite-dimensional

vector spaces focusing on Grassmannians formed by linear subspaces with

inﬁnite dimension and codimension.

3.1 Elementary properties of Grassmann spaces

Let V be an n-dimensional vector space over a division ring and 3 ≤ n < ∞.

Let also k ∈ ¦1, . . . , n−1¦. Recall that two k-dimensional linear subspaces

of V (elements of the Grassmannian (

k

(V )) are adjacent if their intersection

is (k − 1)-dimensional; the latter is equivalent to the fact that the sum of

these linear subspaces is (k + 1)-dimensional.

Now suppose that M and N are incident linear subspaces of V and

dimM < k < dimN

(possible M = 0 or N = V ). We deﬁne

[M, N]

k

:= ¦ S ∈ (

k

(V ) : M ⊂ S ⊂ N ¦.

If M = 0 or N = V then instead of [M, N]

k

we will write

¸N]

k

or [M)

k

,

respectively. In the case when

dimM = k −1 and dimN = k + 1,

the set [M, N]

k

is a line of (

k

(V ). This line contains at least three points:

we choose two linearly independent vectors x, y ∈ N such that N is spanned

by M and x, y then

¸M, x), ¸M, y), ¸M, x +y)

are three distinct points on the line.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 71

The set of all such lines will be denoted by L

k

(V ). Two distinct elements

of (

k

(V ) are collinear (joined by a line) if and only if they are adjacent;

and for any adjacent S, U ∈ (

k

(V ) there is the unique line

[S ∩ U, S +U]

k

containing them. The pair

G

k

(V ) := ((

k

(V ), L

k

(V ))

is a partial linear space. It is clear that G

1

(V ) = Π

V

and G

n−1

(V ) = Π

∗

V

;

moreover,

G

k

(V

∗

) and G

n−k

(V )

are canonically isomorphic (by the annihilator mapping).

The partial linear spaces G

k

(V ), k ∈ ¦1, . . . , n − 1¦, are the Grass-

mann spaces of the building ∆(V ). The associated Grassmann graphs (the

collinearity graphs of G

k

(V )) will be denoted by Γ

k

(V ).

Recall that three distinct mutually collinear points of a partial linear

space form a triangle if they are not collinear (in other words, these points

span a plane).

Lemma 3.1. Let 1 < k < n−1. For any triangle S

1

, S

2

, S

3

in G

k

(V ) only

one of the following two possibilities is realized:

(1) a star-triangle: there is a (k−1)-dimensional linear subspace contained

in each S

i

,

(2) a top-triangle: there is a (k+1)-dimensional linear subspace containing

all S

i

.

Proof. The fulﬁllment of both (1) and (2) implies that S

1

, S

2

, S

3

are

collinear. Therefore, one of these conditions does not hold. Suppose that

S

3

is not contained in the (k+1)-dimensional linear subspace S

1

+S

2

. Then

the dimension of the linear subspace

U := (S

1

+S

2

) ∩ S

3

is not greater than k − 1. Since S

3

is adjacent with S

1

and S

2

, U is a

(k −1)-dimensional linear subspace contained in S

1

and S

2

.

By Proposition 2.1, the Grassmann space G

k

(V ) is connected. Now we

prove the following.

Proposition 3.1. The Grassmann space G

k

(V ) is a connected gamma

space. For any k-dimensional linear subspaces S, U ⊂ V the distance be-

tween S and U is equal to

dim(S +U) −k = k −dim(S ∩ U).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

72 Grassmannians of Classical Buildings

Proof. The case k = 1, n −1 is trivial. Let 1 < k < n −1. If S ∈ (

k

(V )

is adjacent with two distinct points S

1

, S

2

of a certain line [M, N]

k

and S

does not belong to this line then

M = S

1

∩ S

2

⊂ S or S ⊂ S

1

+S

2

= N

(S, S

1

, S

2

form a star-triangle or a top-triangle, respectively). In each of

these cases, S is adjacent with all points of the line [M, N]

k

. Thus the

axiom (Γ) holds (Section 2.5).

For any two linear subspaces S, U ⊂ V there is a base B ⊂ V such that

S and U are spanned by subsets of B (Proposition 1.4). Suppose that the

linear subspaces S and U both are k-dimensional. Let

X ∪ ¦x

1

, . . . , x

2m

¦

be a subset of B such that

S ∩ U = ¸X),

S = ¸X, x

1

, . . . , x

m

),

U = ¸X, x

m+1

, . . . , x

2m

).

The k-dimensional linear subspaces

S

i

:= ¸X, x

1+i

, . . . , x

m+i

), i = 0, . . . , m,

form a path in Γ

k

(V ) connecting S and U. For every path

S = U

0

, U

1

, . . . , U

l

= U

we have

dim(U

0

∩ ∩ U

l

) ≥ k −l.

On the other hand, U

0

∩ ∩ U

l

is contained in the linear subspace S ∩ U

whose dimension is equal to k − m. Thus k − l ≤ k − m and m ≤ l which

means that d(S, U) = m.

Remark 3.1 (Grassmann embedding). Suppose that V is a vector

space over a ﬁeld and consider the exterior power vector space ∧

k

V . If

vectors x

1

, . . . , x

k

and y

1

, . . . , y

k

span the same k-dimensional linear sub-

space of V then

x

1

∧ ∧ x

k

= a(y

1

∧ ∧ y

k

)

for a certain non-zero scalar a. This implies that the Grasmann mapping

g

k

: (

k

(V ) → (

1

(∧

k

V )

¸x

1

, . . . , x

k

) → ¸x

1

∧ ∧ x

k

)

is well-deﬁned. This mapping is injective and the image of every line of

G

k

(V ) is a line of Π

∧

k

V

. Thus g

k

is an embedding of the Grassmann space

G

k

(V ) in the projective space Π

∧

k

V

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 73

Now we describe maximal singular subspaces of G

k

(V ), 1 < k < n −1.

Since G

k

(V ) is a gamma space, the class of maximal singular subspaces of

G

k

(V ) coincides with the class of maximal cliques of the Grassmann graph

Γ

k

(V ) (Proposition 2.7).

Example 3.1. For every (k+1)-dimensional linear subspace N ⊂ V the set

¸N]

k

will be called a top. This is a singular subspace of G

k

(V ) isomorphic

to a k-dimensional projective space. If S ∈ (

k

(V ) does not belong to ¸N]

k

then

dim(S ∩ N) ≤ k −1

and there exist elements of ¸N]

k

which are not adjacent with S. This means

that ¸N]

k

is a maximal singular subspace of G

k

(V ). Every triangle of ¸N]

k

is a top-triangle.

Example 3.2. For every (k−1)-dimensional linear subspace M ⊂ V the set

[M)

k

is said to be a star. This is a singular subspace of G

k

(V ) isomorphic

to an (n − k)-dimensional projective space. Stars are maximal singular

subspaces, since the canonical collineation between G

k

(V ) and G

n−k

(V

∗

)

(the annihilator mapping) sends stars to tops and tops to stars. Every

triangle of a star is a star-triangle.

Proposition 3.2 ([Chow (1949)]). Every maximal singular subspace of

G

k

(V ), 1 < k < n −1, is a star or a top.

Proof. It was noted above that the class of maximal singular subspaces

coincides with the class of maximal cliques of the Grassmann graph. Hence

it is suﬃcient to show that every clique of the Grassmann graph is contained

in a star or a top.

Let A be a clique of Γ

k

(V ) and S

1

, S

2

be distinct elements of A. Suppose

that A is not contained in a star. In this case, there exists S

3

∈ A which

does not contain S

1

∩ S

2

and S

1

, S

2

, S

3

form a top-triangle. Then S

3

is

contained in the (k + 1)-dimensional linear subspace

N := S

1

+S

2

.

If a k-dimensional linear subspace S does not belong to the top ¸N]

k

then

dim(N ∩ S) ≤ k −1

and S is not adjacent with at least one of S

i

. Thus S ,∈ A and our clique

is a subset of ¸N]

k

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

74 Grassmannians of Classical Buildings

Example 3.3. Let M and N be incident linear subspaces of V satisfying

dimM < k < dimN

(possible M = 0 or N = V ). Then [M, N]

k

is a subspace of G

k

(V ); sub-

spaces of such type are called parabolic [Cooperstein, Kasikova and Shult

(2005)]. The parabolic subspace [M, N]

k

is isomorphic to the Grassmann

space G

k−m

(N/M), where m is the dimension of M. This subspace is

singular only in the case when M is (k − 1)-dimensional or N is (k + 1)-

dimensional. By Proposition 3.2, every singular subspace of G

k

(V ) is

parabolic. The canonical collineation between G

k

(V ) and G

n−k

(V

∗

) trans-

fers parabolic subspaces to parabolic subspaces.

The following result [Cooperstein, Kasikova and Shult (2005);

˙

Zynel

(2000)] generalizes Proposition 3.2. It will be proved in Section 3.3.

Theorem 3.1. Every subspace of G

k

(V ), 1 < k < n−1, isomorphic to the

Grassmann space of a ﬁnite-dimensional vector space is parabolic.

Let S and U be adjacent elements of (

k

(V ), 1 < k < n − 1. The set of

all elements of (

k

(V ) adjacent with both S, U is the union of all maximal

cliques of the Grassmann graph (all maximal singular subspaces of G

k

(V ))

containing S and U. By Proposition 3.2, this is

[S ∩ U)

k

∪ ¸S +U]

k

.

A k-dimensional linear subspace of V is adjacent with all elements of this

set if and only if it belongs to the line

[S ∩ U, S +U]

k

joining S and U.

Denote by A

∼

the set consisting of all elements of (

k

(V ) adjacent with

each element of a subset A ⊂ (

k

(V ). We get the following characterization

of lines in terms of the adjacency relation.

Proposition 3.3. For any adjacent S, U ∈ (

k

(V ), 1 < k < n −1, the line

joining S and U coincides with the subset ¦S, U¦

∼∼

.

The intersection of two distinct stars [M)

k

and [M

)

k

contains at most

one element; this intersection is not empty if and only if M, M

are adjacent

elements of (

k−1

(V ). The same holds for the intersection of two distinct

tops. The intersection of the star [M)

k

and the top ¸N]

k

is empty or a line;

the second possibility is realized only in the case when M ⊂ N.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 75

3.2 Collineations of Grassmann spaces

Throughout the section we suppose that V and V

**are n-dimensional vector
**

spaces over division rings and 4 ≤ n < ∞. We will study collineations

between the Grassmann spaces G

k

(V ) and G

k

(V

) for 1 < k < n −1.

3.2.1 Chow’s theorem

For every semilinear isomorphism u : V → V

the bijection

(u)

k

: (

k

(V ) → (

k

(V

)

(Subsection 1.3.2) is a collineation of G

k

(V ) to G

k

(V

). By duality, it can

be considered as the collineation of G

n−k

(V

∗

) to G

n−k

(V

∗

) induced by the

contragradient ˇ u (Subsection 1.3.3).

Let l : V → V

∗

be a semilinear isomorphism. By duality, we can

identify (l)

k

with the collineation of G

k

(V ) to G

n−k

(V

) sending each S to

l(S)

0

. This is a collineation of G

k

(V ) to G

k

(V

) if n = 2k.

Theorem 3.2 ([Chow (1949)]). Let 1 < k < n −1. Then every isomor-

phism of Γ

k

(V ) to Γ

k

(V

) is the collineation of G

k

(V ) to G

k

(V

) induced

by a semilinear isomorphism of V to V

or a semilinear isomorphism of V

to V

∗

; the second possibility can be realized only in the case when n = 2k.

Remark 3.2. In the cases when k = 1, n − 1, every bijection of (

k

(V ) to

(

k

(V

) is an isomorphism of Γ

k

(V ) to Γ

k

(V

).

It follows immediately from Theorem 3.2 that every collineation of

G

k

(V ) to G

k

(V

) is induced by a semilinear isomorphism of V to V

or

V

∗

. This statement generalizes the classical version of the Fundamental

Theorem of Projective Geometry (Corollary 1.2).

Proof. Let f be an isomorphism of Γ

k

(V ) to Γ

k

(V

). By Proposition

3.3, f is a collineation of G

k

(V ) to G

k

(V

). Then f and f

−1

map maximal

singular subspaces to maximal singular subspaces. Recall that maximal

singular subspaces of our Grassmann spaces are stars and tops. The in-

tersection of two distinct maximal singular subspaces is empty, or a single

point, or a line; the third possibility is realized only in the case when these

singular subspaces are of diﬀerent types (one of them is a star and the other

is a top) and the associated (k − 1)-dimensional and (k + 1)-dimensional

linear subspaces are incident.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

76 Grassmannians of Classical Buildings

Now suppose that for a (k −1)-dimensional linear subspace S ⊂ V the

image of the star [S)

k

is a star. Let U be a (k − 1)-dimensional linear

subspace adjacent with S. We choose a (k+1)-dimensional linear subspace

N containing S and U. The stars [S)

k

and [U)

k

intersect the top ¸N]

k

by

lines. Since [S)

k

goes to a star, the image of ¸N]

k

is a top; hence f([U)

k

)

is a star. By connectedness (Proposition 3.1), the same holds for every

(k −1)-dimensional linear subspace U ⊂ V . Similarly, we establish that all

stars go to tops if the image of a certain star of G

k

(V ) is a top. Therefore,

one of the following possibilities is realized:

(A) stars go to stars,

(B) stars go to tops.

The same arguments show that tops go to tops and tops go to stars in the

cases (A) and (B), respectively.

Case (A). In this case, there exists a bijection

f

k−1

: (

k−1

(V ) → (

k−1

(V

)

such that

f([S)

k

) = [f

k−1

(S))

k

for all S ∈ (

k−1

(V ). For every U ∈ (

k

(V )

S ∈ ¸U]

k−1

⇔ U ∈ [S)

k

⇔ f(U) ∈ [f

k−1

(S))

k

⇔ f

k−1

(S) ∈ ¸f(U)]

k−1

and we have

f

k−1

(¸U]

k−1

) = ¸f(U)]

k−1

.

Thus f

k−1

and the inverse mapping send tops to tops which implies that

f

k−1

is an isomorphism of type (A) between the Grassmann graphs Γ

k−1

(V )

and Γ

k−1

(V

**). Step by step, we get a sequence of such isomorphisms
**

f

i

: (

i

(V ) → (

i

(V

), i = k, . . . , 1,

where f

k

= f and

f

i

([S)

i

) = [f

i−1

(S))

i

for all S ∈ (

i−1

(V ) if i > 1. Then

f

i−1

(¸U]

i−1

) = ¸f

i

(U)]

i−1

(3.1)

for each U ∈ (

i

(V ); in particular, f

1

is a collineation of Π

V

to Π

V

. By the

Fundamental Theorem of Projective Geometry, f

1

is induced by a semilin-

ear isomorphism l : V → V

**. We can prove that f
**

i

= (l)

i

induction by i.

Indeed, if f

i−1

is induced by l then for every U ∈ (

i

(V ) we have

f

i−1

(¸U]

i−1

) = ¸l(U)]

i−1

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 77

and (3.1) gives the claim.

Case (B). Stars and tops are projective spaces of dimension n − k and

k, respectively. Since f is a collineation and preserves the dimensions of

singular subspaces, we have n = 2k. By duality, f can be considered as a

colleniation of G

k

(V ) to G

k

(V

∗

) sending stars to stars. Hence it is induced

by a semilinear isomorphism of V to V

∗

.

The following remarkable result is closely related with Chow’s theorem.

Theorem 3.3 ([Huang (1998)]). Let 1 < k < n − 1. If a surjection of

(

k

(V ) to (

k

(V

**) sends adjacent elements to adjacent elements then it is
**

a collineation of G

k

(V ) to G

k

(V

); in particular, every semicollineation of

G

k

(V ) to G

k

(V

) is a collineation.

3.2.2 Chow’s theorem for linear spaces

There is an analogue of Chow’s theorem for linear spaces. Let Π = (P, L)

be a linear space. Two distinct lines of Π are non-intersecting or have a

common point; in the second case, these lines are said to be adjacent. Every

pair of adjacent lines spans a plane. However, in contrast to the projective

case, there are planes containing pairs of non-adjacent lines (for example,

parallel lines in aﬃne planes). The Grassmann graph Γ

1

(Π) is the graph

whose vertex set is L and whose edges are pairs of adjacent lines. It is

not diﬃcult to prove that Γ

1

(Π) is connected and every maximal clique

of this graph is a star or a subset of a top (a star is formed by all lines

passing through a point, a top is the set of all lines contained in a plane).

Every collineation between linear spaces induces an isomorphism between

the associated Grassmann graphs; this isomorphism sends stars to stars.

Theorem 3.4 ([Havlicek (1999)]). Let Π = (P, L) and Π

= (P

, L

) be

linear spaces of the same ﬁnite dimension n ≥ 3. Suppose that f : L → L is

an isomorphism of Γ

1

(Π) to Γ

1

(Π

**). Then one of the following possibilities
**

is realized:

• The isomorphism f maps stars to stars and it is induced by a

collineation of Π to Π

.

• All tops are maximal cliques and f transfers stars to tops and tops to

stars. In this case, Π and Π

**both are 3-dimensional generalized pro-
**

jective spaces and f is induced by a collineation of Π to the generalized

projective space dual to Π

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

78 Grassmannians of Classical Buildings

Remark 3.3. A generalized projective space can be deﬁned as a linear space

where any two lines contained in a plane have a common point (if every

line of a generalized projective space contains at least three points then

it is a projective space). Every ﬁnite-dimensional generalized projective

space is the union of a ﬁnite collection of the following “components”:

points, lines containing more than two points, projective spaces; points from

distinct components are joined by lines of cardinality 2 [Buekenhout and

Cameron (1995)]. Suppose that Π = (P, L) is a generalized projective space

of dimension 3 and denote by P

∗

the set of all planes of Π. A subset of P

∗

is called a line if it consists of all planes containing a certain line L ∈ L; the

set of all such lines is denoted by L

∗

. Then Π

∗

:= (P

∗

, L

∗

) is a generalized

projective space of dimension 3; this generalized projective space is called

dual to Π. There is natural one-to-one correspondence between elements of

L and L

∗

. Since two lines of Π are adjacent if and only if the corresponding

lines of Π

∗

are adjacent, L

∗

can be identiﬁed with L and every collineation

of Π to Π

∗

induces an automorphism of the Grassmann graph Γ

1

(Π).

3.2.3 Applications of Chow’s theorem

Proposition 3.4. Let k, m ∈ ¦1, . . . , n −1¦ and k ,= m. Let also

g

k

: (

k

(V ) → (

k

(V

) and g

m

: (

m

(V ) → (

m

(V

)

be bijections satisfying the following condition: S ∈ (

k

(V ) and U ∈ (

m

(V )

are incident if and only if g

k

(S) and g

m

(U) are incident. Then there exists

a semilinear isomorphism l : V → V

such that

g

k

= (l)

k

and g

m

= (l)

m

,

in other words, g

k

and g

m

are induced by the same semilinear isomorphism

of V to V

.

Proof. By symmetry, we can restrict ourselves to the case when k < m.

It is clear that

g

k

(¸U]

k

) = ¸g

m

(U)]

k

for all U ∈ (

m

(V ). Let us deﬁne

(

k,m

(V ) :=

m

i=k

(

i

(V ) and (

k,m

(V

) :=

m

i=k

(

i

(V

).

For every U ∈ (

k,m

(V ) there exists g(U) ∈ (

k,m

(V

) such that

g

k

(¸U]

k

) = ¸g(U)]

k

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 79

Indeed, we choose m-dimensional linear subspaces U

1

, . . . , U

i

satisfying

U = U

1

∩ ∩ U

i

,

then

g(U) := g

m

(U

1

) ∩ ∩ g

m

(U

i

)

is as required. The mapping

g : (

k,m

(V ) → (

k,m

(V

)

is bijective and its restrictions to (

k

(V ) and (

m

(V ) coincide with g

k

and

g

m

, respectively. For all M, N ∈ (

k,m

(V ) we have

M ⊂ N ⇐⇒ g(M) ⊂ g(N);

in particular, g transfers (

k+1

(V ) to (

k+1

(V

**). This means that g
**

k

is a

collineation of Π

V

to Π

V

if k = 1. In the case when k > 1, the mapping g

k

is an isomorphism of Γ

k

(V ) to Γ

k

(V

) (since g

k

and the inverse mapping

send tops to tops). Therefore, g

k

is induced by a semilinear isomorphism

l : V → V

**; an easy veriﬁcation shows that g
**

m

= (l)

m

.

By duality, we have the following.

Corollary 3.1. Let k, m ∈ ¦1, . . . , n −1¦ and k ,= m. Let also

g

k

: (

k

(V ) → (

n−k

(V

) and g

m

: (

m

(V ) → (

n−m

(V

)

be bijections such that S ∈ (

k

(V ) and U ∈ (

m

(V ) are incident if and only

if g

k

(S) and g

m

(U) are incident. Then g

k

and g

m

are induced by the same

semilinear isomorphism of V to V

∗

.

Let Ω be a non-degenerated reﬂexive form deﬁned on V and ⊥ be the

associated orthogonal relation (Subsection 1.5.2). Consider the bijective

transformation of ((V ) sending each linear subspace S ⊂ V to S

⊥

. Suppose

that f is the restriction of this transformation to (

k

(V ). Then f is a

collineation of G

k

(V ) to G

n−k

(V ) satisfying the following condition:

(P) for any S, U ∈ (

k

(V ) the linear subspaces S, f(U) are incident if and

only if U, f(S) are incident.

If k = 1 then f is a polarity (Subsection 1.5.3). The following result

generalizes Proposition 1.13.

Theorem 3.5 ([Pankov 1 (2004)]). Let n ,= 2k and f be a bijection of

(

k

(V ) to (

n−k

(V ) satisfying the condition (P). Then there exists a non-

degenerate reﬂexive form Ω such that the restriction of the transformation

S → S

⊥

(where ⊥ is the orthogonal relation associated with Ω) to (

k

(V )

coincides with f.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

80 Grassmannians of Classical Buildings

Proof. Since k ,= n −k, we can apply Corollary 3.1 to the bijections

f : (

k

(V ) → (

n−k

(V ) and f

−1

: (

n−k

(V ) → (

k

(V ).

So, f and f

−1

both are induced by a semilinear isomorphism u : V → V

∗

.

Let us consider the bijective transformation h of ((V ) which sends every

linear subspace S to u(S)

0

. The restrictions of h to (

k

(V ) and (

n−k

(V )

coincide with f and f

−1

, respectively. Therefore, the restriction of h

2

to

(

k

(V ) is identity. This implies that h

2

is identity (we leave the details for

the reader). The latter means that the sesquilinear form deﬁned by u is

reﬂexive. This reﬂexive form is as required.

Remark 3.4. In the case when n = 2k, the condition (P) is equivalent to

the fact that f is an involution of (

k

(V ) (f

2

is identity) and the statement

given above fails.

3.2.4 Opposite relation

Let 1 < k < n −1. Two vertices of the Grassmann graph Γ

k

(V ) (elements

of the Grassmannian (

k

(V )) are said to be opposite if the distance between

them is maximal (equal to the diameter of the graph). By Proposition 3.1,

the diameter of Γ

k

(V ) is equal to

k if 2k ≤ n

n −k if 2k > n.

Therefore, in the case when 2k ≤ n, two elements of (

k

(V ) are opposite if

and only if their intersection is zero.

The adjacency relation can be characterized in terms of the relation to

be opposite.

Theorem 3.6. Let 1 < k < n − 1. For any distinct k-dimensional linear

subspaces S

1

, S

2

⊂ V the following conditions are equivalent:

(1) S

1

and S

2

are adjacent,

(2) there exists S ∈ (

k

(V ) ¸ ¦S

1

, S

2

¦ such that every U ∈ (

k

(V ) opposite

to S is opposite to at least one of S

i

.

Proof. Since the annihilator mapping of (

k

(V ) to (

n−k

(V

∗

) is an isomor-

phism between the associated Grassmann graphs, we can restrict ourselves

to the case when 2k ≤ n. In this case, two k-dimensional linear subspaces

are opposite if and only if their intersection is zero.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 81

(1) =⇒ (2). If S

1

and S

2

are adjacent elements of (

k

(Π) then every

S ∈ (

k

(V ) ¸ ¦S

1

, S

2

¦ belonging to the line joining S

1

and S

2

is as required.

Indeed, if U ∈ (

k

(V ) is opposite to S then it intersects S

1

+S

2

in a certain

1-dimensional linear subspace P; since S

1

∩ S

2

⊂ S, this linear subspace is

not contained S

1

∩ S

2

and

(S

1

∩ S

2

) +P

is a unique point of the line joining S

1

and S

2

which is not opposite to U.

(2) =⇒(1). The proof of this implication will be given in several steps.

First we establish that for any 1-dimensional linear subspaces P

1

⊂ S

1

and P

2

⊂ S

2

the sum P

1

+P

2

has a non-zero intersection with S.

Suppose that this intersection is zero. Then P

1

+ P

2

is contained in a

k-dimensional linear subspace opposite to S. By our hypothesis, this linear

subspace is opposite to at least one of S

i

. Thus there is P

i

which is not

contained in S

i

, a contradiction.

In particular, every 1-dimensional linear subspace P ⊂ S

1

∩ S

2

is con-

tained in S (we take P

1

= P

2

= P); hence

S

1

∩ S

2

⊂ S. (3.2)

Our second step is to show that

dim(S ∩ S

1

) = dim(S ∩ S

2

) = k −1. (3.3)

It is suﬃcient to establish that every 2-dimensional linear subspace con-

tained in S

i

(i = 1, 2) has a non-zero intersection with S.

Let us take a 2-dimensional linear subspace U ⊂ S

1

and a 1-dimensional

linear subspace P

2

⊂ S

2

which is not contained in S. By (3.2), P

2

is not

contained in S

1

; hence U+P

2

is 3-dimensional. Let P

1

and Q

1

be distinct 1-

dimensional linear subspaces of U. By the ﬁrst step of the proof, the linear

subspaces P

1

+ P

2

and Q

1

+ P

2

meet S in 1-dimensional linear subspaces

P and Q, respectively. It is clear that P ,= Q. Since U and P + Q are 2-

dimensional linear subspaces contained in the 3-dimensional linear subspace

U + P

2

, their intersection is non-zero. The inclusion P + Q ⊂ S gives the

claim.

Now we choose 1-dimensional linear subspaces P

1

⊂ S

1

and P

2

⊂ S

2

such that

S

i

= (S ∩ S

i

) +P

i

(this is possible by (3.3)). Then P

1

+ P

2

has a non-zero intersection with

S (by the ﬁrst step of the proof) and S +P

1

contains P

2

. Thus S

1

and S

2

both are contained in the (k +1)-dimensional linear subspace S +P

1

which

means that they are adjacent.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

82 Grassmannians of Classical Buildings

Corollary 3.2. Let 1 < k < n − 1. Every bijection of (

k

(V ) to (

k

(V

)

preserving the relation to be opposite (two elements of (

k

(V ) are opposite if

and only if their images are opposite) is a collineation of G

k

(V ) to G

k

(V

).

Remark 3.5. In the case when k = 2, n−2, two distinct elements of (

k

(V )

are adjacent or opposite and the latter statement is trivial.

Theorem 3.6 can be reformulated in the following form.

Theorem 3.7. Let 1 < k < n − 1. For any distinct k-dimensional linear

subspaces S

1

, S

2

⊂ V the following conditions are equivalent:

(1) S

1

and S

2

are adjacent,

(3) there exists S ∈ (

k

(V ) ¸ ¦S

1

, S

2

¦ such that every complement of S is a

complement to at least one of S

i

.

Proof. As in the proof of Theorem 3.6, we can restrict ourselves to the

case when 2k ≤ n. The implication (1) =⇒ (3) is obvious. Since an

element of (

k

(V ) is opposite to S ∈ (

k

(V ) if and only if it is contained in

a complement of S (this is true only in the case when 2k ≤ n), we have

(3) =⇒ (2) and Theorem 3.6 gives the claim.

Remark 3.6. Theorem 3.6 was ﬁrst proved in [Blunck and Havlicek (2005)]

for the case when n = 2k, and it was shown later [Havlicek and Pankov

(2005)] that Blunck–Havlicek’s method works for the general case. In

[Huang and Havlicek (2008)] the same characterization was obtained for

an abstract graph satisfying certain technical conditions; an easy veriﬁca-

tion shows that these conditions hold for the Grassmann graph Γ

k

(V ).

Remark 3.7. Let m be a positive integer which is less than the diameter

of Γ

k

(V ). For a subset A ⊂ (

k

(V ) we deﬁne

A

m

:= ¦ Y ∈ (

k

(V ) : d(X, Y ) ≤ m ∀ X ∈ A ¦.

By [Lim (2010)], for any S, U ∈ (

k

(V ) satisfying 0 < d(S, U) ≤ m one of the

following possibilities is realized: S, U are adjacent and (¦S, U¦

m

)

m

is the

line joining S with U, or they are non-adjacent and (¦S, U¦

m

)

m

coincides

with ¦S, U¦. As a consequence, we have the following result [Lim (2010)]:

if f : (

k

(V ) → (

k

(V

**) is a surjection such that
**

d(S, U) ≤ m ⇐⇒ d(f(S), f(U)) ≤ m

for any S, U ∈ (

k

(V ) then f is an isomorphism of Γ

k

(V ) to Γ

k

(V

).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 83

By the remarks given above, Theorem 3.6 is a partial case of more

general results. On the other hand, our proof of Theorem 3.6 is short and

it can be modiﬁed for Grassmannians of inﬁnite-dimensional vector spaces,

see Section 3.8. The same idea also will be exploited in Section 4.7. By

these reasons, we present this proof here.

3.3 Apartments

3.3.1 Basic properties

Let V be an n-dimensional vector space over a division ring and 3 ≤ n < ∞.

Let also B = ¦x

1

, . . . , x

n

¦ be a base of V . Denote by /

k

the associated

apartment of the Grassmannian (

k

(V ), it consists all k-dimensional linear

subspaces ¸x

i

1

, . . . , x

i

k

) and

[/

k

[ =

n

k

.

Remark 3.8. The apartment of (

k

(V ) deﬁned by a base B

⊂ V coincides

with /

k

if and only if the vectors of B

**are scalar multiples of the vectors
**

of B. Thus there is a one-to-one correspondence between apartments and

bases of the projective space Π

V

.

Exercise 3.1. Show that for any S, U ∈ (

k

(V ) the intersection of all apart-

ments of (

k

(V ) containing S and U coincides with ¦S, U¦.

The annihilator mapping transfers /

k

to the apartment of (

n−k

(V

∗

)

associated with the dual base B

∗

(Subsection 1.1.3).

The restriction of the Grassmann graph Γ

k

(V ) to a subset A ⊂ (

k

(V )

will be denoted by Γ(A). It is clear that Γ(/

k

) is isomorphic to the John-

son graph J(n, k) (the graph whose vertex set is formed by all k-element

subsets of ¦1, . . . , n¦ and two such subsets are connected by an edge if their

intersection consists of k − 1 elements). If k = 1, n − 1 then /

k

is a base

of Π

V

or Π

∗

V

, respectively. If 1 < k < n − 1 then the maximal cliques of

Γ(/

k

) are the intersections of /

k

with the stars

[M)

k

, M ∈ /

k−1

and the tops

¸N]

k

, N ∈ /

k+1

;

they will be called stars and tops of /

k

(respectively). Every maximal

clique of Γ(/

k

) is a base of the corresponding maximal singular subspace

of G

k

(V ).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

84 Grassmannians of Classical Buildings

Proposition 3.5. The Grassmann space G

k

(V ) is spanned by every apart-

ment of (

k

(V ).

Proof. The case k = 1 is trivial and we prove the statement induction by

k. Let B = ¦x

1

, . . . , x

n

¦ be a base of V and / be the associated apartment

of (

k

(V ). Denote by A the subspace of G

k

(V ) spanned by / and deﬁne

P

i

:= ¸x

i

) for every i. Since [P

i

)

k

can be identiﬁed with the Grassmann

space G

k−1

(V/P

i

), the inductive hypothesis implies that [P

i

)

k

is spanned

by /∩ [P

i

)

k

. Therefore,

[P

i

)

k

⊂ A

for every i. Now consider a k-dimensional linear subspace S ⊂ V such that

x

i

,∈ S for all i. We take any (k − 1)-dimensional linear subspace M ⊂ S

and deﬁne

S

i

:= M +P

i

for every i. Then S

1

, . . . , S

n

span the star [M)

k

(otherwise, S

1

+ +S

n

is

a proper subspace of V which contradicts the fact that B is a base). Every

S

i

belongs to A and we get S ∈ [M)

k

⊂ A which completes our proof.

Now we investigate apartments of parabolic subspaces (Example 3.3).

Let M and N be incident linear subspaces of V such that

dimM = m < k < l = dimN.

We take any base of V whose subsets span M and N; the intersection of

the associated apartment of (

k

(V ) with the parabolic subspace [M, N]

k

is

called an apartment of [M, N]

k

. The natural collineation of [M, N]

k

to the

Grassmann space G

k−m

(N/M) establishes a one-to-one correspondence be-

tween apartments of [M, N]

k

and apartments of (

k−m

(N/M). Proposition

3.5 guarantees that [M, N]

k

is spanned by every of its apartments. The

restrictions of the Grassmann graph Γ

k

(V ) to apartments of [M, N]

k

are

isomorphic to the Johnson graph J(l −m, k −m).

Theorem 3.8 ([Cooperstein, Kasikova and Shult (2005)]). Let 1 <

k < n−1 and l, m be natural numbers satisfying k < l ≤ n and 0 ≤ m < k.

Let also A be a subset of (

k

(V ) such that Γ(A) is isomorphic to the John-

son graph J(l −m, k −m) and every maximal clique of Γ(A) is an indepen-

dent subset of the Grassmann space G

k

(V ). Then A is an apartment in a

parabolic subspace of G

k

(V ); in particular, this is an apartment of (

k

(V )

if l = n and m = 0.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 85

The following example shows that the second condition in Theorem 3.8

(concerning the independence of maximal cliques) cannot be dropped.

Example 3.4. Let us take the following ﬁve vectors

(1, 0, 0, 0, 0)

(0, 1, 0, 0, 0)

(0, 0, 1, 0, 0)

(0, 0, 0, 1, 0)

(1, 1, 1, 1, 0)

in R

5

and consider the subset of (

2

(R

5

) consisting of all 2-dimensional

linear subspaces spanned by pairs of these vectors. The restriction of the

Grassmann graph Γ

2

(R

5

) to this subset is isomorphic to J(5, 2). However,

it is not an apartment of (

2

(R

5

).

Theorem 3.1 is a simple consequence of Theorem 3.8 and Proposition

3.5. Let f be a collineation of the Grassmann space G

m

(W) to a subspace

of G

k

(V ). We take any apartment / ⊂ (

m

(W). Then f(/) satisﬁes the

conditions of Theorem 3.8; hence it is an apartment in a parabolic subspace

of G

k

(V ). This parabolic subspace is spanned by f(/) and we get the claim.

3.3.2 Proof of Theorem 3.8

If l = k +1 or m = k −1 then any two distinct elements of A are adjacent

and the statement is trivial.

Suppose that k + 1 < l and m < k −1. Then A contains non-adjacent

elements. Let } be a maximal clique of Γ(A). Then } is contained in

precisely one maximal clique of Γ

k

(V ) (since } is an independent subset of

G

k

(V ) containing more than 2 elements and the intersection of two distinct

maximal cliques of Γ

k

(V ) does not contain a triangle). We say that } is a

star or a top of A if the maximal clique of Γ

k

(V ) containing } is a star or

a top, respectively.

We take an (l − m)-dimensional vector space W and any apartment

/ ⊂ (

k−m

(W). Let f : / → A be an isomorphism of Γ(/) to Γ(A). As in

the proof of Theorem 3.2, one of the following possibilities is realized:

(A) stars go to stars and tops go to tops,

(B) stars go to tops and tops go to stars.

In the case (B), we consider the apartment /

∗

⊂ (

l−k

(W

∗

) formed by

the annihilators of the elements from /; the mapping U → f(U

0

) is an

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

86 Grassmannians of Classical Buildings

isomorphism of Γ(/

∗

) to Γ(A) satisfying (A). By this reason, we restrict

ourselves to the case (A).

Every top } ⊂ A consists of k −m+ 1 elements; suppose that

} = ¦S

1

, . . . , S

k−m+1

¦.

Since this is an independent subset, the linear subspace

M

Y

:= S

1

∩ ∩ S

k−m+1

is m-dimensional, and for every i ∈ ¦1, . . . , k −m+ 1¦

X

i

:=

j=i

S

j

is an (m+1)-dimensional linear subspace. Denote by B

Y

the set formed by

all X

i

. This is a base of [M

Y

, T]

m+1

, where T is the (k + 1)-dimensional

linear subspace corresponding to }. Every element of } is the sum of k−m

elements of B

Y

.

Suppose that a top }

**⊂ A has a non-empty intersection with }. Then
**

the (k + 1)-dimensional linear subspaces corresponding to } and }

are

adjacent and } ∩ }

**consists of unique element S. For every U ∈ } ¸ ¦S¦
**

there exists unique U

∈ }

**¸ ¦S¦ adjacent with U (this follows from the
**

fact that Γ(A) is isomorphic to the Johnson graph J(l −m, k −m)). Since

S, U, U

**form a star-triangle, we have
**

S ∩ U = S ∩ U

.

This implies the following properties:

(1) M

Y

= M

Y

,

(2) if X ∈ B

Y

is contained in S then it belongs to B

Y

; thus B

Y

∩ B

Y

**consists of k − m elements and there is a unique element of B
**

Y

which

does not belong to B

Y

.

By connectedness of Grassmann spaces, M

Y

= M

Y

for any two tops

}, }

**⊂ A; in what follows this m-dimensional linear subspace will be
**

denoted by M.

For every S ∈ A we denote by B(S) the union of all B

Y

such that

S ∈ }. There are precisely l − k distinct tops of A containing S and, by

the property (2), the set B(S) consists of l − m elements. Now, we show

that B(S) coincides with B(U) if S, U ∈ A are adjacent.

There is unique top } ⊂ A containing S and U. By the deﬁnition,

B

Y

is contained in both B(S) and B(U). Consider X ∈ B(S) ¸ B

Y

. Let

}

**⊂ A be a top containing S and such that X ∈ B
**

Y

. We choose unique

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 87

U

∈ }

**adjacent with U and denote by Z the top of A containing U and
**

U

**. Since X is not contained in S (X ,∈ B
**

Y

), we have X ⊂ U

; by the

property (2), this implies X ∈ B

Z

⊂ B(U). Therefore, B(S) ⊂ B(U); the

same arguments give the inverse inclusion.

By connectedness,

B(S) = B(U) ∀ S, U ∈ A;

denote this set by B. Since B is formed by l−m distinct (m+1)-dimensional

linear subspaces containing M, the sum of all elements of B is a linear

subspace N whose dimension is not greater than l; also we have M ⊂ N.

On the other hand, every element of A is the sum of k −m elements from

B; and every star of A is an independent subset consisting of l − k + 1

elements. This implies that dimN ≥ l and N is l-dimensional. Then B is

a base of [M, N]

m+1

and A is the associated apartment of [M, N]

k

(we get

an apartment of (

k

(V ) if l = n and m = 0).

3.4 Apartments preserving mappings

Let V and V

**be n-dimensional vector spaces over division rings and n ≥ 3.
**

In this section we investigate apartments preserving mappings of (

k

(V ) to

(

k

(V

**). Recall that every apartments preserving mapping is injective.
**

3.4.1 Results

All collineations of G

k

(V ) to G

k

(V

**) (if they exist) are apartments preserv-
**

ing (since they are induced by semilinear isomorphisms). Conversely, we

have the following.

Theorem 3.9 (M. Pankov). Every apartments preserving bijection of

(

k

(V ) to (

k

(V

) is a collineation of G

k

(V ) to G

k

(V

).

Remark 3.9. If V is a vector space over a ﬁeld then every base ¦x

1

, . . . , x

n

¦

of V deﬁnes the regular base of Π

∧

k

V

consisting of all

¸x

i

1

∧ ∧ x

i

k

).

Clearly, the projective space Π

∧

k

V

has non-regular bases. The Grassmann

embedding g

k

(Remark 3.1) establishes a one-to-one correspondence be-

tween apartments of (

k

(V ) and regular bases of Π

∧

k

V

. Let f be a bijective

transformation of the projective space Π

∧

k

V

preserving the family of reg-

ular bases (f and f

−1

map regular bases to regular bases). A point of

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

88 Grassmannians of Classical Buildings

Π

∧

k

V

is contained in a regular base if and only if it belongs to the image of

(

k

(V ). Thus f transfers the image of (

k

(V ) to itself and the restriction of

f to this subset can be identiﬁed with an apartments preserving bijective

transformation of (

k

(V ).

Let l : V → V

**be a semilinear embedding. For every k ∈ ¦1, . . . , n−1¦
**

the mapping

(l)

k

: (

k

(V ) → (

k

(V

)

(Subsection 1.3.2) is an apartments preserving embedding of G

k

(V ) in

G

k

(V

). Recall that (l)

k

is bijective if and only if l is a semilinear iso-

morphism (Proposition 1.11).

Let u : V → V

∗

be a semilinear embedding. By duality, we can identify

(u)

k

with the embedding of G

k

(V ) in G

n−k

(V

) sending each S to ¸u(S))

0

.

In the case when n = 2k, this is an apartments preserving embedding of

G

k

(V ) in G

k

(V

).

Theorem 3.10 (M. Pankov). Let 1 < k < n − 1. Then every apart-

ments preserving mapping of (

k

(V ) to (

k

(V

) is the embedding of G

k

(V )

in G

k

(V

) induced by a semilinear embedding of V in V

or a semilinear

embedding of V in V

∗

; the second possibility can be realized only in the

case when n = 2k.

If k = 1, n − 1 then there exist apartments preserving mappings of (

k

(V )

to (

k

(V

**) which cannot be induced by semilinear mappings.
**

Example 3.5 ([Huang and Kreuzer (1995)]). Let us consider the in-

jective mapping α : R → (

1

(R

n

) deﬁned by the formula

t → ¸(t, t

2

, . . . , t

n

)).

For any distinct t

1

, . . . , t

n

∈ R ¸ ¦0¦ we have

t

1

. . . t

n

1

.

.

.

.

.

.

.

.

.

t

n

. . . t

n

n

,= 0.

Hence any subset of α(R) consisting of n distinct points is a base of the pro-

jective space Π

R

n; in particular, any three points of α(R) are non-collinear.

The sets R and (

1

(R

n

) have the same cardinality and we consider any bi-

jection of (

1

(R

n

) to α(R). This mapping transfers bases of Π

R

n to bases

of Π

R

n, but it is not induced by a semilinear mapping (since the images of

three distinct collinear points are non-collinear).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 89

In the case when 1 < k < n −1, Theorem 3.9 is a direct consequence of

Theorem 3.10 and Proposition 1.11.

Let f be an apartments preserving bijection of (

k

(V ) to (

k

(V

). Sup-

pose that k = 1. Then f maps bases of Π

V

to bases of Π

V

. Since three

points of a projective space are non-collinear if and only if they are con-

tained in a certain base of this space, the mapping f sends each triple of

non-collinear points to non-collinear points. This implies that the inverse

mapping transfers triples of collinear points to collinear points. Hence f

−1

is a semicollineation of Π

V

to Π

V

and Corollary 1.3 gives the claim. The

case k = n −1 is similar, by Exercise 1.11.

Remark 3.10. Theorem 3.9 was ﬁrst proved in [Pankov (2002)]. Our proof

of Theorem 3.10 is a modiﬁcation of the proof given in [Pankov 2 (2004);

Pankov 3 (2004)].

3.4.2 Proof of Theorem 3.10: First step

Let B = ¦x

1

, . . . , x

n

¦ be a base of V and / be the associated apartment of

(

k

(V ). Throughout the subsection we suppose that 1 < k ≤ n −k.

We write /(+i) and /(−i) for the sets consisting of all elements of

/ which contain x

i

and do not contain x

i

, respectively. If S is a linear

subspace spanned by a subset of B then we denote by /(S) the set of all

elements of / incident with S; in the case when S is spanned by x

i

and

x

j

, we will write /(+i, +j) instead of /(S). Clearly, /(S) coincides with

a certain /(−i) if S is (n −1)-dimensional.

A subset of / is called exact if it is contained only in one apartment of

(

k

(V ); otherwise, it is said to be inexact. It is trivial that ¹ ⊂ / is exact

if the intersection of all S ∈ ¹ containing x

i

coincides with ¸x

i

) for every i.

Lemma 3.2. A subset ¹ ⊂ / is exact if and only if for each i the inter-

section of all S ∈ ¹ containing x

i

coincides with ¸x

i

).

Proof. Suppose that for a certain number i there exists j ,= i such that

x

j

belongs to every S ∈ ¹ containing x

i

. We choose a vector x = ax

i

+bx

j

with a, b ,= 0. Then

(B ¸ ¦x

i

¦) ∪ ¦x¦

is a base of V which deﬁnes another apartment of (

k

(V ) containing ¹, thus

¹ is inexact. If there are no elements of ¹ containing x

i

then we can take

any vector x which is not a scalar multiple of x

i

and is not contained in the

linear subspace spanned by B ¸ ¦x

i

¦.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

90 Grassmannians of Classical Buildings

By Lemma 3.2,

/(−i) ∪ /(+i, +j), i ,= j,

is an inexact subset.

Lemma 3.3. If ¹ is a maximal inexact subset of / then

¹ = /(−i) ∪ /(+i, +j)

for some distinct i, j.

Proof. For each i there is an element of ¹ containing x

i

(if all elements

of ¹do not contain x

i

then ¹ is a subset of the non-maximal inexact subset

/(−i) which contradicts the fact that our inexact subset is maximal). By

Lemma 3.2, there exist distinct i and j such that x

j

belongs to all elements

of ¹ containing x

i

. Then every S ∈ ¹ is an element of /(−i) or /(+i, +j)

and

¹ ⊂ /(−i) ∪ /(+i, +j).

Since ¹ is a maximal inexact subset, we have the inverse inclusion.

A subset ¹ ⊂ / is said to be complement if /¸ ¹ is a maximal inexact

subset. In this case, Lemma 3.3 implies the existence of distinct i, j such

that

/¸ ¹ = /(−i) ∪ /(+i, +j).

Then

¹ = /(+i) ∩ /(−j);

in what follows this complement subset will be denoted by /(+i, −j).

We say that distinct complement subsets

¹

1

= /(+i

1

, −j

1

), . . . , ¹

k

= /(+i

k

, −j

k

)

form a regular collection if their intersection is a one-element set. In the

general case,

¹

1

∩ ∩ ¹

k

= /(M) ∩ /(N),

where M and N are linear subspaces spanned by subsets of B and

dimM = [¦i

1

, . . . , i

k

¦[ ≤ k ≤ n −k ≤ n − [¦j

1

, . . . , j

k

¦[ = dimN

(note that some of i

1

, . . . , i

k

or j

1

, . . . , j

k

can be coincident). This intersec-

tion is not empty if and only if

¦i

1

, . . . , i

k

¦ ∩ ¦j

1

, . . . , j

k

¦ = ∅; (3.4)

in this case, M is contained in N.

Example 3.6. If (3.4) holds and i

1

, . . . , i

k

are distinct then

¹

1

∩ ∩ ¹

k

= ¦¸x

i

1

, . . . , x

i

k

)¦

and the collection is regular.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 91

Example 3.7. Suppose that n = 2k, the condition (3.4) holds, and

j

1

, . . . , j

k

are distinct. Then

¹

1

∩ ∩ ¹

k

= ¦¸x

i

1

, . . . , x

i

k

)¦,

where

¦i

1

, . . . , i

k

¦ = ¦1, . . . , n¦ ¸ ¦j

1

, . . . , j

k

¦.

The collection is regular.

Lemma 3.4. The collection ¹

1

, . . . , ¹

k

is regular if and only if (3.4) holds

and one of the following possibilities is realized:

(A) i

1

, . . . , i

k

are distinct,

(B) n = 2k and j

1

, . . . , j

k

are distinct.

Proof. If (3.4) is fulﬁlled and the conditions (A) and (B) both do not

hold then

dimM < k < dimN

and /(M) ∩ /(N) contains more than one element.

A collection of k −1 distinct complement subsets ¹

1

, . . . , ¹

k−1

⊂ / is

said to be regular if it can be extended to a regular collection of k distinct

complement subsets; in other words, there exists a complement subset ¹

k

⊂

/ such that

¹

k

,= ¹

1

, . . . , ¹

k−1

and ¹

1

, . . . , ¹

k−1

, ¹

k

is a regular collection.

The adjacency relation can be characterized in terms of regular collec-

tions of complement subsets.

Lemma 3.5. The following conditions are equivalent:

(1) S, U ∈ / are adjacent,

(2) there exists a regular collection of k −1 distinct complement subsets of

/ such that each element of this collection contains S and U.

Proof. (1) =⇒ (2). Suppose that S and U are adjacent and

S ∩ U = ¸x

i

1

, . . . , x

i

k−1

).

We choose

x

i

∈ S ¸ U and x

j

,∈ S +U.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

92 Grassmannians of Classical Buildings

Then

/(+i

1

, −j) ∩ ∩ /(+i

k−1

, −j) ∩ /(+i, −j) = ¦S¦.

Hence these complement subsets form a regular collection and the comple-

ment subsets

/(+i

1

, −j), . . . , /(+i

k−1

, −j)

are as required.

(2) =⇒ (1). Conversely, suppose that each element of a regular collec-

tion

/(+i

1

, −j

1

), . . . , /(+i

k−1

, −j

k−1

)

contains S and U. By deﬁnition, this collection can be extended to a regular

collection of k distinct complement subsets; and it follows from Lemma 3.4

that one of the following possibilities is realized:

(A) i

1

, . . . , i

k−1

are distinct,

(B) n = 2k and j

1

, . . . , j

k−1

are distinct.

Since x

i

1

, . . . , x

i

k−1

belong to S ∩ U, (A) guarantees that the dimension of

S ∩ U is equal to k −1. The vectors x

j

1

, . . . , x

j

k−1

do not belong to S +U

and (B) implies that S + U is (k + 1)-dimensional. In each of these cases,

S and U are adjacent.

Proposition 3.6. Let 1 < k ≤ n −k and f : (

k

(V ) → (

k

(V

) be an apart-

ments preserving mapping. Then f is adjacency preserving: two elements

of (

k

(V ) are adjacent if and only if their images are adjacent.

Proof. First of all we recall that f is injective. Let S and U be k-

dimensional linear subspaces of V and / be an apartment of (

k

(V ) con-

taining them. Then f(/) is an apartment of (

k

(V

**) and f maps inexact
**

subsets of / to inexact subsets of f(/); indeed, if a subset ¹ is contained

in two distinct apartments of (

k

(V ) then f(¹) is contained in their images

which are distinct apartments of (

k

(V

). Since / and f(/) have the same

number of inexact subsets, every inexact subset of f(/) is the image of

a certain inexact subset of /. This implies that an inexact subset of /

is maximal if and only if its image is a maximal inexact subset of f(/).

Therefore, a subset ¹ ⊂ / is complement if and only if f(¹) is a com-

plement subset of f(/); moreover, f[

A

and the inverse mapping transfer

regular collections of complement subsets to regular collections (this follows

immediately from the deﬁnition of a regular collection). By Lemma 3.5, S

and U are adjacent if and only if f(S) and f(U) are adjacent.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 93

3.4.3 Proof of Theorem 3.10: Second step

By duality, every apartments preserving mapping f : (

k

(V ) → (

k

(V

)

can be identiﬁed with an apartments preserving mapping of (

n−k

(V

∗

) to

(

n−k

(V

∗

). Suppose that the latter mapping is induced by a semilinear

embedding u : V

∗

→ V

∗

and consider the mapping

g : ((V ) → ((V

)

S → ¸u(S

0

))

0

.

The restriction of g to (

k

(V ) coincides with f. The mapping g is dimension

preserving and we have

S ⊂ U =⇒ g(S) ⊂ g(U)

for all S, U ∈ ((V ). By Theorem 1.4, the restriction of g to (

1

(V ) is induced

by a semilinear injection l : V → V

**. Since this restriction is apartments
**

preserving, l is a semilinear embedding. For every linear subspace S ⊂ V

we choose P

1

, . . . , P

m

∈ (

1

(S), m = dimS such that

S = P

1

+ +P

m

.

Then

¸l(S)) = ¸l(P

1

)) + +¸l(P

m

)) = g(P

1

) + +g(P

m

) ⊂ g(S)

which implies that ¸l(S)) coincides g(S) (since the dimension of these linear

subspaces is equal to m). Thus g is induced by l and f = (l)

k

.

Therefore, we can prove the theorem only in the case when k ≤ n − k.

Let 1 < k ≤ n − k and f be an apartments preserving mapping of (

k

(V )

to (

k

(V

).

By Proposition 3.6, f is an adjacency preserving injection. Hence it

transfers stars and tops to subsets of stars or tops. For every star there is

an apartment intersecting this star in a set consisting of n−k +1 elements;

on the other hand, the intersections of tops with apartments contain at

most k + 1 elements. Since k ≤ n −k, the image of a star can be a subset

of a top only in the case when n = 2k. In particular, stars go to subsets of

stars if k < n −k.

By the same reason, the image of every maximal singular subspace of

G

k

(V ) is contained in precisely one maximal singular subspace of G

k

(V

).

Indeed, the intersection of two distinct maximal singular subspaces is

empty, or a single point, or a line; thus it intersects apartments in sub-

sets containing at most two elements.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

94 Grassmannians of Classical Buildings

Distinct maximal singular subspaces of G

k

(V ) go to subsets of distinct

maximal singular subspaces of G

k

(V

**) (otherwise there exist non-adjacent
**

elements of (

k

(V ) whose images are adjacent).

Suppose that f transfers stars to subsets of stars. Then, as in the proof

of Theorem 3.2, f induces an injective mapping

f

k−1

: (

k−1

(V ) → (

k−1

(V

)

such that

f([S)

k

) ⊂ [f

k−1

(S))

k

for all S ∈ (

k−1

(V ). Now we show that f

k−1

is apartments preserving.

Consider a base B ⊂ V and the associated apartment /

k

⊂ (

k

(V ). Let

B

be one of the bases of V

associated with the apartment /

k

:= f(/

k

).

Let also

/

k−1

⊂ (

k−1

(V ) and /

k−1

⊂ (

k−1

(V

)

be the apartments deﬁned by the bases B and B

, respectively. Every

S ∈ /

k−1

is the intersection of two adjacent U

1

, U

2

∈ /

k

and

f

k−1

(S) = f(U

1

) ∩ f(U

2

) ∈ /

k−1

.

Since the mapping f

k−1

is injective and [/

k−1

[ = [/

k−1

[ < ∞, we get

f

k−1

(/

k−1

) = /

k−1

.

Step by step, we construct a sequence of apartments preserving map-

pings

f

i

: (

i

(V ) → (

i

(V

), i = k, . . . , 1,

such that f

k

= f,

f

i

([S)

i

) ⊂ [f

i−1

(S))

i

for all S ∈ (

i−1

(V ) and

f

i−1

(¸U]

i−1

) ⊂ ¸f

i

(U)]

i−1

for all U ∈ (

i

(V ) if i > 1. By Theorem 1.4, f

1

is induced by a semilinear

injection l : V → V

. Since f

1

is apartments preserving, l is a semilinear

embedding.

We prove that f

i

= (l)

i

induction by i. If f

i−1

is induced by l then

f

i−1

(¸U]

i−1

) ⊂ ¸f

i

(U)]

i−1

∩ ¸¸l(U))]

i−1

for every U ∈ (

i

(V ). The latter intersection contains more that one element

which implies that f

i

(U) coincides with ¸l(U)).

In the case when the image of a certain star of (

k

(V ) is contained in

a top, we show that each star goes to a subset of a top (the standard

arguments from the proof of Theorem 3.2). Since n = 2k, we can identify

f with an apartments preserving mapping of (

k

(V ) to (

k

(V

∗

) which sends

stars to subsets of stars. The latter mapping is induced by a semilinear

embedding of V in V

∗

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 95

3.5 Grassmannians of exchange spaces

3.5.1 Exchange spaces

We say that a linear space Π = (P, L) satisﬁes the exchange axiom, or

simple, Π is an exchange space if for every subset X ⊂ P and any points

p, q ∈ P ¸ ¸X)

p ∈ ¸X, q) =⇒ q ∈ ¸X, p).

This axiom holds for projective and aﬃne spaces.

Theorem 3.11. In an exchange space every independent subset can be

extended to a base of this space, and any two bases of an exchange space

have the same cardinality.

Proof. See ¸8 in [Karzel, S¨ orensen and Windelberg (1973)].

Every semicollineation between projective spaces of the same ﬁnite di-

mension is a collineation (Corollary 1.3). The same holds for exchange

spaces.

Theorem 3.12 ([Kreuzer (1996)]). Let Π = (P, L) and Π

= (P

, L

)

be ﬁnite-dimensional exchange spaces such that

dimΠ ≤ dimΠ

. (3.5)

Then every semicollineation of Π to Π

is a collineation.

Proof. Let f be a semicollineation of Π to Π

. Consider a subset X ⊂ P

containing at least two points. First, we prove induction by i that

f([X]

i

) ⊂ [f(X)]

i

(Section 2.5). For i = 1 this is trivial; if the inclusion holds for i = k − 1

then

f([X]

k

) = f([[X]

k−1

]

1

) ⊂ [f([X]

k−1

)]

1

⊂ [[f(X)]

k−1

]

1

= [f(X)]

k

.

By Proposition 2.6, we have

f(¸X)) =

∞

i=1

f([X]

i

) ⊂

∞

i=1

[f(X)]

i

= ¸f(X)).

Therefore, if B is a base of Π then

f(¸B)) = f(P) = P

⊂ ¸f(B))

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

96 Grassmannians of Classical Buildings

and Π

**is spanned by f(B). The inequality (3.5) guarantees that f(B) is a
**

base of Π

**. Thus our spaces have the same dimension and f transfers bases
**

of Π to bases of Π

**. This implies that f maps any triple of non-collinear
**

points to non-collinear points (since three distinct points are non-collinear if

and only if there is a base containing them, Theorem 3.11). Then f

−1

sends

triples of collinear points to collinear points, hence f

−1

is a semicollineation

of Π

to Π.

Now suppose that our exchange spaces have the same ﬁnite dimension

and consider a mapping f : P → P

**which sends bases of Π to bases of
**

Π

**. Then f is injective and transfers any triple of non-collinear points to
**

non-collinear points (standard arguments). In the case when f is bijective,

the inverse mapping is a semicollineation of Π

to Π. Theorem 3.12 gives

the following.

Corollary 3.3. Let Π = (P, L) and Π

= (P

, L

**) be exchange spaces of the
**

same ﬁnite dimension. Then every bijection of P to P

sending bases of Π

to bases of Π

is a collineation of Π to Π

.

The following example shows that the condition (3.5) in Theorem 3.12

cannot be dropped.

Example 3.8 ([Kreuzer (1996)]). Let S be a plane in a 3-dimensional

projective space Π = (P, L). Consider the following set of lines

L

:= ¦S¦ ∪ ¦ L ∈ L : L ,⊂ S ¦;

in other words, we remove all lines contained in S and add S as a line.

An easy veriﬁcation shows that Π

:= (P, L

**) is an exchange plane. The
**

identity transformation of P is a semicollineation of Π to Π

which is not a

collineation.

Remark 3.11. A more complicated example of a semicollineation of a 4-

dimensional projective space to a non-Desarguesian projective plane was

given in [Ceccherini (1967)].

3.5.2 Grassmannians

Let Π = (P, L) be an exchange space of ﬁnite dimension n. For every

number k ∈ ¦0, . . . , n−1¦ we denote by (

k

(Π) the Grassmannian consisting

of all k-dimensional subspaces of Π. It follows from Theorem 3.11 that any

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 97

two incident k-dimensional subspaces of Π are coincident; note that for an

arbitrary linear space the same holds only in the case when k = 0, 1.

Two elements of the Grassmannian (

k

(Π) are said to be adjacent if their

intersection is (k −1)-dimensional.

Lemma 3.6. If S, U ∈ (

k

(Π) are adjacent then the subspace ¸S, U) is

(k + 1)-dimensional.

Proof. We take any base X of S ∩ U and any points p ∈ S ¸ U and

q ∈ U ¸ S. The exchange axiom guarantees that X ∪ ¦p¦, X ∪ ¦q¦, and

X ∪ ¦p, q¦ are bases of S, U, and ¸S, U) (respectively).

Remark 3.12. It was noted in Subsection 3.2.2 that there exist pairs of

non-adjacent elements of (

k

(Π) which span (k +1)-dimensional subspaces.

The Grassmann graph Γ

k

(Π) is the graph whose vertex set is (

k

(Π) and

whose edges are pairs of adjacent subspaces. It is not diﬃcult to prove that

this graph is connected and every star is a maximal clique (stars and tops

are deﬁned as for Grassmannians of ﬁnite-dimensional vector spaces).

Proposition 3.7. Every maximal clique of Γ

k

(Π) is a star or a subset of

a top.

Proof. Similar to the proof of Proposition 3.2, see [Pankov (2006)].

For index one Grassmann graphs we have an analogue of Chow’s the-

orem (Theorem 3.4); note that the exchange axiom is not required. For

larger indices such kind results are unknown.

Consider a certain base B of Π. The set consisting of all k-dimensional

subspaces spanned by subsets of B is called the base subset of (

k

(Π) asso-

ciated with (deﬁned by) the base B. In the case when Π is the projective

space of an (n + 1)-dimensional vector space V , the Grassmannian (

k

(Π)

can be identiﬁed with (

k+1

(V ) and base subsets are apartments.

Lemma 3.7. For any adjacent S, U ∈ (

k

(Π) there is a base subset of (

k

(Π)

containing S and U.

Proof. There exists an independent subset Y ⊂ P such that S and U

are spanned by subsets of Y (see the proof of Lemma 3.6). We extend Y

to a base of Π; the associated base subset of (

k

(Π) contains S and U.

Remark 3.13. Two k-dimensional subspaces S and U are contained in a

base subset of (

k

(Π) if and only if

dim¸S, U) = 2k −dim(S ∩ U),

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

98 Grassmannians of Classical Buildings

see [Pankov (2006)]. In the general (non-projective) case, this formula does

not hold and there exist pairs of k-dimensional subspaces which are not

contained in base subsets of (

k

(Π).

Theorem 3.13 ([Pankov (2006)]). Let Π = (P, L) and Π

= (P

, L

) be

exchange spaces of the same ﬁnite dimension n ≥ 2 such that each line of

Π and Π

**contains at least 3 points. Suppose that f : (
**

k

(Π) → (

k

(Π

) is

an injection sending base subsets to base subsets. Then f maps adjacent

elements to adjacent elements.

Proof. The case k = 0 is trivial (any two distinct points are adjacent).

If k = n − 1 then the required statement is a direct consequence of the

following observation: two distinct (n − 1)-dimensional subspaces of Π or

Π

**are adjacent if and only if there is a base subset containing them.
**

Let 0 < k < n − 1. Consider a base B = ¦p

1

, . . . , p

n+1

¦ of Π and

the associated base subset B ⊂ (

k

(Π). We write B(+i) and B(−i) for the

sets consisting of all elements of B which contain p

i

and do not contain p

i

,

respectively. We deﬁne exact, inexact, and complement subsets of B as for

apartments in Grassmannians of ﬁnite-dimensional vector spaces.

Using the fact that every line contains at least three points, we establish

the direct analogue of Lemma 3.2: a subset ¹ ⊂ B is exact if and only if

for every i the intersection of all S ∈ ¹ containing p

i

coincides with p

i

.

We describe maximal inexact subsets of B (cf. Lemma 3.3) and show that

every complement subset of B is

B(+i, −j) := B(+i) ∩ B(−j), i ,= j.

Now deﬁne

m := min¦k, n −k −1¦.

We say that m + 1 distinct complement subsets

B(+i

1

, −j

1

), . . . , B(+i

m+1

, −j

m+1

)

form a regular collection if their intersection is a one-element set. In this

case, one of the following three possibilities is realized (an analogue of

Lemma 3.4):

(1) m = k and i

1

, . . . , i

k+1

are distinct,

(2) m = k = n −k −1 and i

1

, . . . , i

k+1

or j

1

, . . . , j

k+1

are distinct,

(3) m = n −k −1 and j

1

, . . . , j

n−k

are distinct.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 99

A collection of m distinct complement subsets is said to be regular if it can

be extended to a regular collection of m + 1 distinct complement subsets.

We have the standard characterization of the adjacency relation in terms

of complement subsets (cf. Lemma 3.5); in other words, the following

conditions are equivalent:

• S, U ∈ B are adjacent,

• there exists a regular collection of m distinct complement subsets of B

such that each element of this collection contains S and U.

Now consider adjacent k-dimensional subspaces S and U of Π. By Lemma

3.7, there is a base subset B ⊂ (

k

(Π) containing them. The injectivity of f

guarantees that inexact subsets of B go to inexact subsets of f(B). As in

the proof of Theorem 3.10, we establish that a subset of B is complement

if and only if its image is a complement subset of f(B). Therefore, two

elements of B are adjacent if and only if the same holds for their images;

in particular, f(S) and f(U) are adjacent.

Remark 3.14. In the general case, there are pairs of k-dimensional sub-

spaces which are not contained in base subsets of (

k

(Π); thus we cannot

assert that a mapping of (

k

(Π) to (

k

(Π

**) sending base subsets to base sub-
**

sets is injective. By the same reason, the pre-images of adjacent elements

of (

k

(Π

**) need not to be adjacent, see Example 3.9.
**

Theorem 3.14 ([Pankov (2006)]). Let Π = (P, L) and Π

= (P

, L

) be

as in the previous theorem and f be a bijection of (

k

(Π) to (

k

(Π

) such that

f and f

−1

map base subsets to base subsets. Then f is an isomorphism of

Γ

k

(Π) to Γ

k

(Π

**); in the case when 2k+1 < n, it is induced by a collineation
**

of Π to Π

.

Proof. By Theorem 3.13, f is an isomorphism of Γ

k

(Π) to Γ

k

(Π

). There-

fore, f and f

−1

map maximal cliques to maximal cliques. Every maximal

clique is a star or a subset of a top (Proposition 3.7). In the case when

2k +1 < n, stars go to stars (for every star there is a base subset intersect-

ing this star in a subset consisting of n − k + 1 elements, the intersections

of base subsets with tops contain at most k +2 elements). Thus f induces

a bijection

f

k−1

: (

k−1

(Π) → (

k−1

(Π

).

As in the proof of Theorem 3.10, we show that f

k−1

maps base subsets to

base subsets. The inverse mapping is induced by f

−1

, hence it also sends

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

100 Grassmannians of Classical Buildings

base subsets to base subsets. We construct a sequence of bijections

f

i

: (

i

(Π) → (

i

(Π

), i = k, . . . , 0,

where f

k

= f and each f

i

together with the inverse mapping send base

subsets to base subsets. Then f

0

is a collineation of Π to Π

(Corollary 3.3)

and an easy veriﬁcation shows that f is induced by f

0

.

Remark 3.15. Since dual principle does not work for exchange spaces,

we cannot apply our method to the case when 2k + 1 ≥ n. However, if

n = 3, k = 1 and f is as in Theorem 3.14 then, by Theorem 3.4, one of the

following possibilities is realized:

• f is induced by a collineation of Π to Π

;

• Π and Π

**both are projective spaces and f is induced by a collineation
**

of Π to the projective space dual to Π

.

Example 3.9. Suppose that Π

**is a projective space and Π is the linear
**

space obtained from Π

**by removing a certain point p. It is not diﬃcult
**

to prove that Π is an exchange space and dimΠ = dimΠ

. Consider the

bijection of L to L

which sends every L ∈ L to the line of Π

containing

L (this mapping is induced by the natural embedding of Π in Π

). This

bijection transfers base subsets to base subsets. The inverse mapping does

not satisfy this condition, since it is not adjacency preserving (lines of Π

**passing through the point p go to non-intersecting lines of Π).
**

Remark 3.16. Every strong embedding of Π in Π

(an embedding sending

independent subsets to independent subsets) induces an injection of (

k

(Π)

to (

k

(Π

**) which transfers base subsets to base subsets. Using the arguments
**

given above, we can show that every injection f : L → L

transferring base

subsets to base subsets is induced by a strong embedding of Π in Π

[Pankov

(2006)]. In contrast to the projective case, the assumption that f is bijective

does not guarantee that this embedding is a collineation (Example 3.9).

3.6 Matrix geometry and spine spaces

Let V be an n-dimensional vector space over a division ring. Let also W be

an (n −k)-dimensional linear subspace of V . Consider the set A ⊂ (

k

(V )

consisting of all complements of W. There is a one-to-one correspondence

between elements of A and all k (n − k)-matrices over the associated

division ring: we take any M ∈ A and ﬁx bases in M and W, then every

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 101

S ∈ A can be considered as the graph of a linear mapping l : M → W, and

we identify S with the matrix of this linear mapping in the ﬁxed bases. The

distance between two matrices A and B is deﬁned as the rank of A−B; it is

equal to the distance between the corresponding elements of A. Information

concerning adjacency preserving transformations of the set of rectangular

matrices can be found in [Wan (1996)] (Theorem 3.4).

The construction considered above has a natural generalization

[Pra˙ zmowski (2001)]. We ﬁx a linear subspace W ⊂ V and denote by

T

k,m

(W) the set consisting of all S ∈ (

k

(V ) satisfying

dim(W ∩ S) = m

(we assume that m is not greater than dimW and k which guarantees that

the set T

k,m

(W) is not empty). The restriction of the Grassmann space

G

k

(V ) to this set (Subsection 1.2.1) is a partial linear space. Partial linear

spaces of such kind are called spine spaces. In the case when m = 0 and

dimW = n −k, we get the geometry of rectangular matrices. If m is equal

to dimW or k then our spine space is a parabolic subspace of G

k

(V ).

Suppose that 1 < k < n −1 and the line

[M, N]

k

, M ∈ (

k−1

(V ), N ∈ (

k+1

(V ),

induces a line in the spine space (the intersection of [M, N]

k

with T

k,m

(W)

contains at least two points). In the general case, there are the following

three possibilities:

(τ) dim(M ∩ W) = m, dim(N ∩ W) = m + 1;

(α) dim(M ∩ W) = dim(N ∩ W) = m;

(ω) dim(M ∩ W) = m−1, dim(N ∩ W) = m + 1.

The associated line of the spine space is said to be an x-line, x ∈ ¦τ, α, ω¦, if

the corresponding case is realized. For example, the geometry of rectangular

matrices (m = 0 and dimW = n −k) contains only τ-lines.

Every automorphism of the spine space can be extended to an automor-

phism of the Grassmann space G

k

(V ) [Pra˙ zmowski and

˙

Zynel (2002)].

Let B be a base of V such that W is spanned by a subset of B. The

intersection of the associated apartment of (

k

(V ) with T

k,m

(W) is said to

be the base subset of T

k,m

(W) deﬁned by (associated with) the base B.

Two distinct elements of T

k,m

(W) are called x-adjacent, x ∈ ¦τ, α, ω¦,

if they belong to a certain x-line. Two x-adjacent elements of T

k,m

(W) are

contained in a base subset if and only if x ∈ ¦α, ω¦.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

102 Grassmannians of Classical Buildings

In almost all cases, every bijective transformation of T

k,m

(W) preserv-

ing the class of base subsets preserves the α-adjacency and ω-adjacency re-

lations or interchanges them [Pankov, Pra˙ zmowski and

˙

Zynel (2005)]. The

proof of this statement is based on a characterization of the α-adjacency

and ω-adjacency relations in terms of complement subsets (the deﬁnitions

of exact, inexact, and complement subsets are standard).

3.7 Geometry of linear involutions

3.7.1 Involutions and transvections

Suppose that V is an n-dimensional left vector space over a division ring

R, the characteristic of R is not equal to 2 and n ≥ 3.

Let u be an involution of the group GL(V ) (u

2

= 1). Consider the linear

subspaces

S

+

(u) := Ker (1 −u) and S

−

(u) := Im(1 −u).

We have

u(x) = x if x ∈ S

+

(u) and u(x) = −x if x ∈ S

−

(u).

Since the intersection of these linear subspaces is 0 and the sum of their

dimensions is n, we get

S

+

(u) +S

−

(u) = V.

We say that u is a (k, n−k)-involution if the dimension of S

+

(u) and S

−

(u)

is equal to k and n −k, respectively.

Exercise 3.2. Let u be an involution of GL(V ) and x be a vector satisfying

u(x) ∈ ¸x). Show that x belongs to S

+

(u) or S

−

(u). Hint: there exist

unique vectors x

+

∈ S

+

(u) and x

−

∈ S

−

(u) such that x = x

+

+x

−

.

The set of all (k, n −k)-involutions will be denoted by 1

k,n−k

(V ). This

set can be identiﬁed with

(

k,n−k

(V ) := ¦ (S, U) ∈ (

k

(V ) (

n−k

(V ) : S +U = V ¦.

It is clear that v ∈ GL(V ) is conjugate to a (k, n−k)-involution u (there ex-

ists l ∈ GL(V ) such that v = lul

−1

) if and only if v is a (k, n−k)-involution.

Therefore, each 1

k,n−k

(V ) is a maximal set of conjugate involutions.

Lemma 3.8. Involutions u and v commute if and only if

u(S

+

(v)) = S

+

(v) and u(S

−

(v)) = S

−

(v).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 103

Proof. Suppose that uv = vu. If x ∈ S

+

(v) then

vu(x) = uv(x) = u(x)

and u(x) ∈ S

+

(v). Similarly, we show that the linear subspace S

−

(v) is

invariant for u.

Conversely, suppose that the linear subspaces S

+

(v) and S

−

(v) both

are invariant for u. Then uv(x) = vu(x) for every x belonging to S

+

(v)

or S

−

(v). Since V is the sum of these linear subspaces, the latter equality

holds for all x ∈ V .

Let B be a base of V . The base subset of 1

k,n−k

(V ) associated with

(deﬁned by) B is formed by all (k, n−k)-involutions u such that S

+

(u) and

S

−

(u) both are spanned by subsets of the base B. By Lemma 3.8, any two

elements of this subset commute.

Proposition 3.8. The class of maximal commutative subsets of 1

k,n−k

(V )

coincides with the class of base subsets.

Proof. Let u and v be commutative involutions. Then v(S

+

(u)) = S

+

(u)

and the restriction of v to the linear subspace S

+

(u) is an involution; denote

this involution by v

. It is clear that

S

+

(v

) = S

+

(v) ∩ S

+

(u) and S

−

(v

) = S

−

(v) ∩ S

+

(u);

hence

S

+

(v) ∩ S

+

(u) +S

−

(v) ∩ S

+

(u) = S

+

(u).

Similarly, we get the equality

S

+

(v) ∩ S

−

(u) +S

−

(v) ∩ S

−

(u) = S

−

(u)

which implies that

S

+

(v) ∩ S

+

(u) +S

−

(v) ∩ S

+

(u) +S

+

(v) ∩ S

−

(u) +S

−

(v) ∩ S

−

(u) = V.

The same arguments show that for every commutative subset A of

1

k,n−k

(V ) there exists a collection of linear subspaces S

1

, . . . , S

m

satisfying

the following conditions:

• S

1

+ +S

m

= V ,

• S

i

∩ S

j

= 0 if i ,= j,

• for every u ∈ A the linear subspaces S

+

(u) and S

−

(u) can be presented

as the sums of some S

i

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

104 Grassmannians of Classical Buildings

This means that every commutative subset of 1

k,n−k

(V ) is contained in a

certain base subset and every base subset is a maximal commutative subset

of 1

k,n−k

(V ).

Let t ∈ GL(V ). Suppose that Ker (1 − t) is (n − 1)-dimensional. Then

the dimension of Im(1 −t) is equal to 1 and

t(x) = x +α(x)x

0

,

where x

0

is a non-zero vector and α is a linear functional whose kernel

coincides with Ker (1 −t). If

Im(1 −t) ⊂ Ker (1 −t)

(in other words, x

0

∈ Ker α) then t is said to be a transvection. The

subgroup of GL(V ) generated by all transvections is denoted by SL(V ).

Remark 3.17. Note that α(x

0

) ,= −1 (since t(x

0

) ,= 0) and

t

−1

(x) = x −

α(x)

1 +α(x

0

)

x

0

.

If t is a transvection then t

−1

(x) = x −α(x)x

0

also is a transvection.

Remark 3.18. If l ∈ GL(V ) and t commute then

l(Ker (1 −t)) = Ker (1 −t) and l(Im(1 −t)) = Im(1 −t);

but these equalities do not guarantee that lt = tl.

3.7.2 Adjacency relation

Two distinct elements

(S, U), (S

, U

) ∈ (

k,n−k

(V )

are said to be adjacent if one of the following possibilities is realized:

• S = S

and U is adjacent with U

in (

n−k

(V ),

• U = U

and S is adjacent with S

in (

k

(V ).

The Grassmann graph Γ

k,n−k

(V ) is the graph whose vertex set is (

k,n−k

(V )

and whose edges are pairs of adjacent elements.

We use the following lemma to show that this graph is connected.

Lemma 3.9. For any distinct S, S

∈ (

k

(V ) there exists U ∈ (

n−k

(V )

which is a complement to both S and S

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 105

Proof. Consider an independent subset

X ∪ ¦x

1

, . . . , x

m

, y

1

, . . . , y

m

¦

such that

S ∩ S

= ¸X),

S = ¸X, x

1

, . . . , x

m

),

S

= ¸X, y

1

, . . . , y

m

).

If T is a complement to S +S

then

U := T +¸x

1

+y

1

, . . . , x

m

+y

m

)

is as required.

Proposition 3.9. The graph Γ

k,n−k

(V ) is connected.

Proof. Let (S, U) and (S

, U

) be distinct elements of (

k,n−k

(V ). First,

we consider the case when U = U

**and prove induction by the distance
**

d(S, S

) = k −dim(S ∩ S

)

that the graph Γ

k,n−k

(V ) contains a path connecting (S, U) with (S

, U).

This is trivial if d(S, S

) = 1. Suppose that d(S, S

) ≥ 2. We take any

(k −1)-dimensional linear subspace W satisfying

S ∩ S

⊂ W ⊂ S.

Then W + U ,= V and S

**is not contained in W + U. This implies the
**

existence of a 1-dimensional linear subspace P

⊂ S

such that S

:= W+P

**is a complement of U. We have
**

d(S

, S

) = d(S, S

) −1.

By the inductive hypothesis, there exists a path in Γ

k,n−k

(V ) connecting

(S

, U) with (S

, U). Since (S

**, U) is adjacent with (S, U), we get the
**

claim.

The case when S = S

is similar.

In the general case, we take any (n−k)-dimensional linear subspace U

satisfying

S +U

= S

+U

= V

(Lemma 3.9) and construct a path joining

(S, U), (S, U

), (S

, U

), (S

, U

).

**July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding
**

106 Grassmannians of Classical Buildings

For linear subspaces M and N we denote by [M, N]

k,n−k

the set of all

(S, U) ∈ (

k,n−k

(V ) such that S and U are incident to M and N, respec-

tively.

Proposition 3.10. Every maximal clique of Γ

1,n−1

(V ) is of type

[P, V ]

1,n−1

, P ∈ (

1

(V ) or [V, H]

1,n−1

, H ∈ (

n−1

(V ).

Similarly, every maximal clique of Γ

n−1,1

(V ) is of type

[V, P]

n−1,1

, P ∈ (

1

(V ) or [H, V ]

n−1,1

, H ∈ (

n−1

(V ).

In the case when 1 < k < n − 1, the Grassmann graph Γ

k,n−k

(V ) admits

only the following four types of maximal cliques:

• [S, M]

k,n−k

, S ∈ (

k

(V ), M ∈ (

n−k−1

(V ) and S ∩ M = 0;

• [S, N]

k,n−k

, S ∈ (

k

(V ), N ∈ (

n−k+1

(V ) and S +N = V ;

• [Q, U]

k,n−k

, Q ∈ (

k−1

(V ), U ∈ (

n−k

(V ) and Q∩ U = 0;

• [T, U]

k,n−k

, T ∈ (

k+1

(V ), U ∈ (

n−k

(V ) and T +U = V .

Proof. The case when k = 1, n − 1 is trivial. If 1 < k < n − 1 then the

statement easy follows from Proposition 3.2.

We say that two (k, n − k)-involutions are adjacent if the correspond-

ing elements of (

k,n−k

(V ) are adjacent. There is the following algebraic

interpretation of this adjacency relation.

Proposition 3.11. For involutions u, v ∈ 1

k,n−k

(V ) the following condi-

tions are equivalent:

(1) u and v are adjacent,

(2) uv and vu are transvections.

Proof. (1) =⇒ (2). Suppose that S

+

(u) = S

+

(v) (the case when

S

−

(u) = S

−

(v) is similar). Then

Ker (1 −uv) = Ker (1 −vu) = S

+

(u) + (S

−

(u) ∩ S

−

(v)) (3.6)

is (n−1)-dimensional. There exist a linear functional α whose kernel coin-

cides with (3.6) and a non-zero vector x

0

such that

uv(x) = x +α(x)x

0

. (3.7)

Then

v(x) = u(x) +α(x)u(x

0

) (3.8)

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 107

and

x = v

2

(x) = u(u(x) +α(x)u(x

0

)) +α(u(x) +α(x)u(x

0

))u(x

0

)

= x +α(x)x

0

+α(u(x) +α(x)u(x

0

))u(x

0

)

which means that u(x

0

) ∈ ¸x

0

). Then (3.8) implies that v(x

0

) ∈ ¸x

0

).

Using Exercise 3.2 we show that x

0

belongs to the linear subspace (3.6)

which coincides with Ker α. Therefore, uv is a transvection; then vu also

is a transvection (Remark 3.17).

(2) =⇒ (1). If uv is a transvection then (3.7) holds for a certain linear

functional α and a non-zero vector x

0

∈ Ker α. As above, we establish that

u(x

0

) ∈ ¸x

0

). By Exercise 3.2, x

0

belongs to S

+

(u) or S

−

(u).

Suppose that x

0

∈ S

+

(u). For every x ∈ S

+

(v)

u(x) = uv(x) = x +α(x)x

0

and

x = u

2

(x) = u(x) +α(x)u(x

0

) = x + 2α(x)x

0

.

This implies that S

+

(v) ⊂ Ker α. Since u(x) = v(x) for every x ∈ Ker α,

we have S

+

(v) = S

+

(u) and

S

−

(v) ∩ (Ker α) = S

−

(u) ∩ (Ker α)

is (n −k −1)-dimensional.

Suppose that x

0

∈ S

−

(u). For every x ∈ S

−

(v)

u(x) = −uv(x) = −x −α(x)x

0

.

As in the previous case, we get S

−

(v) ⊂ Ker α which implies that S

−

(v)

coincides with S

−

(u) and S

+

(v) is adjacent with S

+

(u).

Let A be a non-empty subset of 1

k,n−k

(V ). Denote by A

c

the set

consisting of all (k, n −k)-involutions commuting with every element of A.

If A

c

is non-empty then A

cc

contains A.

Lemma 3.10 ([Mackey (1942)]). Suppose that k = 1, n − 1. Then two

distinct (k, n−k)-involutions u, v are adjacent if and only if for any distinct

u

, v

∈ ¦u, v¦

cc

we have

¦u, v¦

cc

= ¦u

, v

¦

cc

. (3.9)

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

108 Grassmannians of Classical Buildings

Proof. We restrict ourselves to the case when k = 1 (the case k = n −1

is similar). Recall that every (k, n −k)-involution s is identiﬁed with

(S

+

(s), S

−

(s)) ∈ (

k,n−k

(V ).

Let u = (P, U) and v = (P

, U

). By Lemma 3.8, a (1, n − 1)-involution

(P

, U

**) commute with u if and only if
**

P ⊂ U

and P

⊂ U.

This implies that

¦u, v¦

c

= [U ∩ U

, P +P

]

1,n−1

and

¦u, v¦

cc

= [P +P

, U ∩ U

]

1,n−1

.

If u and v are adjacent (P = P

or U = U

) then any distinct u

, v

∈ ¦u, v¦

cc

are adjacent and (3.9) holds. In the case when P ,= P

and U ,= U

, the

equality (3.9) fails for any adjacent u

, v

∈ ¦u, v¦

cc

.

Problem 3.1. Is there a characterization of the adjacency relation in terms

of the commutativity relation for the case when 1 < k < n −1?

3.7.3 Chow’s theorem for linear involutions

We give a few examples.

Example 3.10. For every semilinear automorphism l : V → V the map-

ping

u → lul

−1

∀ u ∈ GL(V )

is an automorphism of the group GL(V ). It transfers each 1

k,n−k

(V ) to

itself. The associated transformation of (

k,n−k

(V ) is

(S, U) → (l(S), l(U)).

Example 3.11. Recall that the contragradient mapping u → ˇ u is an iso-

morphism of GL(V ) to GL(V

∗

). If u is a (k, n − k)-involution then the

same holds for ˇ u; moreover,

S

+

(ˇ u) = (S

−

(u))

0

and S

−

(ˇ u) = (S

+

(u))

0

(an exercise for the reader). For every semilinear isomorphism s : V → V

∗

the mapping

u → s

−1

ˇ us ∀ u ∈ GL(V )

is an automorphism of the group GL(V ). It transfers each 1

k,n−k

(V ) to

itself. The associated transformation of (

k,n−k

(V ) is

(S, U) → (s

−1

(U

0

), s

−1

(S

0

)).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 109

Automorphisms of the group GL(V ) considered in Examples 3.10 and

3.11 will be called regular.

Example 3.12. In the case when n = 2k, the mapping u → −u is a bijec-

tive transformation of 1

k,k

(V ). The associated transformation of (

k,k

(V )

sends (S, U) to (U, S).

A bijective transformation f of 1

k,n−k

(V ) is called commutativity pre-

serving if f and f

−1

map commutative involutions to commutative invo-

lutions; by Proposition 3.8, this is equivalent to the fact that f and f

−1

transfer base subsets to base subsets.

It is clear that all transformations considered above are commutativity

preserving and the associated transformations of (

k,n−k

(V ) are automor-

phisms of the Grassmann graph Γ

k,n−k

(V ). In the case when n = 2k, there

exist commutativity preserving transformations of 1

k,k

(V ) which do not

induce automorphisms of Γ

k,k

(V ).

Example 3.13. Let n = 2k. Consider any proper subset A ⊂ 1

k,k

(V )

satisfying the following condition:

u ∈ A =⇒ −u ∈ A.

The transformation

u →

−u ∀ u ∈ A

u ∀ u ,∈ A

is commutativity preserving; but the associated transformation of (

k,k

(V )

is not an automorphism of Γ

k,k

(V ) (we leave the details for the reader).

Theorem 3.15 ([Havlicek and Pankov (2005)]). If n ,= 2k then every

automorphism of Γ

k,n−k

(V ) is induced by a semilinear automorphism of V

(Example 3.10) or a semilinear isomorphism of V to V

∗

(Example 3.11).

Let n = 2k and f be an automorphism of Γ

k,k

(V ). Then

f = ig,

where g is the automorphism of Γ

k,k

(V ) induced by a semilinear automor-

phism of V or a semilinear isomorphism of V to V

∗

and i is the transfor-

mation sending every (S, U) to (U, S).

Remark 3.19. Some generalizations of this result were obtained in

[Pra˙ zmowski and

˙

Zynel (2009)].

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

110 Grassmannians of Classical Buildings

Remark 3.20. It must be pointed out that all results concerning the ad-

jacency relation on (

k,n−k

(V ) (Propositions 3.9, 3.10 and Theorem 3.15)

hold in the case of an arbitrary characteristic.

Corollary 3.4 ([Dieudonn´e 1 (1951); Rickart (1950)]). In the case

when k = 1, n − 1, every commutativity preserving bijective transforma-

tion of 1

k,n−k

(V ) can be extended to a regular automorphism of the group

GL(V ).

Proof. Let k = 1, n − 1 and f be a commutativity preserving bijective

transformation of 1

k,n−k

(V ). By Lemma 3.10, the associated transforma-

tion of (

k,n−k

(V ) is an automorphism of Γ

k,n−k

(V ). Theorem 3.15 gives

the claim.

Remark 3.21. Commutativity preserving transformations of 1

k,n−k

(V ),

1 < k < n −1, were investigated in [Pankov (2005)].

3.7.4 Proof of Theorem 3.15

Let f be an automorphism of the graph Γ

k,n−k

(V ). In what follows the

subsets

[S, V ]

k,n−k

, S ∈ (

k

(V ) and [V, U]

k,n−k

, U ∈ (

n−k

(V )

will be called special. Our ﬁrst step is to show that f and f

−1

map special

subsets to special subsets. In the cases when k = 1, n−1, the class of special

subsets coincides with the class of maximal cliques of Γ

k,n−k

(V ) and the

statement is trivial.

Suppose that 1 < k < n − 1. Let S ∈ (

k

(V ). Consider T ∈ (

n−k+1

(V )

satisfying S + T = V . Then [S, T]

k,n−k

is a maximal clique of Γ

k,n−k

(V )

and the same holds for f([S, T]

k,n−k

). By Proposition 3.10, one of the four

possibilities is realized. Assume that our cliques are of the same type; in

other words, there exist

W ∈ (

k

(V ) and Z ∈ (

n−k+1

(V )

such that W +Z = V and

f([S, T]

k,n−k

) = [W, Z]

k,n−k

.

Now we show that

f((S, U)) ∈ [W, V ]

k,n−k

∀ (S, U) ∈ [S, V ]

k,n−k

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 111

Let us take any

(S, U

0

) ∈ [S, T]

k,n−k

and suppose that (S, U) is adjacent with (S, U

0

). Then the linear subspace

P := U ∩ U

0

is (n − k − 1)-dimensional and [P, T]

n−k

is a line of the

Grassmann space G

n−k

(V ). This line contains at least three elements and

only one of them has a non-zero intersection with S. Thus the intersection

of [S, T]

k,n−k

and [S, P]

k,n−k

contains more than one element and the same

holds for their images

f([S, T]

k,n−k

) = [W, Z]

k,n−k

and f([S, P]

k,n−k

).

Since f([S, P]

k,n−k

) is a maximal clique of Γ

k,n−k

(V ), the linear subspace

W is the ﬁrst component in every element of f([S, P]

k,n−k

) and

f((S, U)) ∈ [W, V ]

k,n−k

.

By the connectedness of the graph Γ

k,n−k

(V ), the same holds for every

(S, U) ∈ [S, V ]

k,n−k

. Therefore,

f([S, V ]

k,n−k

) ⊂ [W, V ]

k,n−k

.

We apply the same arguments to the mapping f

−1

and get the inverse

inclusion. In the case when f([S, T]

k,n−k

) is a maximal clique of one of the

remaining three types, the proof is similar.

Similarly, we show that for every U ∈ (

n−k

(V ) the image of the special

subset [V, U]

k,n−k

is a special subset.

The intersection of two distinct special subsets is non-empty if and only

if these special subsets are of diﬀerent types and the sum of the associated

k-dimensional and (n−k)-dimensional linear subspaces coincides with V (it

is clear that this intersection is a one-element set). Since for any distinct

S

i

∈ (

k

(V ) (i = 1, 2) there exists U ∈ (

n−k

(V ) such that S

i

+ U = V

(Lemma 3.9), one of the following possibilities is realized:

(A) f preserves the type of every special subset,

(B) f changes the types of all special subsets.

Case (A). There exist bijective transformations f

and f

of (

k

(V ) and

(

n−k

(V ) (respectively) such that

f((S, U)) = (f

(S), f

(U)) ∀ (S, U) ∈ (

k,n−k

(V ).

Thus for any S ∈ (

k

(V ) and U ∈ (

n−k

(V )

S +U = V ⇐⇒ f

(S) +f

(U) = V.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

112 Grassmannians of Classical Buildings

The mappings f

and f

**are automorphisms of the Grassmann graphs
**

Γ

k

(V ) and Γ

n−k

(V ), respectively (this statement is trivial if k = 1, n − 1,

and it follows immediately from Theorem 3.7 if 1 < k < n −1).

Suppose that k = 1, n − 1. In this case, S ∈ (

k

(V ) and U ∈ (

n−k

(V )

are incident if and only if f

(S) and f

**(U) are incident. By Proposition
**

3.4, f

and f

**are induced by the same semilinear automorphism of V .
**

If 1 < k < n−1 then each of these mappings is induced by a semilinear

automorphism of V or a semilinear isomorphism of V to V

∗

(the second

possibility is realized only in the case when n = 2k). Suppose that f

is induced by a semilinear automorphism l : V → V . If f

is induced

by a semilinear automorphism s : V → V then for any S ∈ (

k

(V ) and

U ∈ (

n−k

(V )

S +U = V ⇐⇒ S +l

−1

s(U) = V ;

this implies that l

−1

(s(U)) = U for every U ∈ (

n−k

(V ) and

f((S, U)) = (l(S), l(U)) ∀ (S, U) ∈ (

k,n−k

(V ).

Now, suppose that n = 2k and f

**is induced by a semilinear isomorphism
**

s : V → V

∗

. As above, we establish that

(sl

−1

(U))

0

= U ∀ U ∈ (

k

(V ).

Let W ∈ (

k−1

(V ). We choose

U

1

, . . . , U

k+1

∈ (

k

(V )

such that

U

1

∩ ∩ U

k+1

= W and U

1

+ +U

k+1

= V.

Then

0 = (sl

−1

(V ))

0

=

k+1

i=1

(sl

−1

(U

i

))

0

=

k+1

i=1

U

i

= W.

The equality W = 0 implies that k = 1 which contradicts the assumption

that n = 2k ≥ 3.

If n = 2k and f

**is induced by a semilinear isomorphism u : V → V
**

∗

then the same arguments show that f

**also is induced by u; thus f is the
**

composition of the transformations

(S, U) → (u(U)

0

, u(S)

0

) and (S, U) → (U, S).

Case (B). There exist bijective mappings

g

: (

k

(V ) → (

n−k

(V ) and g

: (

n−k

(V ) → (

k

(V )

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 113

such that

f((S, U)) = (g

(U), g

(S)) ∀ (S, U) ∈ (

k,n−k

(V ).

By duality, these mappings can be considered as bijections of (

k

(V ) and

(

n−k

(V ) to (

k

(V

∗

) and (

n−k

(V

∗

), respectively. As in the case (A), we

establish that one of the following possibilities is realized: (i) g

and g

are

induced by the same similinear isomorphism u : V → V

∗

and

f((S, U)) = (u(U)

0

, u(S)

0

) ∀ (S, U) ∈ (

k,n−k

(V );

(ii) n = 2k, the mappings g

and g

**are induced by the same similinear
**

automorphism l : V → V

**and f is the composition of the transformations
**

(S, U) → (l(S), l(U)) and (S, U) → (U, S).

3.7.5 Automorphisms of the group GL(V )

Theorem 3.16 ([Dieudonn´e 1 (1951); Rickart (1950)]). If f is an

automorphism of the group GL(V ) then

f(u) = α(u)g(u) ∀ u ∈ GL(V ),

where g is a regular automorphism and α is a homomorphism of GL(V ) to

the center of R.

Proof. The automorphism f preserves the set of all involutions; more-

over, it transfers 1

k,n−k

(V ) to 1

m,n−m

(V ) (because each 1

k,n−k

(V ) can be

characterized as a maximal set of conjugate involutions). By Proposition

3.8, base subsets of 1

k,n−k

(V ) go to base subsets of 1

m,n−m

(V ). Since a

base subset of 1

k,n−k

(V ) consists of

n

k

elements and

n

k

=

n

m

⇐⇒ m = k, n −k,

the image of 1

k,n−k

(V ) coincides with 1

k,n−k

(V ) or 1

n−k,k

(V ). Therefore,

the restriction of f to the set 1

1,n−1

(V ) is a commutativity preserving

bijection to 1

1,n−1

(V ) or 1

n−1,1

(V ). By Corollary 3.4, there exists a regular

automorphism g of GL(V ) such that

f(u) = g(u) ∀ u ∈ 1

1,n−1

(V ) or f(u) = −g(u) ∀ u ∈ 1

1,n−1

(V ).

For every transvection t we take adjacent (1, n−1)-involutions u, v satisfying

t = uv and get f(t) = g(t) in each of these cases. So, f and g are coincident

on SL(V ) and h := g

−1

f sends every element of SL(V ) to itself.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

114 Grassmannians of Classical Buildings

If t is a transvection then ltl

−1

is a transvection for any l ∈ GL(V ) and

h(l)th(l

−1

) = ltl

−1

which means that α(l) := l

−1

h(l) commutes with every transvection. By

Remark 3.18, α(l) preserves every 1-dimensional linear subspace; hence it

is a homothetic transformation. All linear homothetic transformations of

V form the center of the group GL(V ) (this subgroup is isomorphic to the

center of R). The mapping l → α(l) is a homomorphism of GL(V ) to the

center. An easy veriﬁcation shows that f(l) = α(l)g(l).

3.8 Grassmannians of inﬁnite-dimensional vector spaces

Let V and V

**be vector spaces over division rings. Throughout the section
**

we suppose that dimV = dimV

**= α is inﬁnite. In this case, we have the
**

following three types of Grassmannians associated with V :

(

β

(V ) = ¦ S ∈ ((V ) : dimS = β, codimS = α ¦,

(

β

(V ) = ¦ S ∈ ((V ) : dimS = α, codimS = β ¦

for every cardinality β < α and

(

α

(V ) = (

α

(V ) = ¦ S ∈ ((V ) : dimS = codimS = α ¦.

They are orbits of the action of the group GL(V ) on the set ((V ).

3.8.1 Adjacency relation

Let ( be a Grassmannian of V . We say that linear subspaces S, U ∈ ( are

adjacent if

dim(S/(S ∩ U)) = dim(U/(S ∩ U)) = 1

which is equivalent to

dim((S +U)/S) = dim((S +U)/U) = 1.

We deﬁne the associated Grassmann graph as in the ﬁnite-dimensional

case. The Grassmann graphs corresponding to (

β

(V ), (

β

(V ), and (

α

(V )

will be denoted by Γ

β

(V ), Γ

β

(V ), and Γ

α

(V ) (respectively). Semilinear

isomorphisms of V to V

**induce isomorphisms between the Grassmann
**

graphs of the same indices.

Let k ∈ N. The Grassmann graph Γ

k

(V ) is connected and every iso-

morphism of Γ

k

(V ) to Γ

k

(V

**) is induced by a semilinear isomorphism of
**

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 115

V to V

**(the proof is similar to the proof of Theorem 3.2 and we leave it
**

as an exercise for the reader). Since the annihilator mapping deﬁnes an

isomorphism of Γ

k

(V ) to Γ

k

(V

∗

) (see Subsection 1.1.3), the Grassmann

graph Γ

k

(V ) is connected.

The following example shows that the direct analogue of Chow’s theo-

rem does not hold for Grassmannians formed by linear subspaces of inﬁnite

dimension and codimension.

Example 3.14 ([Blunck and Havlicek (2005)]). Let S ∈ (

β

(V ) and

β be inﬁnite. Denote by A the connected component of Γ

β

(V ) containing

S; it consists of all X ∈ (

β

(V ) such that

dim(S/(S ∩ X)) = dim(X/(S ∩ X))

is ﬁnite. We take any S

**∈ A ¸ ¦S¦. There exists U ∈ (
**

β

(V ) ¸ A such that

U ∩ S = 0 and U ∩ S

,= 0.

Let l : V → V be a semilinear automorphism sending S to S

. The associ-

ated automorphism of Γ

β

(V ) transfers A to itself. We deﬁne

f(X) :=

l(X) X ∈ A

X X ∈ (

β

(V ) ¸ A.

This mapping is an automorphism of Γ

β

(V ), but it is not induced by a

semilinear automorphism of V . Indeed, U ∩ S = 0 and f(U) = U has a

non-zero intersection with f(S) = S

.

Problem 3.2. Let A be a connected component of Γ

β

(V ), β < α, and f be

an automorphism of Γ

β

(V ). Is there a semilinear automorphism l : V → V

such that f(S) = l(S) for all S ∈ A?

There is the following inﬁnite-dimensional version of Theorem 3.7.

Theorem 3.17 ([Blunck and Havlicek (2005)]). Let ( be a Grass-

mannian of V . For any distinct S

1

, S

2

∈ ( the following conditions are

equivalent:

(1) S

1

and S

2

are adjacent,

(2) there exists S ∈ ( ¸ ¦S

1

, S

2

¦ such that every complement of S is a

complement to at least one of S

i

.

For every S ∈ (

α

(V ) all complements of S belong to (

α

(V ) (for other

Grassmannians this fails) and we have the following.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

116 Grassmannians of Classical Buildings

Corollary 3.5. Suppose that f : (

α

(V ) → (

α

(V

) is a bijection preserving

the complementary of subspaces: S ∈ (

α

(V ) is a complement of U ∈ (

α

(V )

if and only if f(S) is a complement of f(U). Then f is an isomorphism of

Γ

α

(V ) to Γ

α

(V

).

3.8.2 Proof of Theorem 3.17

If ( = (

1

(V ), (

1

(V ) then any two distinct elements of ( are adjacent and

the implication (2) =⇒ (1) is trivial.

(1) =⇒ (2). If S

1

and S

2

are adjacent then any S ∈ ( ¸ ¦S

1

, S

2

¦

satisfying

S

1

∩ S

2

⊂ S ⊂ S

1

+S

2

is as required (see the proof of Theorem 3.6).

(2) =⇒(1). Suppose that S ∈ (¸¦S

1

, S

2

¦ satisﬁes the condition (2). As

in the proof of Theorem 3.6, we establish that for any 1-dimensional linear

subspaces P

1

⊂ S

1

, P

2

⊂ S

2

the sum P

1

+ P

2

has a non-zero intersection

with S and

S

1

∩ S

2

⊂ S. (3.10)

Now we show that

S ⊂ S

1

+S

2

. (3.11)

This inclusion is trivial if S

1

+ S

2

= V . In the case when S

1

+ S

2

,= V ,

it is suﬃcient to show that S is contained in every H ∈ (

1

(V ) containing

S

1

+ S

2

(since the intersection of all such linear subspaces coincides with

S

1

+S

2

). If S ,⊂ H ∈ (

1

(V ) then H+S = V and H contains a complement

U of S; if S

1

+ S

2

⊂ H then U is not a complement of S

i

(i = 1, 2) which

contradicts (2).

The linear subspaces S

1

and S

2

are not incident. Indeed, if S

1

⊂ S

2

then (3.10) and (3.11) imply that

S

1

⊂ S ⊂ S

2

and every complement of S is not a complement of S

i

(i = 1, 2); similarly,

we get S

2

,⊂ S

1

.

Our next step is to show that

S

i

,⊂ S, i = 1, 2.

Suppose, for example, that S

1

⊂ S. Since S

1

,⊂ S

2

, there exists H ∈ (

1

(V )

which contains S

2

and does not contain S

1

. We have

V = S

1

+H ⊂ S +H;

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 117

thus S + H = V and H contains a certain complement U of S. Then U

is not a complement of S

1

(S

1

is a proper subspace of S) and it is not a

complement of S

2

(U and S

2

both are contained in H), a contradiction.

We take any 2-dimensional linear subspace U ⊂ S

1

and a 1-dimensional

linear subspace of S

2

which is not contained in S. As in the proof of

Theorem 3.6, we establish that U has a non-zero intersection with S ∩ S

1

.

This means that

dim(S

1

/(S ∩ S

1

)) = 1

and the same holds for S

2

. Thus there exist 1-dimensional linear subspaces

P

1

⊂ S

1

and P

2

⊂ S

2

such that

S

i

= (S ∩ S

i

) +P

i

, i = 1, 2.

Note that P

1

,= P

2

(otherwise P

1

= P

2

⊂ S

1

∩ S

2

⊂ S and both S

1

, S

2

are

contained in S).

The linear subspace P

1

+P

2

intersects S in a 1-dimensional linear sub-

space P. We have

S

1

+S

2

⊂ S +P

1

+P

2

= S +P +P

1

= S +P

1

⊂ S

1

+S

2

which implies that

S

1

+S

2

= S +P

1

. (3.12)

By the same arguments,

S

1

+S

2

= S +P

2

. (3.13)

Let U be a complement of S

1

+S

2

. By (3.12), U +P

1

is a complement

of S. Since P

1

⊂ S

1

, the linear subspace U +P

1

is not a complement of S

1

;

hence it is a complement of S

2

. So,

U + (S

1

+S

2

) = V = U + (S

2

+P

1

)

and the inclusion S

2

+P

1

⊂ S

1

+S

2

implies that

S

1

+S

2

= S

2

+P

1

.

Using the equality (3.13), we establish that

S

1

+S

2

= S

1

+P

2

.

Therefore,

dim((S

1

+S

2

)/S

1

) = dim((S

1

+S

2

)/S

2

) = 1

which means that S

1

and S

2

are adjacent.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

118 Grassmannians of Classical Buildings

3.8.3 Base subsets

Let ( be a Grassmannian of V and B be a base of V . The base subset

of ( associated with (deﬁned by) B consists of all elements of ( spanned

by subsets of B. For any two elements of ( there exists a base subsets

containing them (Proposition 1.4).

Every base subset of (

1

(V ) is a base of the projective space Π

V

and

conversely. Base subsets of (

1

(V ) are not bases of the dual projective

space Π

∗

V

= Π

V

∗, since dimV < dimV

∗

(Subsection 1.1.3).

The mappings between Grassmannians of the same indices induced by

semilinear isomorphisms of V to V

**transfer base subsets to base subsets.
**

We prove the following generalization of Theorem 3.9.

Theorem 3.18 ([Pankov 1 (2007)]). Let f be a bijection of (

β

(V ) to

(

β

(V

**), β < α, such that f and f
**

−1

map base subsets to base subsets.

Then f is induced by a semilinear isomorphism of V to V

.

3.8.4 Proof of Theorem 3.18

If β = 1 then, as in Subsection 3.4.1, we show that f and f

−1

are semi-

collineations between the projective spaces associated with V and V

; thus

f is a collineation of Π

V

to Π

V

and the Fundamental Theorem of Projective

Geometry gives the claim.

Let ( be a Grassmannian of V distinct from (

1

(V ) and (

1

(V ). Let

also B = ¦x

i

¦

i∈I

be a base of V and B be the associated base subset of (.

Exact, inexact, and complement subsets of B are deﬁned as for apartments

in Grassmannians of ﬁnite-dimensional vector spaces. We write B(+i) and

B(−i) for the sets of all elements of B which contain x

i

and do not contain

x

i

, respectively. The subsets B(+i) and B(−i) will be called simple subsets

of ﬁrst and second type, respectively. We also deﬁne

B(+i, +j) := B(+i) ∩ B(+j) and B(+i, −j) := B(+i) ∩ B(−j)

for all i, j ∈ I.

As in Subsection 3.4.2, we establish that for every maximal inexact

subset A ⊂ B there exist distinct i, j ∈ I such that

A = B(+i, +j) ∪ B(−i).

The associated complement subset is B(+i, −j). Note that every comple-

ment subset is the intersection of two simple subsets of diﬀerent types.

For two distinct complement subsets B(+i, −j) and B(+i

, −j

) one of

the following possibilities is realized:

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 119

(1) i = i

or j = j

,

(2) i = j

or j = i

**, then the intersection of the complement subsets is
**

empty,

(3) ¦i, j¦ ∩ ¦i

, j

¦ = ∅.

In the ﬁrst case, our complement subsets are said to be adjacent. An easy

veriﬁcation shows that for two distinct complement subsets A, } ⊂ B with

a non-empty intersection the following two conditions are equivalent:

• A and } are adjacent,

• for any distinct complement subsets A

, }

⊂ B satisfying

A ∩ } ⊂ A

∩ }

**the inverse inclusion holds.
**

In other words, the intersection of two distinct complement subsets is max-

imal if and only if these complement subsets are adjacent.

A collection of mutually adjacent complement subsets of B will be called

an A-collection. For each i ∈ I

¦B(+i, −j)¦

j∈I\{i}

and ¦B(+j, −i)¦

j∈I\{i}

are maximal A-collections. It is easy to see that every maximal A-collection

is a collection of such kind. Thus every simple subset of B can be character-

ized as the union of all complement subsets belonging to a certain maximal

A-collection.

Let B

be the base subset of ( associated with other base B

= ¦x

i

¦

i∈I

.

The simple subsets of B

corresponding to i ∈ I will be denoted by B

(+i)

and B

(−i).

A bijection g : B → B

**is said to be special if g and g
**

−1

map inexact

subsets to inexact subsets.

Lemma 3.11. Let g : B → B

**be a special bijection. Then g and g
**

−1

send simple subsets to simple subsets; moreover, there exists a bijective

transformation δ : I → I such that

g(B(+i)) = B

(+δ(i)), g(B(−i)) = B

(−δ(i)) ∀ i ∈ I

or

g(B(+i)) = B

(−δ(i)), g(B(−i)) = B

(+δ(i)) ∀ i ∈ I.

Proof. It is clear that maximal inexact subsets go to maximal inexact

subsets in both directions. Thus g and g

−1

map complement subsets to

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

120 Grassmannians of Classical Buildings

complement subsets. Since two distinct complement subsets are adjacent

if and only if their intersection is maximal, the adjacency relations of com-

plement subsets is preserved and maximal A-collections go to maximal A-

collections (in both directions). Then g and g

−1

transfer simple subsets

to simple subsets. Two distinct simple subsets are of diﬀerent types if and

only if their intersection is empty or a complement subset. This means that

g and g

−1

map simple subsets of diﬀerent types to simple subsets of dif-

ferent types; hence they preserve the types of all simple subsets or change

the type of every simple subset. Since B(−i) = B ¸ B(+i), there exists a

bijective transformation δ : I → I satisfying the required condition.

Let g : B → B

**be a special bijection. We say that g is a special bijection
**

of the ﬁrst type if it preserves the types of all simple subsets; otherwise, g

is said to be a special bijection of the second type.

Let S, U ∈ B. The equality

S ∩ U = ¸x

i

)

implies that the dimension of S and U is not greater than the codimension;

hence ( = (

β

(V ), β ≤ α. This equality is equivalent to the fact that

S, U ∈ B(+i) and for every j ∈ I ¸ ¦i¦ we have S ,∈ B(+j) or U ,∈ B(+j).

Similarly, the equality

S +U = ¸B ¸ ¦x

i

¦)

implies that the dimension of S and U is α; then ( = (

β

(V ), β ≤ α. This

equality is equivalent to the fact that S, U ∈ B(−i) and for every j ∈ I ¸¦i¦

we have S ,∈ B(−j) or U ,∈ B(−j).

We have proved the following.

Lemma 3.12. Let g and δ be as in the previous lemma. Let also S, U ∈ B.

If g is a special bijection of the ﬁrst type and ( = (

β

(V ), β ≤ α, then

S ∩ U = ¸x

i

) ⇐⇒ g(S) ∩ g(U) = ¸x

δ(i)

).

If g is a special bijection of the ﬁrst type and ( = (

β

(V ), β ≤ α, then

S +U = ¸B ¸ ¦x

i

¦) ⇐⇒ g(S) +g(U) = ¸B

¸ ¦x

δ(i)

¦).

If g is a special bijection of the second type then

S ∩ U = ¸x

i

) ⇐⇒ g(S) +g(U) = ¸B

¸ ¦x

δ(i)

¦),

S +U = ¸B ¸ ¦x

i

¦) ⇐⇒ g(S) ∩ g(U) = ¸x

δ(i)

),

and ( = (

α

(V ).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Classical Grassmannians 121

By Lemma 3.12, special bijections of the second type exist only in the

case when ( = (

α

(V ).

For every linear subspace U ⊂ V we denote by [U] the set of all elements

of ( incident with U.

Lemma 3.13. Suppose that ( = (

β

(V ), β ≤ α. Then for every i ∈ I and

every S ∈ [¸x

i

)] ¸ B there exist M, N ∈ B such that

M ∩ N = ¸x

i

) (3.14)

and M, N, S are contained in a base subset of (.

Proof. Consider the set of all X ⊂ B such that S ∩¸X) = 0. Using Zorn

lemma, we establish the existence of a maximal subset X ⊂ B satisfying

this condition (see the proof of Proposition 1.3). Then ¸X) is a complement

of S (for every vector x

j

∈ B¸X the linear subspace ¸X, x

j

) has a non-zero

intersection with S and x

j

belongs to S +¸X)). Therefore, the cardinality

of X is equal to α. This implies the existence of linear subspaces M

, N

**spanned by non-intersecting subsets of X and such that
**

• dimM

= dimN

= β if β is inﬁnite,

• dimM

= dimN

= β −1 if β is ﬁnite.

The linear subspaces

M := ¸M

, x

i

) and N := ¸N

, x

i

)

belong to B and satisfy (3.14). We take any base Y of S containing x

i

. It is

not diﬃcult to prove that X∪Y is a base of V (see the proof of Proposition

1.4). The associated base subset of ( contains M, N, S.

Remark 3.22. In the case when ( = (

β

(V ), β ≤ α, we need the following

“dual” version of Lemma 3.13: if

S

i

:= ¸B ¸ ¦x

i

¦), i ∈ I,

then for every S ∈ [S

i

] ¸ B there exist M, N ∈ B such that

M +N = S

i

and M, N, S are contained in a base subset of (. The dual principles

do not work and this statement cannot be obtained immediately from

Lemma 3.13.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

122 Grassmannians of Classical Buildings

Now we prove the theorem. Let f : (

β

(V ) → (

β

(V

), β < α, be a

bijection such that f and f

−1

map base subsets to base subsets.

Let P ∈ (

1

(V ). We take any base of V which contains a vector belonging

to P and consider the associated base subset B ⊂ (

β

(V ). The restriction

of f to B is a special bijection to the base subset f(B). By Lemma 3.12,

this is a special bijection of ﬁrst type and there exists h(P) ∈ (

1

(V

) such

that

f(B ∩ [P]) ⊂ [h(P)].

Let S ∈ [P] ¸ B. Lemma 3.13 implies the existence of M, N ∈ B ∩ [P]

satisfying M ∩ N = P and a base subset

ˆ

B ⊂ (

β

(V ) containing M, N, S.

We have

f(

ˆ

B ∩ [P]) ⊂ [P

]

for a certain P

∈ (

1

(V

). Lemma 3.12 guarantees that

h(P) = f(M) ∩ f(N) = P

and f(S) belongs to [h(P)]. So,

f([P]) ⊂ [h(P)].

We apply the same arguments to f

−1

and establish the inverse inclusion.

Therefore, there exists a mapping h : (

1

(V ) → (

1

(V

) satisfying

f([P]) = [h(P)] ∀ P ∈ (

1

(V ).

The mapping h is bijective (the inverse mapping is induced by f

−1

) and

the latter equality implies that

h((

1

(S)) = (

1

(f(S)) ∀ S ∈ (

β

(V ). (3.15)

Let B be a base of V and B be the associated base subset of (

β

(V ).

Let also B

be one of the bases of V

**associated with the base subset f(B).
**

We write B

1

and B

1

for the bases of Π

V

and Π

V

deﬁned by B and B

,

respectively. It follows from Lemma 3.12 that h(B

1

) = B

1

.

So, h and h

−1

send projective bases to projective bases. This means

that h is induced by a semilinear isomorphism l : V → V

. Then

h((

1

(S)) = (

1

(l(S))

for every linear subspace S ⊂ V and, by (3.15), we have f(S) = l(S) for

every S ∈ (

β

(V ).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Chapter 4

Polar and Half-Spin Grassmannians

All thick buildings of types C

n

(n ≥ 3) and D

n

(n ≥ 4) can be obtained

from polar spaces of rank n. By [Buekenhout and Shult (1974)], polar spaces

can be deﬁned as partial linear spaces satisfying some natural axioms (each

line contains at least three points, there is no point collinear with all other

points, ...) and Buekenhout–Shult’s well-known property which says that a

point is collinear with one or all points of a line. Using Teirlinck’s character-

ization of projective spaces, we show that this deﬁnition is equivalent to the

classical Tits–Veldkamp deﬁnition of polar spaces (Theorem 4.1); in partic-

ular, all maximal singular subspaces of a polar space are projective spaces

of the same ﬁnite dimension m (the number m+1 is called the rank of the

polar space). One of our main objects is the polar Grassmannian (

k

(Π),

k ∈ ¦0, 1, . . . , n−1¦, consisting of all k-dimensional singular subspaces of a

rank n polar space Π. Basic properties of polar spaces and their Grassman-

nians will be studied in Sections 4.1 and 4.2. In particular, it will be shown

that for every rank n polar space one of the following possibilities is realized:

• each (n−2)-dimensional singular subspace is contained in at least three

distinct maximal singular subspaces (type C

n

),

• each (n−2)-dimensional singular subspace is contained in precisely two

maximal singular subspaces (type D

n

).

Section 4.3 is dedicated to examples: polar spaces associated with re-

ﬂexive sesquilinear forms and quadratic forms, polar spaces of type D

3

(these polar spaces are isomorphic to the index two Grassmann spaces of

4-dimensional vector spaces). We do not consider polar spaces obtained

from pseudo-quadratic forms and polar spaces of type C

3

associated with

Cayley algebras. Remarks concerning embeddings in projective spaces and

classiﬁcation of polar spaces ﬁnish the section.

123

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

124 Grassmannians of Classical Buildings

In Sections 4.4 and 4.5 we consider polar buildings and investigate ele-

mentary properties of their Grassmann spaces. Let Π be a polar space of

rank n and ∆(Π) be the simplicial complex consisting of all ﬂags formed

by singular subspaces of Π. Then ∆(Π) is a building of type C

n

. The

Grassmannians of this building are the polar Grassmannians of Π. The

corresponding Grassmann spaces will be denoted by G

k

(Π). The building

∆(Π) is thick only in the case when Π is a polar space of type C

n

. Sup-

pose that Π is a polar space of type D

n

. Then (

n−1

(Π) can be naturally

decomposed in two disjoint subsets called the half-spin Grassmannians and

denoted by (

+

(Π) and (

−

(Π). As in Example 2.3, the polar space Π deﬁnes

a thick building of type D

n

. The Crassmannians of this building are the

polar Grassmannians (

k

(Π), k ≤ n − 3, and the half-spin Grassmannians.

As in the previous case, the Grassmann space associated with (

k

(Π) is

G

k

(Π). The Grassmann spaces of the half-spin Grassmannians are denoted

by G

δ

(Π), δ ∈ ¦+, −¦.

Let Π and Π

**be polar spaces of the same type X
**

n

, X ∈ ¦C, D¦, and

n ≥ 3; in the case when X = D, we require that n ≥ 4. In Section 4.6 we

investigate collineations of G

k

(Π) to G

k

(Π

**). We also consider collineations
**

of G

δ

(Π) to G

γ

(Π

**), δ, γ ∈ ¦+, −¦, if our polar spaces are of type D
**

n

. In

almost all cases, such collineations are induced by collineations of Π to Π

.

However, if our polar spaces are of type D

4

then their half-spin Grassmann

spaces (the Grassmann spaces of the half-spin Grassmannians) are polar

spaces of type D

4

and there are two additional possibilities:

• the collineations of G

1

(Π) to G

1

(Π

) induced by collineations of Π to

G

δ

(Π

), δ ∈ ¦+, −¦;

• the collineations of G

δ

(Π) to G

γ

(Π

), δ, γ ∈ ¦+, −¦, induced by

collineations of Π to G

−γ

(Π

).

In Section 4.7 we give an example showing that the adjacency relation on

(

n−1

(Π) cannot be characterized in terms of the opposite relation as in

Theorem 3.6; however, the direct analogue of Theorem 3.6 holds for the

half-spin Grassmannians. Characterizations of apartments in terms of the

adjacency relation (analogues of Theorems 3.8) will be obtained only for the

polar Grassmannians formed by maximal singular subspaces and half-spin

Grassmannians (Section 4.8). In Section 4.9 we describe all apartments

preserving mappings of polar and half-spin Grassmannians; in particular,

it will be shown that all apartments preserving bijections are collineations

of the corresponding Grassmann spaces.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 125

4.1 Polar spaces

4.1.1 Axioms and elementary properties

Following [Buekenhout and Shult (1974)] we deﬁne a polar space (of ﬁnite

rank) as a partial linear space Π = (P, L) satisfying the following axioms:

(1) each line contains at least 3 points,

(2) if p ∈ P and L ∈ L then p is collinear with one or all points of the line

L (the Buekenhout–Shult property),

(3) there is no point collinear with all other points,

(4) every ﬂag consisting of singular subspaces is ﬁnite (this implies that

every singular subspace is ﬁnite-dimensional).

The collinearity relation will be denoted by ⊥: we write p ⊥ q if p and q

are collinear points, and p ,⊥ q otherwise. More general, X ⊥ Y means

that every point of X is collinear with every point of Y . For every subset

X ⊂ P we deﬁne

X

⊥

:= ¦ p ∈ P : p ⊥ X ¦.

The axiom (2) guarantees that X

⊥

is a subspace of Π; moreover, for every

point p ∈ P the subspace p

⊥

is a hyperplane of Π (recall that a proper

subspace of a partial linear space is called a hyperplane if it has a non-

empty intersection with every line).

It follows from the axiom (2) that polar spaces are connected gamma

spaces and the distance between non-collinear points is equal to 2.

Let X ⊂ P be a subset satisfying X ⊥ X (a clique of the collinearity

graph of Π). By Section 2.5, X is contained in a certain singular subspace

of Π. Recall that the minimal singular subspace containing X is called

spanned by X and denoted by ¸X). Since for every point p ∈ P

p ⊥ X =⇒ p ⊥ ¸X)

(Corollary 2.3), we have X

⊥

= ¸X)

⊥

.

If S is a singular subspace of Π and p ,⊥ S then the axiom (2) implies

that S ∩ p

⊥

is a hyperplane of S (hyperplanes of a line are points). If S

is a maximal singular subspace then S

⊥

= S (Π is a gamma space and,

by Proposition 2.7, S is a maximal clique of the collinearity graph); in this

case, S ∩ p

⊥

is a hyperplane of S for every point p ∈ P ¸ S.

Theorem 4.1 ([Buekenhout and Shult (1974)]). Suppose that a polar

space Π = (P, L) contains a singular subspace of dimension greater than 1.

Then the following assertions are fulﬁlled:

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

126 Grassmannians of Classical Buildings

(1) all maximal singular subspaces are projective spaces of the same ﬁnite

dimension,

(2) for every maximal singular subspace S there exists a maximal singular

subspace disjoint from S.

Remark 4.1. Polar spaces of rank n ≥ 3 were deﬁned in [Tits (1974)]

and [Veldkamp (1959/1960)] as partial linear spaces satisfying the following

axioms:

• all maximal singular subspaces are (n − 1)-dimensional projective

spaces,

• for every maximal singular subspace S and every point p ,∈ S, all points

of S collinear with p form a hyperplane of S,

• there exist two disjoint maximal singular subspaces.

It follows from Theorem 4.1 that the Tits–Veldkamp and Buekenhout–Shult

deﬁnitions of polar spaces are equivalent.

The original proof given by F. Buekenhout and E. Shult was rather

complicated. In [Buekenhout (1990)] Theorem 4.1 was drawn from the

following result.

Theorem 4.2 ([Teirlinck (1980)]). Suppose that a linear space has a

family of hyperplanes H which satisﬁes the following conditions:

(a) for every distinct hyperplanes H

1

, H

2

∈ H and every point p there is a

hyperplane H ∈ H containing H

1

∩ H

2

and p,

(b) for every point p there exists a hyperplane H ∈ H which does not

contain p.

If every line contains at least 3 points then the linear space is a projective

space.

4.1.2 Proof of Theorem 4.1

Lemma 4.1. Every hyperplane in a linear space is a maximal proper sub-

space of this linear space.

Proof. If H is a hyperplane of a linear space then this linear space is

spanned by H and any point p ,∈ H (since for every point q ,= p the line p q

intersects H).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 127

Proposition 4.1. Every singular subspace of Π whose dimension is not

less than 2 is a projective space.

Proof. Clearly, we can restrict ourselves to maximal singular subspaces.

Let S be a maximal singular subspace whose dimension is assumed to be

not less than 2. It was noted above that

H

p

:= S ∩ p

⊥

is a hyperplane of S for every point p ∈ P ¸ S. Denote by S

p

the singular

subspace spanned by H

p

and p. If q ∈ S

p

¸ H

p

then H

q

coincides with H

p

(it is clear that H

p

⊂ H

q

and Lemma 4.1 gives the claim). We show that

the family of hyperplanes

¦H

p

¦

p∈P\S

satisﬁes the conditions of Theorem 4.2.

(a). Let H

p

and H

q

be distinct hyperplanes of S. Then p ,∈ S

q

and

q ,∈ S

p

. We take any point t ∈ H

q

¸ H

p

. Since t ,⊥ p, the line t q contains

a unique point q

**collinear with p. This point does not belong to H
**

q

.

Therefore, q

∈ S

q

¸ H

q

. Suppose that the line p q

intersects S in a certain

point. This point belongs to the intersection of H

p

and H

q

= H

q

. This

means that the line p q

is contained in S

q

(since S

q

contains q

and H

p

∩H

q

)

which contradicts the fact that p ,∈ S

q

. Thus the line p q

**does not intersect
**

S. Each point of p q

**is collinear with all points of H
**

p

∩H

q

. For every point

u ∈ S we can choose a point v ∈ p q

**collinear with u. The hyperplane H
**

v

contains u and H

p

∩ H

q

.

(b). For every point u ∈ S the axiom (3) implies the existence of a point

p ∈ P ¸S which is not collinear with u. The associated hyperplane H

p

does

not contain u.

Proposition 4.2. All maximal singular subspaces of Π have the same di-

mension.

Proof. Let S and U be maximal singular subspaces of dimension n and

k (respectively) and k ≤ n. Suppose that the dimension of complements of

S ∩ U in U is equal to m (see Remark 4.2). Then

dim(S ∩ U) = k −m−1.

Let t

1

, . . . , t

m+1

be a base of a certain complement of S ∩ U in U. The

subspace S ∩ U

⊥

is the intersection of the hyperplanes

S ∩ t

⊥

1

, . . . , S ∩ t

⊥

m+1

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

128 Grassmannians of Classical Buildings

and

dim(S ∩ U

⊥

) ≥ n −m−1.

Since U = U

⊥

(U is a maximal singular subspace), we get

k −m−1 ≥ n −m−1

and k ≥ n which implies k = n.

Remark 4.2. Let M, N, T be subspaces of a projective space such that

M, N ⊂ T. We say that N is a complement of M in T if

M ∩ N = ∅ and ¸M, N) = T.

Then

dimM + dimN + 1 = dimT

(for projective planes this is trivial; for projective spaces whose dimension

is greater than 2 this follows from Theorem 1.3).

So, if a polar space contains a singular subspace of dimension greater

than 1 then all maximal singular subspaces are projective spaces of a certain

ﬁnite dimension n ≥ 2; the number n+1 is said to be the rank of this polar

space. If a polar space does not satisfy the condition of Theorem 4.1 then

all maximal singular subspaces are lines, and we say that it is a polar space

of rank 2 or a generalized quadrangle.

Lemma 4.2. Let S be a maximal singular subspace in a polar space of rank

n. Let also U be a singular subspace such that the dimension of complements

of S ∩ U in U is equal to m. Then

dim(S ∩ U

⊥

) = n −m−2

and

¸U, S ∩ U

⊥

) (4.1)

is a maximal singular subspace containing U.

Proof. As in the proof of Proposition 4.2, we establish that

dim(S ∩ U

⊥

) ≥ n −m−2.

The subspace (4.1) is spanned by S ∩U

⊥

and a compliment of S ∩U in U.

Since these subspaces are disjoint and the latter subspace is m-dimensional,

the dimension of (4.1) is not less than n − 1. This means that (4.1) is a

maximal singular subspace and its dimension is equal to n −1. The latter

guarantees that S ∩ U

⊥

is (n −m−2)-dimensional.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 129

Proposition 4.3. For every maximal singular subspace S there exists a

maximal singular subspace disjoint from S.

Proof. Let S and U be maximal singular subspaces such that S ∩U ,= ∅.

By the axiom (3), there exists a point p non-collinear with a certain point

t ∈ S ∩ U. Clearly, p ,∈ U and we denote by U

**the maximal singular
**

subspace spanned by p and U ∩ p

⊥

(Lemma 4.2). If U

contains a point

q ∈ S ¸ U then

U ∩ p

⊥

= U ∩ q

⊥

which is impossible (t ∈ S ∩U is collinear with q and non-collinear with p).

Therefore, S ∩ U

is contained in S ∩ U. Since t ,∈ U

,

dim(S ∩ U

) < dim(S ∩ U).

Step by step, we construct a maximal singular subspace disjoint from S.

Theorem 4.1 is the union of Propositions 4.2 and 4.3.

4.1.3 Corollaries of Theorem 4.1

Let Π = (P, L) be a polar space of rank n.

Proposition 4.4. Every non-maximal singular subspace of Π can be pre-

sented as the intersection of two maximal singular subspaces.

Proof. Let S be a singular subspace and U be a maximal singular sub-

space containing S. By Theorem 4.1, there exists a maximal singular sub-

space U

**disjoint from U. Consider the maximal singular subspace
**

U

:= ¸S, U

∩ S

⊥

)

(Lemma 4.2). Note that S and U

∩S

⊥

are disjoint. The subspace U ∩U

does not intersect U

∩ S

⊥

and the inclusion S ⊂ U ∩ U

guarantees that

U ∩ U

coincides with S.

If S is a singular subspace of Π then S

⊥

is the union of all maximal

singular subspaces containing S (recall that Π is a gamma space and the

class of maximal singular subspaces coincides with the class of maximal

cliques of the collinearity graph). By Proposition 4.4, for any pair of singu-

lar subspaces S and U the inclusion S

⊥

⊂ U

⊥

implies that U ⊂ S; hence,

S

⊥

= U

⊥

if and only if S = U.

Now suppose that X is a clique of the collinearity graph of Π. As above,

X

⊥

is the union of all maximal singular subspaces containing X and X

⊥⊥

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

130 Grassmannians of Classical Buildings

is the intersection of all these subspaces. Then ¸X) ⊂ X

⊥⊥

and Proposition

4.4 implies that

X

⊥⊥

= ¸X).

As a consequence, we obtain the following characterization of lines in terms

of the collinearity relation: if p and q are distinct collinear points then

p q = ¦p, q¦

⊥⊥

.

This means that every isomorphism between the collinearity graphs of polar

spaces is a collineation between these polar spaces.

We will need the following result concerning pairs of non-collinear points

in polar spaces.

Lemma 4.3. If n ≥ 3 and p, q are non-collinear points of Π then the

subspace p

⊥

∩ q

⊥

is a polar space of rank n −1.

Proof. The axioms (1), (2), (4) are trivial. We verify (3).

Suppose that there exists a point t ∈ p

⊥

∩q

⊥

collinear with all points of

p

⊥

∩q

⊥

. For every point s ∈ p

⊥

¸¦p¦ the line p s contains two distinct points

collinear with t (one of these point is p and the other is the intersection with

the hyperplane q

⊥

); by the axiom (2), t is collinear with s. Thus p

⊥

⊂ t

⊥

and p = t which contradicts p ,∈ p

⊥

∩ q

⊥

.

So, p

⊥

∩ q

⊥

is a polar space. It is clear that p

⊥

∩ q

⊥

does not contain

(n − 1)-dimensional singular subspaces. Every maximal singular subspace

containing p intersects q

⊥

in an (n − 2)-dimensional subspace. Therefore,

our polar space is of rank n −1.

4.1.4 Polar frames

Let Π = (P, L) be a polar space of rank n. We say that a subset

¦p

1

, . . . , p

2n

¦

is a frame of Π if for every i ∈ ¦1, . . . , 2n¦ there is unique σ(i) ∈ ¦1, . . . , 2n¦

such that

p

i

,⊥ p

σ(i)

.

First of all we show that frames exist.

Let S and U be disjoint maximal singular subspaces of Π. We take

any base B = ¦p

1

, . . . , p

n

¦ of S. By Lemma 4.2, for every i ∈ ¦1, . . . , n¦

there is a unique point of U collinear with all points of B ¸ ¦p

i

¦; we denote

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 131

it by p

n+i

. This point is non-collinear with p

i

(otherwise p

n+i

⊥ S and

the maximal singular subspace S contains p

n+i

which is impossible, since

S ∩ U = ∅). Then ¦p

1

, . . . , p

2n

¦ is a frame of Π.

Proposition 4.5. Every frame is an independent subset.

Proof. If B = ¦p

1

, . . . , p

2n

¦ is a frame of Π then B ¸ ¦p

i

¦ is contained in

the hyperplane p

⊥

σ(i)

; but this hyperplane does not contain p

i

.

Remark 4.3. In some cases, frames are not bases of a polar space (exam-

ples will be given in Subsection 4.3.1).

Since every subset of an independent subset is independent, any k dis-

tinct mutually collinear points in a frame span a (k−1)-dimensional singular

subspace.

Proposition 4.6. Let B be a frame and S, U be singular subspaces spanned

by subsets of B. Then S ∩U is spanned by the set S ∩U ∩B. In particular,

if this set is empty then S and U are disjoint.

Proof. Suppose that B = ¦p

1

, . . . , p

2n

¦. Since B is an independent sub-

set, the subspaces S and U are spanned by S ∩ B and U ∩ B, respectively.

First, we establish that

(S ∩ B) ∩ (U ∩ B) = ∅ =⇒ S ∩ U = ∅.

An easy veriﬁcation shows that there exist disjoint subsets X, Y ⊂ B

such that ¸X), ¸Y ) are maximal singular subspaces and

S ∩ B ⊂ X, U ∩ B ⊂ Y.

Suppose that ¸X) and ¸Y ) have a non-empty intersection and consider a

point p belonging to ¸X) ∩ ¸Y ). The intersection of all

¸X ¸ ¦p

i

¦), p

i

∈ X,

is empty and there exists p

i

∈ X such that ¸X ¸ ¦p

i

¦) does not contain p.

Then

(X ¸ ¦p

i

¦) ∪ ¦p¦

is a base of ¸X). The point p

σ(i)

∈ B ¸ X = Y is collinear with all points

of this base (it is trivial that p

σ(i)

⊥ X ¸ ¦p

i

¦ and we have p

σ(i)

⊥ p, since

p and p

σ(i)

belong to ¸Y )). This means that p

σ(i)

⊥ ¸X) which contradicts

the fact that p

i

∈ ¸X). Therefore, ¸X) and ¸Y ) are disjoint. Since S ⊂ ¸X)

and U ⊂ ¸Y ), we get the claim.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

132 Grassmannians of Classical Buildings

Now suppose that S ∩ U ∩ B ,= ∅. Then S is spanned by the subspaces

¸S ∩ U ∩ B) and ¸(B ∩ S) ¸ (U ∩ B)).

By the ﬁrst part of our proof, these subspaces are disjoint and the second

subspace is disjoint from U; hence it does not intersects S∩U. The inclusion

¸S ∩ U ∩ B) ⊂ S ∩ U ⊂ S

guarantees that ¸S ∩ U ∩ B) coincides with S ∩ U.

Corollary 4.1. If B is a frame of Π then there is no point of Π collinear

with all points of B.

Proof. A point collinear with all points of B is contained in every maxi-

mal singular subspace spanned by a subset of B. By Proposition 4.6, there

exist disjoint maximal singular subspaces spanned by subsets of B.

Proposition 4.7. For any singular subspaces S and U there is a frame of

Π such that S and U are spanned by subsets of this frame.

Proof. We prove the statement induction by n. The case n = 2 is trivial

and we suppose that n ≥ 3.

If S ⊥ U then there exists a maximal singular subspace M containing S

and U. We choose a base of M such that S and U are spanned by subsets

of this base. It was shown above that this base can be extended to a frame

of Π.

Now suppose that S ,⊥ U. In this case, there are non-collinear points

p ∈ S and q ∈ U. The singular subspaces

S

:= S ∩ q

⊥

and U

:= U ∩ p

⊥

are contained in the polar space p

⊥

∩ q

⊥

(Lemma 4.3). By the inductive

hypothesis, there exists a frame B

of p

⊥

∩ q

⊥

such that S

and U

are

spanned by subsets of B

. Since

S = ¸S

, p) and U = ¸U

, q),

the frame B

∪ ¦p, q¦ is as required.

Corollary 4.2. For any singular subspaces S

1

, S

2

there exist maximal sin-

gular subspaces M

1

, M

2

such that

M

1

∩ M

2

= S

1

∩ S

2

and S

i

⊂ M

i

for i = 1, 2.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 133

Proof. Consider a frame B such that S

1

and S

2

are spanned by subsets

of B. There exists a maximal singular subspace

M

1

= ¸X), X ⊂ B,

containing S

1

and intersecting S

2

precisely in S

1

∩S

2

. If M = ¸B¸X) then

the maximal singular subspace

M

2

:= ¸S

2

, M ∩ S

⊥

2

)

contains S

2

and intersects M

1

precisely in S

1

∩ S

2

.

4.2 Grassmannians

4.2.1 Polar Grassmannians

As above, we suppose that Π = (P, L) is a polar space of rank n. For every

k ∈ ¦0, 1, . . . , n−1¦ denote by (

k

(Π) the polar Grassmannian consisting of

all k-dimensional singular subspaces of Π. Then (

0

(Π) = P.

In the case when k ≤ n − 2, we say that S, U ∈ (

k

(Π) are adjacent

if S ⊥ U and their intersection is (k − 1)-dimensional (or, equivalently, if

S and U span a (k + 1)-dimensional singular subspace). If S and U are

distinct elements of (

n−1

(Π) then S ,⊥ U; such subspaces are said to be

adjacent if their intersection belongs to (

n−2

(Π).

Let M and N be incident singular subspaces of Π such that

dimM < k < dimN.

As in Section 3.1, we deﬁne

[M, N]

k

:= ¦ S ∈ (

k

(Π) : M ⊂ S ⊂ N ¦;

if M = ∅ then we will write ¸N]

k

instead of [M, N]

k

. Also denote by

[M)

k

the set of all elements of (

k

(Π) containing M. In the case when

0 ≤ k < n −1, we say that [M, N]

k

is a line of (

k

(Π) if

dimM = k −1 and dimN = k + 1.

The set [M)

n−1

is said to be a line of (

n−1

(Π) if M belongs to (

n−2

(Π).

The set of all lines of (

k

(Π) will be denoted by L

k

(Π).

Two distinct elements of (

k

(Π) are joined by a line if and only if they

are adjacent. For any adjacent S, U ∈ (

k

(Π) there is precisely one line

containing them:

[S ∩ U, ¸S, U)]

k

if k ≤ n −2

and [S ∩ U)

k

if k = n −1.

Exercise 4.1. Show that every line of (

k

(Π), k < n −1, contains at least

three distinct points.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

134 Grassmannians of Classical Buildings

The pairs

G

k

(Π) := ((

k

(Π), L

k

(Π)), k ∈ ¦0, 1, . . . , n −1¦,

are partial linear spaces; they are called the Grassmann spaces of Π. It is

clear that G

0

(Π) = Π. The Grassmann space G

n−1

(Π) also is known as

the dual polar space of Π [Cameron (1982)].

Proposition 4.8. The Grassmann space G

k

(Π) is connected for every k.

The distance between S, U ∈ (

n−1

(Π) is equal to

n −1 −dim(S ∩ U).

Remark 4.4. In the general case, the distance formula is more compli-

cated.

Proof. Let S, U ∈ (

k

(Π). We deﬁne

cd(S, U) := k −dim(S ∩ U)

(if k = n − 1 then cd(S, U) = 1 is equivalent to the fact that S and U are

adjacent).

Suppose that k = n −1 and cd(S, U) > 1. We take any point p ∈ U ¸ S

and denote by S

1

the maximal singular subspace spanned by S ∩p

⊥

and p.

Then S

1

is adjacent with S and

cd(S

1

, U) = cd(S, U) −1.

Step by step, we construct a sequence of maximal singular subspaces

S = S

0

, S

1

, . . . , S

i

= U, i = cd(S, U),

such that S

j−1

and S

j

are adjacent for every j ∈ ¦1, . . . , i¦. As in the proof

of Proposition 3.1, for every path

S = U

0

, U

1

, . . . , U

l

= U

in the collinearity graph of G

n−1

(Π) we have

dim(U

0

∩ U

1

∩ ∩ U

l

) ≥ (n −1) −l.

The trivial inclusion

U

0

∩ U

1

∩ ∩ U

l

⊂ S ∩ U

guarantees that

dim(S ∩ U) ≥ (n −1) −l.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 135

Thus l ≥ cd(S, U) which means that the distance between S and U is equal

to cd(S, U).

Let k ≤ n−2 and S, U be non-adjacent elements of (

k

(Π). If S ⊥ U then

S, U are contained in the singular subspace ¸S ∪U) and a path connecting

S with U can be constructed as in the proof of Proposition 3.1. In the case

when S ,⊥ U, we prove the connectedness induction by cd(S, U).

If cd(S, U) = 1 then there exists a frame ¦p

1

, . . . , p

2n

¦ of Π such that

S = ¸p

i

1

, . . . , p

i

k

, p

m

) and U = ¸p

i

1

, . . . , p

i

k

, p

σ(m)

).

We choose

j ∈ ¦1, . . . , 2n¦ ¸ ¦i

1

, σ(i

1

), . . . , i

k

, σ(i

k

), m, σ(m)¦

(this is possible, since k ≤ n −2). The k-dimensional singular subspace

¸p

i

1

, . . . , p

i

k

, p

j

)

is adjacent with both S and U.

In the case when cd(S, U) > 1, we take a point p ∈ U satisfying p ,⊥ S

and write S

2

for the k-dimensional singular subspace spanned by S ∩ p

⊥

and p. Then

cd(S, S

2

) = 1 and cd(S

2

, U) = cd(S, U) −1.

There exist S

1

∈ (

k

(Π) adjacent with both S, S

2

and the inductive hypo-

thesis implies the existence of a path connecting S

2

with U.

Lemma 4.4. Let 0 < k < n − 1 and U be a (k − 1)-dimensional singular

subspace of Π. The following assertions are fulﬁlled:

(1) [U)

k

is a polar space of rank n −k,

(2) for every frame B of [U)

k

there is a frame B of Π such that U is

spanned by a subset of B and

B = /∩ [U)

k

,

where / consists of all k-dimensional singular subspaces spanned by

subsets of B.

Proof. (1). An easy veriﬁcation shows that [U)

k

is a polar space and

every its maximal singular subspace is [U, M]

k

, where M is a maximal

singular subspace of Π. The latter means that the rank of our polar space

is equal to n −k.

(2). Let ¦S

1

, . . . , S

2n−2k

¦ be a frame of [U)

k

. In each S

i

we take a point

p

i

∈ S

i

¸ U and denote by X the set of all p

i

. For every p

i

there is unique

p

σ(i)

such that p

i

,⊥ p

σ(i)

. By Lemma 4.3, X

⊥

is a polar space of rank k.

This polar space contains U and we choose a frame Y of X

⊥

such that U

is spanned by a subset of Y . Then X ∪ Y is a frame of Π satisfying the

required conditions.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

136 Grassmannians of Classical Buildings

4.2.2 Two types of polar spaces

Theorem 4.3. For a rank n polar space one of the following possibilities

is realized:

(C) every (n−2)-dimensional singular subspace is contained in at least three

distinct maximal singular subspaces,

(D) every (n − 2)-dimensional singular subspace is contained in precisely

two maximal singular subspaces.

We say that a rank n polar space is of type C

n

or D

n

if the corresponding

case is realized.

Proof. Let Π = (P, L) be a polar space of rank n. First consider the case

when n = 2.

Suppose that ¦p

1

, p

2

, p

3

, p

4

¦ is a frame of Π where

p

1

,⊥ p

3

and p

2

,⊥ p

4

.

Every point p

i

lies on precisely two lines from the collection

p

1

p

2

, p

1

p

4

, p

2

p

3

, p

3

p

4

.

Let L be a third line passing through p

1

.

L

p

p

4

p

2

p

1

p

3

q

q'

There is a point p ∈ L collinear with p

3

and pp

3

is a third line passing

through p

3

. Now consider a point q ∈ L ¸ ¦p

1

, p¦ and take a unique point

q

∈ p

3

p

4

collinear with q (it is clear that the point q

is distinct from p

3

and p

4

). A point on the line qq

collinear with p

2

gives a third line passing

through p

2

. This line contains a point collinear with p

4

and we get a third

line through p

4

. Therefore, for every i ∈ ¦1, 2, 3, 4¦ there are at least three

distinct lines passing through p

i

. Since any pair of points is contained in a

certain frame, the polar space Π is of type C

2

if there is a point belonging

to at least three distinct lines.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 137

Now suppose that n ≥ 3. We need to prove the following: if a cer-

tain S ∈ (

n−2

(Π) is contained in at least three distinct maximal singular

subspaces then the same holds for all elements of (

n−2

(Π).

Let N be an (n−3)-dimensional subspace of S. By Lemma 4.4, [N)

n−2

is

a generalized quadrangle. Since S (as a point of this generalized quadrangle)

lies on at least three distinct lines, the generalized quadrangle is of type

C

2

. This means that every element of [N)

n−2

is contained in at least three

distinct maximal singular subspaces.

Therefore, U ∈ (

n−2

(Π) is contained in at least three distinct maximal

singular subspaces if it has an (n − 3)-dimensional intersection with S, in

particular, if S and U are adjacent. The connectedness of the Grassmann

space G

n−2

(Π) gives the claim.

Every line of G

n−1

(Π) contains at least three distinct points if Π is a

polar space of type C

n

. In the case when Π is of type D

n

, the Grassmann

space G

n−1

(Π) is trivial: every line consists of two points; in other words,

lines are edges of the collinearity graph.

Remark 4.5. If Π is a generalized quadrangle of type C

2

then G

1

(Π) is a

generalized quadrangle of type C

2

. If Π is a generalized quadrangle of type

D

2

then G

1

(Π) does not satisfy the polar axiom (1).

Exercise 4.2. Let Π be a generalized quadrangle of type D

2

and

¦p

1

, p

2

, q

1

, q

2

¦ be a frame of Π such that p

i

,⊥ q

i

for i ∈ ¦1, 2¦. Show

that every line of Π intersects the lines p

1

p

2

and q

1

q

2

or the lines p

1

q

2

and

p

2

q

1

. Therefore, if every line of Π consists of three points then Π looks as

in the picture below.

Proposition 4.9. Every frame in a polar space of type D

n

is a base of this

polar space.

Proof. Let Π = (P, L) be a polar space of type D

n

and

B = ¦p

1

, . . . , p

n

, q

1

, . . . , q

n

¦

be a frame of Π such that p

i

,⊥ q

i

for every i ∈ ¦1, . . . , n¦. By Proposition

4.5, B is an independent subset and we need to show that Π is spanned by

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

138 Grassmannians of Classical Buildings

B. In the case when n = 2, this follows from Exercise 4.2. Suppose that

n ≥ 3 and prove the statement induction by n.

By the inductive hypothesis, the polar space p

⊥

1

∩ q

⊥

1

is spanned by

B ¸ ¦p

1

, q

1

¦. For every point t ∈ p

⊥

1

¸ ¦p

1

¦ the line p

1

t intersects p

⊥

1

∩ q

⊥

1

and p

⊥

1

⊂ ¸B). The same inclusion holds for all points of B. Suppose that

p ∈ P is non-collinear with every point of B. We take x ∈ p

1

p

2

collinear

with p and y ∈ q

1

q

2

collinear with x. Then

B

= (B ¸ ¦p

1

, q

2

¦) ∪ ¦x, y¦

is a frame of Π and p ∈ ¸B

) = ¸B).

4.2.3 Half-spin Grassmannians

Throughout this subsection we suppose that Π = (P, L) is a polar space of

type D

n

. We show that the Grassmannian (

n−1

(Π) can be uniquely decom-

posed in two disjoint parts such that the distance between two elements of

(

n−1

(Π) is odd if and only if these elements belong to the diﬀerent parts.

Recall that

d(S, U) = n −1 −dim(S ∩ U).

for all S, U ∈ (

n−1

(Π).

Lemma 4.5. Let S and U be adjacent elements of (

n−1

(Π). Then for any

N ∈ (

n−1

(Π) the distance d(S, N) is odd if and only if d(U, N) is even.

Proof. The statement is trivial if N coincides with S or U and we assume

that N is distinct from S and U. Observe that every point of S ¸ U is non-

collinear with every point of U ¸ S. This means that at least one of the

subspaces S ∩N, U ∩N is contained in S ∩U (otherwise N contains points

x ∈ S ¸ U and y ∈ U ¸ S which cannot be collinear).

Lemma 4.2 implies the existence of a point

p ∈ N ¸ (S ∩ U)

collinear with all points of S ∩ U. If S ∩ N and U ∩ N both are contained

in S ∩U then p does not belong to S ∪U and ¸S ∩U, p) is a third maximal

singular subspace containing S ∩ U which contradicts the assumption that

Π is of type D

n

.

Therefore, only one of these subspaces is contained in S ∩ U. Consider

the case when

S ∩ N ⊂ S ∩ U and U ∩ N ,⊂ S ∩ U.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 139

Then

S ∩ N = S ∩ U ∩ N

is a hyperplane of U ∩ N (since S ∩ U is a hyperplane of U). This implies

that

d(S, N) = d(U, N) + 1.

The second case is similar.

Theorem 4.4. There is a unique pair of disjoint subsets

A, } ⊂ (

n−1

(Π)

satisfying the following conditions:

A ∪ } = (

n−1

(Π),

the distance between two elements of (

n−1

(Π) is odd if and only if one of

them belongs to A and other belongs to }.

Proof. We ﬁx N ∈ (

n−1

(Π) and deﬁne

A := ¦S ∈ (

n−1

(Π) : d(S, N) is even¦,

} := ¦U ∈ (

n−1

(Π) : d(U, N) is odd¦.

These are disjoint subsets whose union is (

n−1

(Π). By the previous lemma,

any two adjacent elements of (

n−1

(Π) belong to the diﬀerent subsets.

Therefore, if the distance between two elements of (

n−1

(Π) is odd then

one of them belongs to A and the other belongs to }; in the case when the

distance is even, the elements both belong to A or }.

Suppose that

A

, }

⊂ (

n−1

(Π)

is another pair of subsets satisfying the same conditions. Then A

intersects

at least one of the subsets A, }. Let A ∩ A

,= ∅ and S be an element of

this intersection. Then d(S, U) is even for every U belonging to A ∪ A

.

This means that A

= A, hence }

= }.

The subsets described in Theorem 4.4 will be denoted by

(

+

(Π), (

−

(Π)

and called the half-spin Grassmannians of Π. The distance between any

two elements of the half-spin Grassmannian (in the collinearity graph of

G

n−1

(Π)) is even.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

140 Grassmannians of Classical Buildings

Let δ ∈ ¦+, −¦. Two elements of (

δ

(Π) are said to be adjacent if their

intersection is (n−3)-dimensional (the distance in the collinearity graph of

G

n−1

(Π) is equal to 2). The intersection

[M)

δ

:= [M)

n−1

∩ (

δ

(Π)

is called a line of (

δ

(Π) if M is (n − 3)-dimensional. The set of all such

lines will be denoted by L

δ

(Π).

If n = 2 then any two distinct elements of (

δ

(Π) are disjoint and there

is only one line which coincides with (

δ

(Π). In what follows we will restrict

ourselves to the case when n ≥ 3.

Two distinct elements of (

δ

(Π) are joined by a line if and only if they

are adjacent. If S, U ∈ (

δ

(Π) are adjacent then [S ∩ U)

δ

is the unique line

containing them.

If δ ∈ ¦+, −¦ then we write −δ for the sing satisfying ¦δ, −δ¦ = ¦+, −¦.

Proposition 4.10. Every line of (

δ

(Π), δ ∈ ¦+, −¦, contains at least three

points.

Proof. Let M ∈ (

n−3

(Π). Consider any (n − 2)-dimensional singular

subspace N

1

containing M. By Proposition 4.4, this is the intersection of

two maximal singular subspaces S

1

and U

1

; one of them belongs to (

δ

(Π)

and the other is an element of (

−δ

(Π). Suppose that U

1

∈ (

−δ

(Π). We

choose two distinct (n − 2)-dimensional singular subspaces N

2

, N

3

,= N

1

contained in U

1

and containing M (in other words, N

1

, N

2

, N

3

are distinct

points on the line [M, U

1

]

n−2

). Proposition 4.4 implies the existence of

S

i

∈ (

δ

(Π) (i = 2, 3) intersecting U

1

precisely in N

i

. Then S

1

, S

2

, S

3

are

distinct points on the line [M)

δ

.

The patrial linear spaces

G

δ

(Π) := ((

δ

(Π), L

δ

(Π)), δ ∈ ¦+, −¦,

are called the half-spin Grassmann spaces of Π.

Exercise 4.3. Shows that every path of length j in the collinearity graph of

G

δ

(Π), δ ∈ ¦+, −¦, can be extended to a path of length 2j in the collinearity

graph of G

n−1

(Π). Hint: if S, U ∈ (

δ

(Π) are adjacent then there exist

p ∈ S ¸ U and q ∈ U ¸ S satisfying p ⊥ q.

Proposition 4.11. The half-spin Grassmann space G

δ

(Π), δ ∈ ¦+, −¦, is

connected. The distance between S, U ∈ (

δ

(Π) (in the collinearity graph of

G

δ

(Π)) is equal to

n −1 −dim(S ∩ U)

2

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 141

Proof. Let S, U ∈ (

δ

(Π). By Proposition 4.8, the collinearity graph of

G

n−1

(Π) contains a path

S = S

0

, S

1

, . . . , S

i

= U, i = n −1 −dim(S ∩ U).

Theorem 4.4 guarantees that i is even and

S

0

, S

2

, . . . , S

i−2

, S

i

is a path in the collinearity graph of G

δ

(Π). The distance formula follows

immediately from Exercise 4.3.

4.3 Examples

4.3.1 Polar spaces associated with sesquilinear forms

Let V be an n-dimensional left vector space over a division ring R and Ω be

a non-degenerate reﬂexive form deﬁned on V . Suppose that Ω has totally

isotropic subspaces of dimensions 2. This implies that n ≥ 4 (the dimension

of a totally isotropic subspace is not greater than the codimension). We

write L(Ω) for the set formed by all lines of Π

V

such that the associated

2-dimensional linear subspaces are totally isotropic. Distinct P, P

∈ (

1

(Ω)

are joined by a such line if and only if P ⊥ P

**(⊥ is the orthogonal relation
**

deﬁned by Ω). We will investigate the pair

Π

Ω

:= ((

1

(Ω), L(Ω)).

Since two “proportional” forms (one of the forms is a scalar multiple of

the other) have the same set of totally isotropic subspaces, we can assume

that Ω is one of the forms given in Theorem 1.6: alternating, symmetric,

or Hermitian.

Suppose that

Ω(x, y) = εσ(Ω(y, x)) ∀ x, y ∈ V,

where σ is an anti-automorphism of R satisfying σ

2

= 1

R

and ε = ±1; in

other words, Ω is one of the forms considered in Examples 1.9 and 1.10.

We deﬁne the trace set

T(σ, ε) := ¦ a + εσ(a) : a ∈ R ¦

and say that the form Ω is trace-valued if

Ω(x, x) ∈ T(σ, ε) (4.2)

for every x ∈ V . This condition holds if the characteristic of R is not equal

to 2. Also every alternating form is trace-valued.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

142 Grassmannians of Classical Buildings

Remark 4.6. In the general case, all vectors x ∈ V satisfying (4.2) form

a linear subspace W ⊂ V . All isotropic vectors belong to W and the

restriction of Ω to W is trace-valued.

Example 4.1. Suppose that R is a ﬁeld of characteristic 2 and the form Ω

is bilinear (σ = 1

R

and ε = 1 = −1). Since the trace set is zero, the form

Ω is trace-valued only in the case when it is alternating.

Lemma 4.6. Let S be a 2-dimensional linear subspace of V containing an

isotropic vector x. If Ω is trace-valued then S contains an isotropic vector

linearly independent with x.

Proof. The statement is trivial if S is totally isotropic. Otherwise, we

choose a vector y ∈ S such that Ω(x, y) = 1 and a scalar a ∈ R satisfying

a + εσ(a) + Ω(y, y) = 0

(this is possible, since our form is trace-valued). Then

Ω(ax + y, ax + y) = Ω(ax, ax) + aΩ(x, y) + εσ(aΩ(x, y)) + Ω(y, y) = 0.

**In other words, if the form Ω is trace-valued and a line of Π
**

V

has a non-

empty intersection with (

1

(Ω) then this intersection contains more than

one point. Using this fact we prove the following.

Lemma 4.7. If Ω is trace-valued then there is a base of V consisting of

isotropic vectors.

Proof. Let B = ¦x

1

, . . . , x

n

¦ be a base of V and x be an isotropic vector.

The intersection of the linear subspaces

¸B ¸ ¦x

i

¦), i = 1, . . . , n,

is empty and we choose j such that x does not belong to ¸B¸¦x

j

¦). Assume

that j = n. Lemma 4.6 implies the existence of isotropic vectors

y

i

∈ ¸x, x

i

) ¸ ¦x¦, i = 1, . . . , n −1.

The base x, y

1

, . . . , y

n−1

is as required.

Proposition 4.12. If Ω is trace-valued then Π

Ω

is a polar space.

Proof. The polar axiom (1) is trivial and we leave the veriﬁcation of the

axiom (2) for the reader. If the axiom (3) fails then there exists an isotropic

vector orthogonal to all isotropic vectors; by Lemma 4.7, this vector is

orthogonal to all vectors of V which is impossible, since our form is non-

degenerate. The axiom (4) follows from the fact that singular subspaces of

Π

Ω

can be identiﬁed with totally isotropic subspaces.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 143

It follows immediately from Theorem 4.1 and Proposition 4.12 that

all maximal totally isotropic subspaces have the same dimension. This

dimension is called the Witt index of the form Ω. The Witt index is not

greater than

n

2

and equal to the rank of the polar space Π

Ω

. For example,

if Ω is alternating then n is even and the Witt index is equal to

n

2

.

Remark 4.7. The veriﬁcation of Tits – Veldkamp axioms is more com-

plicated [Dieudonn´e (1971); Taylor (1992)]. The statement concerning the

dimension of maximal totally isotropic spaces is a part of well-known Witt

theorem.

Lemma 4.8. Let σ be a non-identity anti-automorphism of R satisfying

σ

2

= 1

R

. Then there exists a non-zero scalar a ∈ R such that

a + σ(a) = 0.

Proof. The statement is trivial if the characteristic is equal to 2 (we can

take, for example, a = 1). Suppose that the characteristic is not equal to

2. In this case, the trace set

T(σ, 1) = ¦ a + σ(a) : a ∈ R ¦

coincides with the set of all scalars b ∈ R satisfying σ(b) = b. Let us take

any b ∈ R ¸ T(σ, 1) (such elements exist, since σ is non-identity). Then

c := b + σ(b) ∈ T(σ, 1)

and the equality

c =

c

2

+ σ

c

2

**guarantees that the scalar
**

a := b −

c

2

,= 0

is as required.

By Theorem 1.6, there are precisely the following three types of polar

spaces associated with reﬂexive forms:

• symplectic polar spaces (deﬁned by alternating forms),

• symmetric polar spaces (deﬁned by symmetric forms, the characteristic

is not equal to 2),

• Hermitian polar spaces (deﬁned by trace-valued Hermitian forms).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

144 Grassmannians of Classical Buildings

Proposition 4.13. Let Π

Ω

be one of the polar spaces considered above.

Then Π

Ω

is of type D

m

only in the following case: the characteristic of R

is not equal to 2, the dimension n is even and Ω is a symmetric form of

Witt index

n

2

.

Proof. Let m be the Witt index of the form Ω and S be a totally isotropic

subspace of dimension m− 1. Then S is the intersection of two maximal

totally isotropic subspaces U

1

and U

2

. We take 1-dimensional linear sub-

spaces P

i

(i = 1, 2) such that

U

i

= S + P

i

.

It is clear that P

1

,⊥ P

2

.

Suppose that Ω is alternating. Then for every 1-dimensional linear

subspace P ⊂ P

1

+ P

2

the linear subspace S + P is totally isotropic and

our polar space is of type C

m

.

Consider the case when Ω is a Hermitian form associated with an anti-

automorphism σ : R → R. If x, y ∈ V are isotropic vectors then

Ω(x + by, x + by) = Ω(x, by) + Ω(by, x) = σ(bΩ(y, x)) + bΩ(y, x).

We choose vectors x ∈ P

1

, y ∈ P

2

and a scalar b ∈ R such that a = bΩ(y, x)

satisﬁes the condition of Lemma 4.8. Then ¸x + by) belongs to (

1

(Ω) and

S +¸x + by)

is a third maximal totally isotropic subspace containing S. The polar space

is of type C

m

.

Now suppose that Ω is symmetric and the characteristic of R is not

equal to 2. We have

Ω(x + by, x + by) = 2bΩ(x, y)

for any isotropic vectors x and y. Since P

1

,⊥ P

2

, this means that the linear

subspace P

1

+P

2

does not contain elements of (

1

(Ω) distinct from P

1

and

P

2

.

If m <

n

2

then S + P

1

+ P

2

is a proper linear subspace of S

⊥

, since

dim(S + P

1

+ P

2

) = m+ 1 < n −m+ 1 = dimS

⊥

.

We take any 1-dimensional linear subspace P

⊂ S

⊥

which is not contained

in S +P

1

+P

2

. By Lemma 4.6, P

1

+P

contains P

∈ (

1

(Ω) distinct from

P

1

. Then S + P

**is a third maximal totally isotropic subspace containing
**

S. The polar space is of type C

m

.

Every maximal totally isotropic subspace U containing S is contained in

S

⊥

. If n = 2m then S

⊥

is (m+1)-dimensional and U intersects P

1

+P

2

in

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 145

a 1-dimensional linear subspace P. Since P is totally isotropic, it coincides

with P

1

or P

2

. This implies that there are precisely two maximal totally

isotropic subspaces containing S and the polar space is of type D

m

.

Frames of the polar space Π

Ω

are independent subsets of Π

V

. If the

Witt index is less than

n

2

then every frame spans a proper subspace of

Π

V

whose intersection with (

1

(Ω) is a subspace of Π

Ω

; this subspace is

proper (by Lemma 4.7) and frames are not bases of Π

Ω

. However, if the

characteristic of R is not equal to 2 and Ω is alternating then every frame

is a base of Π

Ω

[Blok and Brouwer (1998); Cooperstein and Shult (1997)].

In the case of characteristic 2, this fails.

Example 4.2. Suppose that n = 4, R is a ﬁeld of characteristic 2 and

the form Ω is alternating. Let B = ¦p

1

, p

2

, q

1

, q

2

¦ be a frame of the polar

space Π

Ω

such that p

i

,⊥ q

i

for each i. There exist vectors x

1

, x

2

, y

1

, y

2

∈ V

satisfying

p

i

= ¸x

i

), q

i

= ¸y

i

) and Ω(x

i

, y

i

) = Ω(y

i

, x

i

) = 1.

Consider two pairs of collinear points

p

∈ p

1

p

2

, q

∈ q

1

q

2

and p

∈ p

1

q

2

, q

∈ p

2

q

1

distinct from the points of B. Then for some non-zero scalars a, b ∈ R we

have

p

= ¸x

1

+ ax

2

), q

= ¸y

2

+ ay

1

),

p

= ¸x

1

+ by

2

), q

= ¸x

2

+ by

1

).

Almost all points of the line p

q

(except the point q

) are of type

¸(x

1

+ ax

2

) + t(y

2

+ ay

1

)), t ∈ R,

and almost all points of the line p

q

(except the point q

) are of type

¸(x

1

+ by

2

) + s(x

2

+ by

1

)), s ∈ R.

These lines have a common point (t = b, s = a). This means that ¸B) is

the grid consisting of all lines which intersect p

1

p

2

and q

1

q

2

or p

1

q

2

and

p

2

q

1

. Therefore, ¸B) is a proper subspace of Π

Ω

; it does not contain, for

example, any line passing through p

1

and distinct from p

1

p

2

and p

1

q

2

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

146 Grassmannians of Classical Buildings

4.3.2 Polar spaces associated with quadratic forms

Let V be an n-dimensional vector space over a ﬁeld F. A non-zero mapping

Q : V → F is called a quadratic form if

Q(ax) = a

2

Q(x)

for all a ∈ F, x ∈ V and there exists a bilinear form Ω : V V → F such

that

Q(x + y) = Q(x) + Q(y) + Ω(x, y)

for all x, y ∈ V . It is clear that Ω is symmetric. The form Q is said to be

non-degenerate if Q(x) ,= 0 for every non-zero vector x belonging to V

⊥

(we

do not require that the associated form Ω is non-degenerate and write V

⊥

for the linear subspace of all vectors orthogonal to V ). By this deﬁnition,

Q is non-degenerate if the associated form Ω is non-degenerate.

A non-zero vector x ∈ V is called singular if Q(x) = 0 and we say that a

linear subspace S is totally singular if every non-zero vector of S is singular.

Every totally singular subspace is totally isotropic for the associated form

Ω; but totally isotropic subspaces of Ω need not to be totally singular. If

the characteristic of F is not equal to 2 then Q can be uniquely recovered

from Ω by the formula

Q(x) =

Ω(x, x)

2

.

This equality guarantees that Q is non-degenerate if and only if Ω is non-

degenerate; moreover, a linear subspace is totally singular (for Q) if and

only if it is totally isotropic (for Ω).

From this moment we assume that Q is non-degenerate and there exist

totally singular subspaces of dimension 2. We write (

1

(Q) for the set of all

1-dimensional totally singular subspaces and denote by L(Q) the set formed

by all lines of Π

V

such that the associated 2-dimensional linear subspaces

are totally singular. Consider the pair

Π

Q

:= ((

1

(Q), L(Q)).

It is clear that P, P

∈ (

1

(Q) are joined in Π

Q

by a line only in the case when

P ⊥ P

**(⊥ is the orthogonal relation deﬁned by Ω). If the characteristic of
**

F is not equal to 2 then Π

Q

coincides with the polar space Π

Ω

.

Suppose that the characteristic of F is equal to 2. The equality

Q(2x) = 2Q(x) + Ω(x, x)

shows that Ω is alternating. We restrict our self to the case when Ω is

non-degenerate and refer [Dieudonn´e (1971)] for the general case. Since

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 147

totally singular subspaces are totally isotropic, Π

Q

is a subspace of the

polar space Π

Ω

. Thus Π

Q

satisﬁes the polar axioms (1), (2) and (4). This

is a proper subspace of Π

Ω

(all vectors are isotropic, but Q ,≡ 0 and there

are non-singular vectors).

Exercise 4.4. Prove the following analogue of Lemma 4.6: for every sin-

gular vector x ∈ V and any vector y ∈ V such that Ω(x, y) ,= 0 there exists

a singular vector z ∈ ¸x, y) independent with x. Hint: take a scalar a ∈ F

satisfying aΩ(x, y) + Q(y) = 0, the vector ax + y is as required.

Exercise 4.5. Prove the analogue of Lemma 4.7: there exists a base of V

consisting of singular vectors. Hint: apply Exercise 4.4 to a singular vector

x ∈ V and a base of V formed by vectors non-orthogonal to x.

Using Exercise 4.5 we establish that Π

Q

satisﬁes the axiom (3). There-

fore, Π

Q

is a polar space. All singular subspaces of Π

Q

can be identiﬁed

with totally singular subspaces of Q.

We do not consider here so-called pseudo-quadratic forms associated

with Hermitian forms (see [Tits (1974)], Section 8.2); as quadratic forms,

they give new examples of polar spaces only in the case of characteristic 2.

4.3.3 Polar spaces of type D

3

Let V be a 4-dimensional vector space over a division ring R.

Exercise 4.6. Show that the Grassmann space G

2

(V ) is a polar space.

Maximal singular subspaces of G

2

(V ) (stars and tops) are projective

planes, thus the rank of the polar space is equal to 3. Since every line is

contained in precisely one star and precisely one top, this is a polar space

of type D

3

.

Remark 4.8. If R is commutative then this polar space can be obtained

from the Klein quadratic form deﬁned on the 6-dimensional vector space

∧

2

V [Cameron (1991); Taylor (1992)].

Conversely, let Π = (P, L) be a polar space of type D

3

. Consider the

associated half-spin Grassmannian space G

δ

(Π), δ ∈ ¦+, −¦. There is

natural one-to-one correspondence between lines of G

δ

(Π) and points of Π

(every line of G

δ

(Π) consists of all elements of (

δ

(Π) passing through a

certain point). Any two distinct elements of (

δ

(Π) are adjacent and G

δ

(Π)

is a linear space.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

148 Grassmannians of Classical Buildings

Lemma 4.9. If Π is a polar space of type D

3

then G

δ

(Π), δ ∈ ¦+, −¦, is

a 3-dimensional projective space.

Proof. Let S

1

, S

2

, S

3

be a triangle in G

δ

(Π). Then

p

1

= S

2

∩ S

3

, p

2

= S

1

∩ S

3

, p

3

= S

1

∩ S

2

form a triangle in Π and span a certain plane U. This plane belongs to

(

−δ

(Π), since each S

i

intersects U in the line p

l

p

m

, where i ,= l, m.

Denote by A the set consisting of all elements of (

δ

(Π) which intersect

U in lines (for every S ∈ (

δ

(Π) the intersection S∩U is empty or a line). If

S and S

are distinct elements of A then S ∩S

**∩U is a certain point p and
**

the associated line [p)

δ

joins S and S

**; it is clear that this line is contained
**

in A. Therefore, A is a subspace of G

δ

(Π) and a line of G

δ

(Π) is contained

in A if and only if the associated point belongs to U.

There is a one-to-one correspondence between elements of A and lines

of U (every line of U is contained in precisely one element of A). This is a

collineation of the subspace A to the projective plane dual to U. Clearly,

S

1

, S

2

, S

3

form a base of the projective plane A and we have established

that every plane of G

δ

(Π) is projective.

If p ∈ P ¸U then U∩p

⊥

is a line in U and the maximal singular subspace

¸U ∩ p

⊥

, p) belongs to (

δ

(Π); in other words, the line of G

δ

(Π) deﬁned by

the point p has a non-empty intersection with A. This means that A is a

hyperplane of G

δ

(Π) and we get the claim.

So, we can suppose that G

+

(Π) = Π

V

, where V is a 4-dimensional

vector space over a division ring. Consider the bijection

f : (

2

(V ) → P

induced by the natural one-to-one correspondence between lines of G

+

(Π)

and points of Π. It was established above that for every plane A of G

+

(Π)

there exists U ∈ (

−

(Π) such that A consists of all elements of (

+

(Π)

intersecting U in lines. This is a one-to-one correspondence and we get

another bijection

g : (

3

(V ) → (

−

(Π).

Exercise 4.7. Show that f is a collineation of G

2

(V ) to Π and g is a

collineation of Π

∗

V

to G

−

(Π).

Exercise 4.8. Let h be a collineation of G

2

(V ) to Π. Show that / ⊂ (

2

(V )

is an apartment if and only if h(/) is a frame of Π. Hint: use Theorem 3.8.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 149

By Exercises 4.7 and 4.8, we get the following.

Proposition 4.14. If Π is a polar space of type D

3

then there exists a

4-dimensional vector space V such that

Π, G

+

(Π), G

−

(Π)

are isomorphic to

G

2

(V ), Π

V

, Π

∗

V

,

respectively. Moreover, every collineation of Π to G

2

(V ) establishes a one-

to-one correspondence between frames of Π and apartments of (

2

(V ).

4.3.4 Embeddings in projective spaces and classiﬁcation

The Veldkamp space V(Π) of a polar space Π (the rank of Π is assumed

to be not less than 3) is the linear space whose points are hyperplanes of

Π and the line joining hyperplanes H

1

and H

2

consists of all hyperplanes

containing H

1

∩ H

2

. The mapping p → p

⊥

is an embedding of Π in V(Π)

and it is not diﬃcult to prove that this embedding transfers lines to lines.

Theorem 4.5 ([Veldkamp (1959/1960)]). Let Π be a polar space of

rank n ≥ 3; in the case when n = 3, we require in addition that Π is

of type C

n

and every 2-dimensional singular subspace is a Desarguesian

projective plane. Then V(Π) is a projective space.

This result was exploited in the following classiﬁcation of polar spaces.

Theorem 4.6 ([Tits (1974)]). Up to isomorphism there are the following

three types of polar spaces whose rank is not less than three:

• the polar spaces associated with non-degenerate reﬂexive sesquilinear,

quadratic, and pseudo-quadratic forms;

• the Grassmann space G

2

(V ), where V is a 4-dimensional vector space,

this polar space is deﬁned by Klein’s quadratic form if V is a vector

space over a ﬁeld;

• the polar spaces of type C

3

associated with Cayley algebras, maxi-

mal singular subspaces of such polar spaces are non-Desarguesian (Mo-

ufang) projective planes.

In particular, every polar space whose rank is greater than 3 is isomorphic

to the polar space associated with a certain reﬂexive sesquilinear, quadratic,

or pseudo-quadratic form.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

150 Grassmannians of Classical Buildings

4.4 Polar buildings

4.4.1 Buildings of type C

n

Let Π = (P, L) be a polar space of rank n ≥ 3. Consider the ﬂag complex

∆(Π) obtained from the set of all proper singular subspaces of Π with the

natural incidence relation. Every frame B of Π deﬁnes the subcomplex

Σ

B

⊂ ∆(Π) consisting of all ﬂags formed by singular subspaces spanned

by subsets of B; this subcomplex is called the apartment associated with

(deﬁned by) the frame B. The complex ∆(Π) together with the set of all

such apartments is a building of type C

n

. This building is thick only in

the case when Π is of type C

n

. The Grassmannians of ∆(Π) are the polar

Grassmannians (

k

(Π), k ∈ ¦0, 1, . . . , n −1¦; the corresponding Grassmann

spaces are G

k

(Π). In what follows the associated Grassmann graphs (the

collinearity graphs of Grassmann spaces) will be denoted by Γ

k

(Π).

By [Tits (1974)], every thick building of type C

n

(n ≥ 3) is isomorphic

to the C

n

-building of a rank n polar space.

4.4.2 Buildings of type D

n

Now let Π = (P, L) be a polar space of type D

n

and n ≥ 4. We write (

∗

(Π)

for the set of all proper singular subspaces whose dimension is not equal to

n −2 and deﬁne the oriﬂamme incidence relation ∗ on (

∗

(Π):

• if S ∈ (

k

(Π), k ≤ n −3, and U ∈ (

∗

(Π) then S ∗ U means that S and

U are incident in the usual sense,

• for S ∈ (

+

(Π) and U ∈ (

−

(Π) we write S ∗ U if their intersection

belongs to (

n−2

(Π).

The associated ﬂag complex will be called the oriﬂamme complex of Π and

denoted by Orif(Π). For every frame B the corresponding apartment Orif

B

is the subcomplex of Orif(Π) consisting of all oriﬂamme ﬂags formed by

singular subspaces spanned by subsets of B. The complex Orif(Π) together

with the set of all such apartments is a thick building of type D

n

. The

Grassmannians are the polar Grassmannians (

k

(Π), k ∈ ¦0, 1, . . . , n − 3¦,

and the half-spin Grassmannians (

δ

(Π), δ ∈ ¦+, −¦. The corresponding

Grassmann spaces are G

k

(Π) and G

δ

(Π); we write Γ

k

(Π) and Γ

δ

(Π) for

the associated Grassmann graphs.

Every thick building of type D

n

(n ≥ 4) is isomorphic to the oriﬂamme

building obtained from a polar space of type D

n

[Tits (1974)].

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 151

4.5 Elementary properties of Grassmann spaces

4.5.1 Polar Grassmann spaces

Let Π = (P, L) be a polar space of rank n ≥ 3. First, we describe triangles

in the Grassmann space G

k

(Π), 1 ≤ k ≤ n −1.

Lemma 4.10. If 1 ≤ k ≤ n − 2 and S

1

, S

2

, S

3

∈ (

k

(Π) form a triangle

in G

k

(Π) then S

i

⊥ S

j

for all i, j ∈ ¦1, 2, 3¦ and only one of the following

possibilities is realized:

(1) a star-triangle: k < n − 2 and there is a (k − 1)-dimensional singular

subspace contained in each S

i

,

(2) a top-triangle: there is a (k + 1)-dimensional singular subspace con-

taining all S

i

.

The Grassmann space G

n−1

(Π) does not contain triangles: any three mutu-

ally collinear points of G

n−1

(Π) are collinear if Π is of type C

n

and G

n−1

(Π)

does not contain triples of mutually collinear points if Π is of type D

n

.

Proof. Let 1 ≤ k ≤ n −2 and S

1

, S

2

, S

3

∈ (

k

(Π) be a triangle in G

k

(Π).

Since S

i

is adjacent with S

j

(i ,= j), we have S

i

⊥ S

j

for all i, j ∈ ¦1, 2, 3¦.

As for Grassmannians of ﬁnite-dimensional vector spaces, if S

3

is not con-

tained in the (k + 1)-dimensional singular subspace spanned by S

1

and S

2

then

¸S

1

, S

2

) ∩ S

3

is a (k − 1)-dimensional singular subspace contained in S

1

and S

2

; in this

case, the singular subspace spanned by S

1

, S

2

, S

3

is (k + 2)-dimensional

which is possible only for k < n −2.

Let S

1

, S

2

, S

3

∈ (

n−1

(Π) be mutually collinear points of G

n−1

(Π). Then

U

1

:= S

2

∩ S

3

, U

2

:= S

1

∩ S

3

, U

3

:= S

1

∩ S

2

belong to (

n−2

(Π). Since

U

1

, U

2

⊂ S

3

, U

1

, U

3

⊂ S

2

, U

2

, U

3

⊂ S

1

,

we have U

i

⊥ U

j

for all i, j ∈ ¦1, 2, 3¦. Suppose that at least two of these

three subspaces are distinct, for example, U

1

,= U

2

. Then

S

3

= ¸U

1

, U

2

)

and U

3

⊥ S

3

. The singular subspace S

3

is maximal and we get the inclusion

S

1

∩ S

2

= U

3

⊂ S

3

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

152 Grassmannians of Classical Buildings

which implies that U

3

is contained in each S

i

. Hence U

1

= U

2

= U

3

, a

contradiction. Therefore, all U

i

are coincident and S

1

, S

2

, S

3

are points on

a line of G

n−1

(Π).

Proposition 4.15. G

k

(Π) is a gamma space.

Proof. By Lemma 4.10, if a point of G

n−1

(Π) is collinear with two dis-

tinct points on a line then it belongs to this line and G

n−1

(Π) is a gamma

space. The case k = 0 was considered in Subsection 4.1.1 and we suppose

that 0 < k < n−1. If S ∈ (

k

(Π) is collinear with two distinct points S

1

, S

2

of a line [M, N]

k

and S does not belong to this line then S ⊥ N (since N

is spanned by S

1

and S

2

) and

M = S

1

∩ S

2

⊂ S or S ⊂ N

(S, S

1

, S

2

form a star-triangle or a top-triangle, respectively). In each of

these cases, S is collinear with all points of the line [M, N]

k

.

It follows from Proposition 4.15 that the class of maximal singular sub-

spaces of G

k

(Π) coincides with the class of maximal cliques of the Grass-

mann graph Γ

k

(Π).

Example 4.3. By the second part of Lemma 4.10, every line of G

n−1

(Π)

is a maximal singular subspace.

Example 4.4. Suppose that k ≤ n −2. For every (k +1)-dimensional sin-

gular subspace N the set ¸N]

k

is called a top. This is a singular subspace of

G

k

(Π) isomorphic to a (k+1)-dimensional projective space. As in Example

3.1, we show that this singular subspace is maximal. Every triangle in a

top is a top-triangle.

Example 4.5. As in the previous example, we suppose that k ≤ n − 2.

Let M be a (k − 1)-dimensional singular subspace and N be a maximal

singular subspace containing M. Then [M, N]

k

is a singular subspace of

G

k

(Π). In the case when k = n − 2, this is a line. If k < n − 2 then the

subspace [M, N]

k

is said to be a star. It can be identiﬁed with a star in the

Grassmann space ¸N]

k

. The star [M, N]

k

is isomorphic to an (n −k − 1)-

dimensional projective space. Let S be a k-dimensional singular subspace

of Π which does not belong to [M, N]

k

. If S ⊂ N then there exists an

element of [M, N]

k

non-adjacent with S (since [M, N]

k

is a maximal clique

in the collinearity graph of ¸N]

k

). If S does not belong to ¸N]

k

then S ,⊥ N

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 153

and we take any point p ∈ N satisfying p ,⊥ S; every element of [M, N]

k

containing p (it is clear that such elements exist) is non-adjacent with S.

Therefore, [M, N]

k

is a maximal singular subspace of G

k

(Π). Every triangle

in a star is a star-triangle.

Proposition 4.16. The following assertions are fulﬁlled:

(1) every maximal singular subspace of G

n−1

(Π) is a line;

(2) if k ≤ n − 2 then every maximal singular subspace of G

k

(Π) is a top

or a star; in particular, all maximal singular subspaces of G

n−2

(Π) are

tops.

Proof. It was noted above that the class of maximal singular subspaces

of G

k

(Π) coincides with the class of maximal cliques of the Grassmann

graph Γ

k

(Π). The statement (1) follows from the second part of Lemma

4.10.

(2). Let A be a maximal clique of Γ

k

(Π) and k ≤ n−2. Suppose that A

is not a star. As in the proof of Proposition 3.2, we choose S

1

, S

2

, S

3

∈ A

which form a top-triangle in G

k

(Π). If S ∈ (

k

(Π) is not contained in the

(k + 1)-dimensional singular subspace

N := ¸S

1

, S

2

)

then S is non-adjacent with at least one of S

i

. This mean that S ,∈ A and

our clique coincides with the top ¸N]

k

.

Example 4.6. Let M be an m-dimensional singular subspace and m < k.

Then [M)

k

is a subspace of G

k

(Π); subspaces of such type are called

parabolic [Cooperstein, Kasikova and Shult (2005)]. This subspace is sin-

gular only in the case when k = n − 1 and m = n − 2. Suppose that

m < n−2. By Lemma 4.4, [M)

m+1

is a polar space of rank n−m−1. The

parabolic subspace [M)

k

can be identiﬁed with the Grassmann space of in-

dex k−m−1 associated with this polar space (every U ∈ [M)

k

corresponds

to [M, U]

m+1

).

Example 4.7. Let n ≥ 4. If 0 < k < n −1 and M, N is a pair of incident

singular subspaces of Π such that

dimM < k < dimN

then [M, N]

k

is a subspace of G

k

(Π) isomorphic to the Grassmann space

of a ﬁnite-dimensional vector space (in the case when dimM = k − 1 and

dimN = k + 1, we get a line). Subspaces of such type are called classical

[Cooperstein (2005)].

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

154 Grassmannians of Classical Buildings

Let S and U be adjacent elements of (

k

(Π). The set of all elements of

(

k

(Π) adjacent with both S and U is the union of all maximal cliques of

Γ

k

(Π) (maximal singular subspaces of G

k

(Π)) containing S and U. In the

case when k ,= n −2, the intersection of all these cliques is the line joining

S and U. For k = n − 2 this fails, since there is only one maximal clique

(the top ¸¸S, U)]

n−2

) containing S and U.

For every subset A ⊂ (

k

(Π) we denote by A

∼

the set consisting of all

elements of (

k

(Π) adjacent with every element of A. As for the Grassmann

spaces of ﬁnite-dimensional vector spaces, we have the following characte-

rization of lines in terms of the adjacency relation if k ,= n −2.

Proposition 4.17. Let k ,= n − 2. Then for any adjacent S, U ∈ (

k

(Π)

the line joining S and U coincides with the set ¦S, U¦

∼∼

.

4.5.2 Half-spin Grassmann spaces

Let Π = (P, L) be a polar space of type D

n

. We investigate the associated

half-spin Grassmann spaces G

δ

(Π), δ ∈ ¦+, −¦ (they are deﬁned only for

n ≥ 3). If n = 3 then both G

δ

(Π) are 3-dimensional projective spaces

(Subsection 4.3.3) and we will suppose that n ≥ 4.

Lemma 4.11. Let S

1

, S

2

, S

3

be a triangle in G

δ

(Π). Then there exist

unique M ∈ (

n−4

(Π) and U ∈ (

−δ

(Π) such that M ⊂ U,

M = S

1

∩ S

2

∩ S

3

and each S

i

has an (n −2)-dimensional intersection with U.

Proof. The subspaces

U

1

:= S

2

∩ S

3

, U

2

:= S

1

∩ S

3

, U

3

:= S

1

∩ S

2

belong to (

n−3

(Π). We have U

i

⊥ U

j

for all i, j ∈ ¦1, 2, 3¦ (since U

i

and

U

j

both are contained in S

l

with l ,= i, j). The equality U

l

= U

j

(l ,= j)

implies that U

l

is contained in all S

i

and we get U

1

= U

2

= U

3

which means

that S

1

, S

2

, S

3

are points on a line, a contradiction. Therefore, all U

i

are

distinct and the dimension of the singular subspace

U := ¸U

1

, U

2

, U

3

)

is not less than n −2.

Suppose that dimU = n −2 or two distinct U

i

, for example, U

1

and U

2

are non-adjacent. In each of these cases, the subspace U is spanned by U

1

and U

2

. Then U ⊂ S

3

(since U

1

and U

2

both are contained in S

3

) and

S

1

∩ S

2

= U

3

⊂ U ⊂ S

3

;

in other words, U

3

is contained in all S

i

which is impossible.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 155

So, U is a maximal singular subspace and U

1

, U

2

, U

3

are mutually ad-

jacent. This means that U

1

, U

2

, U

3

form a star-triangle in G

n−3

(Π). Then

the (n −4)-dimensional singular subspace

M := U

1

∩ U

2

∩ U

3

coincides with S

1

∩S

2

∩S

3

. Each S

i

intersects U in the (n−2)-dimensional

subspace ¸U

l

, U

m

), where l, m ,= i. Hence U belongs to (

−δ

(Π).

Let U

**be a maximal singular subspace of Π intersecting each S
**

i

in an

(n − 2)-dimensional subspace. Since Π is of type D

n

, U

∩ S

i

and U

∩ S

j

(i ,= j) are distinct and their intersection is (n − 3)-dimensional. On the

other hand, this intersection is contained in S

i

∩ S

j

= U

l

(l ,= i, j) which

implies that it coincides with U

l

. Therefore, each U

l

is contained in U

and

U = U

.

Proposition 4.18. G

δ

(Π) is a gamma space.

Proof. Let S

1

, S

2

, S

3

be a triangle in G

δ

(Π). We need to show that S

1

is collinear with all points of the line joining S

2

and S

3

. If S is a point on

this line then S

2

∩S

3

⊂ S; in particular, S contains the (n−4)-dimensional

singular subspace

S

1

∩ S

2

∩ S

3

(Lemma 4.11). This guarantees that S

1

∩ S is (n −3)-dimensional.

We want to describe maximal singular subspaces of G

δ

(Π). By Propo-

sition 4.18, the class of maximal singular subspaces coincides with the class

of maximal cliques of the associated Grassmann graph Γ

δ

(Π).

Let U ∈ (

−δ

(Π). We write [U]

δ

for the set of all elements of (

δ

(Π)

intersecting U in (n −2)-dimensional subspaces.

Proposition 4.19. For every U ∈ (

−δ

(Π) the set [U]

δ

is a maximal sin-

gular subspace of G

δ

(Π) isomorphic to an (n − 1)-dimensional projective

space.

Singular subspaces of such kind will be called special.

Proof. As in the proof of Lemma 4.9, we establish that [U]

δ

is a singular

subspace of G

δ

(Π) and a line [T)

δ

, T ∈ (

n−3

(Π), is contained in [U]

δ

if

and only if T ⊂ U. Every hyperplane of U is contained in precisely one

element of [U]

δ

; this correspondence deﬁnes a collineation between [U]

δ

and

the projective space dual to U. So, [U]

δ

is an (n−1)-dimensional projective

space.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

156 Grassmannians of Classical Buildings

For every S ∈ (

δ

(Π) ¸[U]

δ

there exists a frame B such that S and U are

spanned by subsets of B (Proposition 4.7). It is not diﬃcult to construct

¸X) ∈ [U]

δ

, X ⊂ B,

non-adjacent with S (we leave the details for the reader). This means that

[U]

δ

is a maximal singular subspace of G

δ

(Π).

Proposition 4.20. If M ∈ (

n−4

(Π) then [M)

δ

is a maximal singular sub-

space of G

δ

(Π) isomorphic to a 3-dimensional projective space.

Maximal singular subspace of such kind are said to be stars.

Proof. If S and U are distinct elements of [M)

δ

then they both contain

the (n−4)-dimensional singular subspace M which means that their inter-

section is (n−3)-dimensional and S, U are adjacent; moreover, the inclusion

M ⊂ S ∩ U guarantees that the line [S ∩ U)

δ

joining S and U is contained

in [M)

δ

. Hence [M)

δ

is a singular subspace of G

δ

(Π).

Let S

1

, S

2

, S

3

be a triangle contained in [M)

δ

and U be the associated

element of (

−δ

(Π) (see Lemma 4.11). Consider the subspace

A := [U]

δ

∩ [M)

δ

.

If [M

i

)

δ

(i = 1, 2) are distinct lines of A then

M

1

∩ M

2

= M and M

i

⊂ U, i = 1, 2;

in other words, M

1

, M

2

are adjacent elements of (

n−3

(Π) and the subspace

spanned by them is (n−2)-dimensional. There is a unique element of (

δ

(Π)

intersecting U in ¸M

1

, M

2

); it is a common point of our lines. Thus A is

a projective plane and S

1

, S

2

, S

3

form a base of A. We have proved that

every plane in [M)

δ

is projective.

Suppose that T ∈ (

n−3

(Π) contains M and T ,⊂ U. Then [T)

δ

is a line

of [M)

δ

which is not contained in A. The maximal singular subspace

¸T, U ∩ T

⊥

)

intersects U in the (n −2)-dimensional subspace U ∩ T

⊥

(by Lemma 4.2);

hence it belongs to A and the line [T)

δ

has a non-empty intersection with

A. This means that the plane A is a hyperplane of [M)

δ

. Therefore, [M)

δ

is a 3-dimensional projective space.

Let S ∈ (

δ

(Π) ¸ [M)

δ

. As in the proof of Proposition 4.19, we take a

frame B such that S and M are spanned by subsets of B and construct

¸X) ∈ [M)

δ

, X ⊂ B,

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 157

non-adjacent with S. This guarantees that [M)

δ

is a maximal singular

subspace of G

δ

(Π).

Proposition 4.21. Every maximal singular subspace of G

δ

(Π) is a star or

a special subspace.

Proof. We show that every maximal clique A of the Grassmann graph

Γ

δ

(Π) is a star or a special subspace. Lines are not maximal cliques of

Γ

δ

(Π) (they are contained, for example, in stars), thus A contains a triangle

S

1

, S

2

, S

3

. Let M and U be the associated elements of (

n−4

(Π) and (

−δ

(Π),

respectively (Lemma 4.11). Clearly, [M)

δ

and [U]

δ

are the unique star and

the unique special subspace containing our triangle.

First, we establish that every S ∈ (

δ

(Π) adjacent with all S

i

belongs to

[M)

δ

or [U]

δ

. Since U ∩ S

i

is a hyperplane of S

i

, the subspace S intersects

each U ∩ S

i

in a subspace whose dimension is not less than n − 4. One of

the following two possibilities is realized:

• S ∩ U ∩ S

1

= S ∩ U ∩ S

2

= S ∩ U ∩ S

3

,

• S ∩ U ∩ S

i

,= S ∩ U ∩ S

j

for some i, j.

In the ﬁrst case, the equality

M =

3

i=1

(U ∩ S

i

)

guarantees that

S ∩ U ∩ S

i

= S ∩ M

for each i. Since M is (n −4)-dimensional and the dimension of S ∩U ∩S

i

is not less than n − 4, we have M ⊂ S and S belongs to [M)

δ

. In the

second case, the dimension of S ∩ U is not less than n − 3; thus S ∩ U is

(n −2)-dimensional and S ∈ [U]

δ

.

Now consider

S ∈ [M)

δ

¸ [U]

δ

and S

∈ [U]

δ

¸ [M)

δ

.

Then S ∩ U and S

**∩ U span U (the ﬁrst subspace contains M and the
**

second is a hyperplane of U which does not contain M). Hence every point

of S ∩ S

**is collinear with all points of U and we get
**

S ∩ S

⊂ U

(since U is a maximal singular subspace). The latter inclusion implies that

S ∩ S

= (S ∩ U) ∩ (S

∩ U).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

158 Grassmannians of Classical Buildings

Since S ∩ U = M (this follows from S ∈ [M)

δ

¸ [U]

δ

) and S

∩ U does not

contain M, the subspaces S and S

**are not adjacent.
**

So, every element of [M)

δ

¸ [U]

δ

is non-adjacent with every element of

[U]

δ

¸ [M)

δ

. The inclusion

A ⊂ [M)

δ

∪ [U]

δ

shows that A coincides with [M)

δ

or [U]

δ

.

Example 4.8. Let M be an m-dimensional singular subspace of Π. Then

[M)

δ

is a subspace of G

δ

(Π); subspaces of such type are called parabolic

[Cooperstein, Kasikova and Shult (2005)]. In the case when m < n − 4,

this subspace is non-singular. As in Example 4.6, we consider [M)

m+1

.

This is a polar space of type D

n−m−1

. The parabolic subspace [M)

δ

can be

identiﬁed with one of the half-spin Grassmann spaces of this polar space.

Exercise 4.9. Show that the intersection of two distinct stars of G

δ

(Π) is

empty, or a single point or a line (the second possibility is not realized if

n = 4); moreover, this intersection is a line if and only if the associated

(n −4)-dimensional singular subspaces are adjacent.

Exercise 4.10. Show that the intersection of two distinct special subspaces

of G

δ

(Π) is empty or a line, the second possibility is realized if and only if

the associated elements of (

−δ

(Π) are adjacent.

If [M)

δ

and [U]

δ

are a star and a special subspace satisfying M ⊂ U

then their intersection is a plane (see the proof of Proposition 4.20).

Exercise 4.11. Show that every plane of G

δ

(Π) is contained in precisely

one star and precisely one special subspace. Hint: take any triangle in a

plane and consider the associated elements of (

n−4

(Π) and (

−δ

(Π) (Lemma

4.11).

As above, we write A

∼

for the set consisting of all elements of (

δ

(Π)

adjacent with all elements of A ⊂ (

δ

(Π). We have the standard character-

ization of lines in terms of the adjacency relation.

Proposition 4.22. For any adjacent S, U ∈ (

δ

(Π) the line joining S and

U coincides with the set ¦S, U¦

∼∼

.

Proof. The intersection of all maximal singular subspaces of G

δ

(Π) (stars

and special subspaces) containing both S and U coincides with the line

joining S and U.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 159

A few remarks concerning polar spaces of type D

4

ﬁnish the subsection.

Proposition 4.23. If n = 4 then G

δ

(Π) is a polar space of type D

4

.

Proof. The polar axioms (1), (3) and (4) hold and we verify (2). Let S

be a point of G

δ

(Π) which does not belong to a line [L)

δ

, L ∈ L. If S has

a non-empty intersection with L then it is collinear with all points of [L)

δ

.

If L and S are disjoint then S ∩ L

⊥

is a line of Π (Lemma 4.2) and

¸L, S ∩ L

⊥

)

is the unique point on the line [L)

δ

collinear with S. Therefore, G

δ

(Π)

is a polar space. Maximal singular subspaces of G

δ

(Π) (stars and special

subspaces) are 3-dimensional and every plane is contained in precisely one

star and one special subspace (Exercise 4.11). The polar space is of type

D

4

.

Exercise 4.12. Suppose that n = 4. Show that the natural one-to-one

correspondence between lines of G

δ

(Π) and lines of Π is a collineation of

G

1

(G

δ

(Π)) to G

1

(Π). Consider the half-spin Grassmann spaces of the polar

space G

δ

(Π): one of them consists of stars and the other is formed by special

subspaces; show that they are isomorphic to Π and G

−δ

(Π), respectively.

4.6 Collineations

Throughout the section we suppose that Π = (P, L) and Π

= (P

, L

) are

polar spaces of same type X

n

, X ∈ ¦C, D¦ and n ≥ 3; in the case when

X = D, we will require that n ≥ 4.

4.6.1 Chow’s theorem and its generalizations

Every collineation of Π to Π

induces a collineation of G

k

(Π) to G

k

(Π

) for

each k ∈ ¦1, . . . , n − 1¦. Moreover, if our polar spaces are of type D

n

then

we get a collineation of G

δ

(Π) to G

γ

(Π

) with δ, γ ∈ ¦+, −¦.

There is the following analogue of Theorem 3.2.

Theorem 4.7 ([Chow (1949)]). Every isomorphism of Γ

n−1

(Π) to

Γ

n−1

(Π

) is the collineation of G

n−1

(Π) to G

n−1

(Π

) induced by a

collineation of Π to Π

**. Moveover, if our polar spaces are of type D
**

n

(n ≥ 5) then every isomorphism of Γ

δ

(Π) to Γ

γ

(Π

), δ, γ ∈ ¦+, −¦, is

the collineation of G

δ

(Π) to G

γ

(Π

) induced by a collineation of Π to Π

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

160 Grassmannians of Classical Buildings

Remark 4.9. In [Chow (1949)] this result was established only for the

polar spaces associated with reﬂexive forms, but the method works in the

general case.

Remark 4.10. Suppose that Π and Π

**both are symplectic polar spaces
**

and write V and V

**for the associated (2n)-dimensional vector spaces. It
**

was proved in [Huang (2000)] that every surjection of (

n−1

(Π) to (

n−1

(Π

)

sending adjacent elements to adjacent elements is a collineation of G

n−1

(Π)

to G

n−1

(Π

**); in particular, every semicollineation of G
**

n−1

(Π) to G

n−1

(Π

)

is a collineation. The following more general version of this result was

established in [Huang (2001)]: if a mapping

f : (

n−1

(Π) → (

n−1

(Π

)

sends adjacent elements to adjacent elements and for every S ∈ (

n−1

(Π)

there exists U ∈ (

n−1

(Π) such that the distance between f(S) and f(U)

is maximal then f can be extended to the embedding of G

n

(V ) in G

n

(V

)

induced by a semilinear embedding of V in V

.

Theorem 4.8 ([Pankov, Pra˙ zmowski and

˙

Zynel (2006)]). Let f be

an isomorphism of Γ

k

(Π) to Γ

k

(Π

) and 0 ≤ k ≤ n −2. If

n ,= 4 or k ,= 1

then f is the collineation of G

k

(Π) to G

k

(Π

) induced by a collineation of

Π to Π

(f is a collineation of Π to Π

if k = 0).

In Subsection 4.6.3 we establish that an isomorphism of Γ

k

(Π) to Γ

k

(Π

)

(k ≤ n−3) is induced by a collineation of Π to Π

**if it transfers stars to stars
**

and tops to tops. It follows from elementary properties of triangles that the

latter condition holds for n ≥ 5. Using Chow’s idea, we show that every

isomorphism of Γ

n−2

(Π) to Γ

n−2

(Π

) induces an isomorphism of Γ

n−3

(Π)

to Γ

n−3

(Π

**) which sends stars to stars and tops to tops (Subsection 4.6.4).
**

This is a modiﬁcation of the proof given in [Pankov, Pra˙ zmowski and

˙

Zynel

(2006)]. Theorem 4.7 will be presented as a simple consequence of Theorem

4.8.

Now consider the remaining case n = 4 and k = 1. Suppose that our

polar spaces are of type D

4

. Then the half-spin Grassmann spaces are polar

spaces of type D

4

and there is natural one-to-one correspondence between

their lines and lines of the polar spaces (every line of G

δ

(Π), δ ∈ ¦+, −¦,

consists of all elements of (

δ

(Π) containing a certain line of Π). This means

that a collineation f of Π to G

δ

(Π

**) (if it exists) induces a bijection of L to
**

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 161

L

**; an easy veriﬁcation shows that this is a collineation of G
**

1

(Π) to G

1

(Π

).

It will be shown later that such collineations map all tops to stars and

some stars to tops. The collineation f also induces a collineation between

one of the half-spin Grassmann spaces of Π and G

−δ

(Π

) (since G

−δ

(Π

)

is one of the half-spin Grassmann spaces of the polar space G

δ

(Π

), see

Exercise 4.12); this collineation sends stars to special subspaces and special

subspaces to stars.

Theorem 4.9. If n = 4 then every isomorphism of Γ

1

(Π) to Γ

1

(Π

) is the

collineation of G

1

(Π) to G

1

(Π

) induced by a collineation of Π to Π

or a

collineation of Π to one of the half-spin Grassmann spaces of Π

; the second

possibility can be realized only in the case when the polar spaces are of type

D

4

.

Theorem 4.10. If Π and Π

are of type D

4

then every isomorphism of

Γ

δ

(Π) to Γ

γ

(Π

), δ, γ ∈ ¦+, −¦, is the collineation of G

δ

(Π) to G

γ

(Π

)

induced by a collineation of Π to Π

or a collineation of Π to G

−γ

(Π

).

4.6.2 Weak adjacency on polar Grassmannians

To prove Theorems 4.7 and 4.8 we will use elementary properties of so-called

weak adjacency relation deﬁned on (

k

(Π), 0 < k < n − 1. In Section 4.9

these properties also will be exploited to study apartments preserving

mappings.

Two elements of the Grassmannian (

k

(Π), 0 < k < n−1, are said to be

weakly adjacent if their intersection belongs to (

k−1

(Π). By this deﬁnition,

any two adjacent elements of (

k

(Π) are weakly adjacent; the converse fails.

The weak Grassmann graph Γ

w

k

(Π) is the graph whose vertex set is (

k

(Π)

and whose edges are pairs of weakly adjacent elements. Since Γ

k

(Π) is

a subgraph of Γ

w

k

(Π), the weak Grassmann graph is connected and every

clique of Γ

k

(Π) is a clique in Γ

w

k

(Π).

Example 4.9. As in Example 3.1, we show that every top of (

k

(Π) is a

maximal clique of Γ

w

k

(Π).

Example 4.10. For every M ∈ (

k−1

(Π) the parabolic subspace [M)

k

will

be called a big star. This is a clique of Γ

w

k

(Π) containing non-adjacent

elements. If S

1

and S

2

are distinct non-adjacent elements of [M)

k

and

S ∈ (

k

(Π) is weakly adjacent with both S

1

, S

2

then S belongs to [M)

k

.

Indeed, if S intersects S

1

and S

2

in distinct (k −1)-dimensional subspaces

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

162 Grassmannians of Classical Buildings

then we take points

p

i

∈ (S ∩ S

i

) ¸ (S

1

∩ S

2

), i = 1, 2;

since p

1

⊥ p

2

(these points belong to S) and

S

i

= ¸S

1

∩ S

2

, p

i

),

we get S

1

⊥ S

2

which contradicts the assumption that S

1

and S

2

are non-

adjacent. Therefore, big stars are maximal cliques of Γ

w

k

(Π).

Proposition 4.24. Every maximal clique of Γ

w

k

(Π) is a top or a big star.

Proof. Let A be a maximal clique of Γ

w

k

(Π). If any two distinct elements

of A are adjacent then it is a maximal clique of Γ

k

(Π); hence A is a top

(stars are non-maximal cliques of Γ

w

k

(Π), since they are proper subsets of

big stars). If S

1

and S

2

are distinct non-adjacent elements of A then every

element of A contains S

1

∩ S

2

(see Example 4.10) and A is the big star

corresponding to S

1

∩ S

2

.

4.6.3 Proof of Theorem 4.8 for k < n −2

Let f be an isomorphism of Γ

k

(Π) to Γ

k

(Π

) and k < n − 2. It follows

directly from the characterization of lines in terms of the adjacency relation

(Proposition 4.17) that f is a collineation of G

k

(Π) to G

k

(Π

) and Theorem

4.8 is true for k = 0. If k > 0 then f and f

−1

map maximal singular

subspaces (stars and tops) to maximal singular subspaces.

Lemma 4.12. If f is a collineation of G

k

(Π) to G

k

(Π

), k < n−2, trans-

ferring tops to tops and stars to stars then it is induced by a collineation of

Π to Π

.

Proof. Two stars [M, N]

k

and [M

, N

]

k

are called adjacent if M = M

and N is adjacent with N

. Two distinct stars A, A

**are non-adjacent if
**

and only if their intersection is empty or there exist distinct stars }, }

such

that A ∩ A

is a proper subset of } ∩ }

**. In other words, pairs of adjacent
**

stars can be characterized as pairs with maximal intersections. This means

that f and f

−1

maps adjacent stars to adjacent stars.

For every M ∈ (

k−1

(Π)

[M)

k

=

N∈ [M

n−1

[M, N]

k

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 163

Any two elements of [M)

n−1

can be connected by a path of Γ

n−1

(Π) con-

tained in [M)

n−1

(the parabolic subspace [M)

n−1

is isomorphic to the

Grassmann space consisting of maximal singular subspaces of a certain

polar space, Example 4.6). This implies the existence of a subspace

f

k−1

(M) ∈ (

k−1

(Π

)

such that the associated big star of (

k

(Π

) contains f([M)

k

). It is clear

that

f([M)

k

) = [f

k−1

(M))

k

for every M ∈ (

k−1

(Π) (we apply the same arguments to f

−1

) and the

mapping

f

k−1

: (

k−1

(Π) → (

k−1

(Π

)

is bijective. Since for every S ∈ (

k

(Π)

M ∈ ¸S]

k−1

⇔ S ∈ [M)

k

⇔ f(S) ∈ [f

k−1

(M))

k

⇔ f

k−1

(M) ∈ ¸f(S)]

k−1

,

we have

f

k−1

(¸S]

k−1

) = ¸f(S)]

k−1

.

Therefore, f

k−1

is an isomorphism of Γ

k−1

(Π) to Γ

k−1

(Π

) such that f

k−1

and the inverse mapping send tops to tops (this is a collineation of Π to

Π

if k = 1). If k ≥ 2 then f

k−1

maps stars to stars. As in the proof of

Theorem 3.2, we get a sequence of collineations

f

i

: (

i

(Π) → (

i

(Π

), i = k, . . . , 0,

such that f

k

= f and establish that every f

i

is induced by f

0

.

So, we need to show that f maps stars to stars and tops to tops, except

the case when n = 4 and k = 1. Since the dimension of stars and tops is

equal to

n −k − 1 and k + 1

(respectively), this is true if n ,= 2k + 2.

Suppose that n = 2k + 2. Every top-triangle is contained in precisely

one maximal singular subspace (a top). If S

1

, S

2

, S

3

∈ (

k

(Π) form a star-

triangle then the singular subspace

¸S

1

, S

2

, S

3

)

is (k + 2)-dimensional; in the case when k > 1, this singular subspace

is not maximal and the star-triangle is contained in more than one star.

Therefore, if k > 1 then star-triangles go to star-triangles, top-triangles go

to top-triangles and we get the claim.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

164 Grassmannians of Classical Buildings

4.6.4 Proof of Theorems 4.7 and 4.8

Let f be an isomorphism of Γ

n−2

(Π) to Γ

n−2

(Π

**). Maximal cliques of these
**

graphs are tops and f induces a bijection

g : (

n−1

(Π) → (

n−1

(Π

).

This mapping transfers a line [S)

n−1

, S ∈ (

n−2

(Π), to the line [f(S))

n−1

.

Hence g is a collineation of G

n−1

(Π) to G

n−1

(Π

) and

dim(g(M) ∩ g(N)) = dim(M ∩ N)

for all M, N ∈ (

n−1

(Π) (the latter equality follows from the distance for-

mula given in Proposition 4.8).

Now we show that f is an isomorphism of Γ

w

n−2

(Π) to Γ

w

n−2

(Π

).

Let S

1

, S

2

be weakly adjacent elements of (

n−2

(Π). Let also M

1

, M

2

be

maxima singular subspaces of Π

such that

M

1

∩ M

2

= f(S

1

) ∩ f(S

2

)

and f(S

i

) ⊂ M

i

for i = 1, 2 (Corollary 4.2). If f(S

1

) and f(S

2

) are not

weakly adjacent then

dim(M

1

∩ M

2

) < n −3

and

dim(g

−1

(M

1

) ∩ g

−1

(M

2

)) < n −3.

Since g is induced by f, we have

S

i

⊂ g

−1

(M

i

), i = 1, 2,

and the latter inequality contradicts the fact that S

1

and S

2

are weakly

adjacent. Similarly, we establish that f

−1

preserves the weak adjacency

relation.

Since f and f

−1

transfer tops to tops, big stars go to big stars in both

directions. This means that f induces a bijection

h : (

n−3

(Π) → (

n−3

(Π

).

In the case when n = 3, this is a collineation of Π to Π

which induces f.

If n ≥ 4 then h is a collineation of G

n−3

(Π) to G

n−3

(Π

) preserving the

types of all maximal singular subspaces. By Lemma 4.12, h is induced by

a collineation of Π to Π

**; this collineation induces f.
**

Theorem 4.8 is proved and we use it to prove Theorem 4.7.

Since maximal cliques of Γ

n−1

(Π) and Γ

n−1

(Π

**) are lines of the associ-
**

ated Grassmann spaces, every isomorphism t between these graphs induces

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 165

a bijection

˜

t of (

n−2

(Π) to (

n−2

(Π

**). This bijection is an isomorphism of
**

Γ

n−2

(Π) to Γ

n−2

(Π

**) (two distinct lines have a non-empty intersection if
**

and only if the associated (n − 2)-dimensional singular subspaces are ad-

jacent). By Theorem 4.8,

˜

t is induced by a collineation of Π to Π

; this

collineation induces t.

Now suppose that our polar spaces are of type D

n

and f is an isomor-

phism of Γ

δ

(Π) to Γ

γ

(Π

), δ, γ ∈ ¦+, −¦. Then f is a collineation of G

δ

(Π)

to G

γ

(Π

**) (by Proposition 4.22, lines of half-spin Grassmann spaces can be
**

characterized in terms of the adjacency relation). Hence f maps lines to

lines and induces a bijection of (

n−3

(Π) to (

n−3

(Π

**). Since two distinct
**

(n −3)-dimensional singular subspaces of Π or Π

**are adjacent if and only
**

if the associated lines of the half-spin Grassmann space G

δ

(Π) or G

γ

(Π

)

span a plane, this bijection is an isomorphism of Γ

n−3

(Π) to Γ

n−3

(Π

). In

the case when n ≥ 5, we apply Theorem 4.8 and get the claim.

Similarly, we draw Theorem 4.10 from Theorem 4.9.

4.6.5 Proof of Theorem 4.9

Throughout the subsection we assume that n = 4.

Suppose that f is an isomorphism of Γ

1

(Π) to Γ

1

(Π

). It was noted in

Subsection 4.6.3 that f is a collineation of G

1

(Π) to G

1

(Π

); moreover, f

is induced by a collineation of Π to Π

**if it preserves the types of maximal
**

singular subspaces (Lemma 4.12). Now we suppose that the image of a

certain star is a top; under this assumption, we establish that our polar

spaces are of type D

4

and f is induced by a collineation of Π to one of the

half-spin Grassmann spaces of Π

**. Therefore, if the polar spaces are of type
**

C

4

then f and f

−1

both map stars to stars (hence tops go to tops) and f

is induced by a collineation of Π to Π

.

0. Preliminaries. Our proof will be based on some trivial observations

concerning pairs of triangles. In the present case (n = 4, k = 1), every

triangle is contained in precisely one maximal singular subspace.

We say that triangles

∆ = ¦L

1

, L

2

, L

3

¦ and ∆

= ¦L

1

, L

2

, L

3

¦, L

i

,= L

j

∀ i, j, (4.3)

in G

1

(Π) form a regular pair if L

i

and L

j

are adjacent only in the case when

i ,= j; in other words, every line from each of these triangles is adjacent

with precisely two lines of the other.

Lemma 4.13. If the triangles (4.3) form a regular pair then one of the

following possibilities is realized:

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

166 Grassmannians of Classical Buildings

• There exist a plane S and a point p ,∈ S, p ⊥ S, such that one of the

triangles is a star-triangle in [p, ¸p, S)]

1

and the other is a top-triangle

in ¸S]

1

.

• There exist maximal singular subspaces U, U

intersecting precisely in a

point p and such that ∆ and ∆

**are star-triangles in [p, U]
**

1

and [p, U

]

1

,

respectively.

Proof. Direct veriﬁcation.

We say that triangles

∆ = ¦L, L

1

, L

2

¦ and ∆

= ¦L, L

1

, L

2

¦, L

i

,= L

j

∀ i, j, (4.4)

in G

1

(Π) form a regular pair if L

i

and L

j

are adjacent only in the case when

i = j.

Lemma 4.14. If the triangles (4.4) form a regular pair then one of the

following possibilities is realized:

• There exist two planes S, S

**intersecting precisely in L and such that
**

S ⊥ S

and ∆, ∆

**are top-triangles in ¸S]
**

1

and ¸S

]

1

, respectively.

• There exist two maximal singular subspaces U, U

intersecting precisely

in L and a point p ∈ L such that ∆, ∆

**are star-triangles in [p, U]
**

1

and

[p, U

]

1

, respectively.

Proof. Direct veriﬁcation.

1. Let f be an injection of (

1

(Π) to (

1

(Π

**) preserving the adjacency
**

relation: two elements of (

1

(Π) are adjacent if and only if their images

are adjacent. Then f transfers maximal singular subspaces of G

1

(Π) (stars

and tops) to subsets of maximal singular subspaces of G

1

(Π

). Every max-

imal singular subspace of G

1

(Π

**) contains at most one image of a maximal
**

singular subspace of G

1

(Π) (otherwise there exist non-adjacent elements

of (

1

(Π) whose images are adjacent). We will assume that f satisﬁes the

following condition:

(A) triangles go to triangles.

It is clear that two triangles in G

1

(Π) form a regular pair if and only if

their images form a regular pair in G

1

(Π

**). The condition (A) guarantees
**

that the image of every maximal singular subspace of G

1

(Π) is contained

in precisely one maximal singular subspace of G

1

(Π

**) (since triangles are
**

bases of maximal singular subspaces).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 167

Suppose that the image of a certain star [p, U]

1

, U ∈ (

3

(Π), p ∈ U, is

contained in a top ¸S

]

1

, S

∈ (

2

(Π

**). Our ﬁrst step is the following.
**

Lemma 4.15. There exists a maximal singular subspace U

of Π

contain-

ing S

**and satisfying the following conditions:
**

• for every point q ∈ U there is a plane S(q) ⊂ U

such that

f([q, U]

1

) ⊂ ¸S(q)]

1

,

• for every plane S ⊂ U there is a point q(S) ∈ U

such that

f(¸S]

1

) ⊂ [q(S), U

]

1

.

Proof. Let S be a plane in U which does not contain the point p. Con-

sider any regular pair of triangles

∆

1

⊂ [p, U]

1

and ∆

2

⊂ ¸S]

1

.

Their images also form a regular pair in G

1

(Π

). By our hypothesis, f(∆

1

)

is a top-triangle contained in ¸S

]

1

. Then Lemma 4.13 guarantees that

f(∆

2

) is a star-triangle; moreover, there exist a maximal singular subspace

U

of Π

and a point q(S) ∈ U

such that

q(S) ,∈ S

⊂ U

and f(∆

2

) is contained in the star [q(S), U

]

1

. The latter means that f(¸S]

1

)

is a subset of [q(S), U

]

1

.

Similarly, for another plane T ⊂ U which does not contain p, the image

of the top ¸T]

1

is contained in a certain star [q(T), U

]

1

and

q(T) ,∈ S

⊂ U

.

If U

and U

are distinct then their intersection is S

**. Since the points q(S)
**

and q(T) do not belong to S

,

[q(S), U

]

1

∩ [q(T), U

]

1

= ∅.

On the other hand,

f(¸S]

1

) ⊂ [q(S), U

]

1

and f(¸T]

1

) ⊂ [q(T), U

]

1

;

the tops ¸S]

1

and ¸T]

1

have a non-empty intersection (since S∩T is a line),

a contradiction. Therefore, U

coincides with U

.

Now, let q ∈ U ¸ ¦p¦. We choose a plane S ⊂ U which does not contain

the points q and p. Then f(¸S]

1

) is contained in the star [q(S), U

]

1

(it was

established above). We consider any regular pair of triangles

∆

1

⊂ [q, U]

1

and ∆

2

⊂ ¸S]

1

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

168 Grassmannians of Classical Buildings

Their images form a regular pair and f(∆

2

) is a star-triangle in [q(S), U

]

1

.

If f(∆

1

) is a star-triangle then, by Lemma 4.13, f transfers the star [q, U]

1

to a subset of a certain star [q(S), U

]

1

such that

U

∩ U

= q(S).

Since S

⊂ U

**, the latter equality guarantees that
**

[q(S), U

]

1

∩ ¸S

]

1

= ∅.

However, [p, U]

1

and [q, U]

1

have a non-empty intersection (the line qp)

and their images are contained in ¸S

]

1

and [q(S), U

]

1

, respectively. Thus

f(∆

1

) is a top-triangle. This means that f([q, U]

1

) is contained in a certain

top ¸S(q)]

1

with S(q) ⊂ U

.

Consider a plane S ⊂ U containing the point p and any point q ∈ U ¸S.

Then f([q, U]

1

) is contained in the top ¸S(q)]

1

. Using Lemma 4.13, we

establish that f(¸S]

1

) is a subset of a certain star [q(S), U

]

1

such that

q(S) ,∈ S(q) ⊂ U

.

We take any plane T ⊂ U which does not contain the points p and q. The

image of the top ¸T]

1

is contained in the star [q(T), U

]

1

and q(T) ,∈ S(q).

As above, we show that U

coincides with U

.

Since distinct maximal singular subspaces go to subsets of distinct max-

imal singular subspaces, the mappings

q → S(q) and S → q(S)

(we deﬁne S(p) := S

) are injective.

In the case when f is a collineation of G

1

(Π) to G

1

(Π

), the inclusions

in Lemma 4.15 must be replaced by the equalities. Moreover, the inverse

mapping f

−1

sends every star

[q(S), U

]

1

, S ∈ ¸U]

2

,

to the top ¸S]

1

. We apply the arguments from the proof of Lemma 4.15

to f

−1

and establish that q → S(q) is a one-to-one correspondence be-

tween points of U and planes of U

**; similarly, S → q(S) is a one-to-one
**

correspondence between planes of U and points of U

.

2. Suppose that f is a collineation of G

1

(Π) to G

1

(Π

) and U is as in

the previous step. Let Q be a maximal singular subspace of Π intersecting

U in a certain plane S. The image of the top ¸S]

1

is the star [q(S), U

]

1

.

Let q ∈ Q¸ S. We take any regular pair of triangles

∆

1

⊂ ¸S]

1

and ∆

2

⊂ [q, Q]

1

.

Then f(∆

1

) is a star-triangle in [q(S), U

]

1

. By Lemma 4.13, there are the

following two possibilities for the triangle f(∆

2

):

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 169

• f(∆

2

) is a top-triangle then f([q, Q]

1

) coincides with a certain top ¸T

]

1

with T

⊂ U

,

• f(∆

2

) is a star-triangle then f([q, Q]

1

) is a star [q(S), Q

]

1

such that Q

intersects U

**precisely in the point q(S).
**

The ﬁrst possibility is not realized, since for any plane T

⊂ U

the top

¸T

]

1

is the image of a certain star [t, U]

1

.

Now, suppose that q ∈ S. If the image of the star [q, Q]

1

is a top then,

by the ﬁrst step of our proof, the same holds for all points of Q which is

impossible. Therefore, f([q, Q]

1

) is a star for every point q ∈ Q.

Let M be a third maximal singular subspace containing S. We take

points q ∈ Q ¸ S and t ∈ M ¸ S. Then every element of [q, Q]

1

is non-

adjacent with every element of [t, M]

1

. However,

f([q, Q]

1

) = [q(S), Q

]

1

contains elements adjacent with some elements of

f([t, M]

1

) = [q(S), M

]

1

.

This means that there are no 3 distinct maximal singular subspaces con-

taining S and our polar spaces are of type D

4

.

3. Suppose that the polar spaces are of type D

4

and, as in the ﬁrst

step, f is an injection of (

1

(Π) to (

1

(Π

**) preserving the adjacency relation
**

and satisfying the condition (A). We also assume that the maximal singular

subspaces U and U

belong to (

δ

(Π) and (

γ

(Π

), δ, γ ∈ ¦+, −¦, respectively.

Let Q be an element of (

δ

(Π) adjacent with U. The intersection of U

and Q is a certain line L. We take a point q ∈ L and consider any regular

pair of star-triangles

∆

1

= ¦L, L

1

, L

2

¦ ⊂ [q, U]

1

and ∆

2

= ¦L, L

1

, L

2

¦ ⊂ [q, Q]

1

.

The image of the star [q, U]

1

is contained in the top ¸S(q)]

1

and f(∆

1

) is

a top-triangle. By Lemma 4.14, f(∆

2

) also is a top-triangle which means

that f([q, Q]

1

) is a subset of a certain top.

Lemma 4.15 implies the existence of a maximal singular subspace Q

of Π

**satisfying the following conditions: for every point q ∈ Q there is a
**

plane T(q) ⊂ Q

such that

f([q, Q]

1

) ⊂ ¸T(q)]

1

,

and for every plane S ⊂ Q there is a point t(S) ∈ Q

such that

f(¸S]

1

) ⊂ [t(S), Q

]

1

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

170 Grassmannians of Classical Buildings

Now we choose two planes S ⊂ U and M ⊂ Q intersecting in L and

satisfying S ⊥ M. Consider any regular pair of top-triangles

∆

1

= ¦L, L

1

, L

2

¦ ⊂ ¸S]

1

and ∆

2

= ¦L, L

1

, L

2

¦ ⊂ ¸M]

1

.

Then

f(∆

1

) ⊂ [q(S), U

]

1

and f(∆

2

) ⊂ [t(M), Q

]

1

.

Since these star-triangles form a regular pair, Lemma 4.14 implies that

q(S) = t(M) and U

∩Q

= f(L). In particular, Q

is an element of (

γ

(Π

)

adjacent with U

.

The half-spin Grassmann spaces are connected and the same holds for

every Q ∈ (

δ

(Π). Thus we get a mapping

g : (

δ

(Π) → (

γ

(Π

)

satisfying the following conditions: if M ∈ (

δ

(Π) then for any point q ∈ M

there is a plane S

M

(q) ⊂ g(M) such that

f([q, M]

1

) ⊂ ¸S

M

(q)]

1

,

and for any plane S ⊂ M there is a point q

M

(S) ∈ g(M) such that

f(¸S]

1

) ⊂ [q

M

(S), g(M)]

1

.

Remark 4.11. The mapping g sends adjacent elements to adjacent ele-

ments; but it does not need to be injective.

In the case when f is a collineation of G

1

(Π) to G

1

(Π

), we apply the

same arguments to f

−1

and get the following:

• the latter two inclusions must be replaced by the equalities;

• q → S

M

(q) is a one-to-one correspondence between points of M and

planes of g(M); similarly, S → q

M

(S) is a one-to-one correspondence

between planes of M and points of g(M);

• the mapping g is bijective.

Since every plane of Π is contained in a certain element of (

δ

(Π), the im-

ages of all tops are subsets of stars (in the general case). If f is a collineation

of G

1

(Π) to G

1

(Π

) then f and f

−1

map tops to stars; the maximal singu-

lar subspaces associated with these stars are elements of (

δ

(Π) and (

γ

(Π

),

respectively.

4. As in the previous step, we assume that the polar spaces are of

type D

4

and f is an injection of (

1

(Π) to (

1

(Π

**) preserving the adjacency
**

relation and satisfying the condition (A).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 171

We take any point q ∈ P and consider the associated star [q)

δ

in the

half-spin Grassmann space G

δ

(Π). For every M ∈ [q)

δ

the image of the star

[q, M]

1

is contained in the top ¸S

M

(q)]

1

. Let M and N be distinct elements

of [q)

δ

. The stars [q, M]

1

and [q, N]

1

have a non-empty intersection (the

line M ∩ N). This means that the planes S

M

(q) and S

N

(q) are weakly

adjacent. Thus the set

¦S

M

(q)¦

M∈[q

δ

(4.5)

is a clique of the weak Grassmann graph Γ

w

2

(Π

). Every element of [q, M]

1

is adjacent with a certain element of [q, N]

1

distinct from the line M ∩ N.

Hence every element of ¸S

M

(q)]

1

is adjacent with an element of ¸S

N

(q)]

1

distinct from f(M ∩ N). The latter guarantees that

S

M

(q) ⊥ S

N

(q)

and these planes are adjacent. Therefore, all elements of (4.5) are contained

in a certain maximal singular subspace U

q

(the set (4.5) contains more than

one element and there is only one maximal singular subspace satisfying this

condition). Since each

g(M) ∈ (

γ

(Π

), M ∈ [q)

δ

,

intersects U

q

in the plane S

M

(q), we have U

q

∈ (

−γ

(Π

).

Denote by h the mapping of P to (

−γ

(Π

**) which sends every point q ∈ P
**

to the maximal singular subspace U

q

.

5. Now, assume that f is a collineation of G

1

(Π) to G

1

(Π

).

Let A be a maximal clique of Γ

w

2

(Π

**) containing (4.5). For every element
**

of A the mapping f

−1

transfers the associated top to a star

[t, T]

1

, T ∈ (

δ

(Π).

It has a non-empty intersection with every star

[q, M]

1

, M ∈ [q)

δ

,

which implies that t = q. Hence T belongs to [q)

δ

and A coincides with

(4.5). This maximal clique is the top ¸U

q

]

2

.

The latter guarantees that the mapping h is injective. Show that h is

bijective.

Let T

∈ (

−γ

(Π

) and S

1

, S

2

be distinct planes contained in T

. The

intersection of these planes is a certain line L

. The mapping f

−1

transfers

the tops ¸S

i

]

1

to stars [q

i

, Q

i

]

1

, i = 1, 2; both Q

i

are elements of (

δ

(Π). For

any regular pair of top-triangles

∆

1

= ¦L

, L

1

, L

2

¦ ⊂ ¸S

1

]

1

and ∆

2

= ¦L

, L

1

, L

2

¦ ⊂ ¸S

2

]

1

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

172 Grassmannians of Classical Buildings

the star-triangles

f

−1

(∆

1

) ⊂ [q

1

, Q

1

]

1

and f

−1

(∆

2

) ⊂ [q

2

, Q

2

]

1

form a regular pair. By Lemma 4.14, we have q

1

= q

2

. Therefore, T

= U

q

for a certain point q ∈ P and we get the claim.

Let L be a line of Π. For every point q ∈ L there exists a star

[q, Q]

1

, Q ∈ (

δ

(Π),

containing L. The image of this star is a top ¸M

]

1

with M

⊂ U

q

and we

obtain that f(L) ⊂ U

q

. Conversely, suppose that the line f(L) is contained

in U

q

. Then f(L) is an element of a certain top ¸M

]

1

, M

⊂ U

q

. The

mapping f

−1

sends this top to a star [q, Q]

1

; since L belongs to this star,

we have q ∈ L.

Thus h transfers every line L ∈ L to the line of G

−γ

(Π

) associated with

the line f(L). This means that h is a collineation of Π to G

−γ

(Π

) and f

is induced by this collineation.

Remark 4.12. In the case when f is not a collineation, the mapping h

transfers every line L ∈ L to a subset of the line of G

−γ

(Π

) associated

with f(L).

4.6.6 Remarks

In this subsection we establish some results closely related with Theorems

4.8 and 4.9.

Proposition 4.25. Let k, m ∈ ¦0, . . . , n−1¦ be distinct numbers. Suppose

that

g

k

: (

k

(Π) → (

k

(Π

) and g

m

: (

m

(Π) → (

m

(Π

)

are bijections such that S ∈ (

k

(Π) and U ∈ (

m

(Π) are incident if and only

if g

k

(S) and g

m

(U) are incident. Then g

k

and g

m

are induced by the same

collineation of Π to Π

.

Proof. Similar to Proposition 3.4.

As a consequence, we get the following.

Theorem 4.11 ([Pankov, Pra˙ zmowski and

˙

Zynel (2006)]). Suppose

that a bijection

f : (

k

(Π) → (

k

(Π

), k < n −1,

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 173

preserves the relation ⊥:

S ⊥ U ⇐⇒ f(S) ⊥ f(U).

Then it is the collineation of G

k

(Π) to G

k

(Π

) induced by a collineation of

Π to Π

.

Proof. Every subspaces ¸S]

k

, S ∈ (

n−1

(Π), can be characterized as a

maximal subset A ⊂ (

k

(Π) satisfying the following condition:

S ⊥ U ∀ S, U ∈ A.

This means that f induces a bijection of (

n−1

(Π) to (

n−1

(Π

) and we get

a pair of bijections satisfying the condition of Proposition 4.25.

We will investigate isomorphisms between the weak Grassmann graphs

and need the following lemma.

Lemma 4.16. Two distinct elements S, U ∈ (

k

(Π), 1 ≤ k ≤ n − 2, are

adjacent if and only if they belong to the intersection of two distinct maximal

cliques of Γ

w

k

(Π).

Proof. If S and U are adjacent then the big star [S ∩ U)

k

and the top

¸¸S, U)]

k

are as required. Conversely, suppose that there are two distinct

maximal cliques of Γ

w

k

(Π) containing both S and U. Since the intersection

of two distinct maximal cliques of the same type (two tops or two big stars)

is empty or a single point, one of these cliques is a big star and the other

is a top. This implies that S and U are adjacent.

Theorem 4.12 ([Pankov, Pra˙ zmowski and

˙

Zynel (2006)]). Let 1 ≤

k ≤ n − 2. Then every isomorphism of Γ

w

k

(Π) to Γ

w

k

(Π

) is induced by a

collineation of Π to Π

.

Proof. Let f be an isomorphism of Γ

w

k

(Π) to Γ

w

k

(Π

). Then f and f

−1

transfer maximal cliques (tops and big stars) to maximal cliques. By

Lemma 4.16, f is an isomorphism of Γ

k

(Π) to Γ

k

(Π

). Therefore, it is

induced by a collineation of Π to Π

**if n ,= 4 or k ,= 1. Consider the case
**

when n = 4 and k = 1. The mappings f and f

−1

send tops to tops and big

stars to big stars (since any two distinct elements of a top are adjacent and

a big star contains non-adjacent elements). Hence f induces a bijection of

P to P

. This is the required collineation of Π to Π

.

Remark 4.13. The collineation of G

1

(Π) to G

1

(Π

**) considered in the pre-
**

vious subsection (the case when Π and Π

are of type D

4

) does not preserve

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

174 Grassmannians of Classical Buildings

the relation ⊥ and it is not an isomorphism between the weak Grassmann

graphs. For every point q ∈ P it transfers the big stars [q)

1

to ¸U

q

]

1

, where

U

q

is a maximal singular subspace of Π

.

4.7 Opposite relation

As for Grassmannians of ﬁnite-dimensional vector spaces, two elements of

a polar or half-spin Grassmannian are said to be opposite if the distance

between them is maximal (is equal to the diameter of the associated Grass-

mann graph).

4.7.1 Opposite relation on polar Grassmannians

Let Π be a polar space of rank n. Two points of Π (elements of (

0

(Π))

are opposite if and only if they are non-collinear. Two maximal singular

subspaces are opposite if and only if they are disjoint.

The direct analogue of Theorem 3.6 does not hold in the general case;

in other words, the following two conditions are not equivalent:

(1) S

1

, S

2

∈ (

k

(Π) are adjacent,

(2) there exists S ∈ (

k

(Π) ¸ ¦S

1

, S

2

¦ such that every U ∈ (

k

(Π) opposite

to S is opposite to at least one of S

i

.

Example 4.11 ([Kwiatkowski and Pankov (2009)]). We consider a

(2n)-dimensional vector space V over a ﬁeld R of characteristic 2 and a

non-degenerate alternating form Ω deﬁned on V . Let

p

1

, . . . , p

n

, q

1

, . . . , q

n

be a frame of the polar space Π

Ω

such that p

i

,⊥ q

i

for each i. There exist

vectors x

1

, . . . , x

n

, y

1

, . . . , y

n

∈ V satisfying

p

i

= ¸x

i

), q

i

= ¸y

i

) and Ω(x

i

, y

i

) = Ω(y

i

, x

i

) = 1.

The maximal singular subspaces of Π

Ω

corresponding to the maximal to-

tally isotropic subspaces

¸x

1

, x

2

, x

3

, . . . , x

n

), ¸y

1

, y

2

, x

3

, . . . , x

n

),

and

¸x

1

+ y

2

, x

2

+ y

1

, x

3

, . . . , x

n

)

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 175

will be denoted by S

1

, S

2

, and S (respectively). Their intersection is the

(n−3)-dimensional singular subspace N associated with the totally isotropic

subspace ¸x

3

, . . . , x

n

). We assert that for any collinear points

p

∈ S

1

¸ N and q

∈ S

2

¸ N

the line p

q

intersects S. Indeed,

p

= ¸ax

1

+ bx

2

+ z

) and q

= ¸ay

2

+ by

1

+ z

)

for some a, b ∈ R and z

, z

∈ N; every point of the line p

q

is of type

¸s(ax

1

+ bx

2

) + t(ay

2

+ by

1

) + z)

with z ∈ N and s, t ∈ R; this point belongs to S if s = t = 1. Thus

if U ∈ (

n−1

(Π

Ω

) intersects both S

1

and S

2

then it intersects S. This

means that every element of (

n−1

(Π

Ω

) opposite to S is opposite to at least

one of S

i

. However, S

1

and S

2

are not adjacent. In the case when the

characteristic is not equal to 2, this construction is impossible.

Remark 4.14 (M. Kwiatkowski). Let p and q be distinct points of Π.

We say that the Veldkamp line (the line in the associated Veldkamp space,

Subsection 4.3.4) joining p

⊥

and q

⊥

is thick if it contains t

⊥

, t ,= p, q;

otherwise, it is said to be thin. Clearly, the line is thick if p ⊥ q; and

it is thin if p ,⊥ q and Π is of type D

n

. It is not diﬃcult to show that all

Veldkamp lines are thick if Π is the polar space deﬁned by an alternating or

Hermitian form. The points p and q satisfy (2) if and only if the Veldkamp

line joining p

⊥

and q

⊥

is thick: we take any point t such that t

⊥

belongs

to the Veldkamp line, then every point non-collinear with t is non-collinear

with p or q. Thus the conditions (1) and (2) are not equivalent if k = 0

and Π is, for example, the polar space associated with an alternating or

Hermitian form.

4.7.2 Opposite relation on half-spin Grassmannians

The direct analogue of Theorem 3.6 holds for half-spin Grassmannians.

Theorem 4.13 ([Kwiatkowski and Pankov (2009)]). If Π is a polar

space of type D

n

, n ≥ 4, then for any distinct S

1

, S

2

∈ (

δ

(Π), δ ∈ ¦+, −¦,

the following conditions are equivalent:

(1) S

1

and S

2

are adjacent,

(2) there exists S ∈ (

δ

(Π) ¸ ¦S

1

, S

2

¦ such that every U ∈ (

δ

(Π) opposite

to S is opposite to at least one of S

i

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

176 Grassmannians of Classical Buildings

Corollary 4.3. If Π and Π

**are polar spaces of type D
**

n

, n ≥ 4, then

every bijection of (

δ

(Π) to (

γ

(Π

**), δ, γ ∈ ¦+, −¦, preserving the relation
**

to be opposite (S, U ∈ (

δ

(Π) are opposite if and only if their images are

opposite) is a collineation of G

δ

(Π) to G

γ

(Π

).

Our proof of Theorem 4.13 is a modiﬁcation of the proof of Theorem

3.6 given in Subsection 3.2.4; a veriﬁcation of the conditions proposed in

[Huang and Havlicek (2008)] (see Remark 3.6) is not simpler.

Proof. We will distinguish the following two cases:

• n is even then the dimension of the intersection of two elements of

(

δ

(Π) is odd and S, U ∈ (

δ

(Π) are opposite if and only if S ∩ U = ∅;

• n is odd then the dimension of the intersection of two elements of (

δ

(Π)

is even and S, U ∈ (

δ

(Π) are opposite if and only if S ∩ U is a single

point.

(1) =⇒ (2). Suppose that S

1

, S

2

are adjacent and show that every

S ,= S

1

, S

2

belonging to the line of G

δ

(Π) joining S

1

and S

2

is as required.

Suppose that U ∈ (

δ

(Π) is opposite to S, but it is non-opposite to each S

i

.

If n is even then U intersects S

1

and S

2

in subspaces whose dimensions

are not less than 1. We take lines L

i

contained in U ∩ S

i

, i = 1, 2. Since

S

1

∩ S

2

⊂ S and S is opposite to U, these lines do not intersect S

1

∩ S

2

.

Hence S

i

is spanned by S

1

∩ S

2

and L

i

. The latter means that L

1

,⊥ L

2

which contradicts the fact that these lines are contained in U.

Consider the case when n is odd. According our assumption, the dimen-

sions of U ∩ S

1

and U ∩ S

2

are not less than 2. Let P

i

be planes contained

in U ∩ S

i

, i = 1, 2. Each of these planes has a non-empty intersections

with S

1

∩ S

2

(because S

1

∩ S

2

is (n −3)-dimensional). Since U is opposite

to S and S

1

∩ S

2

⊂ S, these intersections both are 0-dimensional (points).

This implies the existence of lines L

i

⊂ P

i

(i = 1, 2) which do not intersect

S

1

∩ S

2

. As above, we get L

1

,⊥ L

2

.

(2) =⇒(1). Our ﬁrst step is the following statement: for every distinct

collinear points p

i

∈ S

i

, i = 1, 2, the line p

1

p

2

intersects S.

Suppose that n is even. If p

1

p

2

does not intersect S then we take any

frame of Π whose subsets span S and the line p

1

p

2

. Let U be the maximal

singular subspace spanned by points of the frame and disjoint from S. This

is an element of (

δ

(Π) opposite to S. By our hypothesis, U does not

intersect at least one of S

i

which contradicts the inclusion p

1

p

2

⊂ U.

Now consider the case when n is odd. The intersection of S

1

and S

2

is non-empty. We assert that there exists a plane containing p

1

p

2

and

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 177

intersecting each S

i

in a line. If p

1

p

2

does not intersect S

1

∩ S

2

or the

subspace S

1

∩ S

2

contains more than one point (in the second case the

dimension of S

1

∩ S

2

is not less than 2) then we take any point t ∈ S

1

∩S

2

which does not belong to p

1

p

2

; the plane spanned by p

1

p

2

and t is as

required. If S

1

∩ S

2

is a single point belonging to the line p

1

p

2

then this

point coincides with p

1

or p

2

. Suppose that S

1

∩ S

2

= ¦p

1

¦. Then any

point t ∈ S

1

¸ S

2

collinear with p

2

gives the claim (such point exists, since

S

1

∩ p

⊥

2

is (n −2)-dimensional and the dimension of S

1

∩ S

2

is equal to 0).

The case when S

1

∩ S

2

= ¦p

2

¦ is similar.

Let P be a plane containing p

1

p

2

and intersecting each S

i

in a line. If

dim(P ∩ S) ≤ 0 then we take a frame whose subsets span P and S; using

this frame, we construct U ∈ (

δ

(Π) opposite to S and containing P. Then

U is opposite to at least one of S

i

. Since P ⊂ U, the latter contradicts the

fact that P intersects each S

i

in a line. Therefore, P ∩ S contains a line

and this line intersects p

1

p

2

(since P is a plane).

Our second step is to prove the equality

dim(S ∩ S

i

) = n −3, i = 1, 2.

Let us take a point p ∈ S

2

¸ S. Then S

1

∩ p

⊥

is a hyperplane of S

1

or p ∈ S

1

∩ S

2

and S

1

∩ p

⊥

coincides with S

1

. In the ﬁrst case, we set

H := S

1

∩ p

⊥

; in the second case, we take any hyperplane H ⊂ S

1

. Let

u, v be distinct points on a line L ⊂ H. The lines up and vp intersect

S in points u

and v

, respectively. Since p ,∈ S, we have p ,= u

, v

and

the points u

, v

are distinct. The lines L and u

v

**both are contained in
**

the plane ¸L, p); thus they have a non-empty intersection. The inclusion

u

v

**⊂ S guarantees that L intersects S. So, every line of H has a non-
**

empty intersection with S. The subspace H is (n −2)-dimensional and the

dimension of S ∩ H is not less than n − 3. Since S ∩ H ⊂ S ∩ S

1

, the

subspace S ∩ S

1

is (n − 3)-dimensional. Similarly, we show that S ∩ S

2

is

(n −3)-dimensional.

Now we establish the equality

dim(S

1

∩ S

2

) = n −3

which completes our proof. Deﬁne

U := (S ∩ S

1

) ∩ (S ∩ S

2

).

Since S ∩ S

1

and S ∩ S

2

are (n −3)-dimensional subspaces of S, there are

the following three possibilities:

• S ∩ S

1

= S ∩ S

2

and U is (n −3)-dimensional,

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

178 Grassmannians of Classical Buildings

• dimU = n −4,

• dimU = n −5 and S is spanned by S ∩ S

1

and S ∩ S

2

.

By U ⊂ S

1

∩ S

2

, the dimension of S

1

∩ S

2

is equal to n −3 in the ﬁrst and

second cases. Moreover, if U is an (n − 5)-dimensional subspace distinct

from S

1

∩ S

2

then S

1

∩ S

2

is (n −3)-dimensional.

Suppose that U = S

1

∩ S

2

is (n − 5)-dimensional. We take any line

L ⊂ S

1

¸ S and consider the subspace L

⊥

∩ S

2

.

If every point of this subspace belongs to S then L

⊥

∩ S

2

is contained

in S ∩ S

2

. The subspace S ∩ S

2

is (n − 3)-dimensional and the dimension

of L

⊥

∩ S

2

is not less than n −3; thus

S ∩ S

2

= L

⊥

∩ S

2

.

Since S is spanned by S ∩ S

1

and S ∩ S

2

, the later equality implies that

L ⊥ S which is impossible (S is a maximal singular subspace and L ,⊂ S).

Therefore, L

⊥

∩ S

2

contains a point p ,∈ S. Using arguments of the

second step, we show that the intersection of S with the plane ¸L, p) is

a line. This line intersects L which contradicts the inclusion L ⊂ S

1

¸ S.

Hence this case is not realized.

4.8 Apartments

4.8.1 Apartments in polar Grassmannians

Let Π = (P, L) be a polar space of rank n and B = ¦p

1

, . . . , p

2n

¦ be a frame

of Π. Denote by /

k

the associated apartment of (

k

(Π), k ∈ ¦0, 1, . . . , n−1¦.

It consists of all k-dimensional singular subspaces spanned by subsets of the

frame B; in particular, /

0

coincides with B.

Proposition 4.26. [/

k

[ = 2

k+1

n

k+1

.

Proof. The apartment /

k

consists of all k-dimensional singular sub-

spaces ¸p

i

1

, . . . , p

i

k+1

) such that

¦i

1

, . . . , i

k+1

¦ ∩ ¦σ(i

1

), . . . , σ(i

k+1

)¦ = ∅.

There are 2n possibilities to choose p

i

1

, then p

i

2

can be chosen in 2n − 2

ways and so on. Since the order of the points is not taken into account, /

k

consists of

2n (2n −2) (2n −2k)

(k + 1)!

= 2

k+1

n

k + 1

elements.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 179

Exercise 4.13. Show that for any S, U ∈ (

k

(Π) the intersection of all

apartments of (

k

(Π) containing S and U coincides with ¦S, U¦.

Consider the set

J = ¦1, . . . , n, −1, . . . , −n¦.

Recall that a subset X ⊂ J is called singular if

j ∈ X =⇒ −j ,∈ X.

We deﬁne the polar Johnson graph PJ(n, k), k ∈ ¦0, 1, . . . , n − 1¦, as the

graph whose vertex set is formed by all singular subsets consisting of k +1

elements. Two such subsets are connected by an edge if their intersection

consists of k elements, and in the case when k < n−1, we also require that

their union is singular. Then PJ(n, n−1) coincides with the n-dimensional

hypercube graph H

n

. By the deﬁnition, a subset of P = (

0

(Π) is a frame

of Π if the restriction of Γ

0

(Π) to this subset is isomorphic to PJ(n, 0).

For every subset A ⊂ (

k

(Π) we denote by Γ(A) the restriction of the

Grassmann graph Γ

k

(Π) to A. Then Γ(/

k

) is isomorphic to the polar

Johnson graph PJ(n, k). In the case when 0 < k < n − 1, the maximal

cliques of Γ(/

k

) are the intersections of /

k

with the stars

[M, N]

k

, M ∈ /

k−1

, N ∈ /

n−1

,

and the tops

¸T]

k

, T ∈ /

k+1

(if k = n−2 then the ﬁrst possibility is not realized); every maximal clique

of Γ(/

k

) is an independent subset spanning a maximal singular space of

G

k

(Π). The maximal cliques of Γ(/

n−1

) are pairs of adjacent elements; for

every S ∈ /

n−1

there are precisely n elements S

1

, . . . , S

n

∈ /

n−1

adjacent

with S and

S ∩ S

1

, . . . , S ∩ S

n

form a base of the top ¸S]

n−2

.

The following example shows that the Grassmann space G

k

(Π) does not

need to be spanned by apartments of (

k

(Π).

Example 4.12. Suppose that Π = Π

Ω

, where Ω is a non-degenerate trace-

valued reﬂexive form deﬁned on an n-dimensional vector space V . The

Witt index of Ω is assumed to be less than

n

2

. Every frame of Π spans a

proper subspace U in Π

V

. The associated apartment of (

k

(Π) is contained

in the subspace

¦ S ∈ (

k

(Π) : S ⊂ U ¦

By Lemma 4.7, this is a proper subspace of G

k

(Π).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

180 Grassmannians of Classical Buildings

Let M be an m-dimensional singular subspace of Π. Consider the

parabolic subspace [M)

k

, m < k. We take any frame of Π which con-

tains a subset spanning M; the intersection of [M)

k

with the associated

apartment of (

k

(Π) is called an apartment of our parabolic subspace. By

Lemma 4.4, the natural collineation of [M)

k

to the Grassmann space of

index k − m − 1 associated with the polar space [M)

m+1

(Example 4.6)

establishes a one-to-one correspondence between apartments. The restric-

tions of the Grassmann graph Γ

k

(Π) to apartments of [M)

k

are isomorphic

to the polar Johnson graph PJ(n − m− 1, k − m− 1); in particular, the

restriction of Γ

n−1

(Π) to an apartment of [M)

n−1

is isomorphic to the

(n −m−1)-dimensional hypercube graph.

Consider a subset A ⊂ (

n−1

(Π) and for every S ∈ A denote by A

S

the set of all elements of A adjacent with S. We say that A is locally

independent if for every S ∈ A the set of all S∩U, U ∈ A

S

, is an independent

subset of ¸S]

n−2

. Apartments of (

n−1

(Π) and apartments in parabolic

subspaces of G

n−1

(Π) are locally independent.

Theorem 4.14 ([Cooperstein, Kasikova and Shult (2005)]). Let A

be a locally independent subset of (

n−1

(Π) such that Γ(A) is isomorphic

to the m-dimensional hypercube graph, 2 ≤ m ≤ n. If m = n then A is an

apartment of (

n−1

(Π); in the case when m < n, this is an apartment in a

parabolic subspace of G

n−1

(Π).

The proof is similar to the proof of Theorem 3.8.

Proof. Let S ∈ A. There are precisely m elements of A adjacent with

S; we denote them by S

1

, . . . , S

m

. Then

M(S) := S ∩ S

1

∩ ∩ S

m

is an (n−m−1)-dimensional singular subspace and for every i ∈ ¦1, . . . , m¦

the singular subspace

X

i

:=

j=i

(S ∩ S

j

)

is (n−m)-dimensional. The set formed by all X

i

will be denoted by B

0

(S);

this is a base of [M(S), S]

n−m

if m < n and a base of the singular subspace

S if m = n.

Let U be an element of A adjacent with S. For every S

∈ A adjacent

with S and distinct from U there exists unique U

**∈ A adjacent with both
**

S

**, U and non-adjacent with S (this follows directly from the fact that Γ(A)
**

is isomorphic to H

m

). It is not diﬃcult to see that

S

∩ S ∩ U = U

∩ S ∩ U.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 181

This implies the following:

(1) M(S) = M(U),

(2) every X ∈ B

0

(S) satisfying X ⊂ S ∩ U belongs to B

0

(U); therefore,

B

0

(S) ∩B

0

(U) consists of m−1 elements and there is a unique element

of B

0

(S) which does not belong to B

0

(U).

By connectedness of polar Grassmann spaces, we have

M(S) = M(U) ∀ S, U ∈ A;

in what follows this (n − m − 1)-dimensional singular subspace will be

denoted by M.

For every S ∈ A we denote by B(S) the union of all B

0

(U) such that

U ∈ A is adjacent with S. Then B(S) consists of 2m elements; moreover,

this is a frame of the polar space [M)

n−m

if m < n and a frame of Π if

m = n. Now, we show that B(S) coincides with B(U) if S, U ∈ A are

adjacent.

If X ∈ B

0

(S) then X ∈ B(U) (since B

0

(S) is contained in B(U) by

the deﬁnition). Consider the case when X ∈ B

0

(S

) ¸ B

0

(S) for a certain

S

∈ A ¸ ¦U¦ adjacent with S. We take U

∈ A adjacent with both S

and

U. Then S ∩ S

and U

∩ S

are distinct hyperplanes of S

; since X is not

contained in S ∩ S

(X ,∈ B

0

(S)), we have X ⊂ U

∩ S

**which means that
**

X ∈ B

0

(U

) ⊂ B(U).

By connectedness,

B(S) = B(U) ∀ S, U ∈ A;

denote this set by B. In the case when m = n, this is a frame of Π and A

is the associated apartment of (

n−1

(Π). If m < n then B is a frame of the

polar space [M)

n−m

and A is the associated apartment of the parabolic

subspace [M)

n−1

.

Remark 4.15. In the general case, the Grassmann space G

n−1

(Π) and its

parabolic subspaces are not spanned by apartments. By this reason, we

cannot use Theorem 4.14 to prove an analogue of Theorem 3.1. However,

in some special cases this is possible, see [Cooperstein, Kasikova and Shult

(2005)].

4.8.2 Apartments in half-spin Grassmannians

Now suppose that Π is a polar space of type D

n

and n ≥ 4. Let B be a

frame of Π and /

n−1

be the associated apartment of (

n−1

(Π). Then

/

δ

:= /

n−1

∩ (

δ

(Π), δ ∈ ¦+, −¦,

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

182 Grassmannians of Classical Buildings

is the apartment of the half-spin Grassmannian (

δ

(Π) associated with the

frame B.

Proposition 4.27. [/

δ

[ = 2

n−1

.

Proof. This follows from Proposition 4.26, since [/

+

[ = [/

−

[ and /

n−1

is the disjoint union of /

+

and /

−

.

Exercise 4.14. Show that for any S, U ∈ (

δ

(Π) the intersection of all

apartments of (

δ

(Π) containing S and U coincides with ¦S, U¦.

The vertex set of the hypercube graph H

n

(the set of all singular subsets

consisting of n elements) can be decomposed in two disjoint subsets ¸

+

and

¸

−

(Example 2.2). The distance between two vertices is odd if and only

if one of them belongs to ¸

+

and the other belongs to ¸

−

. The distance

between any two vertices from ¸

δ

, δ ∈ ¦+, −¦, is even. The half-cube graph

1

2

H

n

is the graph whose vertex set is ¸

δ

and whose edges are pairs of vertices

at the distance 2 (this construction does not depend on δ ∈ ¦+, −¦). Note

that

1

2

H

4

is isomorphic to PJ(4, 0).

If A ⊂ (

δ

(Π) then we write Γ(A) for the restriction of the Grassmann

graph Γ

δ

(Π) to the set A. It is clear that Γ(/

δ

) is isomorphic to the half-

cube graph

1

2

H

n

. The maximal cliques of Γ(/

δ

) are the intersections of /

δ

with the stars

[S)

δ

, S ∈ /

n−4

,

and the special subspaces

[U]

δ

, U ∈ (

−δ

(Π);

they will be called stars and special subsets of /

δ

(respectively). Every

maximal clique of Γ(/

δ

) is an independent subset spanning a maximal

singular space of G

δ

(Π).

Let M be an m-dimensional singular subspace of Π and m < n − 4.

Consider the parabolic subspace [M)

δ

of the half-spin Grassmann space

G

δ

(Π) (Example 4.8). The deﬁnition of apartments in [M)

δ

is standard.

We identify every U ∈ [M)

δ

with the subspace [M, U]

m+1

in the polar space

[M)

m+1

. In the present case, [M)

m+1

is a polar space of type D

n−m−1

and this correspondence is a collineation of [M)

δ

to one of the half-spin

Grassmann spaces of [M)

m+1

. By Lemma 4.4, this collineation establishes

a one-to-one correspondence between apartments. The restrictions of the

Grassmann graph Γ

δ

(Π) to apartments of [M)

δ

are isomorphic to the half-

cube graph

1

2

H

n−m−1

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 183

Theorem 4.15 ([Cooperstein, Kasikova and Shult (2005)]). Let Π

be a polar space of type D

n

, n ≥ 4, and A be a subset of (

δ

(Π), δ ∈ ¦+, −¦,

such that Γ(A) is isomorphic to the half-cube graph

1

2

H

m

, 4 ≤ m ≤ n,

and every maximal clique of Γ(A) is an independent subset of the half-spin

Grassmann space G

δ

(Π). If m = n then A is an apartment of (

δ

(Π); in the

case when m < n, this is an apartment in a parabolic subspace of G

δ

(Π).

If Π is of type D

4

then the half-spin Grassmann spaces of Π are polar

spaces of type D

4

and every apartment of (

δ

(Π), δ ∈ ¦+, −¦, is a frame

of G

δ

(Π). Since frames are independent subsets of polar spaces, Theorem

4.15 gives the following.

Corollary 4.4. If Π is a polar space of type D

4

then the family of all

frames of the polar space G

δ

(Π), δ ∈ ¦+, −¦, coincides with the family of

all apartments of (

δ

(Π).

Proposition 4.28. If Π is a polar space of type D

n

, n ≥ 4, then half-spin

Grassmann space G

δ

(Π), δ ∈ ¦+, −¦, is spanned by every apartment of

(

δ

(Π).

Proof. In the case when n = 4, this follows immediately from Proposi-

tion 4.9 and Corollary 4.4. Suppose that n ≥ 5 and prove the statement

induction by n. Let B = ¦p

1

, . . . , p

2n

¦ be a frame of Π and / be the asso-

ciated apartment of (

δ

(Π). Denote by A the subspace of G

δ

(Π) spanned

by /.

Every [p

i

)

1

is a polar space of type D

n−1

and [p

i

)

δ

can be identiﬁed with

one of the half-spin Grassmann spaces of this polar space. Then /∩ [p

i

)

δ

is the apartment of [p

i

)

δ

associated with the frame of [p

i

)

1

formed by all

lines p

i

p

j

with j ,= i, σ(i). By the inductive hypothesis, [p

i

)

δ

is spanned by

/∩ [p

i

)

δ

and we have

[p

i

)

δ

⊂ A

for every i. Now consider S ∈ (

δ

(Π) such that p

i

,∈ S for all i. We take any

U ∈ (

−δ

(Π) intersecting S in an (n −2)-dimensional subspace and deﬁne

S

i

:= ¸p

i

, U ∩ p

⊥

i

)

for every i. Then S

i

belongs to [p

i

)

δ

or coincides with U; the second

possibility is realized if p

i

∈ U. We denote by I the set of all i such that

p

i

,∈ U; it is clear that I ,= ∅. The intersection of all U ∩p

⊥

i

, i ∈ I, is empty

(otherwise, there is a point collinear with all points of the frame B which

contradicts Corollary 4.1). This means that the projective space ¸U]

n−2

is

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

184 Grassmannians of Classical Buildings

spanned by the set of all U ∩ p

⊥

i

, i ∈ I. Then the special subspace [U]

δ

is

spanned by the set of all S

i

, i ∈ I. Since S

i

∈ A for every i ∈ I, we get

S ∈ [U]

δ

⊂ A which completes our proof.

Since every parabolic subspace [M)

δ

, M ∈ (

m

(Π), can be identiﬁed with

one of the half-spin Grassmann spaces of the polar space [M)

m+1

such that

apartments correspond to apartments, Proposition 4.28 guarantees that

parabolic subspaces are spanned by apartments. As in Section 3.3, we get

the following.

Corollary 4.5. Every subspace of G

δ

(Π), δ ∈ ¦+, −¦, isomorphic to the

half-spin Grassmann space associated with a polar space of type D

m

, m < n,

is parabolic.

4.8.3 Proof of Theorem 4.15

Let Π

**be a polar space of type D
**

m

. We take any frame B of Π

and consider

the associated apartments

/ ⊂ (

m−1

(Π

), /

δ

⊂ (

δ

(Π

), δ ∈ ¦+, −¦.

Let f : /

+

→ A be an isomorphism of Γ(/

+

) to Γ(A).

Every maximal clique Z of Γ(A) is contained in precisely one maximal

clique of Γ

δ

(Π) (since Z is an independent subset of G

δ

(Π) containing

more than 3 elements and the intersection of two distinct maximal cliques

of Γ

δ

(Π) is not greater than a plane); we say that Z is a star of A or a

special subset of A if the maximal clique of Γ

δ

(Π) containing Z is a star or

a special subspace, respectively. Every special subset o ⊂ /

+

consists of

m elements. If f(o) is a star then m = 4. Therefore, f transfers special

subsets of /

+

to special subsets of A if m ≥ 5.

Consider the case when m = 4. The graphs Γ(/

+

) and Γ(A) are iso-

morphic to PJ(4, 0) and every maximal clique of these graphs consists of 4

elements. For any maximal cliques o and o

of Γ(/

+

) there is a sequence

of maximal cliques

o = o

0

, o

1

, . . . , o

i

= o

such that [o

j−1

∩ o

j

[ = 3 and o

j−1

, o

j

are of diﬀerent types (one of them

is a star and the other is a special subset) for every j ∈ ¦1, . . . , i¦. If the

intersection of two distinct maximal cliques of Γ

δ

(Π) contains a triangle

then these cliques are of diﬀerent types, and it is not diﬃcult to prove that

one of the following possibilities is realized:

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 185

(1) special subsets go to special subsets and stars go to stars,

(2) special subsets go to stars and stars go to special subsets.

In the second case, we take any automorphism g of Γ(/

+

) satisfying (2)

(we leave its construction for the reader); then fg is an isomorphism of

Γ(/

+

) to Γ(A) satisfying (1).

Therefore, we can assume that f transfers special subsets to special

subsets in all cases. Then for every U ∈ /

−

there exists U

∈ (

−δ

(Π) such

that

f(/

+

∩ [U]

+

) = A ∩ [U

]

δ

.

We deﬁne f(U) := U

**and denote by } the subset of (
**

−δ

(Π) consisting of

all f(U), U ∈ /

−

. So, f is extended to a bijection of / to A ∪ }. This

is an isomorphism of Γ(/) to Γ(A ∪ }); indeed, S ∈ A and U ∈ } are

adjacent (as elements of (

n−1

(Π)) if and only if S ∈ [U]

δ

, and the latter is

possible only in the case when f

−1

(S) ∈ [f

−1

(U)]

+

.

So, Γ(A ∪ }) is isomorphic to the m-dimensional hypercube graph and

we want to show that A ∪ } is a locally independent subset of (

n−1

(Π).

Let S ∈ A ∪ }. As in the proof of Theorem 4.14, we consider

S

1

, . . . , S

m

∈ A ∪ }

adjacent with S (all S

i

belong to } if S ∈ A, and they are elements of A

if S ∈ }). The dimension of the subspace

M(S) := S ∩ S

1

∩ ∩ S

m

is not less than n −m−1. In the case when S ∈ },

A ∩ [S]

δ

= ¦S

1

, . . . , S

m

¦

is an independent subset of [S]

δ

. This implies that all S ∩S

i

form an inde-

pendent subset of ¸S]

n−2

and the subspace M(S) is (n−m−1)-dimensional.

Now, suppose that S ∈ A. We take any U ∈ } adjacent with S. For

every S

∈ } ¸ ¦U¦ adjacent with S there exists unique U

∈ A adjacent

with both S

**and U. As in the proof of Theorem 4.14, we have
**

S

∩ S ∩ U = U

∩ S ∩ U.

This implies that

M(S) = M(U).

Hence M(S) is (n − m− 1)-dimensional which guarantees that all S ∩ S

i

form an independent subset of ¸S]

n−2

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

186 Grassmannians of Classical Buildings

Therefore, A ∪ } is locally independent. By Theorem 4.14, this is an

apartment in (

n−1

(Π) (if m = n) or an apartment in a parabolic subspace

of G

n−1

(Π) (if m < n). This gives the claim.

Remark 4.16. Our proof of Theorem 4.15 is diﬀerent from the original

proof given in [Cooperstein, Kasikova and Shult (2005)].

4.9 Apartments preserving mappings

In this section, we suppose that Π = (P, L) and Π

= (P

, L

) are polar

spaces of same type X

n

, X ∈ ¦C, D¦ and n ≥ 3; if X = D then we require

that n ≥ 4.

4.9.1 Apartments preserving bijections

The collineations of G

k

(Π) to G

k

(Π

) induced by collineations of Π to Π

**are apartments preserving. If our polar spaces are of type D
**

n

then the same

holds for the collineations of G

δ

(Π) to G

γ

(Π

), δ, γ ∈ ¦+, −¦, induced by

collineations of Π to Π

.

In the case when Π and Π

**are polar spaces of type D
**

4

, we have also

the following two types of additional collineations:

• the collineations of G

1

(Π) to G

1

(Π

) induced by collineations of Π to

the half-spin Grassmann spaces of Π

,

• the collineations of G

δ

(Π

) to G

γ

(Π

), δ, γ ∈ ¦+, −¦, induced by

collineations of Π to G

−γ

(Π

).

By Corollary 4.4, these collineations are apartments preserving.

Theorem 4.16 (M. Pankov). Every apartments preserving bijection of

(

k

(Π) to (

k

(Π

) is a collineation of G

k

(Π) to G

k

(Π

) (it is a collinearion

of Π to Π

**if k = 0). If our polar spaces are of type D
**

n

, n ≥ 4, then

every apartments preserving bijection of (

δ

(Π) to (

γ

(Π

), δ, γ ∈ ¦+, −¦, is

a collineation of G

δ

(Π

) to G

γ

(Π

).

Remark 4.17. This result was proved in [Pankov 3 (2007)]; for some par-

tial cases it was established in [Pankov 4 (2004); Pankov 2 (2007)].

The proof is based on elementary properties of maximal inexact and

complement subsets; the adjacency relation will be characterized in terms

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 187

of complement subsets. However, in opposite to Grassmannians of ﬁnite-

dimensional vector spaces, certain polar Grassmanians contain two diﬀerent

types of complement subsets; this makes our proof more complicated.

All apartments preserving mappings will be described in Subsection

4.9.7.

4.9.2 Inexact subsets of polar Grassmannians

Let B = ¦p

1

, . . . , p

2n

¦ be a frame of Π and / be the associated apartment

of (

k

(Π). Recall that / consists of all k-dimensional singular subspaces

¸p

i

1

, . . . , p

i

k+1

),

where

¦i

1

, . . . , i

k+1

¦ ∩ ¦σ(i

1

), . . . , σ(i

k+1

)¦ = ∅.

If k = n−1 then every element of / contains precisely one of the points p

i

or p

σ(i)

for each i.

We write /(+i) and /(−i) for the sets consisting of all elements of /

which contain p

i

and do not contain p

i

, respectively. Note that

/(+i) = /(−σ(i))

if k = n − 1. For any i

1

, . . . , i

s

and j

1

, . . . , j

u

belonging to ¦1, . . . , 2n¦ we

deﬁne

/(+i

1

, . . . , +i

s

, −j

1

, . . . , −j

u

)

as the intersection

/(+i

1

) ∩ ∩ /(+i

s

) ∩ /(−j

1

) ∩ ∩ /(−j

u

).

Let ¹ ⊂ /. We say that ¹ is exact if there is only one apartment of

(

k

(Π) containing ¹; otherwise, ¹ is said to be inexact. If ¹ ∩ /(+i) is

not empty then we deﬁne S

i

(¹) as the intersection of all elements of ¹

containing p

i

; we will write S

i

(¹) = ∅ in the case when the intersection of

¹ and /(+i) is empty. If

S

i

(¹) = p

i

for all i then the subset ¹ is exact. However, in contrast to Grassmannians

of ﬁnite-dimensional vector space, the converse fails.

Lemma 4.17. Let ¹ ⊂ /. If there exist distinct i, j such that

p

j

∈ S

i

(¹) and p

σ(i)

∈ S

σ(j)

(¹).

Then ¹ is inexact.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

188 Grassmannians of Classical Buildings

Proof. It is clear that j ,= σ(i). On the line p

i

p

j

we choose a point p

i

distinct from p

i

and p

j

. The point p

i

is collinear with a unique point of the

line p

σ(i)

p

σ(j)

; denote this point by p

σ(j)

. Then

(B ¸ ¦p

i

, p

σ(j)

¦) ∪ ¦p

i

, p

σ(j)

¦

is a frame of Π. This frame deﬁnes a new apartment of (

k

(Π) containing

the subset ¹.

Remark 4.18. In particular, we established that

B ¸ ¦p

i

, p

j

¦, j ,= i, σ(i)

is an inexact subset of B (frames are apartments in the Grassmannian

(

0

(Π) = P).

Lemma 4.18. If Π is a polar space of type C

n

then for each i ∈ ¦1, . . . , 2n¦

there exists a point p

i

,= p

i

, p

σ(i)

such that

(B ¸ ¦p

i

¦) ∪ ¦p

i

¦ and (B ¸ ¦p

σ(i)

¦) ∪ ¦p

i

¦

are frames of Π.

Proof. Let S and U be disjoint (n − 2)-dimensional singular subspaces

spanned by subsets of B ¸ ¦p

i

, p

σ(i)

¦. Since Π is a polar space of type C

n

,

there exists a maximal singular subspace

S

,= ¸S, p

i

), ¸S, p

σ(i)

)

containing S. Lemma 4.2 implies the existence of a point p ∈ S

satisfying

p ⊥ U. This point does not belong to S (for every point q ∈ S we have

q ,⊥ U). Hence p is non-collinear with p

i

and p

σ(i)

. Since p ⊥ (S ∪ U) and

B ¸ ¦p

i

, p

σ(i)

¦ ⊂ S ∪ U,

p is collinear with all points of B ¸ ¦p

i

, p

σ(i)

¦. The point p

i

:= p is as

required.

Lemma 4.19. If S and U are singular subspaces spanned by subsets of B

then S

⊥

∩ U is spanned by a subset of B.

Proof. The case when S ⊥ U is trivial. Suppose that S ,⊥ U and denote

by X the set of all points from U ∩ B which are not contained in S

⊥

.

The singular subspace spanned by (U ∩ B) ¸ X is contained in S

⊥

∩ U.

If this subspace does not coincide with S

⊥

∩ U then S

⊥

has a non-empty

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 189

intersection with ¸X). Let p be a point belonging to this intersection. There

exists p

i

∈ X such that

(X ¸ ¦p

i

¦) ∪ ¦p¦ (4.6)

is a base of ¸X). It is clear that p

σ(i)

belongs to S and it is collinear with

all points of the base (4.6). Then p

σ(i)

⊥ ¸X) which contradicts p

i

∈ ¸X).

Thus S

⊥

does not intersect ¸X) and S

⊥

∩ U is spanned by (U ∩ B) ¸ X.

Proposition 4.29. If k = n − 1 then /(−i) is inexact, but this inexact

subset is not maximal. For every k ∈ ¦0, . . . , n−2¦ the following assertions

are fulﬁlled:

(1) /(−i) is a maximal inexact subset if Π is of type C

n

,

(2) /(−i) is exact if Π is of type D

n

.

Proof. Suppose that k = n −1. In this case, the subset /(−i) coincides

with /(+σ(i)). Let us take any U belonging to /¸ /(−i) = /(+i). Then

S

i

(/(−i) ∪ ¦U¦) = U,

S

σ(i)

(/(−i) ∪ ¦U¦) = S

σ(i)

(/(−i)) = p

σ(i)

.

Since for every j ,= i, σ(i) the intersection of all elements of /(−i) contain-

ing p

j

is the line p

j

p

σ(i)

, we get

S

j

(/(−i) ∪ ¦U¦) = p

j

, p

j

∈ U,

S

j

(/(−i) ∪ ¦U¦) = p

j

p

σ(i)

, p

j

,∈ U.

In the second case, p

σ(j)

belongs to U; then

p

σ(j)

∈ S

i

(/(−i) ∪ ¦U¦) and p

σ(i)

∈ S

j

(/(−i) ∪ ¦U¦).

By Lemma 4.17, the subset

/(−i) ∪ ¦U¦

is inexact. This means that /(−i) is inexact; moreover, this inexact subset

is not maximal.

Consider the case when k ≤ n −2. We have

S

j

(/(−i)) = p

j

for j ,= i (4.7)

(it is not diﬃcult to choose two elements of /(−i) whose intersection is

precisely the point p

j

). Therefore, if /(−i) is contained in the apartment

of (

k

(Π) associated with a frame B

,= B then

B

= (B ¸ ¦p

i

¦) ∪ ¦p¦

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

190 Grassmannians of Classical Buildings

and p ,= p

i

. The point p is collinear with all points of B ¸ ¦p

i

, p

σ(i)

¦ and

non-collinear with p

σ(i)

. Also we have p ,⊥ p

i

(otherwise the point p is

contained in every maximal singular subspace spanned by p

i

and a subset

of B ¸ ¦p

i

, p

σ(i)

¦ which implies p

i

= p, see Proposition 4.6). If S is an

(n−2)-dimensional singular subspace spanned by a subset of B¸¦p

i

, p

σ(i)

¦

then the points p, p

i

, p

σ(i)

deﬁne three distinct maximal singular subspaces

containing S which is impossible if Π is of type D

n

. Thus /(−i) is exact if

Π is of type D

n

.

Suppose that Π is a polar space of type C

n

. By Lemma 4.18, there is a

point p

i

,= p

i

, p

σ(i)

such that

(B ¸ ¦p

i

¦) ∪ ¦p

i

¦

is a frame of Π. The associated apartment of (

k

(Π) contains /(−i). The

subset /(−i) is inexact.

Let U be an element of / ¸ /(−i) = /(+i). Then U is spanned by

p

i

and some points p

i

1

, . . . , p

i

k

. By Lemma 4.19, p

i

is the unique point of

U collinear with all points p

σ(i

1

)

, . . . , p

σ(i

k

)

and (4.7) guarantees that the

subset

/(−i) ∪ ¦U¦

is exact. Since this is true for every U ∈ / ¸ /(−i), the inexact subset

/(−i) is maximal.

Consider the subset

¹

ij

:= /(+i, +j) ∪ /(+σ(i), +σ(j)) ∪ /(−i, −σ(j)), j ,= i, σ(i).

It is clear that

¹

ij

= /(+i, +j) ∪ /(−i) if k = n −1

and

¹

ij

= B ¸ ¦p

i

, p

σ(j)

¦ if k = 0.

Now we prove the following equality

/¸ ¹

ij

= /(+i, −j) ∪ /(+σ(j), −σ(i)). (4.8)

Proof. Let U be an element of / ¸ ¹

ij

. Then U does not belong to

/(−i, −σ(j)) and we have

U ∈ /(+i) or U ∈ /(+σ(j)).

In the ﬁrst case,

U ,∈ /(+i, +j) =⇒ U ∈ /(+i, −j).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 191

In the second case,

U ,∈ /(+σ(i), +σ(j)) =⇒ U ∈ /(+σ(j), −σ(i)).

Therefore,

/¸ ¹

ij

⊂ /(+i, −j) ∪ /(+σ(j), −σ(i)).

Since /(+i, −j) and /(+σ(j), −σ(i)) do not intersect ¹

ij

, the inverse in-

clusion holds.

Proposition 4.30. The subset ¹

ij

is inexact; moreover, it is a maximal

inexact subset, except the case when k = 0 and Π is of type C

n

.

Proof. Let k = 0. Then ¹

ij

= B ¸ ¦p

i

, p

σ(j)

¦ is an inexact subset

(Remark 4.18). If Π is of type C

n

then /(−i) = B ¸ ¦p

i

¦ is an inexact

subset containing ¹

ij

(Proposition 4.29) and the inexact subset ¹

ij

is not

maximal. In the case when Π is of type D

n

, the subsets

/(−i) = B ¸ ¦p

i

¦ and /(−σ(j)) = B ¸ ¦p

σ(j)

¦

are exact (Proposition 4.29) which means that ¹

ij

is a maximal inexact

subset.

Suppose that k ≥ 1. Since

S

i

(¹

ij

) = p

i

p

j

and S

σ(j)

(¹

ij

) = p

σ(j)

p

σ(i)

,

Lemma 4.17 implies that the subset ¹

ij

is inexact.

For every l ,= i, σ(j) we can choose two elements of ¹

ij

whose intersec-

tion is p

l

(we leave the details for the reader). This means that

S

l

(¹

ij

) = p

l

if l ,= i, σ(j).

By (4.8), for every U belonging to /¸ ¹

ij

one of the following possibilities

is realized:

• U ∈ /(+i, −j) intersects S

i

(¹

ij

) = p

i

p

j

precisely in the point p

i

; since

p

σ(j)

is the unique point on the line S

σ(j)

(¹

ij

) = p

σ(j)

p

σ(i)

collinear

with p

i

, the set ¹

ij

∪ ¦U¦ is exact.

• U ∈ /(+σ(j), −σ(i)) intersects S

σ(j)

(¹

ij

) = p

σ(j)

p

σ(i)

precisely in the

point p

σ(j)

; since p

i

is the unique point on the line S

i

(¹

ij

) = p

i

p

j

collinear with p

σ(j)

, the set ¹

ij

∪ ¦U¦ is exact.

Therefore, ¹

ij

∪ ¦U¦ is exact for each U ∈ /¸ ¹

ij

and the inexact subset

¹

ij

is maximal.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

192 Grassmannians of Classical Buildings

If the set /(−i) is inexact (see Proposition 4.29) then it will be called

an inexact subset of ﬁrst type. Similarly, if ¹

ij

is inexact (see Proposition

4.30) then it is said to be an inexact subset of second type.

Proposition 4.31. Every maximal inexact subset is of ﬁrst type or of sec-

ond type. In particular, the following assertions are fulﬁlled:

(1) if k = n −1 then each maximal inexact subset is of second type,

(2) if k ∈ ¦1, . . . , n − 2¦ and Π is of type C

n

then each maximal inexact

subset is of ﬁrst type or of second type,

(3) if k ∈ ¦0, . . . , n − 2¦ and Π is of type D

n

then each maximal inexact

subset is of second type,

(4) if k = 0 and Π is of type C

n

then each maximal inexact subset is of

ﬁrst type.

Proof. Let k = 0. The statement is trivial if Π is of type C

n

.

Suppose that Π is of type D

n

. We need to show that B ¸ ¦p

i

, p

σ(i)

¦ is

exact. Assume that there exist points p

i

, p

σ(i)

such that

(B ¸ ¦p

i

, p

σ(i)

¦) ∪ ¦p

i

, p

σ(i)

¦

is a frame. For every (n − 2)-dimensional singular subspace S spanned by

a subset of B ¸ ¦p

i

, p

σ(i)

¦ the pairs of subspaces

¸S, p

i

), ¸S, p

σ(i)

) and ¸S, p

i

), ¸S, p

σ(i)

)

are coincident (since Π is of type D

n

). This means that ¦p

i

, p

σ(i)

¦ coincides

with ¦p

i

, p

σ(i)

¦ (we leave the details for the reader) and B ¸ ¦p

i

, p

σ(i)

¦ is

exact.

Let k ≥ 1 and ¹ be a maximal inexact subset of /. First, we consider

the case when all S

i

(¹) are non-empty. Denote by I the set of all numbers

i such that the dimension of S

i

(¹) is greater than zero. The subset ¹ is

inexact and I is non-empty. Suppose that for certain l ∈ I the subspace

S

l

(¹) is spanned by p

l

, p

j

1

, . . . , p

j

u

and the subspaces

M

1

:= S

σ(j

1

)

(¹), . . . , M

u

:= S

σ(j

u

)

(¹)

do not contain the point p

σ(l)

. Then p

l

is collinear with all points of the

subspaces M

1

, . . . , M

u

. Since

p

j

1

,⊥ M

1

, . . . , p

j

u

,⊥ M

u

,

p

l

is the unique point of S

l

(¹) collinear with all points of M

1

, . . . , M

u

(Lemma 4.19). If the same holds for every l ∈ I then Lemma 4.19 implies

that ¹ is exact. Therefore, there exist i ∈ I and j ,= i, σ(i) such that

p

j

∈ S

i

(¹) and p

σ(i)

∈ S

σ(j)

(¹).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 193

Then ¹ is contained ¹

ij

. Indeed, for every U ∈ ¹ one of the following

possibilities is realized:

• p

i

∈ U then U ∈ /(+i, +j) ⊂ ¹

ij

,

• p

σ(j)

∈ U then U ∈ /(+σ(i), +σ(j)) ⊂ ¹

ij

,

• p

i

, p

σ(j)

,∈ U then U ∈ /(−i, −σ(j)) ⊂ ¹

ij

.

We have ¹ = ¹

ij

, since the inexact subset ¹ is maximal.

Now suppose that S

i

(¹) = ∅ for certain i. Then ¹ is contained in

/(−i) and we get a maximal inexact subset of ﬁrst type if Π is of type C

n

.

If the polar space is of type D

n

then /(−i) is exact and one of the following

possibilities is realized:

(1) S

j

(¹) = ∅ for certain j ,= i, σ(i). Then ¹ ⊂ /(−j, −i) and ¹ is a

proper subset of ¹

j σ(i)

.

(2) All S

l

(¹), l ,= i, σ(i), are non-empty and there exists j ,= i, σ(i) such

that the subspace S

j

(¹) contains p

σ(i)

. Since every U ∈ ¹ ¸ /(+j)

is contained in /(−j, −i) and every U ∈ ¹ ∩ /(+j) belongs to

/(+j, +σ(i)), we obtain the inclusion

¹ ⊂ /(+j, +σ(i)) ∪ /(−j, −i).

As in the previous case, ¹ is a proper subset of ¹

j σ(i)

.

(3) Each S

j

(¹), j ,= i, σ(i) is non-empty and does not contain p

σ(i)

.

In the cases (1) and (2), ¹ is a proper subset of a maximal inexact subset

which contradicts our assumption.

Consider the case (3). Since the subset ¹is inexact, there exists a frame

B

**,= B such that ¹ is contained in the associated apartment of (
**

k

(Π). We

denote by I the set of all numbers j ,= i, σ(i) such that the dimension of

S

j

(¹) is greater than zero. If I is empty then

B

= (B ¸ ¦p

i

, p

σ(i)

¦) ∪ ¦p

i

, p

σ(i)

¦. (4.9)

As in the case when k = 0, we get ¦p

i

, p

σ(i)

¦ = ¦p

i

, p

σ(i)

¦ which implies

that B = B

. Therefore, I ,= ∅.

Let l ∈ I. If S

l

(¹) is spanned by p

l

, p

j

1

, . . . , p

j

u

(by our assumption,

these points are distinct from p

i

and p

σ(i)

) and the subspaces

M

1

:= S

σ(j

1

)

(¹), . . . , M

u

:= S

σ(j

u

)

(¹)

do not contain p

σ(l)

then p

l

is the unique point of S

l

(¹) collinear with

all points of M

1

, . . . , M

u

. If the same holds for all elements of I then, by

Lemma 4.19, we get (4.9) again. As above, we establish that ¹ is contained

in a certain complement subset ¹

jm

, j, m ,= i, σ(i). The inclusion

¹ ⊂ /(−i) ∩ ¹

jm

means that ¹ is a proper subset of ¹

jm

which is impossible.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

194 Grassmannians of Classical Buildings

4.9.3 Complement subsets of polar Grassmannians

Let / be as in the previous subsection. We say that ¹ ⊂ / is a complement

subset if /¸ ¹ is a maximal inexact subset. A complement subset is said

to be of ﬁrst type or of second type if the corresponding maximal inexact

subset is of ﬁrst type or of second type, respectively. If /(−i) is a maximal

inexact subset (this fails in some cases) then the associated complement

subset is /(+i). By (4.8), the maximal inexact subset ¹

ij

(k > 0 or Π is

of type D

n

) gives the complement subset

(

ij

:= /(+i, −j) ∪ /(+σ(j), −σ(i))

which coincides with /(+i, +σ(j)) if k = n −1. Note that

(

ij

= (

σ(j)σ(i)

.

The follows assertions follow immediately from Proposition 4.31:

(1) if k = n −1 then each complement subset is of second type;

(2) if k ∈ ¦1, . . . , n −2¦ and Π is of type C

n

then each complement subset

is of ﬁrst type or of second type;

(3) if k ∈ ¦0, . . . , n −2¦ and Π is of type D

n

then each complement subset

is of second type; in the case when k = 0, this is a pair of collinear

points;

(4) if k = 0 and Π is of type C

n

then each complement subset is of ﬁrst

type (a single point).

Now we characterize the adjacency and weak adjacency relations of

(

k

(Π), k ≥ 1, in terms of complement subsets of second type. In the case

when k = n − 1 or the polar space is of type D

n

, each complement subset

is of such type. If S, U are distinct elements of / then we write c(S, U) for

the number of compliment subsets of second type in / containing both S

and U. Let us deﬁne

m

c

:= max ¦ c(S, U) : S, U ∈ /, S ,= U ¦.

Lemma 4.20. Let k ≥ 1 and S, U be distinct elements of /. The following

assertions are fulﬁlled:

(1) S, U are adjacent if and only if c(S, U) = m

c

,

(2) if k < n − 1 and n ,= 4 or k ,= 1 then S, U are weakly adjacent if and

only if c(S, U) = m

c

−1.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 195

Proof. Let S, U ∈ / and m be the dimension of S ∩U. If k = n−1 then

the complement subset /(+i, +j) contains both S, U if and only if the line

p

i

p

j

is contained S ∩ U; thus

c(S, U) =

m+ 1

2

**and we get the claim.
**

Suppose that 1 ≤ k ≤ n − 2. If the complement subset (

ij

= (

σ(j)σ(i)

contains S and U then one of the following possibilities is realized:

(A) the subsets /(+i, −j) and /(+σ(j), −σ(i)) both contain S and U;

(B) only one of the subsets /(+i, −j) or /(+σ(j), −σ(i)) contains both

S, U;

(C) each of the subsets /(+i, −j) and /(+σ(j), −σ(i)) contains precisely

one of the subspaces.

The case (A). We have

S, U ∈ /(+i, −j) ∩ /(+σ(j), −σ(i))

if and only if

p

i

, p

σ(j)

∈ S ∩ U.

Thus there are precisely

m+ 1

2

distinct (

ij

= (

σ(j)σ(i)

satisfying (A).

The case (B). If S and U belong to /(+i, −j) then /(+σ(j), −σ(i))

does not contain the subset ¦S, U¦. As above, p

i

∈ S ∩ U and there are

m+ 1 possibilities for i. Also we have

p

j

,∈ S ∪ U and p

σ(j)

,∈ S ∩ U.

Hence there are

2n −2(k −m) −2(m+ 1) = 2(n −k −1)

possibilities for j. This means that there exist precisely

2(m+ 1)(n −k −1)

distinct (

ij

= (

σ(j)σ(i)

satisfying (B).

The case (C). Suppose that

S ∈ /(+i, −j) ¸ /(+σ(j), −σ(i)) and U ∈ /(+σ(j), −σ(i)) ¸ /(+i, −j).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

196 Grassmannians of Classical Buildings

Then p

σ(j)

∈ U. Since U does not belong to /(+i, −j) and p

j

,∈ U, we

have p

i

,∈ U. Similarly, p

i

∈ S implies that p

σ(j)

,∈ S. So,

p

i

∈ S ¸ U and p

σ(j)

∈ U ¸ S.

Since p

σ(i)

,∈ U and p

j

,∈ S,

p

i

⊥ U and p

σ(j)

⊥ S.

The set B¸U contains precisely 2(n−k−1) points collinear with all points

of U and B ∩ (S ¸ U) contains not greater than n − k − 1 such points

(if p

l

∈ B ¸ U is collinear with all points of U then the same holds for

p

σ(l)

∈ B ¸ U, but S ¸ U does not contain at least one of these points).

Similarly, U ¸ S contains not greater than n − k − 1 points of B collinear

with all points of S. Therefore, there exist at most

l(m) := min¦(k −m)

2

, (n −k −1)

2

¦

distinct (

ij

= (

σ(j)σ(i)

satisfying (C). In the case when m = k − 1 (S and

U are adjacent or weakly adjacent), we have

B ∩ (S ¸ U) = ¦p

u

¦ and B ∩ (U ¸ S) = ¦p

v

¦.

If S and U are adjacent then v ,= σ(u) and (

uσ(v)

= (

vσ(u)

is the unique

compliment subset satisfying (C). If S and U are weakly adjacent then

there are no compliment subsets of such kind.

So, we established that

c(S, U) ≤

m + 1

2

+ 2(m+ 1)(n −k −1) + l(m);

moreover,

c(S, U) =

k

2

+ 2k(n −k −1) + 1

if S, U are adjacent and

c(S, U) =

k

2

+ 2k(n −k −1)

if they are weakly adjacent. A direct veriﬁcation shows that

k

2

+ 2k(n −k −1) ≥

m + 1

2

+ 2(m+ 1)(n −k −1) + l(m)

for m ≤ k − 2 and we get the equality only in the case when n = 4 and

k = 1.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 197

Remark 4.19. Suppose that Π is of type C

n

and 1 ≤ k ≤ n − 2. In

this case, the adjacency and weak adjacency relations can be characterized

in terms of maximal inexact subsets of ﬁrst type. For the intersection of

2n−k−2 distinct maximal inexact subsets of ﬁrst type one of the following

possibilities is realized:

• the intersection is empty,

• it consists of k + 2 mutually adjacent elements of /,

• it consists of 2 weakly adjacent elements of /.

Lemma 4.21. Suppose that 1 ≤ k ≤ n − 2 and Π is a polar space of type

C

n

. Let ¹ be a complement subset of /. If ¹ is of ﬁrst type then there are

precisely 2n −1 distinct complement subsets of / disjoint from ¹. If ¹ is

of second type then there are precisely 3 distinct complement subsets of /

disjoint from ¹.

Proof. We take any l ∈ ¦1, . . . , 2n¦ and consider the complement subset

/(+l). There is only one complement subset of ﬁrst type which does not

intersect /(+l); this is /(+σ(l)). If a complement subset

(

ij

= (

σ(j)σ(i)

= /(+i, −j) ∪ /(+σ(j), −σ(i))

is disjoint from /(+l) then

i = σ(l) or j = l;

in other words, this is (

σ(l)m

= (

σ(m)l

and we have 2n − 2 possibilities for

m (since m ,= l, σ(l)). Therefore, there are precisely 2n − 2 + 1 distinct

complement subsets of / disjoint from /(+l).

Now let us ﬁx i, j ∈ ¦1, . . . , 2n¦ such that j ,= i, σ(i) and consider the

associated complement subset (

ij

= (

σ(j)σ(i)

. There are only two comple-

ment subsets of ﬁrst type disjoint from it:

/(+σ(i)) and /(+j).

Suppose that (

i

j

does not intersect (

ij

. Then

/(+i, −j) ∩ /(+i

, −j

) = ∅ =⇒ i

= σ(i) or i

= j or j

= i.

Using the equalities

/(+σ(j), −σ(i)) ∩ /(+i

, −j

) = ∅,

/(+i, −j) ∩ /(+σ(j

), −σ(i

)) = ∅,

/(+σ(j), −σ(i)) ∩ /(+σ(j

), −σ(i

)) = ∅,

we establish that

i

= j, j

= i or i

= σ(i), j

= σ(j).

Thus (

ji

= (

σ(i)σ(j)

is the unique complement subset of second type disjoint

from (

ij

= (

σ(j)σ(i)

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

198 Grassmannians of Classical Buildings

Proposition 4.32. Let f : (

k

(Π) → (

k

(Π

) be an apartments preserving

mapping and k ≥ 1. Then f is adjacency preserving (two elements of

(

k

(Π) are adjacent if and only if their images are adjacent). In the case

when k ≤ n − 2, if k ,= 1 or our polar spaces are not of type D

4

then f

is weak adjacency preserving (two elements of (

k

(Π) are weakly adjacent if

and only if their images are weakly adjacent).

Proof. First of all, recall that the mapping f is injective. Let S and U

be distinct elements of (

k

(Π) and / be an apartment of (

k

(Π) containing

them. Then f(S) and f(U) belong to the apartment f(/). As in the

proof of Proposition 3.6, we show that f transfers inexact subsets of / to

inexact subsets of f(/). Since / and f(/) have the same number of inexact

subsets (Π and Π

**are polar spaces of the same type), each inexact subset
**

of f(/) is the image of a certain inexact subset of /. Then an inexact

subset of / is maximal if and only if the same holds for its image. Thus ¹

is a complement subset of / if and only if f(¹) is a complement subset of

f(/).

In the case when k = n − 1 or our polar spaces are of type D

n

, all

complement subsets are of second type; if 1 ≤ k ≤ n−2 and the polar spaces

are of type C

n

then, by Lemma 4.21, the mapping f preserves the types of

all maximal inexact and complement subsets. It follows from Lemma 4.20

that S and U are adjacent if and only if f(S) and f(U) are adjacent.

Suppose that 1 ≤ k ≤ n −2. If n ,= 4 or k ,= 1 then Lemma 4.20 shows

that S and U are weakly adjacent if and only if their images are weakly

adjacent. In the case when k = 1 and the polar spaces are of type C

4

, this

statement follows immediately from Remark 4.19.

Remark 4.20. If the polar spaces are of type D

4

then there exist col-

ineations of G

1

(Π) to G

1

(Π

**) which do not preserve the weak adjacency
**

relation (Remark 4.13) and the second part of Proposition 4.32 does not

hold in this case.

Remark 4.21. If our polar spaces are of type C

n

then (

k

(Π), 1 ≤ k ≤

n −2, contains complement subsets of both types. Denote by n

1

(n, k) and

n

2

(n, k) the numbers of elements in complement subsets of ﬁrst and second

type, respectively (it is clear that complement subsets of the same type

have the same number of elements). Every apartments preserving mapping

does not change the types of complement subsets if n

1

(n, k) ,= n

2

(n, k).

However, n

1

(n, k) = n

2

(n, k) for some pairs n, k and we need Lemma 4.21

to distinguish complement subsets of diﬀerent types.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 199

4.9.4 Inexact subsets of half-spin Grassmannians

Now we suppose that Π is a polar space of type D

n

and n ≥ 4. Let

B = ¦p

1

, . . . , p

2n

¦ be a frame of Π and / be the associated apartment of

(

δ

(Π), δ ∈ ¦+, −¦. Recall that it consists of all maximal singular subspaces

¸p

i

1

, . . . , p

i

n

) ∈ (

δ

(Π),

where

¦i

1

, . . . , i

n

¦ ∩ ¦σ(i

1

), . . . , σ(i

n

)¦ = ∅.

Every element of / contains precisely one of the points p

i

or p

σ(i)

for every

i. As in Subsection 4.9.2, we deﬁne /(+i), /(−i),

/(+i

1

, . . . , +i

s

, −j

1

, . . . , −j

u

)

and introduce the concepts of exact, inexact, and complement subsets. For

every subset ¹ ⊂ / we denote by S

i

(¹) the intersection of all elements of

/(+i) ∩ ¹; if this subset is empty then we write S

i

(¹) = ∅.

Proposition 4.33. If j ,= i, σ(i) then

/(−i) ∪ /(+i, +j) (4.10)

is an inexact subset.

Proof. It is trivial that

p

j

∈ S

i

(/(−i) ∪ /(+i, +j)) (4.11)

Suppose that S is an element of (4.10) containing p

σ(j)

. Then S does not

contain p

j

and, by (4.11), we have p

i

,∈ S. The latter guarantees that

p

σ(i)

∈ S. Therefore, p

σ(i)

belongs to every element of (4.10) containing

p

σ(j)

and

p

σ(i)

∈ S

σ(j)

(/(−i) ∪ /(+i, +j)).

As in the proof of Lemma 4.17, we construct a frame

(B ¸ ¦p

i

, p

σ(j)

¦) ∪ ¦p

i

, p

σ(j)

¦

which deﬁnes a new apartment of (

δ

(Π) containing (4.10).

Proposition 4.34. If ¹ is a maximal inexact subsets of / then

¹ = /(−i) ∪ /(+i, +j)

for certain i, j such that j ,= i, σ(i).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

200 Grassmannians of Classical Buildings

Proof. Since ¹ is inexact, we have S

i

(¹) ,= p

i

for a certain number i.

If S

i

(¹) is empty then ¹ is contained in the non-maximal inexact subset

/(−i) which contradicts the assumption that the inexact subset ¹ is maxi-

mal. Thus there exists p

j

∈ S

i

(¹) such that j ,= i. An easy veriﬁcation

shows that

¹ ⊂ /(−i) ∪ /(+i, +j).

Since the latter subset is inexact and the inexact subset ¹ is maximal, the

inverse inclusion holds.

The complement subset corresponding to /(−i) ∪ /(+i, +j) is

/(+i, −j) = /(+i, +σ(j)).

By Proposition 4.34, every compliment subset of / coincides with certain

/(+l, +m), m ,= l, σ(l).

Lemma 4.22. Subspaces S, U ∈ / are adjacent if and only if there are

precisely

n −2

2

**distinct complement subsets of / containing S and U.
**

Proof. See the proof of Lemma 4.20 in the case when k = n −1.

Remark 4.22. Suppose that n = 4. Then every complement subset of /

consists of two adjacent elements. On the other hand, G

δ

(Π) is a polar

space of type D

4

and / is a frame of this space; by Proposition 4.31, in

a polar space of type D

n

every complement subset of a frame is a pair of

collinear points.

Proposition 4.35. Suppose that Π is a polar space of type D

n

, n ≥ 4, and

f : (

δ

(Π) → (

γ

(Π

**), δ, γ ∈ ¦+, −¦, is an apartments preserving mapping.
**

Then f is adjacency preserving (two elements of (

δ

(Π) are adjacent if and

only if their images are adjacent).

Proof. Let / be an apartment of (

δ

(Π). As in the proof of Proposition

4.32, we establish that ¹ is a complement subset of / if and only if f(¹) is

a complement subset of the apartment f(/). Then Lemma 4.22 gives the

claim.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 201

4.9.5 Proof of Theorem 4.16

By Subsection 4.6.1, we need to establish that every apartments preserving

bijection is an isomorphism between the associated Grassmann graphs. For

polar Grassmannians of index > 0 and half-spin Grassmannians this follows

immediately from Propositions 4.32 and 4.35. For Grassmannians of index

0 we have the following.

Proposition 4.36. For a mapping f : P → P

**the following conditions are
**

equivalent:

(1) f is apartments (frames) preserving,

(2) f is a collinearity preserving injection (two points are collinear if and

only if their images are collinear).

Proof. The implication (2) =⇒(1) easy follows from the frame deﬁnition.

(1) =⇒ (2). Let f : P → P

**be a mapping which sends frames of
**

Π to frames of Π

**. This mapping is injective. Since any two points are
**

contained in a certain frame, it is suﬃcient to show that two points in a

frame B = ¦p

1

, . . . , p

2n

¦ of Π are collinear if and only if their images are

collinear. As in the proof of Propositions 4.32, we establish that a subset

X ⊂ B is complement if and only if f(X) is a complement subset of the

frame f(B). In the case when Π and Π

are of type D

n

, each complement

subset of B is a pair of collinear points and we get the claim.

Suppose that our polar spaces are of type C

n

. For each i ∈ ¦1, . . . , 2n¦

there exists a point p

i

such that

(B ¸ ¦p

i

¦) ∪ ¦p

i

¦ and (B ¸ ¦p

σ(i)

¦) ∪ ¦p

i

¦

are frames of Π (Lemma 4.18). Then

(f(B) ¸ ¦f(p

i

)¦) ∪ ¦f(p

i

)¦ and (f(B) ¸ ¦f(p

σ(i)

)¦) ∪ ¦f(p

i

)¦ (4.12)

are frames of Π

**. Since f(B) is a frame, there are unique u, v ∈ ¦1, . . . , 2n¦
**

such that

f(p

i

) ,⊥ f(p

u

) and f(p

σ(i)

) ,⊥ f(p

v

). (4.13)

It is clear that u ,= v and f(p

i

) is non-collinear with both f(p

u

), f(p

v

).

Thus there is no frame of Π

**containing the subset
**

¦f(p

i

), f(p

u

), f(p

v

)¦.

We have

¦f(p

i

), f(p

σ(i)

)¦ = ¦f(p

u

), f(p

v

)¦

(otherwise, one of the frames (4.12) contains f(p

i

), f(p

u

), f(p

v

)). Then

(4.13) implies that u = σ(i) and v = i. Therefore, f(p

i

) ,⊥ f(p

σ(i)

) which

guarantees that f(p

i

) is collinear with f(p

j

) if j ,= σ(i).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

202 Grassmannians of Classical Buildings

4.9.6 Embeddings

Proposition 4.37. Every apartments preserving mapping f : P → P

is

an embedding of Π in Π

.

Proof. By Proposition 4.36, f is a collinearity preserving injection. Let

p

1

, p

2

∈ P be distinct collinear points and ¦p

1

, p

2

, . . . , p

2n

¦ be a frame of Π

containing them. We suppose that

p

i

,⊥ p

i+n

∀ i ∈ ¦1, . . . , n¦.

Then all p

i

:= f(p

i

) form a frame of Π

and

p

i

,⊥ p

i+n

∀ i ∈ ¦1, . . . , n¦.

The line p

1

p

2

is a maximal singular subspace of the generalized quadrangle

¦p

3

, . . . , p

n

, p

n+3

, . . . , p

2n

¦

⊥

(Lemma 4.3) and we have

p

1

p

2

= ¦p

1

, p

2

, p

3

, . . . , p

n

, p

n+3

, . . . , p

2n

¦

⊥

.

Similarly, we obtain that

p

1

p

2

= ¦p

1

, p

2

, p

3

, . . . , p

n

, p

n+3

, . . . , p

2n

¦

⊥

.

Thus a point p ∈ P is on the line p

1

p

2

if and only if f(p) is on the line

p

1

p

2

= f(p

1

)f(p

2

). This means that f transfers lines to subsets of lines

and distinct lines go to subsets of distinct lines.

Let f : P → P

**be an apartments preserving mapping. By Proposition
**

4.37, it is an embedding of Π in Π

**. For every singular subspace S of Π
**

there is a frame such that S is spanned by a subset of this frame. This

implies that f(S) consists of mutually collinear points and

dim¸f(S)) = dimS.

So, for every k ∈ ¦1, . . . , n −1¦ we have the mapping

(f)

k

: (

k

(Π) → (

k

(Π

)

S → ¸f(S)).

An easy veriﬁcation shows that this is an apartments preserving embedding

of G

k

(Π) in G

k

(Π

).

Similarly, we construct the apartments preserving embedding (f)

δ

of

G

δ

(Π) in G

γ

(Π

**), δ, γ ∈ ¦+, −¦, if our polar spaces are of type D
**

n

.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 203

Theorem 4.17 (M. Pankov). Let 1 ≤ k ≤ n − 1. If k ,= 1 or the po-

lar spaces are not of type D

4

then every apartments preserving mapping

of (

k

(Π) to (

k

(Π

) is the embedding of G

k

(Π) in G

k

(Π

) induced by an

apartments preserving embedding of Π in Π

**. If our polar spaces are of
**

type D

n

and n ≥ 5 then every apartments preserving mapping of (

δ

(Π) to

(

γ

(Π

), δ, γ ∈ ¦+, −¦, is the embedding of G

δ

(Π) in G

γ

(Π

) induced by an

apartments preserving embedding of Π in Π

.

Remark 4.23. In [Pankov 2 (2007)] a such kind result was proved for

symplectic Grassmannians; in [Pankov 3 (2007)] it was announced for all

Grassmannians associated with polar spaces of type C

n

.

Now suppose that our polar spaces are of type D

4

. In this case, the

associated half-spin Grassmann spaces are polar spaces of type D

4

. Let f

be an apartments preserving embedding of Π in G

δ

(Π

), δ ∈ ¦+, −¦. Then

f induces an embedding of G

1

(Π) in G

1

(Π

**); by Corollary 4.4, this embed-
**

ding is apartments preserving. Also f induces an apartments preserving

embedding of one of the half-spin Grassmann spaces of Π in G

−δ

(Π

).

Theorem 4.18 (M. Pankov). Suppose that the polar spaces are of type

D

4

. Then every apartments preserving mapping of (

1

(Π) to (

1

(Π

) is the

embedding of G

1

(Π) in G

1

(Π

**) induced by an apartments preserving embed-
**

ding of Π in Π

or in one of the half-spin Grassmann spaces of Π

. Similarly,

every apartments preserving mapping of (

δ

(Π) to (

γ

(Π

), δ, γ ∈ ¦+, −¦, is

the embedding of G

δ

(Π) in G

γ

(Π

**) induced by an apartments preserving
**

embedding of Π in Π

or in G

−γ

(Π

).

4.9.7 Proof of Theorems 4.17 and 4.18

Let 1 ≤ k ≤ n − 1 and f : (

k

(Π) → (

k

(Π

) be an apartments preserving

mapping. By Proposition 4.32, it is adjacency preserving (two elements of

(

k

(Π) are adjacent if and only if their images are adjacent). In the case

when k ≤ n−2, the mapping f preserves the weak adjacency relation (two

elements of (

k

(Π) are weakly adjacent if and only if their images are weakly

adjacent) if k ,= 1 or the polar spaces are not of type D

4

.

1. If k = n − 1 then f maps lines to subsets of lines and induces a

mapping

f

n−2

: (

n−2

(Π) → (

n−2

(Π

).

This mapping is injective (otherwise there exists a line of G

n−1

(Π

) contain-

ing the images of two distinct lines of G

n−1

(Π), this implies the existence

of non-adjacent elements of (

n−1

(Π) whose images are adjacent).

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

204 Grassmannians of Classical Buildings

If 1 ≤ k ≤ n − 2 and f preserves the weak adjacency relation then it

transfers maximal cliques of Γ

w

k

(Π) to cliques of Γ

w

k

(Π

). Big stars go to

subsets of big stars (since any two elements of a top are adjacent and a

big star contains non-adjacent elements). The image of every big star is

contained in precisely one big star (the intersection of two distinct big stars

is empty or consists of one element). Therefore, f induces a mapping

f

k−1

: (

k−1

(Π) → (

k−1

(Π

).

It is injective (otherwise there exist weakly adjacent elements of (

k

(Π

)

whose pre-images are not weakly adjacent).

It must be pointed out that the mapping f

k−1

is not deﬁned only in the

case when k = 1 and our polar spaces are of type D

4

.

Let B be a frame of Π and / be the associated apartment of (

k

(Π).

Denote by B

the frame of Π

**corresponding to the apartment f(/). As
**

in the proof of Theorem 3.10 (Subsection 3.4.3), we show that f

k−1

maps

the apartment of (

k−1

(Π) associated with B to the apartment of (

k−1

(Π

)

deﬁned by B

. Therefore, f

k−1

is apartments preserving.

In the case when n ,= 4 or the polar spaces are of type C

4

, we construct

a sequence of apartments preserving mappings

f

i

: (

i

(Π) → (

i

(Π

), i = k, . . . , 0,

such that f

k

= f,

f

i

([S)

i

) ⊂ [f

i−1

(S))

i

for all S ∈ (

i−1

(Π) and

f

i−1

(¸U]

i−1

) ⊂ ¸f

i

(U)]

i−1

for all U ∈ (

i

(Π) if i ≥ 1. An easy veriﬁcation shows that each f

i

, i ≥ 1, is

induced by f

0

.

2. Now suppose that our polar spaces are of type D

4

.

Consider the case k = 2. Since f

1

is apartments preserving, it is adja-

cency preserving and maximal singular subspaces of G

1

(Π) go to subsets of

maximal singular subspaces of G

1

(Π

**). On the other hand, f
**

1

is induced

by f; this implies that f

1

maps tops to subsets of tops. Then stars go to

subsets of stars (for every star there is a top intersecting this star in a line

and the intersection of two distinct tops contains at most one element). In

the present case, two stars of G

1

(Π) are adjacent (Subsection 4.6.3) if and

only if their intersection is a line. As in Subsection 4.6.3, we establish that

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Polar and Half-Spin Grassmannians 205

f

1

transfers big stars to subsets of big stars. The image of every big star is

contained in precisely one big star and f

1

induces a mapping f

0

: P → P

.

It is not diﬃcult to prove that f

0

is an apartments preserving embedding

of Π in Π

which induces f

1

and f.

If k = 3 then f

2

is induced by an apartments preserving embedding of

Π to Π

**(it was established above). It is clear that this embedding also
**

induces f.

Let k = 1. Then f is an adjacency preserving injection of (

1

(Π) to

(

1

(Π

**) satisfying the condition (A) from Subsection 4.6.5. It transfers max-
**

imal singular subspaces of G

1

(Π) to subsets of maximal singular subspaces

of G

1

(Π

**) and distinct maximal singular subspaces go to subsets of distinct
**

maximal singular subspaces. The image of every maximal singular sub-

space of G

1

(Π) is contained in precisely one maximal singular subspace of

G

1

(Π

**). If stars go to subsets of stars then, by the arguments given above,
**

f is induced by an apartments preserving embedding of Π in Π

. In the

case when the image of a certain star is contained in a top, we have the

mapping h : P → (

−γ

(Π

**) (Subsection 4.6.5); it sends every line L ∈ L
**

to a subset of the line of G

−γ

(Π

) corresponding to f(L) (Remark 4.12).

The reader can show that h is an apartments preserving embedding of Π

in G

−γ

(Π

) which induces f.

3. Suppose that our polar spaces are of type D

n

and g is an apartments

preserving mapping of (

δ

(Π) to (

γ

(Π

), δ, γ ∈ ¦+, −¦. By Proposition

4.35, the mapping g is adjacency preserving (two elements of (

δ

(Π) are

adjacent if and only if their images are adjacent). Then it transfers maximal

singular subspaces of G

δ

(Π) (stars and special subspaces) to subsets of

maximal singular subspaces of G

γ

(Π

**). As above, distinct maximal singular
**

subspaces go to subsets of distinct maximal singular subspaces.

There are the following four possibilities for the intersection of two dis-

tinct maximal singular subspaces: the empty set, a single point, a line, and

a plane. For every plane there is an apartment intersecting this plane in a

triangle. In the other cases, the intersections with apartments contain at

most two elements. This means that g transfers every plane to a subset

spanning a plane.

Show that the images of distinct planes are contained in distinct planes.

This guarantees that g maps lines to subsets of lines.

Let A

1

and A

2

be planes of G

δ

(Π) such that g(A

1

) and g(A

2

) are con-

tained in a plane A

of G

γ

(Π

**). Then every element of A
**

1

is adjacent with

every element of A

2

and there exists a maximal singular subspace of G

δ

(Π)

containing both these planes. This implies the existence of an apartment

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

206 Grassmannians of Classical Buildings

/ ⊂ (

δ

(Π) which intersects each A

i

in a triangle; but the apartment g(/)

intersects A

**in a subset containing at most three points. This is possible
**

only in the case when A

1

= A

2

.

Similarly, we establish that distinct lines go to subsets of distinct lines.

Therefore, g induces an injection of (

n−3

(Π) to (

n−3

(Π

). This mapping

is apartments preserving (standard veriﬁcation); hence it is induced by an

apartments preserving embedding of Π in Π

**or in one of the half-spin
**

Grassmann spaces of Π

**(the second possibility can be realized only for
**

n = 4). This embedding induces g.

Remark 4.24. In the case when n = 4, the half-spin Grassmann spaces

G

δ

(Π) and G

γ

(Π

**) are polar spaces of type D
**

4

. By Corollary 4.4 and

Proposition 4.37, g is an embedding of G

δ

(Π) in G

γ

(Π

**). This implies that
**

it transfers lines to subsets of lines and distinct lines go to subsets of distinct

lines.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Bibliography

Abramenko, P. (1996). Twin buildings and applications to S-arithmetic groups,

Lecture Notes in Mathematics 1641 (Springer-Verlag, Berlin).

Abramenko, P. and Van Maldeghem, H. (2000). On opposition in spherical build-

ings and twin buildings, Ann. Comb. 4, 2, pp. 125–137.

Abramenko, P. and Van Maldeghem, H. (2008). Intersections of apartments,

Preprint arXiv:0805.4442, to appear in J. Combin. Theory Ser. A.

Artin, E. (1957). Geometric Algebra (Interscience Publishers, New York-London).

Baer, R. (1952). Linear Algebra and Projective Geometry (Academic Press, New

York).

Blok, R. J. and Brouwer, A. E. (1998). Spanning point-line geometries in buildings

of spherical type, J. Geom. 62, 1-2, pp. 26–35.

Blunck, A. and Havlicek, H. (2005). On bijections that preserve complementarity

of subspaces, Discrete Math. 301, 1, pp. 46–56.

Borovik, A. V., Gelfand, I. M. and White, N. (2003). Coxeter Matroids, Progress

in Mathematics 216 (Birkh¨auser, Boston).

Bourbaki, N. (1968).

´

El´ements de math´ematique. Groupes et alg`ebres de Lie,

Chapters IV–VI, Actualit´es Scientiﬁques et Industrielles 1337 (Hermann,

Paris).

Buekenhout, F. and Cameron, P. (1995). Projective and aﬃne geometry over

division rings, in F. Buekenhout (ed.), Handbook of Incidence Geometry:

Buildings and Foundations (Elsevier, Amsterdam), pp. 27–62.

Buekenhout, F. and Shult, E. (1974). On the foundations of polar geometry,

Geom. Dedicata 3, 2, pp. 155–170.

Buekenhout, F. (1990). On the foundations of polar geometry II, Geom. Dedicata

33, 1, pp. 21–26.

Brown, K. S. (1989). Buildings (Springer-Verlag, New York).

Cameron, P. (1982). Dual polar spaces, Geom. Dedicata 12, 1, pp. 75–85.

Cameron, P. (1991). Projective and polar spaces, QMW Maths Notes 13 (Queen

Mary and Westﬁeld College, London).

Ceccherini, P. V. (1967). Collineazioni e semicollineazioni tra spazi aﬃni o proi-

ettivi, Rend. Mat. e Appl. 26, pp. 309–348.

Chow, W. L. (1949). On the geometry of algebraic homogeneous spaces, Ann. of

207

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

208 Grassmannians of Classical Buildings

Math. 50, 1, pp. 32–67.

Cohen, A. M. (1995). Point-line spaces related to buildings, in F. Buekenhout

(ed.), Handbook of Incidence Geometry: Buildings and Foundations (Else-

vier, Amsterdam), pp. 647–737.

Cooperstein, B. N. and Shult, E.E. (1997). Frames and bases of Lie incidence

geometries, J. Geom. 60, 1-2, pp. 17–46.

Cooperstein, B. N. (2003). Generation of embeddable incidence geometries: a

survey. Topics in diagram geometry, Quad. Mat. 12 (Dept. Math., Seconda

Univ. Napoli, Caserta), pp. 29–57.

Cooperstein, B. N., Kasikova, A. and Shult, E. E. (2005). Witt-type Theorems for

Grassmannians and Lie Incidence Geometries, Adv. Geom. 5, 1, pp. 15–36.

Cooperstein, B. N. (2005). Classical subspaces of symplectic Grassmannians, Bull.

Belg. Math. Soc. Simon Stevin 12, 5, pp. 719–725.

Cooperstein, B. N. (2007). Symplectic subspaces of symplectic Grassmannians,

European. J. Combin. 28, 5, pp. 1442–1454.

Dieudonn´e, J. (1951). On the automorphisms of the classical groups, Mem. Amer.

Math. Soc. 2.

Dieudonn´e, J. (1951). Algebraic homogeneous spaces over ﬁeld of characteristic

two, Proc. Amer. Math. Soc. 2, pp. 295–304.

Dieudonn´e, J. (1971). La G´eom´etrie des Groupes Classiques, Ergebnisse der

Mathematik und ihrer Grenzgebiete 5, 3d edn. (Springer-Verlag, Berlin-

New York).

Garrett, P. (1997). Buildings and Classical groups (Chapman & Hall, London).

Faure, C. A. and Fr¨olicher, A. (1994). Morphisms of projective geometries and

semilinear maps, Geom. Dedicata 53, 3, pp. 237–262.

Faure, C. A. (2002). An elementary proof of the Fundamental Theorem of Pro-

jective Geometry, Geom. Dedicata 90, 1, pp. 145–151.

Fillmore, P. A. and Longstaﬀ, W. E. (1984). On isomorphisms of lattices of closed

subspaces, Canad. J. Math. 36, 5, pp. 820–829.

Havlicek, H. (1994). A generalization of Brauner’s theorem on linear mappings,

Mitt. Math. Sem. Giessen 215, pp. 27–41.

Havlicek, H. (1995). On Isomorphisms of Grassmann Spaces, Mitt. Math. Ges.

Hamburg 14, pp. 117–120.

Havlicek, H. (1999). Chow’s theorem for linear spaces, Discrete Math. 208/209,

pp. 319–324.

Havlicek, H. and Pankov, M. (2005). Transformations on the product of Grass-

mann spaces, Demonstratio Math. 38, 3, pp. 675–688.

Huang, W.-l. and Havlicek, H. (2008). Diameter preserving surjections in the

geometry of matrices, Lin. Algebra Appl. 429, 1, pp. 376–386.

Huang, W.-l. and Kreuzer, A. (1995). Basis preserving maps of linear spaces,

Arch. Math. (Basel) 64, 6, pp. 530–533.

Huang, W.-l. (1998). Adjacency preserving transformations of Grassmann spaces,

Abh. Math. Sem. Univ. Hamburg 68, pp. 65–77.

Huang, W.-l. (2000). Adjacency preserving mappings of invariant subspaces of a

null system, Proc. Amer. Math. Soc. 128, 8, pp. 2451–2455.

Huang, W.-l. (2001). Characterization of the transformation group of the space

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Bibliography 209

of a null system, Results Math. 40, 1-4, pp. 226–232.

Humphreys, J. E. (1975). Linear algebraic groups, Graduate Texts in Mathematics

21 (Springer-Verlag, New York-Heidelberg).

Kakutani, S. and Mackey, G.W (1946). Ring and lattice characterizations of com-

plex Hilbert space, Bull. Amer. Math. Soc. 52, 8, pp. 727–733.

Karzel, H., S¨orensen, K. and Windelberg, D. (1973). Einf¨ uhrung in die Geometrie,

Uni-Taschenb¨ ucher 184 (Vandenhoeck & Ruprecht, G¨ottingen).

Kreuzer, A. (1996). On the deﬁnition of isomorphisms of linear spaces, Geom.

Dedicata 61, 3, pp. 279–283.

Kreuzer, A. (1998). On isomorphisms of Grassmann spaces, Aequationes Math.

56, 3, pp. 243–250.

Kwiatkowski, M. and Pankov, M. (2009). Opposite relation on dual polar spaces

and half-spin Grassmann spaces, Results Math. 54, 3-4, pp. 301–308.

Lim, M. H. (2010). Surjections on Grassmannians preserving pairs of elements

with bounded distance, Linear Algebra Appl. 432, 7, pp. 1703–1707.

Mackey, G. W. (1942). Isomorphisms of normed linear spaces, Ann. of Math. 43,

2, pp. 244–260.

Pankov, M. (2002). Transformations of grassmannians and automorphisms of

classical groups, J. Geom. 75, 1-2, pp. 132–150.

Pankov, M. (2004). Chow’s theorem and projective polarities, Geom. Dedicata

107, pp. 17–24.

Pankov, M. (2004). Transformations of Grassmannians preserving the class of

base sets, J. Geom. 79, 1-2, pp. 169–176.

Pankov, M. (2004). A characterization of geometrical mappings of grassmann

spaces, Results Math. 45, 3-4, pp. 319–327.

Pankov, M. (2004). Mappings of the sets of invariant subspaces of null systems,

Beitr¨age Algebra Geom. 45, 2, pp. 389–399.

Pankov, M., Pra˙ zmowski, K. and

˙

Zynel, M. (2005). Transformations preserving

adjacency and base subsets of spine spaces, Abh. Math. Sem. Univ. Hamb.

75, pp. 21–50.

Pankov, M. (2005). On the geometry of linear involutions, Adv. Geom. 5, 3, pp.

455–467.

Pankov, M., Pra˙ zmowski, K. and

˙

Zynel, M. (2006). Geometry of polar grassmann

spaces, Demonstratio Math. 39, 3, pp. 625–637.

Pankov, M. (2006). Geometrical mappings of Grassmannians of linear spaces

satisfying the exchange axiom, Aequationes Math. 72, 3, pp. 254–268.

Pankov, M. (2007). Base subsets of Grassmannians: inﬁnite-dimensional case,

European. J. Combin. 28, 1, pp. 26–32.

Pankov, M. (2007). Base subsets of symplectic Grassmannians, J. Algebraic Com-

bin. 26, 2, pp. 143–159.

Pankov, M. (2007). Base subsets of polar Grassmannians, J. Combin. Theory Ser.

A 114, 8, pp. 1394–1406.

Pasini, A. (1994). Diagram geometries, Oxford Science Publications (Clarendon

Press, Oxford University Press, New York).

Pra˙ zmowski, K. (2001). On a construction of aﬃne Grassmannians and spine

spaces, J. Geom. 72, 1-2, pp. 172–187.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

210 Grassmannians of Classical Buildings

Pra˙ zmowski, K. and

˙

Zynel, M. (2002). Automorphisms of spine spaces, Abh.

Math. Sem. Univ. Hamb. 72, pp. 59–77.

Pra˙ zmowski, K. and

˙

Zynel, M. (2009). Segre subproduct, its geometry, automor-

phisms and examples, J. Geom. 92, 1-2, pp. 117-142.

Rickart, C. E. (1950). Isomorphic groups of linear transformations, Amer. J.

Math. 72, pp. 451-464.

Ronan, M. (1989). Lectures on buildings, Perspectives in Mathematics 7 (Aca-

demic Press, Boston).

Rudin, W. (1973). Function Analysis, McGraw-Hill Series in Higher Mathematics

(McGraw-Hill, New York-D¨ usseldorf-Johannesburg).

Scharlau, R. (1995). Buildings, in F. Buekenhout (ed.), Handbook of Incidence

Geometry: Buildings and Foundations (Elsevier, Amsterdam), pp. 477–646.

Taylor, D. E. (1992). Geometry of the classical groups, Sigma Series in Pure

Mathematics 9 (Heldermann Verlag, Berlin).

Teirlinck, L. (1980). On projective and aﬃne hyperplanes, J. Combin. Theory

Ser. A 28, 3, pp. 290–306.

Tits, J. (1974). Buildings of spherical type and ﬁnite BN-pairs, Lecture Notes in

Mathematics 386 (Springer-Verlag, Berlin-New York).

Veldkamp, F. D. (1959/1960). Polar geometry I – V, Nederl. Akad. Wet. Proc.

Ser. A 62, pp. 512–551, Nederl. Akad. Wet. Proc. Ser. A 63, pp. 207–212.

Wan, Z.-X. (1996). Geometry of Matrices (World Scientiﬁc, Singapore).

Zick, W. (1983). Lineare Abbildungen in reellen Vektorr¨aumen, Math. Semester-

ber 30, 2, pp. 167–170.

˙

Zynel, M. (2000). Subspaces and embeddings of spaces of pensils, Preprint, Uni-

versity in Bialystok.

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

Index

adjacency relation

on a Grassmannian, 48

on Grassmannians, 70, 97, 114,

133, 140

on the set of chambers, 46

on the set of involutions, 106

weak, 161

annihilator, 16

apartment

in a building, 56

in a Grassmannian, 65, 83, 178, 182

in a parabolic subspace, 84, 180,

182

base

of a partial linear space, 18

of a vector space, 12

base subset

in a Grassmannian of an exchange

space, 97

in a Grassmannian of an

inﬁnite-dimensional vector

space, 118

in a spine space, 101

in the set of involutions, 103

big star, 162

building

abstract, 56

associated with a Tits system, 58

of the linear group, 59

of the orthogonal group, 61

of the symplectic group, 60

polar, 150

spherical, 62

collinearity graph, 18

collineaton, 17

complement subset, 90, 194, 199

complex

chamber, 46

Coxeter, 52

labeled, 46

oriﬂamme, 45, 150

simplicial, 43

contragradient, 25

Coxeter system, 49

diagram, 49

dimension

of a partial linear space, 18

of a vector space, 12

distance, 18

division ring, 8

embedding

Grassmann, 72

of a partial linear space, 17

semilinear, 24

exact subset, 89, 187, 199

form

alternating, 38

Hermitian, 39

quadratic, 146

211

July 2, 2010 14:9 World Scientiﬁc Book - 9in x 6in ClassicalBuilding

212 Grassmannians of Classical Buildings

reﬂexive, 38

sesquilinear, 38

skew-symmetric, 38

symmetric, 38

frame, 130

generalized quadrangle, 128

geodesic, 18

graph

Grassmann, 48, 71, 77, 97, 104,

114, 150

half-cube, 182

hypercube, 179

Johnson, 83

polar Johnson, 179

weak Grassmann, 161

Grassmannian

half-spin, 139

of a chamber complex, 47

of a vector space, 14

of an exchange space, 96

polar, 133

homothetic transformation, 21

hyperplane, 18

independent subset

of a partial linear space, 18

of a vector space, 11

inexact subset, 66, 89, 187, 199

involution, 102

lattice, 28

mapping

apartments preserving, 64, 65

conjugate-linear, 33

linear, 11

semilinear, 20

opposite relation

on Grassmannians, 80, 174

on the set of chambers, 62

panel, 46

polarity, 40

projective plane, 18

rank of a simplicial complex, 45

semicollineaton, 17

space

dual projective, 20

dual vector, 14

exchange, 95

gamma, 67

generalized projective, 78

Grassmann, 48, 71

half-spin Grassmann, 140

linear, 17

normed, 32

partial linear, 17

polar, 125

polar Grassmann, 134

projective, 18

projective(over a division ring), 19

spine, 101

vector, 10

Veldkamp, 149

star, 73, 97, 152, 156

subspace

of a partial linear space, 18

of a vector space, 11

parabolic, 74, 153, 158

singular, 18

special, 155

totally isotropic, 39

top, 73, 97, 152

transvection, 104

triangle, 18, 71, 151, 154

Witt index, 143

Algebra and Discrete Mathematics

Vol. 2

Algebra and Discrete Mathematics

ISSN: 1793-5873

Managing Editor: Rüdiger Göbel (University Duisburg-Essen, Germany) Editorial Board: Elisabeth Bouscaren, Manfred Droste, Katsuya Eda, Emmanuel Dror Farjoun, Angus MacIntyre, H.Dugald Macpherson, José Antonio de la Peña, Luigi Salce, Mark Sapir, Lutz Strüngmann, Simon Thomas The series ADM focuses on recent developments in all branches of algebra and topics closely connected. In particular, it emphasizes combinatorics, set theoretical methods, model theory and interplay between various fields, and their influence on algebra and more general discrete structures. The publications of this series are of special interest to researchers, post-doctorals and graduate students. It is the intention of the editors to support fascinating new activities of research and to spread the new developments to the entire mathematical community.

Vol. 1:

Aspects of Infinite Groups: A Festschrift in Honor of Anthony Gaglione eds. Benjamin Fine, Gerhard Rosenberger & Dennis Spellman Grassmannians of Classical Buildings by Mark Pankov

Vol. 2:

Algebra and Discrete Mathematics Vol. 2

Mark Pankov

University of Warmia and Mazury, Poland

World Scientific

NEW JERSEY

•

LONDON

•

SINGAPORE

•

BEIJING

•

SHANGHAI

•

HONG KONG

•

TA I P E I

•

CHENNAI

GRASSMANNIANS OF CLASSICAL BUILDINGS Algebra and Discrete Mathematics — Vol. Covent Garden. . including photocopying. Ltd. ISBN-13 978-981-4317-56-6 ISBN-10 981-4317-56-X Printed in Singapore. For photocopying of material in this volume. MA 01923. Danvers. Singapore 596224 USA office: 27 Warren Street. All rights reserved. USA. Pte.Published by World Scientific Publishing Co. In this case permission to photocopy is not required from the publisher. 2 Copyright © 2010 by World Scientific Publishing Co. Suite 401-402. electronic or mechanical. NJ 07601 UK office: 57 Shelton Street. without written permission from the Publisher. London WC2H 9HE British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. Inc. or parts thereof. This book. 5 Toh Tuck Link. Hackensack. Pte.. recording or any information storage and retrieval system now known or to be invented. Ltd. may not be reproduced in any form or by any means. please pay a copying fee through the Copyright Clearance Center. 222 Rosewood Drive.

Dedicated to my wife Inna and daughters Sabina and Varvara .

This page intentionally left blank .

A large vii . This term is motivated by the fact that every building of type An is isomorphic to the ﬂag complex of an (n + 1)dimensional vector space and the Grassmannians of the building can be identiﬁed with the Grassmannians of this vector space. All apartments are isomorphic to the simplicial complex obtained from a Coxeter system which deﬁnes the building type. The aim of this book is to present both classical and more recent results on Grassmannians of buildings of classical types (An . The set of all vertices corresponding to the same node is called a Grassmannian (more general objects were investigated in [Pasini (1994)]). These results will be formulated in terms of point-line geometry. Such groups form a suﬃciently wide class which contains classical groups. this partial linear space is called the Grassmann space associated with the Grassmannian.Preface Tits buildings or simple buildings are combinatorial constructions successfully exploited to study various types of groups (classical. The vertex set of a building can be labeled by the nodes of the diagram of the associated Coxeter system. Scharlau (1995). Brown (1989). Kac–Moody). Buildings can be obtained from groups admitting Tits systems. Every building Grassmannian has a natural structure of a partial linear space (point-line geometry). In [Tits (1974)] a building is deﬁned as a simplicial complex with a family of subcomplexes called apartments and satisfying certain axioms. reductive algebraic groups and others. Garrett (1997). Tits (1974)] for various aspects of building theory. Dn ). One of historical backgrounds of this concept is Cartan’s well-known classiﬁcation of simple Lie groups. Bn = Cn . simple algebraic. Ronan (1989). The formal deﬁnition of a building is pure combinatorial and does not depend on a group. We refer to [Abramenko (1996).

Acknowledge The present book contains the main results of my Doctor of Science (habilitation) thesis presented in Institute of Mathematics NASU and I thank A. Kwiatkowski for useful discussions and remarks. Samoilenko and V.viii Grassmannians of Classical Buildings portion of them is a part of the area known as characterizations of geometrical transformations under mild hypotheses. Also I am grateful to H. Sharko for supporting my research. combinatorics and geometry. Kosiorek for drawing the pictures. Mark Pankov . I thank J. Roughly speaking. Havlicek and M. Other results are related with structural properties of apartments. I am grateful to H. Van Maldeghem who read a preliminary version of the book and made a long list of valuable comments. Finally. The book is self-contained and prospective audience includes researchers working in algebra. we want to show that some mappings of Grassmannians can be extended to mappings of the associated buildings. graduate and advanced undergraduate students. The requirement to the reader is knowledge of basics of algebra and graph theory. as well as. the sets of conjugate linear involutions and Grassmannians of exchange spaces. Also we show that our methods work for some geometric constructions non-related with buildings — Grassmannians of inﬁnity-dimensional vector spaces.

. . .1 Linear and partial linear spaces . . . . . . . . . . .4 Proof of Theorem 1.3 Contragradient .4. . . . .1 Main theorem and corollaries . . . .2 Vector spaces over division rings . . . . . 1. . . . . . . . . . . . . 1. .4 1. . . . . . . . .1. . .1 Division rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . .3 Dual vector space . . . . . .4 . .3. . . . . . . . . . .5 . . . . . . . . . . . . . . . . . . Fundamental Theorem of Projective Geometry . 1. . . . . . . . . . 1. . .1 Sesquilinear forms . . . . . . . . . .5. . .2 Proof of Theorem 1. . . . . .3 1.2 Projective spaces over division rings . .3 Polarities . . . .5. . . . . . . Projective spaces . . . . .5. . . .4. .1. . . . Linear Algebra and Projective Geometry 1. . . . .1 Deﬁnitions . . . Introduction 1. . . . . . . Semilinear mappings . .3. . 1. . . Reﬂexive forms and polarities . .2 Mappings of Grassmannians induced by semilinear mappings . . . . .3.Contents Preface 0. . . . . . . . . . . . 1. .2 Reﬂexive forms . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . .2. . .2 1. . . . . .1 Vector spaces .5 . . . ix vii 1 7 8 8 10 14 17 17 19 20 20 21 25 26 26 28 32 33 37 37 38 40 1. . . . . . . . . . . . . . 1.1. . . . 1. . . . . . .4. 1. 1. . . . . . . . . . . . . . .2. . . . . . . . . 1. . .3 Fundamental Theorem for normed spaces . . . 1. 1. . . . . . . . . . . . . . . . . .

. Apartments preserving mappings . . . .4 Spherical buildings . . . 3.3. . . .3 Three examples . . . .5. . . . . 80 . . .3 3. . . . . . . . 2. . . .3. 100 .1 Involutions and transvections .3. . 3. . . . . . 96 . . . . . . . .2 2. . . . . . . . . . . . . . 3. Buildings and Grassmannians 2.1 Deﬁnition and elementary properties . .2 Proof of Theorem 3.2 Elementary properties of Grassmann spaces Collineations of Grassmann spaces . . . 102 . . . . . 2. . . . . . . . . . .2 Coxeter complexes . . . . 102 2. . . .2. . . . . . . . . Appendix: Gamma spaces . . . . . .4. .1 Deﬁnition and examples . . . . . . . . . 3. . . . . 83 . . . . . . .2. . . . .1 Basic properties . . . . 3. . .2.1 3. . . . . . . . . .3 2. . . . . 93 . . . .3 Classical examples . . .7. . . 3. .1.5 3. . . . . . . . .5 3. . . . . . . . . . . . . . . . . . . . . .8 . .2 Proof of Theorem 3. . .7 . . . . . 85 . . . . .2 Grassmannians . . .3 Grassmannians and Grassmann spaces Coxeter systems and Coxeter complexes . 87 . . . . . . 3. . . 2. . . . . 2. . .1. .10: Second step Grassmannians of exchange spaces . . .10: First step . . . .4 3.4 Opposite relation . . 2. . . . . . . . . . . . . . Buildings . . . . 89 . Apartments . . . . . .3. . . . .3. . . . . . . . . . 2. . .4 2. . . .1 Coxeter systems . . . . . . . . . . .1 Results . 3.5. . . 2. . . . . 2. . . . . . . 3.2. .3. . . . . . Mappings of Grassmannians . . . 2. 75 . . . .5 Mappings of the chamber sets . . . . . . . . . . 95 . . . . . . . . . Classical Grassmannians 3. . .4. Geometry of linear involutions . . . . . . . . . . . . .1. . . . . . .6 3. . . . . . . . . . . . . . 78 . . . . 3. . . .x Grassmannians of Classical Buildings 2. . . 43 43 43 46 47 49 49 52 53 55 55 57 59 62 63 65 67 69 . . . 75 . . . . . Matrix geometry and spine spaces . . . . . . . . . . . . . . . . . . . . . 87 . . .2. . 95 . . . . . . 83 . . . . . 3. . . . . . .2. 2. . . . . . . . . . . . . .3. . . . . . . .3 Applications of Chow’s theorem . . . . . . .2. . . . . . . . . . .3 Proof of Theorem 3. . . . . . .2 Buildings and Tits systems . . . . .1 Chow’s theorem . . . . . . . . . . . . . . 70 . . . . . . . . . . . 3. . . . . . . . . . . . . 2. . .4. . . .2 Chow’s theorem for linear spaces . .1 Simplicial complexes . . . . . . .1 Exchange spaces .2 Chamber complexes . . . . . . . . . . . . . . . . 3. . . . . 77 .

4. . . .18 . . . . 3. . . .6 Remarks . . . . . . . Grassmannians . . . . . . . .8. . . . . .1. . .3. . . . . . . . . . . . .Contents xi 3. . . 4. . 4. 3. 4. . . . . . . . 4. . . . . . . .1 Buildings of type Cn . . . .8 . 4. . . 4. . . . . . . . .8. . . . . . . . . . . .1 Chow’s theorem and its generalizations . . . . .5 Automorphisms of the group GL(V ) . . . . . .4. . . . . . . . .4 Proof of Theorem 3. . . . . . . 4. . . . . . . . . . .8.3 Corollaries of Theorem 4. . . . . . . . . . Grassmannians of inﬁnite-dimensional vector spaces 3. . 4. . . . . . . .2 Proof of Theorem 3. . . . . . . . . .8. . . . . 4.1 Polar Grassmann spaces . .6. . 3. . .4 Proof of Theorems 4.2 4. . . . .2 Weak adjacency on polar Grassmannians . . .7. . . .8 3. .6. . . . . . . . . . . . . . . . 4. . . .6. . . . . . . . . . . .1 .4 Embeddings in projective spaces and classiﬁcation Polar buildings . . .1.2. . . . . .3 Half-spin Grassmannians .2 Half-spin Grassmann spaces . . . . . 4. . 4. . . . Examples .2. . . . .7 and 4. . . . . . . . . . . . . . . . . . . 4. . . . .6 4. . . .2 Adjacency relation . . . . . . . . . . . .1 Polar Grassmannians . . . . . . . .9 . . . . . . . .4. . . . . .4 Proof of Theorem 3. . . . . . . .7. . 3.3 4. .6. . . . 4. .3. . 4. .2 Proof of Theorem 4. . . .3 Chow’s theorem for linear involutions . . . . . . . . . . . . . . . . 4. . . . 4. . . . . .6. . . . . . . . . . . . . . . . . 104 108 110 113 114 114 116 118 118 123 125 125 126 129 130 133 133 136 138 141 141 146 147 149 150 150 150 151 151 154 159 159 161 162 164 165 172 174 4. . . . . . . . . . . .3 Base subsets . .2 Polar spaces associated with quadratic forms . . . . . . . . . . . . . . . . . .17 . . . . . . . . . . . . . . . . . . . . . .1 Polar spaces . . . . . . .1.7. 4. . . .3 Proof of Theorem 4. . . .5 Proof of Theorem 4. 3. . . . . . .3. . .5. . .1. . . 3. .1 Adjacency relation . . . . . .1 Axioms and elementary properties . . . . . . . . . Opposite relation .8 for k < n − 2 . . Elementary properties of Grassmann spaces . . . . .2. . . . . . . . . . . . . . . . . . . 4.6. . . . . . .3.4 Polar frames . .15 . . . . .5. . . . . . . . . .3 Polar spaces of type D3 . .7 . . . .2 Buildings of type Dn . . 4. 4. . . . . . . . . . . . . . . . . .2 Two types of polar spaces . . . Collineations .1 Polar spaces associated with sesquilinear forms . . . . Polar and Half-Spin Grassmannians 4. . . . .5 4.4 4. . . . .7. . . . . . .1 . .

8. 4. . . 4. . . . . . . 4. . . . . . . . . .9. . . .8. . . . . . . . .7 Proof of Theorems 4. . . .9.9. . . . . Apartments preserving mappings . . . .16 . .3 Complement subsets of polar Grassmannians . . . . .9.2 Opposite relation on half-spin Grassmannians Apartments . . . 4. . . . . . . . . .2 Apartments in half-spin Grassmannians . . . . . . . . . .7.8. . 4. . .1 Apartments in polar Grassmannians . . . . . .5 Proof of Theorem 4. . . 4. .9. . . .2 Inexact subsets of polar Grassmannians .1 Apartments preserving bijections . .17 and 4. .15 . . . . . . . . . . . . . . . . 4.3 Proof of Theorem 4. . . 4. . . . .18 .9. . .6 Embeddings . . .xii Grassmannians of Classical Buildings 4. . . . . . . 174 175 178 178 181 184 186 186 187 194 199 201 202 203 207 211 Bibliography Index . .9 4. .4 Inexact subsets of half-spin Grassmannians . . . .7. . 4. . . . . . .9. . .1 Opposite relation on polar Grassmannians .8 4. . . . . 4. . . 4.

A Tits system (G. There is a complete description of all ﬁnite Coxeter systems.s ) = 1. (s. S) is a Coxeter system. m(s. The pair (W. N ) is called a BN-pair.Chapter 0 Introduction This short chapter is an informal description of the main objects of the book — buildings and their Grassmannians. s ∈ S are connected by m(s. Similarly. s ) < ∞ . 1 . By one of the basic properties of the Tits systems. B. Let W be a group and S be a set of generators for W such that each element of S is an involution. N. Moreover. s ) ∈ S × S. N which span G and a set of generators S of the quotient group W := N/(B ∩ N ). The simplest buildings are so-called Coxeter complexes — a class of simplicial complexes deﬁned by Coxeter systems. S) is the simplicial complex whose simplices can be identiﬁed with special subsets of type w X with X ⊂ S and w ∈ W . S) is a structure on a group G consisting of two subgroups B. we get a Dynkin diagram without directions. can be reconstructed from elements of W and B. The diagram associated with (W. s ) is the order of ss . The second remarkable property is the fact that every subgroup containing B. The associated Coxeter complex Σ(W. such subgroups are called special. Every Coxeter system can be uniquely (up to an isomorphism) reconstructed from its diagram. note that the pair (B. s ) − 2 edges. where m(s. more complicated buildings can be obtained from the Tits systems. S) is the graph whose vertex set is S and s. All precise deﬁnitions will be given in Chapter 2. In the case when W is ﬁnite. S) is a Coxeter system if the group W has the following presentation S : (ss )m(s. (W.

. . S) be the associated Coxeter system. The vertex set of ∆ can be naturally decomposed in |S| disjoint subsets called Grassmannians. i. . in such buildings maximal simplices form a suﬃciently wide class. S). The simplicial complex ∆(V ) together with the family of all such apartments is a building of type An . If w ∈ W then for every g. S) is the simplicial complex ∆(G. N. n + 1} and S is formed by all transpositions (i. These subcomplexes are isomorphic to Σ(W. there exists a Coxeter system (W. where g ∈ G and P is a special subgroup. i + 1). . S). We restrict ourselves to so-called thick buildings only.2 Grassmannians of Classical Buildings there is a natural one-to-one correspondence between special subgroups and subgroups of W generated by subsets of S. the simplicial complex consisting of all ﬂags of V . Coxeter complexes do not satisfy this condition. we deﬁne an abstract building as a simplicial complex ∆ with a family of subcomplexes called apartments and satisfying the following axioms: • all apartments are Coxeter complexes. S). S). The building associated with the Tits system (G. For every base B of V the subcomplex ΣB which consists of all ﬂags formed by linear subspaces spanned by subsets of B is called the apartment of ∆(V ) associated with the base B. every Coxeter complex is a building with a unique apartment. B. Let V be an (n+ 1)-dimensional vector space and ∆(V ) be the ﬂag complex of V . Note that this building can be obtained from the Tits system of the group GL(V ). denote this special subset by wP . • for any two simplices of ∆ there is an apartment containing both of them. g ∈ G. • a technical condition concerning the existence of “nice” isomorphisms between apartments. So.e. B. By this deﬁnition. S) whose simplices can be identiﬁed with special subsets of type gP . the associated diagram is An . All such special subsets form a subcomplex Σ isomorphic to Σ(W. Following [Tits (1974)]. Let ∆ be a building and (W. Every ΣB is isomorphic to the simplicial complex of the Coxeter system (W. The type of the building ∆ is deﬁned by the diagram of (W. S) such that all apartments of ∆ are isomorphic to Σ(W. g ∈ N belonging to w and every special subgroup P we have gP = g P . N. If the building is associated with a Tits system for a certain group G then the Grassmannians are the orbits of the left action of G on . S) and called apartments. where W is the group of all permutations on the set {1. . The left action of the group G on the building deﬁnes the family of subcomplexes gΣ.

The Grassmannians of the building ∆(V ) are the usual Grassmannians Gk (V ). .. . a set of points together with a family of subsets called lines and satisfying some simple axioms. Two distinct elements a. Note that our adjacency relation coincides with the classical adjacency relation introduced in [Chow (1949)]: two elements of Gk (V ) are adjacent if and only if their intersection is (k − 1)-dimensional. n}. Let G be a Grassmannian of ∆. The Grassmann spaces corresponding to G1 (V ) and Gn (V ) are the projective space associated with V and the dual projective space. Recall that a building is spherical if the associated Coxeter system is ﬁnite. Let V be an (n + 1)-dimensional vector space. the subset formed by all c ∈ G such that P ∪ {c} is a maximal simplex will be called the line joining a and b. this decomposition is related to the fact that the vertex set of ∆ can be labeled by elements of S and this labeling is unique up to a permutation on S.Introduction 3 the vertex set. In the general case. we get a structure known as a partial linear space or a point-line geometry. It contains non-adjacent elements. b ∈ G are said to be adjacent if there exists a simplex P ∈ ∆ such that P ∪ {a} and P ∪ {b} are maximal simplices. In particular. Irreducible thick spherical buildings of rank ≥ 3 were classiﬁed in [Tits (1974)]. respectively. any two distinct elements of these Grassmannians are adjacent. In this case. The Grassmann space of Gk (V ).e. The intersection of G with an apartment of ∆ is called an apartment of the Grassmannian G. There are precisely the following seven types of such buildings: An Bn = C n Dn . So. and it is irreducible if the diagram is connected. This partial linear space is said to be the Grassmann space associated with G. 1 < k < n. i. . k ∈ {1. The term “Grassmannian” is motivated by the following example. . formed by all k-dimensional linear subspaces of V . is more complicated.

Exceptional buildings are related with so-called metasymplectic and parapolar spaces. • embeddings in projective spaces. 2007)]. a survey can be found in [Cohen (1995)]. All thick buildings of types Cn and Dn can be obtained from polar spaces. Kasikova and Shult (2005). see [Cohen (1995)]. Kasikova and Shult (2005). Cooperstein and Shult (1997). Dieudonn´ 2 (1951)] (see also Chapter III in [Dieudonn´ (1971)]) and is e e continued in [Cameron (1982)]. hyperplanes and generalized rank. the remaining four are known as exceptional. Some information concerning Grassmannians of exceptional buildings can be found in [Cohen (1995)]. In the present book we will consider Grassmannians associated with buildings of classical types only.4 Grassmannians of Classical Buildings E6 E7 E8 F4 The ﬁrst three types are called classical. see Chapter 4. Every thick building of type An (n ≥ 3) is isomorphic to the ﬂag complex of a certain (n + 1)-dimensional vector space. Investigation of building Grassmannians goes back to [Chow (1949). Pankov 3 (2007)]. there are several research directions: • axiomatic characterizations of Grassmann spaces. . Currently. • subspaces of Grassmann spaces [Cooperstein. • apartment properties [Blok and Brouwer (1998). see [Cooperstein (2003)] for a survey. Cooperstein. Cooperstein (2005.

The methods are based on deep structural properties of apartments (connections between the adjacency relation and apartments). we work in the latter two directions mentioned above. for example. Also. Pra˙ mowski and z ˙ Zynel (2006)] and the author’s papers refereed in the book. Rickart (1950)]. Pankov. Kreuzer (1998). Huang and Kreuzer (1995).Introduction 5 • characterizations of geometrical transformations of Grassmannians under “mild hypotheses” [Havlicek (1995). we establish similar results for some geometric constructions non-related with buildings. Huang (1998. We describe all apartments preserving mappings of Grassmannians associated with buildings of classical types and collineations (isomorphisms) of the corresponding Grassmann spaces. Actually. 2000. the Fundamental Theorem of Projective Geometry and classical Chow’s theorems [Chow (1949)] are partial cases of our results. Grassmannians of inﬁnitedimensional vector spaces and the sets of conjugate linear involutions. 2001). One of them joins Chow’s theorem with Dieudonn´–Rickart’s classiﬁcation of e e automorphisms of the linear group [Dieudonn´ 1 (1951). Roughly speaking. .

This page intentionally left blank .

1 and 1. The second result of Section 1. The Fundamental Theorem of Projective Geometry can be reformulated in the following form [Baer (1952)]: every isomorphism of the lattices formed by linear subspaces of vector spaces is induced by a semilinear isomorphism.4 we prove a more general result known as Faure–Fr¨licher–Havlicek’s version of the o Fundamental Theorem. The main objects of the chapter are so-called semilinear mappings of vector spaces and the mappings of Grassmannians induced by them (Section 1.2 we consider vector spaces over division rings and the associated projective spaces. In Section 1. In Section 1. This theorem states that all such isomorphisms are induced by linear and conjugate-linear homeomorphisms of the associated normed spaces (the second possibility is realized only in the complex case).4 is Mackey’s theorem concerning isomorphisms of the lattices formed by closed linear subspaces of normed vector spaces (real and complex). Mackey’s result is interpreted as the Fundamental Theorem for normed spaces. it must be pointed out that vector spaces are not assumed to be ﬁnite-dimensional.5 we recall basic facts on reﬂexive sesquilinear forms.Chapter 1 Linear Algebra and Projective Geometry In Sections 1. 7 . By this reason.3). every collineation (isomorphism) of projective spaces is induced by a semilinear isomorphism of the associated vector spaces. By the Fundamental Theorem of Projective Geometry.

see. and we have R∗ = R only in the case when R is a ﬁeld. ·) is a group (the identity element is denoted by 1). for example. c ∈ R hold then we say that (R. σ(1) = 1. It is clear that R∗∗ coincides with R. Proposition 1. If the distributive axioms a(b + c) = ab + ac and (b + c)a = ba + ca ∀ a. the multiplicative operation ∗ on R∗ is deﬁned by the formula a ∗ b = ba. +) is an Abelian group (the identity element is denoted by 0). Show that the equalities σ(0) = 0. Theorem 1. +. H. Every non-zero homomorphism σ : R → R is injective. . We say that a mapping σ : R → R is a homomorphism if σ(a + b) = σ(a) + σ(b) and σ(ab) = σ(a)σ(b) for all a. Every ﬁnite division ring is a ﬁeld.1 (J. Wedderburn). For every division ring R we deﬁne the opposite (or dual ) division ring R∗ as follows: R and R∗ have the same set of elements and the same additive operation. b.1 1.2.1. Theorem 1. • (R \ {0}. A division ring is called a ﬁeld if the multiplicative operation is commutative. Exercise 1.1. The following result is well-known. Exercise 1. σ(−1) = −1 hold for every non-zero homomorphism σ : R → R . Let R and R be division rings. b ∈ R.14 in [Artin (1957)].8 Grassmannians of Classical Buildings 1.1. Show that 0a = a0 = 0 and −1a = −a for every a ∈ R.1 Vector spaces Division rings Let R be a non-empty set with additive and multiplicative operations + and · satisfying the following conditions: • (R. ·) is a division ring. Bijective homomorphisms are called isomorphisms. M.

In particular. for example. For every real number a > 0 there exists b ∈ R such that a = b2 . n n . Hint: the restriction of every non-zero homomorphism σ : C → C to Q is identity and σ(i) = ±i. An isomorphism of a division ring to itself is said to be an automorphism. The group formed by all automorphisms of R is denoted by Aut(R). Every non-zero homomorphism of R to itself is identity.3. On the other hand. The group Aut(C) is not trivial. It contains. Show that every non-zero continuous homomorphism of C to itself is identity or coincides with the complex conjugate mapping. Let σ : R → R be a homomorphism. Proposition 1. Since σ preserves division. ¯ Exercise 1. If σ : R → R is not injective then σ(a) = 0 for a certain non-zero element a ∈ R and we have σ(b) = σ(aa−1 b) = σ(a)σ(a−1 b) = 0 for every b ∈ R. For every n ∈ N we have n = 1 + · · · + 1. An isomorphism of R to the opposite division ring R∗ is called anti-automorphism of R (if R is a ﬁeld then every anti-automorphism is an automorphism). σ sends every section to itself (its restriction to Q is identity) and we get σ(a) = a.Linear Algebra and Projective Geometry 9 Proof. and we have σ(a) = σ(b)2 > 0.2. This implies that σ(m) = m for all m ∈ Z. { r ∈ Q : a < r }. it transfers this section in the section corresponding to σ(a). Since σ is order preserving. n hence σ(n) = σ(1) + · · · + σ(1) = 1 + · · · + 1 = n. Proof. Every a ∈ R \ Q corresponds to unique section of rational numbers { q ∈ Q : q < a }. the restriction of σ to Q is identity. the complex conjugate mapping z → z . This means that σ is order preserving: if a > b then a − b > 0 and σ(a) − σ(b) = σ(a − b) > 0 which implies that σ(a) > σ(b). Aut(R) = {1R }.

Every automorphism of H is inner (x → qxq −1 for a certain q = 0). y ∈ V hold then we say that V is a left vector space over R. Example 1. every right action of R on V satisfying the conditions . but they are noncontinuous. In the general case. our ﬁeld consists of pr elements. its multiplicative group is a cyclic group of order pr − 1. b ∈ R. r) (p is a prime number).1. b ∈ R and x ∈ V . moreover. there exist non-surjective homomorphisms of C to itself.2 Vector spaces over division rings Let R be a division ring and (V. If the distributive axioms a(x + y) = ax + ay and (a + b)x = ax + bx ∀ a. jk = i. This division ring is non-commutative (for example. x. +) be an Abelian group (the identity element is 0).2 (The division ring of real quaternions). ki = j. (a.10 Grassmannians of Classical Buildings There are a lot of other automorphisms of C. ik = i2 j = −j = −ki). i. Example 1. the group of automorphisms is a cyclic group of order r generated by the automorphism a → ap . respectively.1 (Finite ﬁelds). If r = 1 then we get Zp and the group of automorphisms is trivial. Consider the division ring H formed by the real quaternion numbers a + bi + cj + dk. j. Similarly. Elements of V and R are called vectors and scalars. 1. Let also R × V → V. x) → ax be a left action of R on V satisfying the following conditions: (1) 1x = x for all x ∈ V . k) with the multiplicative operation deﬁned by the following conditions i2 = j 2 = k 2 = −1. ij = k. (2) a(bx) = (ab)x for all a. The conjugate mapping a + bi + cj + dk → a − bi − cj − dk is an anti-automorphism. Every ﬁnite ﬁeld coincides with a certain Galois’ ﬁeld GF(p. This is a 4-dimensional vector space over R (the canonical base consists of 1.

this subspace consists of all linear combinations a1 x1 + · · · + ak xk . .Linear Algebra and Projective Geometry 11 (3) x1 = x for all x ∈ V . where ∗ is the multiplicative operation of the opposite division ring. . . . xk ∈ X and a1 . {0} and V are linear subspaces. A subset S ⊂ V is a linear subspace of V if ax + by ∈ S for all x. A subset X ⊂ V is said to be independent if the linear subspace X is not spanned by a proper subset of X. . . then (4) must be rewritten as a(bx) = (ba)x or a(bx) = (a ∗ b)x. xk ∈ X the equality a1 x1 + · · · + ak xk = 0 holds only in the case when all scalars a1 . in other words. Linear bijections are called linear isomorphisms. By this deﬁnition. Clearly. and conversely. b ∈ R and x ∈ V deﬁnes a right vector space over R if the corresponding distributive axioms hold. y ∈ S and a. (4) (xb)a = x(ba) for all a. Therefore. . y ∈ V and a ∈ R. . . every right vector space over R can be presented as a left vector space over R∗ . . . it is called the kernel of l and denoted by Ker l. this action can be considered as a left action. . The intersection of any collection of linear subspaces is a linear subspace. For every subset X ⊂ V the intersection of all linear subspaces containing X is called the linear subspace spanned by X and denoted by X . b ∈ R. . . For every linear mapping l : V → V all vectors x ∈ V satisfying l(x) = 0 form a linear subspace of V . . . for any distinct non-zero vectors x1 . where x1 . The group of all linear automorphisms of V (linear isomorphisms of V to itself) is denoted by GL(V ). Let V and V be left vector spaces over a division ring R. . A mapping l : V → V is said to be linear if l(x + y) = l(x) + l(y) and l(ax) = al(x) for all x. In what follows we restrict ourselves to left vector spaces only. We say that the vector spaces V and V are isomorphic if there exists a linear bijection of V to V . ak ∈ R. ak are zero.

a contradiction. the set X has maximal elements. . . . .12 Grassmannians of Classical Buildings An independent subset X ⊂ V is called a base of V if X coincides with V .4. Every independent subset X ⊂ V can be extended to a base of V . Proposition 1. By Zorn lemma. . Proof. we deﬁne bases of linear subspaces. . The cardinality of bases of V is called the dimension of V and denoted by dim V . ak . consider Y1 . . Theorem 1. . respectively. Every maximal element of X is a base of V containing the subset X. Every one-to-one correspondence between bases of V and V can be uniquely extended to a linear isomorphism between these vector spaces. . Proof. . one of these subsets contains the others and is not independent. thus isomorphic vector spaces have the same dimension. Any two bases of a vector space have the same cardinality (possible inﬁnite). If Y is a linearly ordered subset of X (for any Y. the dimension of linear subspaces also is deﬁned. Let X be the set of all independent subsets containing X. xk ∈ Z and non-zero scalars a1 . Y ∈ Y we have Y ⊂ Y or Y ⊂ Y ) then the subset Z := Y ∈Y Y is independent. U ⊂ V there exists a base of V such that S and U are spanned by subsets of this base.2 in [Baer (1952)].2. . This set is non-empty (X ∈ X) and it is partially ordered by the inclusion relation. . Every independent subset X ⊂ V is a base of the linear subspace X . Every linear isomorphism maps independent subsets to independent subsets and bases to bases. Similarly. respectively. Yk ∈ Y containing x1 . suppose that a1 x1 + · · · + ak xk = 0 holds for non-zero vectors x1 . in particular. For any two linear subspaces S. Suppose that S ∩ U = 0. See Section II. Proof. . Indeed. . . . xk . Show that the subset BS ∪ BU is independent (this is trivial if one of the subspaces is . bases of V exist.3. We take any base X of S ∩ U and extend it to bases BS and BU of S and U . Since every linear subspace can be considered as a vector space. Proposition 1.

3. all xi and yj belong to X = BS ∩BU which contradicts the fact that X is independent. Let S be a linear subspace of V . Let S and U be linear subspaces of V . If these subspaces both are ﬁnite-dimensional then dim(S + U ) = dim S + dim U − dim(S ∩ U ). y1 . . dim U }. all complements of S have the same dimension called the codimension of S and denoted by codim S. By Proposition 1. Therefore. in this case. For any two subsets X. . Since BS and BU both are independent. . Y . . . We leave it as an exercise for the reader. . Y are linear subspaces then the sum X + Y coincides with the linear subspace X. a(x + S) := ax + S. Y ⊂ V we deﬁne X + Y := { x + y : x ∈ X. If X. the subset BS ∪ BU can be extended to a base of V . and we consider the case when S and U are not incident). In the case when S ∩ U = 0. xk ∈ BS .5. ym ∈ BU such that k m ai xi + i=1 k j=1 m bj yj = 0 and all scalars ai . The associated quotient vector space V /S is formed by all subsets x + S (we have x + S = y + S if x − y ∈ S) and the vector space operations on V /S are deﬁned as follows (x + S) + (y + S) := (x + y) + S. the proof is similar. Then ai xi = − i=1 j=1 bj yj ∈ S ∩ U. . If the statement fails then there exist distinct vectors x1 . This base is as required. . y ∈ Y }. we have dim(S + U ) = max{ dim S. . U is isomorphic to V /S (the mapping x → x + S is a linear isomorphism). In the case when at least one of these subspaces is inﬁnite-dimensional. Proposition 1. bj are non-zero. A linear subspace U is called a complement of S if S ∩ U = 0 and S + U = V .Linear Algebra and Projective Geometry 13 contained in the other.

The sum of two cardinalities α and β is the cardinality |X ∪ Y |. respectively. 1. . codim S } = dim V.14 Grassmannians of Classical Buildings This statement is a simple consequence of Proposition 1. In the case when dim V = α is inﬁnite. n − 1} we write Gk (V ) for the Grassmannian consisting of all k-dimensional linear subspaces of V .3 Dual vector space Let V be a left vector space over a division ring R. If dim V = n is ﬁnite then for every k ∈ {1. Recall that the cardinality of a set X is denoted by |X|. • β < α and G β (V ) consists of all linear subspaces of codimension β (the dimension of these linear subspaces is α). . where X and Y are disjoint sets of cardinalities α and β (respectively). in what follows we do not consider two trivial Grassmannians formed by 0 and V . this formula means that max{ dim S.1. . Remark 1. Linear mappings of V to R are called linear functionals.1. For every linear subspace S ⊂ V we have dim S + codim S = dim V . • Gα (V ) = G α (V ) consists of all linear subspaces whose dimension and codimension is α. We deﬁne the sum v + w of linear functionals v and w as follows (v + w)(x) := v(x) + w(x) ∀x∈V. in the inﬁnite-dimensional case. Denote by G(V ) the set of all linear subspaces of V . there are the following three types of Grassmannians: • β < α and Gβ (V ) consists of all linear subspaces of dimension β (the codimension of these linear subspaces is α). If α ≤ β and β is inﬁnite then α + β = β. . . and for any linear functional v and any scalar a ∈ R the linear functional va is deﬁned by the formula (va)(x) := v(x)a ∀ x ∈ V.4 and the following remark. The Grassmannians of V can be deﬁned as the orbits of the action of the group GL(V ) on G(V ).

If dim V = n < ∞ then for every x∗ ∈ V ∗ we have x∗ = ∗ ∗ n i=1 x∗ (xi )x∗ . If V is inﬁnite-dimensional then B ∗ is not a base of V ∗ . Indeed. There exist linearly independent vectors x.4. The associated left vector space over R∗ will be denoted by V ∗ and called the dual vector space. dim V < dim V ∗ (see Section II. If the codimension of Ker v is greater than 1 then consider S ∈ G2 (V ) contained in a complement of Ker v. .3 in [Baer (1952)] for the details). Let B = {xi }i∈I (|I| = dim V ) be a base of V . Proof. In this case. This subset is independent. satisfying i x∗ (xj ) = δij i (δij is Kroneker symbol). . As a consequence establish that codim(S1 ∩ · · · ∩ Sk ) ≤ k for any linear subspaces S1 . y ∈ S satisfying v(x) = v(y).Linear Algebra and Projective Geometry 15 The set of all linear functionals together with the operations deﬁned above is a right vector space other R. . The base B ∗ will be called dual to B. . Lemma 1. Show that codim(S ∩ U ) ≤ codim U + 1 for every linear subspace U ⊂ V and every linear subspace S ⊂ V of codimension 1. There is a one-toone correspondence between linear functionals and mappings of the base B to R. Denote by B ∗ the subset formed by the linear functionals x∗ . Let v ∈ V ∗ \ {0}. i ∈ I. i Thus B is a base of V and dim V = dim V ∗ .1. . Exercise 1. since every linear functional can be uniquely reconstructed from its restriction to B. The kernel of every non-zero linear functional is a linear subspace of codimension 1. Then v(x − y) = 0 which contradicts the fact that S intersects Ker v precisely in 0. Sk ⊂ V of codimension 1. a linear functional cannot be presented as a linear combination of elements from B ∗ if it takes non-zero values on an inﬁnite subset of B.

Let B ∗ be the subset of V ∗ which consists of all linear functionals x∗ . This bijection reverses inclusions: S ⊂ U ⇐⇒ U 0 ⊂ S 0 . denote them by xi1 . If V is ﬁnite-dimensional then this correspondence is a linear isomorphism between V and V ∗∗ . Then U 0 is the intersection of the kernels of v1 . In the inﬁnite-dimensional case.4. . If S and U are linear subspaces of V or linear subspaces of V ∗ then S ⊂ U implies that U 0 ⊂ S 0 . we have S 00 = S. we will identify the second dual vector space V ∗∗ with V . Since S coincides with the intersection of the kernels of x∗1 . The annihilator mapping S → S 0 is a bijection of G ﬁn (V ) to Gﬁn (V ∗ ). . . .6. Every vector x ∈ V deﬁnes the linear functional v → v(x) of V ∗ (an element of V ∗∗ ). n − 1}. x∗k form a base of S 0 i i and dim S 0 = k. Proof.16 Grassmannians of Classical Buildings For every subset X ⊂ V the linear subspace X 0 := { v ∈ V ∗ : v(x) = 0 ∀ x ∈ X } is known as the annihilator of X. . . Since dim U 00 = m (it was established above). In the ﬁnite-dimensional case. . Therefore. . . . if U is a ﬁnite-dimensional linear subspace of V ∗ then codim U 0 = dim U and U 00 = U . Denote by Gﬁn (V ) and G ﬁn (V ) the sets of all proper linear subspaces of V with ﬁnite dimension and ﬁnite codimension. x∗k . Consider a base B = {xi }i∈I of V such that S is spanned by a subset of B. by the second part of Exercise 1.6 gives the following. the inclusion U ⊂ U 00 implies that U 00 = U and m = k. . Similarly. vk } be a base of U . Proposition 1. respectively. . i i Let {v1 . . If S ⊂ V is a linear subspace of ﬁnite codimension then dim S 0 = codim S and S 00 = S. . . . vk and. x∗1 . . Similarly. . i ∈ I. .1. If dim V = n < ∞ then the annihilator mapping transfers Gk (V ) to Gn−k (V ∗ ) for every k ∈ {1. it sends G k (V ) to Gk (V ∗ ) for every natural k. If V is ﬁnitedimensional then Gﬁn (V ) and G ﬁn (V ) both coincide with G(V ). xik . Corollary 1. . The restriction of v ∈ S 0 to B can take nonzero values only on these vectors. i i There are precisely k distinct vectors from B which do not belong to S. Suppose that codim S = k < ∞. . . . . . satisfying x∗ (xj ) = δij . its codimension is equal to m ≤ k. Proposition 1. for every Y ⊂ V ∗ the linear subspace Y 0 := { y ∈ V : w(y) = 0 ∀ w ∈ Y } is called the annihilator of Y .

|L ∩ X| ≥ 2 }. L ) are isomorphic if there exists a bijection f : P → P such that f (L) = L . respectively. 1. w) between two vertices v and w is the smallest number i such that there is a path of length i connecting v and w. this line will be denoted by p q. Let Π = (P. Projective spaces over division rings will be considered in the second subsection.Linear Algebra and Projective Geometry 17 1. the pair (X. this bijection is called a collineaton of Π to Π . A linear space is a partial linear space. We say that the pair Π = (P. For every subset X ⊂ P we deﬁne LX := { L ∩ X : L ∈ L. L) be a partial linear space.2 Projective spaces In the ﬁrst subsection we introduce the point-line geometry language which will be exploited throughout the book. where any two points are collinear. L) and Π = (P . Elements of P and L will be called points and lines. In a connected graph the distance d(v.1 Linear and partial linear spaces Let P be a non-empty set and L be a set consisting of proper subsets of P . every path between v and w consisting of d(v. (2) for any two distinct collinear points p and q there is precisely one line containing them. A bijection of P to P is said to be a semicollineaton of Π to Π if it maps lines to subsets of lines. An injection of P to P sending lines to subsets of lines is called an embedding of Π in Π if distinct lines go to subsets of distinct lines.2. The collinearity graph of Π is the graph whose vertex set is P and whose edges are pairs of distinct collinear points. We say that partial linear spaces Π = (P. Our partial linear space is called connected if the collinearity graph is connected. w) edges is said to be a geodesic. In what follows we will always suppose that a partial linear space contains more than one line. Two or more points are said to be collinear if there is a line containing them. LX ) is called the restriction of the partial linear space Π to the subset X. L) is a partial linear space if the following axioms hold: (1) each line contains at least two points and for every point there is a line containing it. If Π is connected .

the points span a plane). d are distinct points and the lines ab and cd have a non-empty intersection then the lines ac and bd have a nonempty intersection. The minimal subspace containing a subset X ⊂ P (the intersection of all subspaces containing X) will be called the subspace spanned by X and denoted by X . A triple of mutually collinear points is said to be a triangle if these points are not collinear (in other words. A proper subspace of a partial linear space is called a hyperplane if it has a non-empty intersection with every line (it is clear that a hyperplane contains a line or intersects it precisely in a point). q ∈ S the line p q is contained in S. In the general case. We deﬁne the dimension of S as the smallest cardinality α such that S has a base of cardinality α + 1 (if the cardinality α is inﬁnite then α + 1 = α). the subspace spanned by a clique of the collinearity graph (a subset consisting of mutually collinear points) does not need to be singular. A projective space is a linear space where every line contains at least three points and the following axiom holds: if a. We say that S ⊂ P is a subspace of Π if for any two distinct collinear points p. We say that a subset X ⊂ P is independent if the subspace X is not spanned by a proper subspace of X. By this deﬁnition. b. moreover. The dimensions of the empty set and one-point subspaces are equal to −1 and 0 (respectively). Show that maximal singular subspaces exist and every singular subspace is contained in a certain maximal singular subspace. A subspace is said to be singular if any two points of this subspace are collinear (the empty set and one-point subspaces are singular).18 Grassmannians of Classical Buildings then we deﬁne the distance between two points of Π as the distance between the corresponding vertices of the collinearity graph. lines are 1-dimensional subspaces. c. Hint: use Zorn lemma. Recall that in a projective plane each line contains at least three points and any two distinct lines have a non-empty intersection (by the second axiom of partial linear spaces.5. The intersection of any collection of subspaces is a subspace. it is a singular subspace if all subspaces from the collection are singular. this intersection is a point). . every set of mutually non-collinear points is a subspace. Two-dimensional linear spaces and twodimensional singular subspaces of partial linear spaces are called planes. Exercise 1. Projective spaces can be also deﬁned as linear spaces where all planes are projective. An independent subset X ⊂ S is said to be a base of S if X = S. Let S be a subspace of Π (possible S = P ).

It follows from Exercise 1. P2 ∈ G1 (V ) there is the unique line G1 (P1 + P2 ) containing them.6. any two lines have a non-empty intersection and ΠV is a projective plane. Thus ΠV is a linear space. P4 are distinct points of ΠV and P1 + P2 has a non-zero intersection with P3 + P4 then all Pi are contained in a certain 3-dimensional linear subspace. y form a base of S then x . Denote by P ∗ (V ) the Grassmannian consisting of all linear subspaces of codimension 1 in V (in other words. but there exist non-Desarguesian projective planes. If P1 . For projective planes this fails. Remark 1. If the dimension of a projective space is not less than 3 (the dimension can be inﬁnite) then this projective space is isomorphic to the projective space associated with a certain vector space. Exercise 1. ΠV is a projective space of dimension dim V − 1. ΠV is a projective space. For any distinct P1 . Every line G1 (S) contains at least three points (if vectors x. Thus the dimension of G1 (S) is equal to dim S − 1. Denote by L1 (V ) the set of all such lines and consider the pair ΠV := (G1 (V ). in particular. See Chapter VII in [Baer (1952)]. Let X be a subset of G1 (V ).2. Show that S ⊂ G1 (V ) is a subspace of ΠV if and only if there exists a linear subspace S ⊂ V such that S = G1 (S). P2 .3. P ∗ (V ) coincides with Gn−1 (V ) or . P3 . Proof. in other words. Therefore. Hence X is a base of the subspace G1 (S). in particular. For every P ∈ X we choose a non-zero vector xP ∈ P and consider the set X := {xP }P ∈X .2. All projective spaces over division rings satisfy Desargues’ axiom.6 that X is an independent subset of ΠV if and only if X is an independent subset of V .Linear Algebra and Projective Geometry 19 1. In the case when dim V = 3. G1 (S) is called a line of G1 (V ). x + y are three distinct points on the line). y . For every S ∈ G2 (V ) the subset consisting of all 1-dimensional linear subspaces of S. P1 + P3 and P2 + P4 have a non-zero intersection.2 Projective spaces over division rings Let V be a left vector space over a division ring and dim V ≥ 3 (the dimension of V is not assumed to be ﬁnite). L1 (V )). Theorem 1. S ∈ G(V ) if and only if X is a base of the linear subspace S.

semilinear mappings form a more wide class. Let L∗ (V ) be the set of all such lines. The projective space Π∗ is called dual to ΠV .1 Deﬁnitions We say that a mapping l : V → V is semilinear if it is additive: l(x + y) = l(x) + l(y) ∀ x. In the case when R = R and σ is identity.3 Semilinear mappings Throughout the section we suppose that V and V are left vector spaces over division rings R and R . Hint: use Corollary 1. Show that for every ﬁnite-dimensional subspace S of Π∗ V there exists S ∈ G ﬁn (V ) such that S consists of all elements of P ∗ (V ) containing S.1. Semilinear isomorphisms map independent subsets to independent subsets and bases to bases. Semilinear . If l : V → V is a σ-linear isomorphism then the inverse mapping l−1 is a σ −1 -linear isomorphism of V to V .20 Grassmannians of Classical Buildings G 1 (V ) if V is n-dimensional or inﬁnite-dimensional. As for linear mappings.1) and the mapping l is said to be σ-linear. y ∈ V. A subset X ⊂ P ∗ (V ) is said to be a line if there is a linear subspace S ⊂ V of codimension 2 such that X consists of all elements of P ∗ (V ) containing S. 1. all vectors x ∈ V satisfying l(x) = 0 form a linear subspace called the kernel of l and denoted by Ker l. it is isomorphic to ΠV ∗ by the annihilator mapping (Corollary 1. V Exercise 1.3. and there exists a homomorphism σ : R → R such that l(ax) = σ(a)l(x) (1. 1. If l is non-zero (Ker l = V ) then there exists the unique homomorphism σ : R → R satisfying (1. L∗ (V )) V is a linear space.1) for all a ∈ R and x ∈ V . In the general case. respectively. respectively). An easy veriﬁcation shows that Π∗ := (P ∗ (V ).1). A semilinear mapping of V to V is called a semilinear isomorphism if it is bijective and the associated homomorphism of R to R is an isomorphism. moreover. Every semilinear mapping between real vector spaces is linear. we get a linear mapping.7.

x ∈ Ker l. xn . .3. . Example 1. If l : V → V is a σ-linear mapping then for every non-zero scalar a ∈ R the mapping al is σ -semilinear with σ (r) = aσ(r)a−1 ∀ r ∈ R. The following example shows that there exist semilinear bijections over non-surjective homomorphisms of division rings. This transformation is linear only in the case when the scalar a belongs to the center of the division ring R. . . and (al)1 coincides with (l)1 . Show that every semilinear bijection sending independent subsets to independent subsets is a semilinear isomorphism. this mapping is a collineation of ΠV to ΠV . The complexiﬁcation mapping of R2n to Cn : (x1 .4. it is a normal subgroup of ΓL(V ). . y1 .8.3. the associated homomorphism is the natural embedding of R in C. The homothetic transformation x → ax.Linear Algebra and Projective Geometry 21 isomorphisms of V to V exist if and only if dim V = dim V and R is isomorphic to R . The kernel of this homomorphism is GL(V ). . Exercise 1.2 Mappings of Grassmannians induced by semilinear mappings (l)1 : G1 (V ) \ G1 (Ker l) → G1 (V ) Every semilinear mapping l : V → V induces the mapping x → l(x) . 1. . It contains GL(V ) as a subgroup. The group of all semilinear automorphisms of V (semilinear isomorphisms of V to itself) is denoted by ΓL(V ). In the case when l is a semilinear isomorphism. The group of all homothetic transformations H(V ) is isomorphic to the multiplicative group of the division ring. Consider the homomorphism of ΓL(V ) to Aut(R) which sends every σ-linear automorphism of V to σ. . moreover. xn + yn i) is a semilinear bijection. Example 1. Hence GL(V ) is a normal subgroup of ΓL(V ) and the corresponding quotient group is isomorphic to Aut(R). . yn ) → (x1 + y1 i. is a semilinear automorphism of V (the associated automorphism of R is r → ara−1 ). x ∈ V .

for all vectors x ∈ V satisfying l(x) = 0 the scalar ax is a constant. Lemma 1. σ(ab)l(x) = l(abx) = σ(a)σ(b)l(x) which means that σ is a homomorphism. For a non-zero scalar a ∈ R consider the mapping x → l(ax). b ∈ R and any vector x ∈ V we have σ(a + b)l(x) = l(ax + bx) = l(ax) + l(bx) = (σ(a) + σ(b))l(x). If l(y) is a scalar multiple of l(x) then we take any vector z ∈ V such that l(x) and l(z) are linearly independent (by our hypothesis. Proof.7. Let l : V → V and s : V → V be additive mappings such that s(x) ∈ l(x) ∀ x ∈ V.22 Grassmannians of Classical Buildings Proposition 1.2. Let l : V → V be an additive mapping which satisﬁes l( x ) ⊂ l(x) ∀ x ∈ V. l(y) are linearly independent. For any scalars a. Lemma 1. Proof. Let l : V → V be a semilinear mapping such that l(V ) contains two linearly independent vectors. We set σ(0) := 0 and get a mapping σ : R → R . this is possible) and get ax = az = ay . Therefore. For every vector x ∈ V there exists a scalar ax ∈ R (possible zero) such that s(x) = ax l(x).2 implies the existence of a scalar σ(a) ∈ R such that l(ax) = σ(a)l(x) ∀ x ∈ V. This statement is a direct consequence of the following lemma.8 ([Zick (1983)]). If s : V → V is a semilinear mapping satisfying (l)1 = (s)1 then s = al for a certain scalar a ∈ R . Suppose that vectors l(x) and l(y) both are non-zero. Then ax l(x) + ay l(y) = s(x + y) = ax+y (l(x) + l(y)) and ax = ay = ax+y if l(x). If l(V ) contains two linearly independent vectors then l is semilinear. The equality l(x) = 0 implies that s(x) = 0 and we have s(x) = al(x) for every scalar a ∈ R . Proposition 1. If l(V ) contains two linearly independent vectors then s = al for a certain scalar a ∈ R . .

Every semilinear isomorphism of V to V induces bijections between the Grassmannians of the same indices. Since the restriction of (l)1 to the line G1 (S) is a bijection on the line G1 (l(S)). In other words. where σ : R → R is the homomorphism associated with l. l ¯ (l)1 = (s)1 and Proposition 1. the mapping l is a semilinear isomorphism. Let ¯ : G → G and s : G → G be the bijections induced by semilinear isomorphisms l : V → V and s : V → V . . If ¯ = s then s = al for a certain scalar a ∈ R . Let P ∈ G1 (V ). (l)1 is not surjective). hence it is an isomorphism of R to R . then l(V ) = V (otherwise. Then l is a semilinear isomorphism of V to V . Consider the set X consisting of all elements of G containing P . Since Ker l = 0. Then a → x + ay is a bijection of the division ring R to G1 (S) \ { y }. we have the following.9. Proposition 1. Proof. V V Proposition 1. y} be a base of S. s(P ) = S∈¯(X ) s S. in particular. The mapping (l)1 sends x + ay to l(x) + σ(a)l(y) . Let l : V → V be a semilinear mapping such that (l)1 is a collineation of ΠV to ΠV . we have l(P ) = s(P ) for every P ∈ G1 (V ). it deﬁnes a collineation of Π∗ to Π∗ . Since ¯ = s. Then P = S∈X S and l(P ) = S∈¯ ) l(X S. Conversely. l ¯ Proof.Linear Algebra and Projective Geometry 23 It was noted above that for every semilinear isomorphism l : V → V the mapping (l)1 is a collineation of ΠV to ΠV . The latter implies that l(V ) is a linear subspace of V . Let S ∈ G2 (V ) and {x. respectively.10. Suppose that G and G are Grassmannians of V and l ¯ V (respectively) with the same index.7 gives the claim. the homomorphism σ is surjective.

24

Grassmannians of Classical Buildings

A semilinear injection l : V → V is called a semilinear embedding of V in V if it maps independent subsets to independent subsets. Semilinear embeddings of V in V exist only in the case when dim V ≤ dim V and R isomorphic to a certain division subring of R . Example 1.5. The mapping (x1 , . . . , xn ) → (x1 + 0i, . . . , xn + 0i) is a semilinear embedding of Rn in Cn . Similarly, the natural embedding of Qn in Rn is a semilinear embedding. Now we consider the mappings of Grassmannians induced by semilinear embeddings. We restrict ourselves to the case when dim V = n is ﬁnite (the inﬁnite-dimensional case is similar). Let l : V → V be a semilinear embedding. Then for any linear subspace S ⊂ V we have dim l(S) = dim S, and the mapping (l)k : Gk (V ) → Gk (V ) S → l(S) is deﬁned for every k ∈ {1, . . . , n − 1}. Exercise 1.9. Show that the following assertions are fulﬁlled: (1) (2) (3) (4) (l)k is injective for every k ∈ {1, . . . , n − 1}, (l)1 is an embedding of ΠV in ΠV , (l)n−1 is an embedding of Π∗ in Π∗ if dim V = n, V V if s : V → V is a semilinear embedding such that (l)k = (s)k for a certain k ∈ {1, . . . , n − 1} then s is a scalar multiple of l.

Proposition 1.11. Let l : V → V be a semilinear embedding such that (l)k is bijective for a certain k ∈ {1, . . . , n − 1}. Then l is a semilinear isomorphism. Proof. In the case when k = 1, our bijection maps any triple of collinear points of ΠV to collinear points of ΠV and any triple of non-collinear points to non-collinear points. Hence it is a collineation of ΠV to ΠV . By Proposition 1.9, l is a semilinear isomorphism.

Linear Algebra and Projective Geometry

25

Suppose that k > 1. Let S ∈ Gk−1 (V ). We choose Ui ∈ Gk (V ), i = 1, 2, such that S = U1 ∩ U2 . Since (l)k is bijective, there exist Ui ∈ Gk (V ), i = 1, 2, satisfying l(Ui ) = Ui . Then U1 ∩ U2 belongs to Gk−1 (V ) (we leave the details for the reader) and l(U1 ∩ U2 ) = S . Thus (l)k−1 is surjective, and by the statement (1) of Exercise 1.9, it is bijective. Step by step, we establish that (l)1 is bijective. 1.3.3 Contragradient

Now suppose that V and V have the same ﬁnite dimension n. Recall that in this case V = V ∗∗ and V = V ∗∗ . For every x ∈ V and x∗ ∈ V ∗ we will write x∗ · x instead of x∗ (x). Let u : V → V be a σ-linear isomorphism. The adjoint mapping u∗ : V

∗

→ V∗

is deﬁned by the condition u∗ (x∗ ) · x = σ −1 (x∗ · u(x)) ∀ x ∈ V, x∗ ∈ V ∗ . This is a (σ −1 )-linear isomorphism and u∗∗ = u. The inverse mapping u := (u∗ )−1 : V ∗ → V ∗ ˇ is known as the contragradient of u. Since (u∗ )−1 = (u−1 )∗ , the contragradient of the contragradient coincides with u. An easy veriﬁcation shows that u(x∗ ) · u(x) = σ(x∗ · x) ∀ x ∈ V, x∗ ∈ V ∗ . ˇ Therefore, u transfers the annihilator of a linear subspace S ⊂ V to the ˇ annihilator of u(S) and we get ˇ u(U 0 )0 = u(U ) ∗ for every linear subspace U ⊂ V . So, the mapping Gk (V ∗ ) → Gk (V ∗ ) U → u(U 0 )0 coincides with (ˇ)k for every k ∈ {1, . . . , n − 1}. u Exercise 1.10. Show that the contragradient mapping u → u is an isoˇ morphism of ΓL(V ) to ΓL(V ∗ ) transferring GL(V ) to GL(V ∗ ).

26

Grassmannians of Classical Buildings

1.4 1.4.1

Fundamental Theorem of Projective Geometry Main theorem and corollaries

Let V and V be left vector spaces over division rings R and R , respectively. The dimensions of V and V are assumed to be not less than 3. Let also l : V → V be a semilinear injection. The mapping (l)1 : G1 (V ) → G1 (V ) P → l(P ) does not need to be injective, see Example 1.3. For any P, P1 , P2 ∈ G1 (V ) satisfying P ⊂ P1 + P2 (a triple of collinear points of ΠV ) we have l(P ) ⊂ l(P1 ) + l(P2 ) . If l(P1 ) = l(P2 ) then the line G1 (P1 + P2 ) goes to a point. In the case when l(P1 ) and l(P2 ) are distinct, the restriction of (l)1 to G1 (P1 + P2 ) is a bijection on a subset of the line G1 ( l(P1 + P2 ) ). The following generalized version of the Fundamental Theorem of Projective Geometry was proved in [Faure and Frolicher (1994)] and [Havlicek (1994)], independently. Theorem 1.4. Suppose that a mapping f : G1 (V ) → G1 (V ) satisﬁes the following conditions: for any P, P1 , P2 ∈ G1 (V ) P ⊂ P1 + P2 =⇒ f (P ) ⊂ f (P1 ) + f (P2 ) and f (G1 (V )) is not contained in a line. Then f is induced by a semilinear injection of V to V . Theorem 1.4 together with Proposition 1.9 give the classical version of the Fundamental Theorem of Projective Geometry. Corollary 1.2. Every collineation of ΠV to ΠV is induced by a semilinear isomorphism of V to V . Remark 1.3. Denote by PΓL(V ) the group formed by all collineations of the projective space ΠV to itself. Consider the homomorphism of ΓL(V ) to PΓL(V ) which sends l to (l)1 . The Fundamental Theorem of Projective Geometry (Corollary 1.2) states that this homomorphism is surjective. By Proposition 1.7, the kernel of the homomorphism is H(V ) (Example 1.4). Therefore, PΓL(V ) is isomorphic to the quotient group ΓL(V )/H(V ). In the case when V is ﬁnite-dimensional, the contragradient isomorphism of ΓL(V ) to ΓL(V ∗ ) (Exercise 1.10) sends H(V ) to H(V ∗ ); hence it induces an isomorphism between PΓL(V ) and PΓL(V ∗ ).

Linear Algebra and Projective Geometry

27

Corollary 1.3. If V and V have the same ﬁnite dimension then every semicollineation of ΠV to ΠV is a collineation. Proof. Every semicollineation of ΠV to ΠV satisﬁes the conditions of Theorem 1.4; hence it is induced by a semilinear mapping l : V → V . Our semicollineation maps every triple of collinear points to collinear points and it is suﬃcient to show that triples of non-collinear points go to non-collinear points. If this fails then there exist linearly independent vectors x, y, z ∈ V such that l(x), l(y), l(z) are linearly dependent. Let B be a base of V containing x, y, z. Since dim V = dim V is ﬁnite, l(B) spans a proper subspace of V which contradicts l(B) = l(V ) = V . Corollary 1.3 will be generalized in Section 3.5. Problem 1.1. Is there a semicollineation of ΠV to ΠV in the case when dim V > dim V ? It was noted in the previous section that semilinear isomorphisms of V to V induce collineations of Π∗ to Π∗ . We can show that there exist no V V other collineations of Π∗ to Π∗ only in the case when our vector spaces V V are ﬁnite-dimensional. Corollary 1.4. If V and V are ﬁnite-dimensional then every collineation of Π∗ to Π∗ is induced by a semilinear isomorphism of V to V . V V Proof. Let f be a collineation of Π∗ to Π∗ . It can be considered as a V V collineation of ΠV ∗ to ΠV ∗ . By the Fundamental Theorem of Projective Geometry, the latter collineation is induced by a semilinear isomorphism u : V ∗ → V ∗ and f (S) = u(S 0 )0 for every linear subspace S ⊂ V of codimension 1. By Subsection 1.3.3, we have f = (ˇ)1 in the ﬁnite-dimension case. u Exercise 1.11. Show that every semicollineation of Π∗ to Π∗ V V collineation if dim V = dim V < ∞. is a

Let X be a partially ordered set. The supremum of a subset Y ⊂ X is the least element of X which is greater than or equal to all elements of Y . Similarly, the inﬁmum of Y is the greatest element of X which is less than or equal to every element of Y . The supremum and inﬁmum of Y

U ∈ G(V ) S ⊂ U ⇐⇒ f (S) ⊂ f (U ). The classical Fundamental Theorem of Projective Geometry can be reformulated in the following form (see. U } = S ∩ U. If sup Y (inf Y ) exists then it is unique.28 Grassmannians of Classical Buildings are denoted by sup Y and inf Y . Since for all S. . respectively.4 Let f : G1 (V ) → G1 (V ) be a mapping which satisﬁes the conditions of Theorem 1. v3 ∈ V and w1 . w2 . U } = S + U and inf{S. Corollary 1. [Baer (1952)]). Hence the restriction of f to G1 (V ) is a collineation of ΠV to ΠV and there exists a semilinear isomorphism l : V → V such that f (P ) = l(P ) for all P ∈ G1 (V ). The set G(V ) is partially ordered by the inclusion relation. respectively. Proof. Lemma 1. Then for every S ∈ G(V ) f (G1 (S)) = G1 (f (S)) coincides with l(G1 (S)) = G1 (l(S)) and we get f (S) = l(S). for example. Let v1 . Every semilinear isomorphism of V to V induces an order preserving bijection of G(V ) to G(V ) (an isomorphism between the lattices).2 Proof of Theorem 1. The set X is called a lattice if every subset consisting of two elements has the supremum and inﬁmum. it is a lattice. 2.4. It is not diﬃcult to prove that f transfers G1 (V ) and G2 (V ) to G1 (V ) and G2 (V ). P1 .3. Let f : G(V ) → G(V ) be an order preserving bijection: for all S. 3. v2 . U ∈ G(V ) we have sup{S. Then f is induced by a semilinear isomorphism of V to V . w3 ∈ V be vectors satisfying the following conditions: • f ( vi ) = wi for i = 1. This means that the restriction of f to every line is a constant or a bijection to a subset of a line. 1.5. P2 ∈ G1 (V ) P ⊂ P1 + P2 =⇒ f (P ) ⊂ f (P1 ) + f (P2 ) and f (G1 (V )) is not contained in a line.4: for any P.

. x2 . there exist vectors x1 . • w1 . Q ∈ G1 (V ) and x ∈ P + Q with x ∈ Q. y1 ∈ f ( x1 ) ⊂ f ( x1 + x2 ) + f ( x2 ). Then there exists a unique vector y ∈ Q such that P = x + y .4. Since x1 is contained in x1 + x2 + x2 . f ( x2 ). Similarly. Proof. Lemma 1. w3 are linearly independent. Then f ( v1 + v2 + v3 ) is contained in the intersection of w1 + w2 + w3 and w1 + w3 + w2 . y2 . Then f ( v2 + v3 ) = w2 + w3 and f ( v1 + v2 + v3 ) = w1 + w2 + w3 . Lemma 1. w2 . y3 ∈ V satisfying the conditions f ( xi ) = yi and f ( xi + xj ) = yi + yj . It is clear that v1 . Trivial. f ( x3 ) are not contained in a line.Linear Algebra and Projective Geometry 29 • f ( v1 + v2 ) = w1 + w2 and f ( v1 + v3 ) = w1 + w3 . x3 ∈ V such that f ( x1 ).4 implies the existence of y2 ∈ f ( x2 ) such that f ( x1 + x2 ) = y1 + y2 . v2 . The latter intersection coincides with w1 + w2 + w3 and we get the second equality. Let us take any non-zero vector y1 ∈ f ( x1 ). Proof. v3 are linearly independent and v1 +v2 +v3 is the intersection of v1 + v2 + v3 and v1 + v3 + v2 . v2 + v3 = ( v2 + v3 ) ∩ ( v1 + v1 + v2 + v3 ) and f ( v2 + v3 ) is contained ( w2 + w3 ) ∩ ( w1 + w1 + w2 + w3 ) which gives the ﬁrst equality. Let P. Proof. Our ﬁrst step is to establish the existence of vectors y1 . By our hypothesis.

y3 are as required. Thus y2 = 0. We need to show that f ( x2 + x ) = y2 + l(x) . x. The restriction of f to the line associated with the linear subspace x + xi = x + xi + x is injective and we apply Lemma 1. we establish the existence of a non-zero vector y3 ∈ f ( x3 ) satisfying f ( x1 + x3 ) = y1 + y3 . Show that the mapping l : V → V is well-deﬁned (the deﬁnition of the vector l(x) does not depend on the choice of the number i ∈ {1. Similarly.3. 2. the vectors y1 . We choose i ∈ {1. there exists a vector l(x) ∈ f ( x ) such that f ( xi + x ) = yi + l(x) . Proof.3. In the case when f ( x ) ⊂ f ( x1 ) + f ( x2 ). x3 . we consider successively the triples x1 . Suppose that f ( x1 ) = f ( x ) = f ( x2 ).30 Grassmannians of Classical Buildings If y2 is zero then f ( x1 + x2 ) coincides with f ( x1 ) and the restriction of f to the line associated with the linear subspace x1 + x2 + x1 = x1 + x2 is constant. We set l(0) := 0 and assert that l : V → V is a semilinear mapping. So. Since f ( xi + x ) = f ( xi ). 3} such that f ( x ) = f ( xi ). we have l(x) = 0. If f ( x ) is not contained in f ( x1 ) + f ( x2 ) then the required equality follows from Lemma 1. Let x ∈ V \ {0}.4 to yi ∈ f ( xi ) ⊂ f ( xi + x ) + f ( x ). x2 . in this case. y2 . 3}). By Lemma 1. we get f ( x1 ) = f ( x2 ) which is impossible. There exists a non-zero vector l(x) ∈ f ( x ) satisfying f ( x1 + x ) = y1 + l(x) . x and x3 . . 2.

l(w + z) are linearly independent. l is semilinear and f = (l)1 . l(z) are linearly independent and l(v + w + z) = l(v + w) + l(z). By the ﬁrst equality. moreover.3 implies that f ( v + w ) = l(v) + l(w) and f ( xi + v + w ) = yi + l(v) + l(w) . For every non-zero vector x ∈ V and any non-zero scalar a ∈ R we have l(ax) = f ( ax ) = f ( x ) = l(x) . f ( x3 ) are not contained in a line). o .2) Since f ( w ) = f ( z ). the restriction of f to the line associated with the linear subspace w + z is injective and f ( v ) = f ( w ) = f ( w + z ). w ∈ V (the mapping l is additive).2) gives the claim. Therefore. We have f ( xi + v ) = yi + l(v) and f ( xi + w ) = yi + l(w) . Thus l(v + w) = l(v) + l(w) and yi + l(v + w) = yi + l(v) + l(w) . l(v + w) = a(l(v) + l(w)) for a certain scalar a ∈ R . Proof. This proof was taken from the paper [Faure (2002)] dedicated to A. Now suppose that l(w) is a scalar multiple of l(v). l(v+w).Linear Algebra and Projective Geometry 31 Our next step is the equality l(v + w) = l(v) + l(w) for all v. and Lemma 1. f ( x2 ). We choose i ∈ {1. In other words. We take any vector z ∈ V such that l(w). (1. Fr¨licher. l( x ) ⊂ l(x) and the mapping l : V → V satisﬁes the conditions of Proposition 1. This means that l(v). the second equality guarantees that a = 1. 3} such that f ( xi ) is not contained in f ( v ) + f ( w ) (this is possible. since f ( x1 ). l(z) are linearly independent. 2. First consider the case when l(v) and l(w) are linearly independent.4.8. Then f ( v ) = f ( w ) = f ( v + w ). hence l(v + w + z) = l(v) + l(w + z) = l(v) + l(w) + l(z) and (1. Remark 1. Then l(w+z) = l(w)+l(z).

The vector space F n is normed by ||x|| := (|x1 |2 + · · · + |xn |2 )1/2 .6. Note that Gcl (N ) is not a sublattice of the lattice G(N ) if N is inﬁnite-dimensional. • ||ax|| = |a| · ||x|| for every vector x ∈ N and every scalar a ∈ F . in particular. we have S U ∈ Gcl (N ). and ||x|| = 0 implies that x = 0. a matric space (hence it is also a topological space). . For linear subspaces S. x = (x1 . • ||x + y|| ≤ ||x|| + ||y|| for all x. The set Gcl (N ) is partially ordered by the inclusion relation. moreover. U } = S U and inf{S. Example 1.4. . . y ∈ N . The sum of two closed linear subspaces does not need to be closed. Example 1. in a natural way. So.3 Fundamental Theorem for normed spaces Let F be the ﬁeld of real or complex numbers. U ⊂ N we denote by S U the closure of S + U . U } = S ∩ U. the metric is deﬁned by d(x. Let also N be a normed space over F . Let p ∈ N and lp (F ) be the vector space formed by all sequences x = {xn }n∈N ⊂ F satisfying ∞ 1/p ||x||p := n=1 |xn |p < ∞. Gcl (N ) is a lattice. y ∈ N. || · ||p ) is a normed space. this means that N is a vector space (over F ) with a real-valued function x → ||x|| called a norm and satisfying the following conditions: • ||x|| ≥ 0 for all x ∈ N . .32 Grassmannians of Classical Buildings 1. since the closure of a linear subspace is a linear subspace. Our normed space is. (F. 1] }. | · |) is a 1-dimensional normed space. for all S.8. Denote by Gcl (N ) the set of all closed linear subspaces of N . The vector space CF ([0. Then (lp (F ).7. y) := ||x − y|| ∀ x. . 1]) consisting of all continuous functions g : [0. Example 1. Every ﬁnitedimensional linear subspace is closed. xn ). 1] → F is normed by ||g|| := sup{ |g(x)| : x ∈ [0. hence Gcl (N ) coincides with G(N ) if the dimension of N is ﬁnite. U ∈ Gcl (V ) sup{S.

f is induced by a linear or conjugate-linear homeomorphism of N to M . W. Theorem 1. . We will investigate order preserving bijections of Gcl (N ) to Gcl (M ) (isomorphisms between the lattices). U ∈ Gcl (N ). If N and M are ﬁnite-dimensional complex normed spaces then every semilinear isomorphism of N to M induces an order preserving bijection of G(N ) = Gcl (N ) to G(M ) = Gcl (M ). If N and M are real normed spaces then f is induced by a linear homeomorphism of N to M .4. every linear homeomorphism of N to M induces an order preserving bijection of Gcl (N ) to Gcl (M ). If a semilinear mapping l : N → M is continuous then the associated homomorphism σ : C → C is continuous.3. In the case when N and M are inﬁnite-dimensional complex normed spaces.5 was ﬁrst proved in [Mackey (1942)] for real normed spaces. Let N and M be normed spaces over F = R. For a linear mapping l : N → M (if M = F then l is a linear functional) the following conditions are equivalent: • l is continuous. Mackey. W. σ is identity or the complex conjugate mapping. C. A. Every conjugatelinear homeomorphism of N to M induces an order preserving bijection of Gcl (N ) to Gcl (M ). Theorem 1. Remark 1. In what follows a semilinear mapping between complex vector spaces will be called conjugate-linear if the associated homomorphism of C to itself is the complex conjugate mapping. 1. The complex version of this result was given in [Fillmore and Longstaﬀ (1984)]. Clearly. Now suppose that our normed spaces are complex. Recall that a bijection h between two topological spaces is called a homeomorphism if h and h−1 both are continuous. Longstaﬀ ).4 Proof of Theorem 1. By Exercise 1.5 We need some elementary facts concerning linear and conjugate-linear mappings. E. Fillmore.5 (G. P.5.Linear Algebra and Projective Geometry 33 Let M be other normed vector space over F . Let f : Gcl (N ) → Gcl (M ) be an order preserving bijection: S ⊂ U ⇐⇒ f (S) ⊂ f (U ) for all S.

5.5 in [Rudin (1973)]). As in the proof of Corollary 1. By the Fundamental Theorem of Projective Geometry. This means that the restriction of f to G1 (N ) is a collineation of ΠN to ΠM . Let X be a bounded subset of N . Since v is arbitrary . We need to show that l is a homeomorphism of N to M . Then f transfers G1 (N ) and G2 (N ) to G1 (M ) and G2 (M ). We ﬁx z ∈ N satisfying w(z) = 1. Then Ker v is a closed linear subspace of codimension 1 and S := l−1 (Ker v) is a closed linear subspace of codimension 1 in N . the same holds for conjugate-linear mappings. ||x|| ≤ 1 } is ﬁnite and called the norm of l. where y ∈ S. In the complex case.34 Grassmannians of Classical Buildings • l transfers bounded subsets to bounded subsets (a subset X is bounded if there exists a real non-negative number a such that ||x|| ≤ a for all x ∈ X). we establish that f (S) = l(S) ∀ S ∈ Gcl (N ). Let f : Gcl (N ) → Gcl (M ) be an order preserving bijection. For every closed linear subspace of codimension 1 there exists a continuous linear functional whose kernel coincides with this linear subspace (a consequence of the HahnBanach Theorem. Let v : M → R be a non-zero continuous linear functional. Then w(X) is a bounded subset of R and (1. The mapping l is linear. Then v(l(x)) = v(l(y)) + v(w(x)l(z)) = w(x)v(l(z)) (1. for conjugate-linear mappings the proof is similar.32) this statement is proved for linear mappings. Consider a continuous linear functional w : N → R such that Ker w = S. it is clear that ||l(x)|| ≤ ||l|| · ||x|| and ||al|| = |a| · ||l|| for all x ∈ N and a ∈ F . see Section 3. In [Rudin (1973)] (Section 1. Every x ∈ N can be presented in the form x = y + w(x)z. If l : N → M is a continuous linear or conjugate-linear mapping then the number ||l|| := sup{ ||l(x)|| : x ∈ N. it is induced by a certain semilinear isomorphism l : N → M .3) guarantees that the same holds for v(l(X)). Now we start to prove the theorem.3) (since l(y) ∈ Ker v). respectively. Real case.

if zn → z then (zn − z) → 0 and σ(zn − z) → 0 =⇒ σ(zn ) → σ(z). Proof.4) . In a normed space every weakly bounded subset is bounded (Section 3. This fact is a direct consequence of Kakutani–Mackey’s result which will be proved below (Proposition 1.6. it is suﬃcient to verify that σ is continuous in 0. l(X) is weakly bounded. Therefore. nzn → 0. Complex case. Suppose that σ is non-continuous in 0. Lemma 1. If for every sequence of complex numbers zn → 0 the sequence {σ(zn )}n∈N is bounded then σ is continuous.18 in [Rudin (1973)]). Lemma 1.5. The arguments given above work only in the case when σ is identity or the complex conjugate mapping. Indeed. Clearly. we need to show that l is linear or conjugate-linear. the set v(l(X)) is bounded for every continuous linear functional v : M → R. Then there exist a sequence zn → 0 and a real number a > 0 satisfying |σ(zn )| > a ∀ n ∈ N. In this case. Similarly. in other words. we establish that l−1 is continuous. Let σ be the automorphism of C associated with l. The normed spaces are assumed to be inﬁnitedimensional. in other words. |σ(nzn )| = n|σ(zn )| > na ∀n∈N and the sequence {σ(nzn )}n∈N is unbounded. hence it is continuous.3). Since the restriction of σ to N is identity. we have v(l(x)) = σ(w(x))v(l(z)) instead of (1. Thus l transfers bounded subsets to bounded subsets. Let σ be an automorphism of C. a contradiction. {zn }n∈N contains a subsequence {zn }n∈N such that |zn | < 1/n2 for every n ∈ N.12).Linear Algebra and Projective Geometry 35 taken. By the additivity of σ. If N is inﬁnite-dimensional then there exist an independent subset {xn }n∈N ⊂ N and a sequence of continuous linear functionals {vn }n∈N of N such that vi (xj ) = δij . (1.

v : X → F is a continuous linear functional. it can be extended to a continuous linear functional of N .4). For every vector x = v1 (x)x1 + · · · + vn (x)xn ∈ X we deﬁne v(x) := v1 (x)a1 + · · · + vn (x)an . Then vn+1 (xn+1 ) = 1 and vi (xn+1 ) = 0 for all i ≤ n.5 in [Rudin (1973)]). . vn . x2 ∈ N and two continuous linear functionals v1 . . xn and continuous linear functionals v1 . .7.36 Grassmannians of Classical Buildings Proof. Then |v(x)| ≤ |a1 | · ||v1 || · ||x|| + · · · + |an | · ||vn || · ||x|| ≤ a||x||(1/2 + · · · + 1/2n ) < a||x||. We deﬁne n xn+1 := xn+1 − i=1 vi (xn+1 )xi . Proof. Then there exists a continuous linear functional vn+1 such that • all xi belong to Ker vn+1 . . It is not diﬃcult to choose two linearly independent vectors x1 . So. Let X = {xn }n∈N and {vn }n∈N be as in the previous lemma. by the Hahn-Banach Theorem.4) holds for linearly independent vectors x1 . . . . If N is inﬁnite-dimensional then it contains an independent subset {xn }n∈N satisfying the following condition: for every bounded sequence of scalars {an }n∈N there exists a continuous linear functional v : N → F such that v(xn ) = an ∀ n ∈ N. Lemma 1. We can assume that ||vn || = 1/2n ∀n∈N (indeed. v2 of N satisfying (1. for every n ∈ N there exists a scalar bn such that ||bn vn || = 1/2n and we can take xn ∈ xn satisfying bn vn (xn ) = 1). . • vn+1 (xn+1 ) = 1 for a certain vector xn+1 which does not belong to x1 . xn (see Section 3. . . Let {an }n∈N be a bounded sequence of scalars and a = sup |ai |. . . Suppose that (1. .

Linear Algebra and Projective Geometry

37

Proposition 1.12 ([Kakutani and Mackey (1946)]). Suppose that N and M are inﬁnite-dimensional complex normed spaces. Let l : N → M be a semilinear isomorphism which sends closed linear subspaces of codimension 1 to closed linear subspaces. Then l is linear or conjugate-linear. Proof. Let σ be the automorphism of C associated with l. Let also {xn }n∈N be a subset of N with the property described in the previous lemma. Show that for every sequence of complex numbers an → 0 the sequence {σ(an )}n∈N is bounded. If {σ(an )}n∈N is unbounded then {an }n∈N contains a subsequence {bn }n∈N such that |σ(bn )| ≥ n||l(xn )|| ∀ n ∈ N. (1.5)

Let v : N → C be a continuous linear functional satisfying v(xn ) = bn for every n. We take z ∈ N such that v(z) = 1. Then xn = yn + bn z, where yn ∈ Ker v. We have l(xn )/σ(bn ) = l(yn /bn ) + l(z). By (1.5), l(xn )/σ(bn ) → 0 and l(−yn /bn ) → l(z). Hence l(z) belongs to the closure of l(Ker v). Ker v is a closed linear subspace of codimension 1 and the linear subspace l(Ker v) is, by our hypothesis, closed. Thus l(z) belongs to l(Ker v). Since l is bijective, we get z ∈ Ker v which contradicts v(z) = 1. By Lemma 1.5, σ is continuous. This means that it is identity or the complex conjugate mapping. 1.5 1.5.1 Reﬂexive forms and polarities Sesquilinear forms

Let V be a left vector space over a division ring R and σ : R → R be an anti-automorphism (an isomorphism of R to the opposite division ring R∗ ). We say that Ω: V ×V →R is a sesquilinear form over σ or simple a σ-form if Ω(x + y, z) = Ω(x, z) + Ω(y, z), Ω(z, x + y) = Ω(z, x) + Ω(z, y),

38

Grassmannians of Classical Buildings

Ω(ax, by) = aΩ(x, y)σ(b) for all vectors x, y, z ∈ V and all scalars a, b ∈ R; if σ is identity (this is possible only in the commutative case) then Ω is a usual bilinear form. For every y ∈ V the mapping x → Ω(x, y) is a linear functional of V and we get a σ-linear mapping of V to V ∗ . Conversely, every semilinear mapping l : V → V ∗ gives a sesquilinear form (x, y) → l(y) · x. Therefore, there is a one-to-one correspondence between sesquilinear forms on V and semilinear mappings of V to V ∗ . Our form Ω is said to be non-degenerate if the associated semilinear mapping of V to V ∗ is injective (if V is ﬁnite-dimensional then dim V = dim V ∗ and it is a semilinear isomorphism). This condition is equivalent to the fact that for every non-zero y ∈ V there exists x ∈ V such that Ω(x, y) = 0. 1.5.2 Reﬂexive forms

Suppose that dim V = n is ﬁnite. A sesquilinear form Ω : V × V → R is called reﬂexive if Ω(x, y) = 0 =⇒ Ω(y, x) = 0 for every x, y ∈ V . We give a few examples. Example 1.9. A non-zero sesquilinear form Ω : V × V → R is said to be symmetric or skew-symmetric if Ω(x, y) = Ω(y, x) or Ω(x, y) = −Ω(y, x) ∀ x, y ∈ V, respectively. In each of these cases, the associated anti-automorphism is identity (we leave the veriﬁcation to the reader); hence R is commutative. A sesquilinear form Θ : V × V → R is called alternating if Θ(x, x) = 0 ∀ x ∈ V. Every alternating form is skew-symmetric. Conversely, if the characteristic of R is not equal to 2 then every skew-symmetric form on V is alternating. In the case of characteristic 2, the classes of symmetric and skew-symmetric forms are coincident. Non-degenerate alternating forms exist only on evendimensional vector spaces over ﬁelds. ∀ x, y ∈ V

Linear Algebra and Projective Geometry

39

Example 1.10. Let σ be a non-identity anti-automorphism of R. We say that a non-zero sesquilinear form Ω : V × V → R is σ-Hermitian or skew σ-Hermitian if Ω(x, y) = σ(Ω(y, x)) or Ω(x, y) = −σ(Ω(y, x)) ∀ x, y ∈ V, respectively. An easy veriﬁcation shows that Ω is a σ-form and σ 2 = 1R in each of these cases. It is clear that the forms considered above are reﬂexive. Also note that for every reﬂexive σ-form Ω : V × V → R and every non-zero scalar a ∈ R the scalar multiple (x, y) → Ω(x, y)a is a reﬂexive σ -form with σ (b) = a−1 σ(b)a ∀ b ∈ R. ∀ x, y ∈ V

There is a complete description of all non-degenerate reﬂexive forms. Theorem 1.6 (G. Birkhoﬀ, J. von Neumann). If Ω : V × V → R is a non-degenerate reﬂexive form then one of the following possibilities is realized: • R is commutative and Ω is symmetric or alternating, • Ω is a scalar multiple of a Hermitian form. Proof. See, for example, Section 1.6 in [Dieudonn´ (1971)] or Chapter 7 e in [Taylor (1992)]. Let Ω : V ×V → R be a reﬂexive form. We say that two vectors x, y ∈ V are orthogonal and write x ⊥ y if Ω(x, y) = 0. The orthogonality relation is symmetric (by the reﬂexivity). If X and Y are subsets of V then X ⊥ Y means that every x ∈ X is orthogonal to all y ∈ Y ; in this case, we say that the subsets X and Y are orthogonal. The linear subspace consisting of all vectors orthogonal to a subset X ⊂ V is called the orthogonal complement of X and denoted by X ⊥ . A non-zero vector x ∈ V satisfying Ω(x, x) = 0 is said to be isotropic. A linear subspace S is called totally isotropic if the restriction of Ω to S is total zero, in other words, S is contained in S ⊥ . The Grassmannian formed by all k-dimensional totally isotropic subspaces will be denoted by Gk (Ω).

40

Grassmannians of Classical Buildings

From this moment we suppose that the form Ω is non-degenerated. Let u : V → V ∗ be the semilinear isomorphism associated with Ω. Then Ω(x, y) = u(y) · x ∀ x, y ∈ V.

Since the orthogonal relation is symmetric, we have S ⊥ = u(S)0 = u−1 (S 0 ) for every linear subspace S ⊂ V ; in particular, dim S ⊥ = codim S. This implies that dim S ⊥⊥ = dim S; since S ⊂ S ⊥⊥ , we get S ⊥⊥ = S. If S is totally isotropic then S ⊂ S ⊥ and dim S ≤ codim S. The mapping S → S ⊥ is a bijective transformation of G(V ) sending Gk (V ) to Gn−k (V ). This transformation is order reversing: S ⊂ U ⇐⇒ U ⊥ ⊂ S ⊥ . The square of the transformation is identity. Remark 1.6. Taking S 0 = U in (1.6) we obtain u(U 0 )0 = u−1 (U ) which implies that u(U ) = u−1 (U ). ˇ Therefore, (ˇ)k = (u−1 )k for all k. This means that u is a scalar multiple u ˇ ˇ of u−1 . Since u = (u∗ )−1 , we have u∗ = au for non-zero a ∈ R. 1.5.3 Polarities (1.6)

Let V be as in the previous section. A bijection π : G1 (V ) → Gn−1 (V ) is called a polarity if P ⊂ π(Q) ⇐⇒ Q ⊂ π(P ) ∀ P, Q ∈ G1 (V ). (1.7)

Example 1.11. Let be ⊥ be the orthogonal relation associated with a non-degenerate reﬂexive form. Then S ⊂ U ⊥ ⇐⇒ U ⊂ S ⊥ for any S, U ∈ G(V ) (since the orthogonal relation is symmetric). This means that the restriction of the transformation S → S ⊥ to G1 (V ) is a polarity.

8) to the set X . P2 be distinct V elements of G1 (V ) and X be the set formed by all (n− 1)-dimensional linear subspaces containing P1 + P2 . Conversely. For every polarity π : G1 (V ) → Gn−1 (V ) there exists a non-degenerate reﬂexive form Ω such that the restriction of the transformation S → S ⊥ (where ⊥ is the orthogonal relation associated with Ω) to G1 (V ) coincides with π. we have the following.7) that the sesquilinear form deﬁned by u is reﬂexive. Proposition 1. Clearly. P ∈ G1 (V ) belongs to the line G1 (P1 + P2 ) if and only if it is contained in every element of X . The Fundamental Theorem of Projective Geometry guarantees the existence of a semilinear isomorphism u : V → V ∗ such that π(P ) = u(P )0 for every P ∈ G1 (V ). Proof.8). our polarity transfers G1 (π(P1 ) ∩ π(P2 )) (1. Thus P ⊂ P1 + P2 ⇐⇒ π(P1 ) ∩ π(P2 ) ⊂ π(P ) and π is a collineation.7). . By (1. Show that π is a collineation of ΠV to Π∗ .7). the latter is equivalent to the fact that π(P ) contains every element of the set (1.Linear Algebra and Projective Geometry 41 So. Let P1 . By (1.13. It follows from (1. every non-degenerate reﬂexive form deﬁnes a polarity.

This page intentionally left blank .

Chapter 2 Buildings and Grassmannians This chapter is a survey of basic facts concerning buildings and related topics. respectively.1 2.1 Simplicial complexes Deﬁnition and examples Let X be a non-empty set and ∆ be a set consisting of ﬁnite subsets of X (we do not require that X is ﬁnite). 2. Also we show that the same method cannot be applied to apartments preserving mappings of building Grassmannians. If Σ ⊂ ∆ then Σ is called a subcomplex of ∆. Then ∆ is called a simplicial complex. Buildings of classical groups will be considered as examples. Grassmannians of ﬁnite-dimensional vector spaces and Grassmannians formed by totally isotropic subspaces of non-degenerate reﬂexive forms are special cases of this general construction. Let Y be a subset of X and Σ be a simplicial complex whose vertex set is Y . In the second part of the chapter. Suppose that the following conditions hold: (1) every one-element subset belongs to ∆. (2) if A ∈ ∆ then every subset of A belongs to ∆. 43 . elements of X and ∆ are said to be vertices and simplices. we prove Abramenko–Van Maldeghem’s result on apartments preserving mappings of the chamber sets of buildings (maximal simplices of a building are called chambers). We introduce the concepts of building Grassmannians and the associated Grassmann spaces.1.

Example 2. Also note the following remarkable property: for X. respectively. n}. . Then X+ and X− are elements of J+ and J− . respectively. A subset X ⊂ J is said to be singular if j ∈ X =⇒ −j ∈ X. . −n}. . For every k ∈ {1. A subset consisting of pairwise incident elements is said to be a ﬂag. Let us consider the 2n-element set J := {1. . . .2. Y ∈ Jn is an element of J− if and only if n − |X− ∩ Y | is even. Example 2. moreover. Now deﬁne X+ := {1. . −n} and consider the sets J+ := { X ∈ Jn : n − |X+ ∩ X| is even }. The ﬂag complex associated with the natural incidence relation on J will be denoted by Cn . We write J for the set of all singular subsets. . n + 1} with the natural incidence relation (two subsets are incident if one of them is contained in the other). . . . . . . . Consider the set of all proper subsets of {1. Y ∈ Jn the number n − |X ∩ Y | . . . . A bijective morphism f : X → X is an isomorphism if f (∆) = ∆ . X− := {1. Now we give a few examples. The ﬂag complex associated with the relation ∗ is the simplicial complex whose vertex set is X and whose simplices are ﬁnite ﬂags. . . Suppose that X is a set with an incidence relation ∗ (this is a symmetric and reﬂexive binary relation). . . n} we denote by Jk the set of all singular subset consisting of k elements (every one-element subset is singular and we identify J1 with J).1. The associated ﬂag complex is denoted by An . . . A mapping f : X → X is said to be a morphism of ∆ to ∆ if f (∆) ⊂ ∆ (the image of every simplex is a simplex). Isomorphisms of a simplicial complex to itself are called automorphisms. −1. . . J− := Jn \ J+ = { Y ∈ Jn : n − |X+ ∩ Y | is odd }. n} it contains i or −i. .44 Grassmannians of Classical Buildings Let ∆ and ∆ be simplicial complexes whose vertex sets are X and X . Every maximal singular subset consists of n elements and for every i ∈ {1. . n − 1. n.

if X ∈ J+ and Y ∈ J− then X ∗ Y means that X ∩ Y belongs to Jn−1 (for example. Then every (n − 1)-dimensional totally isotropic subspace is contained in precisely two maximal totally isotropic subspaces.Buildings and Grassmannians 45 is odd if and only if one of these subsets belongs to J+ and the other is an element of J− . X+ and X− are incident). We deﬁne an incidence relation ∗ on J ∗ as follows: in the case when X ∈ Jk . k ≤ n − 2. 2. The associated ﬂag complex is called the oriﬂamme complex of Ω and denoted by Orif(Ω). and Y ∈ J ∗ . The rank of a such simplicial complex is the number of vertices in maximal simplices. and U ∈ G ∗ (Ω) then S ∗ U means that S and U are incident in the usual sense. We will restrict ourselves to simplicial complexes satisfying the following condition: every simplex is contained in a certain maximal simplex and all maximal simplices have the same cardinality. The associated ﬂag complex is called the oriﬂamme complex and denoted by Dn . we deﬁne the oriﬂamme incidence relation ∗ on G ∗ (Ω). Example 2. The dimension of maximal totally isotropic subspaces is assumed to be equal to n. Now suppose that V is a (2n)-dimensional vector space over a ﬁeld whose characteristic is not equal to 2 and Ω is a non-degenerate symmetric form on V .3.1. k ≤ n − 2. As in the previous example. . Let J ∗ := J \ Jn−1 . for S ∈ G+ (Ω) and U ∈ G− (Ω) we write S ∗ U if their intersection is (n − 1)-dimensional. Note that for two maximal totally isotropic subspaces S and U the codimension of S ∩ U in S is odd if and only if one of these subspaces belongs to G+ (Ω) and the other is an element of G− (Ω). If S ∈ Gk (Ω). If Ω is a non-degenerate reﬂexive form on V then we write ∆(Ω) for the subcomplex of ∆(V ) consisting of all ﬂags formed by totally isotropic subspaces of Ω. The action of the group SO(Ω) (the group formed by all linear automorphisms of determinant 1 which preserve the form Ω) on the Grassmannian Gn (Ω) is not transitive: there are precisely two orbits which will be denoted by G+ (Ω) and G− (Ω).3 satisfy this condition. Let V be a ﬁnite-dimension vector space over a division ring and ∆(V ) be the ﬂag complex deﬁned by the natural incidence relation on the set of all proper linear subspaces of V . Let G ∗ (Ω) be the set of all totaly isotropic subspaces whose dimension is not equal to n − 1. The ﬂag complexes considered in Examples 2.2 and 2. we write X ∗ Y if X and Y are incident in the usual sense (one of these subsets is contained in the other).

If f and g are isomorphisms between thin chamber complexes ∆ and ∆ such that the restrictions of f and g to a certain chamber are coincident then f = g. Then f (C) and g(C) are chambers containing the panel f (C0 ∩ C) = g(C0 ∩ C) and distinct from the chamber f (C0 ) = g(C0 ). the same holds for every chamber of ∆.46 Grassmannians of Classical Buildings 2. We say that ∆ is labeled by a set S if there exists a mapping α : X → S which satisﬁes the following conditions: • the restriction of α to every maximal simplex is bijective. and we deﬁne the distance d(C.1. If ∆ is a chamber complex labeled by a set S then any two labelings of ∆ by S are coincident up to a permutation on S. In a chamber complex of rank n. A chamber complex is said to be thick if every panel is contained in at least 3 chambers. By connectedness. Proof. In this case. every simplex consisting of n − 1 vertices is called a panel. Lemma 2.1. Let ∆ be a simplicial complex whose vertex set is X. If α is a labeling of ∆ by S then for any permutation t on the set S the mapping tα also is a labeling of ∆.2. We say that ∆ is a chamber complex if the graph Γch (∆) is connected. B ∈ ∆ the inclusion A ⊂ B implies that α(A) ⊂ α(B). C ) between two chambers C and C as the distance between the corresponding vertices in the graph Γch (∆).2 Chamber complexes Let ∆ be a simplicial complex of rank n. and it is called thin in the case when every panel is contained in precisely 2 chambers. in this case. Two maximal simplices C and C are said to be adjacent if |C ∩ C | = n − 1. Suppose that the restrictions of f and g to a chamber C0 are coincident and consider a chamber C adjacent with C0 . . Thus f (C) = g(C) (since our chamber complexes are thin) and the restrictions of f and g to C are coincident. Let Ch(∆) be the set of all maximal simplices of ∆ and Γch (∆) be the graph whose vertex set is Ch(∆) and whose edges are pairs of adjacent elements. Lemma 2. maximal simplices of ∆ are said to be chambers. the mapping α is called a labeling of ∆. • for all A.

Then |S| = n. The equality β(x) = tα(x) (recall that x ∈ C0 ) implies that β(y) = tα(y). Show that β = tα. We take any chamber C adjacent with C0 .s (∆) coincides with a certain Grassmannian Gα. the restrictions of β and tα to C are coincident. Let ∆ be a labeled chamber complex. If C1 and C2 are adjacent chambers C1 \ C2 = {x} and C2 \ C1 = {y} for certain vertices x.Buildings and Grassmannians 47 Proof. then Let α be a labeling of ∆. Let α and β be labelings of ∆ by S. This decomposition does not depend on a labeling of the complex ∆. Then α(x). C2 and it does not contain the panel C1 ∩C2 then x. Since α and β are bijective. 2. If 3 distinct chambers of ∆ are mutually adjacent then their intersection is a panel. by Lemma 2. Since the latter subset consists of unique element. since α(x) = α(y). If α is another labeling of ∆ by the set S then. Therefore. y ∈ C3 . there exists a permutation t on S such that β = tα . there exists a permutation t on the set S such that α = tα and every Grassmannian Gα . Suppose that C0 \ C = {x} and C \ C0 = {y}. C1 ∩ C2 is contained in C3 . Therefore. If a chamber C3 is adjacent with C1 . Proof. the vertex set of ∆ can be decomposed in n disjoint subsets which will be called Grassmannians. . The same holds for the labeling β.s (∆) := α−1 (s). y and α(x) = α(y) (see the proof of the previous lemma). For every s ∈ S we deﬁne the (α. Therefore.2.3 Grassmannians and Grassmann spaces Let ∆ be a chamber complex of rank n and α be a labeling of ∆ by a set S.3.s (∆). we get α(x) = α(y). Then C \ C0 and C0 \ C are one-point sets. The connectedness of Γch (∆) gives the claim.1. Denote by α and β their restrictions to a chamber C0 . s)-Grassmannian Gα. this contradicts the fact that the restriction of α to C3 is bijective. α(y) ∈ S \ α(C ∩ C0 ). Lemma 2.

Two distinct vertices a. The Grassmann graph ΓG is connected. The term “Grassmannian” is motivated by the following example. Proposition 2. b ∈ G are said to be adjacent if there exist adjacent chambers A and B such that a ∈ A and b ∈ B. every line consists of all elements of Gk (V ) “lying between” two incident linear subspaces of dimension k − 1 and k + 1. In the case when 1 < k < n − 1. Let a.1. If k = 1. Let V be an n-dimensional vector space over a division ring. moreover. The Grassmann graph ΓG associated with the Grassmannian G is the graph whose vertex set is G and whose edges are pairs of adjacent vertices. The set consisting of all x ∈ G such that P ∪ {x} is a chamber will be called the line of G associated with (deﬁned by) the panel P . any two distinct elements of this line are adjacent. Clearly. b ∈ G. two elements V of Gk (V ) are adjacent if their intersection is (k − 1)-dimensional (this is equivalent to the fact that the sum of these linear subspaces is (k + 1)dimensional). Example 2. L) is known as the Grassmann space or the shadow space corresponding to the Grassmannian G.4. Every path in Γch (∆) connecting A and B induces a path in ΓG which connects a and b. Then ∆ is labeled and f transfers Grassmannians to Grassmannians (if α is a labeling of ∆ then αf −1 is a labeling of ∆ ). The Grassmannians of ∆(V ) coincide with the Grassmannians of the vector space V . Denote by L the set of all such lines. This is a chamber complex (it will be shown later). We take chambers A and B such that a ∈ A and b ∈ B. Every simplex of ∆ intersects G in at most one vertex. The partial linear space G := (G. . Proof. f induces collineations between the Grassmann spaces. The collinearity relation of G coincides with the adjacency relation and the associated collinearity graph is the Grassmann graph ΓG .48 Grassmannians of Classical Buildings Let G be a Grassmannian of ∆. Let f be an isomorphism of ∆ to a chamber complex ∆ . Consider a panel P which does not intersect G. n−1 then any two distinct elements of Gk (V ) are adjacent and the associated Grassmann spaces are the projective space ΠV and the dual projective space Π∗ . this is equivalent to the existence of a panel P such that P ∪ {a} and P ∪ {b} are chambers. The ﬂag complex ∆(V ) is labeled by the dimension function and its rank is equal to n − 1.

9 in [Bourbaki (1968)]. Subsection IV. S) is called ﬁnite if W is ﬁnite (the group W does not need to be ﬁnite if S is ﬁnite). S). We suppose that m(s. If s.2. s ) = 2 holds if and only if the involutions s and s commute. Remark 2. Note that two generators are not connected by an edge if they commute. S ) are called isomorphic if there exists an isomorphism h : W → W such that h(S) = S . Then each (Wi . s ∈ S then we write m(s.Buildings and Grassmannians 49 2. A Coxeter system is called irreducible if its diagram is connected. Let {Si }i∈I be the connected components of the diagram associated with a Coxeter system (W. S) and (W . s ) − 2 edges (in the case when m(s. (s. s ) ∈ I can be extended to a homomorphism of W to G. s ) = ∞. The diagram associated with the Coxeter system (W. We have m(s. s ) is ﬁnite. S) is the graph whose vertex set is S and any two distinct s. A Coxeter system (W.s ) = 1 ∀ (s.1. The pair (W.s ) = 1. s) and the equality m(s. s) = 1 for every s ∈ S and m(s. Denote by I the set of all pairs (s.1. S) is called a Coxeter system if the group W has the presentation S : (ss )m(s. s ) for the order of the element ss . Si ) is an irreducible Coxeter system and W is the “bounded” direct product of all Wi .2 2. s ) ≥ 2 if s = s . s ) = m(s . s ∈ S are connected by m(s. s ) ∈ I . Coxeter systems (W. s ) ∈ S × S such that m(s.1 in [Bourbaki (1968)]): An Bn = C n . We deﬁne Wi := Si for every i ∈ I. Coxeter systems are isomorphic if and only if they have the same diagram. this means that every mapping h of the set S to a group G satisfying (h(s)h(s ))m(s. There are precisely 11 types of ﬁnite irreducible Coxeter systems (Subsection VI.1 Coxeter systems and Coxeter complexes Coxeter systems Let W be a group generated by a set S.4. we draw one edge labeled by ∞). The ﬁrst condition guarantees that every element of S is an involution.

We say that the pair (W. sn with each si ∈ S. sn is called reduced if n = l(w). . . . An expression w = s1 . S) satisﬁes the exchange condition if for every reduced expression . The length l(w) = lS (w) of an element w ∈ W (with respect to the generating set S) is the smallest number n such that w has an expression w = s1 . i = 3 or i ≥ 5 Let W be a group generated by a set S. .50 Grassmannians of Classical Buildings Dn E6 E7 E8 F4 4 G2 3 H3 3 H4 i l2 (i).

6 in [Bourbaki (1968)]. .1 and Exercise 2. .2. . S) satisﬁes the exchange condition if and only if it is a Coxeter system. For every subset X ⊂ S we deﬁne WX := X . Using Theorem 2. apply the exchange condition to s1 w and s1 w. . .8 in [Bourbaki (1968)]. Let (W. Y ⊂ S. sn = s1 . See Subsection IV. sn }. .1. . . X) is a Coxeter system. Hint: establish that Ssw ⊂ {s} ∪ Sw for every s ∈ S and prove that Sww ⊂ Sw ∪ Sw induction by l(w). . Show that if s1 . .Buildings and Grassmannians 51 w = s1 . . . . sn .1. . Let W be a group and S ⊂ W be a set of involutions which generates W . Exercise 2. . sn (the symbolˆmeans that the corresponding term is omitted). . . . Proof. . . . sn and every s ∈ S satisfying l(sw) ≤ n there exists i ∈ {1. Theorem 2. sn } = {s1 . Then for every subset X ⊂ S the pair (WX . n} such that ˆ sw = s1 . The following properties of a Coxeter system (W. See Subsection IV. . . sn and l(w) = n then {s1 . Proof. . . si . sn ∈ S. for every w ∈ W there exists a subset Sw ⊂ S such that every reduced expression of w is formed by all elements of Sw . S) can be drawn from the exchange condition. and WX ∩ WY = WX∩Y for all subsets X. Exercise 2.2 we establish the following. . By Exercise 2. S) be a Coxeter system.1.2. Then (W.1. Hint: the proof is induction by l(w). . . . w = s1 . s1 .1. Show that WX consists of all w ∈ W satisfying Sw ⊂ X. . Theorem 2. .

.2 Coxeter complexes W k := S \ {sk } . . . . We deﬁne k = 1. . .2. . . . Proposition 2. . n}. We identify this simplex with the special subset X1 ∩ · · · ∩ Xm = wW J . . sjm } . . The Coxeter complex Σ(W. Then wC0 is the maximal simplex corresponding to w ∈ W . W n . . .2. Xm form a simplex if there exists w ∈ W such that X1 = wW j1 . . . . . wW n } will be identiﬁed with the element w ∈ W . Exercise 2. S). C0 can be connected with every element of Ch(∆) and the graph Γch (∆) is connected. Then. . . . n} W J := S \ {sj1 . S = {s1 . C1 = si1 C0 . . . . . C2 = si1 si2 C0 . . . S) is ﬁnite if the Coxeter system is ﬁnite. . Proof. The complex Σ(W. . .3. . . . . Denote by C0 the maximal simplex formed by W 1 . and for every subset J = {j1 . . . special subsets X1 . . Thus wW i = W i implies that w ∈ W i . sik then C0 . S) is the simplicial complex whose vertex set consists of all special subsets wW k with w ∈ W and k ∈ {1. and we have w = 1 if the latter equality holds for all i. . .2. . Xm = wW jm . . . . sik C0 = wC0 is a path in Γch (∆). . . . W J = W j1 ∩ · · · ∩ W jm . jm }. Then every maximal simplex {wW 1 . Let (W. by Theorem 2. .52 Grassmannians of Classical Buildings 2. jm } ⊂ {1. Ck = si1 . . . . n. If w = si1 . Show that wW i = w W j =⇒ i = j and ww −1 ∈ W i . . sn } be a Coxeter system. Therefore. . where J = {j1 . S) is a thin chamber complex. . The Coxeter complex Σ(W.

. it consists of all special subsets wW k . .5 (Type An . The Grassmannians of Σ(W.2. Gelfand and White (2003)]. n ≥ 1). n + 1}. S) preserving the canonical labeling. i = 1. If a chamber wC0 contains Pk then w∈ i=k W i = {1.3).1 to prove the surjectivity.5). n. . . We write Aut0 (Σ(W. Then wPk is contained only in the chambers wC0 and wsk C0 . Let also S be the set formed by the transpositions si = (i. any other labeling of Σ(W. there exist automorphisms of Σ(W. w ∈ W . S) the Grassmannian containing W k . S).2. .3 Three examples In this subsection we consider the Coxeter systems of types An . Grassmannians of Coxeter complexes are closely related with the concept of so-called Coxeter matroids [Borovik. . Cn . Let W = Sn+1 . S) coincide with the orbits of the action of the group W = Aut0 (Σ(W. Thus Pk is contained only in the chambers C0 and sk C0 . In the general case. 2.Buildings and Grassmannians 53 Let Pk be the panel formed by ˆ W 1. Exercise 2. W k. . Hint: use Lemma 2.4. sk } (by Exercise 2. . where Sn+1 is the symmetric group consisting of all permutations of the set I := {1. Then (W. W n (the vertex W k is omitted). S) is a Coxeter system whose diagram is . . Cn . . S) by S is the composition of the canonical labeling with a permutation on S). . . S)). Dn (respectively). . . S)) for the group of all automorphisms of Σ(W. Example 2. i + 1). Denote by Gk (W. . . S) which sends every special subset X to wX. Dn and show that the associated Coxeter complexes are An . S) which do not belong to this group (see Example 2. For every w ∈ W denote by lw the automorphism of Σ(W. The mapping wW k → sk is the canonical labeling of Σ(W. S)) on the vertex set of Σ(W. Show that w → lw is an isomorphism of W to the group Aut0 (Σ(W. S) by the set S (by Lemma 2.

. n. Elements of this group preserve J (the set of all singular subsets. This is an isomorphism of Σ(W. All symplectic permutations form the group denoted by Spn . The automorphism of the complex An sending every subset X ⊂ I to I \ X induces an automorphism of Σ(W. see Example 2. S) to the complex Cn (Example 2. . k} and the mapping wW k → {w(1). S) to Jk (the set formed by all singular subsets consisting of k elements).6 (Type Bn = Cn . w(k)} is well-deﬁned. This isomorphism sends the Grassmannian Gk (W. . . . As in the previous example. n − 1. . w(2). Suppose that W = Spn and S is the set of generators considered above. . . and the transposition sn = (n. We say that a permutation s on the set J := {1. . . . . . . . k} and the mapping wW k → {w(1). . . . n} the subgroup W k is the stabilizer of the subset {1. . Example 2. . −(i + 1)). S) to the set of all kelement subsets of I. . . . −1. . S) which does not preserve the canonical labeling. i + 1)(−i. −n). . . w(k)} is an isomorphism of Σ(W. it transfers the Grassmannian Gk (W. Then (W. S) is a Coxeter system with the diagram S1 S2 S3 Sn-1 Sn i = 1.1). . S) to the complex An (Example 2. n ≥ 2). . w(2).54 Grassmannians of Classical Buildings S1 S2 S3 Sn-1 Sn For every number k ∈ {1. The group Spn is generated by the “symplectic transpositions” si = (i. the subgroup W k is the stabilizer of the subset {1. . .2). −n} is symplectic if s(−j) = −s(j) ∀ j ∈ J.2). . . .

(2) for any two simplices of ∆ there is an element of A containing both of them. . w(n)} is an isomorphism of Σ(W. n). . The reader can show that J+ and J− (Example 2. For example. . . . . In this example we suppose that W is the subgroup SOn ⊂ Spn generated by the set S consisting of s1 . W does not contain permutations transferring X+ := {1. S) to the sets J1 . −n)(−(n − 1). The simplicial complex ∆ is a building if the following additional axiom holds: . w(k)} if k ≤ n − 2. Then (W. sn−1 (deﬁned in the previous example) and the “symplectic transposition” sn = (n − 1.1 Buildings Deﬁnition and elementary properties Let ∆ be a simplicial complex and A be a set of subcomplexes of ∆ satisfying the following axioms: (1) every element of A is isomorphic to a Coxeter complex. . J− . . n − 1. . wW n → {w(1). . . . . .3. . w(2). . .2).2).7 (Type Dn . . S) is a Coxeter system whose diagram is S1 S2 S3 Sn-2 Sn-1 Sn The action of the group W on the set Jn is not transitive. .2) are the orbits of the action of W on Jn and the mapping wW k → {w(1). .3 2. wW n−1 → {w(1). . Jn−2 . Proposition 2. −n} (since every symplectic permutation sending X+ to X− is the composition of (n.Buildings and Grassmannians 55 Example 2. . Then ∆ is a chamber complex (since every Coxeter complex is a chamber complex. . S) to the oriﬂamme complex Dn (Example 2. −n) and an element of W n ). J+ . This isomorphism transfers the Grassmannians of Σ(W. 2. . . . n} to X− := {1. . w(n − 1). n ≥ 4). w(n − 1). . w(−n)}.

C is a labeling of ∆.C and called the retraction of ∆ on Σ centered at C. The axiom (3) guarantees that any two apartments Σ. Σ ∈ AC . Σ ∈ A are isomorphic: we take chambers C ∈ Σ and C ∈ Σ and consider an apartment Σ containing them. Σ ∈ A then for any simplices A. Therefore. By the axiom (3). Remark 2. ﬁt together to give a surjective morphism of ∆ to Σ.8. We write AC for the set of all apartments containing a chamber C. by the axiom (3). Σ ∈ AC there is a unique isomorphism fΣΣ of Σ to Σ whose restriction to C is identity. (2). (3) and (1). elements of A are said to be apartments and A is called a system of apartments. Similar arguments can be used to prove the following property (we leave the details for the reader): (4) if Σ. all isomorphisms fΣ Σ . Every building is a labeled chamber complex. Since the union of all apartments from AC coincides with ∆ (by the axiom (2)). Let ∆ be a building and A be a system of apartments for ∆. by Section 4.3. for any two apartments Σ. Σ ∈ AC (because fΣ Σ fΣ Σ is an isomorphism of Σ to Σ preserving C pointvise).1. we have the following. Example 2. We have fΣ Σ = fΣ Σ fΣ Σ for all Σ.56 Grassmannians of Classical Buildings (3) for all Σ. Let Σ ∈ A and C.3 in [Garrett (1997)]. . Every Coxeter complex can be considered as a building with unique apartment. Then Σ contains every geodesic of the graph Γch (∆) connecting C and C . This morphism will be denoted by ρ Σ. C be chambers of Σ. Now we ﬁx an apartment Σ ∈ AC . B ∈ Σ ∩ Σ there exists an isomorphism of Σ to Σ preserving A and B pointwise.1 in [Brown (1989)]. (4) are equivalent. (2). If α is a labeling of Σ (apartments are labeled by the axiom (1)) then αρ Σ. Section IV. such isomorphism is unique).2. in this case. the axiom (1) can be drawn from the axioms (2) and (3). The axiom systems (1). Proposition 2. Proposition 2. Σ is isomorphic to both Σ and Σ . Σ . Moreover. Σ ∈ A such that Σ ∩ Σ contains a chamber of ∆ there exists an isomorphism of Σ to Σ preserving pointwise all simplices contained in Σ ∩ Σ (by Lemma 2.4.

the building ∆ can admit diﬀerent systems of apartments. . . We require that the following two technical conditions hold: . . there exists a unique (up to isomorphism) Coxeter system (W. Section IV. . ρ(Ck ) = C is a part in Γch (∆) (if A. S) such that all apartments of ∆ are isomorphic to the Coxeter complex Σ(W. The chamber ρ(Ci ) ∈ Σ contains the panel ρ(Ci−1 ∩ Ci ) = Ci−1 ∩ Ci .Buildings and Grassmannians 57 Proof. Let S be a set of generators of the quotient group W := N/(B ∩ N ). Section IV.2 Buildings and Tits systems Suppose that G is a group spanned by proper subgroups B and N . . we have ρ(Ci ) = ρ(C ) = C . . moreover.3 in [Brown (1989)]. . . This path contains at most k − 1 edges which contradicts the assumption that d(C. thus wB and C(w) := BwB are well-deﬁned. . hence it coincides with Ci−1 or C . Ck = C is a geodesic of Γch (∆) which is not contained in Σ. S). In general.3. If w ∈ W then for any two elements g1 and g2 belonging to the class w we −1 have g2 g1 ∈ B ∩ N and g1 B = g2 B. but the union of any collection of apartment systems is again an apartment system. ρ(Ci+1 ). . Ci−1 = ρ(Ci ). C1 . B are adjacent chambers then ρ(A). Suppose that C = C0 . Moreover.C . This implies the existence of a largest system of apartments. C ) = k. Since there is an apartment containing both C and Ci . B ∩ N is a normal subgroup of N . 2. C1 . Consider the retraction ρ = ρ Σ. .4 in [Brown (1989)]. We choose an index i such that Ci−1 ∈ Σ and Ci ∈ Σ. . ρ(B) are adjacent or coincident). So ρ(Ci ) = Ci−1 and C = C0 . There is a unique chamber C ∈ Σ distinct from Ci−1 and containing the panel Ci−1 ∩ Ci .

all P J := BW J B. . . N form a BN-pair of the group G. . . . jm }. . . . For every number k ∈ {1. S) whose vertex set consists of all special subsets gP k with g ∈ G and k ∈ {1. Xm form a simplex if there exists g ∈ G such that X1 = gP j1 . . . We identify this simplex with the special subset X1 ∩ · · · ∩ Xm = gP J . (6) C(s)C(w) = C(sw) if l(sw) ≥ l(w). (2) sBs−1 is not contained in B for every s ∈ S. . . Show that gP i = g P j =⇒ i = j and gg −1 ∈ P i . . . Consider the simplicial complex ∆ = ∆(G. S) is said to be a Tits system and W is called the Weyl group of this Tits system. Now assume that S = {s1 . . . Then (G.5. (5) for every subgroup W = S with S ⊂ S BW B := w∈W C(w) is a subgroup of G. . n}.2 in [Brown (1989)]: (3) every s ∈ S is an involution and (W. and C(s)C(w) = C(w) ∪ C(sw) if l(sw) ≤ l(w).2. N. Section V. are subgroup of G and P J = P j1 ∩ · · · ∩ P jm if J = {j1 . . . n} we deﬁne W k and W J as in Subsection 2. . These subgroups are called special. B. . Section V. special subsets X1 . Xm = gP jm . N. . . sn }. . . (4) C(w) ∩ C(w ) = ∅ if w = w . n}. By the property (5). in particular. . .58 Grassmannians of Classical Buildings (1) C(s)C(w) ⊂ C(w) ∪ C(sw) for all s ∈ S and w ∈ W . . . . also we say that the subgroup B. J ⊂ {1. and BW B coincides with B if W = {1}. Conversely. B.2. . n} and every subset J ⊂ {1. . every subgroup of G containing B is special. Every special subgroup contains B. we have BW B = G. Exercise 2. . S) is a Coxeter system.2 in [Brown (1989)]. The conditions (1) and (2) give the following remarkable properties.

3.3 in [Brown (1989)]. .2). . jm }. Lemma 2. J ⊂ {1. . The Grassmannians of ∆ coincide with the orbits of the left action of the group G on the vertex set of ∆. it does not need to be injective and surjective (Remark 2. given by gP → gP g −1 (it easy follows from the fact that there are no two distinct special subgroups which are conjugate). There is a one-to-one correspondence between simplices of ∆ and parabolic subgroups. and orthogonal groups.Buildings and Grassmannians 59 where J = {j1 . . it consists of all special subsets gP k . Section V. Then {gΣ}g∈G is an apartment system and ∆ is a building. it will be called the fundamental apartment of ∆.4). Denote by Gk (∆) the Grassmannian containing P k . Let G = GL(V ). . . The mapping wW k → wP k is an isomorphism of the Coxeter complex Σ(W. A subgroup of G is called parabolic if it is conjugate to a special subgroup. .9 (The general linear group). Denote by Aut0 (∆) the group formed by all automorphisms of ∆ preserving the canonical labeling (there exist automorphisms of ∆ which do not belong to this group.4). The mapping gP k → sk is the canonical labeling of the building ∆ by the set S (any other labeling of ∆ by S is the composition of the canonical labeling with a permutation on S. Then maximal simplices will be identiﬁed with special subsets gB. are called apartments of ∆. S) to the fundamental apartment Σ. We take any base .3. symplectic. g ∈ G. n}. . .3 Classical examples In this subsection we consider buildings associated with general linear. . The subcomplexes gΣ := lg (Σ). If w ∈ W then for any two elements g1 and g2 of the class w we have g1 P J = g2 P J for every J ⊂ {1. The mapping g → lg is a homomorphism of G to Aut0 (∆). For every g ∈ G denote by lg the automorphism of ∆ transferring every special subset X to gX. see Remark 2. . Remark 2. Let Σ be the subcomplex of ∆ formed by all special subsets wP J with w ∈ W . 2. n} and denote this special subset by wP J . where V is an (n+1)-dimensional vector space over a division ring. . Example 2.

. Let Ω be a non-degenerate alternating form on a (2n)-dimensional vector space V over a ﬁeld. . Then . xk ) is the required isomorphism.10 (The symplectic group). Then (G. Section V.3). .4.5. . for every number k ∈ {1. The associated building ∆ can be identiﬁed with the ﬂag complex ∆(V ) (Example 2. xn+1 . B. . By the Fundamental Theorem of Projective Geometry (Corollary 1. This homomorphism is surjective only in the case when every automorphism of the associated division ring is inner. xk and the mapping gPk → g( x1 . . Remark 2. .4. xn+1 } of this vector space and denote by B the stabilizer of the maximal ﬂag x1 ⊂ x1 . . this subcomplex is isomorphic to the complex An (Example 2.5 in [Brown (1989)]. if l is a σ-linear automorphism of V such that (l)1 = (s)1 for a certain linear automorphism s : V → V then l is a scalar multiple of s which is equivalent to the fact that the automorphism σ is inner). if V is a real vector space or a vector space over the division ring of real quaternion numbers (indeed. This is a normal subgroup of N and the corresponding quotient group is isomorphic to the symmetric group Sn+1 . Example 2. . . . The Grassmannian Gk (∆) (the Grassmannian containing P k ) is Gk (V ). Then B ∩ N consists of all linear automorphisms preserving each xi . . . For every base X ⊂ V denote by ΣX the subcomplex of ∆(V ) consisting of all ﬂags formed by the linear subspaces spanned by subsets of X. .5).1). n} the subgroup P k is the stabilizer of the linear subspace x1 . Indeed. Hence the homomorphism g → lg of GL(V ) to the group Aut0 (∆(V )) is not injective. S) is a Tits system. The associated Grassmann spaces were described in Example 2. Then the fundamental apartment Σ is identiﬁed with ΣX0 and gΣ = Σg(X0 ) for every g ∈ G. For every homothetic transformation the associated automorphism of ∆(V ) is identity.60 Grassmannians of Classical Buildings X0 = {x1 . . Let S be the set of generators of Sn+1 considered in Example 2. . for example. . . Also note that semilinear isomorphisms of V to V ∗ (if they exist) induce automorphisms of ∆(V ) which do not preserve the canonical labeling. N. . . every element of Aut0 (∆(V )) is induced by a semilinear automorphism of V . . . . x2 ⊂ · · · ⊂ x1 . xn in the group G. Let also N ⊂ G be the stabilizer of the projective base x1 . .

xn . . Let V be a (2n)-dimensional vector space over a ﬁeld whose characteristic is not equal to 2. . As in the previous example. . xn . S) is a Tits system. elements S. U ∈ Gk (Ω) are adjacent if S ⊥ U and S ∩ U belongs to Gk−1 (Ω) (this is equivalent to the fact that S + U is an element of Gk+1 (Ω)). y) = 0. We ﬁx an Ω-base X0 = {x1 . Let B ⊂ G be the stabilizer of the ﬂag x1 ⊂ x1 . . n}. Then B ∩ N is a normal subgroup of N and the corresponding quotient group is isomorphic to Spn . N. These apartments are isomorphic to the complex Cn (Example 2. . .2). xn (all linear subspaces in this ﬂag are totally isotropic) and N ⊂ G be the stabilizer of the projective base x1 . We say that a base X ⊂ V is an Ω-base if it consists of isotropic vectors and for each x ∈ X there is precisely one y ∈ X satisfying Ω(x. .11 (The orthogonal group). Such bases form a suﬃciently wide class. every line in Gn (Ω) consists of all maximal totally isotropic subspaces containing a certain element of Gn−1 (Ω). . .6 in [Brown (1989)]. we show that the associated building can be identiﬁed with the subcomplex ∆(Ω) ⊂ ∆(V ) (Example 2.3). these bases form a suﬃciently wide class. yn } satisfying Ω(xi .Buildings and Grassmannians 61 every 1-dimensional linear subspace is totally isotropic and the dimension of maximal totally isotropic subspaces is equal to n. A base X ⊂ V is said to be an Ω-base if for each x ∈ X there is precisely one y ∈ X satisfying Ω(x. In the case when k < n. and every apartment is the subcomplex ∆(Ω) ∩ ΣX . . Two elements of Gn (Ω) are adjacent if their intersection is (n − 1)-dimensional. Suppose that G = Sp(Ω) (the group formed by all linear automorphisms of V preserving the form Ω). Section V. y1 . As in the previous example. . . . . yn . The fundamental apartment is the intersection of ∆(Ω) with ΣX0 . The Grassmannians of this building are Gk (Ω). If S is the set of generators of Spn considered in Example 2. y) = 0. The dimension of maximal totally isotropic subspaces is assumed to be equal to n. k ∈ {1. B. where X is a certain Ω-base. the lines are deﬁned as for Grassmannians of ﬁnite-dimensional vector spaces (by pairs of incident totally isotropic subspaces of dimension k − 1 and k + 1). . x2 ⊂ · · · ⊂ x1 . Let also Ω be a non-degenerate symmetric form deﬁned on V .6 then (G. y1 . yi ) = 0 for all i. Example 2. . .

k ≤ n− 2. Let N ⊂ G be the stabilizer of the projective base x1 . the diameter of the graph Γch (∆) is ﬁnite and equal to the diameter of the graph Γch (Σ). Two elements of Gδ (Ω) are adjacent if their intersection belongs to Gn−2 (Ω). The adjacency relation on Gk (Ω). and Gδ (Ω). where Σ is an apartment of ∆. x2n . Let G be a reductive algebraic group over a ﬁeld and B be a Borel subgroup of G containing a maximal torus T (we refer [Humphreys (1975)] for the precise deﬁnitions). . B.3. Then B and N form a BN-pair of G. See Section IV. this subcomplex is isomorphic to the oriﬂamme complex Dn (Example 2. by Proposition 2. where X is an Ω-base.4. Proof.3). In this case. Suppose that G = SO(Ω) and denote by B the stabilizer of this maximal simplex in G. . . Example 2. C ∈ Σ. 2. . The associated Tits system and building are described in [Tits (1974)] (Chapter 5).12.5. is deﬁned as in the previous example. −}. Let us ﬁx an Ω-base X0 = {x1 .7 in [Brown (1989)]. In a spherical building every apartment Σ is the union of all geodesics connecting two opposite chambers C. δ ∈ {+. Proposition 2. Denote by N the normalizer of T in G.2). . Section V. N. We say that two chambers of a spherical building are opposite if the distance between them is maximal (is equal to the diameter). every line consists of all elements of Gδ (Ω) containing a certain (n − 2)-dimensional totally isotropic subspace. The term “spherical” is motivated by the following. . Standard arguments show that the associated building can be identiﬁed with the oriﬂamme complex Orif(Ω). x2n } and a maximal simplex (a maximal oriﬂamme ﬂag) in the subcomplex Orif X0 . Then B ∩ N is a normal subgroup of N and the corresponding quotient group is isomorphic to SOn . .62 Grassmannians of Classical Buildings Consider the oriﬂamme complex Orif(Ω) (Example 2.5 in [Brown (1989)]. . k ≤ n − 2. Every apartment of this building is Orif X . The Grassmannians of this building are Gk (Ω). . Every Ω-base X deﬁnes the subcomplex Orif X ⊂ Orif(Ω) consisting of all “oriﬂamme ﬂags” formed by the totally isotropic subspaces spanned by subsets of X.4 Spherical buildings A building ∆ is called spherical if the associated Coxeter system is ﬁnite. Then (G. S) is a Tits system.7. Let S be the set of generators of the group SOn from Example 2.

there exists no thick building of type Hi . If the diagram of the Coxeter system associated with ∆ does not contain edges labeled by ∞ then every isomorphism of Γch (∆) to Γch (∆ ) is induced by an isomorphism of ∆ to ∆ .3). A building is called irreducible if the associated Coxeter system is irreducible. A be their systems of apartments. Proof. In particular. Bn = Cn . If ∆ is a thick spherical building then for any two distinct chambers C1 . Conversely. A are apartment systems for ∆ then A = A . Every spherical building admits a unique system of apartments. 2. we have f (A) = A .2.5 Mappings of the chamber sets Let ∆ and ∆ be buildings. 8 (the building type is the type of the associated Coxter system). Every isomorphism between spherical buildings is apartments preserving (sends apartments to apartments). 7. 4.1. where V is an (n + 1)-dimensional vector space over a division ring (this fact is a reformulation of Theorem 1.3. if ∆ is a spherical building and A. C2 ∈ ∆ the following conditions are equivalent: . n ≥ 3. p. Corollary 2. All thick buildings of types Cn and Dn can be obtained from so-called polar spaces (this construction will be considered in Chapter 4). Isomorphisms of ∆ to ∆ (if they exist) induce isomorphisms between the graphs Γch (∆) and Γch (∆ ).1. Theorem 2. and four exceptional types F4 and Ei . Let ∆. ∆ be spherical buildings and A.51. There are three classical types An . We restrict ourselves to buildings of classical types only. Theorem 2. Proof.Buildings and Grassmannians 63 Corollary 2.4 ([Abramenko and Van Maldeghem (2000)]). Every thick building of type An . Dn . Tits). is isomorphic to the ﬂag complex ∆(V ). i = 3. Irreducible thick spherical buildings of rank ≥ 3 were classiﬁed in [Tits (1974)].3 (J. By Corollary 2. If f is an isomorphism of ∆ to ∆ then f (A) is an apartment system for ∆ . See [Tits (1974)]. we have the following. i = 6. in other words.

Conversely. The intersection of an apartment of ∆ with the chamber set Ch(∆) will be called an apartment in Ch(∆).3. C is a geodesic in Γch (∆). Assume that X = {C.64 Grassmannians of Classical Buildings • C1 and C2 are adjacent. this chamber is not adjacent with C (otherwise. since C. We take any geodesic γ of Γch (∆) connecting C with a certain element of X \ {C. γ is contained in X . by Lemma 2.4. We want to show that apartments preserving mappings are adjacency preserving. C . If γ contains C then the ﬁrst possibility is realized. Every apartments preserving mapping is injective (since for any two chambers there is an apartment containing them). our proof will be based on the following characterization of the adjacency relation in terms of apartments. C2 such that no chamber is opposite to a unique member of the set {C. C.4.4. C2 }. Let ∆ be a thick building. We say that a mapping f : Ch(∆) → Ch(∆ ) is apartments preserving if its restriction to every apartment of Ch(∆) is a bijection on a certain apartment of Ch(∆ ). C ∈ ∆ are adjacent if and only if the intersection of all apartments of Ch(∆) containing them coincides with {C. . C contains the panel C ∩ C which is impossible. • there exists a chamber C = C1 . In particular. Proof. C and C are adjacent. C or C . C1 . C . C }. C ∈ X and apartments are thin chamber complexes). If the intersection of all apartments of Ch(∆) containing both C and C coincides with {C. C }. Distinct chambers C. First we establish the existence of C ∈ X \ {C. by Proposition 2. C }. If C does not belong to γ then γ contains a chamber C (C = C ) adjacent with C. C } then. if ∆ also is a thick spherical building then every bijection of Ch(∆) to Ch(∆ ) preserving the relation to be opposite (two chambers of ∆ are opposite if and only if their images are opposite) is an isomorphism of Γch (∆) to Γch (∆ ). Lemma 2. suppose that C and C are adjacent and denote by X the intersection of all apartments of Ch(∆) containing C. C . by Proposition 2. C } such that C.

f (Cn ) ∈ f (A) are adjacent with f (C). Since the rank of ∆ also is equal to n. C . the second is similar. but C does not belong to this apartment which contradicts C ∈ X . Suppose that ∆ and ∆ are thick buildings of the same rank.5 is a modiﬁcation of a result obtained in [Abramenko and Van Maldeghem (2008)]. Then (it follows from Lemma 2. since for any two elements of a building Grassmannian there is an apartment containing them.Buildings and Grassmannians 65 Consider the ﬁrst case (C. . Proof. . It was noted above that f is injective.4 guarantees that f transfers adjacent chambers to adjacent chambers. but require that the buildings are of the same rank. Theorem 2. Apartments preserving mappings are injective. C is a geodesic). C ˆ every apartment containing C and C contains C . ˆ Let C and C be chambers of ∆ and A be an apartment of Ch(∆) containing them. Lemma 2. the apartment f (A) does not contain other chambers adjacent with f (C). ˆ C and C are adjacent if and only if their images are adjacent. Their images f (C1 ). Suppose that the rank of ∆ is equal to n. Therefore. every apartments preserving bijection of Ch(∆) to Ch(∆ ) is an isomorphism of Γch (∆) to Γch (∆ ). The chambers C and C are not adjacent ˆ is a geodesic in Γch (∆). . Hence C. We say that a mapping between two building Grassmannians is apartments preserving if its restriction to every apartment is a bijection to an apartment. We do not assume that the mapping is bijective. Cn . Theorem 2. denote them by C1 . 2. Let f : Ch(∆) → Ch(∆ ) be an apartments preserving mapping.4 Mappings of Grassmannians The intersections of a building Grassmannian with apartments of the associated building are called apartments of this Grassmannian. Hence if ˆ ˆ f (C) and f (C) are adjacent then we have C = Ci for a certain i. . Then every apartments preserving mapping of Ch(∆) to Ch(∆ ) preserves the adjacency relation (two chambers of ∆ are adjacent if and only if their images are adjacent).5. Then A contains precisely n distinct chambers adjacent with C. in particular. .5. . . . ˆ We choose a chamber C containing the panel C ∩ C and distinct from C ˆ and C (our building is thick). .3). Remark 2. C .

V Let V and V be vector spaces of the same ﬁnite dimension n ≥ 3.1. We suppose that G and G both are not projective spaces. Let also G and G be Grassmannians of ∆ and ∆ . Note that such kind theorems were ﬁrst proved in [Chow (1949)] for Grassmannians of ﬁnite-dimensional vector spaces and Grassmannians formed by maximal totally isotropic subspaces of non-degenerate reﬂexive forms. D} and n = 4). this is a base of the projective space ΠV or the dual projective space Π∗ . Denote by G and G the associated Grassmann spaces (Subsection 2. D}. respectively.3) is a collineation of G to G . We say that a subset of an apartment is inexact if there exist other apartments containing this subset. we show that every isomorphism of ΓG to ΓG (Subsection 2. Let ∆ and ∆ be buildings of the same classical type Xn . respectively. every apartments preserving bijection of Gn−1 (V ) to Gn−1 (V ) is a collineation of Π∗ to Π∗ .5 does not work for building Grassmannians. We describe all collineations of G to G and show that they can be extended to isomorphisms of ∆ to ∆ (for projective spaces the latter statement is trivial). n − 1. Every apartment of the Grassmannian Gk (V ) consists of all kdimensional linear subspaces spanned by subsets of a certain base of V . In this case. Our method is based on combinatorial properties of so-called maximal inexact subsets. these results easy follow from elementary properties of maximal singular subspaces of Grassmann spaces. C. In Chapters 3 and V V 4 we establish such kind results for other Grassmannians associated with buildings of classical types. since for any a. For projective spaces this fails (since any two points of a projective space are collinear and every bijection between projective spaces gives an isomorphism of their collinearity graphs). Let V be an n-dimensional vector space over a division ring. b coincides with {a. b}. In the cases when k = 1. One of our main results is the description of all apartments preserving mappings of G to G . X ∈ {A.66 Grassmannians of Classical Buildings Example 2.3). In almost all cases (except the case when X ∈ {C. Certain Grassmannians have .1.13. The idea used to prove Theorem 2. b ∈ G the intersection of all apartments of G containing both a. we establish that every apartments preserving bijection of G to G is a collineation of G to G (hence it can be extended to an isomorphism of ∆ to ∆ ). in particular. similarly. It is not diﬃcult to prove that every apartments preserving bijection of G1 (V ) to G1 (V ) (a bijection which sends bases of ΠV to bases of ΠV ) is a collineation of ΠV to ΠV .

It will be shown later that all Grassmann spaces associated with buildings of classical types are gamma spaces. or coincides with L. p ∈ X such that q is on the line pp ). 2. Exercise 2. For every natural i ≥ 2 we deﬁne [X]i := [[X]i−1 ]1 . Using this characterization we show that every apartments preserving mapping of G to G is adjacency preserving (two elements of G are adjacent if and only if their images are adjacent). Let Π = (P. . Then [X]i is contained in [X]j if i ≤ j. It is clear that X ⊂ [X]1 . L) be a gamma space. or consists of a single point.Buildings and Grassmannians 67 maximal inexact subsets of precisely one type. We characterize the adjacency relation in terms of maximal inexact subsets. This axiom implies that for any point p ∈ P and any line L ∈ L the set of all points on L collinear with p is empty. We write [X]1 for the set formed by all points belonging to the lines joining points of X (a point q belongs to [X]1 if there exist distinct points p. and. Show that [X]1 is a clique of the collinearity graph.6. by Exercise 2.6. Following [Cooperstein. Consider a subset X ⊂ P consisting of mutually collinear points (a clique in the collinearity graph of Π) and suppose that |X| ≥ 2.5 Appendix: Gamma spaces A partial linear space Π = (P. each [X]i is a clique of the collinearity graph. In linear spaces the axiom holds trivially. Kasikova and Shult (2005)] we present results of “opposite nature” (characterizations of apartments in terms of the adjacency relation) for certain Grassmannians associated with buildings of classical types. but there exist other Grassmannians with maximal inexact subsets of two diﬀerent types. Now we establish some elementary properties of gamma spaces. L) is called a gamma space if it satisﬁes the following axiom: (Γ) if a point is collinear with two distinct points of a line then it is collinear with all points of this line.

68 Grassmannians of Classical Buildings Proposition 2. Denote by S the union of all [X]i . Since S is a clique of the collinearity graph. where m = max{i.6. Let S be a maximal singular subspace containing X (Exercise 1. we have X ⊂ S ⊂ X which gives the claim. Proof. If p and q are distinct points of S then p ∈ [X]i and q ∈ [X]j for certain i. Since X contains every [X]i . By Proposition 2. . Proposition 2. j. If X is a clique in the collinearity graph of a certain gamma space and |X| ≥ 2 then ∞ X = i=1 [X]i is a singular subspace. If X is a clique in the collinearity graph of a gamma space Π = (P. Thus S is a singular subspace.5).3. Corollary 2. the line joining these points is contained in [X]m+1 . It is clear that p and q both belong to [X]m . every maximal clique X of the collinearity graph is contained in the singular subspace X . In a gamma space the class of maximal singular subspaces coincides with the class of maximal cliques of the collinearity graph. Every maximal singular subspace U is contained in a certain maximal clique Y of the collinearity graph. j}. we have X = X = S. Using Zorn lemma we can show that every graph has maximal cliques and every clique is contained in a certain maximal clique. L) and a point p ∈ P is collinear with all points of X then p is collinear with all points of X . Proof.7.6. It was established above that Y is a maximal singular subspace and we get U = Y .

The associated Grassmann spaces will be denoted by Gk (V ). It states that every collineation of Gk (V ) to Gk (V ).4 is dedicated to apartments preserving mappings.Chapter 3 Classical Grassmannians It was mentioned in the previous chapter that every thick building of type An−1 (n ≥ 4) is isomorphic to the ﬂag complex ∆(V ). 1 < k < n − 1. The classical Chow’s theorem will be proved in Section 3. is induced by a semilinear isomorphism of V to V or a semilinear isomorphism of V to V ∗ (the second possibility can be realized only in the case when n = 2k). If k = 1. n − 1 then apartments of Gk (V ) are bases of ΠV and Π∗ . where V is an ndimensional vector space over a division ring. Following our programme. V Let V be other n-dimensional vector space over a division ring. Section 3. As an application. Note that G1 (V ) = ΠV and Gn−1 (V ) = Π∗ . 1 < k < n − 1.1.3 we characterize apartments of Gk (V ).2. in terms of the adjacency relation.4) and show that every bijection of Gk (V ) to Gk (V ). The Grassmannians of ∆(V ) are the usual Grassmannians Gk (V ). If 1 < k < n − 1 then every apartments preserving mapping of Gk (V ) to Gk (V ) is induced by a semilinear embedding of V in V or a 69 . 1 < k < n − 1. preserving the opposite relation is a collineation of Gk (V ) to Gk (V ). V respectively. their elementary properties will be studied in Section 3. Also we characterize the adjacency relation in terms of the relation to be opposite (as in Theorem 2. In Section 3. All apartments preserving bijections of Gk (V ) to Gk (V ) are collineations of Gk (V ) to Gk (V ). This characterization follows from a more general result concerning apartments in parabolic subspaces of Gk (V ). we will examine collineations of Gk (V ) to Gk (V ) and apartments preserving mappings of Gk (V ) to Gk (V ). we describe all subspaces of Gk (V ) isomorphic to the Grassmann spaces of ﬁnite-dimensional vector spaces.

. Let also k ∈ {1. e In Section 3. Chow’s theorem and the theorem on apartments preserving bijections are closely related with classical Dieudonn´–Rickart’s results on automorphisms of the group GL(V ). the set [M. M.5–3. . x . the latter is equivalent to the fact that the sum of these linear subspaces is (k + 1)-dimensional.7 we demonstrate how our methods work for some constructions rather similar to Grassmannians of ﬁnite-dimensional vector spaces — spine spaces. In particular. the second possibility can be realized only in the case when n = 2k). . This line contains at least three points: we choose two linearly independent vectors x. M. . y then M. Now suppose that M and N are incident linear subspaces of V and dim M < k < dim N (possible M = 0 or N = V ). In Sections 3. If M = 0 or N = V then instead of [M. Recall that two k-dimensional linear subspaces of V (elements of the Grassmannian Gk (V )) are adjacent if their intersection is (k − 1)-dimensional. 3. N ]k := { S ∈ Gk (V ) : M ⊂ S ⊂ N }. y ∈ N such that N is spanned by M and x.70 Grassmannians of Classical Buildings semilinear embedding of V in V ∗ (as in Chow’s theorem. respectively. N ]k we will write N ]k or [M k . We deﬁne [M. .1 Elementary properties of Grassmann spaces Let V be an n-dimensional vector space over a division ring and 3 ≤ n < ∞. n − 1}. n − 1. There is an example showing that the latter statement fails for k = 1. x + y are three distinct points on the line. Grassmannians of exchange spaces and the sets of conjugate linear involutions.8 we extend some results to the case of inﬁnite-dimensional vector spaces focusing on Grassmannians formed by linear subspaces with inﬁnite dimension and codimension. N ]k is a line of Gk (V ). In the case when dim M = k − 1 and dim N = k + 1. y .

S2 . Two distinct elements of Gk (V ) are collinear (joined by a line) if and only if they are adjacent. Lemma 3. Suppose that S3 is not contained in the (k +1)-dimensional linear subspace S1 +S2 . Since S3 is adjacent with S1 and S2 . .Classical Grassmannians 71 The set of all such lines will be denoted by Lk (V ). S + U ]k containing them. S2 . Therefore.1. V moreover. The partial linear spaces Gk (V ). U is a (k − 1)-dimensional linear subspace contained in S1 and S2 . n − 1}. For any k-dimensional linear subspaces S. The associated Grassmann graphs (the collinearity graphs of Gk (V )) will be denoted by Γk (V ). Lk (V )) is a partial linear space. Let 1 < k < n − 1. Recall that three distinct mutually collinear points of a partial linear space form a triangle if they are not collinear (in other words. . . By Proposition 2. The fulﬁllment of both (1) and (2) implies that S1 .1. k ∈ {1. Then the dimension of the linear subspace U := (S1 + S2 ) ∩ S3 is not greater than k − 1. Proposition 3. For any triangle S1 . and for any adjacent S. .1. one of these conditions does not hold. Gk (V ∗ ) and Gn−k (V ) are canonically isomorphic (by the annihilator mapping). It is clear that G1 (V ) = ΠV and Gn−1 (V ) = Π∗ . (2) a top-triangle: there is a (k +1)-dimensional linear subspace containing all Si . Proof. The pair Gk (V ) := (Gk (V ). are the Grassmann spaces of the building ∆(V ). U ∈ Gk (V ) there is the unique line [S ∩ U. these points span a plane). Now we prove the following. the Grassmann space Gk (V ) is connected. . The Grassmann space Gk (V ) is a connected gamma space. S3 in Gk (V ) only one of the following two possibilities is realized: (1) a star-triangle: there is a (k − 1)-dimensional linear subspace contained in each Si . S3 are collinear. U ⊂ V the distance between S and U is equal to dim(S + U ) − k = k − dim(S ∩ U ).

Let 1 < k < n − 1. . . . . Let X ∪ {x1 . . i = 0. U0 ∩ · · · ∩ Ul is contained in the linear subspace S ∩ U whose dimension is equal to k − m. This mapping is injective and the image of every line of Gk (V ) is a line of Π∧k V . This implies that the Grasmann mapping gk : Gk (V ) → G1 (∧k V ) x1 . S is adjacent with all points of the line [M. x2m } be a subset of B such that S∩U = X . xm . .72 Grassmannians of Classical Buildings Proof. . S2 of a certain line [M. . xk and y1 .5). . . Thus k − l ≤ k − m and m ≤ l which means that d(S. xk → x1 ∧ · · · ∧ xk is well-deﬁned. xm+1 . . . . Suppose that V is a vector space over a ﬁeld and consider the exterior power vector space ∧k V . N ]k and S does not belong to this line then M = S1 ∩ S2 ⊂ S or S ⊂ S1 + S2 = N (S. Thus the axiom (Γ) holds (Section 2. x1+i . . . . The k-dimensional linear subspaces Si := X. On the other hand. If vectors x1 . The case k = 1. N ]k . S2 form a star-triangle or a top-triangle. x1 . . . form a path in Γk (V ) connecting S and U . . . . . x2m .4). Remark 3. . . Suppose that the linear subspaces S and U both are k-dimensional. U ⊂ V there is a base B ⊂ V such that S and U are spanned by subsets of B (Proposition 1. . . In each of these cases. respectively). m. If S ∈ Gk (V ) is adjacent with two distinct points S1 . .1 (Grassmann embedding). . . . n − 1 is trivial. xm+i . . U = X. . . U 1 . . . S = X. . Ul = U we have dim(U0 ∩ · · · ∩ Ul ) ≥ k − l. S1 . For every path S = U 0 . yk span the same k-dimensional linear subspace of V then x1 ∧ · · · ∧ xk = a(y1 ∧ · · · ∧ yk ) for a certain non-zero scalar a. For any two linear subspaces S. . U ) = m. Thus gk is an embedding of the Grassmann space Gk (V ) in the projective space Π∧k V .

For every (k−1)-dimensional linear subspace M ⊂ V the set [M k is said to be a star. S2 be distinct elements of X . If S ∈ Gk (V ) does not belong to N ]k then dim(S ∩ N ) ≤ k − 1 and there exist elements of N ]k which are not adjacent with S. If a k-dimensional linear subspace S does not belong to the top N ]k then dim(N ∩ S) ≤ k − 1 and S is not adjacent with at least one of Si . since the canonical collineation between Gk (V ) and Gn−k (V ∗ ) (the annihilator mapping) sends stars to tops and tops to stars.1. This means that N ]k is a maximal singular subspace of Gk (V ). Thus S ∈ X and our clique is a subset of N ]k .2 ([Chow (1949)]). Stars are maximal singular subspaces. 1 < k < n − 1. the class of maximal singular subspaces of Gk (V ) coincides with the class of maximal cliques of the Grassmann graph Γk (V ) (Proposition 2. Since Gk (V ) is a gamma space.Classical Grassmannians 73 Now we describe maximal singular subspaces of Gk (V ). Example 3. Example 3. S3 form a top-triangle. is a star or a top. Let X be a clique of Γk (V ) and S1 .2. This is a singular subspace of Gk (V ) isomorphic to a k-dimensional projective space. Hence it is suﬃcient to show that every clique of the Grassmann graph is contained in a star or a top. S2 . For every (k+1)-dimensional linear subspace N ⊂ V the set N ]k will be called a top. 1 < k < n − 1. Every triangle of N ]k is a top-triangle. there exists S3 ∈ X which does not contain S1 ∩ S2 and S1 . Then S3 is contained in the (k + 1)-dimensional linear subspace N := S1 + S2 . It was noted above that the class of maximal singular subspaces coincides with the class of maximal cliques of the Grassmann graph. Proof. In this case. Proposition 3. Suppose that X is not contained in a star. This is a singular subspace of Gk (V ) isomorphic to an (n − k)-dimensional projective space. .7). Every triangle of a star is a star-triangle. Every maximal singular subspace of Gk (V ).

Then [M. M are adjacent elements of Gk−1 (V ). S + U ]k joining S and U .2.3. Let M and N be incident linear subspaces of V satisfying dim M < k < dim N (possible M = 0 or N = V ). Let S and U be adjacent elements of Gk (V ). Kasikova and Shult (2005)]. 1 < k < n − 1.2. For any adjacent S. Kasikova and Shult (2005). the second possibility is realized only in the case when M ⊂ N .3. This subspace is singular only in the case when M is (k − 1)-dimensional or N is (k + 1)dimensional. isomorphic to the Grassmann space of a ﬁnite-dimensional vector space is parabolic. Denote by X ∼ the set consisting of all elements of Gk (V ) adjacent with each element of a subset X ⊂ Gk (V ). . this intersection is not empty if and only if M. The intersection of two distinct stars [M k and [M k contains at most one element. It will be proved in Section 3.1. Proposition 3. The set of all elements of Gk (V ) adjacent with both S.74 Grassmannians of Classical Buildings Example 3. U }∼∼ . Every subspace of Gk (V ). U ∈ Gk (V ). Theorem 3. 1 < k < n − 1. 1 < k < n − 1. ˙ The following result [Cooperstein. The parabolic subspace [M.2.3. the line joining S and U coincides with the subset {S. The canonical collineation between Gk (V ) and Gn−k (V ∗ ) transfers parabolic subspaces to parabolic subspaces. N ]k is a subspace of Gk (V ). every singular subspace of Gk (V ) is parabolic. where m is the dimension of M . The intersection of the star [M k and the top N ]k is empty or a line. By Proposition 3. U is the union of all maximal cliques of the Grassmann graph (all maximal singular subspaces of Gk (V )) containing S and U . The same holds for the intersection of two distinct tops. subspaces of such type are called parabolic [Cooperstein. By Proposition 3. this is [S ∩ U k ∪ S + U ]k . We get the following characterization of lines in terms of the adjacency relation. Zynel (2000)] generalizes Proposition 3. N ]k is isomorphic to the Grassmann space Gk−m (N/M ). A k-dimensional linear subspace of V is adjacent with all elements of this set if and only if it belongs to the line [S ∩ U.

2 Collineations of Grassmann spaces Throughout the section we suppose that V and V are n-dimensional vector spaces over division rings and 4 ≤ n < ∞. the third possibility is realized only in the case when these singular subspaces are of diﬀerent types (one of them is a star and the other is a top) and the associated (k − 1)-dimensional and (k + 1)-dimensional linear subspaces are incident. 3.2.2 ([Chow (1949)]). .3. Then every isomorphism of Γk (V ) to Γk (V ) is the collineation of Gk (V ) to Gk (V ) induced by a semilinear isomorphism of V to V or a semilinear isomorphism of V to V ∗ . Proof. or a line. By Proposition 3. By duality. n − 1.3. This is a collineation of Gk (V ) to Gk (V ) if n = 2k. we can identify (l)k with the collineation of Gk (V ) to Gn−k (V ) sending each S to l(S)0 . every bijection of Gk (V ) to Gk (V ) is an isomorphism of Γk (V ) to Γk (V ). or a single point. Let f be an isomorphism of Γk (V ) to Γk (V ). Let 1 < k < n − 1. By duality. Remark 3. the second possibility can be realized only in the case when n = 2k.2) is a collineation of Gk (V ) to Gk (V ). it can be considered as the collineation of Gn−k (V ∗ ) to Gn−k (V ∗ ) induced by the contragradient u (Subsection 1. Then f and f −1 map maximal singular subspaces to maximal singular subspaces.2 that every collineation of Gk (V ) to Gk (V ) is induced by a semilinear isomorphism of V to V or V ∗ . We will study collineations between the Grassmann spaces Gk (V ) and Gk (V ) for 1 < k < n − 1.2). f is a collineation of Gk (V ) to Gk (V ). Recall that maximal singular subspaces of our Grassmann spaces are stars and tops. ˇ Let l : V → V ∗ be a semilinear isomorphism.1 Chow’s theorem For every semilinear isomorphism u : V → V the bijection (u)k : Gk (V ) → Gk (V ) (Subsection 1.3).2.Classical Grassmannians 75 3. Theorem 3. In the cases when k = 1.3. It follows immediately from Theorem 3. The intersection of two distinct maximal singular subspaces is empty. This statement generalizes the classical version of the Fundamental Theorem of Projective Geometry (Corollary 1.

in particular. f1 is induced by a semilinear isomorphism l : V → V . . Case (A). one of the following possibilities is realized: (A) stars go to stars. f1 is a collineation of ΠV to ΠV . The same arguments show that tops go to tops and tops go to stars in the cases (A) and (B). By connectedness (Proposition 3. . For every U ∈ Gk (V ) S ∈ U ]k−1 ⇔ U ∈ [S and we have fk−1 ( U ]k−1 ) = f (U )]k−1 . respectively.76 Grassmannians of Classical Buildings Now suppose that for a (k − 1)-dimensional linear subspace S ⊂ V the image of the star [S k is a star. we establish that all stars go to tops if the image of a certain star of Gk (V ) is a top. Since [S k goes to a star. Thus fk−1 and the inverse mapping send tops to tops which implies that fk−1 is an isomorphism of type (A) between the Grassmann graphs Γk−1 (V ) and Γk−1 (V ). Therefore. The stars [S k and [U k intersect the top N ]k by lines. we get a sequence of such isomorphisms fi : Gi (V ) → Gi (V ). . . hence f ([U k ) is a star. there exists a bijection fk−1 : Gk−1 (V ) → Gk−1 (V ) such that f ([S k ) = [fk−1 (S) for all S ∈ Gk−1 (V ). In this case. Let U be a (k − 1)-dimensional linear subspace adjacent with S.1) for each U ∈ Gi (V ). Similarly. i = k. We can prove that fi = (l)i induction by i. (B) stars go to tops. 1. Indeed. if fi−1 is induced by l then for every U ∈ Gi (V ) we have fi−1 ( U ]i−1 ) = l(U )]i−1 i k k ⇔ f (U ) ∈ [fk−1 (S) k ⇔ fk−1 (S) ∈ f (U )]k−1 . We choose a (k + 1)-dimensional linear subspace N containing S and U . the image of N ]k is a top. By the Fundamental Theorem of Projective Geometry.1). Step by step. Then fi−1 ( U ]i−1 ) = fi (U )]i−1 (3. the same holds for every (k − 1)-dimensional linear subspace U ⊂ V . where fk = f and fi ([S i ) = [fi−1 (S) for all S ∈ Gi−1 (V ) if i > 1.

. In this case. this isomorphism sends stars to stars. Theorem 3. Case (B). f can be considered as a colleniation of Gk (V ) to Gk (V ∗ ) sending stars to stars. L) be a linear space. Let Π = (P.3 ([Huang (1998)]). • All tops are maximal cliques and f transfers stars to tops and tops to stars. It is not diﬃcult to prove that Γ1 (Π) is connected and every maximal clique of this graph is a star or a subset of a top (a star is formed by all lines passing through a point. The Grassmann graph Γ1 (Π) is the graph whose vertex set is L and whose edges are pairs of adjacent lines. L ) be linear spaces of the same ﬁnite dimension n ≥ 3.1) gives the claim. respectively. Let 1 < k < n − 1. Π and Π both are 3-dimensional generalized projective spaces and f is induced by a collineation of Π to the generalized projective space dual to Π . Stars and tops are projective spaces of dimension n − k and k. in particular. Then one of the following possibilities is realized: • The isomorphism f maps stars to stars and it is induced by a collineation of Π to Π . there are planes containing pairs of non-adjacent lines (for example. Two distinct lines of Π are non-intersecting or have a common point. 3. parallel lines in aﬃne planes). However. Every collineation between linear spaces induces an isomorphism between the associated Grassmann graphs.4 ([Havlicek (1999)]). in contrast to the projective case. a top is the set of all lines contained in a plane).Classical Grassmannians 77 and (3. Every pair of adjacent lines spans a plane. in the second case. The following remarkable result is closely related with Chow’s theorem. these lines are said to be adjacent. Theorem 3. Hence it is induced by a semilinear isomorphism of V to V ∗ . every semicollineation of Gk (V ) to Gk (V ) is a collineation. we have n = 2k. Since f is a collineation and preserves the dimensions of singular subspaces. Suppose that f : L → L is an isomorphism of Γ1 (Π) to Γ1 (Π ).2 Chow’s theorem for linear spaces There is an analogue of Chow’s theorem for linear spaces. Let Π = (P. If a surjection of Gk (V ) to Gk (V ) sends adjacent elements to adjacent elements then it is a collineation of Gk (V ) to Gk (V ). By duality.2. L) and Π = (P .

L) is a generalized projective space of dimension 3 and denote by P ∗ the set of all planes of Π. lines containing more than two points. .3 Applications of Chow’s theorem gk : Gk (V ) → Gk (V ) and gm : Gm (V ) → Gm (V ) Proposition 3. . Suppose that Π = (P.m (V ) := i=k Gi (V ).m (V ) there exists g(U ) ∈ Gk. By symmetry. projective spaces.2. Let us deﬁne m m Gk.m (V ) such that gk ( U ]k ) = g(U )]k .m (V ) := i=k Gi (V ) and Gk. There is natural one-to-one correspondence between elements of L and L∗ . . Then Π∗ := (P ∗ . points from distinct components are joined by lines of cardinality 2 [Buekenhout and Cameron (1995)].4. m ∈ {1. L∗ can be identiﬁed with L and every collineation of Π to Π∗ induces an automorphism of the Grassmann graph Γ1 (Π). For every U ∈ Gk. Proof. in other words. Let k. Let also be bijections satisfying the following condition: S ∈ Gk (V ) and U ∈ Gm (V ) are incident if and only if gk (S) and gm (U ) are incident. It is clear that gk ( U ]k ) = gm (U )]k for all U ∈ Gm (V ). A subset of P ∗ is called a line if it consists of all planes containing a certain line L ∈ L. gk and gm are induced by the same semilinear isomorphism of V to V .3. Since two lines of Π are adjacent if and only if the corresponding lines of Π∗ are adjacent. A generalized projective space can be deﬁned as a linear space where any two lines contained in a plane have a common point (if every line of a generalized projective space contains at least three points then it is a projective space). Every ﬁnite-dimensional generalized projective space is the union of a ﬁnite collection of the following “components”: points. Then there exists a semilinear isomorphism l : V → V such that gk = (l)k and gm = (l)m . .78 Grassmannians of Classical Buildings Remark 3. . the set of all such lines is denoted by L∗ . we can restrict ourselves to the case when k < m. 3. L∗ ) is a generalized projective space of dimension 3. n − 1} and k = m. this generalized projective space is called dual to Π.

m (V ) is bijective and its restrictions to Gk (V ) and Gm (V ) coincide with gk and gm .3).m (V ) → Gk. Therefore. Let also gk : Gk (V ) → Gn−k (V ) and gm : Gm (V ) → Gn−m (V ) be bijections such that S ∈ Gk (V ) and U ∈ Gm (V ) are incident if and only if gk (S) and gm (U ) are incident. Let n = 2k and f be a bijection of Gk (V ) to Gn−k (V ) satisfying the condition (P). n − 1} and k = m. f (U ) are incident if and only if U. Let Ω be a non-degenerated reﬂexive form deﬁned on V and ⊥ be the associated orthogonal relation (Subsection 1. By duality. If k = 1 then f is a polarity (Subsection 1. Then f is a collineation of Gk (V ) to Gn−k (V ) satisfying the following condition: (P) for any S. gk is induced by a semilinear isomorphism l : V → V . . Suppose that f is the restriction of this transformation to Gk (V ). For all M. respectively. In the case when k > 1. .13. . we have the following.5. . .5 ([Pankov 1 (2004)]). we choose m-dimensional linear subspaces U1 . . f (S) are incident.2).5. then g(U ) := gm (U1 ) ∩ · · · ∩ gm (Ui ) is as required. Then gk and gm are induced by the same semilinear isomorphism of V to V ∗ .1. . the mapping gk is an isomorphism of Γk (V ) to Γk (V ) (since gk and the inverse mapping send tops to tops). N ∈ Gk. U ∈ Gk (V ) the linear subspaces S. . an easy veriﬁcation shows that gm = (l)m . g transfers Gk+1 (V ) to Gk+1 (V ). . Let k. in particular. Corollary 3.Classical Grassmannians 79 Indeed. The following result generalizes Proposition 1. Consider the bijective transformation of G(V ) sending each linear subspace S ⊂ V to S ⊥ . Then there exists a nondegenerate reﬂexive form Ω such that the restriction of the transformation S → S ⊥ (where ⊥ is the orthogonal relation associated with Ω) to Gk (V ) coincides with f . The mapping g : Gk. This means that gk is a collineation of ΠV to ΠV if k = 1.m (V ) we have M ⊂ N ⇐⇒ g(M ) ⊂ g(N ). Theorem 3. m ∈ {1. Ui satisfying U = U1 ∩ · · · ∩ Ui .

S2 } such that every U ∈ Gk (V ) opposite to S is opposite to at least one of Si . This implies that h2 is identity (we leave the details for the reader).1 to the bijections f : Gk (V ) → Gn−k (V ) and f −1 : Gn−k (V ) → Gk (V ). The latter means that the sesquilinear form deﬁned by u is reﬂexive. For any distinct k-dimensional linear subspaces S1 . Theorem 3. Let us consider the bijective transformation h of G(V ) which sends every linear subspace S to u(S)0 . . Since the annihilator mapping of Gk (V ) to Gn−k (V ∗ ) is an isomorphism between the associated Grassmann graphs. Proof. Let 1 < k < n − 1. 3. The restrictions of h to Gk (V ) and Gn−k (V ) coincide with f and f −1 . in the case when 2k ≤ n. (2) there exists S ∈ Gk (V ) \ {S1 . Two vertices of the Grassmann graph Γk (V ) (elements of the Grassmannian Gk (V )) are said to be opposite if the distance between them is maximal (equal to the diameter of the graph). Therefore. the restriction of h2 to Gk (V ) is identity.4 Opposite relation Let 1 < k < n − 1. So.6. we can restrict ourselves to the case when 2k ≤ n. In the case when n = 2k. respectively.4. two k-dimensional linear subspaces are opposite if and only if their intersection is zero. Therefore. In this case. Since k = n − k. two elements of Gk (V ) are opposite if and only if their intersection is zero. This reﬂexive form is as required.1. S2 ⊂ V the following conditions are equivalent: (1) S1 and S2 are adjacent.80 Grassmannians of Classical Buildings Proof. the diameter of Γk (V ) is equal to k n−k if 2k ≤ n if 2k > n.2. the condition (P) is equivalent to the fact that f is an involution of Gk (V ) (f 2 is identity) and the statement given above fails. The adjacency relation can be characterized in terms of the relation to be opposite. we can apply Corollary 3. By Proposition 3. Remark 3. f and f −1 both are induced by a semilinear isomorphism u : V → V ∗ .

By (3. By our hypothesis. First we establish that for any 1-dimensional linear subspaces P1 ⊂ S1 and P2 ⊂ S2 the sum P1 + P2 has a non-zero intersection with S. Let P1 and Q1 be distinct 1dimensional linear subspaces of U . Suppose that this intersection is zero.3)). The proof of this implication will be given in several steps. S2 } belonging to the line joining S1 and S2 is as required. In particular. hence U +P2 is 3-dimensional. this linear subspace is not contained S1 ∩ S2 and (S1 ∩ S2 ) + P is a unique point of the line joining S1 and S2 which is not opposite to U . 2) has a non-zero intersection with S.2) . Now we choose 1-dimensional linear subspaces P1 ⊂ S1 and P2 ⊂ S2 such that Si = (S ∩ Si ) + Pi (this is possible by (3. Thus S1 and S2 both are contained in the (k + 1)-dimensional linear subspace S + P1 which means that they are adjacent. Our second step is to show that dim(S ∩ S1 ) = dim(S ∩ S2 ) = k − 1. hence S1 ∩ S2 ⊂ S. Thus there is Pi which is not contained in Si . It is clear that P = Q. By the ﬁrst step of the proof. their intersection is non-zero. Let us take a 2-dimensional linear subspace U ⊂ S1 and a 1-dimensional linear subspace P2 ⊂ S2 which is not contained in S. the linear subspaces P1 + P2 and Q1 + P2 meet S in 1-dimensional linear subspaces P and Q. Then P1 + P2 is contained in a k-dimensional linear subspace opposite to S. If S1 and S2 are adjacent elements of Gk (Π) then every S ∈ Gk (V ) \ {S1 . (2) =⇒ (1). since S1 ∩ S2 ⊂ S.Classical Grassmannians 81 (1) =⇒ (2). (3. Indeed. P2 is not contained in S1 . Then P1 + P2 has a non-zero intersection with S (by the ﬁrst step of the proof) and S + P1 contains P2 .2). Since U and P + Q are 2dimensional linear subspaces contained in the 3-dimensional linear subspace U + P2 . The inclusion P + Q ⊂ S gives the claim. respectively. if U ∈ Gk (V ) is opposite to S then it intersects S1 + S2 in a certain 1-dimensional linear subspace P . (3. every 1-dimensional linear subspace P ⊂ S1 ∩ S2 is contained in S (we take P1 = P2 = P ). this linear subspace is opposite to at least one of Si . a contradiction.3) It is suﬃcient to establish that every 2-dimensional linear subspace contained in Si (i = 1.

U }. Theorem 3. For a subset X ⊂ Gk (V ) we deﬁne X m := { Y ∈ Gk (V ) : d(X. we have the following result [Lim (2010)]: if f : Gk (V ) → Gk (V ) is a surjection such that d(S. for any S. or they are non-adjacent and ({S. Remark 3.5. U ) ≤ m one of the following possibilities is realized: S.6. U are adjacent and ({S. As in the proof of Theorem 3.6 was ﬁrst proved in [Blunck and Havlicek (2005)] for the case when n = 2k. Every bijection of Gk (V ) to Gk (V ) preserving the relation to be opposite (two elements of Gk (V ) are opposite if and only if their images are opposite) is a collineation of Gk (V ) to Gk (V ). As a consequence. Let m be a positive integer which is less than the diameter of Γk (V ). Theorem 3. Y ) ≤ m ∀ X ∈ X }. (3) there exists S ∈ Gk (V ) \ {S1 . f (U )) ≤ m for any S. Proof. U ∈ Gk (V ) satisfying 0 < d(S.82 Grassmannians of Classical Buildings Corollary 3.6. we have (3) =⇒ (2) and Theorem 3. Theorem 3. By [Lim (2010)]. U ) ≤ m ⇐⇒ d(f (S).7.2. Let 1 < k < n − 1. Since an element of Gk (V ) is opposite to S ∈ Gk (V ) if and only if it is contained in a complement of S (this is true only in the case when 2k ≤ n). n− 2. U ∈ Gk (V ) then f is an isomorphism of Γk (V ) to Γk (V ). U }m )m coincides with {S. Remark 3. S2 } such that every complement of S is a complement to at least one of Si . The implication (1) =⇒ (3) is obvious. two distinct elements of Gk (V ) are adjacent or opposite and the latter statement is trivial. . an easy veriﬁcation shows that these conditions hold for the Grassmann graph Γk (V ).6 can be reformulated in the following form. S2 ⊂ V the following conditions are equivalent: (1) S1 and S2 are adjacent. Remark 3. In [Huang and Havlicek (2008)] the same characterization was obtained for an abstract graph satisfying certain technical conditions. For any distinct k-dimensional linear subspaces S1 . and it was shown later [Havlicek and Pankov (2005)] that Blunck–Havlicek’s method works for the general case. In the case when k = 2.6 gives the claim. U }m )m is the line joining S with U .7. Let 1 < k < n − 1. we can restrict ourselves to the case when 2k ≤ n.

. Denote by Ak the associated apartment of the Grassmannian Gk (V ). k) (the graph whose vertex set is formed by all k-element subsets of {1. xik and n |Ak | = . Every maximal clique of Γ(Ak ) is a base of the corresponding maximal singular subspace of Gk (V ). they will be called stars and tops of Ak (respectively). U }. see Section 3. our proof of Theorem 3. it consists all k-dimensional linear subspaces xi1 . . .6 is short and it can be modiﬁed for Grassmannians of inﬁnite-dimensional vector spaces.3. U ∈ Gk (V ) the intersection of all apartments of Gk (V ) containing S and U coincides with {S. . By these reasons. The annihilator mapping transfers Ak to the apartment of Gn−k (V ∗ ) associated with the dual base B ∗ (Subsection 1.7. .1.8. .Classical Grassmannians 83 By the remarks given above. The restriction of the Grassmann graph Γk (V ) to a subset X ⊂ Gk (V ) will be denoted by Γ(X ). . . n − 1 then Ak is a base of ΠV or Π∗ . If 1 < k < n − 1 then the maximal cliques of V Γ(Ak ) are the intersections of Ak with the stars [M k . . It is clear that Γ(Ak ) is isomorphic to the Johnson graph J(n. Exercise 3. . M ∈ Ak−1 and the tops N ]k .3 3. The same idea also will be exploited in Section 4. Let also B = {x1 . k Remark 3.8. The apartment of Gk (V ) deﬁned by a base B ⊂ V coincides with Ak if and only if the vectors of B are scalar multiples of the vectors of B.6 is a partial case of more general results. . xn } be a base of V . N ∈ Ak+1 .1. n} and two such subsets are connected by an edge if their intersection consists of k − 1 elements). Theorem 3. we present this proof here. 3. If k = 1. respectively. Thus there is a one-to-one correspondence between apartments and bases of the projective space ΠV . . Show that for any S.1 Apartments Basic properties Let V be an n-dimensional vector space over a division ring and 3 ≤ n < ∞. On the other hand. .3).

5 guarantees that [M. . . Let M and N be incident linear subspaces of V such that dim M = m < k < l = dim N. N ]k . The natural collineation of [M. Now we investigate apartments of parabolic subspaces (Example 3. N ]k is called an apartment of [M. Denote by X the subspace of Gk (V ) spanned by A and deﬁne Pi := xi for every i. We take any base of V whose subsets span M and N . Kasikova and Shult (2005)]). N ]k are isomorphic to the Johnson graph J(l − m. Sn span the star [M k (otherwise. Therefore. The restrictions of the Grassmann graph Γk (V ) to apartments of [M. . Let 1 < k < n − 1 and l. The Grassmann space Gk (V ) is spanned by every apartment of Gk (V ).8 ([Cooperstein. the intersection of the associated apartment of Gk (V ) with the parabolic subspace [M. S1 + · · · + Sn is a proper subspace of V which contradicts the fact that B is a base). in particular. Then X is an apartment in a parabolic subspace of Gk (V ). Let also X be a subset of Gk (V ) such that Γ(X ) is isomorphic to the Johnson graph J(l − m. . N ]k and apartments of Gk−m (N/M ). the inductive hypothesis implies that [Pi k is spanned by A ∩ [Pi k . Every Si belongs to X and we get S ∈ [M k ⊂ X which completes our proof. . . Then S1 .3). k − m). k − m) and every maximal clique of Γ(X ) is an independent subset of the Grassmann space Gk (V ).84 Grassmannians of Classical Buildings Proposition 3. [Pi k ⊂X for every i. Proof. The case k = 1 is trivial and we prove the statement induction by k. Proposition 3. . Now consider a k-dimensional linear subspace S ⊂ V such that xi ∈ S for all i. . We take any (k − 1)-dimensional linear subspace M ⊂ S and deﬁne Si := M + Pi for every i. Since [Pi k can be identiﬁed with the Grassmann space Gk−1 (V /Pi ). N ]k is spanned by every of its apartments. N ]k to the Grassmann space Gk−m (N/M ) establishes a one-to-one correspondence between apartments of [M. this is an apartment of Gk (V ) if l = n and m = 0.5. Theorem 3. m be natural numbers satisfying k < l ≤ n and 0 ≤ m < k. xn } be a base of V and A be the associated apartment of Gk (V ). . Let B = {x1 .

8 (concerning the independence of maximal cliques) cannot be dropped. This parabolic subspace is spanned by f (A) and we get the claim. 0) (0. 0. Then X contains non-adjacent elements. 0.4. (B) stars go to tops and tops go to stars. We take an (l − m)-dimensional vector space W and any apartment A ⊂ Gk−m (W ). 0. The restriction of the Grassmann graph Γ2 (R5 ) to this subset is isomorphic to J(5. 0) (0.2. Let f be a collineation of the Grassmann space Gm (W ) to a subspace of Gk (V ). Suppose that k + 1 < l and m < k − 1.8 and Proposition 3.8 If l = k + 1 or m = k − 1 then any two distinct elements of X are adjacent and the statement is trivial.Classical Grassmannians 85 The following example shows that the second condition in Theorem 3. we consider the apartment A∗ ⊂ Gl−k (W ∗ ) formed by the annihilators of the elements from A.1 is a simple consequence of Theorem 3.3. Theorem 3. 1.2 Proof of Theorem 3. 1.8. 0. Let Y be a maximal clique of Γ(X ). 1. However. it is not an apartment of G2 (R5 ). Let f : A → X be an isomorphism of Γ(A) to Γ(X ). Then f (A) satisﬁes the conditions of Theorem 3. 1. 3. 0. 0) (1. 0) (0. 0. We say that Y is a star or a top of X if the maximal clique of Γk (V ) containing Y is a star or a top. the mapping U → f (U 0 ) is an . In the case (B). Then Y is contained in precisely one maximal clique of Γk (V ) (since Y is an independent subset of Gk (V ) containing more than 2 elements and the intersection of two distinct maximal cliques of Γk (V ) does not contain a triangle). 2). Let us take the following ﬁve vectors (1.5. 0. 1. hence it is an apartment in a parabolic subspace of Gk (V ). Example 3. As in the proof of Theorem 3. 1. 0) in R5 and consider the subset of G2 (R5 ) consisting of all 2-dimensional linear subspaces spanned by pairs of these vectors. 0. respectively. 0. We take any apartment A ⊂ Gm (W ). one of the following possibilities is realized: (A) stars go to stars and tops go to tops.

where T is the (k + 1)-dimensional linear subspace corresponding to Y. . and for every i ∈ {1. By this reason. Since this is an independent subset. Every element of Y is the sum of k − m elements of BY . Y ⊂ X . Consider X ∈ B(S) \ BY . U. There is unique top Y ⊂ X containing S and U . By connectedness of Grassmann spaces. Let Y ⊂ X be a top containing S and such that X ∈ BY . Since S. we have S∩U =S∩U . Denote by BY the set formed by all Xi . Suppose that a top Y ⊂ X has a non-empty intersection with Y. . MY = MY for any two tops Y. We choose unique . (2) if X ∈ BY is contained in S then it belongs to BY . Every top Y ⊂ X consists of k − m + 1 elements. suppose that Y = {S1 . the linear subspace MY := S1 ∩ · · · ∩ Sk−m+1 is m-dimensional. by the property (2). . we show that B(S) coincides with B(U ) if S. Sk−m+1 }. . This implies the following properties: (1) MY = MY . thus BY ∩ BY consists of k − m elements and there is a unique element of BY which does not belong to BY . . .86 Grassmannians of Classical Buildings isomorphism of Γ(A∗ ) to Γ(X ) satisfying (A). in what follows this m-dimensional linear subspace will be denoted by M . By the deﬁnition. This is a base of [MY . we restrict ourselves to the case (A). BY is contained in both B(S) and B(U ). Now. . k − m)). There are precisely l − k distinct tops of X containing S and. . For every U ∈ Y \ {S} there exists unique U ∈ Y \ {S} adjacent with U (this follows from the fact that Γ(X ) is isomorphic to the Johnson graph J(l − m. T ]m+1 . Then the (k + 1)-dimensional linear subspaces corresponding to Y and Y are adjacent and Y ∩ Y consists of unique element S. k − m + 1} Xi := j=i Sj is an (m + 1)-dimensional linear subspace. For every S ∈ X we denote by B(S) the union of all BY such that S ∈ Y. U ∈ X are adjacent. U form a star-triangle. the set B(S) consists of l − m elements.

also we have M ⊂ N . denote this set by B. If V is a vector space over a ﬁeld then every base {x1 .4 Apartments preserving mappings Let V and V be n-dimensional vector spaces over division rings and n ≥ 3. 3.9 (M. Clearly.1) establishes a one-to-one correspondence between apartments of Gk (V ) and regular bases of Π∧k V . Conversely. Theorem 3. B(S) ⊂ B(U ). A point of . Since X is not contained in S (X ∈ BY ). every element of X is the sum of k − m elements from B. we have the following. Recall that every apartments preserving mapping is injective. the sum of all elements of B is a linear subspace N whose dimension is not greater than l. N ]k (we get an apartment of Gk (V ) if l = n and m = 0). this implies X ∈ BZ ⊂ B(U ). . Remark 3. the projective space Π∧k V has non-regular bases. Every apartments preserving bijection of Gk (V ) to Gk (V ) is a collineation of Gk (V ) to Gk (V ). . Then B is a base of [M.4. the same arguments give the inverse inclusion. xn } of V deﬁnes the regular base of Π∧k V consisting of all xi1 ∧ · · · ∧ xik . we have X ⊂ U .Classical Grassmannians 87 U ∈ Y adjacent with U and denote by Z the top of X containing U and U . and every star of X is an independent subset consisting of l − k + 1 elements.9. by the property (2). This implies that dim N ≥ l and N is l-dimensional. Let f be a bijective transformation of the projective space Π∧k V preserving the family of regular bases (f and f −1 map regular bases to regular bases). . By connectedness. U ∈ X . Since B is formed by l−m distinct (m+1)-dimensional linear subspaces containing M .1 Results All collineations of Gk (V ) to Gk (V ) (if they exist) are apartments preserving (since they are induced by semilinear isomorphisms). The Grassmann embedding gk (Remark 3. 3. B(S) = B(U ) ∀ S. On the other hand. N ]m+1 and X is the associated apartment of [M. Therefore. . Pankov). In this section we investigate apartments preserving mappings of Gk (V ) to Gk (V ).

. For every k ∈ {1. Let 1 < k < n − 1. If k = 1. .10 (M. Example 3.88 Grassmannians of Classical Buildings Π∧k V is contained in a regular base if and only if it belongs to the image of Gk (V ). tn . . The sets R and G1 (Rn ) have the same cardinality and we consider any bijection of G1 (Rn ) to α(R). . tn ) . . Theorem 3. tn 1 . but it is not induced by a semilinear mapping (since the images of three distinct collinear points are non-collinear). = 0. Pankov). tn ∈ R \ {0} we have t1 . . . This mapping transfers bases of ΠRn to bases of ΠRn .3. . . this is an apartments preserving embedding of Gk (V ) in Gk (V ). Then every apartments preserving mapping of Gk (V ) to Gk (V ) is the embedding of Gk (V ) in Gk (V ) induced by a semilinear embedding of V in V or a semilinear embedding of V in V ∗ . Let u : V → V ∗ be a semilinear embedding. . By duality. .. . n − 1 then there exist apartments preserving mappings of Gk (V ) to Gk (V ) which cannot be induced by semilinear mappings. . t2 . Let l : V → V be a semilinear embedding. .5 ([Huang and Kreuzer (1995)]). . In the case when n = 2k. Let us consider the injective mapping α : R → G1 (Rn ) deﬁned by the formula t → (t. . Recall that (l)k is bijective if and only if l is a semilinear isomorphism (Proposition 1. the second possibility can be realized only in the case when n = 2k. .11). n − 1} the mapping (l)k : Gk (V ) → Gk (V ) (Subsection 1. . Thus f transfers the image of Gk (V ) to itself and the restriction of f to this subset can be identiﬁed with an apartments preserving bijective transformation of Gk (V ). .2) is an apartments preserving embedding of Gk (V ) in Gk (V ). . any three points of α(R) are non-collinear. we can identify (u)k with the embedding of Gk (V ) in Gn−k (V ) sending each S to u(S) 0 . . . . . in particular. tn n Hence any subset of α(R) consisting of n distinct points is a base of the projective space ΠRn . For any distinct t1 .

A(S) coincides with a certain A(−i) if S is (n − 1)-dimensional. by Exercise 1. Proof. Our proof of Theorem 3. Theorem 3. we will write A(+i. xn } be a base of V and A be the associated apartment of Gk (V ). Clearly.Classical Grassmannians 89 In the case when 1 < k < n − 1. . A subset R ⊂ A is exact if and only if for each i the intersection of all S ∈ R containing xi coincides with xi .2. It is trivial that R ⊂ A is exact if the intersection of all S ∈ R containing xi coincides with xi for every i. thus R is inexact. Hence f −1 is a semicollineation of ΠV to ΠV and Corollary 1.10 is a modiﬁcation of the proof given in [Pankov 2 (2004). in the case when S is spanned by xi and xj . Then f maps bases of ΠV to bases of ΠV .9 was ﬁrst proved in [Pankov (2002)]. the mapping f sends each triple of non-collinear points to non-collinear points. Throughout the subsection we suppose that 1 < k ≤ n − k. Since three points of a projective space are non-collinear if and only if they are contained in a certain base of this space. Suppose that k = 1.11. . Pankov 3 (2004)].10 and Proposition 1.11. . 3.3 gives the claim. Let f be an apartments preserving bijection of Gk (V ) to Gk (V ). A subset of A is called exact if it is contained only in one apartment of Gk (V ).2 Proof of Theorem 3. Remark 3. We choose a vector x = axi + bxj with a. respectively. This implies that the inverse mapping transfers triples of collinear points to collinear points. We write A(+i) and A(−i) for the sets consisting of all elements of A which contain xi and do not contain xi . otherwise. .4. Theorem 3. b = 0. If there are no elements of R containing xi then we can take any vector x which is not a scalar multiple of xi and is not contained in the linear subspace spanned by B \ {xi }. The case k = n − 1 is similar. .10. it is said to be inexact. +j) instead of A(S).9 is a direct consequence of Theorem 3. Lemma 3. Then (B \ {xi }) ∪ {x} is a base of V which deﬁnes another apartment of Gk (V ) containing R.10: First step Let B = {x1 . If S is a linear subspace spanned by a subset of B then we denote by A(S) the set of all elements of A incident with S. Suppose that for a certain number i there exists j = i such that xj belongs to every S ∈ R containing xi .

−j). . where M and N are linear subspaces spanned by subsets of B and dim M = |{i1 . +j) for some distinct i. . .4) {i1 . Example 3. . . ik or j1 .6. . in this case. . . . Since R is a maximal inexact subset. Then R = A(+i) ∩ A(−j). −j1 ). −jk ) form a regular collection if their intersection is a one-element set. This intersection is not empty if and only if (3. . ik are distinct then R1 ∩ · · · ∩ Rk = { xi1 . A(−i) ∪ A(+i. If (3. . In this case. . . +j) and R ⊂ A(−i) ∪ A(+i. . By Lemma 3. j such that A \ R = A(−i) ∪ A(+i. R1 ∩ · · · ∩ Rk = A(M ) ∩ A(N ). Rk = A(+ik . A subset R ⊂ A is said to be complement if A \ R is a maximal inexact subset. . . .90 Grassmannians of Classical Buildings By Lemma 3.4) holds and i1 . . ik } ∩ {j1 . If R is a maximal inexact subset of A then R = A(−i) ∪ A(+i. . . . Proof. +j). . +j). . we have the inverse inclusion. . Then every S ∈ R is an element of A(−i) or A(+i. . Lemma 3.3. xik } and the collection is regular. For each i there is an element of R containing xi (if all elements of R do not contain xi then R is a subset of the non-maximal inexact subset A(−i) which contradicts the fact that our inexact subset is maximal). .2. . . . is an inexact subset. .2. jk }| = dim N (note that some of i1 . . . Lemma 3. . We say that distinct complement subsets R1 = A(+i1 . jk } = ∅. . ik }| ≤ k ≤ n − k ≤ n − |{j1 . In the general case. +j). . .3 implies the existence of distinct i. . j. i = j. jk can be coincident). there exist distinct i and j such that xj belongs to all elements of R containing xi . M is contained in N . in what follows this complement subset will be denoted by A(+i.

. Rk is regular if and only if (3. . xik }. . . Then R1 ∩ · · · ∩ Rk = { xi1 . . Rk is a regular collection. there exists a complement subset Rk ⊂ A such that Rk = R1 . . .4) holds and one of the following possibilities is realized: (A) i1 . . in other words. . Rk−1 ⊂ A is said to be regular if it can be extended to a regular collection of k distinct complement subsets.4. (1) =⇒ (2). . . . The adjacency relation can be characterized in terms of regular collections of complement subsets. . . The following conditions are equivalent: (1) S. . . . U ∈ A are adjacent. . . Suppose that S and U are adjacent and S ∩ U = xi1 . . . .Classical Grassmannians 91 Example 3. Suppose that n = 2k. (B) n = 2k and j1 . The collection R1 . . jk are distinct.4) holds. . . If (3. xik−1 . . . the condition (3. Lemma 3.5. . . . A collection of k − 1 distinct complement subsets R1 . jk }. .4) is fulﬁlled and the conditions (A) and (B) both do not hold then dim M < k < dim N and A(M ) ∩ A(N ) contains more than one element. Rk−1 . where {i1 . Rk−1 and R1 . We choose xi ∈ S \ U and xj ∈ S + U. . . . . Proof. . . . . . . n} \ {j1 . The collection is regular. and j1 . jk are distinct. . Proof. . Lemma 3. (2) there exists a regular collection of k − 1 distinct complement subsets of A such that each element of this collection contains S and U . . . . .7. ik } = {1. . . ik are distinct.

−j). By Lemma 3. −jk−1 ) contains S and U . f |A and the inverse mapping transfer regular collections of complement subsets to regular collections (this follows immediately from the deﬁnition of a regular collection). . . xjk−1 do not belong to S + U and (B) implies that S + U is (k + 1)-dimensional. moreover. Proposition 3.5. −j) ∩ A(+i. This implies that an inexact subset of A is maximal if and only if its image is a maximal inexact subset of f (A). Since A and f (A) have the same number of inexact subsets. . . . . . (A) guarantees that the dimension of S ∩ U is equal to k − 1. First of all we recall that f is injective. The vectors xj1 . (2) =⇒ (1). . every inexact subset of f (A) is the image of a certain inexact subset of A. . Then f (A) is an apartment of Gk (V ) and f maps inexact subsets of A to inexact subsets of f (A). this collection can be extended to a regular collection of k distinct complement subsets. Therefore. . suppose that each element of a regular collection A(+i1 . Conversely. . and it follows from Lemma 3. . In each of these cases. . Let S and U be kdimensional linear subspaces of V and A be an apartment of Gk (V ) containing them. . jk−1 are distinct. a subset R ⊂ A is complement if and only if f (R) is a complement subset of f (A). . Since xi1 . . A(+ik−1 . −j) ∩ · · · ∩ A(+ik−1 .92 Grassmannians of Classical Buildings Then A(+i1 . −j) = {S}. . By deﬁnition. . (B) n = 2k and j1 . ik−1 are distinct. Proof. xik−1 belong to S ∩ U . S and U are adjacent. A(+ik−1 . . if a subset R is contained in two distinct apartments of Gk (V ) then f (R) is contained in their images which are distinct apartments of Gk (V ). −j) are as required. Hence these complement subsets form a regular collection and the complement subsets A(+i1 . S and U are adjacent if and only if f (S) and f (U ) are adjacent. indeed. Let 1 < k ≤ n − k and f : Gk (V ) → Gk (V ) be an apartments preserving mapping. . Then f is adjacency preserving: two elements of Gk (V ) are adjacent if and only if their images are adjacent.4 that one of the following possibilities is realized: (A) i1 . −j1 ). . . . . .6.

. The mapping g is dimension preserving and we have S ⊂ U =⇒ g(S) ⊂ g(U ) for all S. Therefore. on the other hand. Hence it transfers stars and tops to subsets of stars or tops. By Proposition 3. the image of every maximal singular subspace of Gk (V ) is contained in precisely one maximal singular subspace of Gk (V ). or a line. f is an adjacency preserving injection.Classical Grassmannians 93 3. Suppose that the latter mapping is induced by a semilinear embedding u : V ∗ → V ∗ and consider the mapping g : G(V ) → G(V ) S → u(S 0 ) 0 . U ∈ G(V ). The restriction of g to Gk (V ) coincides with f . For every star there is an apartment intersecting this star in a set consisting of n − k + 1 elements. stars go to subsets of stars if k < n − k. Indeed. the restriction of g to G1 (V ) is induced by a semilinear injection l : V → V . .6.4. the intersections of tops with apartments contain at most k + 1 elements. Since this restriction is apartments preserving. . By the same reason. every apartments preserving mapping f : Gk (V ) → Gk (V ) can be identiﬁed with an apartments preserving mapping of Gn−k (V ∗ ) to Gn−k (V ∗ ). Then l(S) = l(P1 ) + · · · + l(Pm ) = g(P1 ) + · · · + g(Pm ) ⊂ g(S) which implies that l(S) coincides g(S) (since the dimension of these linear subspaces is equal to m). the intersection of two distinct maximal singular subspaces is empty.3 Proof of Theorem 3. l is a semilinear embedding. Since k ≤ n − k. Thus g is induced by l and f = (l)k . m = dim S such that S = P1 + · · · + Pm . In particular. the image of a star can be a subset of a top only in the case when n = 2k. . Pm ∈ G1 (S).10: Second step By duality. By Theorem 1. Let 1 < k ≤ n − k and f be an apartments preserving mapping of Gk (V ) to Gk (V ). For every linear subspace S ⊂ V we choose P1 . thus it intersects apartments in subsets containing at most two elements. . or a single point. we can prove the theorem only in the case when k ≤ n − k.4.

The latter intersection contains more that one element which implies that fi (U ) coincides with l(U ) . U2 ∈ Ak and fk−1 (S) = f (U1 ) ∩ f (U2 ) ∈ Ak−1 . we can identify f with an apartments preserving mapping of Gk (V ) to Gk (V ∗ ) which sends stars to subsets of stars. We prove that fi = (l)i induction by i. Every S ∈ Ak−1 is the intersection of two adjacent U1 . Step by step. The latter mapping is induced by a semilinear embedding of V in V ∗ . we construct a sequence of apartments preserving mappings fi : Gi (V ) → Gi (V ). Then.2. . l is a semilinear embedding. fi ([S i ) ⊂ [fi−1 (S) i for all S ∈ Gi−1 (V ) and fi−1 ( U ]i−1 ) ⊂ fi (U )]i−1 for all U ∈ Gi (V ) if i > 1. . Since the mapping fk−1 is injective and |Ak−1 | = |Ak−1 | < ∞.2). Since n = 2k. we get fk−1 (Ak−1 ) = Ak−1 .4. respectively. By Theorem 1. f induces an injective mapping fk−1 : Gk−1 (V ) → Gk−1 (V ) such that f ([S k ) ⊂ [fk−1 (S) k for all S ∈ Gk−1 (V ). Let B be one of the bases of V associated with the apartment Ak := f (Ak ). . i = k. f1 is induced by a semilinear injection l : V → V . In the case when the image of a certain star of Gk (V ) is contained in a top.94 Grassmannians of Classical Buildings Distinct maximal singular subspaces of Gk (V ) go to subsets of distinct maximal singular subspaces of Gk (V ) (otherwise there exist non-adjacent elements of Gk (V ) whose images are adjacent). . Let also Ak−1 ⊂ Gk−1 (V ) and Ak−1 ⊂ Gk−1 (V ) be the apartments deﬁned by the bases B and B . Since f1 is apartments preserving. we show that each star goes to a subset of a top (the standard arguments from the proof of Theorem 3. Consider a base B ⊂ V and the associated apartment Ak ⊂ Gk (V ). 1. as in the proof of Theorem 3. Now we show that fk−1 is apartments preserving. . If fi−1 is induced by l then fi−1 ( U ]i−1 ) ⊂ fi (U )]i−1 ∩ l(U ) ]i−1 for every U ∈ Gi (V ). such that fk = f . Suppose that f transfers stars to subsets of stars.

Let f be a semicollineation of Π to Π . Let Π = (P. we have ∞ ∞ (3.12 ([Kreuzer (1996)]).5 3. Consider a subset X ⊂ P containing at least two points. Proof. i=1 Therefore. Proof. and any two bases of an exchange space have the same cardinality. we prove induction by i that f ([X]i ) ⊂ [f (X)]i (Section 2. Π is an exchange space if for every subset X ⊂ P and any points p. L) satisﬁes the exchange axiom. L) and Π = (P . First. The same holds for exchange spaces.5. if the inclusion holds for i = k − 1 then f ([X]k ) = f ([[X]k−1 ]1 ) ⊂ [f ([X]k−1 )]1 ⊂ [[f (X)]k−1 ]1 = [f (X)]k . This axiom holds for projective and aﬃne spaces. See §8 in [Karzel. L ) be ﬁnite-dimensional exchange spaces such that dim Π ≤ dim Π .5) f( X ) = i=1 f ([X]i ) ⊂ [f (X)]i = f (X) . q ∈ P \ X p ∈ X. For i = 1 this is trivial. o Every semicollineation between projective spaces of the same ﬁnite dimension is a collineation (Corollary 1.5). p .Classical Grassmannians 95 3. q =⇒ q ∈ X.6.3). Theorem 3.1 Grassmannians of exchange spaces Exchange spaces We say that a linear space Π = (P. By Proposition 2. S¨rensen and Windelberg (1973)]. or simple.11. if B is a base of Π then f ( B ) = f (P ) = P ⊂ f (B) . Then every semicollineation of Π to Π is a collineation. Theorem 3. In an exchange space every independent subset can be extended to a base of this space.

Let S be a plane in a 3-dimensional projective space Π = (P. L ) is an exchange plane.8 ([Kreuzer (1996)]). In the case when f is bijective. Example 3. A more complicated example of a semicollineation of a 4dimensional projective space to a non-Desarguesian projective plane was given in [Ceccherini (1967)]. Thus our spaces have the same dimension and f transfers bases of Π to bases of Π . the inverse mapping is a semicollineation of Π to Π. L) and Π = (P . This implies that f maps any triple of non-collinear points to non-collinear points (since three distinct points are non-collinear if and only if there is a base containing them. Consider the following set of lines L := {S} ∪ { L ∈ L : L ⊂ S }.96 Grassmannians of Classical Buildings and Π is spanned by f (B). hence f −1 is a semicollineation of Π to Π.5) in Theorem 3. Corollary 3. The following example shows that the condition (3. Then f −1 sends triples of collinear points to collinear points.12 gives the following. 3.3. in other words. Let Π = (P.5) guarantees that f (B) is a base of Π .12 cannot be dropped. The inequality (3. Then f is injective and transfers any triple of non-collinear points to non-collinear points (standard arguments). An easy veriﬁcation shows that Π := (P.11).2 Grassmannians Let Π = (P. For every number k ∈ {0. .5. Theorem 3. L) be an exchange space of ﬁnite dimension n. Now suppose that our exchange spaces have the same ﬁnite dimension and consider a mapping f : P → P which sends bases of Π to bases of Π . .11. n−1} we denote by Gk (Π) the Grassmannian consisting of all k-dimensional subspaces of Π.11 that any . The identity transformation of P is a semicollineation of Π to Π which is not a collineation. . L). . It follows from Theorem 3. we remove all lines contained in S and add S as a line. Then every bijection of P to P sending bases of Π to bases of Π is a collineation of Π to Π . Theorem 3. Remark 3. L ) be exchange spaces of the same ﬁnite dimension.

U ∈ Gk (Π) there is a base subset of Gk (Π) containing S and U .12.7. 1.7. and X ∪ {p. If S. Proof. X ∪ {q}. Proposition 3. and S. Similar to the proof of Proposition 3. U = 2k − dim(S ∩ U ). is Proof. the associated base subset of Gk (Π) contains S and U . Remark 3. Lemma 3. There exists an independent subset Y ⊂ P such that S and U are spanned by subsets of Y (see the proof of Lemma 3. The Grassmann graph Γk (Π) is the graph whose vertex set is Gk (Π) and whose edges are pairs of adjacent subspaces. U ∈ Gk (Π) are adjacent then the subspace S. Consider a certain base B of Π.2.4). For index one Grassmann graphs we have an analogue of Chow’s theorem (Theorem 3. see [Pankov (2006)]. It is not diﬃcult to prove that this graph is connected and every star is a maximal clique (stars and tops are deﬁned as for Grassmannians of ﬁnite-dimensional vector spaces). note that for an arbitrary linear space the same holds only in the case when k = 0. Every maximal clique of Γk (Π) is a star or a subset of a top. Remark 3. We take any base X of S ∩ U and any points p ∈ S \ U and q ∈ U \ S. Two k-dimensional subspaces S and U are contained in a base subset of Gk (Π) if and only if dim S.13. For any adjacent S. It was noted in Subsection 3. U (k + 1)-dimensional.6). Proof. Lemma 3. q} are bases of S. U . The set consisting of all k-dimensional subspaces spanned by subsets of B is called the base subset of Gk (Π) associated with (deﬁned by) the base B.6. For larger indices such kind results are unknown. The exchange axiom guarantees that X ∪ {p}. In the case when Π is the projective space of an (n + 1)-dimensional vector space V . the Grassmannian Gk (Π) can be identiﬁed with Gk+1 (V ) and base subsets are apartments.2 that there exist pairs of non-adjacent elements of Gk (Π) which span (k + 1)-dimensional subspaces. .2. Two elements of the Grassmannian Gk (Π) are said to be adjacent if their intersection is (k − 1)-dimensional. U (respectively).Classical Grassmannians 97 two incident k-dimensional subspaces of Π are coincident. We extend Y to a base of Π. note that the exchange axiom is not required.

.98 Grassmannians of Classical Buildings see [Pankov (2006)]. Theorem 3. inexact.4): (1) m = k and i1 . . we establish the direct analogue of Lemma 3. respectively. We write B(+i) and B(−i) for the sets consisting of all elements of B which contain pi and do not contain pi . . Lemma 3. . We deﬁne exact. . . i = j. B(+im+1 . In the general (non-projective) case. . . Consider a base B = {p1 . Let 0 < k < n − 1. . . .2: a subset R ⊂ B is exact if and only if for every i the intersection of all S ∈ R containing pi coincides with pi . jk+1 are distinct. L) and Π = (P . ik+1 or j1 . . . . ik+1 are distinct. We say that m + 1 distinct complement subsets B(+i1 . Using the fact that every line contains at least three points. this formula does not hold and there exist pairs of k-dimensional subspaces which are not contained in base subsets of Gk (Π). −jm+1 ) form a regular collection if their intersection is a one-element set. one of the following three possibilities is realized (an analogue of Lemma 3.13 ([Pankov (2006)]). pn+1 } of Π and the associated base subset B ⊂ Gk (Π). (2) m = k = n − k − 1 and i1 . n − k − 1}. . . (3) m = n − k − 1 and j1 . Let Π = (P. . The case k = 0 is trivial (any two distinct points are adjacent). L ) be exchange spaces of the same ﬁnite dimension n ≥ 2 such that each line of Π and Π contains at least 3 points. In this case. . . . jn−k are distinct. Suppose that f : Gk (Π) → Gk (Π ) is an injection sending base subsets to base subsets. .3) and show that every complement subset of B is B(+i. . . If k = n − 1 then the required statement is a direct consequence of the following observation: two distinct (n − 1)-dimensional subspaces of Π or Π are adjacent if and only if there is a base subset containing them. and complement subsets of B as for apartments in Grassmannians of ﬁnite-dimensional vector spaces. Then f maps adjacent elements to adjacent elements. −j) := B(+i) ∩ B(−j). −j1 ). Now deﬁne m := min{k. Proof. We describe maximal inexact subsets of B (cf. . .

f and f −1 map maximal cliques to maximal cliques. In the case when 2k + 1 < n. see Example 3. there are pairs of k-dimensional subspaces which are not contained in base subsets of Gk (Π). Then f is an isomorphism of Γk (Π) to Γk (Π ). it is induced by a collineation of Π to Π . Every maximal clique is a star or a subset of a top (Proposition 3. Lemma 3. Now consider adjacent k-dimensional subspaces S and U of Π. the pre-images of adjacent elements of Gk (Π ) need not to be adjacent. Therefore. L) and Π = (P . stars go to stars (for every star there is a base subset intersecting this star in a subset consisting of n − k + 1 elements. The injectivity of f guarantees that inexact subsets of B go to inexact subsets of f (B).14. in other words. there is a base subset B ⊂ Gk (Π) containing them. f (S) and f (U ) are adjacent.7. We have the standard characterization of the adjacency relation in terms of complement subsets (cf. As in the proof of Theorem 3. In the general case. • there exists a regular collection of m distinct complement subsets of B such that each element of this collection contains S and U .5).7). Let Π = (P. Therefore. Remark 3. Theorem 3. By Theorem 3. By Lemma 3. the following conditions are equivalent: • S.Classical Grassmannians 99 A collection of m distinct complement subsets is said to be regular if it can be extended to a regular collection of m + 1 distinct complement subsets. The inverse mapping is induced by f −1 .10. f is an isomorphism of Γk (Π) to Γk (Π ). hence it also sends .9.14 ([Pankov (2006)]).13. Thus f induces a bijection fk−1 : Gk−1 (Π) → Gk−1 (Π ). thus we cannot assert that a mapping of Gk (Π) to Gk (Π ) sending base subsets to base subsets is injective. As in the proof of Theorem 3. the intersections of base subsets with tops contain at most k + 2 elements). in the case when 2k +1 < n. we establish that a subset of B is complement if and only if its image is a complement subset of f (B). Proof. By the same reason. two elements of B are adjacent if and only if the same holds for their images. L ) be as in the previous theorem and f be a bijection of Gk (Π) to Gk (Π ) such that f and f −1 map base subsets to base subsets. in particular. we show that fk−1 maps base subsets to base subsets. U ∈ B are adjacent.10.

Consider the bijection of L to L which sends every L ∈ L to the line of Π containing L (this mapping is induced by the natural embedding of Π in Π ). . where fk = f and each fi together with the inverse mapping send base subsets to base subsets. Remark 3. Remark 3. Example 3. Let also W be an (n − k)-dimensional linear subspace of V . 0. i = k. one of the following possibilities is realized: • f is induced by a collineation of Π to Π .9). . since it is not adjacency preserving (lines of Π passing through the point p go to non-intersecting lines of Π). we cannot apply our method to the case when 2k + 1 ≥ n. 3.14 then.3) and an easy veriﬁcation shows that f is induced by f0 . Since dual principle does not work for exchange spaces.15. The inverse mapping does not satisfy this condition. However. • Π and Π both are projective spaces and f is induced by a collineation of Π to the projective space dual to Π . There is a one-to-one correspondence between elements of X and all k × (n − k)-matrices over the associated division ring: we take any M ∈ X and ﬁx bases in M and W . . if n = 3. Suppose that Π is a projective space and Π is the linear space obtained from Π by removing a certain point p. then every .6 Matrix geometry and spine spaces Let V be an n-dimensional vector space over a division ring. k = 1 and f is as in Theorem 3. . Consider the set X ⊂ Gk (V ) consisting of all complements of W . This bijection transfers base subsets to base subsets.100 Grassmannians of Classical Buildings base subsets to base subsets. Every strong embedding of Π in Π (an embedding sending independent subsets to independent subsets) induces an injection of Gk (Π) to Gk (Π ) which transfers base subsets to base subsets. We construct a sequence of bijections fi : Gi (Π) → Gi (Π ). Using the arguments given above.16. It is not diﬃcult to prove that Π is an exchange space and dim Π = dim Π . Then f0 is a collineation of Π to Π (Corollary 3.4. In contrast to the projective case.9. by Theorem 3. we can show that every injection f : L → L transferring base subsets to base subsets is induced by a strong embedding of Π in Π [Pankov (2006)]. the assumption that f is bijective does not guarantee that this embedding is a collineation (Example 3.

For example. induces a line in the spine space (the intersection of [M. dim(N ∩ W ) = m + 1. dim(N ∩ W ) = m + 1. it is equal to the distance between the corresponding elements of X . If m is equal to dim W or k then our spine space is a parabolic subspace of Gk (V ).Classical Grassmannians 101 S ∈ X can be considered as the graph of a linear mapping l : M → W . the geometry of rectangular matrices (m = 0 and dim W = n − k) contains only τ -lines. α.m (W ) contains at least two points). In the general case.m (W ) deﬁned by (associated with) the base B. The intersection of the associated apartment of Gk (V ) with Fk. (α) dim(M ∩ W ) = dim(N ∩ W ) = m. we get the geometry of rectangular matrices.m (W ) are contained in a base subset if and only if x ∈ {α.2.m (W ) is not empty). N ]k with Fk. ω}. Two x-adjacent elements of Fk.m (W ) the set consisting of all S ∈ Gk (V ) satisfying dim(W ∩ S) = m (we assume that m is not greater than dim W and k which guarantees that the set Fk. The distance between two matrices A and B is deﬁned as the rank of A− B. α. The associated line of the spine space is said to be an x-line. M ∈ Gk−1 (V ). if they belong to a certain x-line. x ∈ {τ. ω}. Partial linear spaces of such kind are called spine spaces. Let B be a base of V such that W is spanned by a subset of B. The restriction of the Grassmann space Gk (V ) to this set (Subsection 1.4). ω}. Two distinct elements of Fk. Suppose that 1 < k < n − 1 and the line [M. if the corresponding case is realized. (ω) dim(M ∩ W ) = m − 1. The construction considered above has a natural generalization [Pra˙ mowski (2001)]. Information concerning adjacency preserving transformations of the set of rectangular matrices can be found in [Wan (1996)] (Theorem 3. and we identify S with the matrix of this linear mapping in the ﬁxed bases. Every automorphism of the spine space can be extended to an automor˙ z phism of the Grassmann space Gk (V ) [Pra˙ mowski and Zynel (2002)].1) is a partial linear space.m (W ) is said to be the base subset of Fk. x ∈ {τ. N ∈ Gk+1 (V ). there are the following three possibilities: (τ ) dim(M ∩ W ) = m. . We ﬁx a linear subspace W ⊂ V and denote by z Fk.m (W ) are called x-adjacent. N ]k . In the case when m = 0 and dim W = n − k.

n − k)-involutions will be denoted by Ik. Since the intersection of these linear subspaces is 0 and the sum of their dimensions is n.1 Geometry of linear involutions Involutions and transvections Suppose that V is an n-dimensional left vector space over a division ring R. . Let u be an involution of the group GL(V ) (u2 = 1).7.2. inexact. respectively. Therefore.n−k (V ) is a maximal set of conjugate involutions. Lemma 3. Involutions u and v commute if and only if u(S+ (v)) = S+ (v) and u(S− (v)) = S− (v). n−k)-involution u (there exists l ∈ GL(V ) such that v = lul−1) if and only if v is a (k.n−k (V ). This set can be identiﬁed with Gk.8. The z proof of this statement is based on a characterization of the α-adjacency and ω-adjacency relations in terms of complement subsets (the deﬁnitions of exact. we get S+ (u) + S− (u) = V. 3. Pra˙ mowski and Zynel (2005)].7 3. The set of all (k. U ) ∈ Gk (V ) × Gn−k (V ) : S + U = V }.102 Grassmannians of Classical Buildings In almost all cases. Exercise 3. n− k)-involution if the dimension of S+ (u) and S− (u) is equal to k and n − k.n−k (V ) := { (S. n−k)-involution. and complement subsets are standard). Hint: there exist unique vectors x+ ∈ S+ (u) and x− ∈ S− (u) such that x = x+ + x− . Let u be an involution of GL(V ) and x be a vector satisfying u(x) ∈ x . We say that u is a (k. It is clear that v ∈ GL(V ) is conjugate to a (k. each Ik. We have u(x) = x if x ∈ S+ (u) and u(x) = −x if x ∈ S− (u). Show that x belongs to S+ (u) or S− (u). the characteristic of R is not equal to 2 and n ≥ 3. Consider the linear subspaces S+ (u) := Ker (1 − u) and S− (u) := Im (1 − u).m (W ) preserving the class of base subsets preserves the α-adjacency and ω-adjacency re˙ lations or interchanges them [Pankov. every bijective transformation of Fk.

. Let B be a base of V .n−k (V ) associated with (deﬁned by) B is formed by all (k. Then uv(x) = vu(x) for every x belonging to S+ (v) or S− (v). The base subset of Ik. denote this involution by v . . Similarly. the latter equality holds for all x ∈ V .n−k (V ) there exists a collection of linear subspaces S1 . n − k)-involutions u such that S+ (u) and S− (u) both are spanned by subsets of the base B. Sm satisfying the following conditions: • S1 + · · · + Sm = V .8. we show that the linear subspace S− (v) is invariant for u. The class of maximal commutative subsets of Ik. If x ∈ S+ (v) then vu(x) = uv(x) = u(x) and u(x) ∈ S+ (v).n−k (V ) coincides with the class of base subsets.8. suppose that the linear subspaces S+ (v) and S− (v) both are invariant for u. The same arguments show that for every commutative subset X of Ik. By Lemma 3.Classical Grassmannians 103 Proof. Similarly. It is clear that S+ (v ) = S+ (v) ∩ S+ (u) and S− (v ) = S− (v) ∩ S+ (u). . . hence S+ (v) ∩ S+ (u) + S− (v) ∩ S+ (u) = S+ (u). Let u and v be commutative involutions. Suppose that uv = vu. • Si ∩ Sj = 0 if i = j. Proposition 3. • for every u ∈ X the linear subspaces S+ (u) and S− (u) can be presented as the sums of some Si . any two elements of this subset commute. Since V is the sum of these linear subspaces. . we get the equality S+ (v) ∩ S− (u) + S− (v) ∩ S− (u) = S− (u) which implies that S+ (v) ∩ S+ (u) + S− (v) ∩ S+ (u) + S+ (v) ∩ S− (u) + S− (v) ∩ S− (u) = V. Conversely. Proof. Then v(S+ (u)) = S+ (u) and the restriction of v to the linear subspace S+ (u) is an involution.

(S .2 Adjacency relation Two distinct elements (S.18.n−k (V ) is the graph whose vertex set is Gk.n−k (V ). Then the dimension of Im (1 − t) is equal to 1 and t(x) = x + α(x)x0 . 1 + α(x0 ) If t is a transvection then t−1 (x) = x − α(x)x0 also is a transvection. Let t ∈ GL(V ).104 Grassmannians of Classical Buildings This means that every commutative subset of Ik. . Suppose that Ker (1 − t) is (n − 1)-dimensional. For any distinct S.n−k (V ) are said to be adjacent if one of the following possibilities is realized: • S = S and U is adjacent with U in Gn−k (V ).n−k (V ) and whose edges are pairs of adjacent elements.17. where x0 is a non-zero vector and α is a linear functional whose kernel coincides with Ker (1 − t). S ∈ Gk (V ) there exists U ∈ Gn−k (V ) which is a complement to both S and S . x0 ∈ Ker α) then t is said to be a transvection. If l ∈ GL(V ) and t commute then l(Ker (1 − t)) = Ker (1 − t) and l(Im (1 − t)) = Im (1 − t).7. • U = U and S is adjacent with S in Gk (V ). Lemma 3. 3. If Im (1 − t) ⊂ Ker (1 − t) (in other words.9. Remark 3. U ). The Grassmann graph Γk. The subgroup of GL(V ) generated by all transvections is denoted by SL(V ). We use the following lemma to show that this graph is connected.n−k (V ) is contained in a certain base subset and every base subset is a maximal commutative subset of Ik. but these equalities do not guarantee that lt = tl. U ) ∈ Gk. Note that α(x0 ) = −1 (since t(x0 ) = 0) and t−1 (x) = x − α(x) x0 . Remark 3.

(S . If T is a complement to S + S then U := T + x1 + y1 . U ) is adjacent with (S. we take any (n − k)-dimensional linear subspace U satisfying S+U =S +U =V (Lemma 3. ym . S ) = d(S. .9. . S ) − 1. S ) = 1. U ) be distinct elements of Gk. U ) with (S . . . Let (S. . U ) with (S .9) and construct a path joining (S. . Then W + U = V and S is not contained in W + U . U ). This implies the existence of a 1-dimensional linear subspace P ⊂ S such that S := W +P is a complement of U . This is trivial if d(S. . S = X. . . By the inductive hypothesis. S = X. . .n−k (V ) contains a path connecting (S. U ).n−k (V ) is connected. . xm . there exists a path in Γk. .Classical Grassmannians 105 Proof. The graph Γk. (S . . .n−k (V ) connecting (S . y1 . First. U ). U ). U ). . U ) and (S . U ). Suppose that d(S.n−k (V ). we consider the case when U = U and prove induction by the distance d(S. We take any (k − 1)-dimensional linear subspace W satisfying S ∩ S ⊂ W ⊂ S. . S ) = k − dim(S ∩ S ) that the graph Γk. We have d(S . S ) ≥ 2. Proof. In the general case. . xm . ym } such that S∩S = X . U ). . Consider an independent subset X ∪ {x1 . Proposition 3. (S. Since (S . The case when S = S is similar. . we get the claim. y1 . xm + ym is as required. . x1 .

Then Ker (1 − uv) = Ker (1 − vu) = S+ (u) + (S− (u) ∩ S− (v)) (3. N ∈ Gn−k+1 (V ) and S + N = V . Suppose that S+ (u) = S+ (v) (the case when S− (u) = S− (v) is similar).1 (V ) is of type [V. Proposition 3. P ]n−1. (2) uv and vu are transvections. N ]k.1 . H ∈ Gn−1 (V ). U ]k.n−k . U ]k. (1) =⇒ (2). M ∈ Gn−k−1 (V ) and S ∩ M = 0. every maximal clique of Γn−1. T ∈ Gk+1 (V ). There exist a linear functional α whose kernel coincides with (3.n−k (V ) such that S and U are incident to M and N . H]1.n−k . N ]k. V ]n−1. We say that two (k.n−k (V ) are adjacent.106 Grassmannians of Classical Buildings For linear subspaces M and N we denote by [M.n−k (V ) the following conditions are equivalent: (1) u and v are adjacent. respectively. If 1 < k < n − 1 then the statement easy follows from Proposition 3. [Q. U ∈ Gn−k (V ) and Q ∩ U = 0. U ) ∈ Gk. There is the following algebraic interpretation of this adjacency relation. Q ∈ Gk−1 (V ). The case when k = 1. S ∈ Gk (V ). [S.6) is (n − 1)-dimensional. For involutions u. H ∈ Gn−1 (V ).n−1 . P ∈ G1 (V ) or [V. Proof.8) (3. the Grassmann graph Γk.n−k the set of all (S.n−1 (V ) is of type [P. M ]k.n−k . V ]1. n − 1 is trivial. Proof. n − k)-involutions are adjacent if the corresponding elements of Gk. v ∈ Ik.n−k (V ) admits only the following four types of maximal cliques: • • • • [S.1 . P ∈ G1 (V ) or [H. Every maximal clique of Γ1.6) and a non-zero vector x0 such that uv(x) = x + α(x)x0 . Similarly. Proposition 3.n−1 .2. Then v(x) = u(x) + α(x)u(x0 ) (3. [T.7) . U ∈ Gn−k (V ) and T + U = V .10.n−k . S ∈ Gk (V ). In the case when 1 < k < n − 1.11.

As above. Then (3. This implies that S+ (v) ⊂ Ker α. (3. we get S− (v) ⊂ Ker α which implies that S− (v) coincides with S− (u) and S+ (v) is adjacent with S+ (u).8) implies that v(x0 ) ∈ x0 . Suppose that x0 ∈ S+ (u). we have S+ (v) = S+ (u) and S− (v) ∩ (Ker α) = S− (u) ∩ (Ker α) is (n − k − 1)-dimensional. Then two distinct (k. v are adjacent if and only if for any distinct u . then vu also is a transvection (Remark 3. n − 1. x0 belongs to S+ (u) or S− (u). (2) =⇒ (1). uv is a transvection. Therefore. Suppose that x0 ∈ S− (u). Using Exercise 3.Classical Grassmannians 107 and x = v 2 (x) = u(u(x) + α(x)u(x0 )) + α(u(x) + α(x)u(x0 ))u(x0 ) = x + α(x)x0 + α(u(x) + α(x)u(x0 ))u(x0 ) which means that u(x0 ) ∈ x0 . If X c is non-empty then X cc contains X . v ∈ {u. For every x ∈ S+ (v) u(x) = uv(x) = x + α(x)x0 and x = u2 (x) = u(x) + α(x)u(x0 ) = x + 2α(x)x0 .6) which coincides with Ker α. Since u(x) = v(x) for every x ∈ Ker α. Lemma 3.2 we show that x0 belongs to the linear subspace (3. Suppose that k = 1. For every x ∈ S− (v) u(x) = −uv(x) = −x − α(x)x0 .17). n − k)-involutions commuting with every element of X .10 ([Mackey (1942)]). By Exercise 3.2. we establish that u(x0 ) ∈ x0 .n−k (V ). n−k)-involutions u. If uv is a transvection then (3.9) . v}cc we have {u. v}cc = {u .7) holds for a certain linear functional α and a non-zero vector x0 ∈ Ker α. Let X be a non-empty subset of Ik. v }cc . As in the previous case. Denote by X c the set consisting of all (k.

In the case when P = P and U = U . S− (s)) ∈ Gk.n−k (V ).11. Example 3. Is there a characterization of the adjacency relation in terms of the commutativity relation for the case when 1 < k < n − 1? 3. By Lemma 3. U ) → (s−1 (U 0 ). n − k)-involution s is identiﬁed with (S+ (s). For every semilinear automorphism l : V → V the mapping u → lul−1 ∀ u ∈ GL(V ) is an automorphism of the group GL(V ). The associated transformation of Gk. v}cc are adjacent and (3. P + P ]1. v ∈ {u. This implies that {u. U ) and v = (P . U ) commute with u if and only if P ⊂ U and P ⊂ U. moreover. v ∈ {u.10. U ∩ U ]1. If u is a (k.3 Chow’s theorem for linear involutions We give a few examples. If u and v are adjacent (P = P or U = U ) then any distinct u . Recall that the contragradient mapping u → u is an isoˇ ∗ morphism of GL(V ) to GL(V ).n−k (V ) is (S.n−k (V ) to itself. It transfers each Ik. Example 3. n − k)-involution then the same holds for u.1. Problem 3. s−1 (S 0 )). l(U )). We restrict ourselves to the case when k = 1 (the case k = n − 1 is similar).8. a (1.7. It transfers each Ik. v}c = [U ∩ U . Recall that every (k.n−k (V ) is (S. Let u = (P.9) fails for any adjacent u . n − 1)-involution (P . For every semilinear isomorphism s : V → V ∗ the mapping u → s−1 us ∀ u ∈ GL(V ) ˇ is an automorphism of the group GL(V ). U ) → (l(S).n−1 . the equality (3. v}cc = [P + P . v}cc.9) holds.n−1 and {u.n−k (V ) to itself.108 Grassmannians of Classical Buildings Proof. The associated transformation of Gk. ˇ u u S+ (ˇ) = (S− (u))0 and S− (ˇ) = (S+ (u))0 (an exercise for the reader). . U ).

12. Let n = 2k. U ) to (U. In the case when n = 2k. S).k (V ) satisfying the following condition: u ∈ X =⇒ −u ∈ X . It is clear that all transformations considered above are commutativity preserving and the associated transformations of Gk. but the associated transformation of Gk. In the case when n = 2k. U ) to (U.15 ([Havlicek and Pankov (2005)]).k (V ) (we leave the details for the reader). this is equivalent to the fact that f and f −1 transfer base subsets to base subsets.k (V ) is not an automorphism of Γk.k (V ).10) or a semilinear isomorphism of V to V ∗ (Example 3. Some generalizations of this result were obtained in ˙ [Pra˙ mowski and Zynel (2009)].k (V ).19.k (V ) which do not induce automorphisms of Γk.k (V ) sends (S. by Proposition 3. If n = 2k then every automorphism of Γk.11).k (V ) induced by a semilinear automorphism of V or a semilinear isomorphism of V to V ∗ and i is the transformation sending every (S. z .10 and 3.13. Remark 3. the mapping u → −u is a bijective transformation of Ik.Classical Grassmannians 109 Automorphisms of the group GL(V ) considered in Examples 3. Example 3. where g is the automorphism of Γk. Consider any proper subset X ⊂ Ik. S). The transformation u→ −u u ∀u∈X ∀u∈X is commutativity preserving.k (V ).11 will be called regular. Theorem 3. The associated transformation of Gk.n−k (V ) is induced by a semilinear automorphism of V (Example 3.n−k (V ) is called commutativity preserving if f and f −1 map commutative involutions to commutative involutions.8.n−k (V ). Let n = 2k and f be an automorphism of Γk.n−k (V ) are automorphisms of the Grassmann graph Γk. A bijective transformation f of Ik. Then f = ig. Example 3. there exist commutativity preserving transformations of Ik.

V ]k.20.n−k ). 1 < k < n − 1.n−k ) = [W.n−k .15 gives the claim.n−k .4 ([Dieudonn´ 1 (1951).21.9. By Proposition 3. Let S ∈ Gk (V ).n−k (V ). 3. U ]k. the class of special subsets coincides with the class of maximal cliques of Γk.n−k (V ) can be extended to a regular automorphism of the group GL(V ). . V ]k. In the case e when k = 1. there exist W ∈ Gk (V ) and Z ∈ Gn−k+1 (V ) such that W + Z = V and f ([S. T ]k. Our ﬁrst step is to show that f and f −1 map special subsets to special subsets.n−k (V ) (Propositions 3. the associated transformation of Gk. Theorem 3. Assume that our cliques are of the same type. By Lemma 3.n−k (V ) is an automorphism of Γk. Now we show that f ((S. Suppose that 1 < k < n − 1.n−k (V ) and the same holds for f ([S.110 Grassmannians of Classical Buildings Remark 3.10. n − 1.10 and Theorem 3. T ]k. 3. Remark 3.7. In what follows the subsets [S.10. Proof. In the cases when k = 1. U ) ∈ [S.n−k (V ).n−k (V ) and the statement is trivial. It must be pointed out that all results concerning the adjacency relation on Gk.4 Proof of Theorem 3. n − 1 and f be a commutativity preserving bijective transformation of Ik. T ]k.n−k (V ). Let k = 1.15 Let f be an automorphism of the graph Γk. U ∈ Gn−k (V ) will be called special. S ∈ Gk (V ) and [V.n−k . Rickart (1950)]).15) hold in the case of an arbitrary characteristic. one of the four possibilities is realized. Z]k.n−k is a maximal clique of Γk. U )) ∈ [W. n−1. every commutativity preserving bijective transformation of Ik. were investigated in [Pankov (2005)]. Consider T ∈ Gn−k+1 (V ) satisfying S + T = V . in other words. V ]k. Corollary 3. Then [S.n−k (V ). Commutativity preserving transformations of Ik.n−k ∀ (S.n−k .

n−k and f ([S. Then the linear subspace P := U ∩ U0 is (n − k − 1)-dimensional and [P. Since for any distinct Si ∈ Gk (V ) (i = 1. the same holds for every (S.n−k (V ). We apply the same arguments to the mapping f −1 and get the inverse inclusion. V ]k. f ([S. V ]k. the proof is similar.n−k and suppose that (S. Since f ([S.n−k ). By the connectedness of the graph Γk. the linear subspace W is the ﬁrst component in every element of f ([S. P ]k.n−k is a special subset. T ]k. Thus the intersection of [S. U )) ∈ [W. U )) = (f (S). U ) is adjacent with (S. In the case when f ([S. T ]k.n−k ) and f ((S.n−k contains more than one element and the same holds for their images f ([S.n−k and [S. U ) ∈ [S. .Classical Grassmannians 111 Let us take any (S. T ]k. we show that for every U ∈ Gn−k (V ) the image of the special subset [V. The intersection of two distinct special subsets is non-empty if and only if these special subsets are of diﬀerent types and the sum of the associated k-dimensional and (n−k)-dimensional linear subspaces coincides with V (it is clear that this intersection is a one-element set).n−k ) = [W.n−k .n−k ) is a maximal clique of one of the remaining three types.9). Similarly. (B) f changes the types of all special subsets. U0 ). U ]k.n−k (V ). T ]k.n−k .n−k ) ⊂ [W. 2) there exists U ∈ Gn−k (V ) such that Si + U = V (Lemma 3. one of the following possibilities is realized: (A) f preserves the type of every special subset. U ) ∈ Gk. V ]k.n−k ) is a maximal clique of Γk. P ]k. There exist bijective transformations f and f of Gk (V ) and Gn−k (V ) (respectively) such that f ((S. T ]n−k is a line of the Grassmann space Gn−k (V ). U0 ) ∈ [S. V ]k. Z]k. Case (A). Thus for any S ∈ Gk (V ) and U ∈ Gn−k (V ) S + U = V ⇐⇒ f (S) + f (U ) = V. Therefore. P ]k.n−k (V ). f (U )) ∀ (S. This line contains at least three elements and only one of them has a non-zero intersection with S.n−k . P ]k.

U ) ∈ Gk. . If n = 2k and f is induced by a semilinear isomorphism u : V → V ∗ then the same arguments show that f also is induced by u. S ∈ Gk (V ) and U ∈ Gn−k (V ) are incident if and only if f (S) and f (U ) are incident. If 1 < k < n − 1 then each of these mappings is induced by a semilinear automorphism of V or a semilinear isomorphism of V to V ∗ (the second possibility is realized only in the case when n = 2k). Then 0 = (sl−1 (V ))0 = k+1 i=1 k+1 ∀ U ∈ Gk (V ). U ) → (u(U )0 . thus f is the composition of the transformations (S. we establish that (sl−1 (U ))0 = U Let W ∈ Gk−1 (V ). Suppose that f is induced by a semilinear automorphism l : V → V . suppose that n = 2k and f is induced by a semilinear isomorphism s : V → V ∗ . . U ) → (U. this implies that l−1 (s(U )) = U for every U ∈ Gn−k (V ) and f ((S. . Uk+1 ∈ Gk (V ) such that U1 ∩ · · · ∩ Uk+1 = W and U1 + · · · + Uk+1 = V. U )) = (l(S). As above. .112 Grassmannians of Classical Buildings The mappings f and f are automorphisms of the Grassmann graphs Γk (V ) and Γn−k (V ). n − 1. In this case. f and f are induced by the same semilinear automorphism of V . Now. Suppose that k = 1. If f is induced by a semilinear automorphism s : V → V then for any S ∈ Gk (V ) and U ∈ Gn−k (V ) S + U = V ⇐⇒ S + l−1 s(U ) = V . n − 1. (sl−1 (Ui ))0 = Ui = W.n−k (V ). and it follows immediately from Theorem 3. There exist bijective mappings g : Gk (V ) → Gn−k (V ) and g : Gn−k (V ) → Gk (V ) . Case (B). i=1 The equality W = 0 implies that k = 1 which contradicts the assumption that n = 2k ≥ 3. respectively (this statement is trivial if k = 1. By Proposition 3. u(S)0 ) and (S.7 if 1 < k < n − 1). l(U )) ∀ (S.4. S). We choose U1 .

U ) ∈ Gk. Rickart (1950)]). U ) → (l(S). (ii) n = 2k.5 Automorphisms of the group GL(V ) Theorem 3. v satisfying t = uv and get f (t) = g(t) in each of these cases.16 ([Dieudonn´ 1 (1951).n−m (V ).8. n − k. there exists a regular automorphism g of GL(V ) such that f (u) = g(u) ∀ u ∈ I1. moreover. we establish that one of the following possibilities is realized: (i) g and g are induced by the same similinear isomorphism u : V → V ∗ and f ((S. By Corollary 3. . For every transvection t we take adjacent (1.n−k (V ). the restriction of f to the set I1. U )) = (g (U ). Therefore.n−1 (V ) is a commutativity preserving bijection to I1.7.n−k (V ) can be characterized as a maximal set of conjugate involutions). U )) = (u(U )0 . g (S)) ∀ (S. respectively. it transfers Ik.n−k (V ) consists of n elements and k n k = n m ⇐⇒ m = k. n−1)-involutions u. U ) → (U. f and g are coincident on SL(V ) and h := g −1 f sends every element of SL(V ) to itself. Proof.n−k (V ).n−k (V ) or In−k.n−1 (V ) or In−1.n−m (V ) (because each Ik. base subsets of Ik. where g is a regular automorphism and α is a homomorphism of GL(V ) to the center of R. Since a base subset of Ik. 3. u(S)0 ) ∀ (S.1 (V ). S). U ) ∈ Gk. the image of Ik.Classical Grassmannians 113 such that f ((S. By Proposition 3.n−k (V ) to Im.n−k (V ) coincides with Ik. these mappings can be considered as bijections of Gk (V ) and Gn−k (V ) to Gk (V ∗ ) and Gn−k (V ∗ ). The automorphism f preserves the set of all involutions.n−k (V ) go to base subsets of Im. So. the mappings g and g are induced by the same similinear automorphism l : V → V and f is the composition of the transformations (S. As in the case (A).k (V ).4.n−1 (V ) or f (u) = −g(u) ∀ u ∈ I1. l(U )) and (S.n−1 (V ). If f is an e automorphism of the group GL(V ) then f (u) = α(u)g(u) ∀ u ∈ GL(V ). By duality.

codim S = α }.8.18. 3. All linear homothetic transformations of V form the center of the group GL(V ) (this subgroup is isomorphic to the center of R).8 Grassmannians of inﬁnite-dimensional vector spaces Let V and V be vector spaces over division rings. and Gα (V ) will be denoted by Γβ (V ). 3.1 Adjacency relation Let G be a Grassmannian of V . By Remark 3. The Grassmann graphs corresponding to Gβ (V ). The mapping l → α(l) is a homomorphism of GL(V ) to the center. hence it is a homothetic transformation. G β (V ). In this case. Throughout the section we suppose that dim V = dim V = α is inﬁnite. An easy veriﬁcation shows that f (l) = α(l)g(l). codim S = β } for every cardinality β < α and Gα (V ) = G α (V ) = { S ∈ G(V ) : dim S = codim S = α }. α(l) preserves every 1-dimensional linear subspace. We say that linear subspaces S. We deﬁne the associated Grassmann graph as in the ﬁnite-dimensional case. The Grassmann graph Γk (V ) is connected and every isomorphism of Γk (V ) to Γk (V ) is induced by a semilinear isomorphism of . U ∈ G are adjacent if dim(S/(S ∩ U )) = dim(U/(S ∩ U )) = 1 which is equivalent to dim((S + U )/S) = dim((S + U )/U ) = 1.114 Grassmannians of Classical Buildings If t is a transvection then ltl−1 is a transvection for any l ∈ GL(V ) and h(l)th(l−1 ) = ltl−1 which means that α(l) := l−1 h(l) commutes with every transvection. Let k ∈ N. we have the following three types of Grassmannians associated with V : Gβ (V ) = { S ∈ G(V ) : dim S = β. Γβ (V ). They are orbits of the action of the group GL(V ) on the set G(V ). Semilinear isomorphisms of V to V induce isomorphisms between the Grassmann graphs of the same indices. and Γα (V ) (respectively). G β (V ) = { S ∈ G(V ) : dim S = α.

Theorem 3.2. Let X be a connected component of Γβ (V ). Is there a semilinear automorphism l : V → V such that f (S) = l(S) for all S ∈ X ? There is the following inﬁnite-dimensional version of Theorem 3. We take any S ∈ X \ {S}. This mapping is an automorphism of Γβ (V ). We deﬁne f (X) := l(X) X ∈ X X X ∈ Gβ (V ) \ X . S2 ∈ G the following conditions are equivalent: (1) S1 and S2 are adjacent.17 ([Blunck and Havlicek (2005)]). Let l : V → V be a semilinear automorphism sending S to S .1. S2 } such that every complement of S is a complement to at least one of Si . The following example shows that the direct analogue of Chow’s theorem does not hold for Grassmannians formed by linear subspaces of inﬁnite dimension and codimension. Denote by X the connected component of Γβ (V ) containing S.Classical Grassmannians 115 V to V (the proof is similar to the proof of Theorem 3. Let S ∈ Gβ (V ) and β be inﬁnite. but it is not induced by a semilinear automorphism of V . Indeed.3). For any distinct S1 . the Grassmann graph Γk (V ) is connected. β < α. .2 and we leave it as an exercise for the reader). Problem 3. For every S ∈ Gα (V ) all complements of S belong to Gα (V ) (for other Grassmannians this fails) and we have the following. Since the annihilator mapping deﬁnes an isomorphism of Γk (V ) to Γk (V ∗ ) (see Subsection 1. and f be an automorphism of Γβ (V ). The associated automorphism of Γβ (V ) transfers X to itself.14 ([Blunck and Havlicek (2005)]). Example 3.7. it consists of all X ∈ Gβ (V ) such that dim(S/(S ∩ X)) = dim(X/(S ∩ X)) is ﬁnite. U ∩ S = 0 and f (U ) = U has a non-zero intersection with f (S) = S . Let G be a Grassmannian of V . (2) there exists S ∈ G \ {S1 . There exists U ∈ Gβ (V ) \ X such that U ∩ S = 0 and U ∩ S = 0.

If S ⊂ H ∈ G 1 (V ) then H + S = V and H contains a complement U of S.11) imply that S1 ⊂ S ⊂ S2 and every complement of S is not a complement of Si (i = 1. for example. 3.2 Proof of Theorem 3.17 If G = G1 (V ). i = 1. (3.6. Our next step is to show that Si ⊂ S. Suppose that f : Gα (V ) → Gα (V ) is a bijection preserving the complementary of subspaces: S ∈ Gα (V ) is a complement of U ∈ Gα (V ) if and only if f (S) is a complement of f (U ).8. we get S2 ⊂ S1 . The linear subspaces S1 and S2 are not incident. Suppose. If S1 and S2 are adjacent then any S ∈ G \ {S1 . we establish that for any 1-dimensional linear subspaces P1 ⊂ S1 . if S1 ⊂ S2 then (3. (2) =⇒ (1). S2 } satisfying S1 ∩ S2 ⊂ S ⊂ S1 + S2 is as required (see the proof of Theorem 3. 2).10) . As in the proof of Theorem 3. (1) =⇒ (2). G 1 (V ) then any two distinct elements of G are adjacent and the implication (2) =⇒ (1) is trivial.11) This inclusion is trivial if S1 + S2 = V .6). if S1 + S2 ⊂ H then U is not a complement of Si (i = 1. In the case when S1 + S2 = V . We have V = S1 + H ⊂ S + H. Now we show that S ⊂ S1 + S2 . 2) which contradicts (2). that S1 ⊂ S.5. there exists H ∈ G 1 (V ) which contains S2 and does not contain S1 . 2. (3. similarly. it is suﬃcient to show that S is contained in every H ∈ G 1 (V ) containing S1 + S2 (since the intersection of all such linear subspaces coincides with S1 + S2 ). Then f is an isomorphism of Γα (V ) to Γα (V ). Since S1 ⊂ S2 .10) and (3. Indeed. S2 } satisﬁes the condition (2). Suppose that S ∈ G \{S1 .116 Grassmannians of Classical Buildings Corollary 3. P2 ⊂ S2 the sum P1 + P2 has a non-zero intersection with S and S1 ∩ S2 ⊂ S.

hence it is a complement of S2 . As in the proof of Theorem 3. . S2 are contained in S).12) Let U be a complement of S1 + S2 . Using the equality (3. We have S1 + S2 ⊂ S + P1 + P2 = S + P + P1 = S + P1 ⊂ S1 + S2 which implies that S1 + S2 = S + P1 . We take any 2-dimensional linear subspace U ⊂ S1 and a 1-dimensional linear subspace of S2 which is not contained in S. U + (S1 + S2 ) = V = U + (S2 + P1 ) and the inclusion S2 + P1 ⊂ S1 + S2 implies that S1 + S2 = S2 + P1 . i = 1. 2. we establish that S1 + S2 = S1 + P2 . (3. So. S1 + S2 = S + P2 . By the same arguments. The linear subspace P1 + P2 intersects S in a 1-dimensional linear subspace P . U + P1 is a complement of S. Therefore.13). By (3. the linear subspace U + P1 is not a complement of S1 .12). This means that dim(S1 /(S ∩ S1 )) = 1 and the same holds for S2 . Since P1 ⊂ S1 .13) (3. Note that P1 = P2 (otherwise P1 = P2 ⊂ S1 ∩ S2 ⊂ S and both S1 . we establish that U has a non-zero intersection with S ∩ S1 .Classical Grassmannians 117 thus S + H = V and H contains a certain complement U of S. Then U is not a complement of S1 (S1 is a proper subspace of S) and it is not a complement of S2 (U and S2 both are contained in H). a contradiction. dim((S1 + S2 )/S1 ) = dim((S1 + S2 )/S2 ) = 1 which means that S1 and S2 are adjacent. Thus there exist 1-dimensional linear subspaces P1 ⊂ S1 and P2 ⊂ S2 such that Si = (S ∩ Si ) + Pi .6.

inexact. Then f is induced by a semilinear isomorphism of V to V . respectively. We also deﬁne B(+i. such that f and f −1 map base subsets to base subsets.18 ([Pankov 1 (2007)]). since dim V < dim V ∗ (Subsection 1. Base subsets of G 1 (V ) are not bases of the dual projective space Π∗ = ΠV ∗ . Let f be a bijection of Gβ (V ) to Gβ (V ).9. Let G be a Grassmannian of V distinct from G1 (V ) and G 1 (V ).18 If β = 1 then. respectively.4). We prove the following generalization of Theorem 3. −j) and B(+i . For any two elements of G there exists a base subsets containing them (Proposition 1. j ∈ I.1. The subsets B(+i) and B(−i) will be called simple subsets of ﬁrst and second type. −j ) one of the following possibilities is realized: .8. Exact. Let also B = {xi }i∈I be a base of V and B be the associated base subset of G. Every base subset of G1 (V ) is a base of the projective space ΠV and conversely. thus f is a collineation of ΠV to ΠV and the Fundamental Theorem of Projective Geometry gives the claim.8. as in Subsection 3. −j) := B(+i) ∩ B(−j) for all i.4 Proof of Theorem 3. We write B(+i) and B(−i) for the sets of all elements of B which contain xi and do not contain xi .3). we show that f and f −1 are semicollineations between the projective spaces associated with V and V . 3.1.4. +j) ∪ B(−i).3 Base subsets Let G be a Grassmannian of V and B be a base of V . Note that every complement subset is the intersection of two simple subsets of diﬀerent types.4. For two distinct complement subsets B(+i. Theorem 3. The base subset of G associated with (deﬁned by) B consists of all elements of G spanned by subsets of B. −j). and complement subsets of B are deﬁned as for apartments in Grassmannians of ﬁnite-dimensional vector spaces.2. As in Subsection 3. we establish that for every maximal inexact subset X ⊂ B there exist distinct i. V The mappings between Grassmannians of the same indices induced by semilinear isomorphisms of V to V transfer base subsets to base subsets. The associated complement subset is B(+i. +j) := B(+i) ∩ B(+j) and B(+i.118 Grassmannians of Classical Buildings 3. β < α. j ∈ I such that X = B(+i.

Let g : B → B be a special bijection. g(B(−i)) = B (+δ(i)) ∀ i ∈ I. g(B(−i)) = B (−δ(i)) or g(B(+i)) = B (−δ(i)).11. It is clear that maximal inexact subsets go to maximal inexact subsets in both directions. Thus g and g −1 map complement subsets to . Let B be the base subset of G associated with other base B = {xi }i∈I . our complement subsets are said to be adjacent. The simple subsets of B corresponding to i ∈ I will be denoted by B (+i) and B (−i). In other words. A collection of mutually adjacent complement subsets of B will be called an A-collection. An easy veriﬁcation shows that for two distinct complement subsets X . j } = ∅. −j)}j∈ I\{i} and {B(+j. moreover. Thus every simple subset of B can be characterized as the union of all complement subsets belonging to a certain maximal A-collection. j} ∩ {i . Y ⊂ B satisfying X ∩Y ⊂X ∩Y the inverse inclusion holds. the intersection of two distinct complement subsets is maximal if and only if these complement subsets are adjacent. Then g and g −1 send simple subsets to simple subsets. In the ﬁrst case. (2) i = j or j = i . For each i ∈ I {B(+i. (3) {i. −i)}j∈ I\{i} are maximal A-collections. there exists a bijective transformation δ : I → I such that g(B(+i)) = B (+δ(i)). ∀i∈I Proof. Y ⊂ B with a non-empty intersection the following two conditions are equivalent: • X and Y are adjacent.Classical Grassmannians 119 (1) i = i or j = j . Lemma 3. It is easy to see that every maximal A-collection is a collection of such kind. A bijection g : B → B is said to be special if g and g −1 map inexact subsets to inexact subsets. • for any distinct complement subsets X . then the intersection of the complement subsets is empty.

If g is a special bijection of the second type then S ∩ U = xi ⇐⇒ g(S) + g(U ) = B \ {xδ(i) } . S + U = B \ {xi } ⇐⇒ g(S) ∩ g(U ) = xδ(i) . . otherwise. Two distinct simple subsets are of diﬀerent types if and only if their intersection is empty or a complement subset. then G = G β (V ). The equality S ∩ U = xi implies that the dimension of S and U is not greater than the codimension. U ∈ B. If g is a special bijection of the ﬁrst type and G = G β (V ). the equality S + U = B \ {xi } implies that the dimension of S and U is α. hence G = Gβ (V ). If g is a special bijection of the ﬁrst type and G = Gβ (V ). U ∈ B(+i) and for every j ∈ I \ {i} we have S ∈ B(+j) or U ∈ B(+j). Let also S. and G = Gα (V ). Let g : B → B be a special bijection. U ∈ B. β ≤ α. Let S. This equality is equivalent to the fact that S. β ≤ α. U ∈ B(−i) and for every j ∈ I \ {i} we have S ∈ B(−j) or U ∈ B(−j). This equality is equivalent to the fact that S. then S + U = B \ {xi } ⇐⇒ g(S) + g(U ) = B \ {xδ(i) } . the adjacency relations of complement subsets is preserved and maximal A-collections go to maximal Acollections (in both directions). Since two distinct complement subsets are adjacent if and only if their intersection is maximal. β ≤ α. Since B(−i) = B \ B(+i).12. there exists a bijective transformation δ : I → I satisfying the required condition. We say that g is a special bijection of the ﬁrst type if it preserves the types of all simple subsets. hence they preserve the types of all simple subsets or change the type of every simple subset. then S ∩ U = xi ⇐⇒ g(S) ∩ g(U ) = xδ(i) .120 Grassmannians of Classical Buildings complement subsets. Let g and δ be as in the previous lemma. Then g and g −1 transfer simple subsets to simple subsets. This means that g and g −1 map simple subsets of diﬀerent types to simple subsets of different types. g is said to be a special bijection of the second type. Similarly. β ≤ α. Lemma 3. We have proved the following.

N ∈ B such that M + N = Si and M . Using Zorn lemma.14) belong to B and satisfy (3. For every linear subspace U ⊂ V we denote by [U ] the set of all elements of G incident with U . In the case when G = G β (V ).13. β ≤ α. i ∈ I.13: if Si := B \ {xi } . N. xi and N := N . The dual principles do not work and this statement cannot be obtained immediately from Lemma 3. Remark 3. Therefore. The linear subspaces M := M . N. the cardinality of X is equal to α. Consider the set of all X ⊂ B such that S ∩ X = 0.4). N . xj has a non-zero intersection with S and xj belongs to S + X ). This implies the existence of linear subspaces M . . S are contained in a base subset of G. Lemma 3. The associated base subset of G contains M. Then for every i ∈ I and every S ∈ [ xi ] \ B there exist M.14). • dim M = dim N = β − 1 if β is ﬁnite.22.12. xi (3. we need the following “dual” version of Lemma 3. It is not diﬃcult to prove that X ∪ Y is a base of V (see the proof of Proposition 1. special bijections of the second type exist only in the case when G = Gα (V ).Classical Grassmannians 121 By Lemma 3. S are contained in a base subset of G. N ∈ B such that M ∩ N = xi and M. N spanned by non-intersecting subsets of X and such that • dim M = dim N = β if β is inﬁnite. We take any base Y of S containing xi . Proof.13. Suppose that G = Gβ (V ). β ≤ α. then for every S ∈ [Si ] \ B there exist M.3). we establish the existence of a maximal subset X ⊂ B satisfying this condition (see the proof of Proposition 1. Then X is a complement of S (for every vector xj ∈ B \ X the linear subspace X. S.

f ([P ]) ⊂ [h(P )]. Then h(G1 (S)) = G1 (l(S)) for every linear subspace S ⊂ V and. Let also B be one of the bases of V associated with the base subset f (B). It follows from Lemma 3.122 Grassmannians of Classical Buildings Now we prove the theorem. h and h−1 send projective bases to projective bases. Therefore. Let f : Gβ (V ) → Gβ (V ). We write B1 and B1 for the bases of ΠV and ΠV deﬁned by B and B . by (3. The restriction of f to B is a special bijection to the base subset f (B). Let P ∈ G1 (V ). We have ˆ f (B ∩ [P ]) ⊂ [P ] for a certain P ∈ G1 (V ).12 guarantees that h(P ) = f (M ) ∩ f (N ) = P and f (S) belongs to [h(P )].12 that h(B1 ) = B1 . By Lemma 3. N. This means that h is induced by a semilinear isomorphism l : V → V . N ∈ B ∩ [P ] ˆ satisfying M ∩ N = P and a base subset B ⊂ Gβ (V ) containing M. respectively.15). S. Lemma 3. we have f (S) = l(S) for every S ∈ Gβ (V ). . We apply the same arguments to f −1 and establish the inverse inclusion. β < α. The mapping h is bijective (the inverse mapping is induced by f −1 ) and the latter equality implies that h(G1 (S)) = G1 (f (S)) ∀ S ∈ Gβ (V ). Lemma 3.13 implies the existence of M. be a bijection such that f and f −1 map base subsets to base subsets. So. So. there exists a mapping h : G1 (V ) → G1 (V ) satisfying f ([P ]) = [h(P )] ∀ P ∈ G1 (V ). this is a special bijection of ﬁrst type and there exists h(P ) ∈ G1 (V ) such that f (B ∩ [P ]) ⊂ [h(P )]. We take any base of V which contains a vector belonging to P and consider the associated base subset B ⊂ Gβ (V ). Let S ∈ [P ] \ B. (3.12.15) Let B be a base of V and B be the associated base subset of Gβ (V ).

. . • each (n− 2)-dimensional singular subspace is contained in precisely two maximal singular subspaces (type Dn ). all maximal singular subspaces of a polar space are projective spaces of the same ﬁnite dimension m (the number m + 1 is called the rank of the polar space). consisting of all k-dimensional singular subspaces of a rank n polar space Π. it will be shown that for every rank n polar space one of the following possibilities is realized: • each (n−2)-dimensional singular subspace is contained in at least three distinct maximal singular subspaces (type Cn ). Basic properties of polar spaces and their Grassmannians will be studied in Sections 4.1 and 4. n − 1}. we show that this deﬁnition is equivalent to the classical Tits–Veldkamp deﬁnition of polar spaces (Theorem 4. in particular.) and Buekenhout–Shult’s well-known property which says that a point is collinear with one or all points of a line.2. Using Teirlinck’s characterization of projective spaces. polar spaces of type D3 (these polar spaces are isomorphic to the index two Grassmann spaces of 4-dimensional vector spaces). .Chapter 4 Polar and Half-Spin Grassmannians All thick buildings of types Cn (n ≥ 3) and Dn (n ≥ 4) can be obtained from polar spaces of rank n. One of our main objects is the polar Grassmannian Gk (Π). polar spaces can be deﬁned as partial linear spaces satisfying some natural axioms (each line contains at least three points. 123 . . k ∈ {0.3 is dedicated to examples: polar spaces associated with reﬂexive sesquilinear forms and quadratic forms. . We do not consider polar spaces obtained from pseudo-quadratic forms and polar spaces of type C3 associated with Cayley algebras. 1. . In particular.1). Remarks concerning embeddings in projective spaces and classiﬁcation of polar spaces ﬁnish the section. there is no point collinear with all other points. By [Buekenhout and Shult (1974)].. Section 4.

k ≤ n − 3. δ. if our polar spaces are of type D4 then their half-spin Grassmann spaces (the Grassmann spaces of the half-spin Grassmannians) are polar spaces of type D4 and there are two additional possibilities: • the collineations of G1 (Π) to G1 (Π ) induced by collineations of Π to Gδ (Π ). D}. δ ∈ {+. However. As in the previous case. The Grassmannians of this building are the polar Grassmannians of Π. Suppose that Π is a polar space of type Dn . we require that n ≥ 4. δ ∈ {+. Characterizations of apartments in terms of the adjacency relation (analogues of Theorems 3.4 and 4. In Section 4. As in Example 2. We also consider collineations of Gδ (Π) to Gγ (Π ).6 holds for the half-spin Grassmannians. the polar space Π deﬁnes a thick building of type Dn . γ ∈ {+. the direct analogue of Theorem 3. The Crassmannians of this building are the polar Grassmannians Gk (Π). The Grassmann spaces of the half-spin Grassmannians are denoted by Gδ (Π).7 we give an example showing that the adjacency relation on Gn−1 (Π) cannot be characterized in terms of the opposite relation as in Theorem 3. In Section 4. in particular. In Section 4. −}. if our polar spaces are of type Dn . Then ∆(Π) is a building of type Cn .8). In almost all cases. Then Gn−1 (Π) can be naturally decomposed in two disjoint subsets called the half-spin Grassmannians and denoted by G+ (Π) and G− (Π). the Grassmann space associated with Gk (Π) is Gk (Π). The corresponding Grassmann spaces will be denoted by Gk (Π).9 we describe all apartments preserving mappings of polar and half-spin Grassmannians. in the case when X = D. .6 we investigate collineations of Gk (Π) to Gk (Π ).6. • the collineations of Gδ (Π) to Gγ (Π ).3. such collineations are induced by collineations of Π to Π . Let Π be a polar space of rank n and ∆(Π) be the simplicial complex consisting of all ﬂags formed by singular subspaces of Π. induced by collineations of Π to G−γ (Π ). however. −}. −}. δ. and the half-spin Grassmannians.8) will be obtained only for the polar Grassmannians formed by maximal singular subspaces and half-spin Grassmannians (Section 4.124 Grassmannians of Classical Buildings In Sections 4. it will be shown that all apartments preserving bijections are collineations of the corresponding Grassmann spaces.5 we consider polar buildings and investigate elementary properties of their Grassmann spaces. The building ∆(Π) is thick only in the case when Π is a polar space of type Cn . γ ∈ {+. and n ≥ 3. Let Π and Π be polar spaces of the same type Xn . X ∈ {C. −}.

It follows from the axiom (2) that polar spaces are connected gamma spaces and the distance between non-collinear points is equal to 2.1 4. X is contained in a certain singular subspace of Π. More general. L) satisfying the following axioms: (1) each line contains at least 3 points. Theorem 4.1 ([Buekenhout and Shult (1974)]).Polar and Half-Spin Grassmannians 125 4. The collinearity relation will be denoted by ⊥: we write p ⊥ q if p and q are collinear points. by Proposition 2. Suppose that a polar space Π = (P. The axiom (2) guarantees that X ⊥ is a subspace of Π. for every point p ∈ P the subspace p⊥ is a hyperplane of Π (recall that a proper subspace of a partial linear space is called a hyperplane if it has a nonempty intersection with every line). (3) there is no point collinear with all other points. L) contains a singular subspace of dimension greater than 1. moreover. in this case. X ⊥ Y means that every point of X is collinear with every point of Y . S ∩ p⊥ is a hyperplane of S for every point p ∈ P \ S. (4) every ﬂag consisting of singular subspaces is ﬁnite (this implies that every singular subspace is ﬁnite-dimensional). Since for every point p ∈ P p ⊥ X =⇒ p ⊥ X (Corollary 2.3). By Section 2.1 Polar spaces Axioms and elementary properties Following [Buekenhout and Shult (1974)] we deﬁne a polar space (of ﬁnite rank) as a partial linear space Π = (P. If S is a singular subspace of Π and p ⊥ S then the axiom (2) implies that S ∩ p⊥ is a hyperplane of S (hyperplanes of a line are points). S is a maximal clique of the collinearity graph). Let X ⊂ P be a subset satisfying X ⊥ X (a clique of the collinearity graph of Π).5. If S is a maximal singular subspace then S ⊥ = S (Π is a gamma space and. we have X ⊥ = X ⊥ . Recall that the minimal singular subspace containing X is called spanned by X and denoted by X . (2) if p ∈ P and L ∈ L then p is collinear with one or all points of the line L (the Buekenhout–Shult property). and p ⊥ q otherwise. Then the following assertions are fulﬁlled: .7. For every subset X ⊂ P we deﬁne X ⊥ := { p ∈ P : p ⊥ X }.1.

(2) for every maximal singular subspace S there exists a maximal singular subspace disjoint from S. Remark 4.1. (b) for every point p there exists a hyperplane H ∈ H which does not contain p.1 was drawn from the following result. If every line contains at least 3 points then the linear space is a projective space.2 Proof of Theorem 4.1 that the Tits–Veldkamp and Buekenhout–Shult deﬁnitions of polar spaces are equivalent. Buekenhout and E.2 ([Teirlinck (1980)]). The original proof given by F.126 Grassmannians of Classical Buildings (1) all maximal singular subspaces are projective spaces of the same ﬁnite dimension. . Suppose that a linear space has a family of hyperplanes H which satisﬁes the following conditions: (a) for every distinct hyperplanes H1 . Theorem 4. Shult was rather complicated. • for every maximal singular subspace S and every point p ∈ S. In [Buekenhout (1990)] Theorem 4. Polar spaces of rank n ≥ 3 were deﬁned in [Tits (1974)] and [Veldkamp (1959/1960)] as partial linear spaces satisfying the following axioms: • all maximal singular subspaces are (n − 1)-dimensional projective spaces. It follows from Theorem 4.1 Lemma 4. Every hyperplane in a linear space is a maximal proper subspace of this linear space.1.1. • there exist two disjoint maximal singular subspaces. all points of S collinear with p form a hyperplane of S. Proof. H2 ∈ H and every point p there is a hyperplane H ∈ H containing H1 ∩ H2 and p. 4. If H is a hyperplane of a linear space then this linear space is spanned by H and any point p ∈ H (since for every point q = p the line p q intersects H).

Let S be a maximal singular subspace whose dimension is assumed to be not less than 2. . S ∩ t⊥ 1 m+1 . All maximal singular subspaces of Π have the same dimension. (a). we can restrict ourselves to maximal singular subspaces. Each point of p q is collinear with all points of Hp ∩ Hq .1 gives the claim). Let t1 . Since t ⊥ p. . We take any point t ∈ Hq \ Hp . The hyperplane Hv contains u and Hp ∩ Hq . This point belongs to the intersection of Hp and Hq = Hq . This means that the line p q is contained in Sq (since Sq contains q and Hp ∩Hq ) which contradicts the fact that p ∈ Sq . For every point u ∈ S the axiom (3) implies the existence of a point p ∈ P \ S which is not collinear with u.2. The subspace S ∩ U ⊥ is the intersection of the hyperplanes S ∩ t ⊥ . Proof. Let Hp and Hq be distinct hyperplanes of S. It was noted above that Hp := S ∩ p⊥ is a hyperplane of S for every point p ∈ P \ S. tm+1 be a base of a certain complement of S ∩ U in U . Therefore. . Denote by Sp the singular subspace spanned by Hp and p. Proposition 4. (b). Every singular subspace of Π whose dimension is not less than 2 is a projective space. This point does not belong to Hq . . If q ∈ Sp \ Hp then Hq coincides with Hp (it is clear that Hp ⊂ Hq and Lemma 4. We show that the family of hyperplanes {Hp }p∈P \S satisﬁes the conditions of Theorem 4. .Polar and Half-Spin Grassmannians 127 Proposition 4.1. the line t q contains a unique point q collinear with p. Suppose that the line p q intersects S in a certain point. Proof. Thus the line p q does not intersect S. The associated hyperplane Hp does not contain u. q ∈ Sq \ Hq . . Clearly. Then p ∈ Sq and q ∈ Sp . . Then dim(S ∩ U ) = k − m − 1.2).2. Let S and U be maximal singular subspaces of dimension n and k (respectively) and k ≤ n. For every point u ∈ S we can choose a point v ∈ p q collinear with u. . Suppose that the dimension of complements of S ∩ U in U is equal to m (see Remark 4.

2. the dimension of (4. Remark 4. We say that N is a complement of M in T if M ∩ N = ∅ and M. Then dim(S ∩ U ⊥ ) = n − m − 2 and U.2. and we say that it is a polar space of rank 2 or a generalized quadrangle.1) is spanned by S ∩ U ⊥ and a compliment of S ∩ U in U . Let also U be a singular subspace such that the dimension of complements of S ∩ U in U is equal to m. the number n + 1 is said to be the rank of this polar space.1) . S ∩ U ⊥ is a maximal singular subspace containing U . This means that (4.1 then all maximal singular subspaces are lines. if a polar space contains a singular subspace of dimension greater than 1 then all maximal singular subspaces are projective spaces of a certain ﬁnite dimension n ≥ 2. Proof. Let S be a maximal singular subspace in a polar space of rank n. Let M . (4. N ⊂ T .1) is not less than n − 1. Then dim M + dim N + 1 = dim T (for projective planes this is trivial. Since U = U ⊥ (U is a maximal singular subspace). T be subspaces of a projective space such that M. N = T. we get k−m−1≥n−m−1 and k ≥ n which implies k = n.3). As in the proof of Proposition 4. So.1) is a maximal singular subspace and its dimension is equal to n − 1. If a polar space does not satisfy the condition of Theorem 4. for projective spaces whose dimension is greater than 2 this follows from Theorem 1. we establish that dim(S ∩ U ⊥ ) ≥ n − m − 2.2. N . Lemma 4. The subspace (4. The latter guarantees that S ∩ U ⊥ is (n − m − 2)-dimensional. Since these subspaces are disjoint and the latter subspace is m-dimensional.128 Grassmannians of Classical Buildings and dim(S ∩ U ⊥ ) ≥ n − m − 1.

dim(S ∩ U ) < dim(S ∩ U ).2 and 4.3.2).1. S ⊥ = U ⊥ if and only if S = U . Therefore.3. As above.4.1 Let Π = (P. hence. Note that S and U ∩ S ⊥ are disjoint. we construct a maximal singular subspace disjoint from S. p ∈ U and we denote by U the maximal singular subspace spanned by p and U ∩ p⊥ (Lemma 4. Let S and U be maximal singular subspaces such that S ∩ U = ∅. for any pair of singular subspaces S and U the inclusion S ⊥ ⊂ U ⊥ implies that U ⊂ S. Consider the maximal singular subspace U := S. By the axiom (3).3 Corollaries of Theorem 4. Step by step.1 is the union of Propositions 4. S ∩ U is contained in S ∩ U . Clearly. Since t ∈ U . By Proposition 4. Theorem 4.4. Every non-maximal singular subspace of Π can be presented as the intersection of two maximal singular subspaces. Proof. Let S be a singular subspace and U be a maximal singular subspace containing S. Now suppose that X is a clique of the collinearity graph of Π. 4. Proposition 4. there exists a point p non-collinear with a certain point t ∈ S ∩ U . The subspace U ∩ U does not intersect U ∩ S ⊥ and the inclusion S ⊂ U ∩ U guarantees that U ∩ U coincides with S. For every maximal singular subspace S there exists a maximal singular subspace disjoint from S.Polar and Half-Spin Grassmannians 129 Proposition 4. Proof. L) be a polar space of rank n.2). X ⊥ is the union of all maximal singular subspaces containing X and X ⊥⊥ . there exists a maximal singular subspace U disjoint from U .1. U ∩ S ⊥ (Lemma 4. If U contains a point q ∈ S \ U then U ∩ p⊥ = U ∩ q ⊥ which is impossible (t ∈ S ∩ U is collinear with q and non-collinear with p). By Theorem 4. If S is a singular subspace of Π then S ⊥ is the union of all maximal singular subspaces containing S (recall that Π is a gamma space and the class of maximal singular subspaces coincides with the class of maximal cliques of the collinearity graph).

by the axiom (2).4 Polar frames Let Π = (P. We will need the following result concerning pairs of non-collinear points in polar spaces. n} there is a unique point of U collinear with all points of B \ {pi }. This means that every isomorphism between the collinearity graphs of polar spaces is a collineation between these polar spaces. . We take any base B = {p1 . .2. First of all we show that frames exist. q}⊥⊥. for every i ∈ {1.130 Grassmannians of Classical Buildings is the intersection of all these subspaces. (2). . . . We verify (3). p⊥ ∩ q ⊥ is a polar space. . Therefore. . For every point s ∈ p⊥ \{p} the line p s contains two distinct points collinear with t (one of these point is p and the other is the intersection with the hyperplane q ⊥ ). . . . . t is collinear with s. 2n} such that pi ⊥ pσ(i) . By Lemma 4. As a consequence. If n ≥ 3 and p. L) be a polar space of rank n. our polar space is of rank n − 1. Let S and U be disjoint maximal singular subspaces of Π. .3. . .4 implies that X ⊥⊥ = X . we denote .1. Proof. p2n } is a frame of Π if for every i ∈ {1. Lemma 4. The axioms (1). We say that a subset {p1 . . . . . Suppose that there exists a point t ∈ p⊥ ∩ q ⊥ collinear with all points of ⊥ p ∩q ⊥ . Thus p⊥ ⊂ t⊥ and p = t which contradicts p ∈ p⊥ ∩ q ⊥ . pn } of S. (4) are trivial. Then X ⊂ X ⊥⊥ and Proposition 4. q are non-collinear points of Π then the subspace p⊥ ∩ q ⊥ is a polar space of rank n − 1. . So. . It is clear that p⊥ ∩ q ⊥ does not contain (n − 1)-dimensional singular subspaces. 2n} there is unique σ(i) ∈ {1. 4. we obtain the following characterization of lines in terms of the collinearity relation: if p and q are distinct collinear points then p q = {p. Every maximal singular subspace containing p intersects q ⊥ in an (n − 2)-dimensional subspace.

. X and Y are disjoint. Then S ∩ U is spanned by the set S ∩ U ∩ B. An easy veriﬁcation shows that there exist disjoint subsets X.6. is empty and there exists pi ∈ X such that X \ {pi } does not contain p. . In some cases. Proposition 4. p2n }. but this hyperplane does not contain pi . pi ∈ X. the subspaces S and U are spanned by S ∩ B and U ∩ B. Y are maximal singular subspaces and S ∩ B ⊂ X.Polar and Half-Spin Grassmannians 131 it by pn+i . . since S ∩ U = ∅). In particular. This means that pσ(i) ⊥ X which contradicts the fact that pi ∈ X . . Proof. .1). Y ⊂ B such that X . Let B be a frame and S. Since S ⊂ X and U ⊂ Y . we get the claim. Proof. . Proposition 4. . respectively. U be singular subspaces spanned by subsets of B. Every frame is an independent subset. if this set is empty then S and U are disjoint. . p2n } is a frame of Π then B \ {pi } is contained in the hyperplane p⊥ . we establish that (S ∩ B) ∩ (U ∩ B) = ∅ =⇒ S ∩ U = ∅. Since B is an independent subset. any k distinct mutually collinear points in a frame span a (k−1)-dimensional singular subspace. Suppose that B = {p1 . The point pσ(i) ∈ B \ X = Y is collinear with all points of this base (it is trivial that pσ(i) ⊥ X \ {pi } and we have pσ(i) ⊥ p. Then {p1 . . . This point is non-collinear with pi (otherwise pn+i ⊥ S and the maximal singular subspace S contains pn+i which is impossible. Therefore. If B = {p1 . p2n } is a frame of Π.3. The intersection of all X \ {pi } . Then (X \ {pi }) ∪ {p} is a base of X . U ∩ B ⊂ Y. σ(i) Remark 4. . Since every subset of an independent subset is independent. First.5.3. since p and pσ(i) belong to Y ). . Suppose that X and Y have a non-empty intersection and consider a point p belonging to X ∩ Y . frames are not bases of a polar space (examples will be given in Subsection 4. .

Proposition 4. For any singular subspaces S1 .7. In this case. M2 such that M 1 ∩ M 2 = S1 ∩ S2 and Si ⊂ Mi for i = 1. . q .132 Grassmannians of Classical Buildings Now suppose that S ∩ U ∩ B = ∅.3). the frame B ∪ {p. p and U = U . By Proposition 4. A point collinear with all points of B is contained in every maximal singular subspace spanned by a subset of B. q} is as required. Proof. hence it does not intersects S∩U . By the inductive hypothesis. these subspaces are disjoint and the second subspace is disjoint from U . The inclusion S∩U ∩B ⊂S∩U ⊂S guarantees that S ∩ U ∩ B coincides with S ∩ U . there are non-collinear points p ∈ S and q ∈ U . The case n = 2 is trivial and we suppose that n ≥ 3. 2. S2 there exist maximal singular subspaces M1 . Now suppose that S ⊥ U . Proof. By the ﬁrst part of our proof. Corollary 4. We prove the statement induction by n.2. Then S is spanned by the subspaces S∩U ∩B and (B ∩ S) \ (U ∩ B) . Since S = S . It was shown above that this base can be extended to a frame of Π. there exist disjoint maximal singular subspaces spanned by subsets of B. Corollary 4. If S ⊥ U then there exists a maximal singular subspace M containing S and U . For any singular subspaces S and U there is a frame of Π such that S and U are spanned by subsets of this frame. We choose a base of M such that S and U are spanned by subsets of this base. If B is a frame of Π then there is no point of Π collinear with all points of B.1.6. The singular subspaces S := S ∩ q ⊥ and U := U ∩ p⊥ are contained in the polar space p⊥ ∩ q ⊥ (Lemma 4. there exists a frame B of p⊥ ∩ q ⊥ such that S and U are spanned by subsets of B .

Exercise 4. As in Section 3. For any adjacent S. L) is a polar space of rank n. n − 1} denote by Gk (Π) the polar Grassmannian consisting of all k-dimensional singular subspaces of Π.1 Grassmannians Polar Grassmannians As above. S. k < n − 1. Show that every line of Gk (Π). N ]k . . In the case when k ≤ n − 2. 4. If S and U are distinct elements of Gn−1 (Π) then S ⊥ U . Then G0 (Π) = P .2.1. Two distinct elements of Gk (Π) are joined by a line if and only if they are adjacent.Polar and Half-Spin Grassmannians 133 Proof. N ]k is a line of Gk (Π) if dim M = k − 1 and dim N = k + 1. we suppose that Π = (P. For every k ∈ {0. we say that S. . The set [M n−1 is said to be a line of Gn−1 (Π) if M belongs to Gn−2 (Π).1. contains at least three distinct points. U ∈ Gk (Π) are adjacent if S ⊥ U and their intersection is (k − 1)-dimensional (or. . we say that [M. containing S1 and intersecting S2 precisely in S1 ∩ S2 . U ∈ Gk (Π) there is precisely one line containing them: [S ∩ U. Consider a frame B such that S1 and S2 are spanned by subsets of B. X ⊂ B. Also denote by [M k the set of all elements of Gk (Π) containing M . if M = ∅ then we will write N ]k instead of [M. In the case when 0 ≤ k < n − 1. we deﬁne [M. M ∩ S2 contains S2 and intersects M1 precisely in S1 ∩ S2 . equivalently. if S and U span a (k + 1)-dimensional singular subspace). There exists a maximal singular subspace M1 = X . Let M and N be incident singular subspaces of Π such that dim M < k < dim N. The set of all lines of Gk (Π) will be denoted by Lk (Π). 1. U ]k if k ≤ n − 2 and [S ∩ U k if k = n − 1. N ]k := { S ∈ Gk (Π) : M ⊂ S ⊂ N }. If M = B \ X then the maximal singular subspace ⊥ M2 := S2 .2 4. . . such subspaces are said to be adjacent if their intersection belongs to Gn−2 (Π).

As in the proof of Proposition 3. Proposition 4. . We take any point p ∈ U \ S and denote by S1 the maximal singular subspace spanned by S ∩ p⊥ and p. 1. . The Grassmann space Gk (Π) is connected for every k. i}. U ). Ul = U in the collinearity graph of Gn−1 (Π) we have dim(U0 ∩ U1 ∩ · · · ∩ Ul ) ≥ (n − 1) − l. .1. such that Sj−1 and Sj are adjacent for every j ∈ {1. . U ∈ Gk (Π).4. . We deﬁne cd(S. . The distance between S. S1 . In the general case. . It is clear that G0 (Π) = Π. we construct a sequence of maximal singular subspaces S = S0 . Lk (Π)). U ∈ Gn−1 (Π) is equal to n − 1 − dim(S ∩ U ). Proof. The trivial inclusion U0 ∩ U1 ∩ · · · ∩ Ul ⊂ S ∩ U guarantees that dim(S ∩ U ) ≥ (n − 1) − l. U ) > 1. Suppose that k = n − 1 and cd(S. . are partial linear spaces. k ∈ {0. i = cd(S. .8. n − 1}. . Si = U. Let S. the distance formula is more complicated. . for every path S = U 0 . The Grassmann space Gn−1 (Π) also is known as the dual polar space of Π [Cameron (1982)]. . . U ) = 1 is equivalent to the fact that S and U are adjacent).134 Grassmannians of Classical Buildings The pairs Gk (Π) := (Gk (Π). Then S1 is adjacent with S and cd(S1 . . Remark 4. . U ) − 1. U 1 . Step by step. they are called the Grassmann spaces of Π. U ) := k − dim(S ∩ U ) (if k = n − 1 then cd(S. U ) = cd(S. . .

p2n } of Π such that S = pi1 . S2 ) = 1 and cd(S2 . U ). . There exist S1 ∈ Gk (Π) adjacent with both S. The latter means that the rank of our polar space is equal to n − k. . . . S2n−2k } be a frame of [U k .3. In the case when S ⊥ U . U ) − 1. U ) > 1. By Lemma 4. . . σ(i1 ). . The k-dimensional singular subspace pi1 . pσ(m) . Lemma 4. pik .Polar and Half-Spin Grassmannians 135 Thus l ≥ cd(S. Then X ∪ Y is a frame of Π satisfying the required conditions. We choose j ∈ {1. pik . . If S ⊥ U then S. where M is a maximal singular subspace of Π. . . . we prove the connectedness induction by cd(S. For every pi there is unique pσ(i) such that pi ⊥ pσ(i) . pj is adjacent with both S and U . . . ik . In each Si we take a point pi ∈ Si \ U and denote by X the set of all pi . . . Let 0 < k < n − 1 and U be a (k − 1)-dimensional singular subspace of Π. . pm and U = pi1 . This polar space contains U and we choose a frame Y of X ⊥ such that U is spanned by a subset of Y . U ) = cd(S. If cd(S.1. . In the case when cd(S. (2) for every frame B of [U k there is a frame B of Π such that U is spanned by a subset of B and B = A ∩ [U k . . pik . U be non-adjacent elements of Gk (Π). Let k ≤ n−2 and S. . . σ(m)} (this is possible. Proof. Then cd(S. m. . (1). 2n} \ {i1 . (2). Let {S1 . since k ≤ n − 2). An easy veriﬁcation shows that [U k is a polar space and every its maximal singular subspace is [U. we take a point p ∈ U satisfying p ⊥ S and write S2 for the k-dimensional singular subspace spanned by S ∩ p⊥ and p. . X ⊥ is a polar space of rank k.4. . . U are contained in the singular subspace S ∪ U and a path connecting S with U can be constructed as in the proof of Proposition 3. U ) which means that the distance between S and U is equal to cd(S. . . S2 and the inductive hypothesis implies the existence of a path connecting S2 with U . U ) = 1 then there exists a frame {p1 . σ(ik ). . . . M ]k . where A consists of all k-dimensional singular subspaces spanned by subsets of B. The following assertions are fulﬁlled: (1) [U k is a polar space of rank n − k. U ).

. for every i ∈ {1.3. 2. Every point pi lies on precisely two lines from the collection p1 p2 . the polar space Π is of type C2 if there is a point belonging to at least three distinct lines. p3 . Suppose that {p1 . Proof.2 Two types of polar spaces Theorem 4. L) be a polar space of rank n. Let L be a third line passing through p1 . p3 p4 . p} and take a unique point q ∈ p3 p4 collinear with q (it is clear that the point q is distinct from p3 and p4 ). Now consider a point q ∈ L \ {p1 . First consider the case when n = 2. p2 L p q p3 q' p4 p1 There is a point p ∈ L collinear with p3 and pp3 is a third line passing through p3 . (D) every (n − 2)-dimensional singular subspace is contained in precisely two maximal singular subspaces. p4 } is a frame of Π where p1 ⊥ p3 and p2 ⊥ p4 . Therefore. This line contains a point collinear with p4 and we get a third line through p4 . p2 . A point on the line qq collinear with p2 gives a third line passing through p2 . For a rank n polar space one of the following possibilities is realized: (C) every (n−2)-dimensional singular subspace is contained in at least three distinct maximal singular subspaces. We say that a rank n polar space is of type Cn or Dn if the corresponding case is realized.136 Grassmannians of Classical Buildings 4. Let Π = (P.2. p2 p3 . 4} there are at least three distinct lines passing through pi . Since any pair of points is contained in a certain frame. 3. p1 p4 .

p2 . L) be a polar space of type Dn and B = {p1 .5. pn . In the case when Π is of type Dn . Exercise 4. n}. Let Π = (P. . By Lemma 4. q1 . Proposition 4. . If Π is a generalized quadrangle of type C2 then G1 (Π) is a generalized quadrangle of type C2 . . . If Π is a generalized quadrangle of type D2 then G1 (Π) does not satisfy the polar axiom (1). . lines are edges of the collinearity graph. qn } be a frame of Π such that pi ⊥ qi for every i ∈ {1. q2 } be a frame of Π such that pi ⊥ qi for i ∈ {1. Every frame in a polar space of type Dn is a base of this polar space. This means that every element of [N n−2 is contained in at least three distinct maximal singular subspaces.5. 2}. the Grassmann space Gn−1 (Π) is trivial: every line consists of two points. Let N be an (n−3)-dimensional subspace of S. B is an independent subset and we need to show that Π is spanned by . in other words. . the generalized quadrangle is of type C2 . We need to prove the following: if a certain S ∈ Gn−2 (Π) is contained in at least three distinct maximal singular subspaces then the same holds for all elements of Gn−2 (Π).2. q1 . Remark 4. . if every line of Π consists of three points then Π looks as in the picture below. Show that every line of Π intersects the lines p1 p2 and q1 q2 or the lines p1 q2 and p2 q1 . Therefore. . . Therefore. . [N n−2 is a generalized quadrangle. Every line of Gn−1 (Π) contains at least three distinct points if Π is a polar space of type Cn . . By Proposition 4.9. in particular. . The connectedness of the Grassmann space Gn−2 (Π) gives the claim. Since S (as a point of this generalized quadrangle) lies on at least three distinct lines. if S and U are adjacent. U ∈ Gn−2 (Π) is contained in at least three distinct maximal singular subspaces if it has an (n − 3)-dimensional intersection with S. Let Π be a generalized quadrangle of type D2 and {p1 . Proof.Polar and Half-Spin Grassmannians 137 Now suppose that n ≥ 3.4.

Then B = (B \ {p1 . ⊥ By the inductive hypothesis.2. In the case when n = 2. Lemma 4. U ) = n − 1 − dim(S ∩ U ). We show that the Grassmannian Gn−1 (Π) can be uniquely decomposed in two disjoint parts such that the distance between two elements of Gn−1 (Π) is odd if and only if these elements belong to the diﬀerent parts. Lemma 4. . N ) is odd if and only if d(U. y} is a frame of Π and p ∈ B = B . L) is a polar space of type Dn . Recall that d(S. U ∈ Gn−1 (Π). This means that at least one of the subspaces S ∩ N. Suppose that n ≥ 3 and prove the statement induction by n. Observe that every point of S \ U is noncollinear with every point of U \ S. only one of these subspaces is contained in S ∩ U . 4. If S ∩ N and U ∩ N both are contained in S ∩ U then p does not belong to S ∪ U and S ∩ U. Therefore.138 Grassmannians of Classical Buildings B.5.2. Suppose that p ∈ P is non-collinear with every point of B. q2 }) ∪ {x. The statement is trivial if N coincides with S or U and we assume that N is distinct from S and U .3 Half-spin Grassmannians Throughout this subsection we suppose that Π = (P. Consider the case when S ∩ N ⊂ S ∩ U and U ∩ N ⊂ S ∩ U. the polar space p⊥ ∩ q1 is spanned by 1 ⊥ ⊥ B \ {p1 . Let S and U be adjacent elements of Gn−1 (Π). for all S. N ) is even. U ∩ N is contained in S ∩ U (otherwise N contains points x ∈ S \ U and y ∈ U \ S which cannot be collinear).2 implies the existence of a point p ∈ N \ (S ∩ U ) collinear with all points of S ∩ U . this follows from Exercise 4. q1 }. The same inclusion holds for all points of B. p is a third maximal singular subspace containing S ∩ U which contradicts the assumption that Π is of type Dn . We take x ∈ p1 p2 collinear with p and y ∈ q1 q2 collinear with x. For every point t ∈ p1 \ {p1 } the line p1 t intersects p⊥ ∩ q1 1 ⊥ and p1 ⊂ B . Proof. Then for any N ∈ Gn−1 (Π) the distance d(S.

Y ⊂ Gn−1 (Π) is another pair of subsets satisfying the same conditions. N ) = d(U. There is a unique pair of disjoint subsets X . hence Y = Y. Proof.4. We ﬁx N ∈ Gn−1 (Π) and deﬁne X := {S ∈ Gn−1 (Π) : d(S. N ) is even}. This means that X = X . G− (Π) and called the half-spin Grassmannians of Π. Y := {U ∈ Gn−1 (Π) : d(U. Therefore.Polar and Half-Spin Grassmannians 139 Then S∩N =S∩U ∩N is a hyperplane of U ∩ N (since S ∩ U is a hyperplane of U ). Then d(S.4 will be denoted by G+ (Π). N ) is odd}. Let X ∩ X = ∅ and S be an element of this intersection. the distance between two elements of Gn−1 (Π) is odd if and only if one of them belongs to X and other belongs to Y. in the case when the distance is even. N ) + 1. Then X intersects at least one of the subsets X . The distance between any two elements of the half-spin Grassmannian (in the collinearity graph of Gn−1 (Π)) is even. . By the previous lemma. Suppose that X . These are disjoint subsets whose union is Gn−1 (Π). any two adjacent elements of Gn−1 (Π) belong to the diﬀerent subsets. U ) is even for every U belonging to X ∪ X . Y. The subsets described in Theorem 4. Y ⊂ Gn−1 (Π) satisfying the following conditions: X ∪ Y = Gn−1 (Π). the elements both belong to X or Y. Theorem 4. The second case is similar. This implies that d(S. if the distance between two elements of Gn−1 (Π) is odd then one of them belongs to X and the other belongs to Y.

N3 = N1 contained in U1 and containing M (in other words. U1 ]n−2 ). one of them belongs to Gδ (Π) and the other is an element of G−δ (Π). The half-spin Grassmann space Gδ (Π). 2 . Proposition 4. −} then we write −δ for the sing satisfying {δ. Then S1 . Let M ∈ Gn−3 (Π). Shows that every path of length j in the collinearity graph of Gδ (Π). Exercise 4. Hint: if S. The intersection [M δ := [M n−1 ∩ Gδ (Π) is called a line of Gδ (Π) if M is (n − 3)-dimensional.140 Grassmannians of Classical Buildings Let δ ∈ {+. −δ} = {+.11. this is the intersection of two maximal singular subspaces S1 and U1 . are called the half-spin Grassmann spaces of Π. The distance between S. N2 . Proof.3. Lδ (Π)). δ ∈ {+.4. 3) intersecting U1 precisely in Ni . If δ ∈ {+. Two elements of Gδ (Π) are said to be adjacent if their intersection is (n − 3)-dimensional (the distance in the collinearity graph of Gn−1 (Π) is equal to 2). S2 . −}. δ ∈ {+. The patrial linear spaces Gδ (Π) := (Gδ (Π). Suppose that U1 ∈ G−δ (Π). N1 . We choose two distinct (n − 2)-dimensional singular subspaces N2 .4 implies the existence of Si ∈ Gδ (Π) (i = 2. By Proposition 4. −}. Consider any (n − 2)-dimensional singular subspace N1 containing M . S3 are distinct points on the line [M δ . The set of all such lines will be denoted by Lδ (Π). δ ∈ {+. U ∈ Gδ (Π) (in the collinearity graph of Gδ (Π)) is equal to n − 1 − dim(S ∩ U ) . contains at least three points. −}. is connected. U ∈ Gδ (Π) are adjacent then there exist p ∈ S \ U and q ∈ U \ S satisfying p ⊥ q. Two distinct elements of Gδ (Π) are joined by a line if and only if they are adjacent. U ∈ Gδ (Π) are adjacent then [S ∩ U δ is the unique line containing them. Proposition 4.10. δ ∈ {+. can be extended to a path of length 2j in the collinearity graph of Gn−1 (Π). Every line of Gδ (Π). N3 are distinct points on the line [M. In what follows we will restrict ourselves to the case when n ≥ 3. If S. −}. If n = 2 then any two distinct elements of Gδ (Π) are disjoint and there is only one line which coincides with Gδ (Π). −}. −}. Proposition 4.

S1 . We will investigate the pair ΠΩ := (G1 (Ω). Distinct P. in other words. . Theorem 4.3 4.9 and 1. x) ∈ T (σ. x)) ∀ x. i = n − 1 − dim(S ∩ U ). U ∈ Gδ (Π). L(Ω)). Si is a path in the collinearity graph of Gδ (Π).4 guarantees that i is even and S0 . P ∈ G1 (Ω) are joined by a such line if and only if P ⊥ P (⊥ is the orthogonal relation deﬁned by Ω). y ∈ V.3. . This implies that n ≥ 4 (the dimension of a totally isotropic subspace is not greater than the codimension). Suppose that Ω(x. The distance formula follows immediately from Exercise 4. . Suppose that Ω has totally isotropic subspaces of dimensions 2. By Proposition 4. Let S. Ω is one of the forms considered in Examples 1. . 4. ε) := { a + εσ(a) : a ∈ R } and say that the form Ω is trace-valued if Ω(x. S2 . y) = εσ(Ω(y.1 Examples Polar spaces associated with sesquilinear forms Let V be an n-dimensional left vector space over a division ring R and Ω be a non-degenerate reﬂexive form deﬁned on V . Also every alternating form is trace-valued. the collinearity graph of Gn−1 (Π) contains a path S = S0 . or Hermitian. where σ is an anti-automorphism of R satisfying σ 2 = 1R and ε = ±1. . This condition holds if the characteristic of R is not equal to 2.Polar and Half-Spin Grassmannians 141 Proof.3. we can assume that Ω is one of the forms given in Theorem 1. ε) (4. symmetric. We deﬁne the trace set T (σ. .8.10. . Si−2 . .2) for every x ∈ V . . We write L(Ω) for the set formed by all lines of ΠV such that the associated 2-dimensional linear subspaces are totally isotropic. Since two “proportional” forms (one of the forms is a scalar multiple of the other) have the same set of totally isotropic subspaces.6: alternating. Si = U.

6. since our form is trace-valued).2) form a linear subspace W ⊂ V . y1 . is empty and we choose j such that x does not belong to B \{xj } . y)) + Ω(y. Then Ω(ax + y. Example 4. .7. . if the form Ω is trace-valued and a line of ΠV has a nonempty intersection with G1 (Ω) then this intersection contains more than one point. The axiom (4) follows from the fact that singular subspaces of ΠΩ can be identiﬁed with totally isotropic subspaces. . ax) + aΩ(x. Proof. Lemma 4. If Ω is trace-valued then S contains an isotropic vector linearly independent with x. Proof. . y) = 0 (this is possible. since our form is nondegenerate. . yn−1 is as required. . The intersection of the linear subspaces B \ {xi } . xn } be a base of V and x be an isotropic vector. Suppose that R is a ﬁeld of characteristic 2 and the form Ω is bilinear (σ = 1R and ε = 1 = −1).6. all vectors x ∈ V satisfying (4. . this vector is orthogonal to all vectors of V which is impossible. .142 Grassmannians of Classical Buildings Remark 4. . i = 1. If the axiom (3) fails then there exists an isotropic vector orthogonal to all isotropic vectors. . i = 1. Otherwise. All isotropic vectors belong to W and the restriction of Ω to W is trace-valued. Let S be a 2-dimensional linear subspace of V containing an isotropic vector x. . Assume that j = n. . Since the trace set is zero. xi \ {x}. Using this fact we prove the following. The base x.6 implies the existence of isotropic vectors yi ∈ x. .12.7. ax + y) = Ω(ax. Let B = {x1 . The statement is trivial if S is totally isotropic. If Ω is trace-valued then there is a base of V consisting of isotropic vectors. y) = 1 and a scalar a ∈ R satisfying a + εσ(a) + Ω(y. n. the form Ω is trace-valued only in the case when it is alternating. . If Ω is trace-valued then ΠΩ is a polar space. we choose a vector y ∈ S such that Ω(x.1. by Lemma 4. Proof. Proposition 4. The polar axiom (1) is trivial and we leave the veriﬁcation of the axiom (2) for the reader. y) + εσ(aΩ(x. . y) = 0. Lemma 4. In the general case. . In other words. . Lemma 4. n − 1.

c =0 2 c c +σ 2 2 .7. The statement concerning the e dimension of maximal totally isotropic spaces is a part of well-known Witt theorem. a = 1).1 and Proposition 4. for example. Taylor (1992)]. Then c := b + σ(b) ∈ T (σ. In this case. Proof. the trace set T (σ. By Theorem 1. 1) = { a + σ(a) : a ∈ R } coincides with the set of all scalars b ∈ R satisfying σ(b) = b.8. Let σ be a non-identity anti-automorphism of R satisfying σ 2 = 1R . since σ is non-identity). The Witt index is not greater than n and equal to the rank of the polar space ΠΩ . The veriﬁcation of Tits – Veldkamp axioms is more complicated [Dieudonn´ (1971). 2 if Ω is alternating then n is even and the Witt index is equal to n . • symmetric polar spaces (deﬁned by symmetric forms. Lemma 4.6. there are precisely the following three types of polar spaces associated with reﬂexive forms: • symplectic polar spaces (deﬁned by alternating forms). the characteristic is not equal to 2). This dimension is called the Witt index of the form Ω. 1) and the equality c= guarantees that the scalar a := b − is as required. Let us take any b ∈ R \ T (σ. Suppose that the characteristic is not equal to 2.12 that all maximal totally isotropic subspaces have the same dimension.Polar and Half-Spin Grassmannians 143 It follows immediately from Theorem 4. The statement is trivial if the characteristic is equal to 2 (we can take. 2 Remark 4. For example. 1) (such elements exist. Then there exists a non-zero scalar a ∈ R such that a + σ(a) = 0. • Hermitian polar spaces (deﬁned by trace-valued Hermitian forms).

x). x) satisﬁes the condition of Lemma 4. 2 Proof. P1 + P contains P ∈ G1 (Ω) distinct from P1 . We take 1-dimensional linear subspaces Pi (i = 1.144 Grassmannians of Classical Buildings Proposition 4. By Lemma 4.8. Now suppose that Ω is symmetric and the characteristic of R is not equal to 2. The polar space is of type Cm .6. the dimension n is even and Ω is a symmetric form of Witt index n . Every maximal totally isotropic subspace U containing S is contained in S ⊥ . We take any 1-dimensional linear subspace P ⊂ S ⊥ which is not contained in S + P1 + P2 . x + by) = Ω(x. Then S + P is a third maximal totally isotropic subspace containing S. Let m be the Witt index of the form Ω and S be a totally isotropic subspace of dimension m − 1. Then x + by belongs to G1 (Ω) and S + x + by is a third maximal totally isotropic subspace containing S. this means that the linear subspace P1 + P2 does not contain elements of G1 (Ω) distinct from P1 and P2 .13. x + by) = 2bΩ(x. Since P1 ⊥ P2 . Let ΠΩ be one of the polar spaces considered above. If x. The polar space is of type Cm . Consider the case when Ω is a Hermitian form associated with an antiautomorphism σ : R → R. by) + Ω(by. Then for every 1-dimensional linear subspace P ⊂ P1 + P2 the linear subspace S + P is totally isotropic and our polar space is of type Cm . Then S is the intersection of two maximal totally isotropic subspaces U1 and U2 . since 2 dim(S + P1 + P2 ) = m + 1 < n − m + 1 = dim S ⊥ . We have Ω(x + by. y ∈ P2 and a scalar b ∈ R such that a = bΩ(y. Then ΠΩ is of type Dm only in the following case: the characteristic of R is not equal to 2. y ∈ V are isotropic vectors then Ω(x + by. It is clear that P1 ⊥ P2 . 2) such that Ui = S + Pi . If m < n then S + P1 + P2 is a proper linear subspace of S ⊥ . If n = 2m then S ⊥ is (m + 1)-dimensional and U intersects P1 + P2 in . x)) + bΩ(y. Suppose that Ω is alternating. x) = σ(bΩ(y. We choose vectors x ∈ P1 . y) for any isotropic vectors x and y.

y1 . Almost all points of the line p q (except the point q ) are of type (x1 + ax2 ) + t(y2 + ay1 ) . If the Witt index is less than n then every frame spans a proper subspace of 2 ΠV whose intersection with G1 (Ω) is a subspace of ΠΩ . Cooperstein and Shult (1997)]. Let B = {p1 .Polar and Half-Spin Grassmannians 145 a 1-dimensional linear subspace P . p = x1 + by2 . this fails. . q ∈ p2 q1 distinct from the points of B. p2 . it does not contain. There exist vectors x1 . b ∈ R we have p = x1 + ax2 . s ∈ R. Since P is totally isotropic. q = y2 + ay1 . However. any line passing through p1 and distinct from p1 p2 and p1 q2 . q2 } be a frame of the polar space ΠΩ such that pi ⊥ qi for each i. Suppose that n = 4. B is a proper subspace of ΠΩ . This implies that there are precisely two maximal totally isotropic subspaces containing S and the polar space is of type Dm . In the case of characteristic 2.7) and frames are not bases of ΠΩ . q = x2 + by1 . y2 ∈ V satisfying pi = xi . Then for some non-zero scalars a. for example. yi ) = Ω(yi . s = a).2. Frames of the polar space ΠΩ are independent subsets of ΠV . it coincides with P1 or P2 . Consider two pairs of collinear points p ∈ p1 p2 . qi = yi and Ω(xi . This means that B is the grid consisting of all lines which intersect p1 p2 and q1 q2 or p1 q2 and p2 q1 . R is a ﬁeld of characteristic 2 and the form Ω is alternating. and almost all points of the line p q (except the point q ) are of type (x1 + by2 ) + s(x2 + by1 ) . q1 . x2 . Example 4. Therefore. xi ) = 1. if the characteristic of R is not equal to 2 and Ω is alternating then every frame is a base of ΠΩ [Blok and Brouwer (1998). t ∈ R. q ∈ q1 q2 and p ∈ p1 q2 . this subspace is proper (by Lemma 4. These lines have a common point (t = b.

a linear subspace is totally singular (for Q) if and only if it is totally isotropic (for Ω). By this deﬁnition. Consider the pair ΠQ := (G1 (Q). We restrict our self to the case when Ω is non-degenerate and refer [Dieudonn´ (1971)] for the general case. P ∈ G1 (Q) are joined in ΠQ by a line only in the case when P ⊥ P (⊥ is the orthogonal relation deﬁned by Ω). The form Q is said to be non-degenerate if Q(x) = 0 for every non-zero vector x belonging to V ⊥ (we do not require that the associated form Ω is non-degenerate and write V ⊥ for the linear subspace of all vectors orthogonal to V ). x ∈ V and there exists a bilinear form Ω : V × V → F such that Q(x + y) = Q(x) + Q(y) + Ω(x. Q is non-degenerate if the associated form Ω is non-degenerate. Q(x) = 2 This equality guarantees that Q is non-degenerate if and only if Ω is nondegenerate. x) shows that Ω is alternating.146 Grassmannians of Classical Buildings 4. From this moment we assume that Q is non-degenerate and there exist totally singular subspaces of dimension 2. x) . A non-zero vector x ∈ V is called singular if Q(x) = 0 and we say that a linear subspace S is totally singular if every non-zero vector of S is singular. Since e . Suppose that the characteristic of F is equal to 2. It is clear that P. We write G1 (Q) for the set of all 1-dimensional totally singular subspaces and denote by L(Q) the set formed by all lines of ΠV such that the associated 2-dimensional linear subspaces are totally singular. The equality Q(2x) = 2Q(x) + Ω(x. L(Q)). but totally isotropic subspaces of Ω need not to be totally singular. A non-zero mapping Q : V → F is called a quadratic form if Q(ax) = a2 Q(x) for all a ∈ F. Every totally singular subspace is totally isotropic for the associated form Ω. If the characteristic of F is not equal to 2 then Q can be uniquely recovered from Ω by the formula Ω(x.2 Polar spaces associated with quadratic forms Let V be an n-dimensional vector space over a ﬁeld F . It is clear that Ω is symmetric. If the characteristic of F is not equal to 2 then ΠQ coincides with the polar space ΠΩ .3. moreover. y) for all x. y ∈ V .

y independent with x. Taylor (1992)]. the vector ax + y is as required. All singular subspaces of ΠQ can be identiﬁed with totally singular subspaces of Q. Prove the analogue of Lemma 4. ΠQ is a subspace of the polar space ΠΩ .3 Polar spaces of type D3 Let V be a 4-dimensional vector space over a division ring R.5. but Q ≡ 0 and there are non-singular vectors). Consider the associated half-spin Grassmannian space Gδ (Π). thus the rank of the polar space is equal to 3.2). Show that the Grassmann space G2 (V ) is a polar space. y) + Q(y) = 0. (2) and (4). Section 8. . We do not consider here so-called pseudo-quadratic forms associated with Hermitian forms (see [Tits (1974)].6. let Π = (P. −}. Prove the following analogue of Lemma 4. Thus ΠQ satisﬁes the polar axioms (1).3. L) be a polar space of type D3 .6: for every singular vector x ∈ V and any vector y ∈ V such that Ω(x. This is a proper subspace of ΠΩ (all vectors are isotropic. y) = 0 there exists a singular vector z ∈ x.8. Exercise 4. Therefore. Hint: take a scalar a ∈ F satisfying aΩ(x. δ ∈ {+. this is a polar space of type D3 . Maximal singular subspaces of G2 (V ) (stars and tops) are projective planes. as quadratic forms. Using Exercise 4. Conversely. ΠQ is a polar space. Exercise 4.4. Since every line is contained in precisely one star and precisely one top.Polar and Half-Spin Grassmannians 147 totally singular subspaces are totally isotropic. Exercise 4. 4. they give new examples of polar spaces only in the case of characteristic 2.5 we establish that ΠQ satisﬁes the axiom (3).7: there exists a base of V consisting of singular vectors. Hint: apply Exercise 4. There is natural one-to-one correspondence between lines of Gδ (Π) and points of Π (every line of Gδ (Π) consists of all elements of Gδ (Π) passing through a certain point). If R is commutative then this polar space can be obtained from the Klein quadratic form deﬁned on the 6-dimensional vector space ∧2 V [Cameron (1991).4 to a singular vector x ∈ V and a base of V formed by vectors non-orthogonal to x. Any two distinct elements of Gδ (Π) are adjacent and Gδ (Π) is a linear space. Remark 4.

−}. Therefore. . Consider the bijection f : G2 (V ) → P induced by the natural one-to-one correspondence between lines of G+ (Π) and points of Π. where V is a 4-dimensional vector space over a division ring. Exercise 4. S2 . V Exercise 4. Proof.9. This means that X is a hyperplane of Gδ (Π) and we get the claim. δ ∈ {+. since each Si intersects U in the line pl pm . If Π is a polar space of type D3 then Gδ (Π). the line of Gδ (Π) deﬁned by the point p has a non-empty intersection with X . This plane belongs to G−δ (Π).8. If p ∈ P \U then U ∩p⊥ is a line in U and the maximal singular subspace U ∩ p⊥ . S3 be a triangle in Gδ (Π). it is clear that this line is contained in X .8. S2 . Clearly. Show that A ⊂ G2 (V ) is an apartment if and only if h(A) is a frame of Π.148 Grassmannians of Classical Buildings Lemma 4. m. This is a collineation of the subspace X to the projective plane dual to U . Let h be a collineation of G2 (V ) to Π. X is a subspace of Gδ (Π) and a line of Gδ (Π) is contained in X if and only if the associated point belongs to U . Hint: use Theorem 3. in other words. S3 form a base of the projective plane X and we have established that every plane of Gδ (Π) is projective. where i = l. p belongs to Gδ (Π).7. It was established above that for every plane X of G+ (Π) there exists U ∈ G− (Π) such that X consists of all elements of G+ (Π) intersecting U in lines. Show that f is a collineation of G2 (V ) to Π and g is a collineation of Π∗ to G− (Π). p 3 = S1 ∩ S2 form a triangle in Π and span a certain plane U . p 2 = S1 ∩ S3 . If S and S are distinct elements of X then S ∩ S ∩ U is a certain point p and the associated line [p δ joins S and S . Denote by X the set consisting of all elements of Gδ (Π) which intersect U in lines (for every S ∈ Gδ (Π) the intersection S ∩ U is empty or a line). There is a one-to-one correspondence between elements of X and lines of U (every line of U is contained in precisely one element of X ). So. Then p 1 = S2 ∩ S3 . This is a one-to-one correspondence and we get another bijection g : G3 (V ) → G− (Π). Let S1 . S1 . we can suppose that G+ (Π) = ΠV . is a 3-dimensional projective space.

quadratic.7 and 4. every polar space whose rank is greater than 3 is isomorphic to the polar space associated with a certain reﬂexive sesquilinear. This result was exploited in the following classiﬁcation of polar spaces.8. Then V(Π) is a projective space. In particular. • the polar spaces of type C3 associated with Cayley algebras. this polar space is deﬁned by Klein’s quadratic form if V is a vector space over a ﬁeld. where V is a 4-dimensional vector space.3. Theorem 4. The mapping p → p⊥ is an embedding of Π in V(Π) and it is not diﬃcult to prove that this embedding transfers lines to lines.14. quadratic. and pseudo-quadratic forms. we get the following. 4. V respectively. in the case when n = 3. • the Grassmann space G2 (V ).Polar and Half-Spin Grassmannians 149 By Exercises 4. Theorem 4. If Π is a polar space of type D3 then there exists a 4-dimensional vector space V such that Π.5 ([Veldkamp (1959/1960)]). Let Π be a polar space of rank n ≥ 3.6 ([Tits (1974)]). we require in addition that Π is of type Cn and every 2-dimensional singular subspace is a Desarguesian projective plane. Moreover. or pseudo-quadratic form. G− (Π) are isomorphic to G2 (V ). Up to isomorphism there are the following three types of polar spaces whose rank is not less than three: • the polar spaces associated with non-degenerate reﬂexive sesquilinear. Proposition 4. every collineation of Π to G2 (V ) establishes a oneto-one correspondence between frames of Π and apartments of G2 (V ). .4 Embeddings in projective spaces and classiﬁcation The Veldkamp space V(Π) of a polar space Π (the rank of Π is assumed to be not less than 3) is the linear space whose points are hyperplanes of Π and the line joining hyperplanes H1 and H2 consists of all hyperplanes containing H1 ∩ H2 . G+ (Π). ΠV . Π∗ . maximal singular subspaces of such polar spaces are non-Desarguesian (Moufang) projective planes.

In what follows the associated Grassmann graphs (the collinearity graphs of Grassmann spaces) will be denoted by Γk (Π).1 Polar buildings Buildings of type Cn Let Π = (P. • for S ∈ G+ (Π) and U ∈ G− (Π) we write S ∗ U if their intersection belongs to Gn−2 (Π). −}. The complex ∆(Π) together with the set of all such apartments is a building of type Cn .2 Buildings of type Dn Now let Π = (P. Every thick building of type Dn (n ≥ 4) is isomorphic to the oriﬂamme building obtained from a polar space of type Dn [Tits (1974)]. . n − 1}. . and the half-spin Grassmannians Gδ (Π). . every thick building of type Cn (n ≥ 3) is isomorphic to the Cn -building of a rank n polar space. the corresponding Grassmann spaces are Gk (Π). k ≤ n − 3. The corresponding Grassmann spaces are Gk (Π) and Gδ (Π). k ∈ {0. By [Tits (1974)]. 4. We write G ∗ (Π) for the set of all proper singular subspaces whose dimension is not equal to n − 2 and deﬁne the oriﬂamme incidence relation ∗ on G ∗ (Π): • if S ∈ Gk (Π). L) be a polar space of rank n ≥ 3. The Grassmannians of ∆(Π) are the polar Grassmannians Gk (Π). 1. Consider the ﬂag complex ∆(Π) obtained from the set of all proper singular subspaces of Π with the natural incidence relation. . Every frame B of Π deﬁnes the subcomplex ΣB ⊂ ∆(Π) consisting of all ﬂags formed by singular subspaces spanned by subsets of B. 1.4. δ ∈ {+. The Grassmannians are the polar Grassmannians Gk (Π). we write Γk (Π) and Γδ (Π) for the associated Grassmann graphs.4 4. . .150 Grassmannians of Classical Buildings 4.4. For every frame B the corresponding apartment Orif B is the subcomplex of Orif(Π) consisting of all oriﬂamme ﬂags formed by singular subspaces spanned by subsets of B. n − 3}. This building is thick only in the case when Π is of type Cn . The associated ﬂag complex will be called the oriﬂamme complex of Π and denoted by Orif(Π). . . . and U ∈ G ∗ (Π) then S ∗ U means that S and U are incident in the usual sense. k ∈ {0. this subcomplex is called the apartment associated with (deﬁned by) the frame B. The complex Orif(Π) together with the set of all such apartments is a thick building of type Dn . L) be a polar space of type Dn and n ≥ 4.

Polar and Half-Spin Grassmannians 151 4. 2. Proof. S3 ∈ Gk (Π) form a triangle in Gk (Π) then Si ⊥ Sj for all i. if S3 is not contained in the (k + 1)-dimensional singular subspace spanned by S1 and S2 then S 1 . (2) a top-triangle: there is a (k + 1)-dimensional singular subspace containing all Si . S3 ∈ Gn−1 (Π) be mutually collinear points of Gn−1 (Π). The singular subspace S3 is maximal and we get the inclusion S1 ∩ S2 = U 3 ⊂ S3 . 3}. Let 1 ≤ k ≤ n − 2 and S1 . As for Grassmannians of ﬁnite-dimensional vector spaces. If 1 ≤ k ≤ n − 2 and S1 . Since Si is adjacent with Sj (i = j). j ∈ {1. 3} and only one of the following possibilities is realized: (1) a star-triangle: k < n − 2 and there is a (k − 1)-dimensional singular subspace contained in each Si . we have Si ⊥ Sj for all i. U 2 and U3 ⊥ S3 . 2. S3 is (k + 2)-dimensional which is possible only for k < n − 2. Then U1 := S2 ∩ S3 . U 3 ⊂ S1 . Since U 1 . Then S3 = U 1 . 2. U3 := S1 ∩ S2 belong to Gn−2 (Π). for example.5. the singular subspace spanned by S1 . in this case. U 2 ⊂ S3 . we have Ui ⊥ Uj for all i. Lemma 4. S2 . j ∈ {1. we describe triangles in the Grassmann space Gk (Π). U 1 . 3}. j ∈ {1. L) be a polar space of rank n ≥ 3.5 4. S2 ∩ S3 is a (k − 1)-dimensional singular subspace contained in S1 and S2 . 1 ≤ k ≤ n − 1. U 2 . S2 . U1 = U2 . Let S1 . S2 . Suppose that at least two of these three subspaces are distinct.10. S3 ∈ Gk (Π) be a triangle in Gk (Π). First. S2 . U2 := S1 ∩ S3 . U 3 ⊂ S2 . The Grassmann space Gn−1 (Π) does not contain triangles: any three mutually collinear points of Gn−1 (Π) are collinear if Π is of type Cn and Gn−1 (Π) does not contain triples of mutually collinear points if Π is of type Dn .1 Elementary properties of Grassmann spaces Polar Grassmann spaces Let Π = (P.

N ]k is isomorphic to an (n − k − 1)dimensional projective space. Then [M. N ]k is a maximal clique in the collinearity graph of N ]k ).10.4.10. If k < n − 2 then the subspace [M. S is collinear with all points of the line [M.5. Therefore. For every (k + 1)-dimensional singular subspace N the set N ]k is called a top. Proof. As in the previous example. N ]k . Every triangle in a top is a top-triangle. N ]k . every line of Gn−1 (Π) is a maximal singular subspace. S2 of a line [M. we suppose that k ≤ n − 2.15. In the case when k = n − 2. S2 .3. S2 form a star-triangle or a top-triangle. Let S be a k-dimensional singular subspace of Π which does not belong to [M. Example 4. As in Example 3. The case k = 0 was considered in Subsection 4. a contradiction. Gk (Π) is a gamma space. S1 . Suppose that k ≤ n − 2. N ]k and S does not belong to this line then S ⊥ N (since N is spanned by S1 and S2 ) and M = S1 ∩ S2 ⊂ S or S ⊂ N (S. Example 4. if a point of Gn−1 (Π) is collinear with two distinct points on a line then it belongs to this line and Gn−1 (Π) is a gamma space.1. Example 4. It follows from Proposition 4. By the second part of Lemma 4. N ]k is said to be a star. The star [M. In each of these cases.1 and we suppose that 0 < k < n − 1. If S ⊂ N then there exists an element of [M. Hence U1 = U2 = U3 . respectively). we show that this singular subspace is maximal. N ]k is a singular subspace of Gk (Π). all Ui are coincident and S1 . By Lemma 4. N ]k non-adjacent with S (since [M. This is a singular subspace of Gk (Π) isomorphic to a (k + 1)-dimensional projective space.15 that the class of maximal singular subspaces of Gk (Π) coincides with the class of maximal cliques of the Grassmann graph Γk (Π). Proposition 4. If S ∈ Gk (Π) is collinear with two distinct points S1 . It can be identiﬁed with a star in the Grassmann space N ]k .1. S3 are points on a line of Gn−1 (Π). this is a line.152 Grassmannians of Classical Buildings which implies that U3 is contained in each Si . Let M be a (k − 1)-dimensional singular subspace and N be a maximal singular subspace containing M . If S does not belong to N ]k then S ⊥ N .

Proof. Let X be a maximal clique of Γk (Π) and k ≤ n−2. all maximal singular subspaces of Gn−2 (Π) are tops. Example 4. Therefore. every element of [M. [M m+1 is a polar space of rank n − m − 1. [M.Polar and Half-Spin Grassmannians 153 and we take any point p ∈ N satisfying p ⊥ S. Subspaces of such type are called classical [Cooperstein (2005)]. (2). we get a line). .2. S2 then S is non-adjacent with at least one of Si . If 0 < k < n − 1 and M. This mean that S ∈ X and our clique coincides with the top N ]k . (2) if k ≤ n − 2 then every maximal singular subspace of Gk (Π) is a top or a star. Let n ≥ 4. U ]m+1 ). Suppose that m < n − 2. Kasikova and Shult (2005)]. in particular.16. If S ∈ Gk (Π) is not contained in the (k + 1)-dimensional singular subspace N := S1 . Suppose that X is not a star. The parabolic subspace [M k can be identiﬁed with the Grassmann space of index k − m− 1 associated with this polar space (every U ∈ [M k corresponds to [M. S3 ∈ X which form a top-triangle in Gk (Π). N ]k containing p (it is clear that such elements exist) is non-adjacent with S. Let M be an m-dimensional singular subspace and m < k. Proposition 4. Every triangle in a star is a star-triangle. we choose S1 . This subspace is singular only in the case when k = n − 1 and m = n − 2. Example 4. It was noted above that the class of maximal singular subspaces of Gk (Π) coincides with the class of maximal cliques of the Grassmann graph Γk (Π). S2 .7.4. The following assertions are fulﬁlled: (1) every maximal singular subspace of Gn−1 (Π) is a line. N is a pair of incident singular subspaces of Π such that dim M < k < dim N then [M. N ]k is a subspace of Gk (Π) isomorphic to the Grassmann space of a ﬁnite-dimensional vector space (in the case when dim M = k − 1 and dim N = k + 1. By Lemma 4. As in the proof of Proposition 3. N ]k is a maximal singular subspace of Gk (Π).10. subspaces of such type are called parabolic [Cooperstein.6. Then [M k is a subspace of Gk (Π). The statement (1) follows from the second part of Lemma 4.

U3 := S1 ∩ S2 belong to Gn−3 (Π). the subspace U is spanned by U1 and U2 . S2 . j). The set of all elements of Gk (Π) adjacent with both S and U is the union of all maximal cliques of Γk (Π) (maximal singular subspaces of Gk (Π)) containing S and U . U2 := S1 ∩ S3 . 4. a contradiction. S3 be a triangle in Gδ (Π). For every subset X ⊂ Gk (Π) we denote by X ∼ the set consisting of all elements of Gk (Π) adjacent with every element of X . U }∼∼ . For k = n − 2 this fails. As for the Grassmann spaces of ﬁnite-dimensional vector spaces. Then for any adjacent S. L) be a polar space of type Dn . the intersection of all these cliques is the line joining S and U . We have Ui ⊥ Uj for all i. The equality Ul = Uj (l = j) implies that Ul is contained in all Si and we get U1 = U2 = U3 which means that S1 . U3 is not less than n − 2. Suppose that dim U = n − 2 or two distinct Ui .11. Proof. for example. In each of these cases.3. δ ∈ {+. S3 are points on a line. since there is only one maximal clique (the top S. all Ui are distinct and the dimension of the singular subspace U := U1 . we have the following characterization of lines in terms of the adjacency relation if k = n − 2. in other words. Therefore. Proposition 4. Lemma 4. j ∈ {1. Then there exist unique M ∈ Gn−4 (Π) and U ∈ G−δ (Π) such that M ⊂ U . Let S1 . Then U ⊂ S3 (since U1 and U2 both are contained in S3 ) and S1 ∩ S2 = U 3 ⊂ U ⊂ S3 . 2. In the case when k = n − 2.2 Half-spin Grassmann spaces Let Π = (P. U ∈ Gk (Π) the line joining S and U coincides with the set {S. U2 . We investigate the associated half-spin Grassmann spaces Gδ (Π). . Let k = n − 2.5. −} (they are deﬁned only for n ≥ 3).154 Grassmannians of Classical Buildings Let S and U be adjacent elements of Gk (Π). If n = 3 then both Gδ (Π) are 3-dimensional projective spaces (Subsection 4. U ]n−2 ) containing S and U . S2 .3) and we will suppose that n ≥ 4. U1 and U2 are non-adjacent.17. The subspaces U1 := S2 ∩ S3 . 3} (since Ui and Uj both are contained in Sl with l = i. U3 is contained in all Si which is impossible. M = S1 ∩ S2 ∩ S3 and each Si has an (n − 2)-dimensional intersection with U .

Polar and Half-Spin Grassmannians

155

So, U is a maximal singular subspace and U1 , U2 , U3 are mutually adjacent. This means that U1 , U2 , U3 form a star-triangle in Gn−3 (Π). Then the (n − 4)-dimensional singular subspace M := U1 ∩ U2 ∩ U3 coincides with S1 ∩ S2 ∩ S3 . Each Si intersects U in the (n − 2)-dimensional subspace Ul , Um , where l, m = i. Hence U belongs to G−δ (Π). Let U be a maximal singular subspace of Π intersecting each Si in an (n − 2)-dimensional subspace. Since Π is of type Dn , U ∩ Si and U ∩ Sj (i = j) are distinct and their intersection is (n − 3)-dimensional. On the other hand, this intersection is contained in Si ∩ Sj = Ul (l = i, j) which implies that it coincides with Ul . Therefore, each Ul is contained in U and U =U . Proposition 4.18. Gδ (Π) is a gamma space. Proof. Let S1 , S2 , S3 be a triangle in Gδ (Π). We need to show that S1 is collinear with all points of the line joining S2 and S3 . If S is a point on this line then S2 ∩ S3 ⊂ S; in particular, S contains the (n − 4)-dimensional singular subspace S1 ∩ S2 ∩ S3 (Lemma 4.11). This guarantees that S1 ∩ S is (n − 3)-dimensional. We want to describe maximal singular subspaces of Gδ (Π). By Proposition 4.18, the class of maximal singular subspaces coincides with the class of maximal cliques of the associated Grassmann graph Γδ (Π). Let U ∈ G−δ (Π). We write [U ]δ for the set of all elements of Gδ (Π) intersecting U in (n − 2)-dimensional subspaces. Proposition 4.19. For every U ∈ G−δ (Π) the set [U ]δ is a maximal singular subspace of Gδ (Π) isomorphic to an (n − 1)-dimensional projective space. Singular subspaces of such kind will be called special. Proof. As in the proof of Lemma 4.9, we establish that [U ]δ is a singular subspace of Gδ (Π) and a line [T δ , T ∈ Gn−3 (Π), is contained in [U ]δ if and only if T ⊂ U . Every hyperplane of U is contained in precisely one element of [U ]δ ; this correspondence deﬁnes a collineation between [U ]δ and the projective space dual to U . So, [U ]δ is an (n− 1)-dimensional projective space.

156

Grassmannians of Classical Buildings

For every S ∈ Gδ (Π) \ [U ]δ there exists a frame B such that S and U are spanned by subsets of B (Proposition 4.7). It is not diﬃcult to construct X ∈ [U ]δ , X ⊂ B, non-adjacent with S (we leave the details for the reader). This means that [U ]δ is a maximal singular subspace of Gδ (Π). Proposition 4.20. If M ∈ Gn−4 (Π) then [M δ is a maximal singular subspace of Gδ (Π) isomorphic to a 3-dimensional projective space. Maximal singular subspace of such kind are said to be stars. Proof. If S and U are distinct elements of [M δ then they both contain the (n − 4)-dimensional singular subspace M which means that their intersection is (n−3)-dimensional and S, U are adjacent; moreover, the inclusion M ⊂ S ∩ U guarantees that the line [S ∩ U δ joining S and U is contained in [M δ . Hence [M δ is a singular subspace of Gδ (Π). Let S1 , S2 , S3 be a triangle contained in [M δ and U be the associated element of G−δ (Π) (see Lemma 4.11). Consider the subspace X := [U ]δ ∩ [M δ . If [Mi

δ

(i = 1, 2) are distinct lines of X then M1 ∩ M2 = M and Mi ⊂ U, i = 1, 2;

in other words, M1 , M2 are adjacent elements of Gn−3 (Π) and the subspace spanned by them is (n−2)-dimensional. There is a unique element of Gδ (Π) intersecting U in M1 , M2 ; it is a common point of our lines. Thus X is a projective plane and S1 , S2 , S3 form a base of X . We have proved that every plane in [M δ is projective. Suppose that T ∈ Gn−3 (Π) contains M and T ⊂ U . Then [T δ is a line of [M δ which is not contained in X . The maximal singular subspace T, U ∩ T ⊥ intersects U in the (n − 2)-dimensional subspace U ∩ T ⊥ (by Lemma 4.2); hence it belongs to X and the line [T δ has a non-empty intersection with X . This means that the plane X is a hyperplane of [M δ . Therefore, [M δ is a 3-dimensional projective space. Let S ∈ Gδ (Π) \ [M δ . As in the proof of Proposition 4.19, we take a frame B such that S and M are spanned by subsets of B and construct X ∈ [M δ , X ⊂ B,

Polar and Half-Spin Grassmannians

157

non-adjacent with S. This guarantees that [M subspace of Gδ (Π).

δ

is a maximal singular

Proposition 4.21. Every maximal singular subspace of Gδ (Π) is a star or a special subspace. Proof. We show that every maximal clique X of the Grassmann graph Γδ (Π) is a star or a special subspace. Lines are not maximal cliques of Γδ (Π) (they are contained, for example, in stars), thus X contains a triangle S1 , S2 , S3 . Let M and U be the associated elements of Gn−4 (Π) and G−δ (Π), respectively (Lemma 4.11). Clearly, [M δ and [U ]δ are the unique star and the unique special subspace containing our triangle. First, we establish that every S ∈ Gδ (Π) adjacent with all Si belongs to [M δ or [U ]δ . Since U ∩ Si is a hyperplane of Si , the subspace S intersects each U ∩ Si in a subspace whose dimension is not less than n − 4. One of the following two possibilities is realized: • S ∩ U ∩ S1 = S ∩ U ∩ S2 = S ∩ U ∩ S3 , • S ∩ U ∩ Si = S ∩ U ∩ Sj for some i, j. In the ﬁrst case, the equality

3

M=

i=1

(U ∩ Si )

guarantees that S ∩ U ∩ Si = S ∩ M for each i. Since M is (n − 4)-dimensional and the dimension of S ∩ U ∩ Si is not less than n − 4, we have M ⊂ S and S belongs to [M δ . In the second case, the dimension of S ∩ U is not less than n − 3; thus S ∩ U is (n − 2)-dimensional and S ∈ [U ]δ . Now consider S ∈ [M

δ

\ [U ]δ and S ∈ [U ]δ \ [M δ .

Then S ∩ U and S ∩ U span U (the ﬁrst subspace contains M and the second is a hyperplane of U which does not contain M ). Hence every point of S ∩ S is collinear with all points of U and we get S∩S ⊂U (since U is a maximal singular subspace). The latter inclusion implies that S ∩ S = (S ∩ U ) ∩ (S ∩ U ).

158

Grassmannians of Classical Buildings

Since S ∩ U = M (this follows from S ∈ [M δ \ [U ]δ ) and S ∩ U does not contain M , the subspaces S and S are not adjacent. So, every element of [M δ \ [U ]δ is non-adjacent with every element of [U ]δ \ [M δ . The inclusion X ⊂ [M shows that X coincides with [M

δ δ

∪ [U ]δ

or [U ]δ .

Example 4.8. Let M be an m-dimensional singular subspace of Π. Then [M δ is a subspace of Gδ (Π); subspaces of such type are called parabolic [Cooperstein, Kasikova and Shult (2005)]. In the case when m < n − 4, this subspace is non-singular. As in Example 4.6, we consider [M m+1 . This is a polar space of type Dn−m−1 . The parabolic subspace [M δ can be identiﬁed with one of the half-spin Grassmann spaces of this polar space. Exercise 4.9. Show that the intersection of two distinct stars of Gδ (Π) is empty, or a single point or a line (the second possibility is not realized if n = 4); moreover, this intersection is a line if and only if the associated (n − 4)-dimensional singular subspaces are adjacent. Exercise 4.10. Show that the intersection of two distinct special subspaces of Gδ (Π) is empty or a line, the second possibility is realized if and only if the associated elements of G−δ (Π) are adjacent. If [M δ and [U ]δ are a star and a special subspace satisfying M ⊂ U then their intersection is a plane (see the proof of Proposition 4.20). Exercise 4.11. Show that every plane of Gδ (Π) is contained in precisely one star and precisely one special subspace. Hint: take any triangle in a plane and consider the associated elements of Gn−4 (Π) and G−δ (Π) (Lemma 4.11). As above, we write X ∼ for the set consisting of all elements of Gδ (Π) adjacent with all elements of X ⊂ Gδ (Π). We have the standard characterization of lines in terms of the adjacency relation. Proposition 4.22. For any adjacent S, U ∈ Gδ (Π) the line joining S and U coincides with the set {S, U }∼∼. Proof. The intersection of all maximal singular subspaces of Gδ (Π) (stars and special subspaces) containing both S and U coincides with the line joining S and U .

If S has a non-empty intersection with L then it is collinear with all points of [L δ .1 Chow’s theorem and its generalizations Every collineation of Π to Π induces a collineation of Gk (Π) to Gk (Π ) for each k ∈ {1.7 ([Chow (1949)]). (3) and (4) hold and we verify (2). γ ∈ {+. L ) are polar spaces of same type Xn . Exercise 4. Gδ (Π) is a polar space. Proposition 4. Therefore.2) and L. γ ∈ {+.6 Collineations Throughout the section we suppose that Π = (P. Consider the half-spin Grassmann spaces of the polar space Gδ (Π): one of them consists of stars and the other is formed by special subspaces. −}. in the case when X = D. If L and S are disjoint then S ∩ L⊥ is a line of Π (Lemma 4. Proof. D} and n ≥ 3. −}. Maximal singular subspaces of Gδ (Π) (stars and special subspaces) are 3-dimensional and every plane is contained in precisely one star and one special subspace (Exercise 4.Polar and Half-Spin Grassmannians 159 A few remarks concerning polar spaces of type D4 ﬁnish the subsection. 4. we will require that n ≥ 4.12. There is the following analogue of Theorem 3. Moreover. Every isomorphism of Γn−1 (Π) to Γn−1 (Π ) is the collineation of Gn−1 (Π) to Gn−1 (Π ) induced by a collineation of Π to Π .2. If n = 4 then Gδ (Π) is a polar space of type D4 .6. . Suppose that n = 4. Theorem 4. The polar space is of type D4 . L) and Π = (P . Show that the natural one-to-one correspondence between lines of Gδ (Π) and lines of Π is a collineation of G1 (Gδ (Π)) to G1 (Π). 4. δ. Let S be a point of Gδ (Π) which does not belong to a line [L δ .11).23. Moveover. S ∩ L⊥ is the unique point on the line [L δ collinear with S. X ∈ {C. . is the collineation of Gδ (Π) to Gγ (Π ) induced by a collineation of Π to Π . . if our polar spaces are of type Dn then we get a collineation of Gδ (Π) to Gγ (Π ) with δ. L ∈ L. show that they are isomorphic to Π and G−δ (Π). if our polar spaces are of type Dn (n ≥ 5) then every isomorphism of Γδ (Π) to Γγ (Π ). . n − 1}. . respectively. The polar axioms (1).

160 Grassmannians of Classical Buildings Remark 4. −}.9.4). Then the half-spin Grassmann spaces are polar spaces of type D4 and there is natural one-to-one correspondence between their lines and lines of the polar spaces (every line of Gδ (Π). but the method works in the general case. Suppose that our polar spaces are of type D4 . Theorem 4. Using Chow’s idea. In Subsection 4.6. In [Chow (1949)] this result was established only for the polar spaces associated with reﬂexive forms. Pra˙ mowski and Zynel (2006)]).6. Pra˙ mowski and Zynel z (2006)]. It was proved in [Huang (2000)] that every surjection of Gn−1 (Π) to Gn−1 (Π ) sending adjacent elements to adjacent elements is a collineation of Gn−1 (Π) to Gn−1 (Π ). every semicollineation of Gn−1 (Π) to Gn−1 (Π ) is a collineation. in particular.8 ([Pankov. ˙ Theorem 4. δ ∈ {+. This means that a collineation f of Π to Gδ (Π ) (if it exists) induces a bijection of L to . we show that every isomorphism of Γn−2 (Π) to Γn−2 (Π ) induces an isomorphism of Γn−3 (Π) to Γn−3 (Π ) which sends stars to stars and tops to tops (Subsection 4. Remark 4.10. consists of all elements of Gδ (Π) containing a certain line of Π). Now consider the remaining case n = 4 and k = 1. If n = 4 or k = 1 then f is the collineation of Gk (Π) to Gk (Π ) induced by a collineation of Π to Π (f is a collineation of Π to Π if k = 0). It follows from elementary properties of triangles that the latter condition holds for n ≥ 5. Let f be z an isomorphism of Γk (Π) to Γk (Π ) and 0 ≤ k ≤ n − 2.8. ˙ This is a modiﬁcation of the proof given in [Pankov.3 we establish that an isomorphism of Γk (Π) to Γk (Π ) (k ≤ n−3) is induced by a collineation of Π to Π if it transfers stars to stars and tops to tops. Suppose that Π and Π both are symplectic polar spaces and write V and V for the associated (2n)-dimensional vector spaces.7 will be presented as a simple consequence of Theorem 4. The following more general version of this result was established in [Huang (2001)]: if a mapping f : Gn−1 (Π) → Gn−1 (Π ) sends adjacent elements to adjacent elements and for every S ∈ Gn−1 (Π) there exists U ∈ Gn−1 (Π) such that the distance between f (S) and f (U ) is maximal then f can be extended to the embedding of Gn (V ) in Gn (V ) induced by a semilinear embedding of V in V .

γ ∈ {+. Since Γk (Π) is a subgraph of Γw (Π). 4. see Exercise 4. k Example 4.12). If S1 and S2 are distinct non-adjacent elements of [M k and S ∈ Gk (Π) is weakly adjacent with both S1 . By this deﬁnition. Theorem 4. If Π and Π are of type D4 then every isomorphism of Γδ (Π) to Γγ (Π ).2 Weak adjacency on polar Grassmannians To prove Theorems 4. In Section 4. δ. this collineation sends stars to special subspaces and special subspaces to stars.9.10.1. The collineation f also induces a collineation between one of the half-spin Grassmann spaces of Π and G−δ (Π ) (since G−δ (Π ) is one of the half-spin Grassmann spaces of the polar space Gδ (Π ).9. It will be shown later that such collineations map all tops to stars and some stars to tops. 0 < k < n − 1.10.6. we show that every top of Gk (Π) is a maximal clique of Γw (Π).8 we will use elementary properties of so-called weak adjacency relation deﬁned on Gk (Π). If n = 4 then every isomorphism of Γ1 (Π) to Γ1 (Π ) is the collineation of G1 (Π) to G1 (Π ) induced by a collineation of Π to Π or a collineation of Π to one of the half-spin Grassmann spaces of Π .Polar and Half-Spin Grassmannians 161 L . are said to be weakly adjacent if their intersection belongs to Gk−1 (Π). S2 then S belongs to [M k . the weak Grassmann graph is connected and every k clique of Γk (Π) is a clique in Γw (Π). 0 < k < n − 1. Theorem 4. is the collineation of Gδ (Π) to Gγ (Π ) induced by a collineation of Π to Π or a collineation of Π to G−γ (Π ). Two elements of the Grassmannian Gk (Π). an easy veriﬁcation shows that this is a collineation of G1 (Π) to G1 (Π ). if S intersects S1 and S2 in distinct (k − 1)-dimensional subspaces . k Example 4. The weak Grassmann graph Γw (Π) is the graph whose vertex set is Gk (Π) k and whose edges are pairs of weakly adjacent elements. the converse fails. the second possibility can be realized only in the case when the polar spaces are of type D4 . This is a clique of Γw (Π) containing non-adjacent k elements. Indeed.9 these properties also will be exploited to study apartments preserving mappings. −}. As in Example 3. For every M ∈ Gk−1 (Π) the parabolic subspace [M k will be called a big star. any two adjacent elements of Gk (Π) are weakly adjacent.7 and 4.

Let f be an isomorphism of Γk (Π) to Γk (Π ) and k < n − 2. k Proof. Two distinct stars X . If k > 0 then f and f −1 map maximal singular subspaces (stars and tops) to maximal singular subspaces. since p1 ⊥ p2 (these points belong to S) and Si = S1 ∩ S2 . It follows directly from the characterization of lines in terms of the adjacency relation (Proposition 4. Two stars [M. k Proposition 4. 2. p i . Let X be a maximal clique of Γw (Π).8 for k < n − 2 i = 1.8 is true for k = 0. Lemma 4. big stars are maximal cliques of Γw (Π). Proof. Y such that X ∩ X is a proper subset of Y ∩ Y . we get S1 ⊥ S2 which contradicts the assumption that S1 and S2 are nonadjacent. In other words. N ]k and [M .6. pairs of adjacent stars can be characterized as pairs with maximal intersections.17) that f is a collineation of Gk (Π) to Gk (Π ) and Theorem 4. N ]k . . X are non-adjacent if and only if their intersection is empty or there exist distinct stars Y.3 Proof of Theorem 4. If any two distinct elements k of X are adjacent then it is a maximal clique of Γk (Π). since they are proper subsets of k big stars). Every maximal clique of Γw (Π) is a top or a big star. If f is a collineation of Gk (Π) to Gk (Π ). transferring tops to tops and stars to stars then it is induced by a collineation of Π to Π . If S1 and S2 are distinct non-adjacent elements of X then every element of X contains S1 ∩ S2 (see Example 4. hence X is a top (stars are non-maximal cliques of Γw (Π). k < n − 2.162 Grassmannians of Classical Buildings then we take points pi ∈ (S ∩ Si ) \ (S1 ∩ S2 ). This means that f and f −1 maps adjacent stars to adjacent stars.12. 4. Therefore.24. For every M ∈ Gk−1 (Π) [M k = N ∈ [M n−1 [M. N ]k are called adjacent if M = M and N is adjacent with N .10) and X is the big star corresponding to S1 ∩ S2 .

S2 . if k > 1 then star-triangles go to star-triangles. we get a sequence of collineations fi : Gi (Π) → Gi (Π ). Therefore. we need to show that f maps stars to stars and tops to tops. S2 . Every top-triangle is contained in precisely one maximal singular subspace (a top). If k ≥ 2 then fk−1 maps stars to stars. If S1 . . Since the dimension of stars and tops is equal to n − k − 1 and k + 1 (respectively). . fk−1 is an isomorphism of Γk−1 (Π) to Γk−1 (Π ) such that fk−1 and the inverse mapping send tops to tops (this is a collineation of Π to Π if k = 1). Suppose that n = 2k + 2. in the case when k > 1. As in the proof of Theorem 3.6).2. except the case when n = 4 and k = 1. S3 ∈ Gk (Π) form a startriangle then the singular subspace S1 . .Polar and Half-Spin Grassmannians 163 Any two elements of [M n−1 can be connected by a path of Γn−1 (Π) contained in [M n−1 (the parabolic subspace [M n−1 is isomorphic to the Grassmann space consisting of maximal singular subspaces of a certain polar space. 0. k ⇔ f (S) ∈ [fk−1 (M ) k ⇔ fk−1 (M ) ∈ f (S)]k−1 . S3 is (k + 2)-dimensional. top-triangles go to top-triangles and we get the claim. Therefore. this singular subspace is not maximal and the star-triangle is contained in more than one star. this is true if n = 2k + 2. Example 4. It is clear that f ([M k ) = [fk−1 (M ) k for every M ∈ Gk−1 (Π) (we apply the same arguments to f −1 ) and the mapping fk−1 : Gk−1 (Π) → Gk−1 (Π ) is bijective. such that fk = f and establish that every fi is induced by f0 . . This implies the existence of a subspace fk−1 (M ) ∈ Gk−1 (Π ) such that the associated big star of Gk (Π ) contains f ([M k ). i = k. . Since for every S ∈ Gk (Π) M ∈ S]k−1 ⇔ S ∈ [M we have fk−1 ( S]k−1 ) = f (S)]k−1 . So.

By Lemma 4. This mapping transfers a line [S n−1 . If f (S1 ) and f (S2 ) are not weakly adjacent then dim(M1 ∩ M2 ) < n − 3 and dim(g −1 (M1 ) ∩ g −1 (M2 )) < n − 3. . M2 be maxima singular subspaces of Π such that M1 ∩ M2 = f (S1 ) ∩ f (S2 ) and f (Si ) ⊂ Mi for i = 1. S2 be weakly adjacent elements of Gn−2 (Π). Since f and f −1 transfer tops to tops.164 Grassmannians of Classical Buildings 4.8). this collineation induces f . to the line [f (S) Hence g is a collineation of Gn−1 (Π) to Gn−1 (Π ) and dim(g(M ) ∩ g(N )) = dim(M ∩ N ) for all M. Theorem 4. h is induced by a collineation of Π to Π . 2 (Corollary 4.12.6.8 is proved and we use it to prove Theorem 4. we establish that f −1 preserves the weak adjacency relation.7. big stars go to big stars in both directions.2). In the case when n = 3.7 and 4.8 Let f be an isomorphism of Γn−2 (Π) to Γn−2 (Π ). Since g is induced by f .4 Proof of Theorems 4. Since maximal cliques of Γn−1 (Π) and Γn−1 (Π ) are lines of the associated Grassmann spaces. we have Si ⊂ g −1 (Mi ). This means that f induces a bijection h : Gn−3 (Π) → Gn−3 (Π ). n−2 n−2 Let S1 . and the latter inequality contradicts the fact that S1 and S2 are weakly adjacent. If n ≥ 4 then h is a collineation of Gn−3 (Π) to Gn−3 (Π ) preserving the types of all maximal singular subspaces. Similarly. Now we show that f is an isomorphism of Γw (Π) to Γw (Π ). S ∈ Gn−2 (Π). i = 1. this is a collineation of Π to Π which induces f . N ∈ Gn−1 (Π) (the latter equality follows from the distance formula given in Proposition 4. Let also M1 . every isomorphism t between these graphs induces n−1 . Maximal cliques of these graphs are tops and f induces a bijection g : Gn−1 (Π) → Gn−1 (Π ). 2.

9. under this assumption. Hence f maps lines to lines and induces a bijection of Gn−3 (Π) to Gn−3 (Π ). 4. if the polar spaces are of type C4 then f and f −1 both map stars to stars (hence tops go to tops) and f is induced by a collineation of Π to Π . moreover. Then f is a collineation of Gδ (Π) to Gγ (Π ) (by Proposition 4.3 that f is a collineation of G1 (Π) to G1 (Π ). Our proof will be based on some trivial observations concerning pairs of triangles. 0. It was noted in Subsection 4.3) form a regular pair then one of the following possibilities is realized: .12). every line from each of these triangles is adjacent with precisely two lines of the other.6. every triangle is contained in precisely one maximal singular subspace. t is induced by a collineation of Π to Π . Li = Lj ∀ i. We say that triangles ∆ = {L1 . this collineation induces t. L2 . k = 1).8 and get the claim.22. L3 } and ∆ = {L1 . In the present case (n = 4. (4. f is induced by a collineation of Π to Π if it preserves the types of maximal singular subspaces (Lemma 4.13. we draw Theorem 4. Suppose that f is an isomorphism of Γ1 (Π) to Γ1 (Π ). Therefore. lines of half-spin Grassmann spaces can be characterized in terms of the adjacency relation). Preliminaries. δ. Similarly. we establish that our polar spaces are of type D4 and f is induced by a collineation of Π to one of the half-spin Grassmann spaces of Π . L3 }.8. Now we suppose that the image of a certain star is a top. Since two distinct (n − 3)-dimensional singular subspaces of Π or Π are adjacent if and only if the associated lines of the half-spin Grassmann space Gδ (Π) or Gγ (Π ) span a plane. Lemma 4.10 from Theorem 4. this bijection is an isomorphism of Γn−3 (Π) to Γn−3 (Π ). L2 . in other words. γ ∈ {+. j.6. By Theorem 4.5 Proof of Theorem 4. In the case when n ≥ 5.9 Throughout the subsection we assume that n = 4.Polar and Half-Spin Grassmannians 165 ˜ a bijection t of Gn−2 (Π) to Gn−2 (Π ). Now suppose that our polar spaces are of type Dn and f is an isomorphism of Γδ (Π) to Γγ (Π ). we apply Theorem 4.3) in G1 (Π) form a regular pair if Li and Lj are adjacent only in the case when i = j. If the triangles (4. This bijection is an isomorphism of Γn−2 (Π) to Γn−2 (Π ) (two distinct lines have a non-empty intersection if and only if the associated (n − 2)-dimensional singular subspaces are ad˜ jacent). −}.

166

Grassmannians of Classical Buildings

• There exist a plane S and a point p ∈ S, p ⊥ S, such that one of the triangles is a star-triangle in [p, p, S ]1 and the other is a top-triangle in S]1 . • There exist maximal singular subspaces U, U intersecting precisely in a point p and such that ∆ and ∆ are star-triangles in [p, U ]1 and [p, U ]1 , respectively. Proof. Direct veriﬁcation.

We say that triangles ∆ = {L, L1, L2 } and ∆ = {L, L1 , L2 }, Li = Lj ∀ i, j, (4.4)

in G1 (Π) form a regular pair if Li and Lj are adjacent only in the case when i = j. Lemma 4.14. If the triangles (4.4) form a regular pair then one of the following possibilities is realized: • There exist two planes S, S intersecting precisely in L and such that S ⊥ S and ∆, ∆ are top-triangles in S]1 and S ]1 , respectively. • There exist two maximal singular subspaces U, U intersecting precisely in L and a point p ∈ L such that ∆, ∆ are star-triangles in [p, U ]1 and [p, U ]1 , respectively. Proof. Direct veriﬁcation.

1. Let f be an injection of G1 (Π) to G1 (Π ) preserving the adjacency relation: two elements of G1 (Π) are adjacent if and only if their images are adjacent. Then f transfers maximal singular subspaces of G1 (Π) (stars and tops) to subsets of maximal singular subspaces of G1 (Π ). Every maximal singular subspace of G1 (Π ) contains at most one image of a maximal singular subspace of G1 (Π) (otherwise there exist non-adjacent elements of G1 (Π) whose images are adjacent). We will assume that f satisﬁes the following condition: (A) triangles go to triangles. It is clear that two triangles in G1 (Π) form a regular pair if and only if their images form a regular pair in G1 (Π ). The condition (A) guarantees that the image of every maximal singular subspace of G1 (Π) is contained in precisely one maximal singular subspace of G1 (Π ) (since triangles are bases of maximal singular subspaces).

Polar and Half-Spin Grassmannians

167

Suppose that the image of a certain star [p, U ]1 , U ∈ G3 (Π), p ∈ U , is contained in a top S ]1 , S ∈ G2 (Π ). Our ﬁrst step is the following. Lemma 4.15. There exists a maximal singular subspace U of Π containing S and satisfying the following conditions: • for every point q ∈ U there is a plane S(q) ⊂ U such that f ([q, U ]1 ) ⊂ S(q)]1 , • for every plane S ⊂ U there is a point q(S) ∈ U such that f ( S]1 ) ⊂ [q(S), U ]1 . Proof. Let S be a plane in U which does not contain the point p. Consider any regular pair of triangles ∆1 ⊂ [p, U ]1 and ∆2 ⊂ S]1 . Their images also form a regular pair in G1 (Π ). By our hypothesis, f (∆1 ) is a top-triangle contained in S ]1 . Then Lemma 4.13 guarantees that f (∆2 ) is a star-triangle; moreover, there exist a maximal singular subspace U of Π and a point q(S) ∈ U such that q(S) ∈ S ⊂ U and f (∆2 ) is contained in the star [q(S), U ]1 . The latter means that f ( S]1 ) is a subset of [q(S), U ]1 . Similarly, for another plane T ⊂ U which does not contain p, the image of the top T ]1 is contained in a certain star [q(T ), U ]1 and q(T ) ∈ S ⊂ U . If U and U are distinct then their intersection is S . Since the points q(S) and q(T ) do not belong to S , [q(S), U ]1 ∩ [q(T ), U ]1 = ∅. On the other hand, f ( S]1 ) ⊂ [q(S), U ]1 and f ( T ]1 ) ⊂ [q(T ), U ]1 ; the tops S]1 and T ]1 have a non-empty intersection (since S ∩ T is a line), a contradiction. Therefore, U coincides with U . Now, let q ∈ U \ {p}. We choose a plane S ⊂ U which does not contain the points q and p. Then f ( S]1 ) is contained in the star [q(S), U ]1 (it was established above). We consider any regular pair of triangles ∆1 ⊂ [q, U ]1 and ∆2 ⊂ S]1 .

168

Grassmannians of Classical Buildings

Their images form a regular pair and f (∆2 ) is a star-triangle in [q(S), U ]1 . If f (∆1 ) is a star-triangle then, by Lemma 4.13, f transfers the star [q, U ]1 to a subset of a certain star [q(S), U ]1 such that U ∩ U = q(S). Since S ⊂ U , the latter equality guarantees that [q(S), U ]1 ∩ S ]1 = ∅. However, [p, U ]1 and [q, U ]1 have a non-empty intersection (the line qp) and their images are contained in S ]1 and [q(S), U ]1 , respectively. Thus f (∆1 ) is a top-triangle. This means that f ([q, U ]1 ) is contained in a certain top S(q)]1 with S(q) ⊂ U . Consider a plane S ⊂ U containing the point p and any point q ∈ U \ S. Then f ([q, U ]1 ) is contained in the top S(q)]1 . Using Lemma 4.13, we establish that f ( S]1 ) is a subset of a certain star [q(S), U ]1 such that q(S) ∈ S(q) ⊂ U . We take any plane T ⊂ U which does not contain the points p and q. The image of the top T ]1 is contained in the star [q(T ), U ]1 and q(T ) ∈ S(q). As above, we show that U coincides with U . Since distinct maximal singular subspaces go to subsets of distinct maximal singular subspaces, the mappings q → S(q) and S → q(S) (we deﬁne S(p) := S ) are injective. In the case when f is a collineation of G1 (Π) to G1 (Π ), the inclusions in Lemma 4.15 must be replaced by the equalities. Moreover, the inverse mapping f −1 sends every star [q(S), U ]1 , S ∈ U ]2 , to the top S]1 . We apply the arguments from the proof of Lemma 4.15 to f −1 and establish that q → S(q) is a one-to-one correspondence between points of U and planes of U ; similarly, S → q(S) is a one-to-one correspondence between planes of U and points of U . 2. Suppose that f is a collineation of G1 (Π) to G1 (Π ) and U is as in the previous step. Let Q be a maximal singular subspace of Π intersecting U in a certain plane S. The image of the top S]1 is the star [q(S), U ]1 . Let q ∈ Q \ S. We take any regular pair of triangles ∆1 ⊂ S]1 and ∆2 ⊂ [q, Q]1 . Then f (∆1 ) is a star-triangle in [q(S), U ]1 . By Lemma 4.13, there are the following two possibilities for the triangle f (∆2 ):

Polar and Half-Spin Grassmannians

169

• f (∆2 ) is a top-triangle then f ([q, Q]1 ) coincides with a certain top T ]1 with T ⊂ U , • f (∆2 ) is a star-triangle then f ([q, Q]1 ) is a star [q(S), Q ]1 such that Q intersects U precisely in the point q(S). The ﬁrst possibility is not realized, since for any plane T ⊂ U the top T ]1 is the image of a certain star [t, U ]1 . Now, suppose that q ∈ S. If the image of the star [q, Q]1 is a top then, by the ﬁrst step of our proof, the same holds for all points of Q which is impossible. Therefore, f ([q, Q]1 ) is a star for every point q ∈ Q. Let M be a third maximal singular subspace containing S. We take points q ∈ Q \ S and t ∈ M \ S. Then every element of [q, Q]1 is nonadjacent with every element of [t, M ]1 . However, f ([q, Q]1 ) = [q(S), Q ]1 contains elements adjacent with some elements of f ([t, M ]1 ) = [q(S), M ]1 . This means that there are no 3 distinct maximal singular subspaces containing S and our polar spaces are of type D4 . 3. Suppose that the polar spaces are of type D4 and, as in the ﬁrst step, f is an injection of G1 (Π) to G1 (Π ) preserving the adjacency relation and satisfying the condition (A). We also assume that the maximal singular subspaces U and U belong to Gδ (Π) and Gγ (Π ), δ, γ ∈ {+, −}, respectively. Let Q be an element of Gδ (Π) adjacent with U . The intersection of U and Q is a certain line L. We take a point q ∈ L and consider any regular pair of star-triangles ∆1 = {L, L1 , L2 } ⊂ [q, U ]1 and ∆2 = {L, L1 , L2 } ⊂ [q, Q]1 . The image of the star [q, U ]1 is contained in the top S(q)]1 and f (∆1 ) is a top-triangle. By Lemma 4.14, f (∆2 ) also is a top-triangle which means that f ([q, Q]1 ) is a subset of a certain top. Lemma 4.15 implies the existence of a maximal singular subspace Q of Π satisfying the following conditions: for every point q ∈ Q there is a plane T (q) ⊂ Q such that f ([q, Q]1 ) ⊂ T (q)]1 , and for every plane S ⊂ Q there is a point t(S) ∈ Q such that f ( S]1 ) ⊂ [t(S), Q ]1 .

Since every plane of Π is contained in a certain element of Gδ (Π).170 Grassmannians of Classical Buildings Now we choose two planes S ⊂ U and M ⊂ Q intersecting in L and satisfying S ⊥ M . the maximal singular subspaces associated with these stars are elements of Gδ (Π) and Gγ (Π ). L2 } ⊂ S]1 and ∆2 = {L.14 implies that q(S) = t(M ) and U ∩ Q = f (L). L1 . 4. Consider any regular pair of top-triangles ∆1 = {L. • q → SM (q) is a one-to-one correspondence between points of M and planes of g(M ). Thus we get a mapping g : Gδ (Π) → Gγ (Π ) satisfying the following conditions: if M ∈ Gδ (Π) then for any point q ∈ M there is a plane SM (q) ⊂ g(M ) such that f ([q. we assume that the polar spaces are of type D4 and f is an injection of G1 (Π) to G1 (Π ) preserving the adjacency relation and satisfying the condition (A). g(M )]1 . similarly. Lemma 4. L1 .11. The half-spin Grassmann spaces are connected and the same holds for every Q ∈ Gδ (Π). Remark 4. respectively. In the case when f is a collineation of G1 (Π) to G1 (Π ). and for any plane S ⊂ M there is a point qM (S) ∈ g(M ) such that f ( S]1 ) ⊂ [qM (S). As in the previous step. In particular. The mapping g sends adjacent elements to adjacent elements. S → qM (S) is a one-to-one correspondence between planes of M and points of g(M ). Since these star-triangles form a regular pair. Then f (∆1 ) ⊂ [q(S). • the mapping g is bijective. M ]1 ) ⊂ SM (q)]1 . If f is a collineation of G1 (Π) to G1 (Π ) then f and f −1 map tops to stars. but it does not need to be injective. . L2 } ⊂ M ]1 . we apply the same arguments to f −1 and get the following: • the latter two inclusions must be replaced by the equalities. Q ]1 . Q is an element of Gγ (Π ) adjacent with U . U ]1 and f (∆2 ) ⊂ [t(M ). the images of all tops are subsets of stars (in the general case).

Now. Every element of [q. N ]1 distinct from the line M ∩ N . The stars [q. The mapping f −1 transfers the tops Si ]1 to stars [qi . T ∈ Gδ (Π).5) ). Since each g(M ) ∈ Gγ (Π ). Therefore. L1 .5) are contained in a certain maximal singular subspace Uq (the set (4. assume that f is a collineation of G1 (Π) to G1 (Π ). M ]1 and [q. 2. Show that h is bijective. L1 . M ]1 is a clique of the weak Grassmann graph is adjacent with a certain element of [q. This maximal clique is the top Uq ]2 . Denote by h the mapping of P to G−γ (Π ) which sends every point q ∈ P to the maximal singular subspace Uq .5) contains more than one element and there is only one maximal singular subspace satisfying this condition). L2 } ⊂ S1 ]1 and ∆2 = {L . Let M and N be distinct elements of [q δ . Qi ]1 . The latter guarantees that SM (q) ⊥ SN (q) and these planes are adjacent. Hence T belongs to [q δ and X coincides with (4. all elements of (4. M ]1 . N ]1 have a non-empty intersection (the line M ∩ N ). This means that the planes SM (q) and SN (q) are weakly adjacent. T ]1 . S2 be distinct planes contained in T . Let T ∈ G−γ (Π ) and S1 .Polar and Half-Spin Grassmannians 171 We take any point q ∈ P and consider the associated star [q δ in the half-spin Grassmann space Gδ (Π). It has a non-empty intersection with every star [q. which implies that t = q.5).5). For any regular pair of top-triangles ∆1 = {L . For every element 2 of X the mapping f −1 transfers the associated top to a star [t. intersects Uq in the plane SM (q). 5. The intersection of these planes is a certain line L . both Qi are elements of Gδ (Π). M ]1 is contained in the top SM (q)]1 . L2 } ⊂ S2 ]1 Γw (Π 2 . The latter guarantees that the mapping h is injective. Let X be a maximal clique of Γw (Π ) containing (4. M ∈ [q δ . we have Uq ∈ G−γ (Π ). M ∈ [q δ . For every M ∈ [q δ the image of the star [q. i = 1. Hence every element of SM (q)]1 is adjacent with an element of SN (q)]1 distinct from f (M ∩ N ). Thus the set {SM (q)}M∈ [q δ (4.

Q1 ]1 and f −1 (∆2 ) ⊂ [q2 . Suppose z that a bijection f : Gk (Π) → Gk (Π ).172 Grassmannians of Classical Buildings the star-triangles f −1 (∆1 ) ⊂ [q1 . Thus h transfers every line L ∈ L to the line of G−γ (Π ) associated with the line f (L). T = Uq for a certain point q ∈ P and we get the claim. Let L be a line of Π. Similar to Proposition 3. For every point q ∈ L there exists a star [q. n − 1} be distinct numbers.9. ˙ Theorem 4. Q ∈ Gδ (Π).6 Remarks In this subsection we establish some results closely related with Theorems 4. Proposition 4. since L belongs to this star. . . M ⊂ Uq .25.12. Pra˙ mowski and Zynel (2006)]). m ∈ {0. . suppose that the line f (L) is contained in Uq . 4. This means that h is a collineation of Π to G−γ (Π ) and f is induced by this collineation. Let k. Conversely. Suppose that gk : Gk (Π) → Gk (Π ) and gm : Gm (Π) → Gm (Π ) are bijections such that S ∈ Gk (Π) and U ∈ Gm (Π) are incident if and only if gk (S) and gm (U ) are incident. As a consequence. we have q1 = q2 . In the case when f is not a collineation. Therefore. . The mapping f −1 sends this top to a star [q.8 and 4. Q]1 . Proof. Then gk and gm are induced by the same collineation of Π to Π . we get the following. By Lemma 4. Q]1 . . Remark 4. containing L. The image of this star is a top M ]1 with M ⊂ Uq and we obtain that f (L) ⊂ Uq .4. Q2 ]1 form a regular pair.11 ([Pankov. Then f (L) is an element of a certain top M ]1 . the mapping h transfers every line L ∈ L to a subset of the line of G−γ (Π ) associated with f (L). we have q ∈ L.6.14. k < n − 1.

The collineation of G1 (Π) to G1 (Π ) considered in the previous subsection (the case when Π and Π are of type D4 ) does not preserve . Hence f induces a bijection of P to P . Pra˙ mowski and Zynel (2006)]). U ∈ X . Consider the case when n = 4 and k = 1.16. can be characterized as a maximal subset X ⊂ Gk (Π) satisfying the following condition: S⊥U ∀ S. Remark 4. Proof.Polar and Half-Spin Grassmannians 173 preserves the relation ⊥: S ⊥ U ⇐⇒ f (S) ⊥ f (U ). By Lemma 4. If S and U are adjacent then the big star [S ∩ U k and the top S. This implies that S and U are adjacent. Proof. Then it is the collineation of Gk (Π) to Gk (Π ) induced by a collineation of Π to Π . This is the required collineation of Π to Π . This means that f induces a bijection of Gn−1 (Π) to Gn−1 (Π ) and we get a pair of bijections satisfying the condition of Proposition 4.12 ([Pankov. ˙ Theorem 4. Two distinct elements S. U ]k are as required. U ∈ Gk (Π).16. We will investigate isomorphisms between the weak Grassmann graphs and need the following lemma.13. Then every isomorphism of Γk (Π) to Γw (Π ) is induced by a k collineation of Π to Π . The mappings f and f −1 send tops to tops and big stars to big stars (since any two distinct elements of a top are adjacent and a big star contains non-adjacent elements). f is an isomorphism of Γk (Π) to Γk (Π ). Let f be an isomorphism of Γw (Π) to Γw (Π ). 1 ≤ k ≤ n − 2. Let 1 ≤ z w k ≤ n − 2. Conversely. Then f and f −1 k k transfer maximal cliques (tops and big stars) to maximal cliques.25. Lemma 4. it is induced by a collineation of Π to Π if n = 4 or k = 1. suppose that there are two distinct maximal cliques of Γw (Π) containing both S and U . S ∈ Gn−1 (Π). are adjacent if and only if they belong to the intersection of two distinct maximal cliques of Γw (Π). k Proof. Since the intersection k of two distinct maximal cliques of the same type (two tops or two big stars) is empty or a single point. Therefore. Every subspaces S]k . one of these cliques is a big star and the other is a top.

7. 4. .11 ([Kwiatkowski and Pankov (2009)]). . . in other words. xn . xn . x2 . . the following two conditions are not equivalent: (1) S1 . . x3 . . x3 . . qn be a frame of the polar space ΠΩ such that pi ⊥ qi for each i. . . xn y1 . Two points of Π (elements of G0 (Π)) are opposite if and only if they are non-collinear. qi = yi and Ω(xi . two elements of a polar or half-spin Grassmannian are said to be opposite if the distance between them is maximal (is equal to the diameter of the associated Grassmann graph). where Uq is a maximal singular subspace of Π . .174 Grassmannians of Classical Buildings the relation ⊥ and it is not an isomorphism between the weak Grassmann graphs. We consider a (2n)-dimensional vector space V over a ﬁeld R of characteristic 2 and a non-degenerate alternating form Ω deﬁned on V . . The direct analogue of Theorem 3. y1 .7 Opposite relation As for Grassmannians of ﬁnite-dimensional vector spaces. . 4. .6 does not hold in the general case. xi ) = 1. . yi ) = Ω(yi . . S2 } such that every U ∈ Gk (Π) opposite to S is opposite to at least one of Si . Two maximal singular subspaces are opposite if and only if they are disjoint. For every point q ∈ P it transfers the big stars [q 1 to Uq ]1 . . q1 . . . . . . The maximal singular subspaces of ΠΩ corresponding to the maximal totally isotropic subspaces x1 . x3 . (2) there exists S ∈ Gk (Π) \ {S1 . . .1 Opposite relation on polar Grassmannians Let Π be a polar space of rank n. . . . yn ∈ V satisfying pi = xi . . S2 ∈ Gk (Π) are adjacent. and x1 + y2 . y2 . . Example 4. Let p1 . x2 + y1 . xn . There exist vectors x1 . . pn .

S2 ∈ Gδ (Π). p = ax1 + bx2 + z and q = ay2 + by1 + z for some a. and S (respectively). Remark 4. Their intersection is the (n−3)-dimensional singular subspace N associated with the totally isotropic subspace x3 .2 Opposite relation on half-spin Grassmannians The direct analogue of Theorem 3. .14 (M.7. δ ∈ {+.6 holds for half-spin Grassmannians. Clearly. otherwise. This means that every element of Gn−1 (ΠΩ ) opposite to S is opposite to at least one of Si . b ∈ R and z . this construction is impossible. and it is thin if p ⊥ q and Π is of type Dn . Kwiatkowski). . . z ∈ N . It is not diﬃcult to show that all Veldkamp lines are thick if Π is the polar space deﬁned by an alternating or Hermitian form. In the case when the characteristic is not equal to 2. the line is thick if p ⊥ q. Thus if U ∈ Gn−1 (ΠΩ ) intersects both S1 and S2 then it intersects S. . the polar space associated with an alternating or Hermitian form.3. Let p and q be distinct points of Π. −}. S2 . S1 and S2 are not adjacent. However.Polar and Half-Spin Grassmannians 175 will be denoted by S1 . .13 ([Kwiatkowski and Pankov (2009)]). We say that the Veldkamp line (the line in the associated Veldkamp space. Thus the conditions (1) and (2) are not equivalent if k = 0 and Π is. then for any distinct S1 . Indeed. 4. this point belongs to S if s = t = 1. Theorem 4. Subsection 4. q. then every point non-collinear with t is non-collinear with p or q. n ≥ 4.4) joining p⊥ and q ⊥ is thick if it contains t⊥ . every point of the line p q is of type s(ax1 + bx2 ) + t(ay2 + by1 ) + z with z ∈ N and s. it is said to be thin. t = p. t ∈ R. the following conditions are equivalent: (1) S1 and S2 are adjacent. We assert that for any collinear points p ∈ S1 \ N and q ∈ S2 \ N the line p q intersects S. If Π is a polar space of type Dn . for example. S2 } such that every U ∈ Gδ (Π) opposite to S is opposite to at least one of Si . (2) there exists S ∈ Gδ (Π) \ {S1 . xn . The points p and q satisfy (2) if and only if the Veldkamp line joining p⊥ and q ⊥ is thick: we take any point t such that t⊥ belongs to the Veldkamp line.

We will distinguish the following two cases: • n is even then the dimension of the intersection of two elements of Gδ (Π) is odd and S. δ. Our ﬁrst step is the following statement: for every distinct collinear points pi ∈ Si . (2) =⇒ (1). This implies the existence of lines Li ⊂ Pi (i = 1. 2. U ∈ Gδ (Π) are opposite if and only if their images are opposite) is a collineation of Gδ (Π) to Gγ (Π ). U ∈ Gδ (Π) are opposite if and only if S ∩ U = ∅.6 given in Subsection 3. If n is even then U intersects S1 and S2 in subspaces whose dimensions are not less than 1. Consider the case when n is odd.2. Since S1 ∩ S2 ⊂ S and S is opposite to U . U does not intersect at least one of Si which contradicts the inclusion p1 p2 ⊂ U . Let U be the maximal singular subspace spanned by points of the frame and disjoint from S. these intersections both are 0-dimensional (points). Now consider the case when n is odd. (1) =⇒ (2). these lines do not intersect S1 ∩ S2 . the line p1 p2 intersects S. S2 are adjacent and show that every S = S1 . According our assumption. i = 1. If Π and Π are polar spaces of type Dn . −}. Suppose that U ∈ Gδ (Π) is opposite to S. We assert that there exists a plane containing p1 p2 and . By our hypothesis.4. As above. We take lines Li contained in U ∩ Si . • n is odd then the dimension of the intersection of two elements of Gδ (Π) is even and S. S2 belonging to the line of Gδ (Π) joining S1 and S2 is as required. Since U is opposite to S and S1 ∩ S2 ⊂ S. preserving the relation to be opposite (S. U ∈ Gδ (Π) are opposite if and only if S ∩ U is a single point. n ≥ 4. The latter means that L1 ⊥ L2 which contradicts the fact that these lines are contained in U . i = 1. Let Pi be planes contained in U ∩ Si . Our proof of Theorem 4.176 Grassmannians of Classical Buildings Corollary 4. Proof.3. 2) which do not intersect S1 ∩ S2 . The intersection of S1 and S2 is non-empty.13 is a modiﬁcation of the proof of Theorem 3. then every bijection of Gδ (Π) to Gγ (Π ).6) is not simpler. 2. If p1 p2 does not intersect S then we take any frame of Π whose subsets span S and the line p1 p2 . Suppose that S1 . the dimensions of U ∩ S1 and U ∩ S2 are not less than 2. 2. Hence Si is spanned by S1 ∩ S2 and Li . we get L1 ⊥ L2 . This is an element of Gδ (Π) opposite to S. but it is non-opposite to each Si . a veriﬁcation of the conditions proposed in [Huang and Havlicek (2008)] (see Remark 3. Suppose that n is even. Each of these planes has a non-empty intersections with S1 ∩ S2 (because S1 ∩ S2 is (n − 3)-dimensional). γ ∈ {+. i = 1.

the latter contradicts the fact that P intersects each Si in a line. If p1 p2 does not intersect S1 ∩ S2 or the subspace S1 ∩ S2 contains more than one point (in the second case the dimension of S1 ∩ S2 is not less than 2) then we take any point t ∈ S1 ∩ S2 which does not belong to p1 p2 . since S1 ∩ p⊥ is (n − 2)-dimensional and the dimension of S1 ∩ S2 is equal to 0). P ∩ S contains a line and this line intersects p1 p2 (since P is a plane). Since P ⊂ U . Let P be a plane containing p1 p2 and intersecting each Si in a line. Then any point t ∈ S1 \ S2 collinear with p2 gives the claim (such point exists. In the ﬁrst case. the subspace S ∩ S1 is (n − 3)-dimensional. Our second step is to prove the equality dim(S ∩ Si ) = n − 3. i = 1. we set H := S1 ∩ p⊥ . v are distinct. The lines L and u v both are contained in the plane L. 2. we take any hyperplane H ⊂ S1 . The inclusion u v ⊂ S guarantees that L intersects S. v and the points u . thus they have a non-empty intersection. Deﬁne U := (S ∩ S1 ) ∩ (S ∩ S2 ). Since S ∩ S1 and S ∩ S2 are (n − 3)-dimensional subspaces of S. Since S ∩ H ⊂ S ∩ S1 . v be distinct points on a line L ⊂ H. The subspace H is (n − 2)-dimensional and the dimension of S ∩ H is not less than n − 3. Suppose that S1 ∩ S2 = {p1 }. using this frame. The lines up and vp intersect S in points u and v . Therefore. we have p = u . we show that S ∩ S2 is (n − 3)-dimensional. So. p . . every line of H has a nonempty intersection with S. Let u. respectively.Polar and Half-Spin Grassmannians 177 intersecting each Si in a line. we construct U ∈ Gδ (Π) opposite to S and containing P . the plane spanned by p1 p2 and t is as required. Since p ∈ S. Let us take a point p ∈ S2 \ S. Now we establish the equality dim(S1 ∩ S2 ) = n − 3 which completes our proof. in the second case. Then S1 ∩ p⊥ is a hyperplane of S1 or p ∈ S1 ∩ S2 and S1 ∩ p⊥ coincides with S1 . Similarly. If S1 ∩ S2 is a single point belonging to the line p1 p2 then this point coincides with p1 or p2 . 2 The case when S1 ∩ S2 = {p2 } is similar. If dim(P ∩ S) ≤ 0 then we take a frame whose subsets span P and S. Then U is opposite to at least one of Si . there are the following three possibilities: • S ∩ S1 = S ∩ S2 and U is (n − 3)-dimensional.

p2n } be a frame of Π. 1. k ∈ {0. Hence this case is not realized. .178 Grassmannians of Classical Buildings • dim U = n − 4. p is a line. . There are 2n possibilities to choose pi1 . We take any line L ⊂ S1 \ S and consider the subspace L⊥ ∩ S2 . . The apartment Ak consists of all k-dimensional singular subspaces pi1 . . . • dim U = n − 5 and S is spanned by S ∩ S1 and S ∩ S2 . . .1 Apartments Apartments in polar Grassmannians Let Π = (P. the later equality implies that L ⊥ S which is impossible (S is a maximal singular subspace and L ⊂ S). Since S is spanned by S ∩ S1 and S ∩ S2 . Ak consists of n 2n · (2n − 2) · · · (2n − 2k) = 2k+1 k+1 (k + 1)! elements. . if U is an (n − 5)-dimensional subspace distinct from S1 ∩ S2 then S1 ∩ S2 is (n − 3)-dimensional. in particular. n−1}. the dimension of S1 ∩ S2 is equal to n − 3 in the ﬁrst and second cases. . . Proposition 4. . 4. pik+1 such that {i1 . . . L⊥ ∩ S2 contains a point p ∈ S. ik+1 } ∩ {σ(i1 ). Since the order of the points is not taken into account. . By U ⊂ S1 ∩ S2 . σ(ik+1 )} = ∅. we show that the intersection of S with the plane L. . . If every point of this subspace belongs to S then L⊥ ∩ S2 is contained in S ∩ S2 . It consists of all k-dimensional singular subspaces spanned by subsets of the frame B. . Therefore. . then pi2 can be chosen in 2n − 2 ways and so on.8 4. thus S ∩ S2 = L ⊥ ∩ S2 . Proof.26. Moreover. This line intersects L which contradicts the inclusion L ⊂ S1 \ S. The subspace S ∩ S2 is (n − 3)-dimensional and the dimension of L⊥ ∩ S2 is not less than n − 3. . |Ak | = 2k+1 n k+1 . Using arguments of the second step. Suppose that U = S1 ∩ S2 is (n − 5)-dimensional. Denote by Ak the associated apartment of Gk (Π). .8. A0 coincides with B. L) be a polar space of rank n and B = {p1 . .

k ∈ {0. n − 1}. k). −1. In the case when 0 < k < n − 1. The maximal cliques of Γ(An−1 ) are pairs of adjacent elements. . U }. Every frame of Π spans a 2 proper subspace U in ΠV . . 1. Show that for any S. we also require that their union is singular. U ∈ Gk (Π) the intersection of all apartments of Gk (Π) containing S and U coincides with {S. k). T ∈ Ak+1 (if k = n − 2 then the ﬁrst possibility is not realized). The associated apartment of Gk (Π) is contained in the subspace { S ∈ Gk (Π) : S ⊂ U } By Lemma 4. Then P J(n. −n}. . Two such subsets are connected by an edge if their intersection consists of k elements. this is a proper subspace of Gk (Π). Consider the set J = {1. 0). . where Ω is a non-degenerate tracevalued reﬂexive form deﬁned on an n-dimensional vector space V . M ∈ Ak−1 . . Sn ∈ An−1 adjacent with S and S ∩ S1 . . The following example shows that the Grassmann space Gk (Π) does not need to be spanned by apartments of Gk (Π). Example 4. and the tops T ]k . . n. N ∈ An−1 . . a subset of P = G0 (Π) is a frame of Π if the restriction of Γ0 (Π) to this subset is isomorphic to P J(n. n − 1) coincides with the n-dimensional hypercube graph Hn . . We deﬁne the polar Johnson graph P J(n. . . every maximal clique of Γ(Ak ) is an independent subset spanning a maximal singular space of Gk (Π). and in the case when k < n − 1. . Recall that a subset X ⊂ J is called singular if j ∈ X =⇒ −j ∈ X. . the maximal cliques of Γ(Ak ) are the intersections of Ak with the stars [M.Polar and Half-Spin Grassmannians 179 Exercise 4. .12. . The Witt index of Ω is assumed to be less than n . . N ]k . . S ∩ Sn form a base of the top S]n−2 . .7. for every S ∈ An−1 there are precisely n elements S1 .13. . For every subset X ⊂ Gk (Π) we denote by Γ(X ) the restriction of the Grassmann graph Γk (Π) to X . as the graph whose vertex set is formed by all singular subsets consisting of k + 1 elements. Then Γ(Ak ) is isomorphic to the polar Johnson graph P J(n. Suppose that Π = ΠΩ . . . By the deﬁnition.

S]n−m if m < n and a base of the singular subspace S if m = n.14 ([Cooperstein. Sm . in particular. If m = n then X is an apartment of Gn−1 (Π). this is a base of [M (S). Then M (S) := S ∩ S1 ∩ · · · ∩ Sm is an (n−m−1)-dimensional singular subspace and for every i ∈ {1. By Lemma 4. . . the natural collineation of [M k to the Grassmann space of index k − m − 1 associated with the polar space [M m+1 (Example 4. Consider a subset X ⊂ Gn−1 (Π) and for every S ∈ X denote by XS the set of all elements of X adjacent with S.4. Kasikova and Shult (2005)]). we denote them by S1 . . U ∈ XS . The set formed by all Xi will be denoted by B0 (S). the intersection of [M k with the associated apartment of Gk (Π) is called an apartment of our parabolic subspace. in the case when m < n. the restriction of Γn−1 (Π) to an apartment of [M n−1 is isomorphic to the (n − m − 1)-dimensional hypercube graph. Let X be a locally independent subset of Gn−1 (Π) such that Γ(X ) is isomorphic to the m-dimensional hypercube graph. The proof is similar to the proof of Theorem 3. U and non-adjacent with S (this follows directly from the fact that Γ(X ) is isomorphic to Hm ). It is not diﬃcult to see that S ∩ S ∩ U = U ∩ S ∩ U. Apartments of Gn−1 (Π) and apartments in parabolic subspaces of Gn−1 (Π) are locally independent. m < k.6) establishes a one-to-one correspondence between apartments. m} the singular subspace (S ∩ Sj ) Xi := j=i is (n − m)-dimensional. Theorem 4. . . The restrictions of the Grassmann graph Γk (Π) to apartments of [M k are isomorphic to the polar Johnson graph P J(n − m − 1. this is an apartment in a parabolic subspace of Gn−1 (Π). is an independent subset of S]n−2 . Proof. Consider the parabolic subspace [M k . . For every S ∈ X adjacent with S and distinct from U there exists unique U ∈ X adjacent with both S . Let S ∈ X . .180 Grassmannians of Classical Buildings Let M be an m-dimensional singular subspace of Π. There are precisely m elements of X adjacent with S. Let U be an element of X adjacent with S. k − m − 1). We take any frame of Π which contains a subset spanning M .8. We say that X is locally independent if for every S ∈ X the set of all S∩U . . . 2 ≤ m ≤ n.

since X is not contained in S ∩ S (X ∈ B0 (S)). Kasikova and Shult (2005)]. we have X ⊂ U ∩ S which means that X ∈ B0 (U ) ⊂ B(U ).14 to prove an analogue of Theorem 3. in what follows this (n − m − 1)-dimensional singular subspace will be denoted by M . If X ∈ B0 (S) then X ∈ B(U ) (since B0 (S) is contained in B(U ) by the deﬁnition). moreover. this is a frame of Π and X is the associated apartment of Gn−1 (Π). By connectedness. 4. In the general case. If m < n then B is a frame of the polar space [M n−m and X is the associated apartment of the parabolic subspace [M n−1 . However. therefore. B(S) = B(U ) ∀ S. By this reason. the Grassmann space Gn−1 (Π) and its parabolic subspaces are not spanned by apartments. Consider the case when X ∈ B0 (S ) \ B0 (S) for a certain S ∈ X \ {U } adjacent with S. denote this set by B.1. this is a frame of the polar space [M n−m if m < n and a frame of Π if m = n. We take U ∈ X adjacent with both S and U . By connectedness of polar Grassmann spaces. B0 (S) ∩ B0 (U ) consists of m − 1 elements and there is a unique element of B0 (S) which does not belong to B0 (U ). In the case when m = n. . U ∈ X . U ∈ X are adjacent.2 Apartments in half-spin Grassmannians Now suppose that Π is a polar space of type Dn and n ≥ 4. (2) every X ∈ B0 (S) satisfying X ⊂ S ∩ U belongs to B0 (U ). −}.15. Let B be a frame of Π and An−1 be the associated apartment of Gn−1 (Π).8. in some special cases this is possible. Remark 4.Polar and Half-Spin Grassmannians 181 This implies the following: (1) M (S) = M (U ). see [Cooperstein. we have M (S) = M (U ) ∀ S. Then B(S) consists of 2m elements. U ∈ X . we show that B(S) coincides with B(U ) if S. δ ∈ {+. Then S ∩ S and U ∩ S are distinct hyperplanes of S . For every S ∈ X we denote by B(S) the union of all B0 (U ) such that U ∈ X is adjacent with S. Then Aδ := An−1 ∩ Gδ (Π). Now. we cannot use Theorem 4.

It is clear that Γ(Aδ ) is isomorphic to the halfcube graph 1 Hn . Proposition 4. The vertex set of the hypercube graph Hn (the set of all singular subsets consisting of n elements) can be decomposed in two disjoint subsets J+ and J− (Example 2. The maximal cliques of Γ(Aδ ) are the intersections of Aδ 2 with the stars [S δ . U ∈ Gδ (Π) the intersection of all apartments of Gδ (Π) containing S and U coincides with {S. Exercise 4. −}.26. U ]m+1 in the polar space [M m+1 . Show that for any S. By Lemma 4.14. This follows from Proposition 4. Note that 1 H4 is isomorphic to P J(4.182 Grassmannians of Classical Buildings is the apartment of the half-spin Grassmannian Gδ (Π) associated with the frame B. U }.8). δ ∈ {+. The deﬁnition of apartments in [M δ is standard.27. We identify every U ∈ [M δ with the subspace [M. since |A+ | = |A− | and An−1 is the disjoint union of A+ and A− . U ∈ G−δ (Π). The distance between any two vertices from Jδ . [M m+1 is a polar space of type Dn−m−1 and this correspondence is a collineation of [M δ to one of the half-spin Grassmann spaces of [M m+1 . Let M be an m-dimensional singular subspace of Π and m < n − 4. The half-cube graph 1 2 Hn is the graph whose vertex set is Jδ and whose edges are pairs of vertices at the distance 2 (this construction does not depend on δ ∈ {+. this collineation establishes a one-to-one correspondence between apartments. |Aδ | = 2n−1 . 2 . is even. 0).2). and the special subspaces [U ]δ . In the present case. The distance between two vertices is odd if and only if one of them belongs to J+ and the other belongs to J− . they will be called stars and special subsets of Aδ (respectively). 2 If X ⊂ Gδ (Π) then we write Γ(X ) for the restriction of the Grassmann graph Γδ (Π) to the set X .4. −}). Every maximal clique of Γ(Aδ ) is an independent subset spanning a maximal singular space of Gδ (Π). Consider the parabolic subspace [M δ of the half-spin Grassmann space Gδ (Π) (Example 4. Proof. The restrictions of the Grassmann graph Γδ (Π) to apartments of [M δ are isomorphic to the halfcube graph 1 Hn−m−1 . S ∈ An−4 .

If m = n then X is an apartment of Gδ (Π). The intersection of all U ∩ p⊥ . Proof. and X be a subset of Gδ (Π). U ∩ p⊥ i for every i. it is clear that I = ∅. If Π is of type D4 then the half-spin Grassmann spaces of Π are polar spaces of type D4 and every apartment of Gδ (Π). n ≥ 4. . Corollary 4. If Π is a polar space of type Dn . −}. then half-spin Grassmann space Gδ (Π). Then Si belongs to [pi δ or coincides with U . −}. . is a frame of Gδ (Π). δ ∈ {+. . this follows immediately from Proposition 4. Then A ∩ [pi δ is the apartment of [pi δ associated with the frame of [pi 1 formed by all lines pi pj with j = i. Now consider S ∈ Gδ (Π) such that pi ∈ S for all i. there is a point collinear with all points of the frame B which contradicts Corollary 4. δ ∈ {+.9 and Corollary 4.Polar and Half-Spin Grassmannians 183 Theorem 4.4. If Π is a polar space of type D4 then the family of all frames of the polar space Gδ (Π). δ ∈ {+. is empty i (otherwise. this is an apartment in a parabolic subspace of Gδ (Π). Every [pi 1 is a polar space of type Dn−1 and [pi δ can be identiﬁed with one of the half-spin Grassmann spaces of this polar space. 2 and every maximal clique of Γ(X ) is an independent subset of the half-spin Grassmann space Gδ (Π). . such that Γ(X ) is isomorphic to the half-cube graph 1 Hm .15 gives the following. n ≥ 4.28. is spanned by every apartment of Gδ (Π). Proposition 4. i ∈ I. δ ∈ {+. Kasikova and Shult (2005)]).4. Let Π be a polar space of type Dn . −}. We take any U ∈ G−δ (Π) intersecting S in an (n − 2)-dimensional subspace and deﬁne Si := pi . coincides with the family of all apartments of Gδ (Π). Theorem 4. Suppose that n ≥ 5 and prove the statement induction by n. σ(i). We denote by I the set of all i such that pi ∈ U . in the case when m < n. the second possibility is realized if pi ∈ U . Denote by X the subspace of Gδ (Π) spanned by A.15 ([Cooperstein. [pi δ is spanned by A ∩ [pi δ and we have [pi δ ⊂X for every i. This means that the projective space U ]n−2 is . By the inductive hypothesis. p2n } be a frame of Π and A be the associated apartment of Gδ (Π). Let B = {p1 . In the case when n = 4. Since frames are independent subsets of polar spaces. 4 ≤ m ≤ n.1). −}.

we get S ∈ [U ]δ ⊂ X which completes our proof.3 Proof of Theorem 4. −}.28 guarantees that parabolic subspaces are spanned by apartments. We take any frame B of Π and consider the associated apartments A ⊂ Gm−1 (Π ). Then the special subspace [U ]δ is i spanned by the set of all Si . The graphs Γ(A+ ) and Γ(X ) are isomorphic to P J(4.8. is parabolic. . .3. . and it is not diﬃcult to prove that one of the following possibilities is realized: . Since Si ∈ X for every i ∈ I. Let f : A+ → X be an isomorphism of Γ(A+ ) to Γ(X ). Every maximal clique Z of Γ(X ) is contained in precisely one maximal clique of Γδ (Π) (since Z is an independent subset of Gδ (Π) containing more than 3 elements and the intersection of two distinct maximal cliques of Γδ (Π) is not greater than a plane). 4. Proposition 4. 0) and every maximal clique of these graphs consists of 4 elements. Corollary 4. Sj are of diﬀerent types (one of them is a star and the other is a special subset) for every j ∈ {1. respectively. i}. Aδ ⊂ Gδ (Π ). M ∈ Gm (Π). m < n. −}. For any maximal cliques S and S of Γ(A+ ) there is a sequence of maximal cliques S = S0 . we get the following. As in Section 3. Every subspace of Gδ (Π). Every special subset S ⊂ A+ consists of m elements.184 Grassmannians of Classical Buildings spanned by the set of all U ∩ p⊥ . Si = S such that |Sj−1 ∩ Sj | = 3 and Sj−1 . i ∈ I. isomorphic to the half-spin Grassmann space associated with a polar space of type Dm . we say that Z is a star of X or a special subset of X if the maximal clique of Γδ (Π) containing Z is a star or a special subspace. If f (S) is a star then m = 4. S1 . . can be identiﬁed with one of the half-spin Grassmann spaces of the polar space [M m+1 such that apartments correspond to apartments. Therefore. . .15 Let Π be a polar space of type Dm . Consider the case when m = 4. . i ∈ I. δ ∈ {+.5. δ ∈ {+. If the intersection of two distinct maximal cliques of Γδ (Π) contains a triangle then these cliques are of diﬀerent types. . f transfers special subsets of A+ to special subsets of X if m ≥ 5. Since every parabolic subspace [M δ .

U ∈ A− .14. For every S ∈ Y \ {U } adjacent with S there exists unique U ∈ X adjacent with both S and U . . . . This implies that M (S) = M (U ). . In the second case. This is an isomorphism of Γ(A) to Γ(X ∪ Y). Sm ∈ X ∪ Y adjacent with S (all Si belong to Y if S ∈ X . X ∩ [S]δ = {S1 . and the latter is possible only in the case when f −1 (S) ∈ [f −1 (U )]+ . The dimension of the subspace M (S) := S ∩ S1 ∩ · · · ∩ Sm is not less than n − m − 1. S ∈ X and U ∈ Y are adjacent (as elements of Gn−1 (Π)) if and only if S ∈ [U ]δ . Sm } is an independent subset of [S]δ . we have S ∩ S ∩ U = U ∩ S ∩ U.14. Now. As in the proof of Theorem 4. and they are elements of X if S ∈ Y). suppose that S ∈ X .Polar and Half-Spin Grassmannians 185 (1) special subsets go to special subsets and stars go to stars. we consider S 1 . we can assume that f transfers special subsets to special subsets in all cases. So. Therefore. indeed. We deﬁne f (U ) := U and denote by Y the subset of G−δ (Π) consisting of all f (U ). . Let S ∈ X ∪ Y. . Γ(X ∪ Y) is isomorphic to the m-dimensional hypercube graph and we want to show that X ∪ Y is a locally independent subset of Gn−1 (Π). then f g is an isomorphism of Γ(A+ ) to Γ(X ) satisfying (1). . we take any automorphism g of Γ(A+ ) satisfying (2) (we leave its construction for the reader). So. Then for every U ∈ A− there exists U ∈ G−δ (Π) such that f (A+ ∩ [U ]+ ) = X ∩ [U ]δ . In the case when S ∈ Y. Hence M (S) is (n − m − 1)-dimensional which guarantees that all S ∩ Si form an independent subset of S]n−2 . f is extended to a bijection of A to X ∪ Y. . This implies that all S ∩ Si form an independent subset of S]n−2 and the subspace M (S) is (n−m−1)-dimensional. . We take any U ∈ Y adjacent with S. (2) special subsets go to stars and stars go to special subsets. As in the proof of Theorem 4.

L ) are polar spaces of same type Xn . this is an apartment in Gn−1 (Π) (if m = n) or an apartment in a parabolic subspace of Gn−1 (Π) (if m < n). L) and Π = (P .15 is diﬀerent from the original proof given in [Cooperstein. n ≥ 4. Pankov 2 (2007)]. δ. δ. • the collineations of Gδ (Π ) to Gγ (Π ). induced by collineations of Π to G−γ (Π ).17. γ ∈ {+. we have also the following two types of additional collineations: • the collineations of G1 (Π) to G1 (Π ) induced by collineations of Π to the half-spin Grassmann spaces of Π . then every apartments preserving bijection of Gδ (Π) to Gγ (Π ). By Theorem 4. the adjacency relation will be characterized in terms . Theorem 4. By Corollary 4. if X = D then we require that n ≥ 4. 4. we suppose that Π = (P. induced by collineations of Π to Π . X ∈ {C. γ ∈ {+. −}. Remark 4. Our proof of Theorem 4. δ. Remark 4.16 (M. Every apartments preserving bijection of Gk (Π) to Gk (Π ) is a collineation of Gk (Π) to Gk (Π ) (it is a collinearion of Π to Π if k = 0).1 Apartments preserving bijections The collineations of Gk (Π) to Gk (Π ) induced by collineations of Π to Π are apartments preserving. This result was proved in [Pankov 3 (2007)].9 Apartments preserving mappings In this section.16. D} and n ≥ 3. The proof is based on elementary properties of maximal inexact and complement subsets. If our polar spaces are of type Dn . these collineations are apartments preserving. Pankov). −}. In the case when Π and Π are polar spaces of type D4 .4.14. Kasikova and Shult (2005)]. 4. for some partial cases it was established in [Pankov 4 (2004). −}.9. γ ∈ {+. This gives the claim. If our polar spaces are of type Dn then the same holds for the collineations of Gδ (Π) to Gγ (Π ). is a collineation of Gδ (Π ) to Gγ (Π ). X ∪ Y is locally independent.186 Grassmannians of Classical Buildings Therefore.

. . . . R is said to be inexact. . Let R ⊂ A. in opposite to Grassmannians of ﬁnitedimensional vector spaces. this makes our proof more complicated. If k = n − 1 then every element of A contains precisely one of the points pi or pσ(i) for each i.2 Inexact subsets of polar Grassmannians Let B = {p1 . . . If there exist distinct i. . σ(ik+1 )} = ∅. However. . . we will write Si (R) = ∅ in the case when the intersection of R and A(+i) is empty.7. . . For any i1 . We write A(+i) and A(−i) for the sets consisting of all elements of A which contain pi and do not contain pi . . . Note that A(+i) = A(−σ(i)) if k = n − 1. . If R ∩ A(+i) is not empty then we deﬁne Si (R) as the intersection of all elements of R containing pi . where {i1 . . All apartments preserving mappings will be described in Subsection 4. ik+1 } ∩ {σ(i1 ). certain polar Grassmanians contain two diﬀerent types of complement subsets.9. +is . However. . p2n } be a frame of Π and A be the associated apartment of Gk (Π). . pik+1 . otherwise. Lemma 4. ju belonging to {1. . respectively. j such that pj ∈ Si (R) and pσ(i) ∈ Sσ(j) (R). in contrast to Grassmannians of ﬁnite-dimensional vector space. Then R is inexact. . . . −j1 . . . . is and j1 . 2n} we deﬁne A(+i1 . . Let R ⊂ A. . −ju ) as the intersection A(+i1 ) ∩ · · · ∩ A(+is ) ∩ A(−j1 ) ∩ · · · ∩ A(−ju ). Recall that A consists of all k-dimensional singular subspaces pi1 . If Si (R) = pi for all i then the subset R is exact. . 4. . . . .9.17.Polar and Half-Spin Grassmannians 187 of complement subsets. . the converse fails. . . . . We say that R is exact if there is only one apartment of Gk (Π) containing R.

pσ(i) containing S. . pi . pσ(i) }. 2n} there exists a point pi = pi . Hence p is non-collinear with pi and pσ(i) . there exists a maximal singular subspace S = S. The point pi is collinear with a unique point of the line pσ(i) pσ(j) . pσ(i) such that (B \ {pi }) ∪ {pi } and (B \ {pσ(i) }) ∪ {pi } are frames of Π. Then (B \ {pi . . The case when S ⊥ U is trivial. Let S and U be disjoint (n − 2)-dimensional singular subspaces spanned by subsets of B \ {pi . Lemma 4. pσ(i) }.19. j = i. pσ(j) }) ∪ {pi . Proof. S. .18. If Π is a polar space of type Cn then for each i ∈ {1. On the line pi pj we choose a point pi distinct from pi and pj . In particular. If S and U are singular subspaces spanned by subsets of B then S ⊥ ∩ U is spanned by a subset of B. Remark 4. pσ(j) } is a frame of Π. . Lemma 4. Proof. pσ(i) } ⊂ S ∪ U. Suppose that S ⊥ U and denote by X the set of all points from U ∩ B which are not contained in S ⊥ . Since p ⊥ (S ∪ U ) and B \ {pi . Since Π is a polar space of type Cn . The singular subspace spanned by (U ∩ B) \ X is contained in S ⊥ ∩ U . σ(i) is an inexact subset of B (frames are apartments in the Grassmannian G0 (Π) = P ). Lemma 4. It is clear that j = σ(i).188 Grassmannians of Classical Buildings Proof. denote this point by pσ(j) . If this subspace does not coincide with S ⊥ ∩ U then S ⊥ has a non-empty . we established that B \ {pi .2 implies the existence of a point p ∈ S satisfying p ⊥ U . The point pi := p is as required. pj }. p is collinear with all points of B \ {pi . This frame deﬁnes a new apartment of Gk (Π) containing the subset R. This point does not belong to S (for every point q ∈ S we have q ⊥ U ).18.

For every k ∈ {0. Proposition 4. In this case. the subset A(−i) coincides with A(+σ(i)). this inexact subset is not maximal. pj ∈ U. Thus S ⊥ does not intersect X and S ⊥ ∩ U is spanned by (U ∩ B) \ X.Polar and Half-Spin Grassmannians 189 intersection with X .29. we get Sj (A(−i) ∪ {U }) = pj . If k = n − 1 then A(−i) is inexact. Suppose that k = n − 1. n − 2} the following assertions are fulﬁlled: (1) A(−i) is a maximal inexact subset if Π is of type Cn .7) (it is not diﬃcult to choose two elements of A(−i) whose intersection is precisely the point pj ). Therefore. Sj (A(−i) ∪ {U }) = pj pσ(i) . . (2) A(−i) is exact if Π is of type Dn . . then pσ(j) ∈ Si (A(−i) ∪ {U }) and pσ(i) ∈ Sj (A(−i) ∪ {U }).6). σ(i) the intersection of all elements of A(−i) containing pj is the line pj pσ(i) .17. Then pσ(i) ⊥ X which contradicts pi ∈ X .6) is a base of X . There exists pi ∈ X such that (X \ {pi }) ∪ {p} (4. pσ(j) belongs to U . Proof. By Lemma 4. pj ∈ U. Let p be a point belonging to this intersection. This means that A(−i) is inexact. but this inexact subset is not maximal. the subset A(−i) ∪ {U } is inexact. We have Sj (A(−i)) = pj for j = i (4. In the second case. Let us take any U belonging to A \ A(−i) = A(+i). . Then Si (A(−i) ∪ {U }) = U. It is clear that pσ(i) belongs to S and it is collinear with all points of the base (4. Consider the case when k ≤ n − 2. moreover. Since for every j = i. Sσ(i) (A(−i) ∪ {U }) = Sσ(i) (A(−i)) = pσ(i) . if A(−i) is contained in the apartment of Gk (Π) associated with a frame B = B then B = (B \ {pi }) ∪ {p} . .

Let U be an element of A \ A(−i) = A(+i). pσ(i) } which implies pi = p. −σ(i)).19. −j) ∪ A(+σ(j). pσ(i) } then the points p. σ(i). Then U does not belong to A(−i.18. pσ(i) such that (B \ {pi }) ∪ {pi } is a frame of Π.6). Now we prove the following equality A \ Rij = A(+i. pi is the unique point of U collinear with all points pσ(i1 ) . −σ(j)) and we have U ∈ A(+i) or U ∈ A(+σ(j)). pσ(i) deﬁne three distinct maximal singular subspaces containing S which is impossible if Π is of type Dn . there is a point pi = pi . see Proposition 4. the inexact subset A(−i) is maximal. . . Then U is spanned by pi and some points pi1 .7) guarantees that the subset A(−i) ∪ {U } is exact. +σ(j)) ∪ A(−i. . By Lemma 4. . pσ(i) } and non-collinear with pσ(i) . . pσ(ik ) and (4. The subset A(−i) is inexact. It is clear that Rij = A(+i.190 Grassmannians of Classical Buildings and p = pi . j = i. pi . Since this is true for every U ∈ A \ A(−i). U ∈ A(+i. −j). Consider the subset Rij := A(+i. In the ﬁrst case. . . The point p is collinear with all points of B \ {pi . Let U be an element of A \ Rij .8) Proof. Thus A(−i) is exact if Π is of type Dn . . (4. +j) ∪ A(−i) if k = n − 1 and Rij = B \ {pi . The associated apartment of Gk (Π) contains A(−i). +j) ∪ A(+σ(i). pik . Suppose that Π is a polar space of type Cn . Also we have p ⊥ pi (otherwise the point p is contained in every maximal singular subspace spanned by pi and a subset of B \ {pi . By Lemma 4. pσ(j) } if k = 0. +j) =⇒ U ∈ A(+i. −σ(j)). . If S is an (n − 2)-dimensional singular subspace spanned by a subset of B \ {pi .

29) which means that Rij is a maximal inexact subset. This means that Sl (Rij ) = pl if l = i. since pi is the unique point on the line Si (Rij ) = pi pj collinear with pσ(j) .Polar and Half-Spin Grassmannians 191 In the second case. By (4. −σ(i)) do not intersect Rij . Lemma 4.29) and the inexact subset Rij is not maximal. −j) intersects Si (Rij ) = pi pj precisely in the point pi . Since A(+i. The subset Rij is inexact. −σ(i)) intersects Sσ(j) (Rij ) = pσ(j) pσ(i) precisely in the point pσ(j) . A \ Rij ⊂ A(+i. since pσ(j) is the unique point on the line Sσ(j) (Rij ) = pσ(j) pσ(i) collinear with pi . Rij ∪ {U } is exact for each U ∈ A \ Rij and the inexact subset Rij is maximal. the set Rij ∪ {U } is exact. Since Si (Rij ) = pi pj and Sσ(j) (Rij ) = pσ(j) pσ(i) . the set Rij ∪ {U } is exact. Suppose that k ≥ 1. σ(j) we can choose two elements of Rij whose intersection is pl (we leave the details for the reader). the subsets A(−i) = B \ {pi } and A(−σ(j)) = B \ {pσ(j) } are exact (Proposition 4. except the case when k = 0 and Π is of type Cn . If Π is of type Cn then A(−i) = B \ {pi } is an inexact subset containing Rij (Proposition 4. σ(j). Therefore. +σ(j)) =⇒ U ∈ A(+σ(j). −σ(i)). Let k = 0. U ∈ A(+σ(i). In the case when Π is of type Dn . −j) ∪ A(+σ(j). Proof. for every U belonging to A \ Rij one of the following possibilities is realized: • U ∈ A(+i. For every l = i. pσ(j) } is an inexact subset (Remark 4. .18). • U ∈ A(+σ(j).17 implies that the subset Rij is inexact. Therefore. it is a maximal inexact subset. −σ(i)).30. moreover. the inverse inclusion holds. Then Rij = B \ {pi . Proposition 4. −j) and A(+σ(j).8).

pi . we consider the case when all Si (R) are non-empty. Therefore. . pσ(i) and S. . In particular.192 Grassmannians of Classical Buildings If the set A(−i) is inexact (see Proposition 4. . the following assertions are fulﬁlled: (1) if k = n − 1 then each maximal inexact subset is of second type. . pσ(i) }) ∪ {pi . Let k ≥ 1 and R be a maximal inexact subset of A. Mu := Sσ(ju ) (R) do not contain the point pσ(l) . Then pl is collinear with all points of the subspaces M1 . Let k = 0. S. The statement is trivial if Π is of type Cn . . (4) if k = 0 and Π is of type Cn then each maximal inexact subset is of ﬁrst type. Suppose that for certain l ∈ I the subspace Sl (R) is spanned by pl . pju and the subspaces M1 := Sσ(j1 ) (R). . . . First. there exist i ∈ I and j = i. S.19 implies that R is exact. n − 2} and Π is of type Cn then each maximal inexact subset is of ﬁrst type or of second type. . . .29) then it will be called an inexact subset of ﬁrst type. pσ(i) } (we leave the details for the reader) and B \ {pi . pi . (3) if k ∈ {0. . .30) then it is said to be an inexact subset of second type. We need to show that B \ {pi . σ(i) such that pj ∈ Si (R) and pσ(i) ∈ Sσ(j) (R). . pσ(i) } the pairs of subspaces S. The subset R is inexact and I is non-empty. . pσ(i) } is a frame. . . Assume that there exist points pi . pl is the unique point of Sl (R) collinear with all points of M1 . pj1 . .19). pσ(i) are coincident (since Π is of type Dn ). Proposition 4. pσ(i) } is exact. . pσ(i) } coincides with {pi . if Rij is inexact (see Proposition 4. . pσ(i) } is exact. . Mu . . This means that {pi . Mu (Lemma 4. (2) if k ∈ {1. Proof. Since pj1 ⊥ M1 . . . Similarly. Every maximal inexact subset is of ﬁrst type or of second type. For every (n − 2)-dimensional singular subspace S spanned by a subset of B \ {pi . Denote by I the set of all numbers i such that the dimension of Si (R) is greater than zero. Suppose that Π is of type Dn . pju ⊥ Mu . If the same holds for every l ∈ I then Lemma 4. . n − 2} and Π is of type Dn then each maximal inexact subset is of second type. pσ(i) such that (B \ {pi . .31. . . .

As in the case when k = 0. pσ(i) }. σ(i). σ(i) is non-empty and does not contain pσ(i) . (2) All Sl (R). σ(i) such that the subspace Sj (R) contains pσ(i) . pj1 . . Let l ∈ I. R is a proper subset of a maximal inexact subset which contradicts our assumption. . . pσ(i) } which implies that B = B . these points are distinct from pi and pσ(i) ) and the subspaces M1 := Sσ(j1 ) (R). pju (by our assumption. +j) ⊂ Rij . . σ(i). +σ(i)) ∪ A(−j. If I is empty then (4. for every U ∈ R one of the following possibilities is realized: • pi ∈ U then U ∈ A(+i. −i). since the inexact subset R is maximal. Then R ⊂ A(−j. If Sl (R) is spanned by pl . are non-empty and there exists j = i. I = ∅. Then R is contained in A(−i) and we get a maximal inexact subset of ﬁrst type if Π is of type Cn . • pi . We have R = Rij . . Since the subset R is inexact. j = i. R is a proper subset of Rj σ(i) .19. l = i.Polar and Half-Spin Grassmannians 193 Then R is contained Rij . We denote by I the set of all numbers j = i. +σ(j)) ⊂ Rij . Since every U ∈ R \ A(+j) is contained in A(−j. As in the previous case. . +σ(i)). j. As above. we get {pi .9) B = (B \ {pi . . Therefore. If the same holds for all elements of I then. m = i. Mu . we get (4. there exists a frame B = B such that R is contained in the associated apartment of Gk (Π). −σ(j)) ⊂ Rij . The inclusion R ⊂ A(−i) ∩ Rjm means that R is a proper subset of Rjm which is impossible. In the cases (1) and (2). . we establish that R is contained in a certain complement subset Rjm . σ(i) such that the dimension of Sj (R) is greater than zero. we obtain the inclusion R ⊂ A(+j. (3) Each Sj (R). −i) and every U ∈ R ∩ A(+j) belongs to A(+j. Consider the case (3). σ(i). by Lemma 4. . Indeed. • pσ(j) ∈ U then U ∈ A(+σ(i). −i) and R is a proper subset of Rj σ(i) . Now suppose that Si (R) = ∅ for certain i. pσ(j) ∈ U then U ∈ A(−i. . pσ(i) }) ∪ {pi .9) again. Mu := Sσ(ju ) (R) do not contain pσ(l) then pl is the unique point of Sl (R) collinear with all points of M1 . . . pσ(i) } = {pi . . If the polar space is of type Dn then A(−i) is exact and one of the following possibilities is realized: (1) Sj (R) = ∅ for certain j = i.

U be distinct elements of A. U ) = mc . The follows assertions follow immediately from Proposition 4. U ) : S. each complement subset is of such type. . (2) if k < n − 1 and n = 4 or k = 1 then S. (3) if k ∈ {0. this is a pair of collinear points. Note that Cij = Cσ(j)σ(i) . In the case when k = n − 1 or the polar space is of type Dn .8). n − 2} and Π is of type Cn then each complement subset is of ﬁrst type or of second type. . . −σ(i)) which coincides with A(+i. (2) if k ∈ {1. By (4.194 Grassmannians of Classical Buildings 4. U are adjacent if and only if c(S. If S. in terms of complement subsets of second type. +σ(j)) if k = n − 1. U ∈ A.20. U ) for the number of compliment subsets of second type in A containing both S and U . . (4) if k = 0 and Π is of type Cn then each complement subset is of ﬁrst type (a single point). If A(−i) is a maximal inexact subset (this fails in some cases) then the associated complement subset is A(+i).31: (1) if k = n − 1 then each complement subset is of second type. U ) = mc − 1. respectively. Lemma 4. .9. . . k ≥ 1. . the maximal inexact subset Rij (k > 0 or Π is of type Dn ) gives the complement subset Cij := A(+i. S = U }. n − 2} and Π is of type Dn then each complement subset is of second type. in the case when k = 0. −j) ∪ A(+σ(j). A complement subset is said to be of ﬁrst type or of second type if the corresponding maximal inexact subset is of ﬁrst type or of second type. We say that R ⊂ A is a complement subset if A \ R is a maximal inexact subset. . U are weakly adjacent if and only if c(S. Now we characterize the adjacency and weak adjacency relations of Gk (Π). U are distinct elements of A then we write c(S. The following assertions are fulﬁlled: (1) S. Let k ≥ 1 and S.3 Complement subsets of polar Grassmannians Let A be as in the previous subsection. Let us deﬁne mc := max { c(S.

U ∈ A and m be the dimension of S ∩ U . −j) and A(+σ(j). Let S. U }. Suppose that S ∈ A(+i. thus c(S. U . U if and only if the line pi pj is contained S ∩ U . . −σ(i)) if and only if pi . −j) \ A(+σ(j). −σ(i)) contains precisely one of the subspaces. U ∈ A(+i. (C) each of the subsets A(+i. U ) = m+1 2 and we get the claim. pσ(j) ∈ S ∩ U. −j) and A(+σ(j). −j). Thus there are precisely m+1 2 distinct Cij = Cσ(j)σ(i) satisfying (A). If S and U belong to A(+i. (B) only one of the subsets A(+i. −j) ∩ A(+σ(j). We have S. If k = n − 1 then the complement subset A(+i. −j) or A(+σ(j). Hence there are 2n − 2(k − m) − 2(m + 1) = 2(n − k − 1) possibilities for j. −σ(i)) \ A(+i. +j) contains both S.Polar and Half-Spin Grassmannians 195 Proof. The case (B). Suppose that 1 ≤ k ≤ n − 2. pi ∈ S ∩ U and there are m + 1 possibilities for i. −σ(i)) does not contain the subset {S. −σ(i)) and U ∈ A(+σ(j). Also we have pj ∈ S ∪ U and pσ(j) ∈ S ∩ U. As above. −j) then A(+σ(j). −σ(i)) contains both S. −σ(i)) both contain S and U . The case (C). The case (A). This means that there exist precisely 2(m + 1)(n − k − 1) distinct Cij = Cσ(j)σ(i) satisfying (B). If the complement subset Cij = Cσ(j)σ(i) contains S and U then one of the following possibilities is realized: (A) the subsets A(+i.

pi ∈ S implies that pσ(j) ∈ S. So. The set B \ U contains precisely 2(n − k − 1) points collinear with all points of U and B ∩ (S \ U ) contains not greater than n − k − 1 such points (if pl ∈ B \ U is collinear with all points of U then the same holds for pσ(l) ∈ B \ U . 2 if they are weakly adjacent.196 Grassmannians of Classical Buildings Then pσ(j) ∈ U . U are adjacent and c(S. In the case when m = k − 1 (S and U are adjacent or weakly adjacent). So. Similarly. U \ S contains not greater than n − k − 1 points of B collinear with all points of S. (n − k − 1)2 } distinct Cij = Cσ(j)σ(i) satisfying (C). pi ∈ S \ U and pσ(j) ∈ U \ S. Therefore. If S and U are weakly adjacent then there are no compliment subsets of such kind. we have B ∩ (S \ U ) = {pu } and B ∩ (U \ S) = {pv }. c(S. pi ⊥ U and pσ(j) ⊥ S. but S \ U does not contain at least one of these points). Similarly. we have pi ∈ U . we established that c(S. A direct veriﬁcation shows that k + 2k(n − k − 1) ≥ 2 m+1 + 2(m + 1)(n − k − 1) + l(m) 2 for m ≤ k − 2 and we get the equality only in the case when n = 4 and k = 1. . there exist at most l(m) := min{(k − m)2 . Since pσ(i) ∈ U and pj ∈ S. Since U does not belong to A(+i. If S and U are adjacent then v = σ(u) and Cuσ(v) = Cvσ(u) is the unique compliment subset satisfying (C). U ) ≤ moreover. −j) and pj ∈ U . U ) = if S. U ) = k + 2k(n − k − 1) 2 k + 2k(n − k − 1) + 1 2 m+1 + 2(m + 1)(n − k − 1) + l(m).

• it consists of k + 2 mutually adjacent elements of A. j = σ(j). −σ(i )) = ∅. −σ(i)) is disjoint from A(+l) then i = σ(l) or j = l. Then A(+i. .19. −j) ∪ A(+σ(j). σ(l)). Suppose that 1 ≤ k ≤ n − 2 and Π is a polar space of type Cn . . We take any l ∈ {1. For the intersection of 2n − k − 2 distinct maximal inexact subsets of ﬁrst type one of the following possibilities is realized: • the intersection is empty. in other words. .Polar and Half-Spin Grassmannians 197 Remark 4.21. A(+σ(j). Now let us ﬁx i. Therefore. . σ(i) and consider the associated complement subset Cij = Cσ(j)σ(i) . −j) ∩ A(+i . j = i or i = σ(i). If R is of second type then there are precisely 3 distinct complement subsets of A disjoint from R. There are only two complement subsets of ﬁrst type disjoint from it: A(+σ(i)) and A(+j). . Suppose that Ci j does not intersect Cij . . Proof. the adjacency and weak adjacency relations can be characterized in terms of maximal inexact subsets of ﬁrst type. Suppose that Π is of type Cn and 1 ≤ k ≤ n − 2. −j ) = ∅ =⇒ i = σ(i) or i = j or j = i. 2n} and consider the complement subset A(+l). −j ) = ∅. −σ(i )) = ∅. we establish that i = j. −σ(i)) ∩ A(+i . this is A(+σ(l)). If R is of ﬁrst type then there are precisely 2n − 1 distinct complement subsets of A disjoint from R. There is only one complement subset of ﬁrst type which does not intersect A(+l). −σ(i)) ∩ A(+σ(j ). this is Cσ(l)m = Cσ(m)l and we have 2n − 2 possibilities for m (since m = l. • it consists of 2 weakly adjacent elements of A. −j) ∩ A(+σ(j ). If a complement subset Cij = Cσ(j)σ(i) = A(+i. . j ∈ {1. Using the equalities A(+σ(j). there are precisely 2n − 2 + 1 distinct complement subsets of A disjoint from A(+l). Let R be a complement subset of A. In this case. A(+i. . . Lemma 4. 2n} such that j = i. Thus Cji = Cσ(i)σ(j) is the unique complement subset of second type disjoint from Cij = Cσ(j)σ(i) .

Thus R is a complement subset of A if and only if f (R) is a complement subset of f (A).21 to distinguish complement subsets of diﬀerent types.21. In the case when k = 1 and the polar spaces are of type C4 . we show that f transfers inexact subsets of A to inexact subsets of f (A).20 that S and U are adjacent if and only if f (S) and f (U ) are adjacent. Remark 4. k) = n2 (n. Suppose that 1 ≤ k ≤ n − 2. n1 (n. k).20 shows that S and U are weakly adjacent if and only if their images are weakly adjacent. Then an inexact subset of A is maximal if and only if the same holds for its image. by Lemma 4. Every apartments preserving mapping does not change the types of complement subsets if n1 (n. Remark 4. if 1 ≤ k ≤ n−2 and the polar spaces are of type Cn then. As in the proof of Proposition 3. Let S and U be distinct elements of Gk (Π) and A be an apartment of Gk (Π) containing them. 1 ≤ k ≤ n − 2.6. this statement follows immediately from Remark 4. Denote by n1 (n.19. If the polar spaces are of type D4 then there exist colineations of G1 (Π) to G1 (Π ) which do not preserve the weak adjacency relation (Remark 4. Let f : Gk (Π) → Gk (Π ) be an apartments preserving mapping and k ≥ 1. Since A and f (A) have the same number of inexact subsets (Π and Π are polar spaces of the same type). if k = 1 or our polar spaces are not of type D4 then f is weak adjacency preserving (two elements of Gk (Π) are weakly adjacent if and only if their images are weakly adjacent). each inexact subset of f (A) is the image of a certain inexact subset of A. recall that the mapping f is injective. .21. k) and n2 (n.32. respectively (it is clear that complement subsets of the same type have the same number of elements). It follows from Lemma 4. k) = n2 (n. the mapping f preserves the types of all maximal inexact and complement subsets. In the case when k ≤ n − 2. Proof. all complement subsets are of second type. k) for some pairs n. If n = 4 or k = 1 then Lemma 4. However.32 does not hold in this case. k and we need Lemma 4. k) the numbers of elements in complement subsets of ﬁrst and second type. If our polar spaces are of type Cn then Gk (Π). In the case when k = n − 1 or our polar spaces are of type Dn . contains complement subsets of both types.13) and the second part of Proposition 4. First of all. Then f is adjacency preserving (two elements of Gk (Π) are adjacent if and only if their images are adjacent).20.198 Grassmannians of Classical Buildings Proposition 4. Then f (S) and f (U ) belong to the apartment f (A).

+j) for certain i. we have pi ∈ S. . .9. δ ∈ {+. where {i1 .2. pσ(j) }) ∪ {pi . . and complement subsets. . −ju ) and introduce the concepts of exact. Let B = {p1 . we construct a frame (B \ {pi . +j) is an inexact subset. Proposition 4.17. . . +j)) (4. A(+i1 . σ(in )} = ∅.9. . Recall that it consists of all maximal singular subspaces pi1 .34.4 Inexact subsets of half-spin Grassmannians Now we suppose that Π is a polar space of type Dn and n ≥ 4. . .Polar and Half-Spin Grassmannians 199 4. . If j = i. inexact. . . For every subset R ⊂ A we denote by Si (R) the intersection of all elements of A(+i) ∩ R.33. pσ(j) } which deﬁnes a new apartment of Gδ (Π) containing (4. As in the proof of Lemma 4. in } ∩ {σ(i1 ). if this subset is empty then we write Si (R) = ∅. A(−i). .10) containing pσ(j) and pσ(i) ∈ Sσ(j) (A(−i) ∪ A(+i. . . . by (4. . pσ(i) belongs to every element of (4. Then S does not contain pj and. Therefore. It is trivial that pj ∈ Si (A(−i) ∪ A(+i.10).11). The latter guarantees that pσ(i) ∈ S. −}.10) containing pσ(j) . Every element of A contains precisely one of the points pi or pσ(i) for every i. . . +is . .10) Suppose that S is an element of (4. If R is a maximal inexact subsets of A then R = A(−i) ∪ A(+i. pin ∈ Gδ (Π). σ(i) then A(−i) ∪ A(+i. Proposition 4. σ(i). As in Subsection 4. j such that j = i. . p2n } be a frame of Π and A be the associated apartment of Gδ (Π). −j1 . Proof.11) (4. +j)). . . . we deﬁne A(+i). .

35. On the other hand. m = l. Suppose that n = 4. Since R is inexact.32. by Proposition 4. Lemma 4. Proof. Subspaces S. See the proof of Lemma 4.31. −j) = A(+i.200 Grassmannians of Classical Buildings Proof. the inverse inclusion holds. δ. U ∈ A are adjacent if and only if there are precisely n−2 2 distinct complement subsets of A containing S and U .34. By Proposition 4. every compliment subset of A coincides with certain A(+l. An easy veriﬁcation shows that R ⊂ A(−i) ∪ A(+i. σ(l). If Si (R) is empty then R is contained in the non-maximal inexact subset A(−i) which contradicts the assumption that the inexact subset R is maximal. . The complement subset corresponding to A(−i) ∪ A(+i.22. we have Si (R) = pi for a certain number i. γ ∈ {+. As in the proof of Proposition 4. +j). Suppose that Π is a polar space of type Dn . Thus there exists pj ∈ Si (R) such that j = i. Proof.22 gives the claim. n ≥ 4. Proposition 4. +j) is A(+i. −}. and f : Gδ (Π) → Gγ (Π ). Let A be an apartment of Gδ (Π). Gδ (Π) is a polar space of type D4 and A is a frame of this space.20 in the case when k = n − 1. in a polar space of type Dn every complement subset of a frame is a pair of collinear points. we establish that R is a complement subset of A if and only if f (R) is a complement subset of the apartment f (A). +σ(j)). Then f is adjacency preserving (two elements of Gδ (Π) are adjacent if and only if their images are adjacent). Then every complement subset of A consists of two adjacent elements.22. is an apartments preserving mapping. Remark 4. Then Lemma 4. +m). Since the latter subset is inexact and the inexact subset R is maximal.

f (pu ).1. . . For polar Grassmannians of index > 0 and half-spin Grassmannians this follows immediately from Propositions 4. .13) implies that u = σ(i) and v = i. For each i ∈ {1. . The implication (2) =⇒ (1) easy follows from the frame deﬁnition.35. it is suﬃcient to show that two points in a frame B = {p1 . This mapping is injective. For a mapping f : P → P the following conditions are equivalent: (1) f is apartments (frames) preserving. f (pv )} (otherwise. p2n } of Π are collinear if and only if their images are collinear.18). f (pu ).32. . Since f (B) is a frame. f (pσ(i) )} = {f (pu ).6. f (pv )}. . one of the frames (4. Thus there is no frame of Π containing the subset {f (pi ). Therefore. v ∈ {1. For Grassmannians of index 0 we have the following.13) f (pi ) ⊥ f (pu ) and f (pσ(i) ) ⊥ f (pv ). In the case when Π and Π are of type Dn . we establish that a subset X ⊂ B is complement if and only if f (X) is a complement subset of the frame f (B). Let f : P → P be a mapping which sends frames of Π to frames of Π . we need to establish that every apartments preserving bijection is an isomorphism between the associated Grassmann graphs.5 Proof of Theorem 4. . It is clear that u = v and f (pi ) is non-collinear with both f (pu ). f (pv ). f (pi ) ⊥ f (pσ(i) ) which guarantees that f (pi ) is collinear with f (pj ) if j = σ(i). (1) =⇒ (2). . Suppose that our polar spaces are of type Cn .9. Then (f (B) \ {f (pi )}) ∪ {f (pi )} and (f (B) \ {f (pσ(i) )}) ∪ {f (pi )} (4. 2n} such that (4. . . (2) f is a collinearity preserving injection (two points are collinear if and only if their images are collinear). Since any two points are contained in a certain frame. Then (4. 2n} there exists a point pi such that (B \ {pi }) ∪ {pi } and (B \ {pσ(i) }) ∪ {pi } are frames of Π (Lemma 4.12) contains f (pi ). Proposition 4. . Proof. f (pv )). We have {f (pi ). As in the proof of Propositions 4. .32 and 4.36.Polar and Half-Spin Grassmannians 201 4.16 By Subsection 4. there are unique u. .12) are frames of Π . each complement subset of B is a pair of collinear points and we get the claim.

. . pn+3 . By Proposition 4. . . p2 . . pn+3 . . p2n }⊥ . . for every k ∈ {1. Similarly. γ ∈ {+. pn+3 . Every apartments preserving mapping f : P → P is an embedding of Π in Π . . p2n } be a frame of Π containing them. Then all pi := f (pi ) form a frame of Π and ∀ i ∈ {1. . . . Let f : P → P be an apartments preserving mapping.3) and we have p1 p2 = {p1 . . δ. . . −}. p2 ∈ P be distinct collinear points and {p1 . we obtain that p1 p2 = {p1 . . p3 . This means that f transfers lines to subsets of lines and distinct lines go to subsets of distinct lines. . p2 . .37. Similarly. . . pn . . . pn . . n − 1} we have the mapping (f )k : Gk (Π) → Gk (Π ) S → f (S) . . . . . p2n }⊥ .37. . n}.36. . . p2 . The line p1 p2 is a maximal singular subspace of the generalized quadrangle {p3 . . . p2n }⊥ (Lemma 4. This implies that f (S) consists of mutually collinear points and dim f (S) = dim S. p3 . f is a collinearity preserving injection. it is an embedding of Π in Π . So. Thus a point p ∈ P is on the line p1 p2 if and only if f (p) is on the line p1 p2 = f (p1 )f (p2 ). if our polar spaces are of type Dn . n}. . . . . . We suppose that pi ⊥ pi+n pi ⊥ pi+n ∀ i ∈ {1. pn .6 Embeddings Proposition 4. By Proposition 4.9. Let p1 . . . An easy veriﬁcation shows that this is an apartments preserving embedding of Gk (Π) in Gk (Π ). . we construct the apartments preserving embedding (f )δ of Gδ (Π) in Gγ (Π ). . Proof.202 Grassmannians of Classical Buildings 4. For every singular subspace S of Π there is a frame such that S is spanned by a subset of this frame. .

If k = 1 or the polar spaces are not of type D4 then every apartments preserving mapping of Gk (Π) to Gk (Π ) is the embedding of Gk (Π) in Gk (Π ) induced by an apartments preserving embedding of Π in Π . this embedding is apartments preserving. the associated half-spin Grassmann spaces are polar spaces of type D4 . δ. Then f induces an embedding of G1 (Π) in G1 (Π ). In this case. Similarly. is the embedding of Gδ (Π) in Gγ (Π ) induced by an apartments preserving embedding of Π in Π . −}. Suppose that the polar spaces are of type D4 . is the embedding of Gδ (Π) in Gγ (Π ) induced by an apartments preserving embedding of Π in Π or in G−γ (Π ). −}. γ ∈ {+. δ ∈ {+. Then every apartments preserving mapping of G1 (Π) to G1 (Π ) is the embedding of G1 (Π) in G1 (Π ) induced by an apartments preserving embedding of Π in Π or in one of the half-spin Grassmann spaces of Π . Also f induces an apartments preserving embedding of one of the half-spin Grassmann spaces of Π in G−δ (Π ). it is adjacency preserving (two elements of Gk (Π) are adjacent if and only if their images are adjacent). 1.23. .17 (M.18 Let 1 ≤ k ≤ n − 1 and f : Gk (Π) → Gk (Π ) be an apartments preserving mapping. Let f be an apartments preserving embedding of Π in Gδ (Π ). δ. This mapping is injective (otherwise there exists a line of Gn−1 (Π ) containing the images of two distinct lines of Gn−1 (Π). 4. Remark 4.18 (M.17 and 4. every apartments preserving mapping of Gδ (Π) to Gγ (Π ). Let 1 ≤ k ≤ n − 1. Theorem 4. Now suppose that our polar spaces are of type D4 . If our polar spaces are of type Dn and n ≥ 5 then every apartments preserving mapping of Gδ (Π) to Gγ (Π ). Pankov). In [Pankov 2 (2007)] a such kind result was proved for symplectic Grassmannians. If k = n − 1 then f maps lines to subsets of lines and induces a mapping fn−2 : Gn−2 (Π) → Gn−2 (Π ). the mapping f preserves the weak adjacency relation (two elements of Gk (Π) are weakly adjacent if and only if their images are weakly adjacent) if k = 1 or the polar spaces are not of type D4 . By Proposition 4. Pankov).4. this implies the existence of non-adjacent elements of Gn−1 (Π) whose images are adjacent).Polar and Half-Spin Grassmannians 203 Theorem 4. in [Pankov 3 (2007)] it was announced for all Grassmannians associated with polar spaces of type Cn . γ ∈ {+. by Corollary 4.32.9. −}. In the case when k ≤ n − 2.7 Proof of Theorems 4.

3) if and only if their intersection is a line. we show that fk−1 maps the apartment of Gk−1 (Π) associated with B to the apartment of Gk−1 (Π ) deﬁned by B . As in the proof of Theorem 3. fi ([S i ) ⊂ [fi−1 (S) for all S ∈ Gi−1 (Π) and fi−1 ( U ]i−1 ) ⊂ fi (U )]i−1 for all U ∈ Gi (Π) if i ≥ 1. f1 is induced by f . 0. two stars of G1 (Π) are adjacent (Subsection 4.6.6. Denote by B the frame of Π corresponding to the apartment f (A). . Now suppose that our polar spaces are of type D4 . . It is injective (otherwise there exist weakly adjacent elements of Gk (Π ) whose pre-images are not weakly adjacent). such that fk = f . . 2.3).3. An easy veriﬁcation shows that each fi . Consider the case k = 2. f induces a mapping fk−1 : Gk−1 (Π) → Gk−1 (Π ).4. this implies that f1 maps tops to subsets of tops. . it is adjacency preserving and maximal singular subspaces of G1 (Π) go to subsets of maximal singular subspaces of G1 (Π ). Since f1 is apartments preserving. In the present case. Big stars go to k k subsets of big stars (since any two elements of a top are adjacent and a big star contains non-adjacent elements). we construct a sequence of apartments preserving mappings fi : Gi (Π) → Gi (Π ). As in Subsection 4. In the case when n = 4 or the polar spaces are of type C4 . Therefore. On the other hand. Therefore. i ≥ 1. Let B be a frame of Π and A be the associated apartment of Gk (Π).10 (Subsection 3. . Then stars go to subsets of stars (for every star there is a top intersecting this star in a line and the intersection of two distinct tops contains at most one element). It must be pointed out that the mapping fk−1 is not deﬁned only in the case when k = 1 and our polar spaces are of type D4 . fk−1 is apartments preserving. The image of every big star is contained in precisely one big star (the intersection of two distinct big stars is empty or consists of one element).204 Grassmannians of Classical Buildings If 1 ≤ k ≤ n − 2 and f preserves the weak adjacency relation then it transfers maximal cliques of Γw (Π) to cliques of Γw (Π ). is induced by f0 . we establish that i i = k.

Then f is an adjacency preserving injection of G1 (Π) to G1 (Π ) satisfying the condition (A) from Subsection 4. It is clear that this embedding also induces f . It is not diﬃcult to prove that f0 is an apartments preserving embedding of Π in Π which induces f1 and f . −}. a single point. Then it transfers maximal singular subspaces of Gδ (Π) (stars and special subspaces) to subsets of maximal singular subspaces of Gγ (Π ). a line. γ ∈ {+. As above.6.5. In the case when the image of a certain star is contained in a top. Let X1 and X2 be planes of Gδ (Π) such that g(X1 ) and g(X2 ) are contained in a plane X of Gγ (Π ). It transfers maximal singular subspaces of G1 (Π) to subsets of maximal singular subspaces of G1 (Π ) and distinct maximal singular subspaces go to subsets of distinct maximal singular subspaces. This means that g transfers every plane to a subset spanning a plane. If k = 3 then f2 is induced by an apartments preserving embedding of Π to Π (it was established above). δ. Show that the images of distinct planes are contained in distinct planes.Polar and Half-Spin Grassmannians 205 f1 transfers big stars to subsets of big stars. Let k = 1. By Proposition 4. and a plane.5). distinct maximal singular subspaces go to subsets of distinct maximal singular subspaces. There are the following four possibilities for the intersection of two distinct maximal singular subspaces: the empty set. 3. by the arguments given above. This implies the existence of an apartment . Suppose that our polar spaces are of type Dn and g is an apartments preserving mapping of Gδ (Π) to Gγ (Π ). we have the mapping h : P → G−γ (Π ) (Subsection 4. Then every element of X1 is adjacent with every element of X2 and there exists a maximal singular subspace of Gδ (Π) containing both these planes. the mapping g is adjacency preserving (two elements of Gδ (Π) are adjacent if and only if their images are adjacent).35. In the other cases. If stars go to subsets of stars then. f is induced by an apartments preserving embedding of Π in Π . it sends every line L ∈ L to a subset of the line of G−γ (Π ) corresponding to f (L) (Remark 4. The image of every big star is contained in precisely one big star and f1 induces a mapping f0 : P → P . The image of every maximal singular subspace of G1 (Π) is contained in precisely one maximal singular subspace of G1 (Π ).12). For every plane there is an apartment intersecting this plane in a triangle. The reader can show that h is an apartments preserving embedding of Π in G−γ (Π ) which induces f .6. the intersections with apartments contain at most two elements. This guarantees that g maps lines to subsets of lines.

g is an embedding of Gδ (Π) in Gγ (Π ). we establish that distinct lines go to subsets of distinct lines. This mapping is apartments preserving (standard veriﬁcation). This embedding induces g. This is possible only in the case when X1 = X2 . but the apartment g(A) intersects X in a subset containing at most three points. Remark 4.37. the half-spin Grassmann spaces Gδ (Π) and Gγ (Π ) are polar spaces of type D4 .24.4 and Proposition 4. In the case when n = 4. Therefore. By Corollary 4. g induces an injection of Gn−3 (Π) to Gn−3 (Π ). This implies that it transfers lines to subsets of lines and distinct lines go to subsets of distinct lines. .206 Grassmannians of Classical Buildings A ⊂ Gδ (Π) which intersects each Xi in a triangle. Similarly. hence it is induced by an apartments preserving embedding of Π in Π or in one of the half-spin Grassmann spaces of Π (the second possibility can be realized only for n = 4).

Spanning point-line geometries in buildings of spherical type. Buekenhout. 26. Coxeter Matroids. A. Chow. Projective and polar spaces. and Cameron. Geom. Ceccherini. Gelfand. Comb. (2003). J. Ann. F. Blok. (1949).Bibliography Abramenko. Geometric Algebra (Interscience Publishers. 4. K. (1967). (1957). (1991). Twin buildings and applications to S-arithmetic groups. J. New York). Dedicata 12. A. Projective and aﬃne geometry over division rings. pp. N. P. N. and Brouwer. On the foundations of polar geometry. (2008). P. V. Handbook of Incidence Geometry: Buildings and Foundations (Elsevier. Blunck. and Shult. Dedicata 33. e e Chapters IV–VI. M. R. Buekenhout. pp. P. pp. (1982). 1. 27–62. Groupes et alg`bres de Lie. (2000). Abramenko. 125–137. Geom. a ´e Bourbaki. 62. Rend. (1974). (1990). E. Collineazioni e semicollineazioni tra spazi aﬃni o proiettivi. 26–35. 75–85. (1995). H. Discrete Math. (2005). A. E. and Van Maldeghem. On opposition in spherical buildings and twin buildings. (1968). Progress in Mathematics 216 (Birkh¨user. in F.4442. Actualit´s Scientiﬁques et Industrielles 1337 (Hermann. London). Preprint arXiv:0805. e Appl. F. and Havlicek. Intersections of apartments. Linear Algebra and Projective Geometry (Academic Press. I. to appear in J. Lecture Notes in Mathematics 1641 (Springer-Verlag. Baer. 21–26. 46–56. 1. Combin. On the foundations of polar geometry II. 1. Borovik. Artin. S. pp. Cameron. pp. P. e Paris). El´ments de math´matique. pp. Buildings (Springer-Verlag. Dual polar spaces. 309–348. Abramenko.. Boston). H. (1989). Theory Ser. New York). Cameron. Mat. (1952). On bijections that preserve complementarity of subspaces. Geom. 1-2. pp. (1996). P. L. V. P. (1998). E. and White. 155–170. 301. Berlin). Dedicata 3. F. of 207 . Buekenhout. On the geometry of algebraic homogeneous spaces. Buekenhout (ed. QMW Maths Notes 13 (Queen Mary and Westﬁeld College. H. W. Ann. Brown. pp. New York-London). P. Amsterdam). and Van Maldeghem. Geom. 2. A. R.). 2.

Math.). E. BerlinNew York). pp. Havlicek. B. B. Abh.-l. Algebra Appl. Cooperstein. An elementary proof of the Fundamental Theorem of Projective Geometry. 2451–2455. A. Adjacency preserving mappings of invariant subspaces of a null system. (1997). W. Adv. (2003). 15–36. N. Cooperstein. 2. Dieudonn´. Cooperstein. 5. pp. Mitt. Havlicek. 319–324. Handbook of Incidence Geometry: Buildings and Foundations (Elsevier. Discrete Math. 65–77. 1-2. and Longstaﬀ. Frames and bases of Lie incidence geometries. pp. M. (1994). Giessen 215. Amsterdam). Cohen. 28. Huang. A. Napoli. Soc. 36. 117–120. 3d edn. (1995). 12 (Dept. (1994). London). European. Cooperstein.-l. pp. Hamburg 14. 50. Sem. and Pankov. Point-line spaces related to buildings. pp. P. Transformations on the product of Grassmann spaces. 237–262. A. pp. Fillmore. 128. 820–829. (1951).. (2000). Buildings and Classical groups (Chapman & Hall. N. E. A. 429. Math. (2007). W. and Fr¨licher. Dieudonn´. On isomorphisms of lattices of closed subspaces. pp. Adjacency preserving transformations of Grassmann spaces. (1998). B. P.. 60. A. and Shult. W. J. Univ. (2005). 530–533. Lin. (Basel) 64. Classical subspaces of symplectic Grassmannians. 32–67. W. Basis preserving maps of linear spaces. 3. 6. Dieudonn´. Math.E. 5. Math. J. and Havlicek. H. (1951). 8. (1999). A. Diameter preserving surjections in the geometry of matrices. and Shult. J. Simon Stevin 12. pp. (2008). Huang. Huang. Mat. pp. Soc. N. pp. Chow’s theorem for linear spaces. (1995). W. Mitt. 295–304. pp. 145–151. B. Belg. Faure. Havlicek. Hamburg 68. 5. Math. M. On the automorphisms of the classical groups. pp. Topics in diagram geometry. pp.-l. pp. H. Symplectic subspaces of symplectic Grassmannians. A generalization of Brauner’s theorem on linear mappings. Math. Geom. H. and Kreuzer. 17–46. (1984). Canad. J. e Math. (2005). A. Morphisms of projective geometries and o semilinear maps. (2005). Garrett. Arch. Algebraic homogeneous spaces over ﬁeld of characteristic e two. Kasikova. Soc. 2.-l. Huang. Cooperstein. Demonstratio Math. Dedicata 53. (2001). 719–725. 675–688. Characterization of the transformation group of the space . W. Ges. E. C. Witt-type Theorems for Grassmannians and Lie Incidence Geometries. Proc. Combin. Seconda Univ. Amer. Quad. Havlicek. (1997). 5. Math. B. 3. N. Mem. 1. pp. 1. J. N. in F. Caserta). H. (1995). (2002). Buekenhout (ed. 1. Ergebnisse der e e e Mathematik und ihrer Grenzgebiete 5. Faure. J. (1971).208 Grassmannians of Classical Buildings Math. Soc. Math. Geom. C. On Isomorphisms of Grassmann Spaces. H. 376–386. 27–41. pp. 29–57. Generation of embeddable incidence geometries: a survey. pp. pp. Proc.-l. 1442–1454. E. (Springer-Verlag. Bull. pp. Sem. 208/209. 1. La G´om´trie des Groupes Classiques. Dedicata 90. Geom. Amer. 647–737. Geom. Amer. 38. Math. Huang.

Math. Dedicata 61. New York-Heidelberg). J. A characterization of geometrical mappings of grassmann spaces. (2006). M. (1975). Geom. 1-2. Sem. 172–187. (2001). M. pp. pp. 43. 727–733. 3-4. Surjections on Grassmannians preserving pairs of elements with bounded distance. Pankov. Kwiatkowski. Transformations of Grassmannians preserving the class of base sets. 279–283. Transformations of grassmannians and automorphisms of classical groups. S¨rensen. 1-2. (1994). Aequationes Math. 26–32. Linear algebraic groups. (2005).. Pankov.. 3. Oxford University Press. pp. (1996). Geom. Demonstratio Math. K. 1. pp. Combin. 8. Mackey. 45. 75. (2007). pp. pp. M. Amer. Beitr¨ge Algebra Geom. 79. Pankov. J. Isomorphisms of normed linear spaces. M. a ˙ Pankov. 625–637. D. Algebraic Combin. Geometry of polar grassmann z spaces. W. Geom. (2010). 45. Bull. Ann. 3-4. 254–268. European. Pankov. pp. Math. pp. J. M. pp. Pankov. Karzel. and Pankov. 26. (1973). Graduate Texts in Mathematics 21 (Springer-Verlag. H. Geometrical mappings of Grassmannians of linear spaces satisfying the exchange axiom. of Math. K.. Pankov. J. (2005). M. M. Opposite relation on dual polar spaces and half-spin Grassmann spaces. pp. pp. pp. (2004). M. Univ. Humphreys.Bibliography 209 of a null system. A. 39. ˙ Pankov. (1942). On the geometry of linear involutions. Mappings of the sets of invariant subspaces of null systems. 1-4. 389–399. Einf¨hrung in die Geometrie. 56. 319–327. G. 75. Linear Algebra Appl. On the deﬁnition of isomorphisms of linear spaces. Diagram geometries. Pra˙ mowski. 455–467. Pankov. (2002). Base subsets of symplectic Grassmannians. pp. 1703–1707. 40. M. Geom. 243–250. Pankov. M. 143–159. (2007). J. 28. Chow’s theorem and projective polarities. 1-2. M. 3. 3. (2007). Pra˙ mowski. 72. pp. Dedicata 107. 52. 7. 3. Theory Ser. H. S. New York). M. pp. Results Math. and Windelberg. Aequationes Math. Adv. pp. M. (2004). K. pp. 17–24. On isomorphisms of Grassmann spaces. 54. Geom. pp. J. 5. Base subsets of polar Grassmannians. 2. 8. 1394–1406. G. G¨ttingen). Transformations preserving z adjacency and base subsets of spine spaces. A. Oxford Science Publications (Clarendon Press. 3. 301–308. Pankov. M. (2009). Lim. J. 21–50. 132–150. Pasini. and Zynel. On a construction of aﬃne Grassmannians and spine z spaces. pp. (2004). 226–232. E. Pra˙ mowski. (1998). K. M. 2. 432. 169–176. 2. Geom. A 114. o u Uni-Taschenb¨cher 184 (Vandenhoeck & Ruprecht. 244–260. Kakutani. pp. and Zynel. Ring and lattice characterizations of complex Hilbert space. . Results Math. (2004). and Mackey. (2006). Pankov. 72. Combin. M. Base subsets of Grassmannians: inﬁnite-dimensional case.W (1946). Kreuzer. u o Kreuzer. Abh. M. Hamb. A. Results Math. Soc.

pp. 167–170. Abh. (2002). . ˙ Zynel. E. Segre subproduct. Subspaces and embeddings of spaces of pensils. Univ. 117-142. Amsterdam). J. (2009). Math. A 28. On projective and aﬃne hyperplanes. Veldkamp. C. University in Bialystok. ˙ Pra˙ mowski. Rickart. Nederl. Automorphisms of spine spaces. 1-2. Tits. M. (1989). Handbook of Incidence Geometry: Buildings and Foundations (Elsevier. Perspectives in Mathematics 7 (Academic Press. Singapore). Lineare Abbildungen in reellen Vektorr¨umen. M. Taylor. Hamb. D.210 Grassmannians of Classical Buildings ˙ Pra˙ mowski. Lectures on buildings. its geometry. u Scharlau. pp. Akad. pp. Boston). Nederl. Geometry of Matrices (World Scientiﬁc. Z. Buekenhout (ed. Zick. Berlin). Ser. M. McGraw-Hill Series in Higher Mathematics (McGraw-Hill. Rudin.-X. Ser. Combin. Wet. K. Theory Ser. (1992). pp. Ronan. pp. and Zynel. and Zynel. Proc. Proc. Buildings. (2000). Lecture Notes in Mathematics 386 (Springer-Verlag. M. Function Analysis. (1973). E. 92. J. D. K. z Math. pp. pp. A 63. F. Sigma Series in Pure Mathematics 9 (Heldermann Verlag. Wan. J. Amer. Akad. Sem. automorz phisms and examples. Isomorphic groups of linear transformations. (1996). (1974). J. (1995). Geometry of the classical groups. 3. 451-464. New York-D¨sseldorf-Johannesburg). Teirlinck. (1983). (1950). A 62. Polar geometry I – V. Buildings of spherical type and ﬁnite BN-pairs. R. W. 477–646. Geom. 72. Semestera ber 30. pp. 512–551.). W. in F. L. 72. (1959/1960). 290–306. 59–77. Wet. Preprint. Math. (1980). 2. 207–212. Berlin-New York).

83. 72 of a partial linear space. 48 on Grassmannians. 25 Coxeter system. 199 form alternating. 199 complex chamber. 46 on the set of involutions. 150 spherical. 187. 46 Coxeter. 16 apartment in a building. 12 distance. 182 base of of base in polar. 60 . 49 diagram. 52 labeled. 146 211 a partial linear space. 56 associated with a Tits system. 106 weak. 59 of the orthogonal group. 140 on the set of chambers. 90. 18 a vector space. 161 annihilator. 65. 24 exact subset. 61 of the symplectic group. 38 Hermitian. 103 big star. 97. 39 quadratic.Index adjacency relation on a Grassmannian. 114. 150 simplicial. 70. 18 division ring. 58 of the linear group. 8 embedding Grassmann. 101 in the set of involutions. 18 collineaton. 43 contragradient. 133. 118 in a spine space. 18 of a vector space. 182 in a parabolic subspace. 56 in a Grassmannian. 162 building abstract. 46 oriﬂamme. 45. 178. 97 in a Grassmannian of an inﬁnite-dimensional vector space. 12 subset a Grassmannian of an exchange space. 84. 17 complement subset. 180. 17 semilinear. 194. 62 collinearity graph. 89. 49 dimension of a partial linear space.

152 transvection. 97. 96 polar. 18 projective(over a division ring). 62 panel. 20 dual vector. 80. 161 Grassmannian half-spin. 101 vector. 140 linear. 71 half-spin Grassmann. 151. 156 subspace of a partial linear space. 21 hyperplane. 14 exchange. 38 sesquilinear. 71. 179 weak Grassmann. 133 homothetic transformation. 114.212 Grassmannians of Classical Buildings reﬂexive. 48. 152. 32 partial linear. 182 hypercube. 18 special. 14 of an exchange space. 20 opposite relation on Grassmannians. 143 . 97. 154 Witt index. 174 on the set of chambers. 104 triangle. 11 semilinear. 155 totally isotropic. 78 Grassmann. 95 gamma. 139 of a chamber complex. 40 projective plane. 38 symmetric. 102 lattice. 17 space dual projective. 28 mapping apartments preserving. 150 half-cube. 77. 48. 18 independent subset of a partial linear space. 130 generalized quadrangle. 71. 10 Veldkamp. 64. 74. 17 normed. 158 singular. 73. 39 top. 66. 125 polar Grassmann. 67 generalized projective. 18 graph Grassmann. 33 linear. 11 inexact subset. 104. 179 Johnson. 45 semicollineaton. 18 of a vector space. 19 spine. 153. 38 frame. 46 polarity. 47 of a vector space. 187. 199 involution. 97. 38 skew-symmetric. 65 conjugate-linear. 149 star. 18 rank of a simplicial complex. 18 of a vector space. 134 projective. 83 polar Johnson. 73. 17 polar. 128 geodesic. 11 parabolic. 18. 89.