Trends in Logic 50

Leo Esakia

Heyting
Algebras
Duality Theory
Edited by Guram Bezhanishvili
Wesley H. Holliday
Translated by Anton Evseev
Trends in Logic

Volume 50
 TRENDS IN LOGIC
Studia Logica Library

VOLUME 50

Editor-in-Chief
Heinrich Wansing, Department of Philosophy, Ruhr University Bochum,
Bochum, Germany

Editorial Board
Arnon Avron, Department of Computer Science, University of Tel Aviv, Tel Aviv, Israel
Katalin Bimbó, Department of Philosophy, University of Alberta, Edmonton, AB, Canada
Giovanna Corsi, Department of Philosophy, University of Bologna, Bologna, Italy
Janusz Czelakowski, Institute of Mathematics and Informatics, University of Opole,
Opole, Poland
Roberto Giuntini, Department of Philosophy, University of Cagliari, Cagliari, Italy
Rajeev Goré, Australian National University, Canberra, ACT, Australia
Andreas Herzig, IRIT, University of Toulouse, Toulouse, France
Wesley Holliday, UC Berkeley, Lafayette, CA, USA
Andrzej Indrzejczak, Department of Logic, University of Lodz, Lódz, Poland
Daniele Mundici, Mathematics and Computer Science, University of Florence, Firenze, Italy
Sergei Odintsov, Sobolev Institute of Mathematics, Novosibirsk, Russia
Ewa Orlowska, Institute of Telecommunications, Warsaw, Poland
Peter Schroeder-Heister, Wilhelm-Schickard-Institut, Universität Tübingen, Tübingen,
Baden-Württemberg, Germany
Yde Venema, ILLC, Universiteit van Amsterdam,
Amsterdam, Noord-Holland, The Netherlands
Andreas Weiermann, Vakgroep Zuivere Wiskunde en Computeralgebra, University of Ghent,
Ghent, Belgium
Frank Wolter, Department of Computing, University of Liverpool, Liverpool, UK
Ming Xu, Department of Philosophy, Wuhan University, Wuhan, China
Jacek Malinowski, Institute of Philosophy and Sociology, Polish Academy of Sciences,
Warszawa, Poland

Assistant Editor
Daniel Skurt, Ruhr-University Bochum, Bochum, Germany

Founding Editor
Ryszard Wojcicki, Institute of Philosophy and Sociology, Polish Academy of Sciences,
Warsaw, Poland

The book series Trends in Logic covers essentially the same areas as the journal Studia Logica, that is,
contemporary formal logic and its applications and relations to other disciplines. The series aims at publishing
monographs and thematically coherent volumes dealing with important developments in logic and presenting
significant contributions to logical research.
Volumes of Trends in Logic may range from highly focused studies to presentations that make a subject
accessible to a broader scientific community or offer new perspectives for research. The series is open to
contributions devoted to topics ranging from algebraic logic, model theory, proof theory, philosophical logic,
non-classical logic, and logic in computer science to mathematical linguistics and formal epistemology. This
thematic spectrum is also reflected in the editorial board of Trends in Logic. Volumes may be devoted to specific
logical systems, particular methods and techniques, fundamental concepts, challenging open problems, different
approaches to logical consequence, combinations of logics, classes of algebras or other structures, or
interconnections between various logic-related domains. Authors interested in proposing a completed book or a
manuscript in progress or in conception can contact either [email protected] or one of the Editors of the
Series.

More information about this series at http://www.springer.com/series/6645

Leo Esakia (Deceased)
Author

Guram Bezhanishvili Wesley H. Holliday

•

Editors

Heyting Algebras
Duality Theory

123
Author Editors
Leo Esakia (Deceased) Guram Bezhanishvili
Tbilisi, Georgia Department of Mathematical Sciences
New Mexico State University
Las Cruces, NM, USA

Wesley H. Holliday
Department of Philosophy and Group in
Logic and the Methodology of Science
University of California
Berkeley, CA, USA

Translated by
Anton Evseev (Deceased)
Birmingham, UK

ISSN 1572-6126 ISSN 2212-7313 (electronic)

Trends in Logic
ISBN 978-3-030-12095-5 (hardcover) ISBN 978-3-030-12096-2 (eBook)
ISBN 978-3-030-12098-6 (softcover)
https://doi.org/10.1007/978-3-030-12096-2

Library of Congress Control Number: 2018968096

© Springer Nature Switzerland AG 2019

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Leo Esakia (1934–2010)
Foreword

This translation of Leo Esakia’s book on Heyting algebras has been in the making
for a long time. The book was originally published in 1985 by the Georgian
publishing house Metsniereba (Science). It was the first volume of Esakia’s planned
two volume monograph on Heyting algebras. The book turned out to be popular
among Soviet logicians, and Esakia had begun work on the second volume, an
outline of which is presented in the Appendix of the first volume. But after the
break of the Soviet Union, the publishing house ceased to exist, and the project died
with it.
Logicians and mathematicians in the West were aware of the existence of the
book, and there were many requests to translate the book into English. But as far as
we know, there was no formal contract to translate the book by any Western
publishing house.
Hilary Priestley was among the Western mathematicians interested in the book,
having received a copy from Anna Romanowska. In 2003, Hilary enlisted a Russian
student, Anton Evseev, to translate the book into English. At the time, Anton was
an undergraduate studying Mathematics at the University of Oxford. His
hand-written translation was not widely circulated, but Hilary mentioned the
translation to Mai Gehrke, who in turn mentioned it to Nick Bezhanishvili and
David Gabelaia. After Esakia’s death in 2010, several tributes were planned: a
special issue of Studia Logica (Vol. 100, No. 1–2) dedicated to him appeared in
2012, and a special volume of Outstanding Contributions to Logic in his honor
appeared in 2014 (Leo Esakia on Duality in Modal and Intuitionistic Logics,
Springer). In addition, it was decided that the English translation of Esakia’s book
be edited for publication. Hilary scanned and emailed Anton’s hand-written
translation to Nick and David. With the help of Mamuka Jibladze, Nick and David
used some funds from Esakia’s last grant to hire staff from the A. Razmadze
Mathematical Institute to type up the translation. The first round of editing of the
translation occurred in the summer of 2012 by Guram Bezhanishvili and the second
round in the summer of 2017 by Guram and Wesley Holliday.

vii
viii Foreword

At last the final product is before your eyes. The initial work by Anton Evseev
made a valuable contribution toward bringing Leo Esakia’s classic monograph to a
wider audience, and we were very grateful when Anton willingly agreed that the
translation could form the basis of a version edited for publication. It is with great
sadness that we report Anton’s untimely death in February 2017 at the age of 33.
The mathematical community has been robbed of an exceptionally talented col-
league, and we regret that Anton himself will not see this English version of
Heyting Algebras in print.
Although many of the main results of Esakia’s book have already made their
way into the mathematical literature, there is no better way to see them developed
than through Esakia’s concise and lucid presentation. We hope the publication of
this translation will make Esakia’s intellectual legacy accessible to a wider
audience.

Guram Bezhanishvili
Nick Bezhanishvili
David Gabelaia
Mai Gehrke
Wesley H. Holliday
Mamuka Jibladze
Hilary Priestley
Contents

1 Preliminary Notions and Necessary Facts . . . . . . . . . . . . . . . . . . . . . 1

1.1 Universal Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Ordered Sets and Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Heyting Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2 Heyting Algebras and Closure Algebras . . . . . . . . . . . . . . . . . . . . . . 15
2.1 Heyting Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Closure Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Modal Systems and Superintuitionistic Logics . . . . . . . . . . . . . . . 21
2.4 Filters and Congruences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Skeletal Closure Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3 Duality Theory: Hybrids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.1 The Hybrid of Topology (Stone) and Order (Kripke) . . . . . . . . . . 41
3.2 Fundamental Properties of Hybrids . . . . . . . . . . . . . . . . . . . . . . . 47
3.3 The Category of Hybrids and Hybrid Maps . . . . . . . . . . . . . . . . . 54
3.4 The Category of Heyting Algebras and the Category of Strict
Hybrids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 61
3.5 Grzegorczyk Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 68
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 75
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

ix
Editors’ Note

The original title of Esakia’s book, translated into English, was ‘Heyting Algebras I.
Duality Theory’. As explained in the Foreword, volume II did not materialize, so
we have removed ‘I’ from the title. The planned contents of volume II are discussed
in detail in the Appendix.
Esakia was unhappy that a large number of typos and mathematical mistakes
were introduced in the Russian version of his book, which he could not correct, as
he was never given an opportunity to do a final proofreading. We took the liberty to
make the corrections without flagging them, as doing so would distract the reader.
We have, however, added specially marked editorial footnotes when we felt that
further explanation was in order. In addition, we filled in gaps in some proofs. Two
cases are noteworthy. First, we edited proofs at the end of Chap. 2, where we drew
from the paper “Scattered and hereditarily irresolvable spaces in modal logic” by
Guram Bezhanishvili and Patrick J. Morandi, Archive for Mathematical Logic, Vol.
49, 2010, pp. 343–365, as well as further communication between Guram and
Patrick. Second, Julia Ilin pointed out a gap in Esakia’s original proof of what is
now Theorem 5.13, so we replaced it by an algebraic version of the proof from pp.
158–9 of G. Boolos, The Logic of Provability (Cambridge University Press), 1993.
Since the original Russian publication of Esakia’s book, some of the terminol-
ogy in the area has changed. In particular, several objects and results are now
named after Esakia: e.g., Esakia spaces, Esakia’s lemma, the Blok-Esakia theorem,
etc. At the end of this note, we provide a table comparing some of Esakia’s original
terminology with modern terminology. In some cases, we have opted not to give the
most direct translation of a Russian term, opting instead for a more natural English
substitute (e.g., we use the term ‘skeletal Heyting algebra’ instead of the more
directly translated ‘stencil Heyting algebra’).
Throughout we have made light changes to notation for readability and con-
sistency. Parentheses are often omitted after functions and inverses, but they are
often added around meets and conjunctions. In some cases, we use modern symbols
in place of the original symbols (e.g., for sum and product) or adopt modern
notational practices (e.g., in the notation for functors in Sect. 3.3 of Chap. 3).
However, we retain Esakia’s convention that bold typeface signals a definition.

xi
xii Editors’ Note

Esakia’s original bibliography referred to Russian translations of some English

books. We have changed these entries to refer to the original English versions. In
addition, we changed some references from Russian to English translations, even
where the original work was in Russian.
The original Russian version of Esakia’s book had no index. We thank Evgeny
Kuznetsov for providing an index for this English translation. Each page number in
the index indicates the first page on which a phrase appears in bold typeface with its
definition.
Finally, we thank Luca Carai for carefully reading the manuscript and pointing
out a number of typos, and Patrick J. Morandi and Julia Ilin for the help noted
above.

Esakia’s notation Modern notation

Strict hybrid Esakia space
Hybrid Quasi-ordered Esakia space
Strongly isotone map p-morphism or bounded morphism
Skeletal filter Open filter or h-filter
Upper cone Upset
Lower cone Downset
Sub X PðXÞ or }ðXÞ
Con X UpðXÞ
X XðXÞ or OðXÞ
In Int or IPC
G GL
Grz Grz or S4:Grz

Guram Bezhanishvili
Wesley H. Holliday
Introduction

Although many results in the theory of Heyting lattices were discovered by several
authors, the subject owes its existence to the work of Alfred Tarski. Due to his
seminal work at the end of the 1930s and in the beginning of the 1940s, the theory
of Heyting lattices (or Brouwerian lattices, as they were called at the time) became
an independent area with its internal problems, playing an important role in modern
mathematics. G. Birkhoff, the founder of lattice theory, expressed the following
opinion about Heyting lattices at one of the symposia on lattice theory:
One easily shows that the open subsets of any T1 -space X form a (dually atomic, complete)
Brouwerian lattice L(X). Moreover, this lattice determines X up to homeomorphism.
On the other hand, the notion of a Brouwerian lattice was originally invented by
Brouwer, and formalized by Heyting, to provide a model for a logical system in which
proofs by contradiction are excluded (“intuitionist logic”). Thus, intuitionist logic and a
question in topology give rise to exactly the same class of lattices.
Finally, as was shown by Nakayama and Funayama, the (complete, algebraic) “structure
lattice” HðLÞ of all congruence relations on any given lattice L is also a Brouwerian lattice.
Here, we have a single technical lattice-theoretic concept, which plays an important role in
logic, set theory, and algebra alike! This illustrates the unifying power of lattice theory—
and the essential unity of all mathematics. [1, pp. 20–21]

In addition to the quote of Birkhoff, today it is worthwhile to mention the invasion

of Heyting algebras into category theory: the subobject classifier of any topos forms
a Heyting algebra. Finally, in the theory of continuous Scott lattices, which has
many applications, every distributive continuous lattice is a Heyting lattice.
This little book is dedicated to Heyting algebras and closure algebras, which are
closely related to Heyting algebras (by an embedding procedure). The author
attempted to develop duality theory for Heyting algebras, connecting the remark-
able Stone duality with Kripke models. Measures were taken to make the book
accessible to those with little knowledge of logic, especially with such “exotic”
areas of it as modality and intuitionism. The result is far from being perfect.
It is well known that the class of Heyting lattices, being a special class of
distributive lattices containing all Boolean lattices, is situated strictly in between
distributive lattices and Boolean lattices. The theory of distributive lattices is

xiii
xiv Introduction

covered in many books. For example, in the encyclopedic monograph by Birkhoff

[2], in the book by Grätzer [3], where the theory of distributive lattices is discussed
especially carefully and completely, and in the book by Skornjakov [4]. The theory
of Boolean algebras is covered in great detail, with many examples, in the books by
Sikorski [5] and Vladimirov [6]; I would also like to mention an interesting small
book by Halmos [7] and an elegant “miniature” by Dwinger [8] (only 60 pages!).
Regrettably, there are no monographs on Heyting lattices except the book by
Rasiowa and Sikorski [9], which has certainly played an important role. There is a
special chapter in the book, called “Pseudo-Boolean algebras,” that is dedicated to
Heyting algebras. The book, however, is very metamathematical.
My attempt to present some results on Heyting algebras (and closure algebras)
cannot fill this gap. One can hope that the material in the book enriches the chapter
“Heyting Algebras” of the very interesting—in both content and presentation—
monograph Distributive Lattices by Balbes and Dwinger [10]. This may be the only
book in which special chapters are dedicated to the algebras of logical “origin”:
Heyting algebras (Ch. IX), Post algebras (Ch. X), De Morgan algebras and
Lukasiewicz algebras (Ch. XI). This opinion, however, may be subjective.
The distribution of content among the chapters of the book is as follows.
Chapter 1, as indicated by its title, contains preliminary material and develops the
adopted system of notation. One can find necessary background from universal
algebra (Sect. 1.1), category theory (Sect. 1.2), and topology (Sect. 1.3). The content
of Sect. 1.4 on ordered sets and clusters is slightly less traditional. The definition of
Heyting lattices, with the most important examples, is given in Sect. 1.5.
Chapter 2 discusses the theory of Heyting algebras and closure algebras. The
connection with modal systems and intuitionistic logic is made in Sect. 2.3.
Important notions of skeletons and skeletal closure algebras are introduced, and the
equivalence of the category of Heyting algebras and the category of skeletal closure
algebras is proved in Sect. 2.5.
The first section of Chap. 3 introduces the key notion of a hybrid that
“cross-breeds” topology (Stone space) and order (Kripke frame). The main prop-
erties of hybrids are given in Sect. 3.2. The category of hybrids and hybrid maps is
defined in Sect. 3.3. A detailed proof of duality between the category of Heyting
algebras and the category of strict hybrids is given in Sect. 3.4. From this follows
the representation theorem that an arbitrary Heyting algebra is isomorphic to the
ring of cones of a suitable partially ordered set. Finally, Sect. 3.5 is almost entirely
dedicated to Grzegorczyk algebras, which have an interesting origin, curious
properties, and are close relatives of Heyting lattices. All sections conclude with
brief bibliographic notes.
The Appendix contains a very brief survey of the contents of the second part
of the monograph entitled “Heyting algebras II. Additional chapters.” The author
hopes that, despite its brevity, the appendix will allow the reader to have an idea
Introduction xv

of the content of Part II. The book concludes with a list of references, with no
claims of completeness.1
In conclusion, I would like to note that the book is based on the special course
given by the author in the 1970s at the Faculty of Mechanics and Mathematics of
Tbilisi State University. The opportunity to give such a course was kindly offered to
me by the head of the Department of Algebra and Geometry and a fellow of the
Georgian Academy of Sciences, George Chogoshvili, whose expertise I have been
exploiting for a quarter of a century. I consider it a pleasant duty and honor to
express publicly my deep gratitude to Chogoshvili.

References

1. Abbott, J. C. (Eds.). (1970). Trends in lattice theory. Symposium held at the United States
Naval Academy in May of 1966. Van Nostrand Reinhold Mathematical Studies, No. 31. Van
Nostrand Reinhold Co., New York.
2. Balbes, R., & Dwinger, Ph. (1974). Distributive lattices. Columbia, MO: University of
Missouri Press.
3. Birkhoff, G. (1967). Lattice theory (3rd ed., Vol. XXV). American Mathematical Society
Colloquium Publications, American Mathematical Society, Providence, R.I.
4. Dwinger, Ph. (1961). Introduction to Boolean algebras. Hamburger Mathematische
Einzelschriften, Heft 40. Würzburg: Physica-Verlag.
5. Grätzer, G. (1978). General lattice theory. Basel: Birkhäuser Verlag.
6. Halmos, P. R. (1963). Lectures on Boolean algebras. Van Nostrand Mathematical Studies,
No. 1. Princeton, N.J.: D. Van Nostrand Co., Inc.
7. Rasiowa, H., & Sikorski, R. (1963). The mathematics of metamathematics. Monografie
Matematyczne, Tom 41. Warsaw: Państwowe Wydawnictwo Naukowe.
8. Sikorski, R. (1960). Boolean algebras. Ergebnisse der Mathematik und ihrer Grenzgebiete, N.
F., Heft 25. Berlin: Springer.
9. Skornjakov, L. A. (1977). Elements of lattice theory (Translated from the Russian by V.
Kumar, Adam Hilger, Ltd., Bristol). Delhi: Hindustan Publishing Corp.
10. Vladimirov, D. A. (1969). Boolean algebras (Russian). Moscow: Izdat. “Nauka”.

1
Editorial note: In this edition, references appear after each chapter instead of at the end of the
book.
Chapter 1
Preliminary Notions and Necessary Facts

1.1 Universal Algebra

For two sets A, B, let A ∪ B (resp. A ∩ B) denote the union (resp. intersection) of
these sets. Let ∅ denote the empty set. If A ∩ B = ∅, we say that the sets A, B are
disjoint. The difference of sets A and B, i.e., the set of elements of A that do not
belong to B, will be denoted by A − B. If f is a map from a set X to a set Y and
A ⊆ X , B ⊆ Y , then f (A) (the f -image of A) denotes the set of all elements f (x)
where x ∈ A, and f −1 (B) (the f -preimage of B) denotes the set of all x ∈ X such
that f (x) ∈ B. If f maps X to Y (i.e., f : X → Y ), then f is said to be surjective if
f (X ) = Y , injective if f (x) = f (y) implies x = y, and bijective if f is surjective
and injective. If X is an arbitrary set  and F a family of nonempty subsets of X ,
then F is said to be a partition of X if {A : A ∈ F} = X and A ∩ A = ∅ for any
A, A ∈ F such that A = A .
Let K be an arbitrary class of algebras of a given fixed signature and A1 , A2
two algebras in the class K . A map h : A1 → A2 is said to be a homomorphism
if h preserves all signature operations of these algebras. A homomorphism h is
said to be surjective (resp. injective, bijective) if h is surjective (resp. injective,
bijective) as a map. The terms epimorphism, monomorphism, bimorphism will be
used in their categorial meaning (see below). The notions of subalgebra, quotient
algebra, homomorphic image, congruence relation, and direct product have their
usual meaning. Let K be an arbitrary class of algebras of a given fixed signature. We
adopt the following notation:
I(K ) is the class of all isomorphic copies of algebras from the class K ;
H(K ) is the class of all homomorphic images of algebras from K ;
S(K ) is the class of all subalgebras of algebras from K ;
P(K ) is the class of all direct products of nonempty families of algebras belonging
to K .
Recall that a hereditary (S(K ) ⊆ K ), homomorphically (H(K ) ⊆ K ), and multi-
plicatively (P(K ) ⊆ K ) closed class K of algebras is said to be a variety. It is known

© Springer Nature Switzerland AG 2019 1

G. Bezhanishvili and W. H. Holliday (eds.), Heyting Algebras,
Trends in Logic 50, https://doi.org/10.1007/978-3-030-12096-2_1
2 1 Preliminary Notions and Necessary Facts

that a class K of algebras is a variety if and only if it is equationally definable.1

For any class K , the class HSP(K ) is the smallest variety containing K . The class
HSP(K ) is said to be the variety generated by K . If K = {A}, we write HSP(A).
As Tarski showed, a class K is equationally definable if and only if K = HSP(A)
for some algebra A.
An algebra A is said to be a subdirect product of a family of algebras{A j : j ∈ J}
if there exists an injective homomorphism (i.e., an embedding) h : A → {A j :
j ∈ J } of the algebra A into the direct product of the family {A j: j ∈ J } such that for
each j ∈ J , the composition p j ◦ h is surjective, where p j : {A j : j ∈ J } → A j
is the projection map. An algebra A is said to be subdirectly irreducible if A being
a subdirect product of a family {A j : j ∈ J } implies that p j ◦ h is an isomorphism
for some j ∈ J , where h : A → {A j : j ∈ J } is the corresponding embedding.
1.1.1 Theorem An algebra is subdirectly irreducible if and only if the set of all its
nontrivial congruences has a least element.
1.1.2 Theorem If K is a variety, then each algebra in K can be decomposed into a
subdirect product of subdirectly irreducible algebras from K .
According to this theorem, each variety is completely determined by its class
of subdirectly irreducible algebras. One can ‘make’ this class even smaller, as the
following theorem shows.
1.1.3 Theorem A variety K is generated by its subdirectly irreducible finitely gen-
erated algebras.
If a variety is generated by its finite algebras, then it is said to be finitely approx-
imable. If all finitely generated algebras of a variety are finite, the variety is said to
be locally finite.
A class K of algebras is said to have the congruence extension property if for
each subalgebra A0 of an algebra A ∈ K and each congruence 0 on A0 , there is
a congruence  on A such that  ∩ A20 = 0 . If K is a variety, then K has the
congruence extension property if and only if for each embedding f : A1 → A2 and
each surjective homomorphism g : A1 → A3 , there exist a surjective homomorphism
h : A2 → A4 and an embedding k : A3 → A4 such that h ◦ f = k ◦ g.
A class K of algebras is said to be congruence distributive if the lattice of
congruences (A) of each algebra A in K is distributive.
Bibliographic notes. For details the reader is referred to any monograph on universal
algebra, e.g., [4, 9, 12]. But Part II of the monograph [1] is certainly sufficient.

1.2 Categories

A category K 1 is said to be a subcategory of a category K 2 if

1 Editorial note: A class K of algebras is equationally definable if there is a set of identities (see

Sect. 26 of [9]) such that K is exactly the class of algebras satisfying those identities.
1.2 Categories 3

(1) each object in K 1 is an object in K 2 ;

(2) for arbitrary objects A, B in K 1 , each morphism f : A → B in K 1 is a morphism
in K 2 .
The subcategory K 1 is said to be full if, in addition,
(3) for arbitrary objects A, B in K 1 , each morphism f : A → B in K 2 is a morphism
in K 1 .
A morphism f in a category K is said to be a monomorphism if f ◦ g = f ◦ h
implies g = h for arbitrary morphisms g and h in K . A morphism f is said to be
an epimorphism if g ◦ f = h ◦ f implies g = h. A morphism f is said to be a
bimorphism if f is an epimorphism and a monomorphism. Finally, a morphism
f : A → B is said to be an isomorphism if there is a morphism g : B → A such
that g ◦ f = 1 A and f ◦ g = 1 B . In this case, we write g = f −1 and say that A
and B are isomorphic. Every isomorphism is a bimorphism. The converse is not
true in general. If morphisms in a category K are simply maps, then every injective
morphism is a monomorphism and every surjective morphism is an epimorphism.
An object A in a category K is said to be injective if for any monomorphism
f : B → C and any morphism g : B → A there exists a morphism h : C → A such
that h ◦ f = g. An object A in K is said to be projective if for any epimorphism
f : C → B and any morphism g : A → B there is a morphism h : A → C such that
f ◦ h = g.
Categories of algebras. Let K be a class of algebras. Unless stated otherwise, we
will associate with the class K the category K (denoted by the same letter and called
an algebraic category), in which objects are exactly the algebras of the class K and
morphisms are the homomorphisms, i.e., maps preserving the signature operations.
It is important to note that if K is an algebraic category, then the notions of algebraic
and categorical isomorphisms coincide.
Equational categories. An algebraic category K is said to be equational if its class
of objects forms a variety. If K is an equational category, then monomorphisms
coincide with injective homomorphisms, and a direct product of algebras is their
product in the categorical sense, but not every epimorphism is surjective.

1.2.1 Definition Let K be an equational category and A an object in K . A subalgebra

A0 of the algebra A is said to be dense in A if the inclusion map h : A0 → A is an
epimorphism.

A subalgebra A0 of A is dense if and only if for any two morphisms f, g : A → B,

if the restrictions of f, g to A0 coincide, then f = g. A category is said to be balanced
if each bimorphism is an isomorphism. If K is an equational category, then the
following three conditions are equivalent:
(1) K is balanced.
(2) Each epimorphism is surjective.
(3) Each dense subalgebra A0 of an algebra A ∈ K coincides with A.
4 1 Preliminary Notions and Necessary Facts

The property of being balanced is rather strong: even equational categories are
often not balanced, and being balanced implies strong additional properties.
An equational category K is said to be amalgamable if for any monomor-
phisms f 1 : A0 → A1 , f 2 : A0 → A2 , there are monomorphisms g1 : A1 → B,
g2 : A2 → B such that g1 ◦ f 1 = g2 ◦ f 2 .
Let K 1 and K 2 be arbitrary categories. A covariant functor (resp. contravariant
functor) F : K 1 → K 2 is an assignment of an object F(A) in K 2 to any object A in
K 1 and of a morphism F( f ) : F(A) → F(B) in K 2 (resp. F( f ) : F(B) → F(A))
to any morphism f : A → B in K 1 such that
(1) F(1 A ) = 1 F(A) for any object A in K 1 ;
(2) F( f ◦ g) = F( f ) ◦ F(g) (resp. F( f ◦ g) = F(g) ◦ F( f )) for any morphisms
f, g in K 1 for which f ◦ g is defined.

1.2.2 Definition
(1) A covariant functor F : K 1 → K 2 is said to be an equivalence and the categories
K 1 and K 2 are said to be equivalent if
(a) for any object A2 in K 2 , there is an object A1 in K 1 such that F(A1 ) and A2
are isomorphic;
(b) for any objects A, B in K 1 , the map from the set of all morphisms from A
to B to the set of all morphisms from F(A) to F(B) induced by the functor
is bijective.
(2) A contravariant functor F : K 1 → K 2 is said to be a coequivalence (or duality)
and K 1 , K 2 are said to be dual if
(a) for any object A2 in K 2 , there is an object A1 in K 1 such that F(A1 ) and A2
are isomorphic;
(b) for any objects A, B in K 1 , the map from the set of all morphisms from A to
B to the set of all morphisms from F(B) to F(A) induced by F is bijective.

Bibliographic notes. The reader is referred to any book on category theory. Part III
of the monograph [1] is certainly sufficient.

1.3 Topologies

A family K of subsets of an arbitrary set X is said to be a ring of sets (over X )

if A, B ∈ K imply A ∪ B, A ∩ B ∈ K. A ring of sets is said to be a field of sets if
A ∈ K implies −A ∈ K (i.e., X − A ∈ K). A ring of sets K (resp. a field of sets F)
is said to be reduced if for any x, y ∈ X with x = y, there is A ∈ K (resp. A ∈ F)
such that x ∈ A and y ∈ / A or vice versa. Let Sub X (resp. Subω X ) denote the field
of all subsets of X (resp. the field of all finite subsets of X and their complements).
1.3 Topologies 5

A topology on a set X is a family  of subsets of X closed under arbitrary unions

and finite intersections. Sets from  are said to be open.2 A set X equipped with a
topology is said to be a topological space. A family B of open subsets of a space
(X, ) is said to be a basis if any set from  is a union of some sets from B. A class
B0 of open sets is said to be a subbasis if the family B consisting of ∅, X and all
finite intersections of elements of B0 is a basis of X .
A topological space (X, ) is said to  be compact if for any family
{U j : j ∈ J } ⊆  of open  sets such that X = {U j : j ∈ J } there exists a finite
set J0 ⊆ J such that X = {U j : j ∈ J0 }. A topological space X is said to be a
Kolmogoroff space (or a T0 -space) if for any two distinct points, one of the points
belongs to an open set not containing the other point. A topological space is said to
be a T1 -space if every singleton {x} is a closed set. A topological space X is said to
be Hausdorff if for any two distinct x, y ∈ X , there exist disjoint open sets U1 and
U2 such that x ∈ U1 and y ∈ U2 .

1.3.1 Definition A space X is said to be zero-dimensional if the small inductive

dimension of X is zero (ind X = 0), i.e., for any point x ∈ X and any open set U
containing x, there is a closed open set A such that x ∈ A ⊆ U .3

Note that if a space X is zero-dimensional, then the family of all closed open sets
of X is a basis for the space. For the purposes of brevity and convenience, following
P. Halmos, we call closed and open sets clopen.

1.3.2 Proposition If X is a compact Hausdorff space and F is a reduced field of

clopen sets of X , then F coincides with the field of all clopen sets of X .4

1.3.3 Definition A Stone space is a compact Hausdorff zero-dimensional space.

Note that every Stone space X is totally disconnected, i.e., for any two distinct
points x, y ∈ X , there is a partition of X into two open sets U1 and U2 such that
x ∈ U1 , y ∈ U2 . A space X is said to be extremally disconnected if the closure of
each open set is open, i.e., if each regular open set is clopen.5
Let X be a topological space. Let exp X denote the family of all nonempty closed
subsets of X .

1.3.4 Definition The exponential topology (Vietoris topology) on exp X

is the topology whose subbasis is the family of all sets of the form
F1 (U ) = {F ∈ exp X : F ∩ U = ∅} and F2 (U ) = {F ∈ exp X : F ⊆ U }, where U
is an arbitrary open subset of X .

2 Editorial note: As usual, a set is closed if its complement is open.

3 Editorial note: For the general definition of small inductive dimension, see, e.g., Sect. 7.1 of R.
Engelking, General Topology, 2nd ed. (Heldermann-Verlag), 1989.
4 Editorial note: For a proof, see Lemma 1 on p. 74 of [10].
5 Editorial note: Recall that a set is regular open if it is equal to the interior of its closure.
6 1 Preliminary Notions and Necessary Facts

1.3.5 Proposition If X is a Stone space, then so is exp X .6

If X is a Stone space, then the sets of the form F1 (A) and F2 (A), where A is a
clopen subset of X , form a subbasis of exp X .
A subset U of a topological space X is said to be dense if the closure of U
coincides with X .

1.3.6 Definition
(1) A map f from a topological space X to a space Y is said to be continuous if the
f -preimage of each open set in Y is open in X .
(2) A bijective map f : X → Y is said to be a homeomorphism, and the spaces X
and Y are said to be homeomorphic, if f and f −1 are continuous.
(3) A map f : X → Y is said to be closed (resp. open) if the f -image of each closed
(resp. open) set in X is closed (resp. open) in Y .

Any continuous map from a compact space to a Hausdorff space is closed. If, in
addition, f is bijective, then f is a homeomorphism.
 Let F = {(X j ,  j ) : j ∈ J } be a nonempty family of topological spaces, X =
{X j : j ∈ J } the direct (cartesian) product of the sets X j ( j ∈ J ), and p j :
X → X j the projection of X onto X j (i.e., p j (x) = x( j) ∈ X j for each x ∈ X ).
Equip the set X with the topology  by defining a subbasis to be the family of all
sets p −1
j (U ) or, equivalently, all sets of the form {x ∈ X : x( j) ∈ U } where U is an
arbitrary open set in X j . The topology  is said to be the product topology, and
the space X is said to be the (topological) product of the X j ( j ∈ J ). The product
topology on X is the weakest topology making all projections continuous.
Bibliographic notes. [3, 11].

1.4 Ordered Sets and Clusters

1.4.1 Definition A binary relation R on a set X is said to be

(1) reflexive if (∀x)(x Rx);
(2) transitive if (∀x, y, z)((x Ry & y Rz) ⇒ x Rz);
(3) connected if (∀x, y)(x Ry ∨ y Rx);
(4) symmetric if (∀x, y)(x Ry ⇒ y Rx);
(5) anti-symmetric if (∀x, y)((x Ry & y Rx) ⇒ x = y);
(6) upward directed (resp. downward directed) if (∀x, y)(∃z)(x Rz & y Rz) (resp.
(∀x, y)(∃z)(z Rx & z Ry)).

6 Editorialnote: This well-known result is due to E. Michael, “Topologies on spaces of subsets,”

Transactions of the American Mathematical Society 71: 152–182, 1951. For a proof using the same
definitions as above, see Lemma 2.12 of C. Kupke, A. Kurz, and Y. Venema, “Stone coalgebras,”
ILLC Report PP-2003-15.
1.4 Ordered Sets and Clusters 7

1.4.2 Definition A pair (X, R) is said to be a quasi-ordered set if X is a set and R

is a reflexive and transitive relation on X . If, in addition, R is anti-symmetric, then
(X, R) is said to be a partially ordered set.

1.4.3 Definition A subset A of a quasi-ordered set (X, R) is an upper (resp. lower)

cone if x ∈ A and x Ry (resp. y Rx) imply y ∈ A.7
We will often say ‘cone’ omitting the word ‘upper’. It is also convenient to say
‘A is an upper (resp. lower) set’ instead of ‘ A is an upper (resp. lower) cone’.

1.4.4 Proposition Let (X, R) be a quasi-ordered set and A ⊆ X . The following

conditions are equivalent:
(1) A is an upper cone;
(2) R(x) ⊆ A for each x ∈ A, where
 R(x) = {y ∈ X : x Ry};
(3) R(A) ⊆ A, where R(A) = {R(x) : x ∈ A};
(4) A = R(A0 ) for some A0 ⊆ X .

It is also
 obvious that any set of the form R −1 (A) is a lower cone, where
R (A) = {R (x) : x ∈ A} and R −1 (x) = {y ∈ X : y Rx}. Throughout the fol-
−1 −1

lowing notation will be used interchangeably:

x Ry, R(x, y), (x, y) ∈ R, y ∈ R(x), x ∈ R −1 (y).

1.4.5 Definition Upper (resp. lower) cones of the form R(x) (resp. R −1 (x)), where
x ∈ X , will be called principal.

The following statement is easy to verify.

1.4.6 Proposition Let (X, R) be a quasi-ordered set. Then:

(1) the intersection (resp. union) of any family of cones is a cone;
(2) the set-theoretic complement of a lower (resp. upper) cone is an upper (resp.
lower) cone;
(3) each upper (resp. lower) cone A is the union of all principal upper (resp. lower)
cones contained in A.

1.4.7 Definition A cluster of a quasi-ordered set (X, R) is a set of the form R(x) ∩
R −1 (x), where x ∈ X .

Obviously each cluster A is a set of mutually connected points, i.e., if x, y ∈ A,

then x Ry and y Rx. A cluster A is said to be nontrivial if the set A contains at
least two distinct points. Note that the classes E(x) = {y ∈ X : x E y} of the natural
partition of X induced by the equivalence relation

x E y ⇔ x Ry & y Rx

7 Editorial
note: Upper (resp. lower) cones are now commonly called ‘upward closed sets’ (resp.
‘downward closed sets’) or simply ‘upsets’ (resp. ‘downsets’).
8 1 Preliminary Notions and Necessary Facts

are exactly the clusters of the quasi-ordered set (X, R), and

E(x) = R(x) ∩ R −1 (x) for x ∈ X.

The corresponding partially ordered quotient set (X/E, R E ), where

E(x)R E E(y) ⇔ ∃x  ∈ E(x) ∃y  ∈ E(y)(x  Ry  ),

is called the skeleton of the quasi-ordered set (X, R). Due to the importance of
clusters for us, we make a few additional remarks. A quasi-ordered set is partially
ordered if and only if it has no non-trivial clusters. Each quasi-ordered set X can be
turned into a closely related partially ordered set X/E (the skeleton) by “gluing” each
cluster of X to a point. The converse procedure of “blowing up” points of a partially
ordered set Y to clusters (i.e., replacing a point x ∈ Y with a corresponding set A x )
gives a quasi-ordered set X , whose skeleton X/E coincides (up to isomorphism, of
course) with Y . Moreover, each quasi-ordered set can be obtained this way. Therefore,
the prefix ‘quasi’ indicates the possibility of existence of non-trivial clusters. The
following can be considered as an unwritten law of the theory of ordered sets:
If one adds the prefix “quasi-” to any notion, property, etc., which is applicable to
a partially ordered set Y , the resulting notion, property, etc. is applicable to any quasi-
ordered set X whose skeleton X/E coincides with Y . For example, the statement “Z
is a quasi-chain of X ” means that the skeleton of the set Z (i.e., Z /E) is a chain of
the skeleton X/E of the quasi-ordered set X (see below).
1.4.8 Definition Let (X, R) be a quasi-ordered set and E an equivalence relation
(i.e., a reflexive, transitive, and symmetric relation) on X . Then the partition of X
induced by E is the family of classes E(x) = {y ∈ X : x E y} for x ∈ X . A set A
contained in X is said to be E-saturated if A = E(A) = {E(x) : x ∈ A}, i.e., if
A is a union of equivalence classes of the partition. Obviously E-saturated sets are
exactly the sets of the form E(B), where B is a subset of X .
1.4.9 Definition Let (X, R) be a quasi-ordered set. A point x ∈ A is said to be a
maximal (resp. quasi-maximal) point of A ⊆ X if for any y ∈ A, from x Ry it
follows that x = y (resp. y Rx). The set of all maximal points of A will be denoted
by max A. Note that the notions of maximal and quasi-maximal points coincide in
partially ordered sets (X, R).8
If the relation R of a quasi-ordered set (X, R) is connected, then (X, R) is said
to be a quasi-linearly ordered set (or a quasi-chain). If, in addition, the relation R
is anti-symmetric, then (X, R) is said to be a linearly ordered set (or a chain). A
subset Y of a quasi-ordered set (X, R) is said to be a quasi-chain (resp. chain) if
Y with the inherited order, i.e., (Y, RY ), is a quasi-chain (resp. chain). We will need
the following principle, which is equivalent to the axiom of choice.

8 Editorialnote: The concept of a minimal (resp. quasi-minimal) point is defined dually. The set
of all minimal points of A will be denoted by min A.
1.4 Ordered Sets and Clusters 9

1.4.10 Proposition (Hausdorff’s Principle) Any chain of a partially ordered set is

contained in a maximal chain.

1.4.11 Definition Let (X i , Ri ) be quasi-ordered sets, i = 1, 2. A map f : X 1 → X 2

is said to be strongly isotone9 if for any x ∈ X 1 , y ∈ X 2 ,

f (x)R2 y ⇔ ∃x  ∈ X 1 (x R1 x  & f (x  ) = y),

or equivalently,
f (x) ∈ R2−1 (y) ⇔ R1 (x) ∩ f −1 (y) = ∅.

Obviously if a map f is strongly isotone, then it is isotone, i.e., x R1 y implies

f (x)R2 f (y) for each x, y ∈ X 1 . The following simple but useful proposition holds.

1.4.12 Proposition Let (X i , Ri ) be quasi-ordered sets (i = 1, 2) and f : X 1 → X 2

a map. The following conditions are equivalent:
(1) f is strongly isotone;
(2) the f -image and f -preimage of a cone are cones; more precisely, if A is a cone
in (X 1 , R1 ) (resp. in (X 2 , R2 )), then f (A) (resp. f −1 (A)) is a cone in (X 2 , R2 )
(resp. in (X 1 , R1 ));
(3) R1−1 ( f −1 (y)) = f −1 (R2−1 (y)) for each y ∈ X 2 ;
(4) R2 ( f (x)) = f (R1 (x)) for each x ∈ X 1 .

Proof (1 ⇒ 2) Let f : X 1 → X 2 be a strongly isotone map and A a cone in X 1 . We

show that f (A) is a cone in X 2 . Let y ∈ f (A), i.e., f (x) = y for some x ∈ A, and let
y R2 y  . Since f is strongly isotone, there is x  such that x R1 x  and f (x  ) = y  . As A is
a cone, x  ∈ A, and so f (x  ) = y  ∈ f (A). Therefore, f (A) is a cone in X 2 . Now we
suppose that A is a cone in X 2 and show that f −1 (A) is a cone in X 1 . Let x ∈ f −1 (A),
i.e., f (x) ∈ A, and let x R1 x  . We show that x  ∈ f −1 (A), i.e., f (x  ) ∈ A. Since f
is isotone, x R1 x  implies f (x)R2 f (x  ), and as A is a cone, f (x  ) ∈ A.
(2 ⇒ 1) Suppose (2) holds. Let y ∈ R2 ( f (x)). Since R1 (x) is a cone in X 1 ,
f (R1 (x)) is a cone in X 2 . As x ∈ R1 (x), f (x) ∈ f (R1 (x)). Since f (x)R2 y and
f (R1 (x)) is a cone in X 2 , y ∈ f (R1 (x)). Therefore, there is x  ∈ X 1 such that
x R1 x  and f (x  ) = y. Conversely, let x  ∈ X 1 , x R1 x  , and f (x) = y. We show
that f (x)R2 y, i.e., that y ∈ R2 ( f (x)). Consider the cone f −1 (R2 ( f (x))). Clearly
x ∈ f −1 (R2 ( f (x))). Since x R1 x  , x  ∈ f −1 (R2 ( f (x))), i.e., f (x  ) ∈ R2 ( f (x)). But
f (x  ) = y. Consequently, conditions (1) and (2) are equivalent.
(1 ⇔ 3) The equation R1−1 ( f −1 (y)) = f −1 (R2−1 (y)) is equivalent to the
following:
x ∈ R1−1 ( f −1 (y)) ⇔ x ∈ f −1 (R2−1 (y)).

Transform the right-hand side to obtain:

9 Editorial note: Strongly isotone maps are now commonly called ‘p-morphisms’ or ‘bounded mor-

phisms’.
10 1 Preliminary Notions and Necessary Facts

x ∈ f −1 (R2−1 (y)) ⇔ f (x) ∈ R2−1 (y) ⇔ f (x)R2 y.

Transform the left-hand side to obtain:

x ∈ R1−1 ( f −1 (y)) ⇔ (∃x  )(x R1 x  & f (x  ) = y).

Thus, f (x)R2 y ⇔ (∃x  )(x R1 x  & f (x  ) = y).

(1 ⇔ 4) The equation R2 ( f (x)) = f (R1 (x)) is equivalent to the following:

y ∈ R2 ( f (x)) ⇔ y ∈ f (R1 (x)),

i.e., to f (x)R2 y ⇔ (∃x  )(x R1 x  & f (x  ) = y). 

1.4.13 Proposition Let (X, R) be a quasi-ordered set, (X/E, R E ) its skeleton, and
f : X → X/E the canonical map, i.e., f (x) = E(x) for each x ∈ X . Then f is a
strongly isotone map.
Proof Let A bea cone in (X, R). We  show that f (A) is a cone in (X/E, R E ). We
have f (A) = { f (x) : x ∈ A} = {E(x) : x ∈ A}. Suppose that E(y) ∈ f (A)
and E(y)R E E(z). Then y ∈ A and y Rz, and since A is a cone, z ∈ A. Therefore,
E(z) ∈ f (A). Thus, f (A) is a cone. Now let B be a cone in (X/E, R E ). We show that
f −1 (B) is a cone in (X, R). Suppose y ∈ f −1 (B) and y Rz. Then f (y) ∈ B. There-
fore, E(y) ∈ B. Since y Rz, we have E(y)R E E(z). As B is a cone in (X/E, R E ), we
have E(z) ∈ B, so f (z) ∈ B. Thus, z ∈ f −1 (B), and so f −1 (B) is a cone in (X, R).
Consequently, f is strongly isotone by Proposition 1.4.12(2). 
1.4.14 Definition Let (X i , Ri ) be quasi-ordered sets (i = 1, 2). A map f : X 1 →
X 2 is said to be an isomorphism (and (X 1 , R1 ), (X 2 , R2 ) are said to be isomorphic)
if f is a bijection and f, f −1 are isotone maps.
The following proposition is easy to check.
1.4.15 Proposition A map f : X 1 → X 2 is an isomorphism if and only if f is a
bijection satisfying
x R1 y ⇔ f (x)R2 f (y)

for each x, y ∈ X 1 .
1.4.16 Proposition Let f : X 1 → X 2 be a strongly isotone bijection. Then f is an
isomorphism.
Proof Since f is strongly isotone, x R1 y ⇒ f (x)R2 f (y). Let f (x)R2 f (y).
Then there is x  such that x R1 x  and f (x  ) = f (y). As f is injective, x  = y. Thus,
x R1 y. 
1.4.17 Proposition Let (X i , Ri ) be quasi-ordered sets (i = 1, 2) and f : X 1 → X 2
an isotone map. Then (1) implies (2), and if f is a bijection, then (1) and (2) are
equivalent, where:
1.4 Ordered Sets and Clusters 11

(1) f is a strongly isotone map;

(2) f (x)R2 f (y) implies the existence of x  , y  ∈ X 1 such that f (x) = f (x  ), f (y) =
f (y  ), and x  R1 y  (i.e., f is a strong homomorphism of the model (X 1 , R1 ) to
the model (X 2 , R2 ) in the sense of model theory).

Proof (1 ⇒ 2) Let f (x)R2 f (y). By (1), there is y  such that x R1 y  and f (y) = f (y  ).
Therefore, x R1 y  , f (x) = f (x), and f (y) = f (y  ). Thus, (2) holds.
(2 ⇒ 1) Let f be a bijection and suppose f (x)R2 z. Since f −1 (z) = ∅, there
exists y ∈ f −1 (z), i.e., f (y) = z and, by assumption, f (x)R2 f (y). Therefore, by
(2), there exist x  , y  such that f (x) = f (x  ), f (y) = f (y  ), and x  R1 y  . As f is
also an injection, x = x  and y = y  . Thus, x R1 y and f (y) = z. 

Bibliographic notes. See the first chapter of [2]. Strongly isotone maps were
introduced in [5].10

1.5 Heyting Lattices

1.5.1 Definition A lattice (L , ≤) is said to be a lattice with relative pseudocom-

plement if for any elements a, b ∈ L, the set {d ∈ L : a ∧ d ≤ b} has a greatest
element, which is denoted by a → b.

Note that a lattice with relative pseudocomplement → always has a greatest

element 1, since for each a, we have a → a = 1.

1.5.2 Definition A lattice L with relative pseudocomplement → is said to be a

Heyting lattice if L has a least element 0. The element a → 0 of a Heyting lattice
(L , ≤) is called the pseudocomplement of a and is denoted by ¬a.

The existence of a relative pseudocomplement (implication for short) yields a

number of important properties.

1.5.3 Proposition A Heyting lattice L is a bounded distributive lattice.11

Proof That L is bounded is obvious. Let a, b, c ∈ L and let d denote (a ∧ b) ∨ (a ∧

c). Since a ∧ b ≤ d and a ∧ c ≤ d, we have b ≤ a → d and c ≤ a → d. Therefore,
b ∨ c ≤ a → d, and so

a ∧ (b ∨ c) ≤ a ∧ (a → d) ≤ d = (a ∧ b) ∨ (a ∧ c).

Thus, a ∧ (b ∧ c) ≤ (a ∧ b) ∨ (a ∧ c). Since a ∧ (b ∨ c) ≥ (a ∧ b) ∨ (a ∧ c) holds

in every lattice, we have a ∧ (b ∨ c) = (a ∧ b) ∨ (a ∧ c). 

10 Editorial note: Also see D.H.J. de Jongh and A.S. Troelstra, “On the connection of partially
ordered sets with some pseudo-Boolean algebras,” Indigationes Mathematicae 28: 317–328, 1966.
11 Editorial note: For the definition of bounded distributive lattices, see Sect. 2.1 of Chap. 2.
12 1 Preliminary Notions and Necessary Facts

Complete Heyting lattices are characterized by the following.

1.5.4 Proposition A complete lattice L is a Heyting lattice if and only if the following
infinite distributive law holds in L:
  
a∧ ai = (a ∧ ai ) for any set I .
i∈I i∈I


Proof Let L be a complete Heyting lattice. Let b denote the element (a ∧ ai ).
i∈I
Clearly
 a ∧ ai ≤ b for each i ∈ I . Therefore, for each i ∈ I , ai ≤ a → b, so
ai ≤ a → b. Since a ∧ (a → b) ≤ b, we have a ∧ ai ≤ b = (a ∧ ai ). Note
i∈I   i∈I i∈I 
that (a ∧ ai ) ≤ a ∧ ai holds in every complete lattice. Thus, a ∧ ai =
 i∈I i∈I i∈I
(a ∧ ai ).
i∈I 
Conversely, let a, b ∈ L. We show that a → b = {d ∈ L : a ∧ d ≤ b}. By the
infinite distributive law, we have
 
a ∧ {d ∈ L : a ∧ d ≤ b} = {a ∧ d : a ∧ d ≤ b}.

Therefore, a ∧ {d ∈ L : a ∧ d ≤ b} ≤ b. If a∧ d  ≤ b for some d  ∈ L,
then
 d  ∈ {d ∈ L : a ∧ d ≤ b}, and so d  ≤ {d ∈ L : a ∧ d ≤ b}. Thus,
{d ∈ L : a ∧ d≤ b} is the greatest element of the set {d ∈ L : a ∧ d ≤ b}, and
hence a → b = {d ∈ L : a ∧ d ≤ b}. 
1.5.5 Corollary A finite distributive lattice is a Heyting lattice.
1.5.6 Definition A bounded distributive lattice L is said to be a Boolean lattice if
for each a ∈ L there exists b ∈ L such that a ∧ b = 0 and a ∨ b = 1 (b is called the
complement of a and is denoted by −a).12
Every Boolean lattice is a Heyting lattice in which a → b = −a ∨ b.
Conversely, a Heyting lattice is a Boolean lattice if and only if for each a ∈ L,
we have a ∨ ¬a = 1.
1.5.7 Proposition The following hold in every Heyting lattice:
(1) a → b = 1 ⇔ a ≤ b;
(2) a → a = 1;
(3) a → 1 = 1;
(4) 0 → b = 1;
(5) a ∧ (a → b) ≤ b;
(6) a ∧ (a → b) = a ∧ b;
(7) (a → b) ∧ b = b;

12 Editorial note: It is easy to see that if an element of a bounded distributive lattice has a complement,

then it has a unique complement. See, e.g., Sect. II.6 of [1].

1.5 Heyting Lattices 13

(8) (a → b) ∧ (a → c) = a → (b ∧ c);
(9) (a → c) ∧ (b → c) = (a ∨ b) → c;
(10) (a → b) ∧ c = ((c ∧ a) → (c ∧ b)) ∧ c;
(11) ((a ∧ b) → a) ∧ c = c;
(12) a ≤ b ⇒ ¬b ≤ ¬a;
(13) a ≤ ¬¬a;
(14) ¬¬¬a = ¬a;
(15) ¬(a ∨ b) = ¬a ∧ ¬b;
(16) ¬a ∨ ¬b ≤ ¬(a ∧ b);
(17) ¬a ∨ b ≤ a → b;
(18) ¬¬(a ∧ b) = ¬¬a ∧ ¬¬b;
(19) a ∧ b = 0 ⇔ a ≤ ¬b;
(20) a ∧ ¬a = 0.
1.5.8 Main examples of Heyting lattices.
(1) Topology. The family  of all open sets of a topological space X , ordered by
inclusion ⊆, is a Heyting lattice(the lattice  is complete
 and the infinite dis-
tributive law holds in it: A ∩ {A j : j ∈ J } = {A ∩ A j : j ∈ J } for each
A ∈  and A j ∈  where J is an arbitrary set).
(2) Lattices. The family (L) of congruences of each lattice L is a Heyting lattice
(the Funayama–Nakayama theorem).
(3) Distributive lattices. The family I (L) of ideals of each distributive lattice is
a Heyting lattice with respect to inclusion (the Stone theorem).13
(4) Continuous lattices. Every continuous distributive lattice is a Heyting lattice.
(Let L be a complete lattice. An element a ∈ L is said to be compact relative
to b ∈ L if for each family {ai ∈ L : i ∈ I }, from b ≤ {ai ∈ L : i ∈ I } it
follows that there is a finite set I0 ⊆ I such that a ≤ {ai ∈ L : i ∈ I0 }.14 A
lattice L is said to be continuous if every b ∈ L is the least upper bound of all
elements that are compact relative to b.)
(5) Categories. The subobject classifier  of any topos forms a Heyting lattice.
(A topos is a cartesian closed bicomplete category in which the subobject
classifier exists.)
(6) Logic. Propositions ordered by the intuitionistic provability relation form a
Heyting lattice.
(7) Ordered sets. The family of cones of any partially ordered set is a Heyting
lattice (with respect to inclusion).
(8) Any chain with 0 and 1 is a Heyting lattice (for any elements a and b, a →
b = 1 if a ≤ b, and a → b = b otherwise).
(9) Any Boolean lattice is a Heyting lattice, where a → b = −a ∨ b for any ele-
ments a, b.

13 Editorial note: An ideal of a lattice L is a nonempty subset I of L such that (i) if a ∈ I and b ≤ a,

then b ∈ I , and (ii) if a, b ∈ I , then a ∨ b ∈ I .

14 Editorial note: The relation of a being compact relative to b is more standardly called the way

below relation (see, e.g., Sect. I.1 of [7]).

14 1 Preliminary Notions and Necessary Facts


(10) Any finite distributive lattice L is a Heyting lattice (a → b = {d ∈ L : a ∧
d ≤ b}).

Bibliographic notes. The proof of Proposition 1.5.7 can be found in [1, 13]. Exam-
ples (1)–(3) of 1.5.8 can be found in [2]. Example (4) of 1.5.8 is in the book [7], which
is entirely dedicated to the theory of continuous lattices. For Example (5), see [6]
or [8]. The most complete exposition of the relationship between Heyting algebras
and intuitionistic logic (Example (6)) is in the book [13]. Examples (7)–(10) can be
found in [2].

References

1. Balbes, R., & Dwinger, Ph. (1974). Distributive lattices. Columbia: University of Missouri
Press.
2. Birkhoff, G. (1967). Lattice theory (3rd ed., Vol. XXV). Providence: American Mathematical
Society Colloquium Publications, American Mathematical Society.
3. Bourbaki, N. (1966). Elements of mathematics. General topology. Paris, Reading: Hermann,
Addison-Wesley Publishing Co.
4. Cohn, P. M. (1965). Universal algebra. New York: Harper & Row Publishers.
5. Esakia, L. L. (1974). Topological Kripke models. Soviet Mathematics Doklady, 15, 147–151.
6. Freyd, P. (1972). Aspect of topoi. Bulletin of the Australian Mathematical Society, 7, 1–76.
7. Gierz, G., Hofmann, K. H., Keimel, K., Lawson, J. D., Mislove, M. W., & Scott, D. S. (1980).
A compendium of continuous lattices. Berlin: Springer.
8. Goldblatt, R. I. (1979). Topoi: The categorical analysis of logic (Vol. 98). Studies in logic and
the foundations of mathematics. Amsterdam: North-Holland Publishing Co.
9. Grätzer, G. (1968). Universal algebra. Princeton: D. Van Nostrand Co., Inc.
10. Halmos, P. R. (1963). Lectures on Boolean algebras (Vol. 1). Van Nostrand mathematical
studies. Princeton: D. Van Nostrand Co., Inc.
11. Kelley, J. L. (1955). General topology. Toronto: D. Van Nostrand Company, Inc.
12. Mal’cev, A. I. (1973). In D. Smirnov & M. Taı̆clin (Eds.), Algebraic systems (Posthumous
ed.). Translated from the Russian by B. D. Seckler & A. P. Doohovskoy, Die Grundlehren der
mathematischen Wissenschaften (Vol. 192). New York: Springer.
13. Rasiowa, H., & Sikorski, R. (1963). The mathematics of metamathematics (Vol. 41). Monografie
Matematyczne. Warsaw: Państwowe Wydawnictwo Naukowe.
Chapter 2
Heyting Algebras and Closure Algebras

2.1 Heyting Algebras

2.1.1 A bounded distributive lattice (L , ≤) can be considered as a universal algebra

(L , ∧, ∨, 0, 1) with two binary operations ∨, ∧ and two constants 0, 1 satisfying the
following axioms:
(L1) (a ∨ b) ∨ c = a ∨ (b ∨ c), (a ∧ b) ∧ c = a ∧ (b ∧ c),
(L2) a ∨ b = b ∨ a, a ∧ b = b ∧ a,
(L3) a ∨ a = a, a ∧ a = a,
(L4) a ∨ (a ∧ b) = a, a ∧ (a ∨ b) = a,
(L5) a ∧ (b ∨ c) = (a ∧ b) ∨ (a ∧ c), a ∨ (b ∧ c) = (a ∨ b) ∧ (a ∨ c),
(L6) a ∧ 0 = 0, a ∨ 1 = 1.
Let DL denote the variety (and the corresponding equational category) of bounded
distributive lattices.

2.1.2 Definition The center of a bounded distributive lattice L is the set of all
elements that have a complement.1 Let CT(L) denote the center of a lattice L ∈ DL.
The elements of the center are called central.

2.1.3 Proposition The center CT(L) of any lattice L ∈ DL is a bounded sublattice

which is a Boolean lattice.2

2.1.4 Proposition From the point of view of universal algebra (which we will gener-
ally adhere to), Heyting lattices can be defined as algebras (H, ∨, ∧, →, 0, 1) with
three binary operations and two constants satisfying the following axioms:
(H1) (H, ∨, ∧, 0, 1) ∈ DL;
(H2) a ∧ (a → b) = a ∧ b;

1 Editorial note: Recall the definition of a complement from Definition 1.5.6.

2 Editorial note: For a proof, see, e.g., p. 67 of [2].
© Springer Nature Switzerland AG 2019 15
G. Bezhanishvili and W. H. Holliday (eds.), Heyting Algebras,
Trends in Logic 50, https://doi.org/10.1007/978-3-030-12096-2_2
16 2 Heyting Algebras and Closure Algebras

(H3) a ∧ (b → c) = a ∧ ((a ∧ b) → (a ∧ c));

(H4) ((a ∧ b) → a) ∧ c = c.
We denote the resulting variety (and the corresponding category) by HA. The alge-
bras (H, ∨, ∧, →, 0, 1) ∈ HA are called Heyting algebras (or pseudo-Boolean
algebras).
Proof If H is a Heyting lattice, then its lattice operations ∨, ∧ (together with the
greatest element 1 and the least element 0) satisfy Eqs. (L1)–(L6) (i.e., condition
(H1)) by Proposition 1.5.3, and the implication operation → satisfies Eqs. (H2)–
(H4) by Eqs. (6), (10), (11) of Proposition 1.5.7.
Conversely, suppose (H, ∨, ∧, →, 0, 1) is a Heyting algebra. Let a, b ∈ H .
By (H2), a ∧ (a → b) = a ∧ b ≤ b, i.e., a ∧ (a → b) ≤ b. If a ∧ c ≤ b for some
c ∈ H , then by (H3),

c ∧ (a → b) = c ∧ ((c ∧ a) → (c ∧ b)) = c ∧ ((a ∧ b ∧ c) → (b ∧ c)).

Applying (H4) yields

c ∧ ((a ∧ b ∧ c) → (b ∧ c)) = c.

So c ∧ (a → b) = c, i.e., c ≤ a → b. Thus, a → b is the greatest of the elements c

satisfying a ∧ c ≤ b. 
2.1.5 Proposition A Heyting algebra (H, ∨, ∧, →, 0, 1) is a Boolean algebra if
and only if any one of the following conditions holds:
(B1) The law of excluded middle: a ∨ ¬a = 1;
(B2) Double negation law: ¬¬a = a;
(B3) Peirce’s law: (a → b) → a ≤ a;
(B4) a → b ≤ ¬a ∨ b.
Boolean algebras are usually considered as algebras (B, ∨, ∧, −, 0, 1) where −a
denotes the complement of an element a. Clearly if (B, ∨, ∧, −, 0, 1) is a Boolean
algebra, then (B, ∨, ∧, →, 0, 1) is a Heyting algebra, where a → b = −a ∨ b.
Therefore, the equational category BA of Boolean algebras is a full subcategory
of the category HA of Heyting algebras. We will often denote a Boolean algebra
(B, ∨, ∧, −, 0, 1) by B, without any danger of confusion.
2.1.6 The algebra of open sets of a topological space. Let X be a topological
space and  the family of all its open sets. For any A, B ∈ , we have A ∪ B ∈ ,
A ∩ B ∈ , and ∅, X ∈ , so (, ∪, ∩, ∅, X ) is a bounded distributive lattice. For
any A, B ∈ , define A → B = −C(A − B), where A − B = A ∩ −B and C is
the operator of topological closure in the space X .3 We show that → is a relative
pseudocomplement.

3 Editorialnote: The operation → may be equivalently defined by A → B = I(−A ∪ B), where I

is the operation of topological interior in the space X .
2.1 Heyting Algebras 17

The condition A ∩ U ⊆ B is obviously equivalent to the condition

U ⊆ −(A − B), i.e., (A − B) ⊆ −U . Since U is an open set, −U is closed. There-
fore, A − B ⊆ −U is equivalent to C(A − B) ⊆ −U . Thus, U ⊆ −C(A − B) is
equivalent to A ∩ U ⊆ B. So the algebra (, ∪, ∩, → ∅, X ) of open sets of any
topological space X is a Heyting algebra.

2.1.7 Definition An element a of a Heyting lattice H is said to be prime if:

(1) a = 1;
(2) a = b ∧ c implies a = b or a = c.
Note that condition (2) is equivalent to condition

(20 ) b ∧ c ≤ a implies b ≤ a or c ≤ a.
Indeed, suppose (2) holds and b ∧ c ≤ a. Then a = (b ∧ c) ∨ a = (b ∨ a) ∧ (c ∨ a)
and by (2), a = b ∨ a or a = c ∨ a. Therefore, b ≤ a or c ≤ a. Conversely, suppose
(20 ) holds and a = b ∧ c. Then a ≤ b, a ≤ c and b ∧ c ≤ a. Applying (20 ) yields
a = b or a = c.4

2.1.8 Proposition A Heyting algebra H is isomorphic to the lattice  of open sets

of a suitable topological space X if and only if:

(1) H is a complete lattice;

(2) every element a is a greatest lower bound of prime elements of H .

Bibliographic notes. See Chap. IX of [1] or Chap. IV of [19]. For Boolean algebras,
see [6, 20]. For 2.1.6, see [19]. Proposition 2.1.8 appears in [5].

2.2 Closure Algebras

2.2.1 Definition An algebra (B, ∨, ∧, −, 0, 1, C), or (B, C) for short, is said to be

a closure algebra if (B, ∨, ∧, −, 0, 1) is a Boolean algebra and the closure operator
C satisfies Kuratowski’s axioms:
(C1) a ≤ Ca;
(C2) CCa ≤ Ca;
(C3) C(a ∨ b) = Ca ∨ Cb;
(C4) C0 = 0
for all a, b ∈ B.

4 Editorial note: It is now common to call elements of a lattice satisfying (1) and (2) meet irreducible

and elements satisfying (1) and (20 ) meet prime. The dual concepts are join irreducible and join
prime.
18 2 Heyting Algebras and Closure Algebras

The interior I, the dual operator to C, is introduced, as usual, by Ia = −C−a

for each a ∈ B. Obviously Ca = −I−a, so interior algebras (B, ∨, ∧, −, 0, 1, I) are
term equivalent to closure algebras. This is a useful fact.
Let CA denote the class (and the category) of closure algebras. Since every
Boolean algebra can be equipped with the “discrete” closure operator C by set-
ting Ca = a for each a ∈ B, the category BA of Boolean algebras can be considered
as a full subcategory of CA.

2.2.2 Proposition Let (B, ∨, ∧, −, 0, 1, C) ∈ CA. Then:

(1) C1 = 1;
(2) a ≤ b ⇒ Ca ≤ Cb;
(3) I0 = 0 and I1 = 1;
(4) I(a ∧ b) = Ia ∧ Ib;
(5) IIa = Ia;
(6) Ia ≤ a;
(7) a ≤ b ⇒ Ia ≤ Ib.

2.2.3 Definition An element a of a closure algebra (B, ∨, ∧, −, 0, 1, C) is said to

be open (resp. closed) if Ia = a (resp. Ca = a).

It is easy to see that an element a ∈ B is open (resp. closed) if and only if a = Ib

(resp. a = Cb) for some element b ∈ B.
Let H = H (B, C) be the set of all open elements of an algebra (B, C) ∈ CA,
i.e., let H = {Ia : a ∈ B}.

2.2.4 Proposition The algebra (H, ∨, ∧, →, 0, 1) is a Heyting algebra. Moreover:

(1) (H, ∨, ∧, 0, 1) is a sublattice of the Boolean lattice (B, ∨, ∧, 0, 1);
(2) for any a, b ∈ H ,

a → b = −C(a ∧ −b) = I(−a ∨ b).

Proof (1) is obvious. In order to prove (2), it is sufficient to show that

a ∧ d ≤ b ⇔ d ≤ I(−a ∨ b) for any a, b, d ∈ H.

The condition a ∧ d ≤ b is equivalent to d ≤ −a ∨ b. Sine d is an open ele-

ment, d ≤ −a ∨ b implies d = Id ≤ I(−a ∨ b) (by Proposition 2.2.2(7)). Since
I(−a ∨ b) ≤ −a ∨ b (Proposition 2.2.2(6)), from d ≤ I(−a ∨ b) it follows that
d ≤ −a ∨ b. So a ∧ d ≤ b is equivalent to d ≤ I(−a ∨ b). Note that for any ele-
ment a ∈ H , ¬a = (a → 0) = −Ca = I−a. 

So with each algebra (B, C) ∈ CA is associated the Heyting algebra

H = H (B, C) of its open elements, which we call the skeleton of the algebra (B, C).
Moreover, as we will see below, every Heyting algebra is (up to isomorphism) the
skeleton of some closure algebra.
2.2 Closure Algebras 19

The indicated assignment of objects of category HA to objects of category CA can

easily be extended to a (covariant) functor  from the category CA to the category
HA. Indeed, if (B, C), (B , C ) ∈ CA, H , H are their  skeletons, and h : B → B
is a morphism in the category CA, then (h) = h  (the restriction of the map
 H
h to H ) is a morphism h H : H → H in the category HA. Since h is a bounded
lattice homomorphism, h  H preserves lattice operations and constants 0, 1; since h
preserves the operations of complement and interior, we have
  
h  H (a → b) = h(I(−a ∨ b)) = I (−h(a) ∨ h(b)) = h  H (a) → h  H (b).

Obviously if h is injective, then so is h  H . Let h be surjective, and let a ∈ H . Then
h −1 (a ) = ∅. If a ∈ h −1 (a ), then Ia ∈ h −1 (a ). Indeed, h(Ia) = I h(a) = I a = a .
 −1
So h(Ia) = a , and hence h  H (Ia) = a , i.e., h  H (a ) = ∅.
Let J be a set and for any j ∈  J , let (B j , C j ) ∈ CA and let H j be its
 skeleton.
Form the direct product (B, C) = (B j , C j ) : j ∈ J of the family (B j , C j ) :
j ∈ J . It  that the skeleton H of (B, C) coincides with the direct
is easy to check
product H j : j ∈ J , where H j is the skeleton of (B j , C j ). Therefore, the
following simple but useful theorem holds.

2.2.5 Theorem The correspondence  is a functor from the category CA to the

category HA that satisfies:
(1)  preserves injections and surjections;
(2)  commutes with the operation of direct product, i.e.,
    
 (B j , C j ) : j ∈ J = (B j , C j ) : j ∈ J .

2.2.6 Corollary

(a) If K is a subvariety
 of the variety HA, then −1 (K ) = (B, C) ∈ CA :
(B, C) ∈ H is a subvariety of the variety CA.  
(b) If K is a subvariety of the variety CA, then (K ) = (B, C) : (B, C) ∈ K
is a subvariety of the variety HA.

2.2.7 Algebras associated with quasi-orders. With every quasi-ordered set (X, R),
one can associate the closure algebra (Sub X, ∪, ∩, −, ∅, X, R −1 ) and the Heyting
algebra (Con X, ∪, ∩, →, ∅, X ), where Sub X (resp. Con X ) is the family of all
subsets (resp. all cones) of X , operations −, ∪, ∩ are set-theoretic complement, union,
intersection, R −1 is the operation assigning to a set A its R-preimage R −1 (A), and
the operation of relative pseudocomplement (implication) → is defined as follows:
 
(1) A1 → A2 = x ∈ X : R(x) ∩ (A1 − A2 ) = ∅ for any A1 , A2 ∈ Con X ;
Then the pseudocomplement of a cone A is:
 
(2) ¬A = x ∈ X : R(x) ∩ A = ∅ .
20 2 Heyting Algebras and Closure Algebras

2.2.8 Proposition If (X, R) is a quasi-ordered set, then:

(1) the algebra (Sub X, ∪, ∩, −, ∅, X, R −1 ) is a closure algebra;
(2) the algebra (Con X, ∪, ∩, →, ∅, X ) is a Heyting algebra;
(3) Con X is the skeleton of the algebra Sub X .
Proof (1) Obviously (Sub X, ∪, ∩, −, ∅, X ) is a Boolean algebra. We show that
R −1 is a closure operator. Axiom (C1) (see Definition 2.1) that A ⊆ R −1 (A) fol-
lows from the reflexivity of R. The transitivity of R gives R −1 (R −1 (A)) ⊆ R −1 (A)
(Axiom (C2)). Since R −1 (A) = {R −1 (x) : x ∈ A}, we have (C3). Axiom (C4),
that R −1 (∅) = ∅, is obvious.
(2) By Proposition 1.4.6(1), (Con X, ∪, ∩, ∅, X ) is a bounded distributive lat-
tice. It remains to show that → is an operation of relative pseudocomplement.
Let A1 , A2 ∈ Con X . First we show that A1 → A2 is a cone. Let x ∈ A1 → A2
and x Ry. Then A1 ∩ R(x) ⊆ A2 and y ∈ R(x). Since R(y) ⊆ R(x), we have
A1 ∩ R(y) ⊆ A2 , so y ∈ A1 → A2 . Next  we show that A1 ∩ (A1 → A2 ) ⊆ A2 ,
i.e., that A1 ∩ x ∈ X : A1 ∩ R(x) ⊆ A2 ⊆ A2 . Let y ∈ A1 and A1 ∩ R(y) ⊆ A2 .
Since R is reflexive, y ∈ R(y), and hence y ∈ A2 . Lastly, let A1 , A2 , A ∈ Con X
and A1 ∩ A ⊆ A2 . We show that A ⊆ A1 → A2 . Let x ∈ A. We must show that
A1 ∩ R(x) ⊆ A2 . If z ∈ A1 ∩ R(x), then z ∈ A1 and z ∈ R(x). Since x ∈ A and A
is a cone, z ∈ A. Thus, z ∈ A1 ∩ A, and by assumption, z ∈ A2 .
(3) We show that Con X is the skeleton of the algebra Sub X . For any
A ⊆ X , by Proposition 1.4.4, we have A ∈ Con X if and only if for any x ∈ A,
R(x) ⊆ A. Consider the set {x ∈ X : R(x) ⊆ A}. We have {x ∈ X : R(x) ⊆ A} =
x ∈ X : (∀ y)(x Ry ⇒ y ∈ A) = −R −1 (−A). Since R −1 is a closure operator,
−R −1 (−A) = {x ∈ X : R(x) ⊆ A} is the “interior” of the set A. So A ∈ Con X
if and only if −R −1 (−A) = A. Let A1 , A2 ∈ Con X . We show that A1 → A2 =
−R −1 (A1 − A2 ):
 
A1 → A2 = x ∈ X : A1 ∩ R(x) ⊆ A2 = −R −1 (A1 ∩ −A2 ).

This completes the proof. 

2.2.9 Definition Let (X, R) be a quasi-ordered set.
(1) A ring of cones K over
 (X, R) is a ring K of sets, each a cone, such that for any
A1 , A2 ∈ K, the set x ∈ X : R(x) ∩ (A1 − A2 ) = ∅ belongs to K.
(2) A closure field F over (X, R) is a field F of subsets of X such that for any A ∈ F,
the set {x ∈ X : R(x) ∩ A = ∅} belongs to F.
Clearly every ring of cones K over (X, R) forms a Heyting algebra in which the
implication is given by
 
A1 → A2 = x ∈ X : R(x) ∩ (A1 − A2 ) = ∅ .

In Chap. 3 we will prove a representation theorem stating that every Heyting

algebra (resp. closure algebra) is isomorphic to a ring of cones (resp. a closure field)
over a suitable quasi-ordered set.
2.2 Closure Algebras 21

2.2.10 Closure algebras associated with topological spaces. Let X be a

topological space and C the operation of topological closure in this space. Obvi-
ously (Sub X, ∪, ∩, −, ∅, X, C) is a closure algebra. The algebra obtained in this
way will be called the closure algebra over the topological space X . Note that
the Heyting algebra (, ∪, ∩, →, ∅, X ) of open sets in X is the skeleton of
(Sub X, ∪, ∩, −, ∅, X, C) (see 2.1.6).
Bibliographic notes. The most complete presentation of closure algebras is given
in [19], Chap. III; a very brief one is given in Chap. IX of [1]. A proof of Proposition
2.2.2 can be found in [19]. The subject owes its existence to [17]. Theorem 2.2.5 and
its corollary can be found in [10] and in another form in [4, 16], as well as in the
interesting monograph [3].

2.3 Modal Systems and Superintuitionistic Logics

As usual, formulae of intuitionistic propositional calculus are constructed from

propositional variables using connectives ∨, &, →, ¬.
2.3.1 Definition A superintuitionistic logic is a set of formulae containing all the-
orems of the intuitionistic propositional calculus In and closed under the rules of
substitution and modus ponens ( p,  p → q ⇒  q).5 Let  denote the family
of all superintuitionistic logics.
Formulae of the modal propositional system S4 of Lewis are constructed from
propositional variables using the connectives ∨, &, ¬, and ♦. Classical implication
→ and the necessity operator  are defined in a standard way: p → q = ¬ p ∨ q
and  p = ¬♦¬ p.
2.3.2 Definition The modal system S4 is defined as the smallest set of formulae
containing all classical tautologies, all axioms  p → p,  p →  p,
( p → q) → ( p → q), and closed under the rules of substitution, modus
ponens, and Gödel’s rule ( p ⇒   p).6 Any set of formulae containing the axioms
and closed under the rules mentioned above is said to be a normal extension of the
system S4. Let  denote the family of all normal extensions of S4; members of 
will be called modal systems and will be denoted by σ, sometimes with an index.
2.3.3 Let λ be a superintuitionistic logic, i.e., λ ∈ , p a formula, and
(H, ∨, ∧, →, 0, 1) a Heyting algebra. It is known that p can be considered as an alge-
braic polynomial if propositional connectives are understood as the corresponding
algebraic operations of H and propositional variables are understood as variables

5 Editorial note: The intuitionistic propositional calculus is more commonly denoted by ‘Int’ or
‘IPC’. Esakia uses the letters p, q for arbitrary (possibly complex) formulae. He has no special
notation for propositional variables. The rule of substitution states that if  p and q is obtained
from p by uniformly substituting formulae for propositional variables in p, then  q.
6 Editorial note: The term ‘necessitation rule’ is now more common than ‘Gödel’s rule’.
22 2 Heyting Algebras and Closure Algebras

with range H . With each algebra H ∈ HA one can associate the superintuitionistic
logic λ = λ(H ) ∈ , namely the set of the formulae that (when considered as poly-
nomials) are identically equal to 1 ∈ H . The same holds (mutatis mutandis) for modal
systems σ ∈ , so we may adopt the notation σ = σ(B, C), where (B, C) ∈ CA.
Moreover, it is well known (and useful for the algebraic approach) that for any logic
λ ∈  there is a Heyting algebra H ∈ HA (for example, the Lindenbaum–Tarski
algebra7 of λ) such that λ = λ(H ). Similarly, σ ∈  if and only if σ = σ(B, C) for
some closure algebra (B, C) ∈ CA.
If λ = λ(H ), we say that the logic λ is characterized by the algebra H or that H
is an exact, or adequate, algebraic model of λ. If λ ⊆ λ(H ), the algebra H is said
to be an algebraic model of λ ∈ . This terminology can obviously be extended to
classes K ⊆ HA. Namely, the notation λ = λ(K ) means that λ is the set of formulae
that are true (identically equal to 1) in any algebra H ∈ K . In this case we say that
the class K defines (or characterizes) the logic λ = λ(K ). Note also that logics
λ ∈  are in bijective correspondence with varieties K λ of those Heyting algebras
in which all formulae ρ ∈ λ are true. Clearly this terminology is applicable (mutatis
mutandis) to modal systems σ ∈ .
Starting with the fundamental work of McKinsey and Tarski, closure algebras
and Heyting algebras are the main tool in the study of modal systems and superin-
tuitionistic logics. Nowadays the theory of algebras associated with modal systems
(closure algebras, Halmos’ monadic algebras, Grzegorczyk algebras).8 as well as the
theory of algebras associated with superintuitionistic logics (Heyting algebras, Stone
lattices, Boolean algebras) form an important and independent area of research with
its own internal problems and interesting applications.
2.3.4 Kripke models – the semantics of “possible worlds.” In the “post-Kripkean”
period, it became possible to formulate the semantics of a number of superintuitionis-
tic logics and modal systems in terms of Kripke models (X, R),9 where points x ∈ X
of a quasi-ordered set (X, R) are interpreted as “possible worlds,” “situations,” or
“moments of time” and the relation R as the relation of “reachability” (of one world
from another), the “ordering of moments of time,” etc. Under this approach, the
truth values t (truth) and f (falsehood) are assigned not to a sentence p but to a
pair ( p, x) where x ∈ X . In other words, one associates with each sentence p the
set A p ⊆ X of those “situations” in which p is true. For superintuitionistic logics,
A p is assumed to be a cone of the quasi-ordered set (X, R), and for modal systems,
A p is any subset of X . Clearly the statement ‘the modal formula p is true (under all

7 Editorial note: For the definition of the Lindenbaum-Tarski algebra, see, e.g., Sect. VI.10 of [19].
8 Editorial note: Closure algebras provide the algebraic semantics for S4 and its extensions. The
algebraic semantics for arbitrary normal modal logics is based on the more general concept of a
Boolean algebra with operators (see, e.g., P. Blackburn, M. de Rijke, and Y. Venema, Modal Logic
(Cambridge University Press), 2001).
9 Editorial note: The term ‘Kripke frame’ is now standard. In current terminology, a Kripke model

is a Kripke frame together with a valuation function assigning to each propositional variable a subset
of X , which must be a cone in the case of Kripke models for superintuitionistic logics. Note that
the binary relation R in a Kripke frame need not be reflexive and transitive, but these conditions are
required to give a semantics for S4.
2.3 Modal Systems and Superintuitionistic Logics 23

valuations) in the model (X, R)’ is equivalent to the statement ‘the modal formula
p (more precisely, the corresponding polynomial) is identically 1 in the closure
algebra (Sub X, ∪, ∩, −, ∅, X, R −1 ).’ The same holds for formulae of superintu-
itionistic logics.
So Kripke models, when viewed algebraically, are represented by the closure
algebras (Sub X, ∪, ∩, −, ∅, X, R −1 ). It is known that a closure algebra (B, C) is
isomorphic to the closure algebra of the above kind, i.e., (B, ∨, ∧, −, 0, 1, C) 
(Sub X, ∪, ∩, −, ∅, X, R −1 ), if and only if the Boolean algebra (B, ∨, ∧, −, 0, 1)
 and atomic and the closure operator C is completely additive, i.e.,
is complete
C ai = Cai for any set I and ai ∈ B for i ∈ I .
i∈I i∈I
Clearly not all closure algebras satisfy these properties. Recent results show that
there exist the so-called unmodellable modal systems, i.e., systems that have no
adequate Kripke models.10 In other words, the class  ∗ of those modal systems
σ ∈  for which an adequate algebra (B, C) ∈ CA of the above special kind exists
(σ = σ(B, C)) by no means coincides with the class . Moreover, the cardinality of
the difference  −  ∗ is that of continuum. Having eliminated the artificial and non-
heuristic nature of the algebraic approach, the semantics of “possible worlds” (Kripke
models) has however lost the Lindenbaum–Tarksi kind of guarantee: existence of an
adequate model for every system.
A similar situation occurs in the case of superintuitionistic logics. Under the
algebraic approach, Kripke models (X, R) are represented by Heyting algebras of
the special kind (Con X, ∪, ∩, →, ∅, X ).
2.3.5 Definition An element a of a Heyting algebra H is said to be strictly com-
pact if for any set K ⊆ H , if a ≤ {b ∈ H : b ∈ K } then a ≤ b for some b ∈ K .
A Heyting algebra H is said to be strictly compactly generated if every a ∈ H
is equal to the least upper bound of all strictly compact elements b ∈ H such that
b ≤ a.11
2.3.6 Proposition A Heyting algebra (H, ∨, ∧, →, 0, 1) is isomorphic to the alge-
bra Con X of all cones of a suitable quasi-ordered set (X, R) if and only if the lattice
H is complete and strictly compactly generated.
Proof Clearly every lattice of cones Con X is complete (see Proposition 1.4.6(1)).
It is easy to check that a cone A is strictly compact if and only if it is principal, and
every cone is the union of principal cones contained in it (see Proposition 1.4.6(3)).
So the algebra Con X is strictly compactly generated.
Let (H, ∨, ∧, →, 0, 1) be complete and strictly compactly generated, let X be the
set of all strictly compact elements of H , and let a Rb ⇔ b ≤ a for any a, b ∈ H .
Define a map f : H → Con X by f (a) = R(a) ∩ X . It is easy to check that f is an
isomorphism from the Heyting algebra H onto the algebra Con X of cones of the
quasi-ordered set (X, R). 

10 Editorial note: Such systems are now commonly called Kripke incomplete or Kripke frame
incomplete.
11 Editorial note: Strictly compact elements are now usually called completely join prime, and

strictly compactly generated lattices are called completely join prime generated.
24 2 Heyting Algebras and Closure Algebras

Clearly not every Heyting algebra possesses the properties mentioned in

Proposition 2.3.6. The notion of being unmodellable can be applied to superintu-
itionistic logics as well. However, we shall not go deep into this subject. To end
this section, note that S. Thomason has axiomatically defined a very simple modal
system which is not only unmodellable, but has no Kripke models at all!
Bibliographic notes. A fundamental work on the connection between intuitionistic
logic and Lewis’s modal system S4 is [18]. A.V. Kuznetsov and his students obtained
very interesting and deep results on superintuitionistic logics; we only mention the
survey paper [15]. Results connecting modal systems and superintuitionistic logics
can be found in [3, 7–9, 13, 16]. For S. Thomason’s system “without models,”
see [21].

2.4 Filters and Congruences

2.4.1 Let L be a bounded distributive lattice, i.e., let L ∈ DL. A nonempty subset F
of L is said to be a filter in L if a ∧ b ∈ F is equivalent to a ∈ F and b ∈ F. Recall
that F is a filter in L if and only if:
(a) a, b ∈ F ⇒ a ∧ b ∈ F;
(b) a ∈ F and a ≤ b ⇒ b ∈ F.
A filter is said to be principal if it is of the form [a) = {b ∈ L : a ≤ b}, where
a ∈ L. A filter F in L is said to be proper if F = L. A filter F is proper if and only
/ F. A filter F is said to be prime if it is proper and a ∨ b ∈ F implies a ∈ F or
if 0 ∈
b ∈ F. A filter F in L is said to be maximal if F is proper and for any proper filter
F , from F ⊆ F it follows that F = F . In a distributive lattice, every maximal filter
is prime. We recall the following fundamental theorem of M. Stone for distributive
lattices.12
2.4.2 Prime filter theorem. Let L be a distributive lattice, I an ideal in L, and F
a filter in L such that I ∩ F = ∅. Then there exists a prime filter F in L such that
F ⊆ F and I ∩ F = ∅.
2.4.3 Corollary In a distributive lattice L, if a ≤ b, then there is a prime filter F in
L that contains a and does not contain b.
2.4.4 Proposition Let L be a bounded distributive lattice. Then
(1) every proper filter F in L is contained in a maximal filter;
(2) every filter is an intersection of prime filters;
(3) if F ∪ I = L and F ∩ I = ∅, then F is a prime filter if and only if I is a prime
ideal.

12 Editorial
note: See Theorem 6 of M. Stone, “Topological representations of distributive lattices
and Brouwerian logics,” Časopis pro pěstování matematiky a fysiky 67(1): 1–25, 1937.
2.4 Filters and Congruences 25

Let H be a Heyting algebra and F a subset of H .

2.4.5 Proposition F is a filter in H if and only if:
(1) 1 ∈ F;
(2) a ∈ F and a → b ∈ F imply b ∈ F.
2.4.6 Proposition Let F be a filter in H . Then a → b ∈ F if and only if b ∈ [F, a],
where [F, a] is the filter generated by F and a.13
Proof If a → b ∈ F, then since a ∧ (a → b) ≤ b, we have b ∈ [F, a]. If b ∈ [F, a],
then there exists d ∈ F such that a ∧ d ≤ b. Then d ≤ a → b, so d ∈ F implies
a → b ∈ F. 
2.4.7 Proposition Let F be a proper filter in H and ¬a ∈
/ F. Then the filter [F, a]
is proper.
Proof Follows from ¬a = a → 0 and Proposition 2.4.6. 
For a Heyting algebra H , let (H ) denote the lattice of all congruences and
F (H ) the lattice of all filters of H .
2.4.8 Let F ∈ F (H ). Define a binary relation θ(F) on H as follows: for any
a, b ∈ H , (a, b) ∈ θ(F) if and only if there is an element d ∈ F such that a ∧ d =
b ∧ d.14
2.4.9 Proposition Let H ∈ HA and F ∈ F (H ). Then:
(a) θ(F) is a congruence of the Heyting algebra H , i.e., θ(F) ∈ (H );
(b) the map θ : F (H ) → (H ) is an isomorphism of the lattices F (H ) and (H ).
Recall that the inverse map is defined as follows: if θ ∈ (H ), then
F(θ) = {a ∈ H : (a, 1) ∈ θ}.
2.4.10 Corollary Let H, H ∈ HA and h : H → H be a homomorphism. Then h
is injective if and only if h(a) = 1 implies a = 1.
2.4.11 Theorem A Heyting algebra H is subdirectly irreducible (in HA) if and only
if there is an element d ∈ H such that d = 1 and a ≤ d for each a ∈ H such that
a = 1.
Proof Let d ∈ H satisfy the indicated condition. Then the filter F = {d, 1} = [d)
is obviously the smallest of all filters distinct from the unit filter [1). By Proposi-
tion 2.4.9, there is a smallest nontrivial congruence of H . Therefore, by Birkhoff’s
Theorem (Theorem 1.1.1), H is subdirectly irreducible.

13 Editorial note: The filter [F, a] generated by F and a is the smallest filter F such that F ⊆ F
and a ∈ F. An element b belongs to [F, a] iff there is d ∈ F such that a ∧ d ≤ b.
14 Editorial note: Equivalently, (a, b) ∈ θ(F) if and only if a → b, b → a ∈ F (see, e.g., Sect. I.13

of [19]).
26 2 Heyting Algebras and Closure Algebras

Let H be subdirectly irreducible. Then (by Birkhoff’s Theorem and Proposition

2.4.9) there exists a filter F which is the smallest of all filters distinct from [1).
The filter F is principal, i.e., F = [d) for some d ∈ H . Indeed, otherwise for each
a ∈ F, there must be b ∈ F such that a = b and b ≤ a. Therefore, [a) ⊆ [b) ⊆ F
and [a) = F, contradicting the fact that F is the smallest nonunit filter. Now let
a ∈ H and a = 1. Then since F = [d) is the smallest nonunit filter, [d) ⊆ [a) and
so a ≤ d. 
Let (B, C) be a closure algebra, F a filter of the Boolean algebra B, and H the
skeleton of (B, C).
 
2.4.12 Proposition The set F 0 = a ∈ B : (∃ b)(b ∈ F ∩ H & b ≤ a) satisfies:
(1) F 0 is a filter of the Boolean algebra B;
(2) for each a ∈ F 0 , Ia ∈ F 0 ;
(3) F 0 ⊆ F;
(4) F 0 is the greatest of all filters of the Boolean algebra B satisfying (2) and (3).
Proof (1) Since F ∩ H is a filter of H , F ∩ H generates the filter F 0 .
(2) Let a ∈ F 0 . Then b ≤ a for some b ∈ F ∩ H . Clearly Ib = b and since Ib ≤ Ia,
we have Ia ∈ F 0 .
(3) Obvious.
(4) Let F1 be a filter of the Boolean algebra B such that F1 ⊆ F and a ∈ F1
implies Ia ∈ F1 for any a ∈ B. We show that F1 ⊆ F 0 . Let a ∈ F1 . Then Ia ∈
F1 , and since F1 ∩ H ⊆ F ∩ H , we have Ia ∈ F 0 . As Ia ≤ a, we conclude that
a ∈ F 0. 
2.4.13 Definition A filter F of the Boolean algebra B is said to be a skeletal filter
of the algebra (B, C) if it coincides with F 0 , i.e., F = F 0 .15
2.4.14 Proposition Let (B, C) be a closure algebra and F a filter of the Boolean
algebra B. The following conditions are equivalent:
(1) For any a ∈ B, a ∈ F implies Ia ∈ F;
(2) F is a skeletal filter of the algebra (B, C).
Proof (1 ⇒ 2) Suppose (1) holds. To show that F ⊆ F 0 , let a ∈ F. Then by (1),
Ia ∈ F. Therefore, Ia ∈ F ∩ H . Thus, Ia ∈ F 0 , and as Ia ≤ a, we have a ∈ F 0 .
(2 ⇒ 1) Follows directly from Proposition 2.4.12. 
2.4.15 Proposition Let (B, C) be a closure algebra, and let F1 and F2 be skeletal
filters of (B, C). Then
F1 ⊆ F2 ⇔ F1 ∩ H ⊆ F2 ∩ H.

2.4.16 Notation Let F s (B, C) denote the family of all skeletal filters of a closure
algebra (B, C).

15 Editorial note: Skeletal filters are also commonly called ‘I-filters’ or ‘open filters’.
2.4 Filters and Congruences 27

2.4.17 Theorem The family F s (B, C) of all skeletal filters of a closure algebra
(B, C) ordered by inclusion is isomorphic to the family of all filters of the skeleton
H of (B, C).

Proof Using Proposition 2.4.12, it is easy to verify that the following maps are
isotone and injective: the map assigning to each skeletal filter F of (B, C) the filter
F ∩ H of H , and the map assigning to each filter G of H the smallest filter F of the
Boolean algebra B such that F ∩ H = G. Consequently, the congruences of (B, C)
are in bijective correspondence with the skeletal filters. 

2.4.18 Proposition Let (B, C) be a closure algebra, H its skeleton, and F a skeletal
filter. Then:
(1) the filter F ∩ H of the algebra H is prime if and only if for any a, b ∈ B,
Ia ∨ Ib ∈ F implies Ia ∈ F or Ib ∈ F;
(2) the filter F ∩ H is maximal if and only for any a ∈ B, either a ∈ F or C−a ∈ F.

Proof (1) Let F ∩ H be a prime filter and Ia ∨ Ib ∈ F. Suppose that Ia ∈ /F

and Ib ∈ / F. Then Ia ∈ / F ∩ H and Ib ∈ / F ∩ H . Since the filter F ∩ H is prime,
Ia ∨ Ib ∈ / F ∩ H , so Ia ∨ Ib ∈
/ F, a contradiction. Conversely, let a ∨ b ∈ F ∩ H for
a, b ∈ H . Then Ia = a, Ib = b, and Ia ∨ Ib ∈ F. Hence, either Ia ∈ F or Ib ∈ F,
i.e., either Ia = a ∈ F ∩ H or Ib = b ∈ F ∩ H .
(2) Let F ∩ H be a maximal filter in H and a ∈ / F. Then Ia ∈ / F and hence
Ia ∈/ F ∩ H . Since Ia ∈ / F ∩ H , the maximality of F ∩ H implies that Ia ∧ b =
0 for some b ∈ F ∩ H . From Ia ∧ b = 0 it follows that b ≤ −Ia = C−a. Since
b ∈ F, we conclude that C−a ∈ F. Conversely, let F be a filter in H , F ∩ H ⊆ F
and F ∩ H = F . Then there exists an element a ∈ F such that a ∈ / F ∩ H , so
a∈ / F. Therefore, C−a = −Ia ∈ F, and hence I−Ia ∈ F. Thus, I−Ia ∈ F ∩ H , and
so I−Ia ∈ F . Since a = Ia and a ∈ F , we have Ia ∧ I−Ia ∈ F . But Ia ∧ I−Ia =
I(Ia ∧ −Ia) = I0 = 0. Consequently, 0 ∈ F , i.e., F is not a proper filter. This proves
that the filter F ∩ H is maximal. 

It is well known that the lattice of filters of a Boolean algebra is isomorphic to the
lattice of its congruences.

2.4.19 Theorem Let (B, C) be a closure algebra and F a filter of the Boolean
algebra B. The following conditions are equivalent:
(1) F is a skeletal filter;
(2) the congruence θ(F) of the Boolean algebra B corresponding to the filter F is
a congruence of the closure algebra (B, C).

Proof Let θ(F) be the congruence of the Boolean algebra B corresponding to the
filter F, i.e., (a, b) ∈ θ(F) if and only if there exists d ∈ F such that a ∧ d = b ∧ d.
(1 ⇒ 2). Let F be a skeletal filter and (a, b) ∈ θ(F), i.e., a ∧ d = b ∧ d for some
d ∈ F. Then Id ∈ F and Ia ∧ Id = Ib ∧ Id. Therefore, (Ia, Ib) ∈ θ(F). Thus, θ(F)
is a congruence of (B, C).
28 2 Heyting Algebras and Closure Algebras

(2 ⇒ 1). Let θ(F) be a congruence of (B, C), and let a ∈ F, i.e., (a, 1) ∈ θ(F).
Then (Ia, I1) ∈ θ(F). Therefore, Ia ∧ d = I1 ∧ d for some d ∈ F. But Ia ∧ d =
I1 ∧ d = 1 ∧ d = d, so d ≤ Ia. Thus, Ia ∈ F. 
Consequently, the congruences of (B, C) are in bijective correspondence with the
skeletal filters.
2.4.20 Corollary The lattice of congruences (B, C) of a closure algebra (B, C)
is isomorphic to the lattice of congruences of the skeleton of (B, C).
2.4.21 Corollary A closure algebra (B, C) is subdirectly irreducible (in CA) if and
only if its skeleton is subdirectly irreducible (in HA).
2.4.22 Corollary The variety of Heyting algebras HA and the variety of closure
algebras CA are congruence-distributive and have the congruence extension prop-
erty.
Bibliographic notes. For filters and congruences of Heyting algebras, see Chap. IX
of [1] and Chap. IV of [19]. Theorem 2.4.2 is due to M. Stone. For congruences of
closure algebras, see Chap. III of [19]. The connection between Heyting algebras
and closure algebras has been investigated independently by a number of authors;
see [3, 4, 12].

2.5 Skeletal Closure Algebras

2.5.1 Definition Let (B, ∨, ∧, 0, 1) be a Boolean lattice and (H, ∨, ∧, 0, 1) a

bounded sublattice. The lattice (H, ∨, ∧, 0, 1) is said to be a relatively complete
sublattice of the Boolean lattice B if for any a ∈ B, the set {b ∈ H : b ≤ a} has a
greatest element (in H ).16
2.5.2 Proposition Relatively complete sublattices H of a Boolean lattice B are in
bijective correspondence with closure operators C on B.
Proof Let C be a closure operator on a Boolean algebra B. Consider the set H =
{Ia : a ∈ B} (i.e., the skeleton of the algebra (B, C)). It is easy to check that H is a
relatively complete sublattice of B, where for a ∈ B, the greatest element of the set
{b ∈ H : b ≤ a} is Ia = −C−a ∈ H . Indeed, if a ∈ B, then clearly a ≥ Ia ∈ H ;
and if b ≤ a and b ∈ H , then b = Ib ≤ Ia.
Conversely, let H be a relatively complete sublattice of a Boolean lattice B.
Define a map I : B → B by setting Ia to be the greatest element of the set
{b ∈ H : b ≤ a}. It is easy to check that I is an interior operator: I1 = 1 since 1 ∈ H ;
it is immediate from the definition of Ia that Ia ≤ a; Ia ≤ IIa follows from transi-
tivity of ≤; and since H is a sublattice of B, I(a ∧ b) = Ia ∧ Ib. So Ca = −I−a is
a closure operator on the Boolean algebra B and hence (B, C) ∈ CA. It is easy to
show that this correspondence is bijective. 

16 Editorial
note: For H to be a relatively complete sublattice of B is equivalent to the inclusion of
H into B having a right adjoint.
2.5 Skeletal Closure Algebras 29

2.5.3 Corollary Let (H, ∨, ∧, 0, 1) be a relatively complete sublattice of a Boolean

lattice (B, ∨, ∧, 0, 1). Then for each a, b ∈ H , there exists the implication a → b ∈
H , i.e., (H, ∨, ∧, →, 0, 1) is a Heyting algebra.
Proof Let a, b ∈ H and d = −a ∨ b ∈ B. Since H is relatively complete, the set
{e ∈ H : e ≤ d} has a greatest element g. It is easy to check that g = a → b. 

2.5.4 Corollary Let H be a relatively complete sublattice of B ∈ BA. Then H is

the skeleton of the closure algebra (B, C), where Ca is the least element of the set
{−b : b ∈ H and a ∧ b = 0}.
Proof This follows from Ca = −I−a and the fact that Ia is the greatest element of
{b ∈ H : b ≤ a}. 
2.5.5 Remark Let (H, ∨, ∧, →, 0, 1) be a Heyting algebra. The class of closure
algebras whose skeleton “coincides with” (is isomorphic to) H contains a “privi-
leged” closure algebra (B(H ), C H ) whose Boolean reduct is generated, as a Boolean
algebra, by its skeleton H . The algebras (B(H ), C H ), where H ∈ HA, are said to be
skeletal algebras, i.e., generated from their skeletons H .

2.5.6 Definition Let (B, C) be a closure algebra and H its skeleton. The algebra
(B, C) is said to be a skeletal algebra if the smallest Boolean subalgebra B (of B)
containing H coincides with B.
The construction of (B(H ), C H ) from a given skeleton H is well known. We
recall one of the methods of the construction.
2.5.7 Construction We start with a more general situation. Let H ∈ DL. It is known
that there exists a Boolean algebra (the Boolean envelope) B(H ) containing H as
a sublattice and generated by H as a Boolean algebra.17 The algebra B(H ) can be
constructed as follows. Let X be the set of all prime filters of the lattice H . Define a
map ϕ : H → Sub X , where Sub X is the Boolean algebra (the field) of all subsets
of X , by setting ϕ(a) = { p ∈ X : a ∈ p} for a ∈ H . Then ϕ is an injective lattice
homomorphism from H into Sub X . For simplicity, identify the lattice H with its ϕ-
image in Sub X , i.e., with the ring of sets {ϕ(a) : a ∈ H }. Let B(H ) be the smallest
Boolean subalgebra of Sub X that contains H , i.e., let B(H ) be the field of sets
generated by the ring H . Clearly the smallest and greatest elements 0, 1 of the lattice
H and of B(H ) are the same (and are equal to ∅ and X , respectively) and B(H ) is
determined up to an isomorphism that is the identity on H .
Now let the lattice H be a Heyting algebra, i.e., let H ∈ HA. Then H is a rel-
atively complete sublattice of its envelope B(H ). Indeed, any element a ∈ B(H )
can be represented as a = {−ak ∨ bk : k < n}, where ak , bk ∈ H .18 The great-
est element (in H ) of the set {b ∈ H : b ≤ a} is equal to {ak → bk : k < n},

17 Editorial note: This Boolean algebra is also called the free Boolean extension of H (see, e.g.,
Sect. V.4 of [1]).
18 Editorial note: For a proof of this fact, see, e.g., Sect. IV.3 of [19].
30 2 Heyting Algebras and Closure Algebras

where → is the operation of relative pseudocomplement in the Heyting algebra

H . Relative completeness of the lattice H allows us (see Proposition 2.5.2) to equip
the algebra B(H ) with the closure operator C H , and the skeleton of the resulting
closure algebra (B(H ), C H ) coincides with H . So we have a map  : HA → CA
that assigns to an algebra H ∈ HA a skeletal algebra (H ) = (B(H ), C H ), which
is constructed canonically from the skeleton H .

Let SA denote the class (and the corresponding category) of skeletal closure
algebras. Clearly the category SA is a full subcategory of CA.

2.5.8 Proposition The map  can be extended to a functor from the category HA to
the category SA of skeletal closure algebras.

Proof Let h : H1 → H2 be a Heyting homomorphism, where (Hi ) = (B(Hi ), C Hi )

(i = 1, 2). Extend the map h to B(H1 ) as follows: if a ∈ B(H1 ) and
a= (−ak ∨ bk ), where m ∈ ω and ak , bk ∈ H1 , let h(a) = −h(ak ) ∨ h(bk ) .
k<m k<m
Then h : B(H1 ) → B(H2 ) is a well-defined Boolean homomorphism.19 Let I1 a =
−C1 −a. We have
  

h(I1 a) = h I1 (−ak ∨ bk )
k<m
   
 
=h (ak → bk ) = h (ak → bk )
k<m k<m
 
 
= (h(ak ) → h(bk )) = I2 (−h(ak ) ∨ h(bk )) = I2 h(a).
k<m k<m

Thus, h = (h) is a morphism in the category SA.20 

2.5.9 Proposition The functor  has the following properties:

(1)  preserves injections and surjections;
A closure algebra ( {H j : j ∈ J }) can be embedded into the closure algebra
(2) 
{ H j : j ∈ J };  
(3) If J is a finite set, ( {H j : j ∈ J }) is isomorphic to { H j : j ∈ J }.

Let (K ) = {(H ) : H ∈ K }, where K is a class of algebras contained in HA.

19 Editorial note: To see that this is well-defined, we use the fact (see, e.g., Lemma 1 on p. 97 of
[1]) that h extends uniquely to a Boolean homomorphism
 h from B(H1 ) to B(H2 ), which implies
that h( (−ak ∨ bk )) = −h(ak ) ∨ h(bk ) .
k<m k<m
20 Editorial note: The verification that  preserves the identity and composition is straightforward.
2.5 Skeletal Closure Algebras 31

2.5.10 Corollary
(1) (HK ) = H( K );
(2) (SK ) = S( K );
(3) (H1 × H2 )  (H1 ) × (H2 ), where H1 × H2 is the direct product of the
Heyting algebras H1 and H2 .

2.5.11 Theorem The functors  and  establish the equivalence of the category
HA of Heyting algebras and the category SA of skeletal algebras.

Proof One can use Proposition 2.5.8 to verify the conditions for equivalence from
Definition 1.2.2. However, we prefer to give a different proof based on the Duality
Theorem (see below, Theorems 3.3.4 and 3.4.6). 

We now turn to a classification of the elements in a closure algebra.

2.5.12 Definition Let (B, C) be a closure algebra and a ∈ B. Set:

(1) ρa = a ∧ C(Ca − a);
(2) πa = a − ρa.

2.5.13 Definition Let (B, C) be a closure algebra. For a natural number k ∈ ω and
a ∈ B, define:
(1) ρ0 a = a, ρ1 a = ρa;
(2) ρk+1 a = ρρk a;
(3) π k a = a − ρk a.

2.5.14 Lemma For k ∈ ω, we have ρk a ≥ ρk+1 a, ρk a ≤ a, and π k a ≤ a.

Proof Easy verification. 

Using the operator ρ, one can classify elements of closure algebras as follows.

2.5.15 Definition Let (B, C) be a closure algebra and a ∈ B.

(1) We say that a has finite rank if there exists m ∈ ω such that m = 0 and ρm a = 0.
In this case the rank of a is the smallest natural number k = 0 such that ρk a = 0.
(2) We call a cyclic if a = 0 and ρk a = ρk+1 a = 0 for some k ∈ ω.
(3) We say that a has infinite rank if a is not cyclic and does not have finite rank.

2.5.16 Proposition
(1) If (B, C) is a closure algebra and a ∈ B, then a has one (and only one) of the
three properties given in Definition 2.5.15. Moreover, if a = 0 or a = 1, then the
rank of a is equal to 1, i.e., ρa = 0.
(2) There exists a closure algebra (B, C) which has a cyclic element a ∈ B.
(3) There exists a closure algebra (B, C) in which for any k ≤ ω, there is an element
of rank k.
(4) There is a closure algebra (B, C) which has an element a of infinite rank.
32 2 Heyting Algebras and Closure Algebras

Proof (1) Obvious.

(2) Let B = {0, a, b, 1} be the four-element Boolean algebra and C be the closure
operator on B defined as follows: C0 = 0 and Cd = 1 if d = 0. It is easy to see that
b is cyclic.
(3) and (4). Let X be the set of all natural numbers, i.e., X = {0, 1, 2, . . .}, and let
R be the dual of the natural order on X , i.e., n Rm ⇔ n ≥ m. Let B0 = Sub0 X be
the field of all finite subsets of X and their complements. Let d be the set of all odd
numbers. Finally, let B = B(X ) be the smallest field of subsets of X containing d
and all sets of the field B0 , i.e., the smallest field generated by the family B0 ∪ {d}.
It is easy to check that elements of B can be represented as (a ∩ d) ∪ (b ∩ −d) for
suitable a, b ∈ B0 and that (B, R −1 ) is a closure algebra. Let am = {k : k < m},
where m is an even number and m = 0, and let bm = am ∩ d. A simple calculation
shows that:
(a) bm ∈ B0 ⊆ B, and the rank of bm is m
2
;
(b) d ∈ B, and d has infinite rank. 

The following useful lemma holds.

2.5.17 Lemma Let (B, C) be a closure algebra, a ∈ B, and m ∈ ω. Then

a= (ρk a − ρk+1 a) ∨ ρm a.
k<m

Proof By induction on m. Let m = 0. Then a = ρ0 a. Let


a= (ρk a − ρk+1 a) ∨ ρm a.
k<m

We show that a = (ρk a − ρk+1 a) ∨ ρm+1 a. Clearly

k<m+1


(ρk a − ρk+1 a)
k<m+1

= (ρk a − ρk+1 a) ∨ (ρm a − ρm+1 a)
k<m

= (ρk a − ρk+1 a) ∨ (ρm a ∧ −ρm+1 a)
k<m
   
 
= (ρ a − ρ
k k+1
a) ∨ ρ a ∧
m
(ρ a − ρ
k k+1
a) ∨ −ρ
m+1
a
k<m k<m
 

=a∧ (ρ a − ρ
k k+1
a) ∨ −ρ m+1
a .
k<m

Therefore,
2.5 Skeletal Closure Algebras 33

(ρk a − ρk+1 a) ∨ ρm+1 a
k<m+1
  

= a∧ (ρ a − ρ
k k+1
a) ∨ −ρ m+1
a ∨ ρm+1 a
k<m
 

= (a ∨ ρm+1 a) ∧ (ρk a − ρk+1 a) ∨ −ρm+1 a ∨ ρm+1 a
k<m
= (a ∨ ρm+1 a) ∧ 1
= a ∨ ρm+1 a
=a

because, by Lemma 2.5.14, ρm+1 a ≤ a. 

2.5.18 Lemma Let (B, C) ∈ CA and a ∈ B. Then

ρk a − ρk+1 a = Cρk a − C(Cρk a − ρk a).

Proof Let b = ρk a. We show that b − ρb = Cb − C(Cb − b), i.e., that

(1) b − (b ∧ C(Cb − b)) = Cb − C(Cb − b).

Since
 
b ∧ −(b ∧ C(Cb − b)) = b ∧ −b ∨ −C(Cb − b)
 
= (b ∧ −b) ∨ b ∧ −C(Cb − b)
= b ∧ −C(Cb − b),

(1) is equivalent to
(2) b − C(Cb − b) = Cb − C(Cb − b).
Since b ≤ Cb, we clearly have b − C(Cb − b) ≤ Cb − C(Cb − b). It remains to
show that
(3) Cb − C(Cb − b) ≤ b − C(Cb − b).
From Cb − b ≤ C(Cb − b) it follows that
(4) Cb − C(Cb − b) ≤ b.
Moreover, since (Cb − C(Cb − b)) ∧ C(Cb − b) = 0, we have
(5) Cb − C(Cb − b) ≤ −C(Cb − b).
From (4) and (5) we have

Cb − C(Cb − b) ≤ b ∧ −C(Cb − b) = b − C(Cb − b). 

34 2 Heyting Algebras and Closure Algebras

2.5.19 Lemma Let (B, C) be a closure algebra and a ∈ B. Then


a − ρm a = (ρk a − ρk+1 a) − ρm a.
k<m

Proof By Lemma 2.5.17, a = (ρk a − ρk+1 a) ∨ ρm a. Therefore,

k<m

 

a ∧ −ρ a =
m
(ρ a − ρ
k k+1
a) ∨ ρ a ∧ −ρm a
m

k<m
 

= (ρ a − ρ
k k+1
a) ∧ −ρ a ∨ (ρm a ∧ −ρm a)
m

k<m

= (ρk a − ρk+1 a) − ρm a. 
k<m

2.5.20 Theorem Let (B, C) be a closure algebra and a an element of B of rank

m ∈ ω. Then 
a= (ρk a − ρk+1 a).
k<m

Proof Since the rank of a is equal to m, ρm a = 0. Now apply Lemma 2.5.17. 

2.5.21 Corollary Under the conditions of Theorem 2.5.20,

 
a= Cρk a − C(Cρk a − ρk a) .
k<m

Proof Apply Lemma 2.5.18. 

2.5.22 Lemma Let (B, C) ∈ CA and a ∈ B. Then:

(1) Cρk a ≥ C(Cρk a − ρk a) (k < m);
(2) C(Cρk a − ρk a) ≥ Cρk+1 a (k < m − 1).

Proof (1) Since Cρk a − ρk a ≤ Cρk a,

C(Cρk a − ρk a) ≤ Cρk a.

(2) Since ρk a ∧ C(Cρk a − ρk a) ≤ C(Cρk a − ρk a),

 
C ρk a ∧ C(Cρk a − ρk a) ≤ C(Cρk a − ρk a).
2.5 Skeletal Closure Algebras 35

As
ρk a ∧ C(Cρk a − ρk a) = ρρk a = ρk+1 a,

we obtain Cρk+1 a ≤ C(Cρk a − ρk a). 

2.5.23 Remark For any element a ∈ B of order m, there is a chain of elements
ρ0 a ≥ ρ1 a ≥ · · · ≥ ρk a ≥ · · · ≥ ρm−1 a such that a= (ρk a − ρk+1 a)
k<m
(Theorem 2.5.20). By Corollary 2.5.21 and Lemma 2.5.22, there is a chain a0 ≥
b0 ≥ a1 ≥ b1 ≥ · · · ≥ am−1 ≥ bm−1 consisting of closed elements of (B, C) such
that a = (ak − bk ). If H is the skeleton of (B, C), then since −ak , −bk ∈ H , it
k<m
is clear that the element a belongs to the skeletal subalgebra (B(H ), C H ) of (B, C).
2.5.24 Lemma Let (B, C) ∈ CA and for each k < m ∈ ω, let ak , bk be closed ele-
ments of (B, C) such that a = (ak − bk ). Then there are closed elements ck , dk
k<m
of (B, C) such that a = (ck − dk ) and ck−1 ≥ dk−1 ≥ ck ≥ dk .21
k<m

Proof By replacing bk by ak ∧ bk , we may assume that ak ≥ bk for each k. For each

t with 0 ≤ t < m, set cm = 0 and

ct = {ak0 ∧ · · · ∧ akt | 0 ≤ k0 < k1 < · · · < kt < m},

et = {bk0 ∧ · · · ∧ bkt | 0 ≤ k0 < k1 < · · · < kt < m},
dt = ct+1 ∨ et .

Then c0 ≥ d0 ≥ c1 ≥ d1 ≥ · · · ≥ cm−1 ≥ dm−1 ≥ cm = 0 and each ct and dt is

closed. We claim that:  
(ak − bk ) = (ck − dk ).
k<m k<m

We first prove the inequality k<m (ak − bk ) ≤ k<m (ck − dk ). Since

c0 = a0 ∨ · · · ∨ am−1 , the inequalities between the ck , dk show that

(ak − bk ) ≤ c0 = (c0 − d0 ) ∨ (d0 − c1 ) ∨ · · · ∨ (cm−1 − dm−1 ) ∨ (dm−1 − cm ).
k<m

Consequently, it suffices to show that (as − bs ) ∧ (dt − ct+1 ) = 0 for each s, t. This
in turn is equivalent to as ∧ dt ≤ bs ∨ ct+1 . Since dt = ct+1 ∨ et , this amounts to
proving as ∧ et ≤ bs ∨ ct+1 . For this, by the definition of et , it is sufficient to show
that as ∧ bk0 ∧ · · · ∧ bkt ≤ bs ∨ ct+1 for each k0 < · · · < kt . If s ∈
/ {k0 , . . . , kt }, then
as ∧ bk0 ∧ · · · ∧ bkt ≤ as ∧ ak0 ∧ · · · ∧ akt ≤ ct+1 . If s ∈ {k0 , . . . , kt }, then
as ∧ bk0 ∧ · · · ∧ bkt ≤ bs ≤ bs ∨ ct+1 . This completes the proof of the inequality.

21 Editorial
note: Starting with this lemma, we have made edits to Esakia’s proofs in the rest of this
chapter as mentioned in the Editors’ Note.
36 2 Heyting Algebras and Closure Algebras

We prove the reverse inequality by induction. In the base case, c0 = a0 and d0 =

b0 , so the result is clear. Suppose that m > 1 and that the result holds for m − 1. Set
cm−1 = 0 and for 1 ≤ t < m − 1,

ct = {ak0 ∧ · · · ∧ akt | 0 ≤ k0 < k1 < · · · < kt < m − 1},

et = {bk0 ∧ · · · ∧ bkt | 0 ≤ k0 < k1 < · · · < kt < m − 1},
dt = ct+1 ∨ et .

By induction, we have:
 
(ak − bk ) ≥ (ck − dk ).
k<m−1 k<m−1

Also,
ct = ct ∨ (ct−1 ∧ am−1 ) and et = et ∨ (et−1 ∧ bm−1 ).

Therefore,
dt = ct+1 ∨ (ct ∧ am−1 ) ∨ et ∨ (et−1 ∧ bm−1 ).

From this we get:

dt = dt ∨ (ct ∧ am−1 ) ∨ (et−1 ∧ bm−1 ).

Now,

cm−1 − dm−1 = (a0 ∧ · · · ∧ am−1 ) − (b0 ∧ · · · ∧ bm−1 ) ≤ (ak − bk ).
k<m

Similarly,

c0 − d0 = (a0 ∨ · · · ∨ am−1 ) − (c1 ∨ (b0 ∨ · · · ∨ bm−1 ))

≤ (a0 ∨ · · · ∨ am−1 ) − (b0 ∨ · · · ∨ bm−1 )

≤ (ak − bk ).
k<m

If 1 ≤ t < m − 1, then

ct − dt = (ct ∨ (ct−1 ∧ am−1 )) ∧ ¬dt = (ct ∧ ¬dt ) ∨ (ct−1 ∧ am−1 ∧ ¬dt ).

Since dt ≤ dt , we have ct ∧ ¬dt ≤ ct ∧ ¬dt ≤ k<m−1 (ak − bk ), where the last

inequality uses the induction hypothesis. For the other term, we have:
2.5 Skeletal Closure Algebras 37

ct−1 ∧ am−1 ∧ ¬dt = ct−1 ∧ am−1 ∧ ¬dt ∧ (¬ct ∨ ¬am−1 ) ∧ (¬et−1 ∨ ¬bm−1 )
= ct−1 ∧ am−1 ∧ ¬dt ∧ ¬ct ∧ (¬et−1 ∨ ¬bm−1 )
= [ct−1 ∧ am−1 ∧ ¬dt ∧ ¬ct ∧ ¬et−1 ] ∨
[ct−1 ∧ am−1 ∧ ¬dt ∧ ¬ct ∧ ¬bm−1 ].

In the final expression above, the second term is below am−1 ∧ ¬bm−1 =
am−1 − bm−1 . It remains to consider the first term. It is equal to

ct−1 ∧ am−1 ∧ ¬dt ∧ ¬(ct ∨ et−1 ) = ct−1 ∧ am−1 ∧ ¬dt ∧ ¬dt−1


≤ (ct−1 − dt−1 ) ≤ (ak − bk ),
k<m−1

where the last inequality uses the induction hypothesis. Thus,

 
(ck − dk ) ≤ (ak − bk ).
k<m k<m

Consequently, by induction, the reverse inequality holds. The lemma is therefore

proven. 

2.5.25 Lemma Let (B, C) ∈ CA and for each k < m ∈ ω, let ck , dk be closed ele-
ments of (B, C) such that a = (ck − dk ) and ck−1 ≥ dk−1 ≥ ck ≥ dk . Then there
k<m
are closed elements ak , bk of (B, C) such that ρa = (ak − bk ).
1≤k<m

Proof By definition, ρa = a ∧ C(Ca − a). For simplicity let ek denote

ck − dk (k < m). We have:
 

a ∧ C(Ca − a) = ek ∧ C(Ca − a)
0≤k<m
 

= e0 ∨ ek ∧ C(Ca − a)
1≤k<m
 
  
= e0 ∧ C(Ca − a) ∨ ek ∧ C(Ca − a)
1≤k<m

Since c0 ≥ c1 ≥ · · · ≥ cm−1 , we have a = (ck − dk ) ≤ ck = c0 and hence

k<m k<m
Ca ≤ c0 because c0 is closed. From c0 − d0 ≤ a we obtain −a ≤ −c0 ∨ d0 , which
with c0 ≥ d0 implies Ca − a ≤ c0 ∧ (−c0 ∨ d0 ) = c0 ∧ d0 = d0 . Then since d0 is
closed, we have C(Ca − a) ≤ d0 , which implies e0 ∧ C(Ca − a) = 0.
Consequently,
38 2 Heyting Algebras and Closure Algebras
 

a ∧ C(Ca − a) = ek ∧ C(Ca − a)
1≤k<m
  
= ek ∧ C(Ca − a)
1≤k<m
  
= (ck − dk ) ∧ C(Ca − a)
1≤k<m
  
= (ck ∧ C(Ca − a)) − dk .
1≤k<m

Taking ak to be ck ∧ C(Ca − a) and bk to be dk completes the proof. 

2.5.26 Lemma If a and b are closed elements of (B, C) ∈ CA, then ρ(a − b) = 0.
Proof We have:
 
ρ(a − b) = (a − b) ∧ C C(a − b) − (a − b)
 
= (a − b) ∧ C C(a − b) ∧ −(a ∧ −b)
 
= (a − b) ∧ C C(a − b) ∧ (−a ∨ b)
 
= (a − b) ∧ C (C(a − b) ∧ −a) ∨ (C(a − b) ∧ b)
 
= (a − b) ∧ C(C(a − b) ∧ −a) ∨ C(C(a − b) ∧ b)
   
= (a − b) ∧ C(C(a − b) ∧ −a) ∨ (a − b) ∧ C(C(a − b) ∧ b) .

Since a ∧ −b ≤ a, we have C(a ∧ −b) ≤ Ca = a. From this it follows that

C(a ∧ −b) ∧ −a = 0, and hence (a − b) ∧ C(C(a − b) ∧ −a) = 0.
Since C(a − b) ∧ b ≤ b, we have C(C(a − b) ∧ b) ≤ Cb = b. Thus,
C(C(a − b) ∧ b) ∧ −b = 0. Consequently, (a − b) ∧ C(C(a − b) ∧ b) = 0. 
2.5.27 Proposition Let (B, C) ∈ CA and a, ak , bk ∈ B for k < m ∈ ω such that
ak , bk are closed and a = (ak − bk ). Then ρm a = 0.
k<m

Proof By repeated application of Lemmas 2.5.24 and 2.5.25, ρm−1 a = c − d for

some closed elements c, d ∈ B. Therefore, ρm a = ρρm−1 a = ρ(c − d) = 0 by
Lemma 2.5.26. 
2.5.28 Theorem Let (B, C) be a closure algebra, H its skeleton, and B(H ) the
Boolean envelope of the lattice H , i.e., the smallest Boolean subalgebra of B con-
taining H . Then B(H ) is closed under the closure operator C, i.e., (B(H ), C H ) is a
subalgebra of (B, C).
Proof It is known that each element a ∈ B(H ) can be represented as a =
(ak − bk ), where m ∈ ω and −ak , −bk ∈ H for k < m.22 Since −bk → −ak ∈ H
k<m

22 Editorial note: See, e.g., Sect. IV.3 of [19].

2.5 Skeletal Closure Algebras 39

for k < m, we have (−bk → −ak ) ∈ H . Moreover, using Proposition 2.2.4(2),

k<m

 
(−bk → −ak ) = I(bk ∨ −ak )
k<m k<m

= I−(−bk ∧ ak )
k<m

=− C(ak − bk )
k<m
= −Ca.

Therefore, −Ca ∈ H . Since −1 ∈ H , we have Ca = 1 − (−Ca). Thus,

Ca ∈ B(H ). 

2.5.29 Remark As a consequence of Lemma 2.5.24, without loss of generality,

each a ∈ B(H ) can be represented as a = (ak − bk ), where −ak , −bk ∈ H and
k<m
ak−1 ≥ bk−1 ≥ ak ≥ bk for all k < m.

A convenient characterization of skeletal algebras is given by the following.

2.5.30 Theorem
(1) Let (B, C) be a closure algebra and (B(H ), C H ) its skeletal subalgebra. Then
 
B(H ) = a ∈ B : (∃ m ∈ ω)(ρm a = 0) .

(2) A closure algebra (B, C) is a skeletal algebra if and only if each element a ∈ B
has finite rank.

Proof For part (1), if a ∈ B(H ), then for some m ∈ ω we have a = (ak − bk )
k<m
where −ak , −bk ∈ H and so ak and bk are closed. Then by Proposition 2.5.27,
ρm a = 0. Conversely,
 k if ρm a = 0, then
 a is of rank m, so by Corollary 2.5.21,
a = k<m Cρ a − C(Cρk a − ρk a) . Since −Cρk a, −C(Cρk a − ρk a) ∈ H , we
conclude that a ∈ B(H ).
Part (2) follows from part (1) by the definition of skeletal algebras. 

Bibliographic notes. The notion of a relatively complete subalgebra of a Boolean

algebra was introduced by P. Halmos in [14]. Construction 2.5.7 is due to M. Stone
and G. Birkhoff. Results 2.5.8, 2.5.9, 2.5.11 were obtained by the author [9, 12] and
are based on [10] (see also [11]). The ρ-operator (Definition 2.5.12), classification
of elements (Definition 2.5.15), and the results related to these notions are due to the
author and are published here for the first time.
40 2 Heyting Algebras and Closure Algebras

References

1. Balbes, R., & Dwinger, Ph. (1974). Distributive lattices. Columbia: University of Missouri
Press.
2. Birkhoff, G. (1967). Lattice theory (3rd ed., Vol. XXV). Providence: American Mathematical
Society Colloquium Publications, American Mathematical Society.
3. Blok, W. J. (1976). Varieties of interior algebras. Ph.D. thesis, University of Amsterdam.
4. Blok, W. J., & Dwinger, Ph. (1975). Equational classes of closure algebras. I. Indagationes
Mathematicae, 37, 189–198.
5. Dowker, C. H., & Papert, D. (1966). Quotient frames and subspaces. Proceedings of the London
Mathematical Society, 3(16), 275–296.
6. Dwinger, Ph. (1961). Introduction to Boolean algebras (Vol. 40). Hamburger Mathematische
Einzelschriften. Würzburg: Physica-Verlag.
7. Esakia, L. L. (1976). On modal “companions” of superintuitionistic logics (Russian). In VII
Soviet Symposium on Logic (Kiev, 1976) (pp. 135–136).
8. Esakia, L. L. (1976). On the Dummett-Lemmon conjecture (Russian). In IV Soviet Conference
In Mathematical Logic (Chisinau, 1976) (p. 160).
9. Esakia, L. L. (1979). On the theory of modal and superintuitionistic systems (Russian). Logical
inference (pp. 147–172). Moscow: Nauka.
10. Esakia, L. L. (1974). Topological Kripke models. Soviet Mathematics Doklady, 15, 147–151.
11. Esakia, L. L. (1981). Diagonal constructions, Löb’s formula, and Cantor’s scattered spaces
(Russian). In Studies in Logic and Semantics, Metsniereba, Tbilisi (pp. 128–143).
12. Esakia, L. L. (1984). On the variety of Grzegorczyk algebras. Sel. Sov. Math., 3, 343–366.
13. Esakia, L. L., & Meskhi, S. (1977). Five critical modal systems. Theoria, 43, 52–60.
14. Halmos, P. R. (1962). Algebraic logic. New York: Chelsea Publishing Co.
15. Kuznetsov, A. V. (1971). On superintuitionistic logics (Russian). Mathematical Investigations,
6, 75–122.
16. Maksimova, L. L., & Rybakov, V. V. (1974). The lattice of normal modal logics. Algebra and
Logic, 13, 105–122.
17. McKinsey, J. C. C., & Tarski, A. (1944). The algebra of topology. Annals of Mathematics,
2(45), 141–191.
18. McKinsey, J. C. C., & Tarski, A. (1948). Some theorems about the sentential calculi of Lewis
and Heyting. Journal of Symbolic Logic, 13, 1–15.
19. Rasiowa, H., & Sikorski, R. (1963). The mathematics of metamathematics (Vol. 41). Monografie
Matematyczne. Warsaw: Państwowe Wydawnictwo Naukowe.
20. Sikorski, R. (1960). Boolean algebras (Vol. 25). Ergebnisse der Mathematik und ihrer Gren-
zgebiete, N. F. Berlin: Springer.
21. Thomason, S. K. (1972). Semantic analysis of tense logics. Journal of Symbolic Logic, 37,
150–158.
Chapter 3
Duality Theory: Hybrids

3.1 The Hybrid of Topology (Stone) and Order (Kripke)

Let X be a Stone space. Clearly each binary relation R on X for which the sets
R(x) = {y ∈ X : x Ry}, for each x ∈ X , are nonempty and closed gives rise to (or
can be thought of as) a map ρ : X → exp X from the space X to the space exp X of
nonempty closed subsets of X equipped with the exponential topology.1 It is natural
to regard R as continuous if the corresponding map ρ : X → exp X is continuous.
Let the set X be equipped with a quasi-order R and a topology  making X a
compact Hausdorff zero-dimensional space. The resulting triple (X, , R) is there-
fore a “mixture” of a Kripke model (X, R), i.e., a set of “worlds” X equipped with
a relation of “reachability” R, and a Stone space (X, ). We adopt the following
fundamental definition.
3.1.1 Definition A triple (X, , R) is said to be a hybrid if:
(1) (X, ) is a Stone space;
(2) (X, R) is a quasi-order;
(3) the relation R is continuous (the condition “connecting” topology and order).
We call (X, ) the topological component and (X, R) the order component of the
hybrid (X, , R).
The following theorem characterizes hybrids from various points of view.
3.1.2 Theorem Let (X, ) be a Stone space and (X, R) a quasi-ordered set. The
following conditions are equivalent:
(I) (X, , R) is a hybrid;
(II) (a) R(x) is a closed set for each x ∈ X ;
(b) for each closed (resp. open) set A ⊆ X , the set R −1 (A) = {x ∈ X :
R(x) ∩ A = ∅} is closed (resp. open);

1 Editorial note: For the definition of the exponential (Vietoris) topology, see Definition 1.3.4.
© Springer Nature Switzerland AG 2019 41
G. Bezhanishvili and W. H. Holliday (eds.), Heyting Algebras,
Trends in Logic 50, https://doi.org/10.1007/978-3-030-12096-2_3
42 3 Duality Theory: Hybrids

(III) (a) R(x) is a closed set for each x ∈ X ;

(b) if A is clopen in the space X , then so is R −1 (A).
(IV) Separation condition

(a) If y ∈
/ R(x), then there exists a cone partition of the space X , i.e., a partition
of X into two disjoint open sets U1 and U2 such that x ∈ U1 and U1 is an
upper cone, and y ∈ U2 and U2 is a lower cone.
(b) For any x ∈ X and any open V ⊆ X such that R(x) ∩ V = ∅, there is an
open neighbourhood U of x such that R(z) ∩ V = ∅ for each z ∈ U .

(V) Closed graph condition

(a) The graph of the relation R is closed in the topological space X × X = X 2 ;
(b) the set R −1 (U ) is open for each open set U .
(VI) Cones and topological closure
(a) R −1 (x) is closed for each x ∈ X ;
(b1 ) the smallest closed set containing a cone is a cone;
(b2 ) the smallest cone containing a closed set is closed.
(VII) Condition of commutation of operators R and C
(a) R −1 (x) is closed for each x ∈ X ;
(b) RC A = CR A for each set A ⊆ X , where CA is the topological closure of
A in the space (X, ).2

The theorem is based on a number of lemmas. For brevity and without the danger
of confusion, we will often write X instead of (X, ) and (X, R) instead of (X, , R).

3.1.3 Lemma (X, , R) is a hybrid if and only if:

(1) for each open set U , the set R −1 (U ) is open;

(2) for each closed set F, the set R −1 (F) is closed;
(3) R(x) is closed for each x ∈ X .

Proof Clearly ρ : X → exp X is well defined if and only if R(x) is closed for each
x ∈ X . Recall that a map f from a topological space Y to a topological space
Z is continuous if and only if the f -preimage of each set from an open subba-
sis of Z is open. A subbasis of the space exp X consists of the sets of the form
F1 (U ) = {F ∈ exp X : F ∩ U = ∅} and F2 (U ) = {F ∈ exp X : F ⊆ U }, where
U is an open subset of X . It is easy to check that:

2 Editorialnote: Following the policy for omission of parentheses explained in the Preface, in
contexts involving C, parentheses will sometimes be omitted after R and R −1 for readability.
3.1 The Hybrid of Topology (Stone) and Order (Kripke) 43

(a) ρ−1 (F1 (U )) = {x ∈ X : ρ(x) ∈ F1 (U )}

= {x ∈ X : ρ(x) ∩ U = ∅}
= {x ∈ X : R(x) ∩ U = ∅}
= R −1 (U );
(b) −ρ−1 (F2 (U )) = −{x ∈ X : ρ(x) ⊆ U }
= −{x ∈ X : ρ(x) ∩ −U = ∅}
= {x ∈ X : R(x) ∩ −U = ∅}
= R −1 (−U ).

Let ρ be continuous. Then ρ−1 (F1 (U )) is open for each open set U ⊆ X . By (a),
ρ (F1 (U )) = R −1 (U ). Therefore, R −1 (U ) is open if U is open, i.e., (1) holds. Also,
−1

ρ−1 (F2 (U )) is open, i.e., −ρ−1 (F2 (U )) is closed. By (b), −ρ−1 (F2 (U )) = R −1 (−U ),
so the R-preimage of a closed set is closed, i.e., (2) holds. Conversely, suppose
conditions (1)–(2) of the lemma hold. Then R −1 (U ) is open for each open U ⊆ X .
By (a), R −1 (U ) = ρ−1 (F1 (U )). Moreover, the set −U is closed. By condition (2),
R −1 (−U ) is closed. By (b), R −1 (−U ) = −ρ−1 (F2 (U )). Therefore, ρ−1 (F2 (U )) is
open. The continuity of ρ has thus been established. 

3.1.4 Lemma Let X be a Stone space and R a quasi-order on X such that:

(a) for each x ∈ X , the set R(x) is closed;

(b) for each clopen set A, the set R −1 (A) is clopen.

If y ∈
/ R(x), then there exists a partition of the space X into two open sets U1 and
U2 such that x ∈ U1 and U1 is an upper cone, and y ∈ U2 and U2 is a lower cone.

Proof Let x, y ∈ X and y ∈ / R(x). Then y ∈ −R(x), and −R(x) is an open lower
cone since R(x) is a closed upper cone. Because the space is zero-dimensional,
there exists a clopen set A such that y ∈ A ⊆ −R(x). Let U2 = R −1 (A) and U1 =
−R −1 (A). Clearly y ∈ U2 and U2 is a lower cone. By condition (b) of the lemma,
R −1 (A) is clopen. It is easy to see that x ∈
/ R −1 (A); for otherwise, R(x) ∩ A = ∅,
which is impossible since A ⊆ −R(x). Thus, the set U1 = −R −1 (A) is a clopen
upper cone such that x ∈ U1 . 

3.1.5 Lemma Let X be a Stone space and R a quasi-order on X such that if y ∈ /

R(x), then there exists a partition of the space X into two open sets U1 and U2 such
that x ∈ U1 and U1 is an upper cone, and y ∈ U2 and U2 is a lower cone. Then the
set G = {(x, y) : y ∈ R(x)} is closed in the product X × X = X 2 .

Proof If G is not closed in X 2 , then there is (x, y) ∈ X 2 such that

(1) (x, y) ∈ CG and (x, y) ∈
/ G (i.e., y ∈
/ R(x)).

Let {U1 , U2 } be the partition of X into two open sets such that x ∈ U1 and U1
is an upper cone, and y ∈ U2 and U2 is a lower cone. Since U1 × U2 is an open
44 3 Duality Theory: Hybrids

neighborhood of (x, y), and (x, y) belongs to the closure of G (in X 2 ), we have (U1 ×
U2 ) ∩ G = ∅, i.e., there are x  ∈ U1 and y  ∈ U2 such that (x  , y  ) ∈ G, so x  Ry  .
As U1 is an upper cone, from x  ∈ U1 and x  Ry  it follows that y  ∈ U1 . Therefore,
y  ∈ U1 and y  ∈ U2 , i.e., U1 ∩ U2 = ∅. The obtained contradiction concludes the
proof. 

3.1.6 Lemma Let X be a Stone space and R a quasi-order on X such that the graph
of G is closed in X 2 . Then:
(a) the smallest cone containing a closed set is closed;
(b) R −1 (x) is closed in X for each x ∈ X .

Proof First we show that if a set F is closed, then R(F) is closed. Since G is closed
in X 2 , G is a Stone space with respect to the subspace topology. The projection
p : G → X (where p(x, y) = y for x ∈ G) is a continuous map from the Stone
space G onto the Stone space X and hence is a closed map. Let F be a closed subset
 X . Then (F × X ) ∩ G is closed in G,and as p is closed,
of  the set p((F × X ) ∩ G) =
y ∈ X : (∃ (x0 , y0 ) ∈ (F × X ) ∩ G) p(x0 , y0 ) = y is closed in X . It is easy to
see that the following equalities hold:
  
y ∈ X : (∃ (x0 , y0 ) ∈ (F × X ) ∩ G) p(x0 , y0 ) = y
  
= y ∈ X : (∃ x0 ∈ F) (x0 , y) ∈ (F × X ) ∩ G
  
= y ∈ X : (∃ x0 ∈ F) x0 ∈ R −1 (y) = R(F).

Therefore, R(F) is closed. Thus, if F is a closed set, then the smallest cone containing
F (i.e., the set R(F)) is closed.
A similar argument shows that if F is a closed set, then R −1 (F) is closed. In
particular, as each point of X is closed, so is R −1 (x). 

3.1.7 Lemma Let X be a Stone space and R a quasi-order on X such that:

(1) the smallest closed set containing a cone is a cone;
(2) the smallest cone containing a closed set is closed.
Then RCA = CR A for each subset A of X , where C is the operation of topological
closure in X .

Proof It is easy to see that condition (1) can be written as the following inclusion:
(10 ) RCR A ⊆ CR A.
Since R is reflexive, A ⊆ R(A), so C A ⊆ CR A, and hence RC A ⊆ RCR A. This
together with (10 ) and the transitivity of ⊆ gives RCA ⊆ CR A. Condition (2) can
be rewritten as the following inclusion:
(20 ) CRCA ⊆ RCA.
3.1 The Hybrid of Topology (Stone) and Order (Kripke) 45

Since A ⊆ CA, we have R A ⊆ RCA, so CR A ⊆ CRCA. By (20 ) and the transitivity

of ⊆, we obtain CR A ⊆ RCA. The two inclusions RCA ⊆ CR A and CR A ⊆ RCA
give the required equality. 

3.1.8 Lemma Let X be a Stone space and R a quasi-order on X such that:

(1) R −1 (x) is closed for each x ∈ X ;

(2) RCA = CR A for each subset A of X .

Then:

(a) R(x) is closed for each x ∈ X ;

(b) for each closed (resp. open) set A in X , the set R −1 (A) is closed (resp. open).

Proof First we show that if F is closed, then R(F) is closed. Since F is closed, CF =
F. By (2), CR F = RCF = R(F). Therefore, R(F) is closed. As each singleton {x}
is closed, it follows that R(x) is closed.
Next we show that if U is open, then R −1 (U ) is open. Since R −1 (U ) is a lower
cone, −R −1 (U ) is an upper cone. Moreover, U ∩ −R −1 (U ) = ∅. As U is open,
U ∩ C−R −1 (U ) = ∅. By (2), the closure of each upper cone is an upper cone,
so C−R −1 (U ) is an upper cone. Therefore, R −1 (U ) ∩ C−R −1 (U ) = ∅, and so
C−R −1 (U ) ⊆ −R −1 (U ). This implies that −R −1 (U ) is closed, and hence R −1 (U )
is open.
It is left to show thatif F is closed, then R −1 (F) is closed. For this it is sufficient to
show that R −1 (F) = R −1 (x) is compact. Let F = {U j : j ∈ J } be an open cover
x∈F
of R −1 (F). Clearly F is an open cover of R −1 (x) for each x ∈ F. By (1), R −1 (x) is
closed andhence compact. Therefore, there is a finite subcover, say {U j : j ∈ Jx }.
Let Ux = {U j : j ∈ Jx } and Ux0 = −R−Ux . As we already saw, the R-image of
a closed set is closed. This implies that Ux0 is open. From R −1 (x) ⊆ Ux it follows
that x ∈ Ux0 . Consequently, {Ux0 : x ∈ F} is an open cover of the closed set F, so
there is a finite subcover, say {Ux01 , . . . , Ux0k }. But then {U j : j ∈ Jx1 ∪ · · · ∪ Jxk } is
a finite subcover of R −1 (F). Thus, R −1 (F) is compact and hence closed. 

Proof of Theorem 3.1.2. By Lemma 3.1.3, conditions (I) and (II) are equivalent.
(II ⇒ III) Let A be a clopen subset of X . By (II), the set R −1 (A) is closed since
A is closed, and R −1 (A) is open since A is open.
(III
 ⇒ IV)
 By  Lemma 3.1.4, (III) implies (IVa). We now show (IVb). Since
R −1 Ai = R −1 (Ai ), it follows from (IIIb) and the zero-dimensionality of X
i∈I i∈I
that the R-preimage of each open set is open. Let x ∈ X and V be an open set such that
R(x) ∩ V = ∅. Let U = R −1 (V ). Since V is open, U is open. As R(x) ∩ V = ∅,
we have x ∈ U . If z ∈ U , then there exists y ∈ V such that z ∈ R −1 (y). Therefore,
y ∈ R(z), and so R(z) ∩ V = ∅.
(IV ⇒ V) By Lemma 3.1.5, (Va) follows from (IVa). We now prove (Vb). Let U
be an open set. Consider the set R −1 (U ). Let z ∈ R −1 (U ). By (IVb), there exists an
open set V such that z ∈ V and (∀ y ∈ V )(R(y) ∩ U = ∅). The latter condition can
46 3 Duality Theory: Hybrids

be written as V ⊆ R −1 (U ). Therefore, each point z ∈ R −1 (U ) is an interior point,

and hence R −1 (U ) is open.
(V ⇒ VI) By Lemma 3.1.6, (VIa) and (VIb2 ) follow from (Va). It remains to
prove (VIb1 ). For this it is sufficient to show that CR A is a cone for each A ⊆ X . Let
x ∈ CR A and x Ry. If y ∈ / CR A, then there exists an open set U such that y ∈ U and
U ∩ R(A) = ∅. Since R(A) is a cone, this implies R −1 (U ) ∩ R(A) = ∅. By (Vb),
R −1 (U ) is open. Therefore, R −1 (U ) ∩ CR A = ∅. As x Ry, we have x ∈ R −1 (U ).
Therefore, R −1 (U ) is an open neighborhood of x entirely missing CR A. Thus,
x∈/ CR A, a contradiction. Consequently, CR A is a cone, and so the smallest closed
set containing a cone is a cone.
(VI ⇒ VII) Follows from Lemma 3.1.7.
(VII ⇒ II) Follows from Lemma 3.1.8.
The theorem is proved. 
We give a few “degenerate” examples of hybrids.
(A) Let (X, ) be a Stone space and R the identity relation on the set X , i.e, x ∈
R −1 (y) if and only if x = y. Then (X, , R) is a hybrid. Indeed, since R −1 (x) =
{x} for each x ∈ X and all singletons {x} are closed (as X is Hausdorff), condition
(VIIa) of Theorem 3.1.2 holds. Since for each set A in X , we have R(A) = A,
the commutation condition (VIIb) obviously holds as well.
(B) Let (X, ) be a Stone space and R the relation on X such that x ∈ R(y) and
y ∈ R(x) for each x, y ∈ X (i.e., (X, R) is a cluster). Then (X, , R) is a
hybrid. Indeed, in this case, R(A) = X for each nonempty subset A of X , and
R(∅) = ∅. Clearly condition (VI) of Theorem 3.1.2 holds.
(C) Let (X, R) be a finite quasi-ordered set. Equip X with the discrete topology.
Then it is obvious that (X, , R) is a hybrid.
We conclude this section by fixing some notation and terminology.
3.1.9
(1) The class (and the category) of all hybrids will be denoted by HYB.3
(2) HYB+ is the class (and the category) of all strict (or partially ordered) hybrids,
i.e., hybrids (X, , R) such that R is a partial order.
(3) ST is the class (and the category) of Stone spaces.
(4) QO (resp. QOF) is the category of all (resp. all finite) quasi-ordered sets and
strictly isotone maps.
(5) PO (resp. POF) is the category of all (resp. all finite) partially ordered sets and
strictly isotone maps.
Bibliographic notes. Hybrids (Definition 3.1.1) were introduced originally by the
author in [4] under the name of ‘perfect Kripke models’. The notion was motivated
by the papers of Jónsson and Tarski [10] and Halmos [13].

3 Editorial note: The morphisms of the category HYB will be introduced in Definition 3.3.1.
3.2 Fundamental Properties of Hybrids 47

3.2 Fundamental Properties of Hybrids

3.2.1 Theorem Let (X, R) be a strict hybrid and F a closed subset of X . Then
for each x ∈ F, there exists a maximal (resp. minimal) y ∈ F such that x Ry (resp.
y Rx).

Proof Suppose that there exists x ∈ F such that max(F ∩ R(x)) = ∅. Let Y be
a maximal chain contained in R(x) ∩ F.4 Consider the family F = {R(y) ∩ F :
y ∈ Y } of nonempty closed sets. Clearly F is a chain with respect to inclusion and
by compactness of X , we have {R(y) ∩ F : y ∈ Y } = ∅. Let z ∈ {R(y) ∩ F :
y ∈ Y }, i.e., for each y ∈ Y , we have y Rz and z ∈ F. Then z ∈ max Y because Y is
a maximal chain. Therefore, z ∈ max(R(x) ∩ F), a contradiction. The dual part of
the theorem is proved similarly. 

3.2.2 Corollary Let (X, R) ∈ HYB+ and F be a nonempty closed subset of X . Then
max F = ∅ and min F = ∅.

Remark to Theorem 3.2.1 and Corollary 3.2.2. It is easy to check that Theo-
rem 3.2.1 and Corollary 3.2.2 hold for arbitrary hybrids provided in the statement
the prefix ‘quasi-’ is added to the words ‘maximal’ and ‘minimal’. It is appropriate
here to recall Definition 1.4.7 and especially the informal remarks after it.

3.2.3 Theorem Let (X, R) ∈ HYB+ . Then the set max X is closed in X .

Proof Let y ∈ X and y ∈ / max X . Consider x ∈ max X such that y Rx (see Theo-
rem 3.2.1). Then y ∈/ R(x). By Theorem 3.1.2, there is a clopen upper set A such that
x ∈ A and y ∈ / A. Clearly y ∈ R −1 (A) − A and x ∈ / R −1 (A) − A. Since R −1 (A)
−1 −1
is clopen, so is R (A) − A. Obviously (R (A) − A) ∩ max A = ∅. Therefore, y
has an open neighborhood R −1 (A) − A disjoint from max X . Thus, y ∈ / C max X ,
where C max X is the topological closure of max X . 

However, the dual statement of Theorem 3.2.3 is false as the following theorem
shows.

3.2.4 Theorem There exists a strict hybrid (X, , R) such that the set min X is not
closed.

Proof Let X be the one-point compactification of an infinite discrete space X 0 , and

let z ∈ X − X 0 , i.e., z is the only compactification point. It is known that X is a Stone
space.5 Define a partial order R on X as follows:
(1) x Rx for each x ∈ X (reflexivity of R);
(2) R −1 (x) = {x} for each x ∈ X 0 ;
(3) R −1 (z) = X .

4 Editorial note: The existence of such a maximal chain follows from Proposition 1.4.10.
5 Editorial note: See, e.g., Example D on p. 26 of [18].
48 3 Duality Theory: Hybrids

Therefore, z is the greatest point of X , and min X = X 0 . It is easy to check that

(X, , R) is a strict hybrid and that the set min X = X 0 is not closed in the space X
(recall that z is the only non-isolated point of the space). 

Recall that a set A in a partially ordered set (X, R) is said to be an antichain if

any two distinct points x, y of A are incomparable, i.e., x ∈/ R(y) and y ∈/ R(x).

3.2.5 Corollary There exist a strict hybrid (X, R), an open set U ⊆ X , and an
antichain A ⊆ X such that:
(1) the set R(U ) is not open in the space X ;
(2) the closure of the antichain A (i.e., the set C A) in X is not an antichain.

Proof Let (X, R) be the strict hybrid constructed in the proof of Theorem 3.2.4.
(1) Set U = {x}, where x ∈ X 0 . Clearly U is open, and R(U ) = R(x) = {x, z},
which is not open.
(2) Take the set X 0 as our antichain A. The closure CX 0 of X 0 is equal to X , but
X is not an antichain. 

For a hybrid (X, , R) and a subset Y of X , let (Y, Y , RY ) denote the set Y
equipped with the subspace topology Y and the order RY which is the restriction
of R to Y , i.e., RY (x) = R(x) ∩ Y for each y ∈ Y .

3.2.6 Theorem Let (X, , R) ∈ HYB+ and Y be a clopen set in (X, ). Then
(Y, Y , RY ) is a strict hybrid.

Proof It is well known that (Y, Y ) is a Stone space. Moreover, if x ∈ Y , then the
set RY (x) = R(x) ∩ Y is closed in the space Y . Now let A be a clopen set in (Y, Y ).
Then A = A0 ∩ Y for some clopen A0 in (X, ). Since (X, , R) is a hybrid, by
Theorem 3.1.2, R −1 (A0 ∩ Y ) is clopen in the space (X, ). The remainder of the
proof follows from the following easily verifiable equalities:
 
RY−1 (A0 ∩ Y ) = RY−1 (x) : x ∈ A0 ∩ Y
 
= R −1 (x) ∩ Y : x ∈ A0 ∩ Y
 −1 
=Y∩ R (x) : x ∈ A0 ∩ Y
= Y ∩ R −1 (A0 ∩ Y ). 

3.2.7 Corollary Let Y be a clopen subset of a strict hybrid (X, , R). Then the set
max Y is closed in (X, ).

Proof By Theorem 3.2.6, (Y, Y , RY ) is a hybrid. Therefore, by Theorem 3.2.3, the

set max Y is closed in (Y, Y ). Since Y is closed in (X, ), max Y is closed in (X, ).


If (X, R) is a partially ordered set, then we call a set of the form R(x) ∪ R −1 (x),
for x ∈ X , a bow (with the knot x).
3.2 Fundamental Properties of Hybrids 49

3.2.8 Theorem Let Y be a chain of a strict hybrid (X, R). Then its topological
closure CY is also a chain.

Proof Let Y be a chain in (X, R) and y ∈ CY . To show that y is comparable with each
x ∈ Y, i.e., that y ∈ R(x) ∪ R −1 (x) for each x ∈ Y, suppose
y∈ / R(x) ∪ R (x) for some x ∈ Y . The bow R(x) ∪ R −1 (x) is closed and
−1

Y ⊆ R(x) ∪ R −1 (x). Therefore, y ∈ −(R(x) ∪ R −1 (x)), −(R(x) ∪ R −1 (x)) is open,

and Y ∩ −(R(x) ∪ R −1 (x)) = ∅. Thus, y ∈ / CY , a contradiction. Consequently, y
is comparable with each point of Y .
Let x, y ∈ CY and x, y ∈ / Y . We will show that x and y are comparable. If
not, then x ∈/ R(y) ∪ R −1 (y), i.e., x ∈ −(R(y) ∪ R −1 (y)). Note that Y ⊆ R(y) ∪
R −1 (y). Therefore, x ∈/ CY , a contradiction. Thus, x ∈ R(y) ∪ R −1 (y). 

3.2.9 Corollary Let (X, R) ∈ HYB+ and Y be a maximal chain in (X, R). Then Y
is closed in X .

Proof Since Y is a chain, CY is also a chain by Theorem 3.2.8. Then since Y ⊆ CY

and Y is a maximal chain, we have Y = CY , so Y is closed. 

3.2.10 Theorem Let (X, , R) be a strict hybrid and Y a closed chain. Then
(Y, Y , RY ) is a linear strict hybrid (i.e., a strict hybrid whose order component
(Y, RY ) is a chain).

Proof Since Y is closed in (X, ), (Y, Y ) is a Stone space. As RY (x) = Y ∩ R(x)
for x ∈ Y , the set RY (x) is closed in (Y, Y ). Let A be clopen in (X, ). Then
A ∩ Y is clopen in (Y, Y ). Consider the set A0 = RY−1 (A ∩ Y ) = Y ∩ R −1 (A ∩ Y ).
Clearly A0 is closed in (Y, Y ). We will prove that A0 is open in (Y, Y ). Let
x ∈ max(A ∩ Y ). Since (Y, RY ) is a chain, we have RY−1 (A ∩ Y ) = RY−1 (x). The
set Y0 = (R(x) − {x}) ∩ Y is contained in the closed set R(x) ∩ Y and is open in
(Y, Y ) since Y0 = Y − RY−1 (x). Consider the set CY Y0 , i.e., the topological closure
of Y0 in (Y, Y ). Since Y0 ⊆ R(x) ∩ Y = RY (x) and the set R(x) ∩ Y is closed in
(Y, Y ), we have CY Y0 ⊆ R(x) ∩ Y . Moreover, x ∈ / CY Y0 since A ∩ Y is clopen in
Y , x ∈ A ∩ Y , and (A ∩ Y ) ∩ Y0 = ∅. Let x0 ∈ min CY Y0 . Then x is covered by x0
(i.e., x RY x0 , and if x RY z and z RY x0 , then x = z or z = x0 ). The set RY (x0 ) is closed
since RY (x0 ) = Y ∩ R(x0 ), and so the set Y − RY (x0 ) = RY−1 (x) = RY−1 (A ∩ Y ) is
open in (Y, Y ). The theorem is proved. 

If (X, R) is a partially ordered set and Y ⊆ X , let sup Y (resp. inf Y ) denote the
least (resp. greatest) element of the set {x ∈ X : (∀ y ∈ Y )(y Rx)} (resp. of the set
{x ∈ X : (∀ y ∈ Y )(x Ry)}).

3.2.11 Lemma Let Y be a nonempty chain in a strict hybrid (X, R). Then sup Y
and inf Y exist and sup Y, inf Y ∈ CY , where CY is the closure of Y in X .

Proof Let Y be a nonempty chain in (X, R). If max Y = ∅, then clearly max Y = {x}
and x = sup Y . It is also clear that x ∈ Y , and since Y ⊆ CY , we have x ∈ CY . Let
max Y = ∅. By Theorem 3.2.8, CY is a closed chain in the hybrid (X, R). Therefore,
50 3 Duality Theory: Hybrids

by Theorem 3.2.1, max(CY ) = ∅. Let x ∈ max CY . Then x = sup Y . Indeed, since

Y ⊆ CY and x ∈ max CY , for each y ∈ Y , y Rx. Let z be such that y Rz for all
y ∈ Y . We will show that x Rz, i.e., that z ∈ R(x). Suppose that z ∈ / R(x). Then
the chain Y0 = R −1 (z) ∩ CY is a closed set, x ∈/ Y0 , and Y ⊆ Y0 . Since Y ⊆ Y0 ,
CY ⊆ CY0 = Y0 . As x ∈ / Y0 , we also have x ∈
/ CY , which together with x ∈ CY
yields a contradiction. Thus, z ∈ R(x), and so x = sup Y . The existence of inf Y is
proved similarly. 

If (X, R) is a partially ordered set and Y is a chain in (X, R), let Y denote the
completion of the chain Y , i.e., for each x ∈ X , x ∈ Y if and only if there exists a
subset Y0 of Y such that sup Y0 = x or inf Y0 = x.

3.2.12 Theorem Let (X, R) be a strict hybrid and Y a chain in (X, R). Then the
completion Y of Y coincides with its topological closure CY .

Proof We show that Y = CY . If the chain Y is finite, then obviously Y = Y = CY .

Let Y be infinite, Y0 ⊆ Y , and sup Y0 = x. By Lemma 3.2.11, x ∈ CY0 . Since
CY0 ⊆ CY , we have x ∈ CY . Similarly, if x = inf Y0 , then x ∈ CY . Therefore,
Y ⊆ CY . To show that CY ⊆ Y , let x ∈ CY and x ∈ / Y . Consider the chains
Y0 = R −1 (x) ∩ Y and Y1 = R(x) ∩ Y . Let x0 = sup Y0 and x1 = inf Y1 . Suppose
x0 = x and x = x1 . Obviously x0 ∈ R −1 (x) and x1 ∈ R(x). Let Z 0 = R −1 (x0 ) ∩ CY
and Z 1 = R(x1 ) ∩ CY . Then Z 0 , Z 1 are closed, Y ⊆ Z := Z 0 ∪ Z 1 , and the set Z is
closed and contained in CY . Moreover, x ∈ / Z . Thus, −Z is an open neighborhood
of x, and −Z ∩ Y = ∅. Hence, x ∈ / CY , a contradiction. So x = x0 or x = x1 , and
so CY ⊆ Y . Consequently, Y = CY . 

3.2.13 Definition Let (X, R) be a partially ordered set, x, y ∈ X , x = y, and

x Ry. The ordered pair (x, y) is said to be a gap in (X, R) if y covers x, i.e., if
z ∈ R(x) ∩ R −1 (y) implies z = x or z = y. A point x ∈ X is said to be a gap in
(X, R) if either x ∈ max X or there exists y ∈ X such that (x, y) is a gap in (X, R).
Let Gap(X, R) denote the set of all points of X that are gaps. A partially ordered
set (X, R) is said to have enough gaps if each “interval” R(x) ∩ R −1 (y) (where
x = y and x Ry) contains a gap, i.e., if there is a gap (x  , y  ) in (X, R) such that
x  , y  ∈ R(x) ∩ R −1 (y).

3.2.14 Lemma Let (X, , R) be a linear strict hybrid. Then the linearly ordered
set (X, R) has enough gaps.

Proof Let x = y and x Ry. We will find a gap in the interval R(x) ∩ R −1 (y). Since R
is anti-symmetric, x ∈/ R(y). Therefore, there is a clopen lower set A such that x ∈ A
and y ∈ / A. Clearly −A is a clopen upper set, x ∈/ −A, and y ∈ −A. Let x  ∈ max A
 
and y ∈ min −A. Then (x , y ) is the gap we sought for. 

3.2.15 Lemma Let (X, , R) be a linear strict hybrid. Then the family F of sets of
the form R(x) ∩ −R(y) and R(x) ∪ −R(y), where x, y ∈ Gap(X, R), is a reduced
field of sets.
3.2 Fundamental Properties of Hybrids 51

Proof Let F0 be the smallest field of subsets of X containing all sets of the form
 R). Clearly F ⊆ F0 . Since each element A ∈ F0 can be
R(x), where x ∈ Gap(X,
represented as A = (Ak − Bk ), where m ∈ ω, Ak and Bk are sets of the form
k<m  
R(x) for suitable x ∈ Gap(X, R),6 and (Ak − Bk ) = (Ak ∩ −Bk ), we have
k<m k<m
A ∈ F. Therefore, F = F0 , and so F is a field. Moreover, the field F is reduced.
Indeed, let x, y ∈ X and x = y. Then since (X, R) is linearly ordered, either x Ry or
y Rx. Without loss of generality we may assume that x Ry. By Lemma 3.2.14, there
is a gap (x  , y  ) in the interval R(x) ∩ R −1 (y). We have R(y  ) ∈ F, y ∈ R(y  ), and
x∈ / R(y  ). 

3.2.16 Theorem Any strict hybrid (X, R) has enough gaps.

Proof Let x, y ∈ X , x = y, and x Ry. To find a gap in the interval R(x) ∩ R −1 (y),
let Y be a maximal chain in (X, R) such that x, y ∈ Y . By Lemma 3.2.9, Y is closed
in X . By Theorem 3.2.10, (Y, RY ) is a linear strict hybrid. Let (x  , y  ) be a gap in the
hybrid (Y, RY ), where (x  , y  ) ∈ Y and x  , y  ∈ RY (x) ∩ RY−1 (y). Then (x  , y  ) is a
gap in the hybrid (X, R). Indeed, let z ∈ X be such that z = x  , z = y  , x  Rz, and
z Ry  . Since the chain Y is maximal, z ∈ Y , contradicting the fact that (x  , y  ) is a
gap in (Y, RY ). 

3.2.17 Theorem Let (X, , R) be a strict hybrid and Y a closed chain in X . Then
the subspace topology of the hybrid (Y, Y , RY ) coincides with the classical interval
topology of the chain.

Proof Recall that the “open” intervals of the chain Y are

(1) Y;
(2) −RY (x), where x ∈ Y ;
(3) −RY−1 (x), where x ∈ Y ;
(4) −RY (x) ∩ −RY−1 (y), where y Rx and y = x.

The interval topology 0Y of the chain (Y, RY ) is obtained by taking these open
intervals as a basis for the topology. Note that a subset of the chain (Y, RY ) is open in
its interval topology 0Y if and only if it is a union of open intervals. Also, recall that
in any chain with a least element and a greatest element, the “closed” intervals form
a subbasis for closed sets. By Lemma 3.2.11, the chain (Y, RY ) is a complete lattice,
so the space (Y, 0Y ) is compact. Obviously (Y, 0Y ) is Hausdorff. By Lemma 3.2.14,
the chain (Y, Y ) has enough gaps. Each closed interval in (Y, RY ), i.e., each set of
the form RY (x) ∩ RY−1 (y) where x Ry, is closed in (Y, Y ). Therefore, the interval
topology 0Y is dominated by the topology Y of the hybrid (Y, Y , RY ). But any
Hausdorff topology dominated by a compact topology coincides with it.7 

note: Since (X, R) is a chain, the family {R(x) : x ∈ Gap(X, R)} is a ring of sets.
6 Editorial
7 Editorial
note: See, e.g., Corollary 3.1.14 of R. Engelking, General Topology, 2nd ed.
(Heldermann-Verlag), 1989.
52 3 Duality Theory: Hybrids

3.2.18 Definition A partially ordered set (X, R) is said to be Dedekind complete8

if for each nonempty upward (resp. downward) directed subset Z of X , sup Z (resp.
inf Z ) exists.9

3.2.19 Theorem Each strict hybrid (X, R) is Dedekind complete.

Proof Let (X, R) be a strict hybrid. By Lemma 3.2.11, each nonempty chain Y of X
has sup Y and inf Y . Now let Z be an arbitrary nonempty upward (resp. downward)
directed set. The existence of sup Z (resp. inf Z ) is guaranteed by a theorem of
P. Cohn: if all nonempty chains in a partially ordered set have sup and inf, then each
nonempty upward (resp. downward) directed set has sup (resp. inf).10 

As an immediate consequence of Theorem 3.1.2(IVa), if x ∈/ R −1 (y) in a hybrid

(X, R), then there exists a clopen upper cone A such that x ∈ A and y ∈ / A. The
following lemma states that a stronger principle holds.

3.2.20 Lemma Let (X, R) be a hybrid with x ∈ X and D a closed lower cone
such that x ∈
/ D. Then there exists a clopen upper cone A such that R(x) ⊆ A and
A ∩ D = ∅.

Proof Let x ∈ / D. Then R(x) ⊆ −D as −D is an upper cone. For each y ∈ D, we

have x ∈ / R −1 (y). Therefore, there is a clopen upper cone A y such that x ∈ A y and
y∈/ A y . Thus, F = {A y ∩ D : y ∈ D} is a family of closed subsets of X . Clearly
F = ∅ since for each y ∈ D, we have y ∈ / A y . By the compactness of X , we
have {A y ∩ D : y ∈ D0 } = ∅ for some finite subset D0 of D. Let A = {A y :
y ∈ D0 }. Then A is a clopen upper cone, x ∈ A, so R(x) ⊆ A, and A ∩ D = ∅. 

3.2.21 Definition

(a) Let (X, ) be a Stone space and R a quasi-order on X . We say that the quasi-
ordered space (X, , R) satisfies the Priestley separation axiom if for any
x, y ∈ X such that y ∈ / R(x), there exists a partition of X into two open sets X 1
and X 2 such that X 1 is an upper cone and x ∈ X 1 , and X 2 is a lower cone and
y ∈ X 2.
(b) If in addition the relation R is anti-symmetric (i.e., (X, R) is a partially ordered
set), then (X, , R) is said to be a Priestley space (or more precisely, a partially
ordered Priestley space).

3.2.22 Theorem

(1) Any hybrid (X, , R) satisfies the Priestley separation axiom.

8 Editorial note: This is a nonstandard use of the term ‘Dedekind complete’, which usually means

that each nonempty subset bounded above has a supremum (resp. bounded below has an infimum).
9 Editorial note: A subset Z of X is upward (resp. downard) directed if for all x, y ∈ Z , there is

z ∈ Z such that x Rz and y Rz (resp. z Rx and z Ry).

10 Editorial note: See Proposition 5.9 on p. 33 of [3].
3.2 Fundamental Properties of Hybrids 53

(2) Conversely, if (X, , R) is a Priestley space, then (X, , R) is a strict hybrid if

and only if for any x ∈ X and any open set V of (X, ) such that R(x) ∩ V = ∅,
there exists an open neighborhood U of x such that R(z) ∩ V = ∅ for all z ∈ U .

Proof (1) Follows from Theorem 3.1.2(IVa).

(2) Let (X, , R) be a Priestley space. Then (X, , R) satisfies condition (IVa)
of Theorem 3.1.2. Moreover, R is a partial order. Therefore, by Theorem 3.1.2,
(X, , R) is a strict hybrid if and only if condition (IVb) of Theorem 3.1.2
holds. 

3.2.23 Definition Let (X, , R) ∈ HYB and C be the closure operator of the Stone
space (X, ). Define two operators C1 and C2 as follows: C1 A = CR A and C2 A =
R −1 CA for each subset A of X .

3.2.24 Lemma Let (X, , R) be a hybrid. The operators C1 and C2 satisfy Kura-
towski’s axioms, where A, B ⊆ X and i = 1, 2:

(I) A ⊆ Ci A;
(II) Ci Ci A ⊆ Ci A;
(III) Ci (A ∪ B) = Ci A ∪ Ci B;
(IV) Ci ∅ = ∅.

Proof Since A ⊆ CA, A ⊆ R A, and A ⊆ R −1 A (by reflexivity of R) for any A ⊆

X , both operators satisfy axiom (I). The additivity of topological closure and the
operators R and R −1 implies the additivity of Ci (i = 1, 2), i.e., axiom (III). Since
C∅ = ∅, R∅ = ∅, and R −1 ∅ = ∅, axiom (IV) holds. It remains to show that axiom
(II) holds.
First we show that C1 C1 A ⊆ C1 A, i.e., that CRCR A ⊆ CR A. By
Theorem 3.1.2(VIIb), C and R commute, so CRCR A = CCR R A ⊆ CR A (as R
is transitive).
Next we prove that C2 C2 A ⊆ C2 A, i.e., that R −1 CR −1 CA ⊆ R −1 CA. Since CA
is closed, by Theorem 3.1.2(IIb), the set R −1 CA is also closed, i.e., CR −1 CA =
R −1 CA. Hence R −1 CR −1 CA = R −1 R −1 CA ⊆ R −1 CA, where the last inclusion
follows from the transitivity of R. 

3.2.25 Lemma Let (X, , R) ∈ HYB and C be the operation of topological clo-
sure in (X, ). Then:

(1) C1 A = A if and only if A is a closed upper cone of the hybrid;

(2) C2 A = A if and only if A is a closed lower cone of the hybrid (X, , R).

Proof (1) If A is a closed upper cone, then R(A) = A since A is an upper cone and
CA = A since A is closed. Therefore, C1 A = CR A = A. Conversely, let C1 A = A,
i.e., CR A = A. Since CA ⊆ CR A, we have C A ⊆ A. From R A ⊆ CR A it also
follows that R(A) ⊆ A. Thus, A is a closed upper cone.
(2) If A is a closed lower cone, then C2 A = R −1 CA = A. Conversely, let C2 A =
A. We will show that A is a closed lower cone. By Theorem 3.1.2(IIb), if A0 is
54 3 Duality Theory: Hybrids

a closed set, then so is R −1 (A0 ). Therefore, since CA is closed, so is R −1 CA,

i.e., CR −1 CA = R −1 CA. By assumption, R −1 CA = A. Thus, CA = A, i.e., A is
closed. Using R −1 CA = A and CA = A again yields R −1 (A) = A, i.e., A is a lower
cone. 

3.2.26 Corollary Each of the families 1 of all open lower cones and 2 of all open
upper cones of a hybrid (X, , R) forms a Heyting lattice with respect to inclusion.

Proof By Lemma 3.2.24, C1 and C2 are closure operators. By Lemma 3.2.25, the
family 1 (resp. 2 ) coincides with the family of all open sets of the topology
induced by C1 (resp. C2 ). 

We conclude this section by the following remark.

3.2.27 Remark
(1) Theorem 3.2.3 is true for arbitrary hybrids provided in its statement we replace
the set max X by the set qmax X of quasi-maximal points.
(2) Theorem 3.2.6 still holds if ‘strict hybrid’ is replaced by ‘hybrid’.
(3) Corollary 3.2.7 holds if ‘strict hybrid’ is replaced by ‘hybrid’ and the set max Y
is replaced by qmax Y .
(4) Theorem 3.2.8 is still true if (X, Y ) is a hybrid and Y , CY are quasi-chains.
(5) Lemma 3.2.9 is also true if (X, R) is a hybrid and Y is a maximal quasi-chain.
(6) Theorem 3.2.10 still holds if (X, , R) is a hybrid, Y is a closed quasi-chain,
and (Y, Y , RY ) is a quasi-linear hybrid.

Bibliographic notes. There are many publications dedicated to connections between

topology and order; we only point out the interesting monograph by Nachbin [15].
In connection with Theorem 3.2.16, it is appropriate to mention the early work of
Mostowski and Tarski [14]. For the notion of Dedekind completeness, see [19].
Priestley spaces were introduced (under a different name) in the important paper
[16] dedicated to distributive lattices.

3.3 The Category of Hybrids and Hybrid Maps

3.3.1 Definition Let (X 1 , 1 , R1 ) and (X 2 , 2 , R2 ) be hybrids. A map f : X 1 →

X 2 is said to be hybrid if f is:

(1) continuous with respect to the topological components of these hybrids;

(2) strongly isotone with respect to their order components.
3.3 The Category of Hybrids and Hybrid Maps 55

3.3.2 Lemma Let (X i , i , Ri ) ∈ HYB and Ci be the operation of topological clo-

sure in (X i , i ) (i = 1, 2). The following are equivalent:
(1) the map f : X 1 → X 2 is hybrid;
(2) for each x ∈ X 1 and each subset A of X 1 , the following commutation conditions
hold:
(a) f C1 (A) = C2 f (A) and (b) f R1 (x) = R2 f (x).

Proof Since each continuous map from a compact topological space to a Haus-
dorff space is closed, the continuity of the map f is equivalent to (2a), and by
Proposition 1.4.12, f is strongly isotone if and only if (2b) holds. 

Let HYB (resp. HYB+ ) denote the category of hybrids (resp. strict hybrids) and
hybrid maps (see 3.1.9). Also recall that ST is the category of Stone spaces and
continuous maps.
Our next task is to establish the duality of the category HYB of hybrids and the
category CA of closure algebras, and that of the category HYB+ of strict hybrids and
the category HA of Heyting algebras.

3.3.3 Definition of Functors

(1a) To every closure algebra (B, C) with the skeleton H we assign (B, C)∗ =
(X, , R), where X is the set of all ultrafilters of the Boolean algebra
B. The topology  is generated by the basis F = {ϕ(a) : a ∈ B} (ϕ(a) =
{x ∈ X : a ∈ x}). The binary relation R is given by x Ry ⇔ a ∈ y implies
Ca ∈ x for each a ∈ B.
(1b) To every morphism h : B1 → B2 of the category CA we assign the map h ∗ =
f : X 2 → X 1 given by f (x) = h −1 (x), where x is an ultrafilter of B2 , X 1 =
(B1 )∗ , and X 2 = (B2 )∗ .
(2a) To every hybrid (X, , R) ∈ HYB we assign the algebra (B(X ), ∪, ∩,
−, ∅, X, R −1 ) = (X, , R)∗ of clopen sets, where B(X ) is the field of all
clopen subsets of the Stone space X .
(2b) To every hybrid map f : X 1 → X 2 ((X i , i , Ri ) ∈ HYB, i = 1, 2) we assign
the map f ∗ = h : B(X 2 ) → B(X 1 ) given by h(A) = f −1 (A) for each
A ∈ B(X 2 ).
(3) For every closure algebra (B, C) ∈ CA, let ϕ B = ϕ be the map B → B(X ),
where X = B∗ , given by ϕ(a) = {x ∈ X : a ∈ X } for a ∈ B.
(4) For every hybrid (X, , R) ∈ HYB, let ψ X = ψ be the map X → B(X )∗ given
by ψ(x) = {A ∈ B(X ) : x ∈ A} for x ∈ X .

3.3.4 Duality Theorem The correspondences (B, C) → (B, C)∗ , h → h ∗ and

(X, , R) → (X, , R)∗ , f → f ∗ are contravariant functors establishing the dual-
ity (= coequivalence) of the categories CA and HYB.

The proof of the theorem is based on the following lemmas. Let (B, C) ∈ CA, H
be its skeleton, and (B, C)∗ = (X, , R).
56 3 Duality Theory: Hybrids

3.3.5 Lemma The following are equivalent for x, y ∈ X :

(1) a ∈ y implies Ca ∈ x for each a ∈ B;
(2) Ia ∈ x implies a ∈ y for each a ∈ B;
(3) x ∩ H ⊆ y ∩ H .

Proof (1 ⇔ 2) Condition (1) is equivalent to (∀a ∈ B) (Ca ∈ /x ⇒a∈ / y). Since x

and y are ultrafilters, this condition is equivalent to (∀a ∈ B) (−Ca ∈ x ⇒ −a ∈ y).
Replacing a by −a and using Ia = −C−a, we obtain the equivalent condition
(∀a ∈ B) (Ia ∈ x ⇒ a ∈ y), which is condition (2).
(2 ⇒ 3) Let a ∈ x ∩ H . Since Ia = a, by (2), a ∈ y ∩ H .
(3 ⇒ 2) Let x ∩ H ⊆ y ∩ H and Ia ∈ x. Then Ia ∈ x ∩ H , so Ia ∈ y ∩ H . Since
Ia ≤ a, we obtain a ∈ y. 

The following is an immediate consequence of Lemma 3.3.5.

3.3.6 Lemma The relation R is reflexive and transitive, i.e., (X, R) is a quasi-
ordered set.

3.3.7 Lemma
(1) The family F = {ϕ(a) : a ∈ B} is a basis of the space (X, ) consisting of clopen
sets;
(2) the space (X, ) is compact, Hausdorff, and zero-dimensional;
(3) the map ϕ is a bijective isomorphism of the closure algebra (B, C) onto the
algebra of clopen sets (B(X ), R −1 ) of the hybrid (X, , R).

Proof (1) We have ϕ(1) = X and ϕ(a) ∩ ϕ(b) = ϕ(a ∧ b) for a, b ∈ B since
ϕ(a ∧ b) = {x : a ∧ b ∈ x} = {x ∈ X : a ∈ x} ∩ {x ∈ X : b ∈ x}, so F is a basis.
It follows from the definition of the topology on X that every set ϕ(a), where
a ∈ B, is open. Moreover, ϕ(−a) = X − ϕ(a) since ϕ(−a) = {x ∈ X : −a ∈ x}
= {x ∈ X : a ∈ / x} = X − ϕ(a) because elements of X are ultrafilters. Thus, each
ϕ(a) is clopen.
(2) That X is zero-dimensional follows from (1). To see that X is Hausdorff, if
x, y ∈ X and x = y, then x − y = ∅, i.e., there exists a ∈ x − y. By (1), ϕ(a) is
clopen, x ∈ ϕ(a), and y ∈ / ϕ(a). To see that X is compact, let K ⊆ B, and sup-
pose that {ϕ(a) : a ∈ K } = ∅ but each finite subfamily has a nonempty inter-
section. Then b1 , . . . , bk ∈ K implies b1 ∧ · · · ∧ bk = 0, so the filter generated by
K is proper. Hence, there exists an ultrafilter x ∈ X such that K ⊆ x. Therefore,
x ∈ {ϕ(a) : a ∈ K }, a contradiction.
(3) It follows from the proof of (1) that ϕ : B → B(X ) is a Boolean
homomorphism, and reasoning as in the proof of (2) yields  that ϕ is injective. We
show that ϕ is surjective. If A ∈ F(X ), then by (1), A = {ϕ(a) : a ∈ K } for some
K ⊆ B. Since A is clopen and hence compact, there exists a finite subset K 0 ⊆ K
such that A = {ϕ(a) : a ∈ K 0 }. Therefore, A = ϕ(b) for some b ∈ B. It remains to
show that ϕ(Ca) = R −1 (ϕ(a)) for each a ∈ B. Recall that a ∈ x is
equivalent to x ∈ ϕ(a). The equality ϕ(Ca) = R −1 (ϕ(a)) can be written as
3.3 The Category of Hybrids and Hybrid Maps 57

(∀x ∈ X )(x ∈ ϕ(Ca) ⇔ x ∈ R −1 (ϕ(a))), which by Lemma 3.3.5 is equivalent to:

for each x ∈ X , we have Ca ∈ x ⇔ (∃y ∈ X ) (a ∈ y & x Ry). Let a ∈ y and
x Ry, i.e., (∀b ∈ B) (b ∈ y ⇒ Cb ∈ x). Therefore, Ca ∈ x. Now let Ca ∈ x. We
show that there is an ultrafilter y ∈ X such that a ∈ y and (∀b) (Ib ∈ x ⇒ b ∈ y).
Consider the set K a = {a} ∪ {b : Ib ∈ x}. We claim that the filter generated by
K a is proper. Suppose a ∧ b1 ∧ · · · ∧ bk = 0 where Ibi ∈ x for each bi . Then
b1 ∧ · · · ∧ bk ≤ −a and hence I(b1 ∧ · · · ∧ bk ) ≤ I−a. Therefore, Ib1 ∧ · · · ∧ Ibk ≤
−Ca. Since Ib1 , . . . , Ibk ∈ x and x is a filter, it follows that −Ca ∈ x, which con-
tradicts Ca ∈ x. Now extend K a to an ultrafilter, say y. Clearly a ∈ y and K a ⊆ y.
Let Ib ∈ x. Then b ∈ K a , i.e., b ∈ y. The lemma is proved. 
3.3.8 Lemma If A is a clopen subset of X , then R −1 A is clopen.
Proof Let A be clopen in X . Then, by Lemma 3.3.7(3), A = ϕ(a) for some a ∈ B
and R −1 (ϕa) = ϕ(Ca). Thus, R −1 (ϕa) is clopen. 
As an immediate consequence of Lemma 3.3.7, we obtain the following.
3.3.9 Lemma The algebra (B(X ), ∪, ∩, −, ∅, X, R −1 ), where B(X ) is the field of
all clopen subsets of X , is a closure algebra isomorphic to (B, ∨, ∧, −, 0, 1, C).
In what follows we call the algebra (B(X ), ∪, ∩, −, ∅, X, R −1 ) the algebra of
clopen sets of (X, , R).
3.3.10 Lemma For each x ∈ X , the set R(x) is closed.
Proof Let y ∈ X belong to the closure CR(x) of R(x) in X . Then for each clopen
set A, from y ∈ A it follows that R(x) ∩ A = ∅. We show that y ∈ R(x). Suppose
y∈ / R(x). By Lemmas 3.3.5 and 3.3.7, y ∈ / R(x) is equivalent to the existence of a
clopen set A such that y ∈ A and x ∈/ R −1 (A). Since y ∈ R −1 (A) by the reflexivity of
R, and x ∈/ R −1 (A), we have R −1 (A) ∩ R(x) = ∅. Indeed, if z ∈ R −1 (A) ∩ R(x),
then z ∈ R −1 (A), and as x Rz, we have x ∈ R −1 (A) by the transitivity of R,
a contradiction. Since A is clopen, by Lemma 3.3.8, so is R −1 (A). Therefore, there
is a clopen set A0 = R −1 (A) such that y ∈ A0 and R(x) ∩ A0 = ∅, i.e., y ∈ / CA0 , a
contradiction. Thus, R(x) is closed for each x ∈ X . 
Lemmas 3.3.9 and 3.3.10 together yield the following.
3.3.11 Lemma (X, , R) is a hybrid, and the closure algebra (B, ∨, ∧, −, 0, 1, C)
is isomorphic to the algebra of clopen sets (B(X ), ∪, ∩, −, ∅, X, R −1 ) of the hybrid
(X, , R).
3.3.12 Lemma Let (X, , R) be an arbitrary hybrid and F a downward directed
(with respect to inclusion) family of nonempty closed subsets of (X, ). Then

R −1 {F : F ∈ F} = {R −1 F : F ∈ F}.11

11 Editorial note: This result is now known as Esakia’s lemma; see, e.g., p. 350 of A. Chagrov and

M. Zakharyaschev, Modal Logic (Clarendon Press), 1997. Note that the lemma does not require
that R be reflexive or transitive.
58 3 Duality Theory: Hybrids

Proof The equality in the lemma is equivalent to the following: for each x ∈ X ,

x ∈ R −1 {F : F ∈ F} ⇔ (∀F ∈ F) (x ∈ R −1 F)

or, in other words,

R(x)∩ {F : F ∈ F} = ∅ ⇔ (∀F ∈ F) (R(x) ∩ F = ∅).

Obviously if R(x) ∩ {F : F ∈ F} = ∅, then R(x) ∩ F = ∅. Conversely, suppose

R(x) ∩ F = ∅ for each F ∈ F. Since F is closed and R(x) is closed for each x ∈ X
(Theorem 3.1.2(IIa)), the set R(x) ∩ F is closed. As F is downward directed, so
is the family {R(x) ∩ F : F ∈ F}. Compactness of the space (X, ) implies that
{R(x) ∩ F : F ∈ F} is nonempty, hence R(x) ∩ {F : F ∈ F} = ∅. 

3.3.13 Lemma Let (Bi , Ci ) ∈ CA, h : B1 → B2 be a morphism in the category CA,

(X i , i , Ri ) be the hybrid corresponding to (Bi , Ci ) (i = 1, 2), and f = h ∗ = h −1 .
Then:

(1) f maps the set X 2 to X 1 ;

(2) f is a continuous map from (X 2 , 2 ) to (X 1 , 1 );
(3) f is injective (resp. surjective) if and only if h is surjective (resp. injective);
(4) f is a strongly isotone map from (X 2 , R2 ) to (X 1 , R1 ).

Proof (1) Let x ∈ X 2 . We will show that f (x) ∈ X 1 , i.e., that f (x) is an ultrafil-
ter of B1 . First let a, b ∈ f (x) = h −1 (x), i.e., ha, hb ∈ x. Then ha ∧ hb ∈ x. But
ha ∧ hb = h(a ∧ b), i.e., a ∧ b ∈ h −1 (x) = f (x). Next let a ∈ f (x) and
a ≤ b ∈ B2 . Then ha ∈ x and ha ≤ hb. Therefore, hb ∈ x, i.e., b ∈ h −1 (x) = f (x).
Lastly let a ∨ b ∈ f (x), i.e., h(a ∨ b) ∈ x. Then ha ∨ hb ∈ x, and since x is an ultra-
filter and hence a prime filter, ha ∈ x or hb ∈ x, i.e., a ∈ f (x) or b ∈ f (x).
(2) We show that f −1 (A) is open in (X 2 , 2 ) for each clopen subset A of X 1 .
Since A = ϕ(a) for some a ∈ B1 ,

f −1 (A) = f −1 (ϕa) = {x ∈ X 2 : f (x) ∈ ϕ(a)} = {x ∈ X 2 : a ∈ f (x)}

= {x ∈ X 2 : ha ∈ x} = ϕ(ha) ∈ B(X 2 ).

(3) Let f be injective. If h is not surjective, then there exists a ∈ B2 − h(B1 ).

As h(B1 ) is a subalgebra of B2 , by Makinson’s lemma [11], there exist x, y ∈ X 2
such that a ∈ x, a ∈ / y, and x ∩ h(B1 ) = y ∩ h(B1 ). Therefore, x = y and f (x) =
f (y). Thus, f is not injective, a contradiction. Now let h be surjective. We show
that f (x) = f (y) implies x = y. Suppose x = y. Say a ∈ x and a ∈ / y. Since h
is surjective, h −1 (a) = ∅. Let b ∈ h −1 (a). Then hb = a and so b ∈ f (x). But
hb = a ∈ / y. Therefore, b ∈/ f (y).
Next, let f be surjective. To show that h is injective, let a, b ∈ B1 with a =
b. Without loss of generality we may assume that a ≤ b. Therefore, there exists
x ∈ X 1 such that a ∈ x and b ∈ / x. Since f is surjective, there exists y ∈ X 2 such
that x = f (y). Thus, ha ∈ y and hb ∈ / y, so ha = hb, and hence h is injective. Now
3.3 The Category of Hybrids and Hybrid Maps 59

let h be injective. We show that f is surjective, i.e., that for each x ∈ X 1 there exists
y ∈ X 2 such that f (y) = x, i.e., x = {a ∈ B1 : ha ∈ y}. Let F be the filter of B2
generated by h(x). Since h is injective, F is proper, so there is a y ∈ X 2 extending
F. It is easy to check that {a ∈ B1 : ha ∈ y} = x.
(4) We show that f is strongly isotone, i.e., that f −1 R1−1 (x) = R2−1 f −1 (x) for
x ∈ X 1 (see Proposition 1.4.12). Since X 1 is a Stone space and {ϕ(a) : a ∈ B1 } is
a basis of clopen sets, {x} = {ϕ(a) : a ∈ x}, and {ϕ(a) : a ∈ x} is a downward
directed family of nonempty clopen sets in X 1 . Therefore, the left-hand side of the
equality can be rewritten as follows:

f −1 R1−1 (x) = f −1 R1−1 {ϕ(a) : a ∈ x} = f −1 {R1−1 ϕ(a) : a ∈ x}

= f −1 {ϕ(C1 a) : a ∈ x} = { f −1 ϕ(C1 a) : a ∈ x}

= {ϕ(hC1 a) : a ∈ x},

where we have used Lemmas 3.3.12 and 3.3.7(3) and the proof of (2). The right-hand
side can be rewritten as follows:

R2−1 f −1 (x) = R2−1 f −1 {ϕ(a) : a ∈ x} = R2−1 { f −1 ϕ(a) : a ∈ x}

= R2−1 {ϕ(ha) : a ∈ x} = {R2−1 ϕ(ha) : a ∈ x}

= {ϕ(C2 ha) : a ∈ x},

where again we have used Lemmas 3.3.12 and 3.3.7(3) and the proof of (2). Since h is
a closure algebra homomorphism, hC1 a = C2 ha. Thus, f −1 R1−1 (x) = R2−1 f −1 (x).

3.3.14 Lemma Let (X i , i , Ri ) ∈ HYB, (Bi , Ci ) = (X i , i , Ri )∗ be the corre-
sponding closure algebra (i = 1, 2),12 f : X 1 → X 2 a continuous strongly isotone
map, and h = f ∗ = f −1 . Then:
(1) h maps B2 to B1 ;
(2) h is a morphism of the category CA;
(3) h is injective (resp. surjective) if and only if f is surjective (resp. injective).
Proof (1) Since the f -preimage of a clopen set is clopen, h = f −1 maps B2 to B1 .
(2) Since f −1 (A ∪ B) = f −1 (A) ∪ f −1 (B) and f −1 (−A) = − f −1 (A), we have
that h is a Boolean homomorphism. We prove that h preserves the closure operator,
i.e., that f −1 (R2−1 A)= R1−1 f −1 (A) for each clopen
 set A of (X 2 , 2 ). This equa-
tion is equivalent to { f −1 R2−1 (x) : x ∈ A} = {R1−1 f −1 (x) : x ∈ A}. The latter
equation follows from the fact that f is strongly isotone (see Proposition 1.4.12).
(3) Easy to verify.13 

12 Editorial note: It is straightforward to see that (X ∗ −1

i , i , Ri ) = (B(X i ), Ri ) is indeed a closure
algebra.
13 Editorial note: Indeed, it is easy to see that f being surjective implies that f −1 is injective. For

the converse, given y ∈ Y , we have that {y} is the intersection of the downward directed family of
60 3 Duality Theory: Hybrids

3.3.15 Lemma The map ψ : X → B(X )∗ is:

(1) a homeomorphism of the spaces (X, ) and (X 0 , 0 ) = (X, )∗ ∗ ;
(2) an isomorphism of the quasi-ordered sets (X, R) and (X 0 , R0 ) = (X, R)∗ ∗ .
Proof (1) First we show that ψ is surjective, i.e., that x0 ∈ X 0 = X ∗ ∗ implies
ψ −1 (x0 ) = ∅. Since x0 is an ultrafilter of B(X ) = X ∗ , the family
{A ∈ B(X ) : A ∈ x0 } of clopen sets has the finite intersection property. Then since X
is compact by Lemma 3.3.7(2), there exists x such that x ∈ {A ∈ B(X ) : A ∈ x0 }.
But then x0 = ψ(x). Next we show that ψ is injective. Indeed, if x = y, then since
X is Hausdorff and zero-dimensional by Lemma 3.3.7(2), we have ψ(x) = ψ(y).
Finally, ψ is continuous, i.e., the ψ-preimage of each element of a basis of B(X )∗
is open, i.e., ψ −1 (ϕA) is open for A ∈ B(X ) where ϕ(A) = ϕ B(X ) (A). Indeed,
y ∈ ψ −1 (ϕA) ⇔ ψ(y) ∈ ϕ(A) ⇔ A ∈ ψ(y) ⇔ y ∈ A. Therefore, ψ −1 (ϕA) = A,
and the continuity of ψ has been proved. Since X and X ∗ ∗ are compact Hausdorff
spaces, ψ is a homeomorphism.
(2) We show that ψ is an isomorphism of the quasi-ordered sets (X, R) and
(X 0 , R0 ) = (X, R)∗ ∗ . Note first that if x Ry, then for each clopen subset A of
(X, ), from y ∈ A it follows that x ∈ R −1 (A). The converse is also true. Indeed, if
y∈ / R(x), then y ∈ −R(x). Since R(x) is closed, −R(x) is open. As (X, ) is zero-
dimensional, there exists a clopen set A such that y ∈ A ⊆ −R(x), so x ∈ / R −1 A.
−1
Therefore, x Ry is equivalent to the following: y ∈ A implies x ∈ R A for each
clopen A. (From this, in particular, it follows that if (X, , R1 ) and (X, , R2 ) are
hybrids with the same topological component (X, ), and R1−1 (A) = R2−1 (A) for
each clopen A, then R1 = R2 .)
We are ready to show that x Ry ⇔ ψ(x)R0 ψ(y). By the definition of the relation
R0 on X 0 , we have: ψ(x)R0 ψ(y) ⇔ (A ∈ ψ(y) ⇒ R −1 A ∈ ψ(x)) for each clopen
A ∈ B(X ). By the definition of ψ, the right-hand side is equivalent to the implication
y ∈ A ⇒ x ∈ R −1 (A) for each A ∈ B(X ). This, in turn, is equivalent to x Ry by the
previous paragraph. Therefore, x Ry ⇔ ψ(x)R0 ψ(y). Since ψ is a bijection, ψ is an
isomorphism by Proposition 1.4.15. The lemma is proved. 
The Duality Theorem is now proved.14
Bibliographic notes. The Duality Theorem was first published in [4]. See also [5].

clopens Ai containing y. Since each Ai is nonempty, and f −1 is injective, f −1 (Ai ) is nonempty.

Thus, by compactness, the intersection of the f −1 (Ai ) is nonempty, so f −1 ({y}) is nonempty. This
shows that f is surjective.
To see that f −1 being surjective implies that f is injective, use the fact that distinct points
are distinguished by clopens. For the converse, suppose A is a clopen subset of X . Then f (A) is
closed in Y , so f (A) is the intersection of clopens Bi such that f (A) ⊆ Bi . Thus, f −1 ( f (A)) =
{ f −1 (Bi ) | i ∈ I }, and since f is injective, A = f −1 ( f (A)), so A = { f −1 (Bi ) | i ∈ I }. Using
compactness, it follows that there is one clopen B such that A = f −1 (B).
Note that Lemmas 3.3.13(3) and 3.3.14(3) are inter-derivable using the Duality Theorem (the
proof of which does not require these lemmas).
14 Editorial note: That (·) is a well-defined contravariant functor follows from Lemmas 3.3.11 and
∗
3.3.13 (and the easily verifiable facts that it preserves identity and composition), and that (·)∗ is
a well-defined contravariant functor follows from Lemma 3.3.14 (and the same easily verifiable
facts). That (·)∗ satisfies part (a) of the definition of coequivalence (Definition 1.2.2(2)) follows
from Lemma 3.3.15. That (·)∗ satisfies part (b) is a consequence of Lemmas 3.3.13 and 3.3.14
3.4 The Category of Heyting Algebras and the Category of Strict Hybrids 61

3.4 The Category of Heyting Algebras and the Category

of Strict Hybrids

Recall that a ring of cones over a quasi-ordered set is, by definition, a ring of sets K
such that A1 , A2 ∈ K imply {x : R(x) ∩ (A1 − A2 ) = ∅} ∈ K (Definition 2.2.9).
Let (X, , R) be a hybrid and (B(X ), ∪, ∩, −, ∅, X, R −1 ) the algebra of clopen
sets of the hybrid (X, , R) (Lemma 3.3.9). Let H(X ) denote the set of all clopen
cones of (X, , R).

3.4.1 Proposition
(1) The set H(X ) is a ring of cones over the quasi-ordered set (X, R);
(2) A ∈ H(X ) if and only if A = −R −1 (A0 ) for some clopen set A0 ∈ B(X ).

Proof (1) Obviously ∅, X ∈ H(X ), and if A1 , A2 ∈ H(X ), then A1 ∪ A2 ,

A1 ∩ A2 ∈ H(X ). Let A1 , A2 ∈ H(X ). We will show that the set {x ∈ X :
R(x) ∩(A1 − A2 ) = ∅} belongs to H(X ). Since A1 and A2 are clopen sets of (X, ),
A1 − A2 is clopen, and hence R −1 (A1 − A2 ) is clopen by Theorem 3.1.2(IIIb).
Since R −1 (A1 − A2 ) is a lower cone, −R −1 (A1 − A2 ) is a clopen upper cone,
and so −R −1 (A1 − A2 ) ∈ H(X ). As −R −1 (A1 − A2 ) = {x ∈ X :
R(x) ∩ (A1 − A2 ) = ∅}, the set {x : R(x) ∩ (A1 − A2 ) = ∅} belongs to H(X ).
(2) follows from the proof of (1). 

3.4.2 Corollary Let (X, , R) be a hybrid. Then the set H(X ) of all clopen cones
of (X, , R) forms a Heyting algebra which is the skeleton of the algebra of clopen
sets (B(X ), R −1 ) of the hybrid.

Proof By Proposition 3.4.1(1), (H(X ), ∪, ∩, →, ∅, X ) is a Heyting algebra, where

A1 → A2 = {x : R(x) ∩ (A1 − A2 ) = ∅} for A1 , A2 ∈ H(X ). By Proposition 3.4.1
(2), H(X ) is the skeleton of the algebra. 

We call the Heyting algebra H(X ) the ring of clopen cones of the hybrid
(X, , R).

3.4.3 Convention Based on the Duality Theorem (Theorem 3.3.4), we will often
conveniently identify the closure algebra (B, C) with the algebra of clopen sets
(B(X ), R −1 ) of the corresponding hybrid (X, , R) and its skeleton H with the ring
of clopen cones H(X ) of the hybrid.

3.4.4 Theorem A closure algebra (B, C) is a skeletal closure algebra if and only if
the corresponding hybrid (X, , R) is strict (i.e., (X, R) is a partially ordered set).

Proof Use Convention 3.4.3 and identify (B, C) with (B(X ), R −1 ) and its skeleton
H with H(X ). Let (B(X ), R −1 ) be a skeletal closure algebra. Then the field of clopen
sets B(X ) is the smallest field containing the ring of clopen cones H(X ). In other
words, each clopen A ∈ B(X ) can be obtained from finitely many clopen cones
from H(X ) by applying the set-theoretic operations of union ∪, intersection ∩, and
62 3 Duality Theory: Hybrids

complement −. To show that R is anti-symmetric, i.e., that there are no nontrivial

clusters in (X, R), suppose there are x, y ∈ X such that x = y, x Ry, and y Rx. Since
x = y, there exists a clopen set A such that x ∈ A and y ∈ / A. Clearly A cannot be
either an upper or a lower cone since x Ry and y Rx. The clopen A cannot be obtained
by applying ∪, ∩, and − to a finite number of clopen cones because every clopen
cone A0 has the property that either x, y ∈ A0 or x, y ∈ / A0 , and the operations ∪,
∩, − preserve this property. So A ∈ / B(X ), contradicting the fact that (B, C) is a
skeletal algebra.
Now let (X, R) be a partially ordered set. Then for each x, y ∈ X , from x = y it
follows that either x ∈/ R(y) or y ∈
/ R(x). Without loss of generality we may assume
that x ∈/ R(y). Then x ∈ −R(y). Since R(y) is closed in the space (X, ), the set
−R(y) is open. As (X, ) is zero-dimensional, there exists a clopen set A such
that x ∈ A ⊆ −R(y). Since −R(y) is a lower cone, R −1 (A) ⊆ −R(y). Moreover,
x ∈ R −1 (A), y ∈/ R −1 (A), and as A is clopen, by Theorem 3.1.2(IIIb), R −1 (A) is
clopen. Therefore, x ∈ / −R −1 (A) and y ∈ −R −1 (A). Thus, the ring of clopen cones
separates any two distinct points of the space (X, ), i.e., it is reduced. Hence, the
field F generated by it is reduced. Since (X, ) is a compact Hausdorff space and
F is a reduced field of clopen sets, by Proposition 1.3.2, the field coincides with the
field B(X ) of all clopen sets of (X, , R). 
3.4.5 Construction Let (B, C) be a skeletal closure algebra, (X, , R) the corre-
sponding strict hybrid, and H the skeleton of the algebra. Let X 0 denote the set of all
prime filters of the Heyting algebra H . Define a relation R0 on X 0 by x R0 y ⇔ x ⊆ y
for each x, y ∈ X 0 . Introduce a topology 0 on the set by declaring the family
F0 = {ϕ(a) : a ∈ H } ∪ {−ϕ(a) : a ∈ H }, where ϕ(a) = {x ∈ X 0 : a ∈ x}, as a sub-
basis of open sets. 15
That the two constructions coincide is stated in the following.
3.4.6 Theorem
(1) The space (X, ) is homeomorphic to the space (X 0 , 0 ).
(2) The partially ordered set (X, R) is isomorphic to the partially ordered set
(X 0 , R0 ).
Proof Define a map f : X → X 0 by setting f (x) = x ∩ H for each x ∈ X .
(1) To show that f is injective, suppose f (x) = f (y), i.e., x ∩ H = y ∩ H ,
and x = y. From x = y it follows that there is a ∈ B such that a ∈ x and a ∈ / y.
Since x, yare ultrafilters, −a ∈
/ x and −a ∈ y. As (B, C) is a skeletal alge-
bra, a = (ak ∧ −bk ) for some m ∈ ω, ak , bk ∈ H . Since x is an ultrafilter and
k<m
k0 − bk0 ∈ x for some k0 < m, i.e. ak0 ∈ x and bk0 ∈
a ∈ x, a / x. On the other hand,
−a = −(ak ∧ −bk ) ∈ y, and as y is an ultrafilter, for all k < m, we have
k<m
−(ak ∧ −bk ) ∈ y, i.e., ak ∈
/ y or bk ∈ y. So ak0 ∈ x, bk0 ∈/ x, and (∀k < m)
(ak ∈ y ⇒ bk ∈ y). Since ak , bk ∈ H for all k < m, x ∩ H = y ∩ H , and ak0 ∈ x,

15 Editorial note: It is easy to see that (X 0 , 0 ) is a zero-dimensional Hausdorff space.

3.4 The Category of Heyting Algebras and the Category of Strict Hybrids 63

we have ak0 ∈ y. Therefore, bk0 ∈ y, and as x ∩ H = y ∩ H , we arrive at a contra-

diction: bk0 ∈ x and bk0 ∈ / x. Thus, f is injective.
Next we show that f is surjective. Let p be a prime filter of the Heyting algebra H .
To prove that there exists an ultrafilter x of B such that f (x) = p, consider the family
F p of all proper filters F of the Boolean algebra B such that F ∩ H = p. The family
F p is nonempty because the skeletal filter

F p = {a ∈ B : (∃b ∈ p) (b ≤ a)}

generated by the prime filter p (see Proposition 2.4.12) is proper, and F p ∩ H =

p, so F p ∈ F p . Let K be a nonempty chain in F p . Then F0  = {F : F ∈ K} is a
proper
 filter of the Boolean algebra B, and F0 ∩ H = p since {F : F ∈ K} ∩ H =
{F ∩ H : F ∈ K} = p. Hence, by Zorn’s lemma,16 F p has a maximal element,
say x. To show that x is an ultrafilter of the Boolean algebra B, suppose a ∨ b ∈ x
and a, b ∈ / x for some a, b ∈ B. Since x is a maximal element of F p , we have
the following for the filter Fa = [x, a] (resp. Fb = [x, b]) generated by the filter
x and the element a (resp. b) : (Fa ∩ H ) − p = ∅(resp. (Fb ∩ H ) − p = ∅). Let
d1 ∈ (Fa ∩ H ) − p and d2 ∈ (Fb ∩ H ) − p. Then a ∧ a1 ≤ d1 and b ∧ a2 ≤ d2 for
some a1 , a2 ∈ x. Therefore, (a ∧ a1 ) ∨ (b ∧ a2 ) ≤ d1 ∨ d2 . But

(a ∧ a1 ) ∨ (b ∧ a2 ) = (a ∨ b) ∧ (a ∨ a2 ) ∧ (a1 ∨ b) ∧ (a1 ∨ a2 ).

We have a ∨ b ∈ x; a ∨ a2 ∈ x since a2 ∈ x; a1 ∨ b ∈ x since a1 ∈ x; and

a1 ∨ a2 ∈ x. Thus, d1 ∨ d2 ∈ x. But d1 , d2 ∈ H , so d1 ∨ d2 ∈ p, a contradiction.
Consequently, x is an ultrafilter of B, and f is surjective.
Now we show that the bijection f is continuous. Let A be a subbasic clopen
subset of X 0 . Then either (I) A = { p ∈ X 0 : a ∈ p}, where a ∈ H , or (II) A =
{ p ∈ X0 : a ∈
/ p}.
Case I: We have f −1 (A) = {x ∈ X : f (x) ∈ A} = {x ∈ X : a ∈ f (x)}
= {x ∈ X : a ∈ x ∩ H }. But since a ∈ H, we have {x ∈ X :
a ∈ x ∩ H } = {x : a ∈ x}, and {x ∈ X : a ∈ x} is clopen in X .
Case II: f −1 (A) = {x ∈ X : f (x) ∈ A} = {x : f (x) ∈ { p ∈ X 0 : a ∈ / p}} =
{x ∈ X : a ∈/ f (x)} = {x ∈ X : a ∈ / x ∩ H }. But as a ∈ H , we have {x ∈ X :
a∈ / x ∩ H } = {x ∈ X : a ∈/ x}, and {x ∈ X : −a ∈ x} is clopen in X . Since f is
a continuous bijection of the compact space X onto the Hausdorff space X 0 , f is a
homeomorphism.
(2) We show that the bijection f is an isomorphism of (X, R) onto (X 0 , R0 ).
By Proposition 1.4.15, it is sufficient to prove that x Ry ⇔ f (x)R0 f (y), i.e., that
x Ry ⇔ x ∩ H ⊆ y ∩ H . But the latter follows from Lemma 3.3.5. The lemma is
proved. 

3.4.7 Remark Together the Duality Theorem (Theorem 3.3.4) and Theorem 3.4.6
establish Theorem 2.5.11, stating that the category HA of Heyting algebras and the
category SA of skeletal closure algebras are equivalent.

16 Editorial note: Zorn’s lemma states that a partially ordered set in which every chain has an upper
bound has a maximal element (see, e.g., [11]).
64 3 Duality Theory: Hybrids

We now have the following important corollary.

3.4.8 Corollary

(1) The category SA of skeletal closure algebras is dual to the category HYB+ of
strict hybrids.
(2) The category HA of Heyting algebras is dual to the category HYB+ of strict
hybrids.

Proof (1) Theorems 3.3.4 and 3.4.4.

(2) Theorems 3.3.4, 3.4.4, and 3.4.6. 

3.4.9 Corollary Let H be a Heyting algebra and (X, , R) the corresponding strict
hybrid. Then H is a Boolean algebra if and only if R is the identity relation.

Using Theorem 3.3.4, we prove the following useful lemma generalizing a fact
known for Boolean algebras.

3.4.10 Lemma Let H be a Heyting algebra, H0 a subalgebra of H , and a ∈ H . The

following are equivalent:

(1) a ∈
/ H0 ;
(2) there exist prime filters x, x  of H such that a ∈ x, a ∈
/ x  , and x ∩ H0 = x  ∩ H0 .

Proof (2 ⇒ 1) is obvious. We prove (1 ⇒ 2). Let H0 be a subalgebra of H ,

h : H0 → H the inclusion map, and f = h ∗ : X → X 0 the map of the corresponding
hybrids (X, , R) and (X 0 , 0 , R0 ) dual to h. Since h is injective, f is a surjective
hybrid map.17 Consider the set f (ϕ(a)) = A0 ⊆ X 0 , where ϕ(a) = {x ∈ X : a ∈ x}
is a clopen cone of (X, R). Since f is a hybrid map, f is continuous, and so f is a
closed map preserving cones (Proposition 1.4.12). Hence, f (ϕ(a)) = A0 is a closed
cone. If f −1 (A0 ) = ϕ(a), then (as f is surjective) f (X − ϕ(a)) = X 0 − A0 . Since
X − ϕ(a) is clopen, X 0 − A0 is closed in the space (X 0 , 0 ). Therefore, A0 is a
clopen cone, and so A0 = ϕ(a0 ) for some a0 ∈ H0 . Thus, h(a0 ) = a, a contradic-
tion. It follows that ϕ(a) ⊆ f −1 (A0 ) and f −1 (A0 ) − ϕ(a) = ∅, i.e., there exists
x  ∈ f −1 (A0 ) such that x  ∈
/ ϕ(a). But f (x  ) ∈ A0 = f (ϕ(a)), so there exists
x ∈ ϕ(a) such that f (x) = f (x  ); x and x  are the prime filters we sought. 

3.4.11 Lemma Let (X, , R) be a hybrid and X 0 a closed cone of the hybrid. Let
0 be the subspace topology on X 0 and R0 the restriction of R to X 0 , i.e., R0 (x) =
R(x) ∩ X 0 for all x ∈ X 0 . Then (X 0 , 0 , R0 ) is a hybrid.

Proof Since X 0 is closed in (X, R), (X 0 , 0 ) is a Stone space. Clearly (X 0 , R0 ) is

a quasi-ordered set. For each x ∈ X , R0 (x) = R(x) ∩ X 0 . Hence, R0 (x) is closed in
(X 0 , 0 ). So (IIa) of Theorem 3.1.2 holds. Let A0 = A ∩ X 0 , where A is an arbitrary
subset of X . We have

17 Editorial note: The proof is similar to that of the analogous part of Lemma 3.3.13(3).
3.4 The Category of Heyting Algebras and the Category of Strict Hybrids 65

R0−1 (A0 ) = R0−1 (A ∩ X 0 ) = {R0−1 (x) : x ∈ A ∩ X 0 }

= {R −1 (x) ∩ X 0 : x ∈ A ∩ X 0 }

= X0 ∩ {R −1 (x) : x ∈ A ∩ X 0 } = X 0 ∩ R −1 (A ∩ X 0 )
= X 0 ∩ R −1 (A).

Notice that the fact that X 0 is a cone is needed only for the last equality. Indeed,
we show that X 0 ∩ R −1 (A) ⊆ X 0 ∩ R −1 (A ∩ X 0 ). The reverse inclusion is obvious.
Let x ∈ X 0 ∩ R −1 (A). Then x ∈ X 0 and R(x) ∩ A = ∅. Let y ∈ R(x) ∩ A. Then
x Ry. Since x ∈ X 0 and X 0 is a cone, y ∈ X 0 . Therefore, y ∈ R(x) ∩ A ∩ X 0 , so
R(x) ∩ A ∩ X 0 = ∅. Thus, x ∈ X 0 ∩ R −1 (A ∩ X 0 ). Now let A0 be a closed subset
of the space (X 0 , 0 ). Then since X 0 is closed in (X, ), A0 = A ∩ X for some
set A closed in (X, ). But R0−1 (A0 ) = X 0 ∩ R −1 (A), and since R −1 (A) is closed
in (X, ), the set R0−1 (A0 ) is closed in (X 0 , 0 ). A similar argument shows that
if U0 is open in X 0 , then R0−1 (U0 ) is open in X 0 . Consequently, condition (IIa) of
Theorem 3.1.2 holds, and so (X 0 , 0 , R0 ) is a hybrid. 
3.4.12 Definition The hybrid (X 0 , 0 , R0 ), where X 0 is a closed cone of a hybrid
(X, , R), is called a subhybrid of (X, , R).
Recall that if (X, ) is an arbitrary topological space, where  is the family of
all its open sets, E is an equivalence relation on the set X , and f : X → X/E is the
canonical map ( f (x) = E(x), x ∈ X ), then the set X/ E equipped with the quotient
topology  E = {A ∈ X/E : f −1 (A) ∈ } is said to be a quotient space of (X, ).
3.4.13 Lemma Let (X, , R) be a hybrid, (X E , R/E) the skeleton of the quasi-
ordered set (X, R), and (X/E,  E ) the quotient space. Then (X/E,  E , R E ) is a
strict hybrid, and the canonical map f : X → X/E is a hybrid map.
Proof Obviously f is a continuous map from (X, ) onto (X/E,  E ) and, by Propo-
sition 1.4.13, f is a strongly isotone map from (X, R) onto (X/E, R E ). Note that
since (X/E, R E ) is the skeleton of (X, R), it follows that (X/E, R E ) is a partially
ordered set (see the discussion after Definition 1.4.7).
It remains to show that (X/E,  E , R E ) is a hybrid. Clearly (X/E,  E ) is com-
pact. We show that it is Hausdorff and zero-dimensional. Let f (x) and f (y) be two
distinct points of X/E. Then E(x) = E(y), i.e., x ∈ / E(y). Thus, either y ∈ / R(x)
or x ∈ / R(y). Without loss of generality we may assume that y ∈ / R(x). By Theo-
rem 3.1.2(IVa), there is a partition of X into two open sets A1 and A2 such that x ∈ A1 ,
y ∈ A2 , A1 is an upper cone, and A2 is a lower cone. Clearly f (x) ∈ f (A1 ) and
f (y) ∈ f (A2 ), i.e., E(x) ⊆ E(A1 ) and E(y) ⊆ E(A2 ). Moreover, E(y) ⊆ E(A1 )
and E(x) ⊆ E(A2 ) since A1 and A2 are cones. For the same reason, E(A1 ) = A1
and E(A2 ) = A2 . Thus, f (x) and f (y) are separated by disjoint clopen sets f (A1 )
and f (A2 ).18 So (X/E,  E ) is a Stone space.

18 Editorial note: Since (X/E, 

E ) is compact and distinct points are separated by clopens, it follows
that (X/E,  E ) is zero-dimensional. See, e.g., p. 69 of P. Johnstone, Stone Spaces (Cambridge
University Press), 1982.
66 3 Duality Theory: Hybrids

Now let E(x) = f (x) be a point in X/E. We show that R E (E(x)) is closed in
(X/E,  E ). We have

R E (E(x)) = {E(y) : E(x)R E E(y)} = {E(y) : y ∈ R(x)}.


But E(R(x)) = {E(y) : y ∈ R(x)} = R(R(x) ∩ R −1 (x)). Since the latter set is
closed in (X, ), the set {E(y) : y ∈ R(x)} = R E (E(x)) is closed by the definition of
the quotient topology. Finally, let K = {E(x) : x ∈ A} be a clopen set in (X/E,  E ).
We show that R −1E (K) is clopen. We have

R −1
E (K) = {R −1
E (E(x)) : x ∈ A}

= {{E(y) : E(y) ∈ R −1
E (E(x))} : x ∈ A}

= {E(y) : y ∈ R −1 (A)} = E(R −1 (A)) = R −1 (A).

Since K is clopen in X/E, by the definition of the quotient topology we may assume
that A is clopen in X . Then R −1 (A) is clopen in X by Theorem 3.1.2(IIIb). Therefore,
R −1
E (K) is clopen in X/E. Thus, (X/E,  E , R E ) is a hybrid, and the canonical map
f : X → X/E is a hybrid map. 
3.4.14 Definition If (X, , R) is a hybrid and E is the equivalence relation on X
defined by
x E y ⇔ x Ry & y Rx,

then the strict hybrid (X/E,  E , R E ) is called the skeleton of the hybrid (X, , R).
3.4.15 Proposition Let (X/E,  E , R E ) be the skeleton of a hybrid (X, , R),
(B(X ), R −1 ) and (B(X/E), R −1
E ) the corresponding algebras of clopen sets, and
H(X ) and H(X/E) the corresponding algebras of clopen cones of these hybrids.
Then:
(1) the closure algebra (B(X/E), R −1 E ) is embeddable in the closure algebra
(B(X ), R −1 );
(2) the algebra (B(X/E), R −1
E ) is a skeletal algebra;
(3) the Heyting algebras H(X ) and H(X/E) are isomorphic.
Proof (1) Since the canonical map f : X → X/E is surjective, the dual hybrid map
h = f −1 : B(X/E) → B(X ) is an embedding by Lemma 3.3.14.
(2) That B(X/E) is a skeletal algebra follows from Theorem 3.4.4.
(3) For the map h from (1), we have that A ∈ H(X/E) implies h(A) ∈ H(X ).
Let h  : H(X/E) → H(X ) be the restriction of h to H(X/E). Then h  is the desired
isomorphism. 
Let (B, C) be a closure algebra, H its skeleton, and (H ) (resp. (B, C)) the
lattice of congruences of the Heyting algebra H (resp. of the closure algebra (B, C)).
Let (X, , R) be the hybrid corresponding to (B, C). Let (X, , R) (or simply
X ) denote the family of all closed cones of (X, , R) (recall Definition 3.4.12).
3.4 The Category of Heyting Algebras and the Category of Strict Hybrids 67

3.4.16 Theorem The following pairs of lattices are dually isomorphic:

(1) (B, C) and (X, ⊆);

(2) (H ) and (X, ⊆);
(3) (H ) and ((X/E), ⊆), where (X/E,  E , R E ) is the skeleton of the hybrid
(X, , R).

Proof By Theorem 2.4.17, F (H ) = ∼ F s (B, C) (see Definition 2.4.13). By Propo-

sition 2.4.9(b), (H ) ∼
= F (H ). By Corollary 2.4.20, (H ) ∼ = (B, C). Note that
X ∼ = (X/E) since the existence of clusters (irrespective of their cardinality)
does not affect the “stock” of cones. Indeed, if a cone A contains a point x, then
E(x) ⊆ A. Assign to a skeletal filter F of (B, C) the set A F in X as follows:
A F = {ϕ(Ia) : a ∈ F}, where ϕ is the isomorphism of B onto the algebra B(X )
of clopen sets of the hybrid X , i.e., ϕ(a) = {x ∈ X : a ∈ x} for a ∈ B. We have
ϕ(Ia) = −R −1 −ϕ(a) (recall Lemma 3.3.7(3)). Since ϕ(a) is clopen, −R −1 −ϕ(a)
is a clopen cone. Clearly A F is a closed cone (being an intersection of clopen cones),
that is, A F ∈ X . Obviously F ⊆ F  implies A F  ⊆ A F . It is easy to see that the
indicated correspondence establishes the required dual isomorphism. 

Recall that a principal cone (i.e., a cone of the form R(x), where x ∈ X ) is closed.
Note that the above correspondence establishes also a bijection between the set of
prime skeletal filters of (B, C) (as well as the set of prime filters of the Heyting
algebra H ) and the set of principal cones.
As byproducts of Theorem 3.3.4 (and Corollary 3.4.8), we obtain:

3.4.17 Theorem

(1) The category POF of all finite partially ordered sets is dual to the category FHA
of all finite Heyting algebras.
(2) The category QOF of all finite quasi-ordered sets is dual to the category FCA
of all finite closure algebras.

3.4.18 Representation Theorem Each Heyting algebra is isomorphic to a ring of

cones of a suitable partially ordered set.

3.4.19 Corollary Each finite Heyting algebra (and so each finite distributive lattice)
is isomorphic to the ring of all cones of some partially ordered set.

Bibliographic notes. Lemma 3.4.10 extends the well-known lemma of Makinson

[11] for Boolean algebras to Heyting algebras. Corollary 3.4.19 is well known (see,
e.g., Sec. III.3 of [1]).
68 3 Duality Theory: Hybrids

3.5 Grzegorczyk Algebras

Grzegorczyk’s modal system Grz is probably one of the most interesting systems (at
least among the modal systems in —see Definition 2.3.2). A. Grzegorczyk defined
the modal system Grz axiomatically and showed that the system is a modal com-
panion of intuitionistic logic, i.e., that for each formula p,

In  p ⇔ Grz  T ( p);

here T ( p) is the modal formula obtained by the well-known Gödel translation of p.19
It has been shown that the modal system Grz is the strongest modal companion of
intuitionistic logic. This allows one to construct the following “superintuitionistic”
classification of the family  (see Sect. 2.3) of all normal extensions of Lewis’s
classical modal system S4. For each superintuitionistic logic λ ∈ , let λ denote
the class of all modal companions of λ. That is, λ is the class of all modal systems
σ ∈  such that the superintuitionistic fragment of σ coincides with λ.20 The
classes λ will be called layers. Clearly the classes λ (λ ∈ ) partition  into
layers, which are in one-to-one correspondence with superintuitionistic logics. We
point out the following extreme cases: 21
(a) If λ = CL (classical two-valued logic), then the layer λ consists of all exten-
sions of Lewis’s system S5.22
(b) If λ = In (intuitionistic logic), then λ = {σ ∈  : S4 ⊆ σ ⊆ Grz}; so, intu-
itionistic logic can be embedded (by Gödel’s translation) into any modal system
situated between S4 and Grz.
Furthermore, every layer λ (λ ∈ ) is an infinite closed interval (of a lattice
) of the form [S4 + , Grz + ], i.e., λ = {σ ∈  : S4 +  ⊆ σ ⊆ Grz + },
where  is a set of modal formulae. An important feature of Grzegorczyk’s system is
contained in the following statement: the partially ordered set (, ⊆) is isomorphic
to the partially ordered set of all normal extensions of Grzegorczyk’s modal system
Grz.23
3.5.1 Definition (Grzegorczyk’s modal system) Grz is the normal extension of
Lewis’s system S4 obtained by postulating the formula

(g0 ) (( p →  p) → p) → p

19 Editorial note: The Gödel translation T is defined as follows: if p is a propositional vari-

able, let T ( p) =  p, and for arbitrary formulae p, q, let T ( p ∨ q) = T ( p) ∨ T (q), T ( p ∧ q) =
T ( p) ∧ T (q), T ( p → q) = (T ( p) → T (q)), and T (¬ p) = ¬T ( p).
20 Editorial note: The superintuitionistic fragment of σ is the superintuitionistic logic

{ p | T ( p) ∈ σ}.
21 Editorial note: For a proof, see Corollary 9.64 (and Exercise 9.21) in A. Chagrov and M.

Zakharyaschev, Modal Logic (Clarendon Press), 1997.

22 Editorial note: S5 is obtained by adding to S4 the axiom ♦ p → ♦ p.
23 Editorial note: This result is now known as the Blok–Esakia Theorem; see p. 325 of A. Chagrov

and M. Zakharyaschev, Modal Logic (Clarendon Press), 1997.

3.5 Grzegorczyk Algebras 69

as a new axiom; symbolically, Grz = S4 + g0 . More precisely, Grz is given by

Axiom schemas: A1. ( p → q) → ( p → q),

A2.  p → p,
A3.  p →  p,
A4. (( p →  p) → p) → p;
p, p → q
Inference Rules: (R1) ,
q
p
(R2) .
p

3.5.2 Theorem Grzegorczyk’s modal system Grz can be defined axiomatically as

follows: Grz = S4 + (R3), where

( p →  p) → p
(R3) .
p

In other words, Grz differs from S4 only by the presence of the additional rule of
inference (R3).

The proof of this theorem is based on the theory of hybrids and is deferred for a
bit (see Lemma 3.5.11).
Note that Grzegorczyk’s formula (g0 ) (( p →  p) → p) → p can be rewrit-
ten in terms of the modal operator of “possibility” ♦ (= ¬¬) as follows:

(g1 ) p → ♦( p & ¬♦(¬ p & ♦ p)).

Let (B, C) be a closure algebra. The algebraic polynomial corresponding to Grze-

gorczyk’s formula (g0 ), as is easy to verify, looks as follows:

−a ∨ C(a ∧ −C(−a ∧ Ca)) = 1,

or equivalently,
a ≤ C(a − C(Ca − a)).

3.5.3 Definition The closure algebra (B, C) is called a Grzegorczyk algebra if for
each a ∈ B,
(g2 ) a ≤ C(a − C(Ca − a)).

It is clear from the definition that the class of Grzegorczyk algebras is a variety.
We denote the class (and, as usual, the corresponding equational category) of Grze-
gorczyk algebras by GA. Clearly GA ⊆ CA and the category GA of Grzegorczyk
algebras is a full subcategory of the category CA of closure algebras.
Recall the notation of Sect. 2.5 (see Definition 2.5.12):
70 3 Duality Theory: Hybrids

ρa = a ∧ C(Ca − a) and πa = a − ρa.

In terms of the operator ρ, Grzegorczyk’s “inequality” can be written as follows:

(g3 ) a ≤ Cπa.

Let (B, C) be a Grzegorczyk algebra and (X, , R) the corresponding hybrid. We

use Convention 3.4.3 to identify (B, C) with the algebra of clopen sets (B(X ), R −1 )
of the hybrid (X, , R). Let A be a clopen set in (X, , R), i.e., A ∈ B(X ) = B.
Then π A = A − R −1 (R −1 A − A). In other words, for each x ∈ X ,

x ∈ π A ⇔ (x ∈ A & ¬(∃y)(x Ry & y ∈

/ A & (∃z)(y Rz & z ∈ A)).

3.5.4 Definition If A is clopen in X , then the points in π A are called passive points
of A, and the points in ρA are called active points of A.

This can be explained as follows. A point x ∈ A is a passive point of A (i.e.,

x ∈ π A) if and only if it is impossible, starting from x (see the right-hand side of the
above equivalence) to “leave” the set A and then “return” to A again. On the other
hand, a point x ∈ A is an active point of A if it is possible starting from x to leave A
and return to A again. Indeed,

x ∈ ρA ⇔ (x ∈ A & (∃y ∈ X )(x Ry & y ∈

/ A & (∃z ∈ X )(y Rz & z ∈ A)).

In terms of passive points, axiom (g3 ) can be written as follows: starting from
each point of a clopen set A, a passive point of A is reachable; more formally,

(g3 ) (∀x ∈ X )(x ∈ A ⇒ (∃y ∈ X )(x Ry & y ∈ π A)).

Consequently, we established the following.

3.5.5 Theorem Let (B, C) be a closure algebra and (X, , R) the corresponding
hybrid. Then (B, C) is a Grzegorczyk algebra if and only if for each clopen set A in
(X, , R), we have A ⊆ R −1 (π A) (i.e., the set of all passive points of A is cofinal
in A).

The following theorem is as convenient.

3.5.6 Theorem Let (B, C) be a closure algebra and (X, , R) the corresponding
hybrid. The following are equivalent:
(1) Every quasi-maximal point of each clopen A is maximal;
(2) (B, C) is a Grzegorczyk algebra.

Proof (1 ⇒ 2) Suppose (1) holds. Let us prove (g3 ). Let x ∈ A. By the remark after
Theorem 3.2.1 and Corollary 3.2.2, there is a quasi-maximal point y ∈ A such that
3.5 Grzegorczyk Algebras 71

x Ry. By (1), y is maximal. Clearly one cannot leave A and return to A starting from
y, for this would contradict the maximality of y in A.
(2 ⇒ 1) Suppose (1) is false, i.e., some clopen set A contains a quasi-maximal
point x ∈ A that is not maximal. This means that there exists y ∈ X such that x = y,
x Ry, and y Rx. Since x = y, there exists a clopen set A0 such that x ∈ A0 and y ∈/ A0 .
Let us show that (g3 ) is false for x ∈ A ∩ A0 . Indeed, x ∈ A ∩ A0 , and starting at x,
one can leave A ∩ A0 (i.e., x Ry and y ∈ / A ∩ A0 ) and return to A ∩ A0 again (y Rx
and x ∈ A ∩ A0 ). So, (B, C) is not a Grzegorczyk algebra. 

3.5.7 Corollary Let (B, C) ∈ CA and (X, , R) be the corresponding hybrid. Then
(B, C) is a Grzegorczyk algebra if and only if either of the following holds:

(1) for each clopen subset A of X , we have A ⊆ R −1 (max A);

(2) for each nonempty clopen set A, we have max A = ∅.

A close link between Grzegorczyk algebras and Heyting algebras is established

by the following.

3.5.8 Corollary If a hybrid (X, , R) is strict, then the corresponding algebra

(B, C) is a Grzegorczyk algebra.

3.5.9 Corollary All skeletal closure algebras belong to the variety of Grzegorczyk
algebras, i.e., SA ⊆ GA.

Recall that a closure algebra (B, C) is finite if and only if the corresponding hybrid
(X, , R) is a finite quasi-ordered set (X, R) with the discrete topology . In this
case we forget  (see Theorem 3.4.17(2)).

3.5.10 Corollary Let (B, C) be a finite closure algebra and (X, R) the correspond-
ing quasi-ordered set. The following are equivalent:

(1) (B, C) is a Grzegorczyk algebra;

(2) (B, C) is a skeletal algebra;
(3) (X, R) is a finite partially ordered set.

The following lemma is an algebraic formulation of Theorem 3.5.2.24

3.5.11 Lemma Let (B, C) be a closure algebra. The following are equivalent:

(1) (B, C) is a Grzegorczyk algebra;

(2) for each a ∈ B, if a = 0, then πa = 0;
(3) for each a ∈ B, if a = 0, then ρa = a;
(4) there are no cyclic elements in (B, C).

24 Editorial note: Lemma 3.5.11 is an algebraic formulation of Theorem 3.5.2 provided that the rule
(R3) is reformulated in terms of ♦.
72 3 Duality Theory: Hybrids

Proof (1 ⇒ 2) Let (B, C) be a Grzegorczyk algebra, i.e., assume that for all a ∈ B, we
have a ≤ C(a − (Ca − a)). Let a = 0. Suppose πa = a − C(Ca − a) = 0. Then
Cπa = 0 and 0 = a ≤ Cπa = 0, a contradiction. Hence, πa = 0.
(2 ⇒ 1) Identify the algebra (B, C) with the algebra of clopen sets of its hybrid
(X, , R). Suppose (1) fails, i.e., there exists a clopen set A such that A ⊆ R −1 (π A),
so A − R −1 (π A) = ∅. Then A = ∅. Let A0 = A − R −1 (π A). Since A is clopen
and R −1 maps clopen sets to clopen sets (Theorem 3.2.1(IIIb)), we see that both
π A and π A0 are clopen. Now suppose π A0 = ∅. Then there is x ∈ π A0 , i.e., x
is a passive point of A0 = A − R −1 (π A). We have x ∈ A and R(x) ∩ π A = ∅.
Therefore, x ∈/ π A. But then x is an active point of A, so there must be a point y ∈ X
such that x Ry, y ∈ / A, and (∃z)(y Rz & z ∈ A). Since R is transitive, we obtain
x Rz and z ∈ A. As R(x) ∩ π A = ∅, we have z ∈ / π A. Moreover, z ∈/ R −1 (π A), for
otherwise, R(z) ∩ π A = ∅ and (since x Rz) R(x) ∩ π A = ∅. Furthermore, since
z∈/ R −1 (π A) and z ∈ A, we have z ∈ A0 . So x Ry and y ∈ / A, and hence y ∈ / A0 ,
as well as y Rz and z ∈ A0 . Therefore, starting at x, one can leave A0 (x Ry and
y∈ / A0 ) and return to A0 again (y Rz and z ∈ A0 ). It follows that x is an active
(and not passive) point of A0 , so x ∈ / π A0 , a contradiction. Thus, π A0 = ∅, and
we have found a nonempty clopen A0 such that π A0 = ∅. This disproves (2). By
contraposition, (2) implies (1).
(2 ⇔ 3) Let (2) hold and a = 0. Suppose ρa = a. Then πa = 0 since πa =
a − ρa, a contradiction. Conversely, let (3) hold and a = 0. Then πa = a − ρa = 0
because ρa ≤ a and ρa = a. Thus, πa = 0.
(3 ⇔ 4) Suppose (3) holds and there exists a cyclic element b ∈ B; that is,
ρm (b) = ρm+1 (b) = 0. Then a = ρm (b) = 0 and ρa = a, contradicting (3). Con-
versely, suppose (3) fails, i.e. there exists a = 0 such that ρa = a. Clearly a is
cyclic. 

Grzegorczyk algebras are then characterized by the following.

3.5.12 Corollary A closure algebra (B, C) is a Grzegorczyk algebra if and only if

it has no cyclic elements.

3.5.13 Theorem The variety GA of Grzegorczyk algebras is generated by its finite

algebras.25

Proof First note that if p, q are algebraic polynomials of a closure algebra, then
p = q is equivalent to (− p ∨ q) ∧ (−q ∨ p) = 1. Indeed, p = q ⇔ p ≤ q and
q ≤ p ⇔ (− p ∨ q) = 1 and (−q ∨ p) = 1 ⇔ (− p ∨ q) ∧ (−q ∨ p) = 1.
Now let p be a polynomial. If p = 1 in all Grzegorczyk algebras, then clearly
p = 1 in all finite Grzegorczyk algebras. Now let p = p(v1 , . . . , vk ) = 1 (where
v1 , . . . , vk are the variables of p) in some algebra (B, C) ∈ GA. Then there exist
a1 , . . . , ak ∈ B such that p(a1 , . . . , ak ) = 1. Identify the algebra (B, C) with the

25 Editorial note: Julia Ilin pointed out a gap in Esakia’s original proof, so we have replaced it by an

algebraic version of the proof from pp. 158–9 of G. Boolos, The Logic of Provability (Cambridge
University Press), 1993.
3.5 Grzegorczyk Algebras 73

algebra of clopen sets (B(X ), R −1 ) of its hybrid (X, , R). Let S be the set of
subpolynomials of p(a1 , . . . , ak ) and

T = S ∪ {C(Cq − q) : q ∈ S}.

Let B0 be the Boolean subalgebra of B generated by T . Since T is finite, B0 is finite,

and it induces an equivalence relation ∼ on X given by

x ∼ y iff (∀a ∈ B0 )(x ∈ a ⇔ y ∈ a).

Let [x] be the equivalence class of x, and X 0 = {[x] | x ∈ X }. The equivalence

classes are exactly the atoms of B0 , so every element of B0 is a finite union of
equivalence classes. Thus, B0 is isomorphic to Sub(X 0 ).
Define a relation R0 on X 0 by:

[x]R0 [y] iff (∀a)[(Ca ∈ B0 & a ∈ y) ⇒ Ca ∈ x]

iff (∀a)[(Ca ∈ B0 & Ca ∈ y) ⇒ Ca ∈ x],

where the equivalence of the two definitions follows from the axioms a ≤ Ca and
CCa ≤ Ca, which also imply that R0 is reflexive and transitive. However, R0 is not
in general a partial order, so we define

[x]R1 [y] iff [x]R0 [y] and ([y]R0 [x] ⇒ [x] = [y]).

We claim that R1 is a partial order. Reflexivity is obvious and antisymmetry is imme-

diate from the definition. For transitivity, suppose [x]R1 [y]R1 [z], which implies
[x]R0 [y]R0 [z] and hence [x]R0 [z]. If in addition [z]R0 [x], then [z]R0 [y] by the
transitivity of R0 , which with [y]R1 [z] implies [z] = [y]. Then from [x]R1 [y] we
have [x]R1 [z].
Define C1 on B0 by

C1 a = {[x] : ∃y ∈ a such that [x]R1 [y]}.

Since B0 is isomorphic to Sub(X 0 ), it follows from the definition of C1 that (B0 , C1 )

is isomorphic to (Sub(X 0 ), R1−1 ). Then since (X 0 , R1 ) is a finite partially ordered
set, (B0 , C1 ) is a finite Grzegorczyk algebra by Corollary 3.5.10.
We claim that for each Cq ∈ S, we have Cq = C1 q. First, we show that
x∈ / Cq implies x ∈ / C1 q. Suppose x ∈/ Cq. To show x ∈ / C1 q, we must show that
for any y with [x]R1 [y], we have y ∈ / q. So suppose [x]R1 [y] and hence [x]R0 [y]. If
y ∈ q, then since [x]R0 [y] and Cq ∈ B0 , we have x ∈ Cq, contradicting our assump-
tion. Thus, y ∈ / q, so we conclude that x ∈/ C1 q.
Second, we show that x ∈ Cq implies x ∈ C1 q. Assume x ∈ Cq. If x ∈ q, we are
done, so assume x ∈ / q. Thus, x ∈ Cq − q and hence x ∈ C(Cq − q). Since x ∈ Cq,
74 3 Duality Theory: Hybrids

there is a y ∈ q such that x Ry. Then by (g2 ), y ∈ C(q − C(Cq − q)), so there is
a z ∈ q − C(Cq − q) such that y Rz. Hence x Rz, which implies [x]R0 [z]. Since
Cq ∈ S and hence q ∈ S, we have C(Cq − q) ∈ T . Then given C(Cq − q) ∈ B0 ,
x ∈ C(Cq − q), and z ∈ / C(Cq − q), it is not the case that [z]R0 [x]. Thus [x]R1 [z],
which with z ∈ q implies x ∈ C1 q.
As a consequence of the previous claim and the fact that p(a1 , . . . , ak ) = 1 in
(B, C), we conclude that p(a1 , . . . , ak ) = 1 in (B0 , C1 ). Since (B0 , C1 ) is a finite
Grzegorczyk algebra, this completes the proof. 

3.5.14 Corollary The variety GA of Grzegorczyk algebras is generated by the class

of all skeletal algebras.

Proof By Corollary 3.5.9, SA ⊆ GA. By Corollary 3.5.10, every finite Grzegorczyk

algebra is a skeletal algebra. Thus, the result follows from Theorem 3.5.13. 

We now give a name to the hybrids dual to Grzegorczyk algebras.

3.5.15 Definition A hybrid (X, , R) is well founded if, starting at any point x
of any clopen set A, one can “reach” a maximal point of A; more precisely, if
A ⊆ R −1 (max A).

Let HYB f be the category of all well founded hybrids and hybrid maps.

3.5.16 Theorem The category GA of Grzegorczyk algebras is dual to the category

HYB f of well founded hybrids.

Proof Apply the Duality Theorem (Theorem 3.3.4), Theorem 3.5.6, and
Corollary 3.5.7. 

Using this theorem and Corollary 3.5.8, we easily obtain the following.

3.5.17 Corollary The category HYB+ of strict hybrids is a full subcategory of the
category HYB f of well founded hybrids.

Bibliographic notes. Grzegorczyk algebras were introduced by the author as alge-

braic models for one of the most interesting modal systems (at least among the
systems of the post-Lewis period), namely the system formulated axiomatically by
A. Grzegorczyk in [8]. Some results were reported by the author in [7]; see also
[6]. Segerberg [17] was the first to prove the finite approximability of Grzegorczyk’s
modal system in terms of the semantics of “possible worlds” (cf. Theorem 3.5.13).26
A number of deep results related to Grzegorczyk algebras appeared in Blok’s mono-
graph [2]. Some results obtained by L. Maksimova in the interesting papers [12, 13]
are related to Grzegorczyk algebras.

26 Editorialnote: A logic is finitely approximable if it is characterized (in the sense of Sect. 2.3.3)
by a class of finite algebras.
References 75

References

1. Birkhoff, G. (1967). Lattice theory (Vol. 3), American mathematical society colloquium pub-
lications Providence: American Mathematical Society.
2. Blok, W.J. (1976) Varieties of interior algebras. Ph.D. thesis, University of Amsterdam, Ams-
terdam.
3. Cohn, P. M. (1965). Universal algebra. New York: Harper & Row Publishers.
4. Esakia, L. L. (1974). Topological Kripke models. Soviet Mathematics Doklady, 15, 147–151.
5. Esakia, L. L. (1979). On the theory of modal and superintuitionistic systems (Russian). Logical
inference (pp. 147–172). Moscow: Nauka.
6. Esakia, L. L. (1981). Diagonal constructions, Löb’s formula, and Cantor’s scattered spaces
(Russian). Studies in logic and semantics (pp. 128–143). Tbilis: Metsniereba.
7. Esakia, L. L. (1984). On the variety of Grzegorczyk algebras. Selecta Sovietica Mathematica,
3, 343–366.
8. Grzegorczyk, A. (1967). Some relational systems and the associated topological spaces. Fun-
damenta Mathematicae, 60, 223–231.
9. Halmos, P. R. (1962). Algebraic logic. New York: Chelsea Publishing Co.
10. Jónsson, B., & Tarski, A. (1951). Boolean algebras with operators. I. American Journal of
Mathematics, 73, 891–939.
11. Makinson, D. C. (1969). On the number of ultrafilters of an infinite Boolean algebra. Zeitschrift
für Mathematische Logik und Grundlagen der Mathematik, 15, 121–122.
12. Maksimova, L. L. (1979). Interpolation theorems in modal logics and amalgamable varieties
of topological Boolean algebras. Algebra and Logic, 18, 348–370.
13. Maksimova, L. L., & Rybakov, V. V. (1974). The lattice of normal modal logics. Algebra and
Logic, 13, 105–122.
14. Mostowski, A., & Tarski, A. (1939). Boolesche ringe mit geordneter basis. Fundamenta Math-
ematicae, 32, 69–86.
15. Nachbin, L. (1965) Topology and order. Translated from the Portuguese by Lulu Bechtolsheim.
Van Nostrand Mathematical Studies (Vol. 4). Princeton: D. Van Nostrand Co., Inc.
16. Priestley, H. A. (1972). Ordered topological spaces and the representation of distributive lat-
tices. Proceedings of the London Mathematical Society, 3(24), 507–530.
17. Segerberg, K. (1971) An essay in classical modal logic. Vols. 1, 2, 3. Filosofiska Studier (Vol.
13). Uppsala: Filosofiska Föreningen och Filosofiska Institutionen vid Uppsala Universitet.
18. Sikorski, R. (1960). Boolean algebras (Vol. 25). Ergebnisse der Mathematik und ihrer Gren-
zgebiete, N. F. Berlin: Springer.
19. Wolk, E. S. (1968). The topology of a partially well ordered set. Fundamenta Mathematicae,
62, 255–264.
Appendix

In this appendix, we outline, in general terms, the content of the second part of
the monograph entitled ‘Heyting Algebras II. Additional Chapters’ which is being
prepared for publication.1 Hopefully, the appendix will allow the reader to form an
opinion about the topics covered in part II.

A.1 On Subdirectly Irreducible Heyting Algebras

and Subdirect Products

In Sect. 1.1, we defined subdirect products of algebras and subdirectly irreducible

algebras, and stated a number of known theorems, including G. Birkhoff’s classical
theorem. It was noted that every variety is generated by its subdirectly irreducible
algebras; and, in a sense, every variety is characterized completely by its subdirectly
irreducible members.
Let us give a characterization of subdirectly irreducible Heyting algebras and
closure algebras in terms of hybrids.
A closure algebra (B, C) (resp. a Heyting algebra H ) is called well-connected
[14] if for any a, b ∈ B (resp. a, b ∈ H ) Ca ∧ Cb = 0 (resp. a ∨ b = 1) implies
that either Ca = 0 or Cb = 0 (resp. either a = 1 or b = 1).
It is easy to see that each subdirectly irreducible Heyting algebra (see Theorem
2.4.11) is well-connected; hence, every variety of Heyting algebras is generated
by its well-connected algebras. In light of Corollary 2.4.21, every subdirectly irre-
ducible closure algebra (B, C) is well-connected.2 A characterization of subdirectly
irreducible and well-connected algebras is given by the following.

1 Editorial note: As noted in the Foreword, volume II was never published.

2 Editorial note: In addition to Corollary 2.4.21, this uses the equivalence of (1) and (2) in Proposition

A.1.1.
© Springer Nature Switzerland AG 2019 77
G. Bezhanishvili and W. H. Holliday (eds.), Heyting Algebras,
Trends in Logic 50, https://doi.org/10.1007/978-3-030-12096-2
78 Appendix

A.1.1 Proposition Let (B, C) be a closure algebra and H its skeleton. Then the
following are equivalent:
(1) H is well-connected;
(2) (B, C) is well-connected;
(3) the hybrid (X, , R) that corresponds to the algebra (B, C) is a principal cone,
i.e., there exists x ∈ X such that R(x) = X .

A.1.2 Proposition (Subdirect irreducibility) Let (B, C) be a closure algebra and H

its skeleton. Then the following are equivalent:
(1) H is subdirectly irreducible;
(2) (B, C) is subdirectly irreducible;
(3) the hybrid (X, , R) is super-principal, i.e., there exists a clopen cluster C
such that R(x) = X for each x ∈ C.

Using Theorem 3.4.16 (see also Definition 3.4.12), one can prove:

A.1.3 Proposition (Subdirect products) The following are equivalent:

(1) the Heyting algebra H is decomposable into a subdirect product of algebras Hi
(i ∈ I );
(2) let Ai be the closed cone of the hybrid (X, , R) corresponding to the Heyting
algebra Hi (i ∈ I ); then the family of conesAi is a covering of some dense
subspace X 0 of the space (X, ), i.e., X 0 ⊆ {Ai : i ∈ I }.

A.1.4 Proposition (Cone covering) A Heyting algebra H is finitely approximable

if and only if the corresponding hybrid (X, , R) contains a family  {Ai : i ∈ I } of
finite cones covering some dense subset X 0 of the space X , i.e., X 0 ⊆ {Ai : i ∈ I }.

A.2 Hybrid Formulation of Glivenko’s Theorem

Let H be a Heyting algebra and Rg(H ) the set of all its regular elements, i.e.,

Rg(H ) = {¬a : a ∈ H }.

It is known that Rg(H ) is a Boolean algebra and that the map h : H → Rg(H ),
given by ha = ¬¬a (a ∈ H ), is a surjective homomorphism of the Heyting algebra
H onto the Boolean algebra Rg(H ) (see [16]).

A.2.1 Proposition Let (X, , R) be the hybrid corresponding to the Heyting alge-
bra H . Then the set Y = max X is a Stone space and the Boolean algebra B(Y ) of
clopen sets of Y is isomorphic to the algebra Rg(H ). Note that the hybrid (Y, Y , RY )
is a subhybrid of (X, , R), where Y = max X , Y is the subspace topology, and
RY is the restriction of R to Y . Moreover, RY is the identity relation on Y .
Appendix 79

A.3 Weak Decompositions of Boolean Algebras

and Heyting Algebras

The content of this section is based on [10]. Let {Hi : i ∈ I } be a family

 of Heyting
algebras. For each i ∈ I , let pi be the projection of the direct product Hi onto Hi .
 i∈I
A subalgebra H of the direct product Hi is called a weak product of the
i∈I
 algebras Hi (i ∈ I ) if H satisfies the following absorption condition: if
Heyting
a∈ Hi differs from b ∈ H only by finitely many components, then a is “absorbed”
i∈I 
by H . More precisely, if a ∈ Hi , b ∈ H and the set {i ∈ I : pi (a) = pi (b)} is
i∈I
finite, then a ∈ H .
Note that the concept of the weak product is applicable to bounded distributive
lattices as well. It is easy to check that every weak product of Heyting algebras Hi
(i ∈ I ) is a subdirect product of these algebras. If I is finite,
 then the direct product
of the algebras Hi (i ∈ I ) is their only weak product. Let W Hi denote an arbitrary
i∈I
weak product of the algebras Hi .
A set S = {ai ∈ H : i ∈ I } of non-zero  elements of H is called a partition of
the unit if ai ∧ a j = 0 for all i = j and {ai : i ∈ I } = 1 ∈ H .
An internal characterization of weak decompositions of Heyting algebras is given
by the following statement.

A.3.1 Proposition
 (Weak decompositions) A Heyting algebra H is isomorphic to
a weak product W Hi of Heyting algebras Hi (i ∈ I ) if and only if there exists a
i∈I
partition of the unit {ai : i ∈ I } in H such that H/ai  Hi for all i ∈ I (where H/ai
is the quotient algebra of H by the principal filter [ai )).

The following corollary is of independent interest.

A.3.2 Corollary (Weak decompositions

 for Boolean algebras) A Boolean algebra
B is isomorphic to a weak product W Bi of Boolean algebras Bi (i ∈ I ) if and only
i∈I
if there exists a partition of the unit {ai : i ∈ I } of B such that B/ai  Bi for all
i ∈ I.

Let Hi be the direct sum of Heyting algebras Hi (i ∈ I ), i.e., the subalgebra
i∈I  
of their direct product Hi consisting of all elements a ∈ Hi such that at least
i∈I i∈I
I : pi (a) = 0}, {i ∈ I : pi (a) = 1} is finite. It is easy to check that the
one of {i ∈ 
direct sum Hi is a weak product of the algebras Hi (i ∈ I ).
i∈I

A.3.3 Proposition Weak productsof the algebras Hi (i ∈ I ) are precisely

 the sub-
algebras of their direct product Hi that contain their direct sum Hi as a
i∈I i∈I
subalgebra. In other words, weak products of Heyting algebras Hi (i ∈ I ) are the
80 Appendix

 
algebras H “situated between” the direct sum Hi and the direct product Hi ,
  i∈I i∈I
i.e., Hi ⊆ H ⊆ Hi .
i∈I i∈I

One can describe weak products in terms of duality theory as follows. For sim-
plicity and brevity, we restrict ourselves to Boolean algebras.  Let Bi (i ∈ I ) be a
Boolean algebra and X i the corresponding Stone space. Let X i be the topological
 i∈I
sum of the spaces X i . It is known that X i is zero-dimensional and Hausdorff. Note
i∈I   
that if the set of indices I is infinite, then X i is not compact. Let β X i be the
 i∈I i∈I
Stone–Čech compactification3 of X i . For a space X , let Z(X ) be the set of all
i∈I
zero-dimensional Hausdorff compactifications4 of X . The set Z(X ) can be partially
ordered by means of continuous maps identical on X . Finally, let W be the set of all
weak products of the family {Bi : i ∈ I } of Boolean algebras ordered by inclusion.
A.3.4Proposition
  The partially ordered set W is isomorphic
 to the partially ordered
set Z X i of compactifications of the topological sum X i of Stone spaces X i .
i∈I i∈I

A.3.5 Corollary (Ph. Dwinger) The Stone space X of the direct product Bi of
i∈I
Boolean
  algebras
 Bi (i ∈ I ) is homeomorphic to the Stone–Čech compactification
β i∈I X i of the topological sum of the spaces X i , where X i corresponds to the
Boolean algebra Bi .5

A.3.6 Corollary (G. Day) The Stone space X of the direct sum Bi of Boolean
i∈I
  Bi (i ∈ I ) is homeomorphic
algebras  to6the Alexandroff one-point compactification
α X i of the topological sum Xi .
i∈I i∈I

 Concluding this section, we note that if a Heyting algebra H is the direct product
Hi of algebras Hi (i ∈ I ), then the skeletal closure algebra B(H ) is isomorphic
i∈I

to a weak product W B(Hi ) of skeletal algebras B(Hi ) (cf. Proposition 2.5.9(2)).
i∈I
This weak product coincides with the categorical product of the algebras B(Hi ) in
the category SA (and in general does not coincide with their direct product; recall
that the class SA of skeletal algebras is not a variety). Using the above, one can give
a description of the direct product of Heyting algebras in terms of hybrids.

3 Editorial note: See, e.g., Sect. 3.6 of R. Engelking, General Topology, 2nd ed. (Heldermann-

Verlag), 1989.
4 Editorial note: For the notion of compactification, see, e.g., Sect. 3.5 of R. Engelking, General

Topology, 2nd ed. (Heldermann-Verlag), 1989.

5 Editorial note: See Theorem 3 of Ph. Dwinger, “Remarks on the field representations of Boolean

algebras,” Indagationes Mathematicae 22: 213–217, 1960.

6 Editorial note: See Theorem 4 of G. Day, “Superatomic Boolean algebras,” Pacific Journal of

Mathematics 23(3): 479–489, 1967.

Appendix 81

A.4 The Lattice of Varieties of Heyting Algebras

Using the descriptions of weak products and almost direct products (in the sense
of S. Feferman and R. Vaught)7 of closure algebras and Heyting algebras in terms
of hybrids [7], one can prove the following statement, due to W. Blok [3] and the
author [4].8

A.4.1 Proposition The lattice L(HA) of varieties of Heyting algebras is isomorphic

to the lattice L(GA) of varieties of Grzegorczyk algebras.

A.5 Representable Ordered Sets: Grätzer’s Problems

A partially ordered set (X, R) is said to be hybrid (or representable) if there exists
a hybrid (X ,  , R ) such that the partially ordered set (X, R) is isomorphic to
(X , R ). Let HPO be the class of all hybrid partially ordered sets. Clearly a partially
ordered set (X, R) is hybrid if and only if there exists a Heyting algebra H such that
(X, R) is isomorphic to the partially ordered set of all prime filters of H .

A.5.1 Proposition Let (X, R) be a hybrid partially ordered set (i.e.,

(X, R) ∈ HPO).
(1) The partially ordered set (X, R) has enough gaps.9
(2) The partially ordered set (X, R) is Dedekind complete.10
(3) All finite partially ordered sets belong to HPO.
(4) A linearly ordered set (X, R) belongs to HPO if and only if (X, R) satisfies
conditions (1)–(2).11
(5) The class HPO is closed under direct products of arbitrary families of partially
ordered sets from HPO.
(6) The class HPO is closed under cardinal sums of arbitrary families of partially
ordered sets from HPO (recall that (X, R) is the cardinal sum of partially
ordered sets (X i , Ri ) (i ∈ I ) if X is the (disjoint) union of the sets X i (i ∈ I )
and for all x, y ∈ X , x Ry ⇔ ∃ i ∈ I : x, y ∈ X i & x Ri y).

Unfortunately, the author failed to obtain a “digestable” characterization of the

class HPO.
We conclude the section by quoting Grätzer [12]: “Investigate further the poset
P(L) of prime ideals of a distributive lattice” (Problem II.4) and “Characterize

7 Editorial note: See S. Feferman and R. L. Vaught, “The first order properties of products of

algebraic systems,” Fundamenta Mathematicae 47: 57–103, 1959.

9 Editorial note: Recall Definition 3.2.13.
10 Editorial note: Recall Definition 3.2.18.
11 Editorial note: C.f. Theorem 9 of R. Balbes, “On the Partially Ordered Set of Prime Ideals of a

Distributive Lattice,” Canadian Journal of Mathematics 23(5): 866–874, 1971.

82 Appendix

P(L) under the additional assumption that L has a unit and/or a zero” (Problem
II.5). It is tempting to replace L by H and P(L) by P(H ) in those quotations and
suggest this as a new problem to the reader.

A.6 MacNeille Completions

MacNeille showed, by generalizing the celebrated Dedekind cuts, that every partially
ordered set can be embedded into a complete lattice so that all existing suprema and
infima are preserved (the so-called MacNeille completion; see for example Sect. V.9
of [2]). However, this embedding does not preserve certain important properties: e.g.,
there is an example of a distributive lattice whose MacNeille completion is not even
modular (N. Funayama). The MacNeille completion of a Boolean lattice is a Boolean
lattice (V. Glivenko, M. Stone). Using the Duality Theorem and certain b properties
of hybrids, one can describe a construction dual to MacNeille’s construction and
show that the completion of a Heyting lattice is a Heyting lattice.

A.7 The Category of Heyting Algebras is Balanced

The content of this section is based on [11]. Let (X i , i , Ri ) be hybrids (i = 1, 2).

A hybrid (X, , R) is called the product of the hybrids (X i , i , Ri ) (i = 1, 2) if
the set X = X 1 × X 2 is equipped with the topology  of the product of the spaces
(X i , i ) (i = 1, 2) and the relation R on X is defined componentwise, i.e.,

(x, y)R(x , y ) ⇔ x R1 x & y R2 y .

Let X 0 , X 1 , X 2 be hybrids, f i : X i → X 0 a surjective hybrid map (i = 1, 2), and

X = (x, y) ∈ X 1 × X 2 : f 1 (x) = f 2 (y) .

A.7.1 Proposition Let H0 , H1 , H2 be the Heyting algebras corresponding to the

hybrids X 0 , X 1 , and X 2 , and let h i : H0 → Hi (i = 1, 2) be the inclusions dual to
the maps f i . If H is the Heyting algebra of cones of X , then H amalgamates the
amalgam (H0 , H1 , H2 ).

Using the Duality Theorem (Theorem 3.3.4), Proposition A.7.1, and some prop-
erties of hybrids, we can prove:

A.7.2 Proposition The category HA of Heyting algebras is balanced; that is, every
epimorphism of the category HA is surjective.

We recall that the category BA of Boolean algebras is balanced, but that the
category DL of bounded distributive lattices is not balanced.
Appendix 83

A.8 Equationally Non-conservative Operations

It is well known that every equational class (variety) K of algebras is closed under
the fundamental algebraic operations of taking homomorphic images  (H),
 subalge-
bras (S), direct products (P), subdirect products (PS ), direct limits lim , and inverse
  −→
limits lim . In order to emphasize this conservation property, i.e., the preservation
←−
of identities under the indicated operations, we call such operations equationally
conservative (or simply conservative). The role of such operations in modern alge-
braic research is well known. However, we can mention interesting applications of
non-conservative operations as well.

(A) Tensor sums [8]. Let {X i : i ∈ I } be a family of hybrids (I = ∅) and X = Xi
i∈I
the cartesian product of the sets X i (i ∈ I ). Equip X with the product topology 
and the “componentwise” order relation R, i.e., x Ry ⇔ xi Ri yi for all i ∈ I . Call X
the product of the hybrids X i (i ∈ I ).

A.8.1 Proposition (a) If X i is a hybrid for each i ∈ I , then the product X = Xi
i∈I
is a hybrid;
(b) the hybrid X is strict if and only if X i is strict for each i ∈ I .
Let Bi be the closure algebra (resp. let Hi be the Heyting algebra) corresponding
to the hybrid X i (i ∈ I ). The closure algebra B (resp. the Heyting algebra H ) dual
to the hybrid (resp. the strict hybrid) X is called the tensor sum of closure algebras
Bi (resp. the Heyting algebras Hi ) and is denoted by B = Bi (resp. H = Hi ).
i∈I i∈I

A.8.2 Proposition The variety of Heyting algebras HA and the variety of closure
algebras CA are closed under the operation of tensor sum.
One can illustrate the non-conservativity of the operation on specially chosen
Heyting algebras. For example, the so-called variety LC, i.e., the variety of Heyting
algebras such that the identity (a → b) ∨ (b → a) = 1 holds, is not closed under
tensor sums. However, it is noteworthy that many interesting classes of algebras are
closed under .
A.8.3 Proposition The following classes of algebras are closed under tensor sums:
(1) KC (the class of Heyting algebras such that the law of weak excluded middle
holds, i.e., ¬a ∨ ¬¬a = 1);
(2) SA (the class of skeletal closure algebras);
(3) BA (the class of Boolean algebras);
(4) MA (the class of Halmos’s monadic algebras);
(5) Post (the class of Post algebras);
(6) Luk (the class of Lukasiewicz algebras).12

12 Editorial
note: For discussion of monadic algebras, see [13]. For discussion of Post algebras and
Lukasiewicz algebra, see, e.g., [1].
84 Appendix

Note that every Lukasiewicz algebra and every Post algebra is a Heyting lattice.
Let X k = {0, 1} (for all k ∈ ω) be the two-point space with the discrete topology
k . Define an order Rk on X k as follows: x Rk y ⇔ x = y or x = 0 and y = 1.
Consider the product of hybrids (X, , R) = (X k , k , Rk ).
i∈I

A.8.4 Proposition (X, , R) is a strict hybrid and the corresponding Heyting alge-
bra H is a free distributive lattice on countably many generators. Moreover, the lattice
H is isomorphic to the coproduct (in the category DL) of the three-element Heyting
lattices Hk (k ∈ ω).
Let (X i , Ri ) (i = 1, 2) be hybrids such that (X 2 , R2 ) is a cluster, i.e., for all
x, y ∈ X 2 , x R2 y and y R2 x. Let (X, R) be their product. Figuratively speaking,
(X, R) is obtained from (X 1 , R1 ) by blowing every point x ∈ X 1 up into a cluster A x
of size X 2 (i.e., A x and X 2 are of the same cardinality) and the order among clusters
is the same as the order among points of X 1 ; in other words, the skeleton of the hybrid
(X, R) coincides with (X 1 , R1 ). The blowup procedure has interesting applications;
in particular, it features in the proof of the Dummett–Lemmon conjecture [5].
In conclusion, we note that the operation of tensor sum induces the following
operation on varieties (and hence on modal systems and superintuitionistic logics).
If K 1 , K 2 are varieties of Heyting algebras, then let K = K 1 K 2 be the smallest
variety generated by the class of Heyting algebras of the form H1 H2 , where
H1 ∈ K 1 and H2 ∈ K 2 .
(B) Alternation [8]. Consider another non-conservative operation, which we call the
operation of alternation. Its definition is motivated by S. Jaskowski’s -operation.13
Let (Bi , Ci ) be a closure algebra and Hi its skeleton, indexed by elements of a
partially ordered set (I, ≤).
The alternation # (Bi , Ci ) = (B, C) of closure algebras (Bi , Ci ) (i ∈ I ) is
i∈I
defined as follows:

(a) B = Bi is the direct product of the Boolean algebras Bi ,
i∈I
i = Ci f i if (∀ j)(i < j ⇒ f j = 0) and (C f )i = 1 otherwise; here f ∈
(b) (C f )
B= Bi , and f i is the projection of f onto the component Bi .
i∈I

A.8.5 Proposition The alternation # (Bi , Ci ) of a family of closure algebras

i∈I
{(Bi , Ci ) : i ∈ I }, for a partially ordered set I , is a closure algebra.
Note that if the partially ordered set (I, ≤) is an antichain, i.e., x ≤ y ⇔ x = y,
then it is immediate from the definition that the alternation # (Bi , Ci ) is the direct
 i∈I
product (Bi , Ci ) of the algebras (Bi , Ci ) (i ∈ I ).
i∈I

13 Editorialnote: See S. Jaśkowski, “Recherches sur le Système de la Logique Intuitioniste,” Actes

du Congrès International de Philosophie Scientifique 6: 58–61, 1936. An English translation is
available in S. Jaśkowski, “Investigations into the System of Intuitionist Logic,” Studia Logica
34(2): 117–120, 1975.
Appendix 85

A definition of the operation dual to alternation is as follows. Here we restrict

ourselves to a finite partially ordered set I of indices.
By definition, a hybrid (X, , R) is the ordered sum of hybrids (X i , i , Ri )
(i ∈ I ) if:
(a) the space (X, ) is the topological sum of the spaces (X i , i );
(b) for all x, y ∈ X , x Ry iff either x ∈ X i and y ∈ X j (i ≤ j, i = j) or x, y ∈ X i
and x Ri y (in other words, (X, R) is obtained by “positioning” the quasi-ordered
sets (X i , Ri ) according to the partially ordered set (I, ≤), preserving the original
order Ri inside every (X i , Ri )!).
Note that in the case of (I, =), the quasi-ordered sets (X i , Ri ) are positioned in a
“row” without any link between any of them.
A.8.6 Proposition Let (I, ≤) be a finite partially ordered set and (X i , i , Ri ) a
hybrid for each i ∈ I . Then:
(1) the ordered sum (X, , R) of these hybrids is a hybrid;
(2) if (X i , i , Ri ) is strict for each i ∈ I , then the hybrid (X, , R) is strict.
Using the Duality Theorem (Theorem 3.3.4), we can assume that the operation
of alternation # Hi is also defined for arbitrary Heyting algebras. In particular, if
i∈I
(I, ≤) = ({1, 2}, ≤) is the two-element chain, then in our terminology Jaskowski’s
-operation can be written as follows: (H2 ) = H1 # H2 , where H1 is the two-element
Boolean algebra and H2 is an arbitrary Heyting algebra; or, passing to the correspond-
ing hybrids (X i , i , Ri ) (i = 1, 2), we obtain their ordered sum; here (X 2 , 2 , R2 )
is an arbitrary hybrid and (X 1 , 1 , R1 ) is the one-point hybrid “situated under”
(X 2 , 2 , R2 ).
Even this very special case of the operation of alternation demonstrates its equa-
tional non-conservativity: recall the known method of a characteristic sequence of
Heyting algebras starting from the two-element Boolean algebra by iterating the
operations of direct product and Jaskowski’s -operation.

A.9 Boolean Cascades

The content of this section is based on [10]. In Part II of the book, certain special
classes of Heyting algebras and closure algebras are analyzed in terms of hybrids:
(1) Stone algebras;
(2) Horn’s LC-algebras;
(3) Halmos’s monadic algebras;
(4) Closure algebras corresponding to McKinsey’s interesting modal system S4.1
(“Lions in Alaska”14 );

14 Editorial note: See J.C.C. McKinsey, “On the syntactical construction of systems of modal logic,”

The Journal of Symbolic Logic, Vol. 10, No. 3, 1945, pp. 83–94.
86 Appendix

(5) A complete description of the free cyclic algebra of Makinson’s variety D ∗ ;

(6) Cascade Heyting algebras.
We comment only on this last item because the indicated class of Heyting algebras
corresponds to a superintuitionistic logic which has not been considered previously
in the literature.
Let PWL (called the logic of Peirce’s weak law) be the calculus obtained from
the intuitionistic propositional calculus by postulating the formula
 
(q → p) ∨ (( p → q) → p) → p

as an additional axiom (which we call Peirce’s weak law).

Let CHA be the class of Heyting algebras satisfying Peirce’s weak law
 
(b → a) ∨ ((a → b) → a) → a = 1

(cf. Peirce’s law in Proposition 2.1.5). We call the algebras in this class cascade
Heyting algebras.
The cascade join of a finite sequence of Heyting algebras Hi (1 ≤ i ≤ k; k ∈ ω) is,
by definition, the Heyting algebra H that is the union of the sublattices Hi isomorphic
to the lattices Hi (1 ≤ i ≤ k) such that the intersection of Hi and Hi+1 (1 ≤ i < k)
contains exactly one element that is the least element of Hi+1 and the greatest element
of Hi . If a Heyting algebra H is isomorphic to the cascade join of a finite sequence
of Boolean algebras, then we call H a Boolean cascade.

A.9.1 Proposition
(1) The variety CHA of cascade Heyting algebras is generated by the class of finite
Boolean cascades and is locally finite.
(2) The smallest variety of closure algebras generated by the skeletal closure alge-
bras B(H ), where H ∈ CHA, is the variety of Grzegorczyk algebras satisfying
the additional identity
  
C(a − Cb) ∧ C a − C Ca − C(Cb − Ca) = 0.

This variety is hereditarily finitely approximable.

Let (X, R) be a partially ordered set. We call the following condition the three
point condition: for any three distinct points x, y, z ∈ X , if x and y are incomparable,
then x Rz implies y Rz.

A.9.2 Proposition Let H be a subdirectly irreducible Heyting algebra and X the

corresponding hybrid. Then the following are equivalent:
(1) H is a cascade algebra;
(2) the partially ordered set (X, R) satisfies the three point condition.
Appendix 87

A.10 Cantor’s Scattered Spaces

In this section we recall a few important and fruitful century-old ideas due to Cantor
and, with pleasure, discover their close connection with Grzegorczyk’s and Gödel’s
modal systems.
(1) The problem of representation of functions by trigonometric series was a focus
of attention for many mathematicians when (100 years ago) George Cantor, trying
to extend the uniqueness of representation to a wider class of functions and not yet
having discovered set theory, introduced the concept of derivative of infinite order.
This concept was not only the starting point of research in point-set topology, but
it also lead Cantor to the construction of transfinite ordinal numbers and later to
the discovery of set theory. The celebrated diagonal construction of Cantor and the
concept of scattered space are also of that period. We start by recalling the main
definitions.
A point x of a topological space X is called a limit point of a set A if for
every neighborhood Ux of x we have (A ∩ Ux ) − {x} = ∅. We denote the set
of all limit points of A by d A. The set d A is called the derivative of A. The
derivatives of order α are defined by transfinite induction: A0 = A, Aα+1 = d Aα
and Aλ = {Aα : α < λ} for a limit ordinal λ.
It is known that in introducing ordinal numbers Cantor was motivated by the
following concept. A topological space X is said to be scattered if it does not
contain a nonempty subset that is dense-in-itself (Cantor). Recall that a space X is
scattered if and only if there exists an ordinal α such that X α = ∅ and that every
ordinal is a scattered space.15
(2) It has been recently discovered that the modal systems G of Gödel and Grz of
Grzegorczyk, which are closely related to each other, can be applied to the analysis
of mathematical results related to Peano Arithmetic.16 Both systems allow adequate
arithmetical interpretations: for the former system, interpret the modality  p as
Bew( p), where Bew is the predicate of provability in Peano Arithmetic (R. Solovay);
for the latter system, interpret  p as Bew( p) & p (A.V. Kuznetsov, R. Goldblatt,
G. Boolos).
Now we turn to the connections between (1) and (2).
A.10.1 Proposition Grzegorczyk’s modal system Grz (resp. Gödel’s modal system
G) is characterized by the class of Cantor’s scattered spaces if the operator of pos-
sibility ♦ is interpreted as the operator of topological closure (resp. as the operator
of topological derivative).
The proof of this statement is based on the following new axiomatic definition of
Cantor’s scattered spaces, which, in our point of view, is of independent interest.

15 Editorial note: See, e.g., Sect. 8.5 of Z. Semadeni, Banach Spaces of Continuous Functions, Vol.

1 (PWN – Polish Scientific Publishers), 1971.

16 Editorial note: The system G is the smallest normal modal logic containing the Löb axiom

( p → p) →  p and is now called the Gödel–Löb logic GL. See, e.g., G. Boolos, The Logic
of Provability (Cambridge University Press), 1993.
88 Appendix

A.10.2 Proposition (An axiomatic definition of Cantor’s scattered spaces) The class
of scattered spaces is exactly the class of sets X equipped with an operator d satisfying
the following conditions:

(1) d∅ = ∅;
(2) d(A ∪ B) = d A ∪ d B;
(3) d A = d(A − d A) (Löb’s formula!).

In other words, the class of scattered spaces coincides with the class of topological
spaces X such that the operator of derivative on X satisfies “Löb’s identity” d A =
d(A − d A) (for all sets A). We conclude by referring the reader to [9] for more
details.

A.11 Symmetric Heyting algebras

A number of authors (G. Moisil, A.V. Kuznetsov, A. Muchnik, S. Rauszer, L. Iturioz)

have suggested “symmetric” formulations of intuitionistic propositional calculus
SIn. In this calculus, for each connective &, ∨, →, ¬, there is a dual one ∨, &,
←, ∼, and the duality principle of classical logic is restored. On the other hand,
McKinsey and Tarski [15] had already paid special attention to the difference between
Boolean lattices and Heyting lattices due to the failure of the duality principle in
Heyting lattices. On p. 141 of that work, the authors write: “The problem of dualism
in Brouwerian algebra is not yet clear.” McKinsey and Tarski indicated that the class
HL of Heyting lattices is very different from both the class DL of distributive lattices
and the class BL of Boolean lattices (which is a subclass of HL) because even the
weak duality principle fails in HL. In the quoted paper, they introduced the concept
of bi-Heyting algebra (this concept, as the authors indicate, had been considered by
T. Skolem as early as 1919).
Let (H, ≤) be a bounded distributive lattice. Put the lattice H “upside-down,”
i.e., re-order the set H as follows: a ≤◦ b ⇔ b ≤ a. It is easy to see that (H, ≤◦ ) is a
bounded distributive lattice. So the class of distributive lattices is stable with respect
to the operation of “turning things upside-down.” The class BL of Boolean lattices
has the property of strict stability. Namely, if (B, ≤) ∈ BL, then (B, ≤◦ ) not only
remains in BL but is even isomorphic to the original one, i.e., (B, ≤)  (B, ≤◦ ). On
the other hand, the class of Heyting lattices is not stable: it is known that the family
 of all open sets of any topological space, ordered by inclusion ⊆, is a Heyting
lattice; however, the lattice (, ⊆◦ ) (which is obviously isomorphic to the lattice of
closed subsets of X ) is, in general, not a Heyting lattice.
We call a Heyting lattice (H, ≤) symmetric (resp. strictly symmetric) if (H, ≤◦ )
is a Heyting lattice (resp. a Heyting lattice isomorphic to the original one (H, ≤)).
Appendix 89

A.11.1 Proposition [6] Let H be a Heyting lattice and (X, R) its hybrid. The fol-
lowing are equivalent:
(1) H is symmetric (resp. strictly symmetric);
(2) (X, Ř) is a hybrid (resp. a hybrid isomorphic to the hybrid (X, R)), where
x Ř y ⇔ y Rx for all x, y ∈ X .

References

1. Balbes, R., & Dwinger, Ph. (1974). Distributive lattices. Columbia, MO: University of Missouri
Press.
2. Birkhoff, G. (1967). Lattice theory (3rd ed., Vol. XXV). Providence: American Mathematical
Society Colloquium Publications, American Mathematical Society.
3. Blok, W. J. (1976). Varieties of interior algebras. Ph.D. thesis, University of Amsterdam.
4. Esakia, L. L. (1976). On modal “companions” of superintuitionistic logics (Russian). In VII
Soviet Symposium on Logic (Kiev, 1976) (pp. 135–136).
5. Esakia, L. L. (1976). On the Dummett-Lemmon conjecture (Russian). In IV Soviet Conference
in Mathematical Logic (Chisinau, 1976) (p. 160).
6. Esakia, L. L. (1978). Semantical analysis of bimodal systems (Russian). In Logic, Semantics
and Methodology, Metsniereba, Tbilisi (pp. 87–99).
7. Esakia, L. L. (1979). Duality for almost direct products of Boolean, Heyting and closure
algebras. In VI International Congress of Logic, Methodology and Philosophy of Science, West
Germany (pp. 103–106).
8. Esakia, L. L. (1979). On the theory of modal and superintuitionistic systems (Russian). In
Logical Inference. Moscow: Nauka (pp. 147–172).
9. Esakia, L. L. (1981). Diagonal constructions, Löb’s formula, and Cantor’s scattered spaces
(Russian). In Studies in Logic and Semantics, Metsniereba, Tbilisi (pp. 128–143).
10. Esakia, L. L. (1983). On a locally finite variety of Heyting algebras (Russian). In XVII Soviet
Algebraic Conference. Part II, Minsk (pp. 280–281).
11. Esakia, L. L., & Grigolia, R. (1981). The variety of Heyting algebras is balanced (Russian). In
XVI Soviet Algebraic Conference. Part II, Leningrad (pp. 37–38).
12. Grätzer, G. (1978). General lattice theory. Basel: Birkhäuser Verlag.
13. Halmos, P. R. (1962). Algebraic logic New York: Chelsea Publishing Co.
14. McKinsey, J. C. C., & Tarski, A. (1944). The algebra of topology. Annuals of Mathematics,
2(45), 141–191.
15. McKinsey, J. C. C., & Tarski, A. (1946). On closed elements in closure algebras. Annals of
Mathematics, 2(47), 122–162.
16. Rasiowa, H., & Sikorski, R. (1963). The mathematics of metamathematics. Monografie Matem-
atyczne, Tom 41. Warsaw: Panstwowe Wydawnictwo Naukowe.
Index

A envelope, 29
Active point, 70 free extension, 29
Adequate model (of a logic), 22 lattice, 12
Algebra Bow, 48
Boolean, 16
closure, 17
Grzegorczyk, 69 C
Heyting, 16 Cardinal sum, 81
interior, 18 Cascade
of clopen sets, 57 Boolean, 86
of open sets, 16 Heyting algebra, 86
pseudo-Boolean, 16 join, 86
skeletal, 29 Category, 2
subdirectly irreducible, 2 algebraic, 3
Algebraic category, 3 balanced, 3
equational, 3 dual, 4
Alternation (operation), 84 equivalent, 4
Amalgamable equational category, 4 isomorphic, 3
Antichain, 48 topos, 13
Anti-symmetric (binary relation), 6 Center, 15
Central element, 15
Chain, 8
B CL, 68
Balanced category, 3 Class of algebras
Basis, 5 congruence distributive, 2
Bimorphism, 3 equationally definable, 2
Binary relation variety, 1
anti-symmetric, 6 with congruence extension property, 2
connected, 6 Clopen, 5
downward directed, 6 Closed
reflexive, 6 element, 18
symmetric, 6 map, 6
transitive, 6 set, 5
upward directed, 6 Closure algebra, 17
Boolean over a topological space, 21
algebra, 16 well-connected, 77
cascade, 86 Closure field, 20
© Springer Nature Switzerland AG 2019 91
G. Bezhanishvili and W. H. Holliday (eds.), Heyting Algebras,
Trends in Logic 50, https://doi.org/10.1007/978-3-030-12096-2
92 Index

Closure operator, 17 prime, 17

Cluster, 7 Epimorphism, 3
nontrivial, 7 Equational category, 3
Coequivalence, 4 amalgamable, 4
Compact Equationally
element (relative to another element), 13 conservative operation, 83
strictly (element), 23 definable, 2
topological space, 5 non-conservative operation, 83
Complement, 12 Equivalence, 4
Completely additive, 23 Equivalent, 4
Completion Esakia’s lemma, 57
MacNeille, 82 E-saturated, 8
of a chain, 50 Exact model (of a logic), 22
Cone, 7 Excluded middle, 16
lower, 7 Exponential topology, 5
principal, 7 Extremally disconnected, 5
upper, 7
Congruence distributive, 2
Congruence extension property, 2 F
Connected (binary relation), 6 Field of sets, 4
Conservative operation, 83 reduced, 4
Continuous Filter, 24
lattice, 13 maximal, 24
map, 6 prime, 24
relation, 41 principal, 24
Contravariant functor, 4 proper, 24
Covariant functor, 4 skeletal, 26
Finitely approximable
logic, 74
D variety, 2
Dedekind complete, 52 Free Boolean extension, 29
Dense Function
subalgebra, 3 bijective, 1
subset, 6 closed, 6
Derivative of a set, 87 continuous, 6
Disjoint, 1 homeomorphism, 6
Distributive lattice, 15 injective, 1
Double negation law, 16 isomorphism, 10
Downward directed (binary relation), 6 isotone, 9
Dual, 4 open, 6
Duality, 4 strongly isotone, 9
Duality Theorem (for CA and HYB), 55 surjective, 1
Functor
coequivalence, 4
E contravariant, 4
Element covariant, 4
central, 15 duality, 4
closed, 18 equivalence, 4
join irreducible, 17
join prime, 17
meet irreducible, 17 G
meet prime, 17 G, 87
open, 18 Gap (in an ordered set), 50
Index 93

Gödel Kolmogoroff space, 5

modal system, 87 Kripke
rule, 21 incomplete, 23
translation, 68 model, 22
Grz, 68 Kuratowski’s axioms, 17
Grzegorczyk
algebra, 69
modal system, 68 L
Lattice
Boolean, 12
H bounded, 15
Hausdorff space, 5 continuous, 13
Hausdorff’s Principle, 9 distributive, 15
Heyting algebra, 16 Heyting, 11
cascade, 86 of open sets, 16
well-connected, 77 Layer, 68
Heyting lattice, 11 Limit point, 87
strictly symmetric, 88 Linearly ordered set, 8
symmetric, 88 Löb axiom, 87
Homeomorphic, 6 Locally finite, 2
Homeomorphism, 6 Logic
Homomorphism, 1 characterized by a class of algebras, 22
bijective, 1 characterized by an algebra, 22
injective, 1 CL, 68
surjective, 1
defined by a class of algebras, 22
Hybrid, 41
finitely approximable, 74
map, 54
G, 87
partially ordered set, 81
Grz, 68
super-principal, 78
In, 68
well founded, 74
of Peirce’s weak law, 86
S4, 21
S5, 68
I
superintuitionistic, 21
Ideal, 13
Lower cone, 7
Implication, 11
In, 68
Infinite distributive law, 12
Interior M
algebra, 18 MacNeille completion, 82
operator, 18 Maximal
Isomorphic, 3 chain, 9
quasi-ordered sets, 10 point, 8
Isomorphism, 3 Meet irreducible, 17
between quasi-ordered sets, 10 Meet prime, 17
Isotone, 9 Minimal (point), 8
Modal companion, 68
Modal system, 21
J G (Gödel), 87
Join irreducible, 17 Grz (Grzegorczyk), 68
Join prime, 17 Lewis, 21
S4, 21
S5, 68
K unmodellable, 23
Knot of a bow, 48 Monomorphism, 3
94 Index

N Quasi-linearly ordered set, 8

Non-conservative operation, 83 Quasi-maximal (point), 8
Normal extension, 21 Quasi-minimal (point), 8
Quasi-ordered set, 7
Quotient space, 65
O
Object (of a category), 3
injective, 3 R
projective, 3 Rank
Open finite, 31
element, 18 infinite, 31
map, 6 of an element, 31
set, 5 Reduced
Operation field of sets, 4
alternation, 84 ring of sets, 4
complement, 12 Reflexive (binary relation), 6
difference, 1 Regular
intersection, 1 element, 78
union, 1 open (set), 5
Order component (of a hybrid), 41 Relatively complete sublattice, 28
Ordered sum, 85 Relative pseudocomplement, 11
Representable partially ordered set, 81
Representation Theorem (for Heyting alge-
P bras), 67
Partially ordered set, 7 Ring
hybrid, 81 of clopen cones, 61
representable, 81 of cones, 20
Partition, 1 of sets, 4
induced by E, 8 reduced, 4
of the unit, 79 Rule
Passive point, 70 Gödel’s, 21
Peirce’s law, 16 modus ponens, 21
Peirce’s weak law of substitution, 21
logic of, 86
Point
active, 70 S
passive, 70 S4, 21
Possible world, 22 S5, 68
Priestley Scattered topological space, 87
separation axiom, 52 Separation axiom
space, 52 Hausdorff, 5
Prime, 17 Priestley, 52
Prime filter theorem, 24 T0 , 5
Principal (cone), 7 T1 , 5
Product Skeletal algebra, 29
of hybribs, 83 Skeleton
topological, 6 of a closure algebra, 18
topology, 6 of a hybrid, 66
Pseudo-Boolean algebra, 16 of a quasi-ordered set, 8
Pseudocomplement, 11 Stable, 88
Stone space, 5
Strictly
Q compact element, 23
Quasi-chain, 8 compactly generated, 23
Index 95

symmetric Heyting lattice, 88 Topology, 5

Strictly stable, 88 exponential, 5
Strongly isotone, 9 product, 6
Subbasis, 5 Vietoris, 5
Subcategory, 2 Topos, 13
full, 3 Totally disconnected, 5
Subdirectly irreducible, 2 Transitive (binary relation), 6
Subdirect product, 2 T0 -space, 5
Subhybrid, 65 T1 -space, 5
Sublattice
relatively complete, 28
Superintuitionistic U
fragment, 68 Unmodellable modal system, 23
logic, 21 Upper cone, 7
Super-principal hybrid, 78 Upward directed (binary relation), 6
Symmetric
binary relation, 6
Heyting lattice, 88
V
Variety, 1
finitely approximable, 2
T
locally finite, 2
Tensor sum
of closure algebras, 83 Vietoris topology, 5
of Heyting algebras, 83
Three point condition, 86
Topological component (of a hybrid), 41 W
Topological product, 6 Way below, 13
Topological space, 5 Weak
compact, 5 decomposition, 79
extremally disconnected, 5 product, 79
Hausdorff, 5 Well-connected
Kolmogoroff, 5 closure algebra, 77
scattered, 87 Heyting algebra, 77
Stone, 5 Well founded hybrid, 74
T0 -space, 5
T1 -space, 5
totally disconnected, 5 Z
zero-dimensional, 5 Zero-dimensional, 5