Tese Dimi 2018 09 14

An algebraic framework to
a theory of sets based on

the surreal numbers
Dimi Rocha Rangel
Tese apresentada
ao
Instituto de Matemática e Estatı́stica
da
Universidade de São Paulo
para
obtenção do tı́tulo
de
Doutor em Ciências
Programa: Doutorado em Matemática do IME-USP

Orientador: Prof. Dr. Hugo Luiz Mariano
Durante o desenvolvimento deste trabalho o autor recebeu auxı́lio financeiro da CAPES
Julho de 2018
An algebraic framework to
a theory of sets based on
the surreal numbers
Esta versão da tese contém as correções e alterações sugeridas

pela Comissão Julgadora durante a defesa da versão original do trabalho,
realizada em 17/07/2018. Uma cópia da versão original está disponı́vel no
Instituto de Matemática e Estatı́stica da Universidade de São Paulo.
Comissão Julgadora:
• Prof. Dr. Hugo Luiz Mariano (Presidente) - IME-USP

• Prof. Dr. Odilon Otavio Luciano - IME-USP
• Prof. Dr. Giorgio Venturi - UNICAMP
• Prof. Dr. Rodrigo de Alvarenga Freire - UNB
• Prof. Dr. Andreas Bernhard Michael Brunner - UFBA
Resumo
RANGEL D.R. Um contexto algébrico para uma teoria de conjuntos baseada

nos números surreais. 2018. 116+vi f. Tese (Doutorado) - Instituto de Matemática
e Estatı́stica, Universidade de São Paulo, São Paulo, 2018.
A noção de número surreal foi introduzida por J.H. Conway em meados da década
de 1970: os números surreais constituem uma classe (própria) linearmente ordenada
N o contendo a classe de todos os números ordinais (On) e que, trabalhando dentro da
base conjuntista NBG, pode ser definida por uma recursão na classe On. Desde então,
apareceram muitas construções desta classe e foi isolada uma axiomatização completa
desta noção que tem sido objeto de estudo devido ao grande número de propriedades
interessantes, incluindo entre elas resultados modelos-teóricos. Tais construções sugerem
fortes conexões entre a classe N o de números surreais e as classes de todos os conjuntos
e todos os números ordinais.
Na tentativa de codificar o universo dos conjuntos diretamente na classe de números
surreais, encontramos algumas pistas que sugerem que esta classe não é adequada para
esse fim. O presente trabalho é uma tentativa de se obter uma ”teoria algébrica (de
conjuntos) para números surreais” na linha da Teoria dos Algébrica dos Conjuntos - uma
teoria categorial de conjuntos introduzida nos anos 1990: estabelecer links abstratos e
gerais entre a classe de todos números surreais e um universo de ”conjuntos surreais”
semelhantes às relações entre a classe de todos os ordinais (On) e a classe de todos
os conjuntos (V ), que também respeite e expanda os links entre as classes linearmente
ordenadas de todos ordinais e de todos os números surreais.
Introduzimos a noção de álgebra surreal (parcial) (SUR-álgebra) e exploramos al-
gumas das suas propriedades categoriais, incluindo SUR-álgebras (relativamente) livres
(SA, ST ). Nós estabelecemos links, em ambos os sentidos, entre SUR-álgebras e álgebras
ZF (a pedra angular da Teoria Algébrica dos Conjuntos). Desenvolvemos os primeiros
passos de um determinado tipo de teoria de conjuntos baseada (ou ranqueada) em
números surreais, que expande a relação entre V e On.
Palavras-chave: Números surreais; Teoria Algébrica dos Conjuntos; SUR-álgebras.
i
ii
Abstract
RANGEL D.R. An algebraic framework to a theory of sets based on the sur-

real numbers. 2018. 116+vi pp. PhD thesis- Instituto de Matemática e Estatı́stica,
Universidade de São Paulo, São Paulo, 2018.
The notion of surreal number was introduced by J.H. Conway in the mid 1970’s: the
surreal numbers constitute a linearly ordered (proper) class N o containing the class of
all ordinal numbers (On) that, working within the background set theory NBG, can be
defined by a recursion on the class On. Since then, have appeared many constructions of
this class and was isolated a full axiomatization of this notion that been subject of interest
due to large number of interesting properties they have, including model-theoretic ones.
Such constructions suggests strong connections between the class N o of surreal numbers
and the classes of all sets and all ordinal numbers.
In an attempt to codify the universe of sets directly within the surreal number class,
we have founded some clues that suggest that this class is not suitable for this purpose.
The present work is an attempt to obtain an ”algebraic (set) theory for surreal numbers”
along the lines of the Algebraic Set Theory - a categorial set theory introduced in the
1990’s: to establish abstract and general links between the class of all surreal numbers
and a universe of ”surreal sets” similar to the relations between the class of all ordinals
(On) and the class of all sets (V ), that also respects and expands the links between the
linearly ordered class of all ordinals and of all surreal numbers.
We have introduced the notion of (partial) surreal algebra (SUR-algebra) and we ex-
plore some of its category theoretic properties, including (relatively) free SUR-algebras
(SA, ST ). We have established links, in both directions, between SUR-algebras and ZF-
algebras (the keystone of Algebraic Set Theory). We develop the first steps of a certain
kind of set theory based (or ranked) on surreal numbers, that expands the relation be-
tween V and On.
Keywords: Surreal numbers; Algebraic Set Theory; SUR-algebras.
iii
iv
Contents
Contents v
Introduction 1
1 Preliminaries 9
1.1 Set theoretic backgrounds . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.1 NBG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.2 Well founded relations . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.3 Cuts and densities . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 A glance on Algebraic Set Theory . . . . . . . . . . . . . . . . . . . . . . 14
1.2.1 Set theory and ZF-algebras . . . . . . . . . . . . . . . . . . . . . 14
1.2.2 Standard ZF-algebras . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3 On Surreal Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.1 Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.2 The axiomatic approach . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.3 Ordinals in No . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2 Introducing Surreal Algebras 27

2.1 Axiomatic definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Examples and constructions . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.1 The surreal numbers as SUR-algebras . . . . . . . . . . . . . . . . 29
2.2.2 The free surreal algebra . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.3 The free transitive surreal algebra . . . . . . . . . . . . . . . . . . 36
v
vi CONTENTS
2.2.4 The cut surreal algebra . . . . . . . . . . . . . . . . . . . . . . . . 39

2.3 Partial Surreal Algebras and morphisms . . . . . . . . . . . . . . . . . . 41
2.3.1 Simple examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3.2 Cut partial Surreal Algebras . . . . . . . . . . . . . . . . . . . . . 46
2.3.3 Simple constructions on pSUR . . . . . . . . . . . . . . . . . . . . 47
2.3.4 Directed colimits of partial Surreal Algebras . . . . . . . . . . . . 50
2.4 Universal Surreal Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3 SUR-algebras and ZF-algebras 61

3.1 Anchored Surreal Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2 The main diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.3 The Hereditary Positive members of an equipped SUR-algebra . . . . . . 80
3.4 Spaces of Signs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4 Set Theory in the free Surreal Algebra 93

4.1 Set Theory in the Surreal Algebra SA . . . . . . . . . . . . . . . . . . . . 94
4.1.1 The maps (·)p and (·)n . . . . . . . . . . . . . . . . . . . . . . . . 94
4.1.2 Inclusion, union, intersection and power . . . . . . . . . . . . . . 95
4.1.3 Singletons, doubletons and ordered pairs . . . . . . . . . . . . . . 97
4.1.4 Cartesian product, relations and functions . . . . . . . . . . . . . 98
j∗
4.2 Logical analysis of the functions V SA . . . . . . . . . . . . . . . . . . 99
b∗
4.3 Cat+ (SA) and the functor induced by j + : V → SA . . . . . . . . . . . . 101
5 Conclusions and future works 109

Introduction
The notion of surreal number was introduced by J.H. Conway in the mid 1970’s: the
surreal numbers constitute a linearly ordered (proper) class N o containing the class of
all ordinal numbers (On) that, working within the background set theory NBG, can be
defined by a recursion on the class On. Since then, have appeared many constructions
of this class and was isolated a full axiomatization of this notion.
Surreal numbers have been subject of interest in many areas of Mathematics due to
large number of interesting properties that they have:
- In Algebra, through the concept of field of Hahn series and variants (see for instance
[Mac39], [All62], [Sco69], [vdDE01], [KM15]);
- In Analysis (see the book [All87]);
- In Foundations of Mathematics, particularly in Model Theory, since the surreal number
line is for proper class linear orders what the rational number line Q is for the countable:
surreal numbers are the (unique up to isomorphism) proper class Fraı̈ssé limit of the
finite linearly ordered sets, they are set-homogeneous and universal for all proper class
linear orders.
The plethora of aspects and applications of the surreals maintain the subject as an
active research field. To make a point, the 2016 edition of the ”Joint Mathematics
Meetings AMS” –the largest Math. meeting in the world– have counted 14 talks in its
”AMS-ASL Special Session on Surreal Numbers”.
http://jointmathematicsmeetings.org/meetings/national/jmm2016/2181_program_
ss16.html
In our thesis we try to develop, from scratch, a new (we hope!) and complementary
foundational aspect of the Surreal Number Theory: to establish, in some sense, a set
theory based on the class of surreal numbers.
Set/class theories are one of the few fundamental mathematical theories that holds
the power to base other notions of mathematics (such as points, lines, and real numbers).
This is basically due to two aspects of these theories: the first is that the basic entities
and relations are very simple in nature, relying only on the primitive notions of set/class
and a (binary) membership relation (“X ∈ Y ”), the second aspect is the possibility
that this theory can perform constructions of sets by several methods. This combination
1
2 CONTENTS
of factors allows to achieve a high degree of flexibility, in such a way that virtually all
mathematical objects can be realized as being of some kind of set/class, and it has the
potential to define arrows (category) as entities of the theory. In particular, the set/class
theories traditionally puts as a principle (the Axiom of Infinity) the existence of an
”infinite set” - the smallest of these would be the set of all natural numbers - thus, such
numbers are a derived (or ”a posteriori) notion, which encodes the essence of the notion
of ”to be finite”, that is apparently more intuitive.
The usual set/class theories (as ZFC or NBG) have the power of ”encode” (syntac-
tically) its Model Theory: constructions of models of set theory by the Cohen forcing
method or through boolean valued models method are done by a convenient encoding of
the fundamental binary relations ∈ and =.
Let us list below some other fundamental theories:
• Set theories with additional predicates for non-Standard Analysis, as the E. Nelson’s
set theory named IST.
• P. Aczel’s ”Non-well-founded sets” ([Acz88]), where sets and proper classes are
replaced by directed graphs (i.e., a class of vertices endowed with a binary relation)
• K. Lopez-Escobar ”Second Order Propositional Calculus” ([LE09]), a theory with
three primitive terms, that encodes the full Second Order Intuitionistic Propositional
Calculus also includes Impredicative Set Theory.
• Toposes, a notion isolated in the 1970’s by W. Lawvere and M. Tierney, provide
generalized set theories, from the category-theoretic point of view.
• Algebraic Set Theory (AST), another categorial approach to set/class theory, in-
troduced in the 1990’s by A. Joyal and I. Moerdijk ([JM95]).
Algebraic Set Theory replaces the traditional use of axioms on pertinence by catego-
rial relations, proposing the general study of ”abstract class categories” endowed with
a notion of ”small fibers maps”. In the same way that the notion of “ being finite ”
is given a posteriori in ZFC, after guaranteeing an achievement of the Peano axioms -
which axiomatizes the algebraic notion of free monoid in 1 generator - the notions of
”to be a set” and ”be an ordinal” are given a posteriori in AST. The class of all sets
is determined by a universal property, that of free ZF-algebra, whereas the class of all
ordinals is characterized globally by the property of constituting free ZF-algebra with
inflationary/monotonous successor function. In the same direction, the (small fibers)
rank map, ρ : V → On, is determined by the universal property of V , and the inclusion
map, i : On → V , is given by an adjunction condition.
The main aim of this work is to obtain an ”algebraic theory for surreals” along the
lines of the Algebraic Set Theory: to establish abstract and general links between the
class of all surreal numbers and a universe of ”surreal sets” similar to (but expanding it)
CONTENTS 3
the (ZF-algebra) relations between the classes On and V , giving the first steps towards
a certain kind of (alternative) ”relative set theory” (see [Fre16] for another presentation
of this general notion).
In more details:
We want to perform a construction (within the background class theory NBG) of a
”class of all surreal sets”, V ∗ , that satisfies, as far as possible, the following requirements:
• V ∗ is an expansion of the class of all sets V , via a map j ∗ : V → V ∗ .
• V ∗ is ranked in the class of surreal numbers N o, in an analogous fashion that V
is ranked in the class of ordinal numbers On. I.e., expand, through the injective map
ρ
j : On → N o, the traditional set theoretic relationship V On to a new setting
i
ρ∗
V ∗ N o.
i∗
Noting that:
(i) the (injective) map j : On → N o, is a kind of ”homomorphism”, partially encoding
the ordinal arithmetic;
(ii) the traditional set-theoretic constructions (in V ) keep some relation with itsS(ordinal)
complexity
S (e.g., x ∈ y → ρ(x) < ρ(y), ρ({x}) = ρ(P (x)) = ρ(x) + 1, ρ( i∈I xi ) =
i∈I ρ(xi ));
then we wonder about the possibility of this new structured domain V ∗ determines a
category, by the encoding of arrows (and composition) as objects of V ∗ , in an analogous
fashion to the way that the class V of all sets determines a category, i.e. by the encoding
of some notion of ”function” as certain surreal set (i.e. an objects of V ∗ ); testing, in
particular, the degree of compatibility of such constructions with the map j ∗ : V → V ∗
and examining if this new expanded domain could encode homomorphically the cardinal
arithmetic.
We list below 3 instances of communications that we have founded in our bibliographic
research on possible themes relating surreal numbers and set theory: we believe that they
indicate that we are pursuing a right track.
(I) The Hypnagogic digraph and applications
J. Hamkins have defined in [Ham13] the notion of ”hypnagogic digraph”, (Hg, *), an
acyclic digraph graded on (N o, <), i.e., it is given a ”rank” function v : Hg → N o such
that for each x, y ∈ Hg, if x * y, then v(x) < v(y). The hypnagogic digraph is a
proper-class analogue the countable random Q-graded digraph: it is the Fraı̈ssé limit
of the class of all finite N o-graded digraphs. It is simply the On-saturated N o-graded
class digraph, making it set-homogeneous and universal for all class acyclic digraphs.
Hamkins have applied this structure, and some relativized versions, to prove interesting
results concerning models of ZF set theory.
4 CONTENTS
(II) Surreal Numbers and Set Theory

https://mathoverflow.net/questions/70934/surreal-numbers-and-set-theory
Asked July 21, 2011, by Alex Lupsasca:
I looked through MathOverflow’s existing entries but couldn’t find a satisfactory answer to the
following question:
What is the relationship between N o, Conway’s class of surreal numbers, and V , the Von
Neumann set-theoretical universe?
In particular, does V contain all the surreal numbers? If so, then is there a characterization of
the surreal numbers as sets in V ? And does N o contain large cardinals?
I came across surreal numbers recently, but was surprised by the seeming lack of discussion of
their relationship to traditional set theory. If they are a subclass of V , then I suppose that could
explain why so few people are studying them.
(III) Surreal Numbers as Inductive Type?
https://mathoverflow.net/questions/63375/surreal-numbers-as-inductive-type?rq=1
Asked in April 29, 2011, by Todd Trimble:
Prompted by James Propp’s recent question about surreal numbers, I was wondering whether
anyone had investigated the idea of describing surreal numbers (as ordered class) in terms of a
universal property, roughly along the following lines.
In categorical interpretations of type theories, it is common to describe inductive or recursive
types as so-called initial algebras of endofunctors. The most famous example is the type of
natural numbers, which is universal or initial among all sets X which come equipped with an
element x and an function f : X → X. In other words, initial among sets X which come
equipped with functions 1 + X → X (the plus is coproduct); we say such sets are algebras of the
endofunctor F defined by F (X) = 1 + X. Another example is the type of binary trees, which
can be described as initial with respect to sets that come equipped with an element and a binary
operation, or in other words the initial algebra for the endofunctor F (X) = 1 + X 2 .
In their book Algebraic Set Theory, Joyal and Moerdijk gave a kind of algebraic description
of the cumulative hierarchy V . Under some reasonable assumptions on a background category
(whose objects may be informally regarded as classes, and equipped with a structure which allows
a notion of ”smallness”), they define a ZF-structure as an ordered object which has small
sups (in particular, an empty sup with which to get started) and with a ”successor” function.
Then, against such a background, they define the cumulative hierarchy V as the initial ZF-
structure, and show that it satisfies the axioms of ZF set theory (the possible backgrounds allow
intuitionistic logic). By tweaking the assumptions on the successor function, they are able to
describe other set-theoretic structures; for example, the initial ZF-structure with a monotone
successor gives On, the class of ordinals, relative to the background.
Now it is well-known that surreal numbers generalize ordinals, or rather that ordinals are special
numbers where player R has no options, or in different terms, where there are no numbers past
the ”Dedekind cut” which divides L options from R options. In any case, on account of the
highly recursive nature of surreal numbers, it is extremely tempting to believe that they too
can be described as a recursive type, or as an initial algebraic structure of some sort (in a
background category along the lines given by Joyal-Moerdijk, presumably). But what would it
CONTENTS 5
be exactly?
I suppose that if I knocked my head against a wall for a while, I might be able to figure it out or
at least make a strong guess, but maybe someone else has already worked through the details?
Chronology of the work:

This works begins at the second semester of 2014, dedicated to the study of the basic
results on surreal numbers exposed on books and trying to prospect a new/complementary
approach to this vast theory with connections in many branches of Mathematics.
In the first half of 2015 we wonder if N o is such a fundamental object, it should be
some way a basis for a Set Theory. To aid us to materialize such idea we have dedicated
almost all the semester to the study of the principles of Algebraic Set Theory [JM95].
At the 2nd semester 2015 we begin the preparation for qualification exam, that has
occurred in November 2015, thus we begin the search of articles and the partially study
them. In this meantime we came across Hamkin’s paper [Ham13] and two Mathoverflow
links mentioned above that have indicated that we are pursuing a right track.
In first semester 2016 we focus on the main theme of the thesis according its title
”An algebraic framework to a theory of sets based on the surreal numbers”, and effec-
tively begins the novelties of work, as the definition of the concept of SUR-algebras and
its morphisms and the SA example, that we continue until the first months of 2018,
simultaneously with the begin of the typing process.
Overview of this thesis:

Chapter 1:
This initial chapter establishes the notations and contains the preliminary results needed
for the sequel of this thesis. It begins establishing our set theoretic backgrounds – that
we will use freely in the text without further reference – which is founded in NBG class
theory, and contains mainly the definitions and basic results on some kinds of binary
relations, in particular on well-founded relations, and ”cuts” as certain pairs of subsets
of a class endowed with a binary irreflexive relation. After, in the second section, we
introduce briefly a (categorically naive) version of ZF-algebras, a notion introduced in
the 1990’s in the setting of Algebraic Set Theory, in particular we introduce the concept
of standard ZF-algebra, very useful to developments occurred in the Chapter 3. The
last section is dedicated to introduce the linearly ordered class of surreal numbers under
many equivalent constructions and to present a characterization and some of its main
structure, including its algebraic structure and its relations with the class (or ZF-algebra)
of all ordinal numbers.
Chapter 2:
Motivated by properties of the linearly ordered class (N o, <), we introduce the notion
6 CONTENTS
of Surreal Algebra (SUR-algebra): an structure S = (S, <, ∗, −, t), where < is an acyclic
relation on S, ∗ is a distinguished element of S, − is an involution of S and t is a
function that chooses an intermediary member between each small (Conway) cut in
(S, <), satisfying some additional compatibility properties between them. Every SUR-
algebra turns out to be a proper class. We verify that N o provides naturally a SUR-
algebra and present new relevant examples: the free surreal algebra (SA) and the free
transitive SUR-algebra (ST ). In the sequel, a section is dedicated to a generalization
of this new concept: we introduce the notion of partial SUR-algebra and consider two
kinds on morphisms between them. This relaxation is needed to perform constructions
as products, sub partial-SUR-algebra and certain kinds of directed colimits. As an
application of the latter construction, we are able to prove some universal properties
satisfied by SA and ST (and generalizations), that justifies its names of (relatively) free
SUR-algebras.
Chapter 3:
This chapter is dedicated to established links, in both directions, between certain classes
of (equipped) SUR-algebras and certain classes of ZF-algebras (the keystone of Algebraic
j
Set Theory), that ”explains” and ”expands” the relationship On N o . Even if some
b
general definitions and results are given in this chapter, its main goal is not develop an
extensive and systematic study of the introduced concepts (we intend dedicate to this in
the future) but, instead, to provide means to understand and appreciate the content of
the main diagram presented in the Section 2, that summarizes the relationship between
the structured classes On, V, N o, SA.
Inspired by the fundamental properties of the ”birthday” function on surreal number,
b : N o On, we introduce, in the first section, the notion of anchored SUR-algebra,
(S, b), where b : S → C is a certain function from the SUR algebra S onto a (rooted)
well-founded class C = (C, ≺), satisfying some convenient properties. This induces an
useful well-founded relation ≺b in S and a recursively-defined subclass of HPb (S)⊆S
of hereditary positive members that, under additional conditions, provides an induced
ZF-algebra structure (e.g., HPb∗ (SA) ∼ = V, HPb (N o) ∼= On) and also axiomatizations
results for SA and ST , analogous to the axiomatization of N o, describing them up to
isomorphisms.
In the Section 2 we present the ”main diagram” of this thesis, that summarizes the
relationship between the structured classes On, V, N o, SA and will be the basis for the
development of Chapter 4: (i) the maps ρ : V → On and ρ∗ : SA → N o connect objects
of a same category (of ZF-algebras in the former case and of SUR-algebras in the latter);
j j∗
(ii) the pairs of arrows On N o and V SA are ”Chimera morphisms”, i.e. each pair
b b∗
establishes connections between ZF-algebras and SUR-algebras.
In the two last sections, we introduce the concept of hereditary positive members of
a SUR-algebra endowed with certain well-founded relation and associate to some (well-
founded) ZF-algebras Z corresponding SUR-algebra space of signs Sig(Z) (e.g. Sig(V ) ∼
=
CONTENTS 7
SA, Sig(On) ∼= N o). Every SUR-algebra ”space of signs” is anchored on its underlying
well-founded ZF-algebra, that is its class of hereditary positive members.
Chapter 4:
We develop here the first steps of a certain kind of set theory based (or ranked) on surreal
numbers, that expands the relation between V and On.
For us, there are three main requirements for a theory deserves be named a ”Set Theory”:
(i) have the potential to define arrows (category) as entities of the theory, through a
fundamental binary relation; (ii) be the ”derived set theory” of a free object in a category
(like in ZF-algebra setting); (iii) its ”internal” category is a topos-like category.
In fact, we work out a ”positive set theory” on SA ranked on N o, that expands the
ZF-algebra relationship V → On through the ”positive” map j + : V → SA, given by
j + (X) = j ∗ (X) = hj + [X], ∅i, X ∈ V . Thus the free/initial SUR-algebra SA supports, in
many senses, an expansion of the free/initial ZF-algebra V and its underlying set theory.
To accomplish this, we use the functions j ∗ and b∗ that establishes connections between V
and SA. Under logical and set-theoretical perspective, the map j + : V → SA preserves
and reflects many constructions. On the category-theoretic perspective, the map j +
defines a full, faithful and logical functor j + : Set → Cat+ (SA), from the topos Set
associated to V into the topos Cat+ (SA) associated to SA.
Chapter 5:
In the this last chapter, we present a (non-exhaustive) list of questions that have occurred
to us during the elaboration of this thesis, that we can not be able to deal in the present
work by lack of time and/or of skills, but that we intend to address in the future.
8 CONTENTS
Chapter 1
Preliminaries
This chapter establishes the notations and contains the preliminary results needed for
the sequel of this thesis. It begins establishing our set theoretic backgrounds – that we
will use freely in the text without further reference – which is founded in NBG class
theory, and contains mainly the definitions and basic results on some kinds of binary
relations. After, we introduce briefly a (categorically naive) version of ZF-algebras, a
notion introduced in the 1990’s in the setting of Algebraic Set Theory. Finally, we present
the class of surreal numbers, and some of its main structure, under many equivalent
constructions.
1.1 Set theoretic backgrounds

This preliminary section is devoted to establishing our set theoretic backgrounds which
is founded in NBG class theory1 , and contains mainly the definitions and basic results on
the binary relations that will appear in the sequel of this work as: (strict) partial order
relations, acyclic relations, extensional relations, well founded relations, and ”cuts” as
certain pairs of subsets of a class endowed with a binary irreflexive relation.
1.1.1 NBG
In this work, we will adopt the (first-order, with equality) theory NBG as our background
set theory, where the unique symbol in the language is the binary relation ∈. We will use
freely the results of NBG, in the sequel, we just recall below some notions and notations.
We recall also the basic notions and results on some kinds of binary relations needed for
1
In some parts of the thesis, we will need some category-theoretic tools and reasoning, thus we will
use an expansion of NBG by axioms asserting the existence of Grothendieck universes.
9
10 CHAPTER 1. PRELIMINARIES
the development of this thesis. Standard references of set/class theory are [Jec03] and
[Kun13].
1. On NBG:
Recall that the primitive concept of NBG is the notion of class. A class is improper
when it is a member of some class, otherwise the class is proper. The notion of set in
NBG is defined: a set is a improper class.
We will use V to denote the universal class – whose members are all sets – On will
stand for the class of all ordinal numbers and T r denote the class of all transitive sets.
On⊆T r⊆V and all the three are proper classes.
Given classes C and D, then C is a subclass of D (notation: C⊆D), when all members
of C are also members of D. Classes that have the same members are equal. Every
subclass of a set is a set.
∅ stands for the unique class without members. ∅ is a set.
Given a class C, denote Ps (C) the class whose members are all the subsets of C. If C
is a set, then Ps (C) is a set too. There is no class that has as members all the subclasses
of a proper class2 .
Given classes C and D, and a function f : C → D, then the (direct) image f [C] =
{d ∈ D : ∃c ∈ C, d = f (c)} is a subset of D, whenever C is a set.
Since NBG satisfies the axiom of global choice (i.e., there is a choice function on
V \ {∅}) and then every class (proper or improper) can be well-ordered, which implies
nice cardinality results: as in ZFC, any set X is equipotent to a unique cardinal number
(= initial ordinal), called the its cardinality of X (notation: card(X)); moreover, all the
proper classes are equipotent – we will denote card(C) = ∞ the cardinality of the proper
class C – ∞ can be seen as a representation of the well-ordered the proper class On.

2. Binary relations:
A relation R is a class whose members are ordered pairs. The domain (respect.,
range) of R is the class of all first (respect., second) components of the ordered pair in
the relation. The support of the relation R (notation: supp(R)) is the class obtained
by the reunion of its domain and range. We will say that a binary relation is defined
on/over its support class.
A relation R is reflexive when (x, x) ∈ R for each x in the support of R; on the other
hand, R is irreflexive, when (x, x) ∈
/ R for each x in the support of R. We will use <, ≺, /
to denote general irreflexive relations; ≤, , v will stand for reflexive relations.
2
This is a ”metaclass” in NBG, i.e., an equivalence class of formulae in one free variable, modulo the
N BG-theory: any such formula is not collectivizing.
1.1. SET THEORETIC BACKGROUNDS 11
A pre-order is a reflexive and transitive relation. A partial order is a antisymmetric

pre-order. A strict partial order is a irreflexive and transitive relation. There are well
known processes of: obtain a strict partial order from a partial order and conversely.
Let R be a binary relation and let s, s0 ∈ supp(R). Then s and s0 are R-comparable
when: s = s0 or (s, s0 ) ∈ R or (s0 , s) ∈ R. A relation R is total or linear when every pair
of members of its support are comparable.
Every pre-order relation on a class C gives rise to an equivalence relation ∼ on the
same support: for each c, c0 ∈ C, c ∼ c0 iff c c0 and c0 c.
Let n ∈ N, a n-cycle of the relation R is a n + 1-tuple (x0 , x1 , · · · , xn ) such that
xn = x0 and, for each i < n, (xi , xi+i ) ∈ R. A relation is acyclic when it does not have
cycles. Every acyclic relation is irreflexive. A binary relation is a strict partial order iff
it is a transitive and acyclic relation. Note that a binary relation is acyclic and total iff
it is a strict linear order.

3. Induced binary relations:

Given a binary relation R on a class C. For each c ∈ C, denote cR := {d ∈ C : (d, c) ∈
R}.
Define a new binary relation on C: for each c, c0 ∈ C, c vR c0 iff holds ∀x((x, c) ∈
R → (x, c0 ) ∈ R) or, equivalently, cR ⊆c0 R . Clearly, vR is pre-order relation on C.
Denote ≡R , the equivalence relation associated to the pre-order vR . We will say
that the binary R is extensional when ≡R is the identity relation on C or, equivalently,
when vR is a partial order. The axiom of extensionality in NBG ensures that (V, ∈V ×V )
is a class endowed with an extensional relation and, since members of ordinal numbers
are ordinal numbers3 , then (On, ∈On×On ) is class endowed with an extensional relation.

1.1.2 Well founded relations
In this subsection we recall basic properties and constructions concerning general well-
founded relations. Also, we introduce a special kind of well-founded relation suitable for
our purposes in Chapter 3.
4. On well-founded relations:
Recall that a binary relation ≺ on a class C is well-founded when:
(i) The subclass x≺ = {y ∈ C : y ≺ x} is a set.
3
If α ∈ On, then α∈ = {β ∈ On : β ∈ α} = {x ∈ V : x ∈ α} = α.)
(ii) For each X⊆C that is a non-empty subset, there is u ∈ X that is a ≺-minimal
/ X).4
member of X (i.e., ∀v ∈ C, v ≺ u⇒v ∈
Let ≺ be an well-founded relation on a class C. Since for each n ∈ N, the (non-empty)
subset {x0 , · · · , xn }⊆C has a ≺-minimal member, then ≺ is an acyclic relation and, in
particular, ≺ is irreflexive.
If D⊆C, then (D, ≺D×D ) is an well-founded class.
An well-founded relation that is a strict linear/total order is a well-order relation.
The axiom of regularity in NBG, guarantees that the binary relation ∈ over the
universal class V is an well-founded relation. (On, ∈) is an well-ordered proper class.
Let ≺ be an well-founded relation on a class C. Then it holds:
The induction principle: Let X⊆C be a subclass. If, for each c ∈ C, the inclusion
c≺ ⊆X entails c ∈ C, then X = C.
The recursion theorem: Let H be a (class) function such that H(c, g) is defined for
each c ∈ C and g a (set) function with domain c≺ . Then there is a unique (class) function
F with domain C such that F (c) = H(c, Fc≺ ), for each c ∈ C.

5. Rooted well-founded relations:
Remark: Let (C, ≺) be a well-founded class; the subclass root(C) of its roots has as
members its ≺-minimal members. Note that:
• If C is a non-empty class, then root(C) is a non-empty class.
• If v denotes the pre-order on C associated to ≺ (i.e., c v d iff ∀x ∈ C(x ≺ c⇒x ≺ d),
then: root(C) = {a ∈ C : a v c, for all c ∈ C}.
Definition: A well-founded class (C, ≺) will be called rooted, when it has a unique
root Φ. If it is the case, then the structure (C, ≺, Φ) will be called a rooted well-founded
class.
If ≺ is an extensional well-founded relation on a non-empty class C, then (C, ≺) is
rooted: indeed, if r, r0 ∈ root(C), then r v r0 and r0 v r, thus r = r0 . However, to
emphasize the distinguished element in a structure of rooted well-founded class, we will
employ the redundant expression ”rooted extensional well-founded class”.
Examples and counter-examples:
(V, ∈, ∅) is a rooted extensional well-founded class
(On, ∈, ∅) is a rooted extensional well-ordered class.
Every well-ordered set (X, ≤) gives rise to a rooted extensional well-ordered set (X, <
, Φ), where Φ = ⊥ is the least element of X and the strict relation, <, associated to ≤,
4
By the global axiom of choice (for classes), this condition is equivalent of a apparently stronger one:
(ii)’ For each X⊆C that is a non-empty subclass, there is u ∈ X that is a ≺-minimal member of X.
1.1. SET THEORETIC BACKGROUNDS 13
is an well-founded relation, since for each x, y ∈ X, x< ⊆y < iff x ≤ y.

(N, ≤, 0) is a well-ordered set, thus it gives rise to a rooted extensional well-ordered
set. (N \ {0}, |, 1) is determines a rooted well-founded set that is not extensional. Note
that (N \ {0, 1}, |) is an well-founded set that is not rooted since its subset of minimal
elements is the infinite set of all prime numbers.

1.1.3 Cuts and densities
Many of useful variants of the concept of Dedekind cut were already been defined on the
setting strict linear order on a given set (see for instance [All87]). In this preliminary
subsection we present expansions of these notions in two different directions: we consider
binary relations that are only irreflexive (instead of being a strict linear order) and defined
on general classes instead of improper classes (= sets). We also generalize the notions of
density à la Hausdorff to this new setting.
Through this subsection, C denote a class and < stands for a irreflexive binary relation
whose support is C.
6. Cuts
A Conway cut in (C, <) is a pair (A, B) of arbitrary subclasses5 of C such that
∀a ∈ A, ∀b ∈ B, a < b (notation: A < B). Since < is a irreflexive relation on C, then
A ∩ B = ∅. A Conway cut (A, B) will be called small, when A and B are subsets of C.
We can define in theory NBG the class Cs (C, <) := {(A, B) ∈ Ps (C) × Ps (C) : A < B},
formed by all the small Conway cuts in (C, <).
A Cuesta-Dutari cut in (C, <) is a Conway cut (A, B) such that A ∪ B = C. Note
that (∅, C) and (C, ∅) are always Cuesta-Dutari cuts in (C, <). On the other hand, if C is
a proper class, then the class CDs (C, <) of all small Cuesta-Dutari cuts in (C, <) is the
empty class.
A Dedekind cut in (C, <) is a Cuesta-Dutari cut (A, B) such that A and B are
non-empty subclasses. If C is a set, then (A, B) is a Dedekind cut in (C, <) iff (A, B) is
a Conway Cut such that the set {A, B} is a partition of C.

7. Densities
Let α be an ordinal number. Then (C, <) will be called an ηα -class, when for each
small Conway cut (A, B) in (C, <), such that card(A), card(B) < ℵα , there is some t ∈ C
such that ∀a ∈ A, ∀b ∈ B, (a < t, t < b) (notation: A < t < B).
5
A and/or B could be the empty set.
Let (C, <) be an ηα -class. Taking cuts (∅, {c}) (respectively ({c}, ∅)), for all c ∈ C,
we can conclude that an ηα -class (C, <) does not have <-minimal (respec. <-maximal)
elements. Taking cuts (∅, X) (or (X, ∅)), for all X⊆C such that card(X) < ℵα , we see
that an ηα -class (C, <) has card(C) ≥ ℵα .
An η0 -class (C, <) is just a ”dense and without extremes” class.
If (C, <) is an ηα -class and β ∈ On is such that β ≤ α, then clearly (C, <) is an
ηβ -class.
(C, <) will be called an η∞ -class, when it is an ηα -class for all ordinal number α:
this means that for each small Conway cut (A, B) in (C, <) there is some t ∈ C such
that A < t < B. Every η∞ -class is a proper class. We will see in the Section 3 in this
Chapter that the strictly linearly ordered proper class of all surreal numbers (N o, <) is
η∞ . We will introduce in Chapter 2 the notion of SUR-algebra: every such structure is
a η∞ -class.

From now on, we will use the notion of Conway cut only in the small sense.
1.2 A glance on Algebraic Set Theory

In this section, we introduce briefly a (categorially naive) version of ZF-algebras, a no-
tion introduced in the 1990’s in the setting of the so called ”Algebraic Set Theory”
(AST), see [JM95]. Our main goal here is to provide an intermediary step between
the usual axiomatic set/class theories and our algebraic approach to obtain something
like ”a set theory based on surreal numbers”, materialized by the concept of ”surreal-
algebra” (SUR-algebra), introduced in Chapter 2 and further developed in Chapter 3 (in
connection with ZF-algebras).
An analogy can be useful to understand the point of the AST approach: as in tradi-
tional set/class theories, the notion ”be a infinite set” is defined by a convenient universal
property (ω is the ⊆-least inductive set) and, after that, the notion of ”a set x is finite”
can be formalized (x is equipotent to a member of the set ω), in algebraic set theory,
the notion of ”be a set” or ”be an ordinal” is given a priori, through convenient uni-
versal properties among the ZF-algebras, and the notion of ”membership” is defined a
posteriori.
1.2.1 Set theory and ZF-algebras
The NBG class theory provides, naturally, a grading by size of the objects of the theory:
the proper classes are the ”large” ones (all of them have the same size) and the sets or
1.2. A GLANCE ON ALGEBRAIC SET THEORY 15
improper classes are the ”small” objects6 .

A similar size distinction can be adapted to the arrows in the theory:
8. Small functions in NBG:

Definition: Let f : C → D be a class function.
(i) f is called locally small when, for each subset D0 ⊆D, f −1 [D0 ]⊆C is also a subset.
(ii) f has small fibers when, for each d ∈ D,f −1 [{d}] is a subset of C.
Remark: Since f −1 [D0 ] = d∈D0 f −1 [{d}], a function is locally small iff it has small
S
fibers.
Example: The rank function ρ : V → On has small fibers.
Recall theSrecursively defined Von Neumann cumulative hierarchy
S of sets: for each ordinal
α, Vα = β∈α Ps (Vβ ). By the regularity axiom in NBG, V = α∈On Vα . For each set x,
ρ(x) := min{α ∈ On : x⊆Vα } = min{α ∈ On : x ∈ Vα+1 }, thus ρ−1 [{α}] = Vα+1 \ Vα
is a set, thus ρ : V → On has small fibers. Recall that ρ is a retraction of the (small
fibers) inclusion i : On ,→ V .

Instead of the general (and original) higher technical categorial setting of AST – a
Heyting pretopos with natural numbers object (to represent a category whose objects
are ”big sets”) endowed with a class of arrows satisfying the axioms for small maps –, we
adopt the below described ”naive” approach to algebraic set theory: our base category
is the category whose objects are classes7 and the small maps are just the ones that we
have considered above.
9. ZF-algebras and its morphisms:

V as a ZF-algebra: We can arrive to the concept of ZF-algebra regarding some
distinctive properties of the universal class V :
- Ps (V ) is a ”large” (= proper class) small-complete sup-lattice by the ⊆-relation, where
the small suprema are given by reunions;
- V = Ps (V );
- u : V → Ps (V ) x 7→ {x} is an endofunction.
W W
Definition: A ZF-algebra is a structure L = (L, , s), where L is a class, (L, ) is a
small-complete sup-lattice, and s : L → L is an endofunction, called successor function
of the ZF-algebra L.
6
We will use the index ”s”, as in Ps (X), to indicate the small/set character of an object of the theory.
7
To be more precise, a set theoretical setting to formalize such ”very-large” categories (the category
of classes is a very-large category, since it has a metaclass of objects and a metaclass of arrows) include,
for instance, NBG (or ZFC) plus axioms asserting the existence of at least two distinct Grothendieck
universes U0 and U1 , lets say U0 ∈ U1 , then: the members of U0 are (represent the) ”sets”, the subsets
of U0 are classes (they belong to U1 ) and the ”meta-classes” are the subsets of U1 .
W
Derived binary relations: Let L = (L, , s) be a ZF-algebra, then there are two
useful binary relations on L naturally
W defined:
- partial order relation: x ≤ y iff {x, y} = y, x, y ∈ L;
- ”membership” relation: xεy iff s(x) ≤ y, x, y ∈ L.
ZF-algebra morphisms: Let L = (L, , s) and L0 = (L0 , 0 , s0 ) be ZF-algebras. A
W W
ZF-algebra morphism h : L → L0 is a (class) function h : L → L0 such that h preserves
small suprema and preserves successor functions (i.e., h ◦ s = s0 ◦ h).
The category ZF − alg: Of course, we can define a ”very-large” category ZF − alg,
whose objects are the ZF-algebras and the arrows are the ZF-algebra morphisms, with
obvious composition and identities.

W
10. On special kinds of successor function: Let L = (L, , s) be a ZF-algebra.
Definition: The successor function s : L → L will be called an:
-Wirreducible successor: when, for all x ∈ L and all small family {yi : i ∈ I}⊆L, s(x) ≤
i∈I yi iff ∃i ∈ I, s(x) ≤ yi ;
- inflationary successor: when ∀x ∈ L, x ≤ s(x);
- order-successor: when ∀x ∈ L, x < y iff s(x) ≤ y.
Remark:
(i) If s is an inflationary successor, then ε is a transitive relation. Indeed: suppose that
yεx and let z ∈ y ε , then s(z) ≤ y ≤ s(y) ≤ x, thus z ∈ xε and y ε ⊆ xε .
(ii) Suppose that s represents an order-successor. Then:
- s is inflationary and ε is transitive;
- ∀w, z ∈ L (w < z ⇔ wεz).
(iii) Suppose that s represents an order-successor and let z ∈ L. Then the following
conditions on z ∈ L are equivalent:
(1) z is not a successor (i.e. z 6= s(y), ∀y ∈ L);
(2) z is a limit ∀y ∈ L, (s(y) ≤ z⇒s(y) < z);
(3) ∀y ∈SL, (y < z⇔s(y) < z);
(4) z ε ⊆ Syεz y ε ;
(5) z ε = yεz y ε .
Indeed, for instance (2) ⇒ (4):
Let x ∈ z ε , then s(x) ≤ z and, by (2), s(x) < z. Since s is an order-sucessor, s(s(x)) ≤ z,
and applying again the hypothesis s(s(s(x)))) ≤ S z. Set y := s(s(x)), then yεz (since
ε ε
s(y) ≤ z) and x ∈ y (since s(x) ≤ y), thus x ∈ yεz y .
(iv) Suppose that s represents an order-successor and ≤ is a linear order, then:
- ∀w, z ∈ L (w v z ⇔ wεz or w = z).
- s satisfies the condition: ∀x ∈ L, x < s(y) iff x ≤ y.

Example 11.
S
(i) V = (V, , s), s(x) = {x} is a ZF-algebra, since:
- by the pair axiom, if x is set, then {x} is a set too;
- by the reunion axiom, the reunion of a set of members of V is a set (i.e., is a member
of V ) and is the ⊆-least set that contains all of its members.
(ii) On = (On, , s), s(x) = x+ = x ∪ {x} is a ZF-algebra, since:
S
- if x is a ordinal number, then x+ is a ordinal number too;
- every set of ordinal numbers has a ⊆-supreme in On that coincide with the reunion of
that set.
(iii) T r = (T r, , s), s(x) = x+ = x ∪ {x} is a ZF-algebra, since:
S
- if x is a transitive set, then x+ is a transitive too;
- every set of transitive sets has a ⊆-supreme in T r that coincide with the reunion of
that set.
Note that, the map x 7→ x+ is, in general, ⊆-inflationary (∀x ∈ V , x⊆x+ ) and,
restricted to On, it is ⊆-increasing (∀x, y ∈ On, x⊆y ⇒ x+ ⊆y + , since x ( y iff x ∈ y).

It is clear that, between the inclusion maps On ,→ T r, On ,→ V, T r ,→ V, only the

first one is a ZF-algebra morphism.
It is well-known that the rank functionSρ : V → On can be characterized (or, alter-
natively, ∈-recursively defined) by ρ(x) = {ρ(y)+ : y ∈ x}, ∀x ∈ V . The (proof of the)
general result below shows, in particular, that ρ is a ZF-algebra morphism ρ : V → O
(in fact, it is the unique such morphism).
S
Proposition 12.The ZF-algebra V = (V, , { }) is the initial (or relatively free) ZF-
algebra.
W
Proof. Let L = (L, , s) be a ZF-algebra. We must show, that there is a unique
ZF-algebra morphism h : V → L.
Candidate
S and uniqueness:
Since x = y∈xS{y}, ∀x ∈ VW
, there is at most W morphism h : V → L such
W one ZF-algebra
that h(x) = h( y∈x {y}) = y∈x h({y}) = y∈x s(h(y)) = Range(s ◦ hx∈ ).
Existence:
Define,
W by recursion on the well-founded relation ∈, a function h : V → L by h(x) :=
Range(s ◦ hx∈ ), x ∈ V (this sup exists, since Range(s
S ◦ hx
W∈ ) is a set). Then:
•
W For each small family {x i : i ∈ I} in
W W V , h( x
i∈I i ) = W Range(s ◦ h( i∈I xi )∈ ) =
S
{s(h(y)) : y ∈ xi , for some i ∈ I} = i∈I ( y∈xi s(h(y))) = i∈I h(xi ), thus h preserves
small suprema. W W
• For each y ∈ V , h({y}) = Range(s ◦ h{y}∈ ) = {s(h(y))} = s(h(y)), thus h
preserves successors.

Proposition 13.On = (On, , ( )+ ) is the initial (or relatively free) ZF-algebra in the
S
category of ZF-algebras with inflationary successor map.
W
Proof. Let L = (L, , s) be a ZF-algebra such that ∀l ∈ L, l ≤ s(l), for instance,
L = On, T r. We must show, that there is a unique ZF-algebra morphism h : O → L.
Candidate and uniqueness:
By an analysis of cases of ordinal numbers –zero, limit or successor– (or recalling that
every member of a ordinal number is a ordinal and that ρS: V → On is a retraction of the
inclusion i : On ,→ V ), for every x ∈ On, we have x = y∈x yS+ . Then thereW is at +most
+
one
W ZF-algebra Wmorphism h : On → L such that h(x) = h( y∈x y ) = y∈x h(y ) =
y∈x s(h(y)) = Range(s ◦ hx∈ ).
Existence:
Define,
W by recursion on the well-founded relation ∈, a function h : On → L by h(x) :=
Range(s ◦ hx∈ ), x ∈ On (this sup exists, since S
Range(s ◦ hWx∈ ) is a set). Then:
•W For each small family {xi : i ∈ I}Win On, W h( i∈I xi ) = W Range(s ◦ h( i∈I xi )∈ ) =
S
{s(h(y)) : y ∈ xi , for some i ∈ I} = i∈I ( y∈xi s(h(y))) = i∈I h(xi ), thus h preserves
small suprema.
• For each y ∈ On, h(y + ) = Range(s ◦ h(y+ )∈ ) = z∈y∪{y} s(h(z)) = z∈y s(h(z)) ∨
W W W
W
{s(h(y))} = h(y) ∨ s(h(y)) = s(h(y)) (since s is assumed to be inflationary), thus h
preserves successors.

Remark 14.
(i) In the general categorial setting for algebraic set theory, but adding adequate
hypothesis (see for instance [JM95]), it can be proved the existence of a free ZF-algebra
V satisfying a universal property even stronger than in Proposition 12 and many ZF-
algebras of ”ordinal numbers” (free ZF-algebras with successor map inflationary or in-
creasing). In this upside down (abstract class) theory, there is a ”derived” set theory –
the theory of the objects in free model (= free ZF-algebra) V, where xεy is defined as
s(x) ≤ y, for each x, y ∈ V – it is an intuitionist fragment of ZF set-theory.
(ii) Note that the inclusion map i : On ,→ V is a section of the (unique) ZF-algebra
morphism ρ : V → On (i.e. ρ◦i = IdOn ), that preserves arbitrary suprema (= reunions),
preserves and reflects the binary relations = and ε (= ∈) and such that ∀x ∈ V, ∀β ∈ On,
x ∈ i(β) iff x = i(α) for some and unique α ∈ On. But i : On ,→ V does not preserves
successors. An analogous relation holds for the inclusion i : T r ,→ V (apart from that it
is not a section of ρ). In fact, as a consequence of the universal property of V , there is
no ZF -algebra morphisms h : On → V , g : T r → V .

1.2.2 Standard ZF-algebras
In this subsection we introduce a special kind of ZF-algebra suitable for our purposes in
Chapter 3.
W
Definition 15. A ZF-algebra (A, , s) is a standard ZF-algebra iff it verifies:
(s0) A has a ε-minimal member.
(s1) ε is an well founded relation.
(s2) ∀z ∈ A, z = xεz s(x) 8
W

W
Proposition 16. Let (A, , s) be a ZF-algebra that satisfies (s1), then are equivalent:
W
(i) ∀z ∈ A z = xεz s(x)
(ii) ∀w, z ∈ A (w ≤ z ⇔ w v z)
Proof.
(i) ⇒ (ii). Suppose w ≤ z and take any xεw, then s(x) W
≤ w ≤ z, thus
W xεz: this means
w v z. Conversely, suppose w v z, thus wε ⊆z ε and w = xεw s(x) ≤ xεz s(x) = z.
W (ii) ⇒ (i). Suppose that ε is an well founded W relation. Let z ∈ A and define w :=
xεz s(x). Since s(x) ≤ z, for all xεz, then w = xεz s(x) ≤ z. Let y ∈ A such that yεz,
then s(y) ≤ w, thus yεw: this means W z v w and, by (ii), we have z ≤ w. Since since ≤
is antisymmetric, we have z = w = xεz s(x).

Items (c), (d), (e) below are obtained by specifying 10 to the standard ZF-algebra
setting in the following:
W
Remark 17. Let (A, , s) is a standard ZF-algebra. Then:
(a) (A, ε) is a rooted extensional well-founded class: since ≤ is antisymmetric, by (ii)
in the Proposition 16 above, ε is extensional. Thus, by (s0) and the remarks on 5, (A, ε)
is a rooted (extensional) well-founded class.
(b) Since ε is an acyclic relation, there is no x ∈ A such that s(x) ≤ x.
(c) If s : A → A is inflationary (i.e. ∀x ∈ A, x ≤ s(x)), then ε is a transitive relation
(i.e. yεx⇒y v x, for all x, y ∈ A).
(d) Suppose that s : A → A represents an order-successor (i.e. ∀x ∈ A, x < y iff
s(x) ≤ y), in particular s : A → A is inflationary and ε is transitive. Then:
(d1) ∀w, z ∈ A (w < z ⇔ wεz).
8
Since ε is well-founded, z ε is a subset of A, thus this sup exists.
(d2) Consider the following conditions on z ∈ A:

W a successor (i.e. z 6= s(y), ∀y ∈ A);
(i) z is not
(ii) z = yεz y Then (i)⇒(ii).
W Indeed: if for all x, s(x) ≤ z⇒s(x) < z, then s(x) <
W ≤ z; since z = xεz s(x) and s(x) ≤ z, then y = s(s(x)) ≤ z and we have
s(s(x))
z = yεz y.
Moreover, if s : A → A is an irreducible successor9 , then (i)⇔(ii).
(e) Suppose that s represents an order-successor and ≤ is a linear order, then:
• ∀w, z ∈ A (w v z ⇔ wεz or w = z).
• s : A → A satisfies the condition: ∀x ∈ A, x < s(y) iff x ≤ y.

18. Examples and counter-examples:

S
(i) V = (V, , s), where s(x) = {x}, for all x ∈ V , is a standard ZF-algebra. Indeed:
- ∀x, y ∈ V , xεy iff s(x)⊆y iff {x}⊆y iff x ∈ y;
- (V, ∈)
S is an well-founded (extensional) class;
- z = x∈z {x}, ∀z ∈ V ;
- ∅ ∈ V is the unique root of (V, ∈).
(ii) On = (On, , s), where s(x) = x+ = x ∪ {x}, for all x ∈ On, is a standard
S
ZF-algebra. Indeed:
- ∀x, y ∈ On, xεy iff s(x)⊆y iff x+ ⊆y iff x ∈ y;
- (On,S∈ ) is an well-founded (extensional) class;
- z = x∈ z x+ , ∀z ∈ On;
- ∅ ∈ On is the unique root of (On, ∈ ).
(iii) T r = (T r, , s), where s(x) = x+ = x ∪ {x}, for all x ∈ On, is a ZF-algebra that
S
satisfies (s0) and (s1) but does not satisfies (s2). Indeed:
- ∀x, y ∈ T r, xεy iff s(x)⊆y iff x+ ⊆y iff x ∈ y;
- (T r, ∈ ) is an well-founded (extensional) class;
- ∅ ∈ T r is the unique root of (T r, ∈ );
- (T r, ∈ ) is not extensional. In fact:
3 = {0, {0}, {0, {0}}} and
V3 = {0, {0}, {0, {0}}, {{0}}},
are distinct members of T r with the same transitive members.
- Since (T r, ∈ ) is not extensional thus, by the items above and Remark 17.(a), it does
not satisfies (s2).

9
W
I.e. for all x ∈ L and all small family {yi : i ∈ I}⊆L, s(x) ≤ i∈I yi iff ∃i ∈ I, s(x) ≤ yi .
1.3. ON SURREAL NUMBERS 21
1.3 On Surreal Numbers

This section is dedicated to present the class of surreal numbers – a concept introduced
by J.H. Conway in the mid 1970’s – under many (equivalent) constructions within the
background set class theory NBG, its order and algebraic structure and its relations with
the class (or the ZF-algebra) of all ordinal numbers.
1.3.1 Constructions
1.3.1.1 The Conway’s construction formalized in NBG
We begin with the Conway’s construction following the appendix of his book [Con01],
in which he gave a more formal construction.
We start defining, recursively, the sets Gα in order to define class of games.
(i) G0 = {h∅, ∅i}

[
(ii) Gα = {hA, Bi : A, B ⊆ Gβ }
β<α
S
The class G of Conway games is given by G = α<On Gα .
We can define a preorder 6 in G. Let x = hLx , Rx i and y = hLy , Ry i.
x 6 y iff no xL satisfies xL > y and no y R satisfies x > y R ,
were xL ∈ Lx and y R ∈ Ry .
The second step of the construction is the definition of the class of pre-numbers. We
will again define the ordinal steps Pα recursively:
• P0 = {h∅, ∅i}
[
• Pα = {hA, Bi : A, B ∈ Pβ and B 6> A}
β<α
[
The class P of the pre-numbers is given by P = Pα .
α<On
Finally, the class N o is defined as the quotient of the class of prenumbers by the
equivalence relation induced by 6. To avoid problems with equivalence classes that are
proper classes, we can make a Scott’s Trick.
Following Conway’s notation, we will denote a class (X, Y )/ ∼ by {X|Y } and given
a surreal number x, we will denote x = {Lx |Rx }, where Lx |Rx is a prenumber that
represents x. We will also use the notation xL for an element of Lx and xR for an
element of Rx .
The birth function b is defined as b(x) = min{α : ∃(L, R) ∈ Pα x = {L|R}}
We can also define, for any ordinal α, the sets Oα , Nα and Mα (”old”, ”made” and
”new”):
• Oα = {x ∈ N o : b(x) < α}
• Nα = {x ∈ N o : b(x) = α}
• Mα = {x ∈ N o : b(x) 6 α}
To end this subsection we will now define, recursively, the operations +, −, · in P :
• x + y = {xL + y, x + y L |xR + y, x + y R };
• xy = {xL y + xy L − xL y L , xR y + xy R − xR y R |xL y + xy R − xL y R , xR y + xy L − xR y L };
• −x = {−xR | − xL };
• 0 = {∅|∅};
• 1 = {0|∅}.
Proposition 19.With this operations, N o is a real-closed Field. In addition, every (set)

ordered field has an isomorphic copy inside N o. If Global Choice is assumed, this is valid
also for class ordered fields.
1.3.1.2 The Cuesta-Dutari cuts construction
Given an strict linear order (T, <), we can make a Cuesta ”completion” of T , denoted
by χ(T ), wich is defined by
χ(T ) = (T ∪ CD(T ), <0 ),
with <0 defined as follows:
(i) If x, y ∈ T , then the order is the same as in T ;
(ii) If x = (L, R), y = (L0 , R0 ) ∈ CD(T ), then x <0 y if L ( L0 ;
(iii) If x ∈ T and y = (L, R) ∈ CD(T ), then x <0 y if x ∈ L or y <0 x if x ∈ R.
The idea of that construction is basically the iteration of the Cuesta-Dutari comple-
tion starting from the empty set until the we obtain a η∞ class.
By recursion we define the sets Tα :
• T0 = ∅;
• Tα+1 = χ(Tα );
[
• Tγ = Tβ .
β<γ
And finally we have [

No := Tα
α∈On
In that construction, the birth function b is given by the map that assigns to each
surreal number x, the ordinal b(x) which corresponds to the least set Tb (x) that x belongs.
Note that in this construction the sets ”old”, ”made” and ”new” can be presented in
a simpler[way:
• Oα = Tβ
β<α
• Mα = Tα
• Nα = Tα \ Oα
1.3.1.3 The binary tree construction or the space of signs construction
Consider the class Σ = {f : α → {−, +} : α ∈ On}. We can define in this class an

relation < as follows:
• f < g ⇐⇒ f (α) < g(α), where α is the least ordinal such that f and g differs, with
the convention − < 0 < + (f (α) = 0 iff f is not defined in α).
With this relation, Σ is a strict linearly ordered class isomorphic to (N o, <).
In this construction, the birth function is given by the map b : Σ → On, f 7→ dom f .
1.3.2 The axiomatic approach
It is an well-known fact that the notion of real numbers ordered field can be completely
described (or axiomatized) as a certain structure –of complete ordered field – and every
pair of such kind of structure are isomorphic under a unique ordered field isomorphism
(in fact, there is a unique ordered field morphism between each pair of complete ordered
fields and it is, automatically, an isomorphism). In this subsection, strongly based on
section 3 of the chapter 4 in [All87], we present a completely analogous description for
the ordered class (or ordered field) of surreal numbers.
Definition 20. A full class of surreal numbers is a structure S = (S, <, b) such that:
(i) (S, <) is a strictly linearly ordered class;
(ii) b : S → On is a surjective function;

(iii) For each (small) Conway cut (L, R) in (S, <), the class IS (L, R) = {x ∈ S : L <
{x} < R} is non-empty and its subclass mS (L, R) = {x ∈ IS (L, R) : ∀y ∈ S, b(y) <
b(x) → y ∈ / IS (L, R)} is a singleton;
(iv) For each Conway cut (L, R) in (S, <) and each ordinal number α such that b(z) < α,
∀z ∈ L ∪ R, b({L|R}) 6 α, where {L|R} its unique member of mS (L, R).
Remark 21. Let S = (S, <, b) be a full class of surreal numbers.

• Condition (ii) above entails that S is a proper class.
• Condition (iii) above guarantees that (S, <) is a η∞ -class.
• Since the order relation in On is linear (is an well-order), according the notation in
condition (iii), mS (L, R) = {x ∈ IS (L, R) : ∀y ∈ S, y ∈ IS (L, R) → b(x) ≤ b(y)}.
• By condition (iv), b({∅|∅}) = 0.
As mentioned in section 3 of chapter 4 in [All87], by results proven in Conway’s book

[Con01], the constructions of surreal numbers classes presented in our subsection 3.1 (by
Conway cuts, by Cuesta-Dutari cuts and by the space of sign-expansions), endowed with
natural ”birthday” functions, are all full classes of surreal numbers.
Definition 22. Let S = (S, <, b) and S 0 = (S 0 , <0 , b0 ) be full classes of surreal numbers.
A surreal (mono)morphism f : S → S 0 is a function f : S → S 0 such that:
(i) ∀x, y ∈ S, x < y ⇐⇒ f (x) <0 f (y);
(ii) ∀x ∈ S, b0 (f (x)) = b(x).
Remark 23. Let S = (S, <, b) and S 0 = (S 0 , <0 , b0 ) be full classes of surreal numbers.
• Since < and <0 are linear order, a surreal morphism is always injective and condition
(i) is equivalent to:
(i)’ ∀x, y ∈ S, x < y−→f (x) <0 f (y).
• Naturally, we can define a (”very-large”) category whose objects are the full classes of
surreal numbers and the arrows are surreal morphisms, with obvious composition and
identities. Clearly, an isomorphism in such category is just a surjective morphism.
Proposition 24. Let S = (S, <, b) and S 0 = (S 0 , <0 , b0 ) be full classes of surreal numbers.
Then:
(i) There is a unique surreal (mono)morphism f : S → S 0 and it is an isomorphism.
(ii) For each ordinal number α, b−1 ([0, α)) is a set. Or, equivalently, b is a locally small
function.
t
(iii) The function (L, R) ∈ Cs (S, <) 7→ {L|R} ∈ S is surjective.
In particular, all the constructions of surreal numbers classes presented in our subsec-
tion 3.1, endowed with natural birthday functions, are canonically isomorphic, through
a unique isomorphism.
In the section 4 of chapter 4 in [All87], named ”Subtraction in N o”, we can find the
following result:
Proposition 25. Let S = (S, <, b) be a full class of surreal numbers. Then there is a
unique function − : S → S such that:
(i) b(−x) = b(x), ∀x ∈ S;
(ii) −(−x) = x, ∀x ∈ S;
(iii) x < y ⇐⇒ −y < −x, ∀x, y ∈ S;
(iv) −{L|R} = {−R| − L}, ∀(L, R) ∈ Cs (S, <).
Remark 26. Let S = (S, <, b) be a full class of surreal numbers.

• In the presence of condition (ii), condition (iii) is equivalent to:
(iii)’ x < y → −y < −x, ∀x, y ∈ S.
• By condition (iii), condition (iv) makes sense, since L < R⇒ − R < −L.
• By condition (iv), −{∅|∅} = {∅|∅}.
We finish this Subsection registering the following useful results whose proofs can be
found in [All87], pages 125, 126.
Fact 27. Let S = (S, <, b) be a full class of surreal numbers. Let A, A0 , B, B 0 , {x}, {x0 }⊆S
be subsets such that A < B and A0 < B 0 and x = {A|B}, x0 = {A0 |B 0 }. Then:
(a) If A and A0 are mutually cofinal and B and B 0 are mutually coinitial, then
{A|B} = {A0 |B 0 }.
(b) Suppose that (A, B) and (A0 , B 0 ) are timely representations of x and x0 respec-
tively, i.e b(z) < b(x), ∀z ∈ A ∪ B and b(z 0 ) < b(x0 ), ∀z 0 ∈ A0 ∪ B 0 . If x = x0 then A and
A0 are mutually cofinal and B and B 0 are mutually coinitial.

1.3.3 Ordinals in No
The results presented in this Subsection can be found in the Chapter 4 of [All87].
The ordinals can be embedded in a very natural way in the field N o. The function
that makes this work is recursively defined as follows:
Definition 28.j(α) = {j[α]|∅}, α ∈ On.
The following result establishes a relation between the function j and the birthday
function:
Proposition 29.[[Gon86], p. 41] b ◦ j = idOn

That map j encodes completely the ordinal order into the surreal order:
Proposition 30.α < β iff j(α) < j(β), ∀α, β ∈ On.
We have also that j(0) = 0, j(1) = 1. In fact, that embedding preserves also some
algebraic structure. Although the sum and product of ordinals are not commutative, we
have alternative definitions sum and product in On closely related to the usual operations
that are commutative:
Definition 31.If α and β are ordinals we can define the Hessemberg Sum (respesctively
Hessemberg Product) of α and β as the sum (product) of the normal forms of α and β
as if they are polynomials. (See [Gon86], p. 41)
Fact 32. [[Gon86], p.42] The Hessemberg sum and product of ordinals are mapped by j
to the surreal sum and product.
˙ 0, 1) has an isomorphic copy in N o given by

In other words, the semi-ring (On, +̇, ×,
the image of j
Chapter 2
Introducing Surreal Algebras
Motivated by the structure definable in the class N o of all surreal numbers, we introduce
in this Chapter the notion of surreal algebra (SUR-algebra) as a (higher-order) structure
S = (S, <, ∗, −, t), satisfying some properties were, in particular, < is an acyclic relation
on S where t : Cs (S) → S is a function that gives a coherent choice of witness of
η∞ density of (S, <). Every SUR-algebra turns out to be a proper class. Besides the
verification that N o indeed support the SUR-algebra structure, we have defined two
distinguished SUR-algebras SA and ST , respectively the ”free surreal algebra”’ and
”the free transitive surreal algebra”, that will be useful in the sequel of this work. We
also have introduced the notion of partial SUR-algebra (that can be a improper class)
and describe some examples and constructions in the corresponding categories. We have
provided, by categorical methods, some universal results that characterizes the SUR-
algebras SA and ST , and also some relative versions with base (”urelements”) SA(I),
ST (I 0 ) where I, I 0 are partial SUR-algebra satisfying a few constraints.
2.1 Axiomatic definition

Definition 33. A surreal algebra (or SUR-algebra) is an structure S = (S, <, −, ∗, t)
where:
• < is a binary relation in S;
• ∗ ∈ S is a distinguished element;
• − : S → S is a unary operation;
• t : Cs (S) → S is a function, where Cs (S) = {(A, B) ∈ Ps (S) × Ps (S) : A < B}.
Satisfying the following properties:
(S1) < is an acyclic relation.
(S2) ∀x ∈ S, −(−x) = x.
27
28 CHAPTER 2. INTRODUCING SURREAL ALGEBRAS
(S3) −∗ = ∗.
(S4) ∀a, b ∈ S, a < b iff −b < −a.
(S5) ∀(A, B) ∈ Cs (S), A < t(A, B) < B.
(S6) ∀(A, B) ∈ Cs (S), −t(A, B) = t(−B, −A).
(S7) ∗ = t(∅, ∅).
Remark 34.
• Let (S, <) be the underlying relational structure of a surreal algebra S. Then < is an
irreflexive relation, by condition (S1), and by (S5), (S, <) is a η∞ -relational structure.
As a consequence S is a proper class: see 7 in the Subsection 1.1.3. The other axioms
establish the possibility of choice of witness for the η∞ property satisfying additional
coherent conditions.
• Note that (S3) follows from (S7) and (S6) : −∗ = −t(∅, ∅) = t(−∅, −∅) = t(∅, ∅) = ∗.
• Axiom (S7) establish that the SUR-algebra structure is ”an extension by definitions”
of a simpler (second-order) language: without a symbol for constant ∗.
• In the presence of (S2), condition (S4) is equivalent to:
(S4)’ ∀a, b ∈ S, a < b ⇒ −b < −a.
• By condition (S4), condition (S6) makes sense, since A < B⇒ − B < −A (and if A, B
are sets, then −A, −B are sets).

A morphism of surreal algebras is a function that preserves all the structure on the
nose. More precisely:
Definition 35. Let S = (S, <, −, ∗, t) and S 0 = (S 0 , <0 , −0 , ∗0 , t0 ) be SUR-algebras. A

morphism of SUR-algebras h : S → S 0 is a function h : S → S 0 that satisfies the
conditions below:
(Sm1) h(∗) = ∗0 .
(Sm2) h(−a) = −0 h(a), ∀a ∈ S.
(Sm3) a < b =⇒ h(a) <0 h(b), ∀a, b ∈ S.
(Sm4) h(t(A, B)) = t0 (h[A], h[B]), ∀(A, B) ∈ Cs (S).1

1
Note that, by property (Sm3), (A, B) ∈ Cs (S) =⇒ (h[A], h[B]) ∈ Cs (S 0 ), thus (Sm4) makes sense.
2.2. EXAMPLES AND CONSTRUCTIONS 29
Definition 36. The category of SUR-algebras:

We will denote by SU R − alg the (”very-large”) category such that Obj(SU R − alg)
is the class of all SUR-algebras and M or(SU R − alg) is the class of all SUR-algebras
morphisms, endowed with obvious composition and identities.

Remark 37.
Of course, we have the same ”size issue” in the categories ZF −alg, of all ZF-algebras
(Section 1.2, Chapter 1), and in SU R − alg: each object can be a (respect., is a) proper
class, thus it cannot be represented in NBG background theory this ”very large” category.
The mathematical (pragmatical) treatment of this question, that we will adopt in the
Chapters 1, 2 and 3, is to assume a stronger background theory: NBG (or ZFC) and
also three Grothendieck universes U0 ∈ U1 ∈ U2 . The members of U0 represents ”the
sets”; the members of U1 represents ”the classes”; the members of U2 represents ”the
meta-classes”. Thus a category C is: (i) ”small”, whenever C ∈ U0 ; (ii) ”large”, whenever
C ∈ U1 \ U0 ; (iii) ”very large”, whenever C ∈ U2 \ U1 .

2.2 Examples and constructions
2.2.1 The surreal numbers as SUR-algebras
The structure (N o, <, b) of full surreal numbers class, according the Definition 20 in the
Subsection 1.3.2 in Chapter 1, induces a unique structure of SUR-algebra (N o, <, −, ∗, t),
where:
• The function t : Cs (N o, <) → N o is such that (A, B) 7→ t(A, B) := {A|B};
• The distinguished element ∗ ∈ N o is given by ∗ := {∅|∅};
• The function − : N o → N o is the unique function satisfying the conditions in Propo-
sition 25 and Remark 26.
This SUR-algebra has two distinctive additional properties:

• t is a surjective function;
• < is a strict linear order (equivalently, since < is acyclic, < is a total relation).
2.2.2 The free surreal algebra
We will give now a new example of surreal algebra, denoted SA2 , which is not a linear
order and satisfies a nice universal property on the category of all surreal algebras (see
Section 2.4). The construction, is based is based on a cumulative Conway’s cuts hierarchy
over a family of binary relations. 3 .
We can define recursively the family of sets SAα as follows:
Suppose that, for all β < α, we have
S constructed (α)
the S
sets SAβ and <β , binary
(α)
relations on SAβ , and denote SA = β<α SAβ and < = β<α <β . Then, for α we
define:
• SAα = SA(α) ∪ {hA, Bi : A, B ⊆ SA(α) and A <(α) B}.4
• <α =<(α) ∪{(a, hA, Bi), (hA, Bi, b) : hA, Bi ∈ SAα \ SA(α) and a ∈ A, b ∈ B}.
• The class SA5 is the union SA := α∈On SAα .

S
S
• <:= α∈On <α is a binary relation on SA.
Fact: Note that that:

(a) SA(0) = ∅, SA(1) = SA0 = {h∅, ∅i} and SA1 = {h∅, ∅i, h∅, {h∅, ∅i}i, h{h∅, ∅i}, ∅i}.
By simplicity, we will denote ∗ := h∅, ∅i = 0, −1 := h∅, {∗}i and 1 := h{∗}, ∅i thus
SA1 = {0, −1, 1}.
(b) <0 = ∅ and <1 = {(−1, 0), (0, 1)}.
(c) −1 and 1 are <-incomparable.
(d) SA(α) ⊆SAα , α ∈ On.
(e) SAβ ⊆SAα , β ≤ α ∈ On.
(f) SA(β) ⊆SA(α) , β ≤ α ∈ On.
(g) <(α) =<α ∩SA(α) × SA(α) , α ∈ On (by the definition of <α ).
(h) <β =<(α) ∩SAβ × SAβ , β < α ∈ On.
(i) <β =<α ∩SAβ × SAβ , β ≤ α ∈ On (by items (g) and (h) above).
(j) <α =< ∩SAα × SAα , α ∈ On.
(k) Cs (SAα , <α ) = Cs (SA, <) ∩ (Ps (SAα ) × Ps (SAα )), α ∈ On (by item (j)).
(l) Cs (SA(α) , <(α) ) = Cs (SA, <) ∩ (Ps (SA(α) ) × Ps (SA(α) )), α ∈ On.
2
The ”A” in SA is to put emphasis on acyclic.
3
Starting from the emptyset, and performing a cumulative construction based on Cuesta-Dutari
completion of a linearly ordered set, we obtain N o: see for instance [All87].
4
The expression hA, Bi is just an alternative notation for the ordered pair (A, B), used for the reader’s
convenience.
5
Soon, we will see that SA is a proper class.
We already have defined < and ∗(= h∅, ∅i) in SA, thus we must define t : Cs (SA) →
SA and − : SA → SA to complete the definition of the structure SA: this will be
carry out by recursion on well-founded relations on SA and Cs (SA)6 that will be defined
below.
For each x ∈ SA, we define its rank as r(x) := min{α ∈ On : x ∈ SAα }. Since for
each β < α, SAβ ⊆SA(α) ⊆SAα , it is clear that r(x) = α iff x ∈ SAα \ SA(α) .
Following Conway ([Con01], p.291), we can define for the SA setting the notions of:
”old members”, ”made members” and ”new members”. More precisely, for each ordinal
α:
• The set of old members w.r.t. α is the subset of SA of all members ”born before day
α”. O(SA, α) := SA(α) ;
• The set of made members w.r.t. α is the subset of SA of all members ”born on or
before day α”. M (SA, α) := SAα ;
• The set of new members w.r.t. α is the subset of SA of all members ”born on day
α”. N (SA, α) := SAα \ SA(α) .
We will denote x ≺ y in SA iff r(x) < r(y) in On.
Claim 1: ≺ is an well-founded relation in SA.

Proof. Let y ∈ SA and let α := r(y). Given x ∈ SA, r(x) < α iff x ∈ O(SA, α) = SA(α) .
Therefore, the subclass {x ∈ SA : x ≺ y} is a subset of SA.
Now let X be a non-empty subset of SA. Then r[X] is a non-empty subset of On and let
α := min(r[X]). Consider any a ∈ r−1 [{α}] ∩ X, then clearly a is a ≺-minimal member
of X.
We have an induced ”rank” on the class (of small <-Conway cuts) Cs (SA) = {(A, B) ∈
Ps (A) × Ps (B) : A < B}, R(A, B) := min{α ∈ On : A ∪ B⊆SA(α) }. We can also define
a binary relation on the class Cs (SA):
(A, B) / (C, D) in Cs (SA) iff R(A, B) < R(C, D) in On.
Claim 2: / is an well-founded relation in Cs (SA).

Proof. Let (C, D) ∈ Cs (SA) and let α := R(C, D). Given (A, B) ∈ Cs (SA), R(A, B) <
α iff ∃β < α, A ∪ B⊆O(SA, β) = SA(β) . Therefore, the subclass {(A, B) ∈ Cs (SA) :
6
From now on, we will omit the binary relation on a class when it is clear from the setting.
(A, B) / (C, D)} is a subset of Cs (SA).

Now let Y be a non-empty subset of Cs (SA). Then R[Y ] is a non-empty subset of On
and let α := min(R[Y ]). Consider any (A, B) ∈ R−1 [{α}] ∩ Y , then clearly (A, B) is a
/-minimal member of Y .
Let H be the (class) function H(p, g) where, for each p = (C, D) ∈ Cs (SA) and g a
(set) function with domain p/ := {(A, B) ∈ Cs (SA) : (A, B) / p}, given by H(p, g) :=
hC, Di (i.e., H is just first coordinate projection). Then H is a class function and we can
define by /-recursion a unique (class) function t : Cs (SA) → SA by t(p) = H(p, tp/ ),
i.e. t(C, D) = hC, Di. The range of t is included in SA: since A and B are subsets of
SA such that A < B, there exists α ∈ On such that A, B⊆SA(α) ; since < is the reunion
of the increasing compatible family of binary relations {<β : β ∈ On}, the have that
A <(α) B, thus hA, Bi ∈ SAα ⊆SA.
Claim 3: ∀α ∈ On, M (SA, α) = Cs (O(SA, α)). Thus N (SA, α) = Cs (SA(α) )\SA(α) .

Proof. Since M (SA, α) = O(SA, α) ∪ Cs (O(SA, α)), we just have to prove that,
SA(α) ⊆Cs (SA(α) ), for each α ∈ On. Suppose that SA(β)S⊆Cs (SA(β) ) for
S each β ∈ (β)
On
(α)
such that β < α. By the assumption, we have SA = β<α SAβ = β<α Cs (SA ).
SinceSSA(β) ⊆SA(α) and <(β) =<(α) ∩(SA(β) × SA(β) )7 , we have Cs (SA(β) )⊆Cs (SA(α) ),
thus β<α Cs (SA(β) )⊆Cs (SA(α) ). Summing up, we conclude that SA(α) ⊆Cs (SA(α) ) and
the result follows by induction.
Claim 4: Cs (SA) = SA and t : Cs (SA) → SA is the identity map, thus, in particu-

lar, t is a bijection.
S
Proof. By items (k) and (l) in the Fact above, C s (SA, <) = α∈On Cs (SAα , <α ) =
(α) (α) (α) (α)
S
Sα∈On s C (SA , < ). By Claim 3 above, SA α = C s (SA , < ), ∀α ∈ On, thus
(α) (α)
S
α∈On Cs (SA , < ) = α∈On SAα = SA. Summing up, we obtain Cs (SA) = SA.
Then t : Cs (SA) → SA, given by (A, B) 7→ hA, Bi is the identity map.
For each x ∈ SA, denote (Lx , Rx ) ∈ Cs (SA) the unique representation of x: in fact,
x = hLx , Rx i.
Claim 5: r ◦ t = R.
Proof. The functional equation is equivalent to:
7
The non trivial inclusion <(β) ⊇<(α) ∩(SA(β) × SA(β) ) holds since for every pair (x, y) in the right
side there are δ < β and γ < α (that we can assume γ ≥ δ) such that (x, y) ∈<γ ∩SAδ ×SAδ =<δ ⊆ <β .
∀(A, B) ∈ Cs (SA), ∀γ ∈ On, hA, Bi ∈ SAγ iff A ∪ B⊆SA(γ) .

If (A, B) ∈ Cs (SA) and A ∪ B⊆SA(γ) then, since A < B we have A <(γ) B, thus
hA, Bi ∈ SAγ by the recursive definition of SAγ . On the other hand, if (A, B) ∈ Cs (SA)
and hA, Bi ∈ SAγ , then by Claim 3 above, (A, B) ∈ Cs (SA, <) ∩ Cs (SA(γ) ), <(γ) ) =
Cs (SA(γ) ), <(γ) ), thus A ∪ B⊆SA(γ) .
Claim 6: Let (A, B) ∈ Cs (SA) and α ∈ On, then: ∀a ∈ A, ∀b ∈ B, r(a), r(b) < α
iff r(t(A, B)) ≤ α. In particular: ∀a ∈ A, ∀b ∈ B, r(a), r(b) < r(t(A, B)).
Proof. The equivalence is just a rewriting of the equivalence proved above:
A ∪ B⊆SA(α) iff hA, Bi ∈ SAα .
Claim 7: ∀x, y ∈ SA, x < y ⇒ r(x) 6= r(y). In particular, the relation < in SA is
irreflexive.
Proof. Suppose that there are x, y ∈ SA such that x < y and r(x) = r(y) = α ∈ On.
Thus x, y ∈ SAα \ SA(α) and, since x, y ∈ SAα and (x, y) ∈<, we get (x, y) ∈<α \ <(α) .
Thus (x, y) = (a, hLy , Ry i) for some a ∈ Ly ⊆SA(α) or (x, y) = (hLx , Rx i, d) for some
d ∈ Rx ⊆SA(α) . In both cases we obtain x = a ∈ SA(α) or y = d ∈ SA(α) , contradicting
our hypothesis.
Claim 8: Let A, B⊆SA be subclasses such that A < B, then r[A] ∩ r[B] = ∅.
Proof. Suppose that A < B and that there are a ∈ A and b ∈ B such that r(a) =
r(b) ∈ r[A] ∩ r[B]. Then a < b and r(a) = r(b), contradicting the Claim 7 above.
Claim 9: Let (A, B), (C, D) ∈ Cs (SA). Then hA, Bi < hC, Di iff hA, Bi ∈ C (then
r(hA, Bi) < r(hC, Di)) or hC, Di ∈ B (then r(hC, Di) < r(hA, Bi)).
Proof. (⇐) If hA, Bi = c ∈ C and r(hC, Di) = α, then (c, hC, Di) ∈<α ⊆ <, thus
hA, Bi < hC, Di. The other case is analogous.
(⇒) Suppose that hA, Bi < hC, Di. By Claim 7 above we have α = r(hA, Bi) 6=
r(hC, Di) = γ. If α < γ we have SAα ⊆SA(γ) and hC, Di ∈ SAγ \SA(γ) , thus (hA, Bi, hC, Di) ∈<γ
\ <(γ) and we have hA, Bi ∈ C. If γ < α we conclude, by an analogous reasoning, that
hC, Di ∈ B.
Claim 10: Let (A, B), (C, D) ∈ Cs (SA). Then A < hC, Di < B and R(A, B) ≤
R(C, D) iff A⊆C and B⊆D.
Proof. (⇐) Let R(C, D) = α, then ∀c ∈ C, ∀d ∈ D, (c, hC, Di), (hC, Di, d) ∈<α ⊆ <.
If A⊆C and B⊆D, then A < hC, Di < B and A ∪ B⊆C ∪ D⊆SA(α) , i.e. R(A, B) ≤ α =
R(C, D).
(⇒) Let A < hC, Di < B and suppose that there is a ∈ A\C, then hLa , Ra i = a < hC, Di.
Since hLa , Ra i ∈
/ C, then by Claim 9 above, we have hC, Di ∈ Ra , thus:
R(C, D) =Claim5 r(hC, Di) <Claim6 r(hLa , Ra i) = r(a) < r(hA, Bi) = R(A, B). Analo-
gously, if A < hC, Di < B and B \ D 6= ∅, we obtain R(C, D) < R(A, B).
Claim 11: For each (A, B) ∈ Cs (SA, <), A < hA, Bi < B. In particular, (SA, <) is
a η∞ proper class.
Proof. By Claim 7 above, < is an irreflexive relation. By Claim 10 above, for each
(A, B) ∈ Cs (SA, <), A < hA, Bi < B, thus (SA, <) is a η∞ class. It follows from 7 in
the Subsection 1.1.3, that SA is proper class.
Claim 12: For each (A, B) ∈ Cs (SA, <) and each z ∈ SA such that A < z < B,
then r(t(A, B)) ≤ r(z).
Proof. Suppose that the result is false and let α the least ordinal such that there are
(A, B) ∈ Cs (SA) and z ∈ SA such that A < z < B, but r(z) < r(t(A, B)) = R(A, B) =
α: thus α > 0. By a simple analysis of the cases α ordinal limit and α successor, we
can see that there are A0 ⊆A, B 0 ⊆B such that R(A0 , B 0 ) = α0 < α and A0 < z < B 0 ,
contradicting the minimality of α8 .
Define, by recursion on the well-ordered proper class (On, <), a function s : On → SA
by s(α) := hs[α], ∅i, α ∈ On.
Claim 13: r ◦ s = idOn . In particular, the function r : SA → On is surjective and

SA is a proper class.
Proof. We will prove the result by induction on the well-ordered proper class (On, <).
Let α ∈ On and suppose that r(s(β) = β, for all ordinal β < α. Then:
r(s(α)) = r(hs[α], ∅i) = r(t(s[α], ∅)) =Claim 5 R(s[α], ∅) = min{γ ∈ On : s[α] ∪
∅⊆SA(γ) }.
By the induction hypothesis, we have:
(IH) s(β) ∈ SAβ \ SA(β) , for all ordinal β < α.
Since s(β) ∈ SAβ , we have s(β) ∈ SA(α) , ∀β < α. If s[α]∪∅⊆SA(γ) for some γ < α, then
s(γ) ∈ SA(γ) , in contradiction with (IH). Summing up, we conclude that r(s(α) = α,
and the result follows by induction.
8
Hint: in the case α = γ + 1, use Claim 7.
Claim 14: There is a unique function − : SA → SA, such that:

(i) ∀x ∈ SA, r(−x) = r(x);
(ii) ∀x ∈ SA, −(−x) = x;
(iii) ∀x, y ∈ SA, x < y iff −y < −x;
(iv) ∀(A, B) ∈ Cs (SA), −t(A, B) = t(−B, −A).
Proof. Let z ∈ SA and suppose that a function − is defined for all x, y ∈ SA, such that
x, y ≺ z, satisfying the conditions (i)–(iv) adequately restricted to the subset SA(α) ,
for α := r(z) ∈ On. Then z = hLz , Rz i = t(Lz , Rz ) ∈ SAα \ SA(α) for a unique
(Lz , Rz ) ∈ Cs (SA) (Claim 4) and α is the least γ ∈ On such that Lz ∪ Rz ⊆SA(γ) (Claim
5), thus ∀x ∈ Rz ∪ Lz , r(x) < r(z). Then −x is defined ∀x ∈ Lz ∪ Rz , satisfying the
conditions (i)–(iv) restricted to the subset SA(α) . Since x < y, ∀x ∈ Lz ∀y ∈ Rz , it
holds, by condition (iii), −y < −x then −Rz < −Lz and since −Rz , −Lz are the images
of a function on sets, (−Rz , −Lz ) ∈ Cs (SA). Moreover, by condition (i), α is the least
γ ∈ On such that −Rz ∪ −Lz ⊆SA(γ) , i.e. t(−Rz , −Lz ) = h−Rz , −Lz i ∈ SAα \ SA(α) .
Define −z := t(−Rz , −Lz ).
Now we will prove that the conditions (i)–(iv) still holds for all members in SAα \ SA(α) .
(i) Let x ∈ SAα . If x ∈ SA(α) , this condition holds by hypothesis. If x ∈ SAα \ SA(α) ,
then by the recursive definition above, −x = h−Rx , −Lx i ∈ SAα \ SA(α) , thus r(−x) =
α = r(x). Thus (i) holds in SAα .
(ii) Let x ∈ SAα \ SA(α) , then −x, −(−x) ∈ SAα \ SA(α) (by the validity of con-
dition (i) on SAα established above). −(−x) = −(−t(Lx , Rx )) = −t(−Rx , −Lx ) =
t(−(−Lx ), −(−Rx )) = t(Lx , Rx ) = x, since by hypothesis the conditions (iii) and (ii)
holds for members of SA(α) . Thus (ii) holds in SAα .
(iii) We suppose that ∀x, y ∈ SA(α) , x < y iff −y < −x. Let x, y ∈ SAα such that
x < y. If both x, y ∈ SA(α) then, by hypothesis −y < −x. Otherwise, by Claim 7,
there is exactly one between x, y that is a member of SAα \ SA(α) . By Claim 9: if
r(y) < r(x) = α then y ∈ Rx ; if r(x) < r(y) = α then x ∈ Ly . Thus: if r(y) < r(x) = α
then −y ∈ −Rx = L−x , thus −y < −x; if r(x) < r(y) = α then −x ∈ −Ly = R−y ,
thus −y < −x. Then we have proved that ∀x, y ∈ SAα , x < y ⇒ −y < −x. Since the
conditions (i) and (ii) have already be established on SAα , we also have ∀x, y ∈ SAα ,
−y < −x ⇒ x = −(−x) < −(−y) = y.
(iv) Suppose that t(A, B) = hA, Bi = S h−B, −Ai = t(−B, S−A) holds for (α)
all (A, B) ∈
(β) (β) (β)
S
C
β<α s (SA)∩P s (SA )×P s (SA ) = C
β<α s (SA ) = β<α SA β = SA . We must
prove that the condition still holds for all (C, D) ∈ Cs (SA) ∩ Ps (SA ) × Ps (SA(α) ) =
(α)
Cs (SA(α) = SAα . Let z = (C, D) = hC, Di = t(C, D) ∈ SAα \ SA(α) . Then just by the
recursive definition of −z, we have −z = −t(C, D) = t(−B, −A), as we wish.
Finally, we will prove that SA satisfies all the 7 axioms of SUR-algebra:
(S7) ∗ = t(∅, ∅).

This holds by our definition of ∗.
(S5) ∀(A, B) ∈ Cs (SA), A < t(A, B) < B.

This holds by the Claim 10 above.
(S6) ∀(A, B) ∈ Cs (SA), −t(A, B) = t(−B, −A).

This holds by Claim 14.(iv).
(S3) −∗ = ∗.
Since −∗ = −t(∅, ∅) = t(−∅, −∅) = t(∅, ∅) = ∗.
(S2) ∀x ∈ SA, −(−x) = x.

This holds by Claim 14.(ii).
(S4) ∀a, b ∈ SA, a < b iff −b < −a.

This holds by Claim 14.(iii).
(S1) < is an acyclic relation.

Suppose that < is not acyclic and take x0 < ... < xn < x0 a cycle in (SA, <) of
minimum length n ∈ N. Since < is an irreflexive relation (see the Claim 7), n > 0.
Let α = max{r(xi ) : i ≤ n} and let j be the least i ≤ n such that r(xj ) = α.
If j = 0: Since x0 < x1 and xn < x0 , then by Claim 7, r(x1 ), r(xn ) < r(x0 ). Writing
x0 = hLx0 , Rx0 i (since, by Claim 4, SA = Cs (SA)), we obtain from Claim 9 that
xn ∈ Lx0 and x1 ∈ Rx0 . As Lx0 < Rx0 , we have xn < x1 and then x1 < ... < xn < x1
is a cycle of length n − 1 < n, a contradiction.
If j > 0: Then define j − := j − 1 and j + := j + 1 (respec. j + = 0), if j < n (respec.
j = n).
Then by Claim 7, r(xj − ), r(xj + ) < r(xj ) and by Claim 9: xj − ∈ Lxj and xj + ∈ Rxj .
As Lxj < Rxj , we have xj − < xj + and then we can take a sub-cycle of the original
one omitting xj : this new cycle has of length n − 1 < n, a contradiction.
2.2.3 The free transitive surreal algebra
We will give now a new example of surreal algebra, denoted ST 9 , which is a strict
partial order10 that is not linear and satisfies a nice universal property on the category
of all transitive surreal algebras (see Section 2.4). The construction is similar to the
construction of SA in the previous subsection: it is based on a cumulative Conway’s cuts
hierarchy over a family of binary (transitive) relations.
We can define recursively the family of sets STα as follows:
Suppose that, for all β < α,
Swe have constructedS
the sets STβ and <β , binary relations
on STβ , and denote ST (α) = β<α STβ and <(α) = β<α <β . Then, for α we define:
9
The ”T” in ST is to put emphasis on transitive.
10
Recall that a binary relation is that is a strict partial order iff it is a transitive and acyclic relation.
• STα = ST (α) ∪ {hA, Bi : A, B ⊆ ST (α) and A <(α) B}.
• <α = the transitive closure of the relation <0α , where <0α := (<(α) ∪{(a, hA, Bi), (hA, Bi, b) :
hA, Bi ∈ STα \ ST (α) and a ∈ A, b ∈ B}).
S
• The (proper) class ST is the union ST := α∈On STα .
S
• <:= α∈On <α is a binary (transitive) relation on ST .
The following result is straightforward an completely analogous to the corresponding

items in the Fact in the previous Subsection on SA:
Fact 1: Note that that:
(a) ST (0) = ∅, ST (1) = ST0 = {h∅, ∅i}. By simplicity, we will denote 0 := h∅, ∅i,
1 := h∅, {0}i, −1 := h{0}, ∅i. Thus: ST0 = {0}, SA1 = {0, 1, −1}.
(b) <0 = ∅, <1 = {(−1, 0), (0, 1), (−1, 1)}.
(c) −1 < 0 < 1, −1 < h{−1}, {1}i < 1, but 0, h{−1}, {1}i are <-incomparable.
(d) ST (α) ⊆STα , α ∈ On.
(e) STβ ⊆STα , β ≤ α ∈ On.
(f) ST (β) ⊆ST (α) , β ≤ α ∈ On.
Analogously to in the SA case, we can define rank functions r : ST → On11 and

R : Cs (ST ) → On that induces well-founded relations on ST and on Cs (ST ).
The results below are almost all (the exception are the items (m), (n), (o)) analogous
to corresponding items in the Fact in the previous Subsection on SA. However, the
techniques needed in the proofs are different than in SA case and deserve a careful
presentation.
Fact 2:
(g) <(α) =<α ∩SA(α) × SA(α) , α ∈ On.
(h) <β =<(α) ∩SAβ × SAβ , β < α ∈ On.
(i) <β =<α ∩SAβ × SAβ , β ≤ α ∈ On.
(j) <α =< ∩SAα × SAα , α ∈ On.
(k) Cs (STα , <α ) = Cs (ST, <) ∩ Ps (STα ) × Ps (STα ), α ∈ On.
(l) Cs (ST (α) , <(α) ) = Cs (ST, <) ∩ Ps (ST (α) ) × Ps (ST (α) ), α ∈ On.
(m) ∀α ∈ On, <α is a transitive and a acyclic relation on STα .
(n) < is a transitive and acyclic relation (or, equivalently, it is a strict partial order) on
ST .
11
For each x ∈ ST , r(x) = α ∈ On iff x ∈ STα \ ST (α) .
(o) Let x, y ∈ ST and denote α := max{r(x), r(y)}. Then are equivalent:

• x < y.
• Exists n ∈ N, exists {z0 , · · · , zn+1 }⊆STα such that: x = z0 , y = zn+1 ; zj ∈ Lzj+1 or
zj+1 ∈ Rzj , for all j ≤ n; {z1 , · · · , zn }⊆ST (α) .
Proof. Item (i) follows from items (g) and (h). Items (k) and (l) follows from item (j).
Items (n) and (o) are direct consequences
S of item (m) (for the item (o) is required to
perform induction in α), since <= α∈On <α .
(g) Clearly <(α) ⊆ <α ∩SA(α) × SA(α) . To show the converse inclusion let x, y ∈
SA(α) be such that x <α y and let x = x0 <0α ... <0α xn = y be a sequence in (STα , <0α )
with the number k = card({i 6 n : r(xi ) = α} being minimum. We will show that
k = 0, thus the sequence is just x = x0 <(α) ... <(α) xn = y and then x <(α) y because
<(α) is a transitive relation (since <β , β ∈ On is a transitive relation, by construction).
Suppose, by absurd, that k > 0 and let j be the least i ≤ n such that r(xj ) = α.
By our hypothesis on x, y we have 0 < j < n. Since xj−1 <0α xj <0α xj+1 , we have
r(xj−1 ), r(xj+1 ) < r(xj ) = α and xj−1 ∈ Lxj , xj+1 ∈ Rxj . As Lxj <(α) Rxj , we have
xj−1 <(α) xj+1 and then we can take a sub-cycle of the original one omitting xj : this
new cycle has k − 1 < k members with rank α, a contradiction.
(h) We only prove the non-trivial inclusion. Let x, y ∈ SAβ be such that x <(α) y.
Since <(α) = γ<α <γ , let β 0 be the least γ < α such that x <β 0 y. We will prove that
S
β 0 ≤ β, thus we obtain x <β y, as we wish. Suppose, by absurd, that β 0 > β. Then
0 0 0
(x, y) ∈<β 0 ∩SA(β ) × SA(β ) , and by the item (g) proved above (x, y) ∈<(β ) . Thus there
is some γ < β 0 such that x <γ y, contradicting the minimality of β 0 .
(j) Let x, y ∈ SAα be such that x < y. Since <= γ∈On <γ , let α0 be the least
S
γ ∈ On such that x <α0 y. We will prove that α0 ≤ α, thus we obtain x <α y, as we
0 0
wish. Suppose, by absurd, that α0 > α. Then (x, y) ∈<α0 ∩SA(α ) × SA(α ) , and by
0
the item (g) proved above (x, y) ∈<(α ) . Thus there is some γ < α0 such that x <γ y,
contradicting the minimality of α0 .
(m) By definition of <γ , <γ is a transitive relation, ∀γ ∈ On.
Suppose that the statement is false and let α ∈ On be the least ordinal such that
(STα , <α ) has some cycle. Then ∀β < α, <β is an acyclic relation but <α has some
cycle (or, equivalently, <0α has some cycle). Let x0 <0α ... <0α xn <0α x0 be a cycle in
(STα , <0α ) with the number k = card({i 6 n : r(xi ) = α} being minimum. Note that
k > 0, otherwise x0 , ..., xn ∈ SA(α) and the cycle is x0 <(α) ... <(α) xn <(α) x0 , thus there
is a β < α and a cycle x0 <β ... <β xn <β x0 in (STβ , <β ), contradicting our hypothesis.
Let j be the least i ≤ n such that r(xj ) = α.
If j = 0: Since x0 <0α x1 and xn <0α x0 , then r(x1 ), r(xn ) < r(x0 ) = α. Writing x0 =
hLx0 , Rx0 i, we have that xn ∈ Lx0 and x1 ∈ Rx0 . As Lx0 <(α) Rx0 , we have xn <(α) x1 ,
and then x1 <0α ... <0α xn <0α x1 is a cycle in (STα , <0α ) with k − 2 < k members with
rank α, a contradiction.
If j > 0: Then define j − := j − 1 and j + := j + 1 (respect. j + = 0), if j < n (respect.
j = n).
Then r(xj − ), r(xj + ) < r(xj ) = α and: xj − ∈ Lxj , xj + ∈ Rxj . As Lxj <(α) Rxj , we have
xj − <(α) xj + and then we can take a sub-cycle of the original one omitting xj : this new
cycle has k − 1 < k members with rank α, a contradiction.
Since the harder part was already done, we just sketch the construction of the SUR-
algebra structure (ST, <, −, ∗, t):
• As in the SA case, from the well founded relation on Cs (ST ) we can define recursively
a function with range ST , t : Cs (ST ) → ST by t(A, B) = hA, Bi. We can prove, by
induction, that STα = Cs (ST (α) ), α ∈ On. Thus t is a bijection (is the identity function).
Moreover, if (A, B) ∈ Cs (ST ), then A < t(A, B) < B.
• We define ∗ := 0 = t(∅, ∅).
• As in the SA case, we can define (recursively) the function − : ST → ST by
−hA, Bi := h−B, −Ai.
The verification of the satisfaction of the SUR-algebra axioms (S2)–(S7) are analogous
as in the SA case. The satisfaction of (S1) was proved in item (m) of Fact 2 above.
2.2.4 The cut surreal algebra
In this Subsection we present a generalization of the SA, ST constructions. Given a

surreal algebra S, we can define a new surreal algebra whose domain is Cs (S) with the
following relations and operations:
Definition 38.Let (S, <, −, ∗, t) be a surreal algebra. Consider the following structure
in Cs (S)
• ∗0 = (∅, ∅)
• −0 (A, B) = (−B, −A)
• (A, B) <0 (C, D) ⇐⇒ t(A, B) < t(C, D)
• t0 (Γ, ∆) = (t[Γ], t[∆]), Γ, ∆ ⊆s Cs (S), Γ <0 ∆
Proposition 39.With this operations (Cs (S), <0 , −0 , ∗0 , t0 ) is a surreal algebra.
Proof.
(S1) <0 is acyclic because any cycle (A0 , B0 ) <0 ... <0 (An , Bn ) induces a cycle t(A0 , B0 ) <
... < t(An , Bn ) in S, which is acyclic.
(S2) −0 −0 (A, B)) = −0 (−B, −A) = (− − A, − − B) = (A, B).

(S3) −0 ∗0 = −0 (∅, ∅) = (−∅, −∅) = (∅, ∅) = ∗0 .
(S4) (A, B) <0 (C, D) iff t(A, B) < t(C, D) iff −t(C, D) < −t(A, B) iff t(−D, −C) <
t(−B, −A) iff (−D, −C) <0 (−B, −A) iff −0 (C, D) <0 −0 (A, B).
(S5) Let (Γ, ∆) ∈ Cs (Cs (S)). Then Γ <0 ∆ and thus t[Γ] < t[∆]. Since S satisfies (S5),
t[Γ] < t(t[Γ], t[∆]) < t[∆]. By the definition of <0 , Γ <0 (t[Γ], t[∆]) <0 ∆ and then
Γ <0 t0 (Γ, ∆) <0 ∆.
(S6) −0 t0 (Γ, ∆) = −0 (t[Γ], t[∆]) = (−t[∆], −t[Γ]) = (t[−0 ∆], t[−0 Γ]) = t0 (−0 ∆, −0 Γ)
(S7) t0 (∅, ∅) = (t[∅], t[∅]) = (∅, ∅) = ∗0

Some properties of the sur-algebra S are transferred to Cs (S) as we can see in the
above proposition:
Proposition 40.
(a) If S is transitive then Cs (S) is transitive.
(b) If S is linear then Cs (S) is pre-linear, i.e., denote ∼t the equivalence relation on
Cs (S) given by (A, B) ∼t (C, D) iff t(A, B) = t(C, D). Then it holds exactly one
between of the alternatives: (A, B) <0 (C, D); (A, B) ∼t (C, D); (C, D) <0 (A, B).
Proof.
(a) Suppose that we have (A1 , B1 ), (A2 , B2 ), (A3 , B3 ) ∈ Cs (S) satisfying (A1 , B1 ) <0
(A2 , B2 ) <0 (A3 , B3 ). Then, by definition, t(A1 , B1 ) < t(A2 , B2 ) < t(A3 , B3 ). Since
< is transitive, we have that t(A1 , B1 ) < t(A3 , B3 ) and then (A1 , B1 ) <0 (A3 , B3 ).
(b) Is straightforward.

If follows almost directly by the definition of the structure in Cs (S) that:
Proposition 41.t : Cs (S) → S is a morphism of surreal algebras.
Remark 42. In the case of the three principal examples of SUR-algebras we have that
t : Cs (SA) → SA and t : Cs (ST ) → ST are bijections and t : Cs (N o) → N o is a
surjection.
2.3. PARTIAL SURREAL ALGEBRAS AND MORPHISMS 41
Proposition 43.If f : S → S 0 is a sur-algebra morphism then Cs (f ) : Cs (S) → Cs (S 0 ) :

(A, B) 7→ (f [A], f [B]) is a morphism of sur-algebras.

Proposition 44.Cs determines a functor from SU R to SU R, and t determines a natural
transformation t : IdSU R−alg → Cs

From a direct application of the Propositions 40 and 41, we obtain the following:
Proposition 45.Let S = (S, <, −, ∗, t) a SUR-algebra.
1. If S is universal on the category SU R − alg then the following diagram commutes:

! t id
(S → Cs (S) → S) = (S →S S)
2. If S is an object of the full subcategory SU RT −alg ,→ SU R−alg, of all transitive

SUR-algebra, and is universal on the this category SU RT − alg, then the following
diagram commutes:
! t id
(S → Cs (S) → S) = (S →S S)

Remark 46. Note that: Cs (SA) = SA and Cs (ST ) = ST .
2.3 Partial Surreal Algebras and morphisms

In several recursive constructions, the intermediate stages plays an important role in
the comprehension of the object constructed. As we have seen in the Section 2.1, all
surreal algebra is a proper class but, on the other hand, the intermediate stages of
the constructions of N o, SA, ST are sets. To gain some flexibility and avoid technical
difficulties, we introduce in this Section the (more general and flexible ) notion of partial
surreal algebra: every SUR-algebra is a partial SUR-algebra and this new notion can be
supported on a set. Besides simple examples, that contains in particular the intermediate
stages of N o, SA, ST , and a relativized notion of Cut (partial) SUR-algebra, we are
interest on general constructions of partial SUR-algebras: for that we will consider two
kinds of morphisms between them. We will perform general constructions as products,
sub partial-SUR-algebra and certain kinds of directed colimits. As an application of the
latter construction, we are able to prove some universal properties satisfied by SA and ST
(and natural generalizations), that justifies its names of (relatively) free SUR-algebras.
Definition 47. A partial surreal algebra (pSUR-algebra) is a structure S = (S, <

, −, ∗, t) where S is a class (proper or improper), ∗ ∈ S, − is an unary function in S, <
is a binary relation in S and t : Cst (S) → S is a partial function, i.e., Cst (S) ⊆ Cs (S),
satisfying:
(pS1) < is an acyclic relation.
(pS2) ∀x ∈ S, −(−x) = x
(pS3) −∗ = ∗.
(pS4) ∀a, b ∈ S, a < b iff −b < −a
(pS5) If (A, B) ∈ Cst (S), then A < t(A, B) < B.
(pS6) If (A, B) ∈ Cst (S), then (−B, −A) ∈ Cst (S) and −t(A, B) = t(−B, −A).
(pS7) (∅, ∅) ∈ Cst (S) and ∗ = t(∅, ∅).
Note that (pS1), (pS2), (pS3) and (pS4) coincide, respectively, with the SUR-algebra
axioms (S1), (S2), (S3) and (S4). The statements (pS5), (pS6) and (pS7) are relative
versions of, respectively, the SUR-algebra axioms (S5), (S6) and (S7). SUR-algebras are
precisely the pSUR-algebras S such that Cst (S) = Cs (S).
Definition 48. Let S = (S, <, −, ∗, t) and S 0 = (S 0 , <0 , −0 , ∗0 , t0 ) be partial SUR-algebras.

Let h : S → S 0 be (total) function and consider the conditions below:
(Sm1) h(∗) = ∗0 .
(Sm2) h(−a) = −0 h(a), ∀a ∈ S.
(Sm3) a < b =⇒ h(a) <0 h(b), ∀a, b ∈ S.

0
(pSm4) (A, B) ∈ Cst (S) =⇒ (h[A], h[B]) ∈ Cst (S 0 ) and h(t(A, B)) = t0 (h[A], h[B]),
∀(A, B) ∈ Cst (S).
0
(fpSm4) (A, B) ∈ Cs (S) =⇒ (h[A], h[B]) ∈ Cst (S 0 ) and h(t(A, B)) = t0 (h[A], h[B]),
∀(A, B) ∈ Cst (S).
We will say that h : S → S 0 is:

• a partial SUR-algebra morphism (pSUR-morphism) when it satisfies: (Sm1),
(Sm2), (Sm3) and (pSm4);
• a full partial SUR-algebra morphism (fpSUR-morphism) when it satisfies:
(Sm1), (Sm2), (Sm3) and (fpSm4).

Remark 49.
• Note that the property (Sm3) entails: (A, B) ∈ Cs (S) =⇒ (h[A], h[B]) ∈ Cs (S 0 ).
• The conditions (Sm1), (Sm2) and (Sm3) are already present in the definition of
SUR-algebra morphism. The property:
(Sm4)
h(t(A, B)) = t0 (h[A], h[B]), ∀(A, B) ∈ Cs (S);
completes the definition of SUR-algebra morphism.
• Every full partial SUR-algebra morphism is partial SUR-algebra morphism.
• Let S, S 0 be partial SUR-algebras and h : S → S 0 is a map. Suppose that S or S 0

is a SUR-algebra, then h is a pSUR morphism iff h is a fpSUR-morphism.
• If S is a partial SUR-algebra, then: idS : S → S is a pSUR-morphism and idS :

S → S is a fpSUR-morphism iff S is a SUR-algebra.
• Let f : S → S 0 , f 0 : S 0 → S 00 be pSUR morphisms:

• Then f 0 ◦ f is a pSUR-morphism.
• If f is fpSUR-morphism, then f 0 ◦ f is a fpSUR-morphism. In particular, the
composition of fpSUR-morphisms is a fpSUR-morphism.

Definition 50. The category of partial SUR-algebras:
We will denote by pSU R−alg the (”very-large”) category such that Obj(pSU R−alg)
is the class of all partial SUR-algebras and M or(pSU R − alg) is the class of all partial
SUR-algebras morphisms, endowed with obvious composition and identities.

Remark 51.
(a) Of course, we have in the category pSU R−alg the same ”size issue” present in the
categories of ZF − alg and SU R − alg: we will adopt the same ”solution” explained in
Remark 37. An alternative is to consider only ”small” partial SUR-algebras (and obtain
a ”large” category –instead of very large– pSU Rs −alg, of all small partial SUR-algebras)
since we will see that there are set-size partial SUR-algebras: we will not pursue this
track because our main concern in considering partial SUR-algebras is get flexibility to
make (large indexed) categorial constructions with small partial SUR-algebras to obtain
a total SUR-algebra as a (co)limit process, i.e., we want pSU R ⊇ SU R.
(b) As we saw above that, even if the class of full morphism of partial SUR-algebras
is closed under composition, it does not determines a category under composition, since
it lacks the identities for the small partial SUR-algebras. However this notion will be
useful to perform constructions of total SUR-algebra as colimit of a large diagram small
partial SUR-algebras and fpSUR-morphisms between them (see Subsection 2.3.4).

52. Denote Σ-str the (very large) category such that:
(a) The objects of Σ-str are the structures S = (S, <, −, ∗, t) where S is a class,
∗ ∈ S, − is an unary function in S, < is a binary relation in S and t : Dt → S is a
function such that Dt ⊆ Ps (S) × Ps (S).
(b) Let S = (S, <, −, ∗, t) and S 0 = (S 0 , <0 , −0 , ∗0 , t0 ) be partial SUR-algebras. A
Σ-morphism, h : S → S 0 , is a (total) function h : S → S 0 satisfying the conditions
below:
(Σm1) h(∗) = ∗0 .
(Σm2) h(−a) = −0 h(a), ∀a ∈ S.
(Σm3) a < b =⇒ h(a) <0 h(b), ∀a, b ∈ S.

0
(Σm4) (h × h)[Dt ]⊆Dt and h(t(A, B)) = t0 (h[A], h[B]), ∀(A, B) ∈ Dt .
(c) Endowed with obvious composition and identities, Σ-str is a very large category
and
SU R − alg ,→ pSU R − alg ,→ Σ − str
are inclusions of full subcategories.

2.3.1 Simple examples
In this short Subsection we just present first examples of partial SUR-algebras and its
morphisms.
Example 53. Let (G, +, −, 0, <) be a linearly ordered group. For each a ∈ G such
that a ≥ 0 (respect. a ∈ G ∪ {∞} such that a > 0) then Xa := [−a, a] ⊆ G (respect.
Xa :=] − a, a[⊆G), is a partial SUR-algebra, endowed with obvious definitions of ∗, −, <
and such that:

(1) Cst (Xa ) := {(x< , x> ) : x ∈ Xa }, t(x< , x> ) := x ∈ Xa (t is bijective);
or, alternatively,
(2) Cst (Xa ) := {(L, R) ∈ Cs (Xa ) : ∃(!)x ∈ Xa L≤ = x< , R≥ = x> }, t(L, R) := x ∈ Xa (t
is surjective).
Note that if b ≥ a, then the inclusion Xa ,→ Xb is a pSU R-morphism, if Xa , Xb are
endowed with the second kind of t-map.

Another simple (and useful) class of examples are given by the ordinal steps of the
recursive constructions of the SUR-algebras SA, ST and N o.
Example 54. For any ordinal α we have that the Σ-structure (SAα , <α , −α , ∗α , tα ) is a
partial SUR-algebra with the below definitions:
• ∗α = ∗
• −α = − SAα
• <α =<SAα ×SAα
• Cst (SAα ) = Cs (SA(α) ) and tα = t Cs (SA(α) )
Just like in the previous example, we have:
Example 55. For any ordinal α we have that the Σ-structure (STα , <α , −α , ∗α , tα ) is a
partial SUR-algebra with the below definitions:
• ∗α = ∗
• −α = − STα
• <α =<STα ×STα
• Cst (STα ) = Cs (ST (α) ) and tα = t Cs (ST (α) )
Example 56. For any given α ∈ On, the Σ-structure (N oα , ∗α , −α , <α , tα ) is a partial
SUR-algebra with the operations defined below:
• ∗α = 0
• −α = − N oα
• <α =<N oα ×N oα
• Cst (N oα ) = Cs (N o(α) ) and tα = t Cs (N o(α) )

Remark 57.
• Note that in the three examples above S = SA, ST, N o, the inclusion Sα ,→ Sβ
is a pSUR-morphism, where α ≤ β ≤ ∞ are ”extended” ordinals, with the convention
S∞ := S.
• We can also define partial SUR-algebras on the sets SA(α) , ST (α) , N o(α) , for each
α ∈ On \ {0} (this is useful!).
• Note that iα : SA(α) ,→ SAα is a fpSUR-algebra morphism, for each α ∈ On \ {0}.
It can be established, by induction on α ∈ On \ {0} that for each γ < α iγα : SAγ ,→
SAα is a fpSUR-morphism. An analogous situation occurs to the partial SUR-algebras
STγ ,→ ST (α) ,→ STα .

2.3.2 Cut partial Surreal Algebras
In this short Subsection we present an adaption/generalization of the notion of ”Cut

Surreal Algebra”, introduced in the Subsection 2.2.4, to the realm of partial SUR-algebra.
Definition 58. Let S = (S, <, −, ∗, t) be a partial SUR-algebra. The Cut structure of S
is the Σ-structure S (t) = (S 0 , <0 , −0 , <0 , t0 ), where:
1. S 0 := Cst (S)
2. ∗0 := (∅, ∅)
3. −0 (A, B) := (−B, −A)
4. (A, B) <0 (C, D) ⇐⇒ t(A, B) < t(C, D)
5. ∀Γ, ∆⊆Cst (S), (Γ, ∆) ∈ dom(t0 ) iff Γ <0 ∆ and (t[Γ], t[∆]) ∈ dom(t)
0
6. t0 : Cst (Cst (S)) → Cst (S), (Γ, ∆) 7→ t0 (Γ, ∆) := (t[Γ], t[∆])

The list below is a sequence of results on Cut Partial SUR-algebras that extend the
results presented in the Subsection 2.2.4 on Cut SUR-algebras: its proofs will be omitted.
Proposition 59. Let S = (S, <, −, ∗, t) be a partial SUR-algebra. Then:
(a) S (t) = (S 0 , <0 , −0 , <0 , t0 ) as defined above is a partial SUR-algebra. Moreover, if
0
S is a SUR-algebra, i.e. Cst (S) = Cs (S), then S (t) is a SUR-algebra, i.e. Cst (Cst (S)) =
Cs (Cs (S)).
(b) t : Cst (S) → S is a morphism of partial SUR-algebras. Moreover, if S is a
SUR-algebra, then t is a fpSUR-algebra morphism.

Proposition 60. Let S = (S, <, −, ∗, t) be a partial SUR-algebra. Then:
(a) If S is transitive, then Cst (S) is transitive.
(b) If S is linear, then Cst (S) is pre-linear12 .

Proposition 61.
(a) If f : S → S 0 is a morphism of partial SUR-algebras then Cst (f ) : Cst (S) → Cst (S 0 ),
given by: (A, B) 7→ (f [A], f [B]) is a morphism of partial SUR-algebras.
(b) The cut partial SUR-algebra construction determines a (covariant) functor Cst :
pSU R → pSU R:
f C t (f )
(S → S 0 ) 7→ (Cst (S) →
s
Cst (S 0 ))
(c) The t-map determines a natural transformation between functors on pSU R − alg,
t : IdpSU R−alg → Cst .

2.3.3 Simple constructions on pSUR
In this Section, we will verify the full subcategory pSU R − alg ,→ Σ − str is closed under
some simple categorial constructions: as (Σ-)substructure and non-empty products. We
also present some results on initial objects and (weakly) terminal objects.
We can define a notion of substructure in the categories pSU R and Σ-str:
Definition 62. Let S = (S, <, −, ∗, t) and S 0 = (S 0 , <0 , −0 , ∗0 , t0 ) be Σ-structures. S will
be called a Σ-substructure of S whenever:
12
I.e., denote ∼t the equivalence relation on Cst (S) given by (A, B) ∼t (C, D) iff t(A, B) = t(C, D).
Then it holds exactly one between of the alternatives: (A, B) <0 (C, D); (A, B) ∼t (C, D); (C, D) <0
(A, B).
(s1) S ⊆ S 0 ;
(s2) <=<0S×S ;
(s3) − = −0S×S ;
(s4) ∗ = ∗0 ;
(s5) dom(t) = t0−1 [S]∩(Ps (S)×Ps (S)) := {(A, B) ∈ dom(t0 )∩(Ps (S)×Ps (S)) : t0 (A, B) ∈
S}⊆dom(t0 ) and t = t0 : dom(t) → S.

Remark 63.
(a) The inclusion i : S ,→ S 0 determines a Σ-morphism.
(b) By conditions (s1) and (s2) above note that Cs (S, <) = Cs (S 0 , <0 ) ∩ (Ps (S) ×
Ps (S)).
(c) By item (b): if dom(t0 )⊆Cs (S 0 , <0 ), then dom(t)⊆Cs (S, <).
(d) By the results presented in the Subsections 2.2.2 and 2.2.3, for any two extends
ordinals α ≤ β ≤ ∞ we have:
• SAα is a Σ-substructure of SAβ .
• STα is a Σ-substructure of STβ .
(e) An useful generalization of the notion of Σ-substructure is the notion of Σ-
embedding: a Σ-morphism j : S → S 0 is a Σ-embedding when:
(e1) it is injective;
(e2) ∀a, b ∈ S, (a < b⇔j(a) <0 j(b));
(e3) ∀(A, B) ∈ Ps (S) × Ps (S), ((A, B) ∈ dom(t)⇔t0 (j[A], j[B]) ∈ range(j)).
(e) An inclusion i : S ,→ S 0 determines a Σ-embedding precisely when S is a Σ-
substructure of S 0 . Note that the Σ-embeddings j : S → S 0 are precisely the Σ-
morphisms described (uniquely) as j = i ◦ h, where i : S j ,→ S 0 is a Σ-substructure
inclusion and h : S → S j is a Σ-isomorphism.
(f) For technical reasons, we consider an even more general notion: a Σ-morphism
j : S → S 0 is a Σ − quasi-embedding whenever it satisfies the conditions (e1) and (e3)
above.

By a straightforward verification we obtain the:
Proposition 64. Let j : S → S 0 be a Σ-embedding of Σ-structures. If S 0 is a partial

SUR-algebra, then S is a partial SUR-algebra.

Definition 65. Given a non-empty indexed set of partial Σ-structure S i = (Si , <i
, −i , ∗i , tiQ
), i ∈ I, we define the Σ-structure product S = (S, <, −, ∗, t) as follows:
(a) S = i∈I Si ;
(b) <= {((ai )i∈I , (bi )i∈I ) : ai <i bi , ∀i ∈ I};
(c) −(ai )i∈I = (−i ai )i∈I ;
(d) ∗ = (∗i )i∈IT ;
(e) dom(t) = i∈I (πi × πi )−1 [dom(ti )] = {((Ai )i∈I , (Bi )i∈I ) ∈ Ps (S) × Ps (S)) : (Ai , Bi ) ∈
dom(ti ), ∀i ∈ I} and t((Ai )i∈I , (Bi )i∈I )) = (ti (Ai , Bi ))i∈I .

Note that: For each i ∈ I, the projection πi : S → Si is a Σ-structure morphism.

By a straightforward verification we obtain:
Proposition 66. Keeping the notation above.
(a) The pair (S, (π)i∈I ) above defined constitutes a(the) categorial product in Σ-str.
I.e., for each diagram (S 0 , (fi )i∈I ) in Σ-str such that fi : S 0 → S i , ∀i ∈ I, there is a
unique Σ-morphism f : S 0 → S such that πi ◦ f = fi , ∀i ∈ I.
(b) Suppose that {S i : i ∈ I}⊆ pSUR-alg. Then S ∈ pSUR-alg and (S, (π)i∈I ) is the
product in the category pSUR-alg.

Proposition 67. Let f : S → S 0 be a pSUR-alg morphism. If (S, <) is strictly linearly
ordered, then:
(a) ∀a, b ∈ S, a < b ⇐⇒ f (a) <0 f (b);
(b) f is an injective function.
Proof. If a < b, then f (a) <0 f (b), since f is a Σ-structure morphism. Suppose that
f (a) <0 f (b) but a 6< b, then a = b or b < a, thus f (a) = f (b) or f (b) <0 f (a). In
any case, we get a contradiction with f (a) <0 f (b), since <0 is an acyclic relation. This
establishes item (a). Item (b) is similar, since < satisfies trichotomy and <0 is acyclic.

The result above yields some information concerning the empty product (= terminal
object) in pSUR-algebras.
Proposition 68. If there exists a weakly terminal object13 S 1 in the category pSUR-alg
then S 1 must be a proper class.
Proof. Suppose that S 1 is an weakly terminal object in pSUR-alg. Since the (proper
class) SUR-algebra N o is strictly ordered, then by Proposition 67 above anyone of the
13
Recall that an object in a category is weakly terminal when it is the target of some arrow departing
from each object of the category.
existing morphisms f : N o → S 1 is injective. Then S 1 (and Cst (S 1 )) must be a proper

class.
If we consider the small size version of pSUR, we can guarantee by an another ap-
plication of Proposition 67, that this (large but not very-large) category does not have
(weakly) terminal objects: there are small abelian linearly ordered abelian groups (or
even the additive part of a ordered/real closed field) of arbitrary large cardinality, and
we have seen in Example 53 how to produce small pSUR-algebras from that structures.
Concerning initial objects we have the following:
Proposition 69.
(a) Consider the Σ-structure S 0 = (S0 , <, −, ∗, t) over a singleton set S0 := {∗}, with
<:= ∅, Dt = dom(t) := ∅ (thus S 0 ∈ / pSU R − alg) and with − : S0 → S0 and t : Dt → S0
the unique functions available. Then S 0 is the (unique up to unique isomorphism) initial
object in Σ-str.
(b) Consider the Σ-structure S p0 = (S0 , <, −, ∗, tp ) over a singleton set S0 := {∗}, with
p p
<:= ∅, Dt = dom(tp ) := {(∅, ∅)}⊆Cs (S0 , <) and with − : S0 → S0 and tp : Dt → S0
the unique functions available. Then S p0 is the (unique up to unique isomorphism) initial
object in pSUR-alg.
Proof.
(a) Let S 0 be a Σ-structure and let h : {∗} → S 0 be the unique function such that
h(∗) = ∗0 ∈ S 0 , then clearly h is the unique Σ-structure morphism from S 0 into S 0 : note
that (h × h) : dom(t) = ∅ → dom(t0 ) is such that t0 ◦ (h × h) = h ◦ t.
(b) It is easy to see that S p0 is a partial SUR-algebra. Let S 0 be a partial SUR-algebra
and let h : {∗} → S 0 be the unique function such that h(∗) = ∗0 ∈ S 0 , then clearly h
is the unique Σ-structure morphism from S 0 into S 0 : since (∅, ∅) ∈ dom(t0 ), note that
(h × h) : dom(tp ) := {(∅, ∅)} → dom(t0 ) is such that t0 ◦ (h × h) = h ◦ t.

2.3.4 Directed colimits of partial Surreal Algebras
One of the main general constructions in Mathematics is the colimit of an upward di-
rected diagram. In the realm of partial SUR-algebras this turns out to be essential for
the constructions of SUR-algebras and to obtain general results about them. We can
recognize the utility of this process by the cumulative constructions of our main exam-
ples: N o, SA, ST . Thus we will be concerned only with the colimit of small partial
SUR-algebras, but over a possibly large directed diagram.
This Section is completely technical but its consequences/applications are very inter-
esting: see the entire Section 2.4 and the Theorems at the end of Section 3.1.
Recall that:
• Given a regular ”extended” cardinal κ (where ”card(X) = ∞” means that X is a
proper class14 ), a partially ordered class (I, ≤) will be κ-directed, if every subclass I 0 ⊆I
such that card(I 0 ) < κ admits an upper bound in I.
• pSU Rs − alg denotes the full subcategory of pSU R − alg determined by of all
small partial SUR-algebras and its morphisms (then SU R − alg ∩ pSU Rs − alg = ∅).
Analogously, we will denote Σs -str the full subcategory of Σ-str determined by of all
small partial Σ-structures and its morphisms.
70. The (first-order) directed colimit construction: Let (I, ≤) is a ω-directed

hij
ordered class and consider D : (I, ≤) → Σs − str, (i ≤ j) 7→ (S i → S j ) be a diagram.
Define:
• S∞ := (ti∈I Si )/ ≡, the set-theoretical colimit, i.e. ≡ is the equivalence relation on the
class ti∈I Si such that (ai , i) ≡ (aj , j) iff there is k ≥ i, j such that hik (ai ) = hjk (aj ) ∈ Sk ;
• hj : Sj → S∞ , aj 7→ [(aj , j)];
• ∗ := [(∗i , i)] (= [(∗j , j)], ∀i, j ∈ I);
• −[(ai , i)] = [(−i ai , i)];
• [(ai , i)] < [(aj , j)] iff there is k ≥ i, j such that hik (ai ) <k hjk (aj ) ∈ Sk

With the construction above, it is straightforward to verify that (S∞ , <, −, ∗) is the
colimit in the appropriate category of first-order (but possibly large) structures15 , with
colimit co-cone (hj : Sj → S∞ )j∈I and, if D : (I, ≤) → pSU Rs − alg, then the same
(colimit) co-cone is in the ”first-order part” of the category pSU R − alg, i.e., it satisfies
the properties [pS1]–[pS4] presented in Definition 47. However, to ”complete” the Σ-
structure (respect. pSUR-algebra) we will need some extra conditions below:
hij
Proposition 71. Let D : (I, ≤) → Σs − str, (i ≤ j) 7→ (S i → S j ) be a diagram such
that:
(i) (I, ≤) is a ω-directed ordered class and hij : Si → Sj is a injective Σ-morphism,
whenever i ≤ j;
or;
(ii) (I, ≤) is a ∞-directed ordered class (e.g. (On, ≤)),
then S∞ := (ti∈I Si )/ ≡ is a (possibly large) partial Σ-structure and (hj : Sj → S∞ )j∈I
is a colimit cone in the category Σ − str.
14
Recall that in NBG, all the proper classes are in bijection, by the global form of the axiom of choice.
15
I.e., we drop the second-order part of the Σ-structure: the map t : dom(t)⊆Ps (S∞ )×Ps (S∞ ) → S∞ .
hij
Proposition 72. If D : (I, ≤) → pSU Rs − alg, (i ≤ j) 7→ (S i → S j ) is a diagram,
where:
(i) (I, ≤) is a ω-directed ordered class and hij : Si → Sj is a injective pSUR-morphism,
whenever i ≤ j;
or;
(ii) (I, ≤) is a ∞-directed ordered class (e.g. (On, ≤))
then S∞ is a (possibly large) partial SUR-algebra and (hj : Sj → S∞ )j∈I is a colimit cone
in the category pSU R − alg.

Proposition 73. The subclass of morphisms f pSU R ⊆ pSU R is closed under directed
colimits in the cases (i) and (ii) described in the Proposition above. More precisely: if
D : (I, ≤) → pSU Rs − alg is a directed diagram satisfying (i) or/and (ii)above and
such that hij : Si → Sj is a fpSUR-morphism, whenever i < j, then the colimit co-cone
∀j ∈ I, (hj : Sj → S∞ )j∈I is formed by fpSUR-algebra morphisms. Moreover:
(a) If (I, ≤) is ∞-directed and the transition arrows (hij )i≤j are injective, then S∞ is a
SUR-algebra (thus it is a proper class);
(b) If the transition arrows (hij )i≤j are injective (respect. Σ − quasi-embedding, Σ-
embedding), then the cocone arrows (hj )j∈I are injective (respect. Σ − quasi-embedding,
Σ-embedding);
(c) If ti : Csti (Si ) → Si is injective (respect. surjective/bijective), ∀i ∈ I, then t∞ :
∞
Cst (S ∞ ) → S ∞ is injective (respect. surjective/bijective).

Example 74.
We have noted in Remark 57 that for each sequence of ordinal γ < β < α, iγβ :
SAγ ,→ SAβ is fpSUR-algebra morphism. It is also a Σ-embedding. Then, for each
α > 0, SA(α) ∼
(α)
= colimγ<α SAγ as a pSUR-algebra and iγ : SAγ ,→ SA(α) determines
a colimit co-cone of an ω-directed diagram16 formed by fpSUR-algebras embeddings.
Moreover SA = SA∞ ∼ = colimγ∈On SAγ and i∞ γ : SAγ ,→ SA determines a colimit
co-cone a ∞-directed diagram over formed by fpSUR-algebras embeddings.
Analogous results holds for ST (α) ∼
= colimγ<α STγ , α > 0, and ST ∼
= colimγ∈On STγ .

16
In fact it is κ directed diagram, where κ is any regular cardinal such that κ ≥ α + ω.
2.4. UNIVERSAL SURREAL ALGEBRAS 53
2.4 Universal Surreal Algebras

In this final Section of Chapter 2, we present some categorical-theoretic universal prop-
erties17 concerning SUR-algebras and partial SUR-algebras. We will need notions, con-
structions and results developed in the previous Sections of this Chapter to provide,
for each small partial SUR-algebra I, a ”best” SUR-algebra over I, SA(I), (respect. a
”best” transitive SUR-algebra over I, ST (I)). As a consequence of this result (and its
proof) we will determine the SUR-algebras SA and ST in the category of SUR by uni-
versal properties that characterizes them uniquely up to unique isomorphisms: these will
justify the adopted names ”SA = the free surreal algebra” and ”ST = the free transitive
surreal algebra”.
We start with the following
75. Main construction: Let I = (I, <, −, ∗, t) be a partial SUR-algebra. Consider:

t incl
(a) The set-theoretical pushout diagram over (I ← Cst (I) ,→ Cs (I)):
i1
Cs (I) -
(I t Cs (I))/ ∼
6 6
incl i0
Cst (I) - I
t
Note that:
+
• I = (I tCs (I))/ ∼, is the vertex of the set-theoretical pushout diagram, where ∼ is the
least equivalence relation18 on I t Cs (I) such that (x, 0) ∼ ((A, B), 1) iff (A, B) ∈ Cst (I)
and x = t(A, B).
• If I is small, then I + is small.
• ∀(A, B), (C, D) ∈ Cs (I) \ Cst (I), ((A, B), 1) ∼ ((C, D), 1) iff (A, B) = (C, D) (by in-
duction on the number of steps that witness the transitive closure).
• ∀x, y ∈ I, (x, 0) ∼ (y, 0) iff x = y (by induction on the number of steps needed in the
transitive closure).
• Since Cst (I) ,→ Cs (I) is injective function, then i0 : I → (I t Cs (I))/ ∼, x 7→ [(x, 0)]
17
An analysis of model-theoretic universal properties of the ”first-order part” of (partial) SUR-
algebras, and its possible connections with categorial-theoretic universality presented here, will be theme
of future research, see Chapter 5 for more details.
18
Recall that the least equivalence relation on a set X that contains R⊆X × X is obtained from R
adding the opposite relation R−1 and the diagonal relation ∆X , and then taking the transitive closure
trcl(R ∪ R−1 ∪ ∆X ) = R(eq,X) .
is an injective function (see above) and (i0 )+ : (Ps (I) × Ps (I)) → (Ps (I + ) × Ps (I + )),
(A, B) 7→ (i0 [A], i0 [B]) is an injective function.
(b) Let I + := (I + , ∗+ , −+ , <+ , t+ ) the Σ-structure defined below:

• I + := (I t Cs (I))/ ∼.
• ∗+ := [(∗, 0)] = [((∅, ∅), 1)].
• −+ [(x, 0)] := [(−x, 0)];
−+ [((A, B), 1)] := [((−B, −A), 1)].
• Define <+ by cases (only three):
[(x, 0)] <+ [(y, 0)] iff x < y;
[(x, 0)] <+ [((A, B), 1)] iff x ∈ A, whenever (A, B) ∈ Cs (I) \ Cst (I);
[((A, B), 1)] <+ [(y, 0)] iff y ∈ B, whenever (A, B) ∈ Cs (I) \ Cst (I).
Note that Cs (i0 ) = (i0 )+ : Cs (I) → Cs (I + ), (A, B) 7→ (i0 [A], i0 [B]), is an injective
function with adequate domain and codomain.
• Define Cst (I + ) := range(Cs (i0 ))⊆Cs (I + ) (thus Cs (I) ∼
+ + +
= Cst (I + )) and t+ : Cst (I + ) →
I + , (i0 [A], i0 [B]) 7→ t+ (i0 [A], i0 [B]) := [((A, B), 1)].
Note that (A, B) ∈ dom(t) iff t+ (i0 [A], i0 [B]) ∈ range(i0 ).
Thus we obtain another set-theoretical pushout diagram that is isomorphic to the

previous pushout diagram:
+
t+
Cst (I + ) -
I+
6 6
(i0 × i0 ) i0
Cst (I) - I
t

We describe below the main technical result in this Section:

Lemma 76. Let I = (I, ∗, −, <, t) be a (small) partial SUR-algebra and keep the notation
in 75 above. Then
(a) I + = (I + , ∗+ , −+ , <+ , t+ ) is a (small) partial SUR-algebra.
(b) i0 : I → I + is a Σ-embedding and full morphism of partial SUR-algebras.
+
(c) If t : Cst (I) → I is injective (respect. surjective/bijective), then t+ : Cst (I + ) → I +
is injective (respect. surjective/bijective).
(d) If S 0 = (S 0 , ∗0 , −0 , <0 , t0 ) is a partial SUR-algebra, then for each fpSUR-algebra

morphism f : I → S 0 there is a unique pSUR-algebra morphism f + : I + → S 0 such
that f + ◦ i0 = f . In particular, if S 0 is a SUR-algebra, then f and f + are automatically
fpSUR-algebras morphisms. Moreover:
• If t0 is injective, then f is a Σ − quasi-embedding iff f + is a Σ − quasi-embedding.
Proof. Items (a), (b) and (c) are straightforward verifications. We will just sketch the
proof of the universal property in item (d).
Candidate and uniqueness:

Suppose that there is a pSUR-algebra morphism f + : I + → S 0 such that f + ◦ i0 =
f . Since f : I → S is a full partial SUR-algebra morphism, we have (f × f ) :
0 + 0
Cs (I) → Cst (S 0 ). Then (f + × f + ) : Cst (I + ) → Cst (S 0 ) : (Γ, ∆) = (i0 [A], i0 [B]) 7→
0
(f + [Γ], f + [∆]) = (f [A], f [B]) ∈ Cst (S 0 ) and f + (t+ (Γ, ∆)) = t0 (f + [Γ], f + [∆]) =
t0 ((f [A], f [B])) ∈ S 0 . Since range(i0 ) ∪ range(i1 ) = I + , the function f + is determined
by f :
• f + (z) = f (x) ∈ S 0 , whenever z = [(x, 0)] ∈ range(i0 );
• f + (z) = t0 ((f [A], f [B])) ∈ S 0 , whenever z = ([(A, B), 1)] ∈ range(i1 ).
Existence:
Since f : I → S is a full partial SUR-algebra morphism, we have (f × f ) : Cs (I) →
0
Cst (S 0 ), then the arrows
t0 ◦(f ×f ) f
(Cs (I) −→ S 0 ←− I)
yields a commutative co-cone over the diagram
t incl
(I ← Cst (I) ,→ Cs (I)).
By the universal property of set-theoretical pushout, there is a unique function f + :

I + → S 0 such that:
• f + ◦ i0 = f ;
• f + ◦ i1 = t0 ◦ (f × f ) .
Thus it remains only to check that f + : I + → S 0 is a pSUR-algebra morphism:
• f + (∗+ ) = f + ([∗, 0]) = f (∗) = ∗0 ;
• f + (−+ [(x, 0)]) = f (−x) = −0 f (x) = −0 f + ([(x, 0)]);
f + (−+ [((A, B), 1)]) = t0 ((f × f ) (−B, −A)) = t0 (f [−B], f [−A]) = −0 t0 (f [B], f [A]) =
−0 f + ([((A, B), 1)]).
• If [(x, 0)] <+ [(y, 0)], then x < y thus f + ([(x, 0)]) = f (x) <0 f (y) = f + ([(y, 0)]);
If (A, B) ∈ Cs (I) \ Cst (I):
0
- if [(x, 0)] <+ [((A, B), 1)], then x ∈ A and f (x) ∈ f [A]. Since(f [A], f [B]) ∈ Cst (S 0 ),
thus f + ([(x, 0)]) = f (x) <0 t0 (f [A], f [B]) = f + ([((A, B), 1)]);
0
- if [((A, B), 1)] <+ [(y, 0)], then y ∈ B and f (y) ∈ f [B]. Since (f [A], f [B]) ∈ Cst (S 0 ),
thus f + ([((A, B), 1)]) = t0 (f [A], f [B]) <0 f (y) = f + ([(y, 0)]).
+ 0
• If (Γ, ∆) = (i0 [A], i0 [B]) ∈ Cst (I + ), then (f + [Γ], f + [∆]) = (f [A], f [B]) ∈ Cst (S 0 ) and
f + (t+ (Γ, ∆)) = f + (t+ ((i0 [A], i0 [B])) = f + ([((A, B), 1)]) = t0 ◦ (f × f ) ([((A, B), 1)]) =
t0 (f [A], f [B]) = t0 (f + [i0 [A]], f + [i0 [B]]) = t0 (f + [Γ], f + [∆]).

Remark 77.
In the setting above, we can interpret the Conway’s notions in a very natural way:
• Old(I) := i0 [I] ∼
= I;
+
• M ade(I) := I ;
• N ew(I) := I + \ i0 [I].
Note that if t : Cst (I) → I is surjective (e.g. I = N o(α) , SA(α) , ST (α) , α ∈ On \ {0}),
+
then t+ : Cst (I + ) → I + and every ”made member” is represented by a Conway cut in of
”old members”. This representation is unique, whenever t : Cst (I) → I is bijective (e.g.,
I = SA(α) , ST (α) , α ∈ On \ {0}).
When I = SA(α) , α ∈ On \ {0} and Cst (I) = {(A, B) ∈ Cs (SA(α) , <(α) ) : t(A, B) =
+
hA, Bi ∈ SA(α) (t : Cs (I) → I is bijective), then t+ : Cst (I + ) → I + can be identified
with the (bijective) map Cs (SA(α) , <(α) ) → SAα .

A slight modification in the construction of the Σ-structure presented in 75 above,

just replacing <+ by <+ +
(tc) := trcl(< ), yields the following:
Lemma 78. Let I = (I, ∗, −, <, t) be a (small) partial SUR-algebra and keep the notation
in 75 above. Then
(a) I + + + + + +
(tc) = (I , ∗ , − , <(tc) , t ) is a (small) transitive partial SUR-algebra.
+
(b) i0 : I → I(tc) is a Σ − quasi-embedding (see Remark 63.(f )) and full morphism of
+
partial SUR-algebras. Moreover, if I is a transitive SUR-algebra, then i0 : I → I(tc) is a
Σ-embedding.
+
(c) If t : Cst (I) → I is injective (respect. surjective/bijective), then t+ : Cst (I + ) → I +
is injective (respect. surjective/bijective).
(d) If S 0 = (S 0 , ∗0 , −0 , <0 , t0 ) is a partial transitive SUR-algebra, then for each fpSUR-
algebra morphism f : I → S 0 there is a unique pSUR-algebra morphism f + : I + → S 0
such that f + ◦ i0 = f . In particular, if S 0 is a transitive SUR-algebra, then f and f + are
automatically fpSUR-algebras morphisms.

When I = ST (α) , α ∈ On \ {0} and Cst (I) = {(A, B) ∈ Cs (ST (α) , <(α) ) : t(A, B) =
+ + +
hA, Bi ∈ ST (α) } (t : Cs (I) → I is bijective), then t+ : Cst (I(tc) ) → I(tc) can be identified
with the (bijective) map Cs (ST (α) , <(α) ) → STα .
Remark 79.
Note that applying the construction ( )+ to the SUR-algebra SA we obtain (SA)+ ∼
=
Cs (SA) = SA.
Applying both constructions ( )+ and ( )+
(tc) to the SUR-algebra ST we obtain
(ST )+ = (ST )(tc) ∼
+
= Cs (ST ) = ST .

Now we are ready to state and prove the main result of this Section:
Theorem 80. Let I be any small partial SUR-algebra. Then there exists SUR-algebras
denoted by SA(I) and ST (I), and pSUR-morphisms jIA : I → SA(I) and jIT : I →
ST (I) such that:
(a)
(a1) jIA is a fpSUR-morphism and a Σ-embedding;
∞
(a2) If t : Cst (I) → I is injective (respect. surjective/bijective), then t∞ : Cst (SA(I)) →
SA(I) is injective (respect. surjective/bijective);
(a3) jIA : I → SA(I) satisfies the universal property: for each SUR-algebra S and each
pSUR-morphism h : I → S, there is a unique SUR-morphism hA : SA(I) → S such that
hA ◦ jIA = h. Moreover:
• If t0 is injective, then h is a Σ − quasi-embedding iff hA is a Σ − quasi-embedding.
(b)
(b1) jIT is a fpSUR-morphism and a Σ−quasi-embedding, that is a Σ-embedding whenever
I is transitive;
∞
(b2) If t : Cst (I) → I is injective (respect. surjective/bijective), then t∞ : Cst (ST (I)) →
ST (I) is injective (respect. surjective/bijective);
(b3) jIT : I → ST (I) satisfies the universal property: for each transitive SUR-algebra S
and each pSUR-morphism h : I → S, there is a unique SUR-morphism hT : ST (I) → S
such that hT ◦ jIT = h.
Proof.
Item (a): based on Lemma 76 and Proposition 73, we can define, by transfinite recur-
sion, a convenient increasing (compatible) family of diagrams Dα : [0, α] → pSU R − alg,
α ∈ On, where:
(D0) D0 ({0}) = I;
(D1) For each 0 ≤ γ < β < α, Dα (γ, β) = Dβ (γ, β) : Dβ (γ) → Dβ (β) is Σ-embedding
and a fpSUR-morphism;
(α) (α)
Just define Dα (α) = (Dα )+ , where Dα := colimβ<α Dα (β) and take, for β <
α, Dα (β, α) = (hαβ )+ : Dα (β) → (colimβ<α Dα (β))+ be the unique pSUR-morphism –
that is automatically a fpSUR-morphism and a Σ-embedding, whenever hαβ satisfies this

conditions (see Lemma 76.(d))– such that (hαβ )+ ◦i0 = hαβ , where i0 : (colimβ<α Dα (β)) →
(colimβ<α Dα (β))+ and where hαβ : Dα (β) → (colimβ<α Dα (β)) is the colimit co-cone
arrow: by the recursive construction and by Proposition 73 hαβ is a fpSUR-morphism
and a Σ-embedding. This completes the recursion.
Gluing this increasing family of diagrams we obtain a diagram D∞ : On → pSU R −
alg.
By simplicity we will just denote:
• SA(I)α := D∞ (α), α ∈ On;
• SA(I)∞ := colimα∈On SA(I)α ;
A
• Dα (β, α) = jβ,α , for each 0 ≤ β ≤ α ≤ ∞ (since the family (Dα )α is increasing, we
just have to introduce notation for ”new arrows”).
Then we set: SA(I) := SA(I)∞ and jIA := j0,∞
A
.
The verification that SA(I) is a SUR-algebra that satisfies the property in item (a2)
and that jIA satisfies item (a1)19 , follows the recursive construction of the diagram and
from a combination of Proposition 73 and Lemma 76.
By the same Lemma and Proposition combined, it can be checked by induction
that for each α ∈ On, there is a unique pSUR-morphism hα : SA(I)α → S such that
A
hα ◦ j0,α = h and such that hα is injective (respect. Σ − quasi-embedding, Σ-embedding),
whenever h is injective (respect. Σ − quasi-embedding, Σ-embedding). By applying one
more time the colimit construction, we can guarantee that there is a unique pSUR-
morphism hA := h∞ : SA(I)∞ → S such that hA ◦ jIA = h and that it satisfies the
additional conditions.
The proof of item (b) is analogous to the proof of item (a): basically we just have to
replace to use of technical Lemma 76 by other technical Lemma 78. In general, we can on
T
guarantee that jβ,α is a Σ-embedding and a fpSUR-morphism only for 0 < β < α ≤ ∞.

In particular, taking I = S 0 as the initial object in pSUR-alg (see Proposition 69

in Subsection 2.3.3), we have that SA ∼ = SA(I) and ST ∼ = ST (I), and they satisfy
corresponding universal properties:
Corollary 81.
(a) SA is universal (= initial object) over all SUR-algebras, i.e. for each SUR-algebra
S, there is a unique SUR-algebra morphism fS : SA → S.
19 A A
In fact, jβ,α is Σ-embedding whenever 0 ≤ β ≤ α ≤ ∞ and jβ,α is a fpSUR-morphism whenever
0 ≤ β < α ≤ ∞.
(b) ST is universal (= initial object) over all transitive SUR-algebras, i.e. for each
transitive SUR-algebra S 0 , there is a unique SUR-algebra morphism hS 0 : ST → S 0 .
Proof. Item (a): Since for each each SUR-algebra S there is a unique pSUR-morphism
uS : S 0 → S then, by Theorem 80.(a) above, SA(S 0 ) is a SUR-algebra that has the
required universal property, thus we only have to guarantee that SA ∼
= SA(S 0 ). Taking
into account the Remark 77 and the constructions performed in the proof of the item
(a) in Theorem above, that we have a (large) family of compatible pSUR-isomorphisms
SAα = SA(I)α , ∀α ∈ On. Thus SA = α∈On SAα ∼
∼ S
= colimα∈On SA(S 0 )α = SA(S 0 )∞ =
SA(S 0 ).
For item (b) the reasoning is similar: note that I = S 0 = {∗} is a transitive
partial SUR-algebra to conclude that ST (S 0 ) has the required universal S
property and
note that by the proof of item (b) in Theorem 80 above, that ST = α∈On STα ∼ =
colimα∈On ST (S 0 )α = ST (S 0 )∞ = ST (S 0 ).

This Corollary describes, in particular, that SA and ST are ”rigid” as Σ-structures

and :
• SA and Cs (SA) are isomorphic SUR-algebras and the universal map SA → Cs (SA) is
the unique iso from SA to Cs (SA);
• ST and Cs (ST ) are isomorphic SUR-algebras and the universal map ST → Cs (ST ) is
the unique iso from ST to Cs (ST ).
We finish this Section with an application of the Corollary above: we obtain some
non-existence results.
Corollary 82.
(i) Let L be a linear SUR-algebra, i.e., < is a total relation (for instance take L = N o).
Then there is no SUR-algebra morphism h : L → ST .
(ii) Let T be a transitive SUR-algebra, i.e., < is a transitive relation (for instance take
T = ST, N o). Then there is no SUR-algebra morphism h : T → SA.
Proof. (i) Suppose that there is a SUR-algebra morphism h : L → ST . Since the

binary relation < in L is acyclic and total, it is a strictly linear order, in particular it
is transitive. Let a, b ∈ L, since L is linear, a < b in L ⇔ h(a) < h(b) in ST . Now,
by the universal property of ST (see Theorem above) there is a unique SUR-algebra
morphism u : ST → L and then h ◦ u = idST . Summing up, h : (L, <) → (ST, <) is
an isomorphism of structures, thus (ST, <) is a strictly ordered class, but the members
of ST 0 and h{−1}, {1}i are not comparable by Fact 1.(c) in the Subsection 2.2.3, a
contradiction.
(ii) Suppose that there is a SUR-algebra morphism h : T → SA. Since the binary
relation < in T is transitive, by the universal property of ST there is a (unique) SUR-

algebra morphism v : ST → T , thus we get a SUR-algebra morphism g = h ◦ v :
ST → SA. By the universal property of SA there is a unique SUR-algebra morphism
u : SA → ST (u is a inclusion) and then g ◦ u = idSA . Thus, for each a, b ∈ SA, a < b
in SA ⇔ u(a) < u(b) in ST , but the members of SA denoted by −1 and 1 are not
related (see Fact.(c) in the Subsection 2.2.2) and u(−1) < u(1) in ST (by Fact 1.(c) in
the Subsection 2.2.3), a contradiction.

Chapter 3
SUR-algebras and ZF-algebras
In this Chapter, we will establish relations, in both directions, between certain classes of
(equipped) SUR-algebras and certain classes of (equipped) ZF-algebras, that ”explains”
j
and ”expands” the relations On N o. We introduce here the following concepts:
b
anchored SUR-algebras (Section 1), the hereditary positive subclass of an equipped SUR-
algebra (Section 3), the space of signs associated to some standard ZF-algebra (Section
4).
Even if we present some general definitions, constructions and results, the main goal
of this Chapter is not develop an extensive and systematic study of the introduced con-
cepts1 , but provide means to appreciate the content of the Section 2, named ”The main
diagram”, that summarizes the relationship between the structured classes On, V, N o, SA
and will provide a theoretical basis for the development of the next Chapter on ”set the-
ory based on surreal numbers”.
3.1 Anchored Surreal Algebras

Motivated by the axiomatization of the class of all surreal number (N o, <, b), presented
in the Subsection 1.3.2 of Chapter 1, where b : N o → On is the ”birthday function”,
we begin the present Section introducing a generalization: the concept of anchor on a
SUR-algebra. We develop some general results on anchored SUR-algebras, provide two
distinct anchors on the free SUR-algebra SA and another anchor on the free transitive
SUR-algebra ST and establish characterization results of these anchored SUR-algebras
in the same vein that the class of all surreal numbers can be axiomatized by its birthday
function.
Definition 83. Let C = (C, ≺, Φ) be a rooted well-founded class (see 5 in the Subsection
1
We intend develop these themes in a future research project.
61
62 CHAPTER 3. SUR-ALGEBRAS AND ZF-ALGEBRAS
1.1.2, Chapter 1). Denote v the pre-order relation on C associated with ≺, according 3
in the Subsection 1.1.1, Chapter 1. Let S be a SUR-algebra. An anchor b of S in C
(notation: b : S → C) is a function b : S → C that satisfies the conditions (b1)–(b5)
below:
(b1) b is a surjective function.
(b2) b has small fibers (see 8 in the Subsection 1.2.1, Chapter 1).
(b3) b(∗) = Φ.
(b4) b(−a) = b(a), ∀a ∈ S.
(b5) For each subsets A, B⊆S such that A < B and each c ∈ C such that ∀a ∈
A, ∀b ∈ B, b(a), b(b) ≺ c, then b(t(A, B)) v c.
The binary relations ≺, v in C determines binary relations ≺b , vb in S by:

• a ≺b b iff b(a) ≺ b(b), ∀a, b ∈ S;
• a vb b iff b(a) v b(b), ∀a, b ∈ S.
For each subsets A, B⊆S such that A < B, consider M umb (A, B) the subclass of
all u ∈ S such that A < u < B and that is a vb -minimum for that property (i.e., if
A < z < B, then u vb z).
The anchor b : S → C will be called strict if it also satisfy (b6) and (b7) below:
(b6) ∀(A, B) ∈ Cs (S), t(A, B) ∈ M umb (A, B).
(b7) ∀(A, B) ∈ Cs (S), if s, s0 ∈ M umb (A, B) are <-comparable (i.e., s = s0 or s < s0
or s0 < s), then s = s0 .
• The pair (S, b), where b : S → C is an anchor, is called an anchored SUR-algebra.

• If b : S → C and b0 : S 0 → C are anchor over the same rooted well-founded class C,
a C-morphism of anchored SUR-algebras h : (S, b) → (S 0 , b0 ) is a SUR-algebra morphism
h : S → S 0 such that b0 ◦ h = b.
• We denote C − ancSU R − alg the category2 naturally obtained from the setting
described just above.

Remark 84. Let b : S → C be an anchor. Note that:

On sizes: Since S is a proper class (see Remark 34) then C is Sa proper class too:
this follows from the conditions (b1) and (b2) and the equality S = c∈C b−1 [{c}].
2
Apart from the usual questions concerning sizes that occurs in category-theory.
3.1. ANCHORED SURREAL ALGEBRAS 63
On (b2): As we saw in 8 in Subsection 1.2.1 in Chapter 1, condition (b2) is equivalent

to:
(b2)’ b is a locally small function.
On (b3): The condition b(∗) = Φ is inessential, since it follows from (b5) and the
hypothesis that Φ is the unique root in (C, ≺) (see 5, Chapter 1).
On (b4): The condition b(−a) = b(a), ∀a ∈ S, means that ”the members of S born
in pairs, at the same time”.
On (b5): A stronger version of condition (b5) will appear in some examples:
(b5)strong For each subsets A, B⊆S such that A < B, then:
for each c ∈ C: ∀a ∈ A, ∀b ∈ B, b(a), b(b) ≺ c (in C) iff b(t(A, B)) v c;
or, equivalently:
b(t(A, B)) = minv {c ∈ C : ∀a ∈ A, ∀b ∈ B, b(a), b(b) ≺ c}.
This strong condition entails:
• t−1 [b−1 [{Φ}]] = {(∅, ∅)}; moreover, if t : Cs (S) → S is a surjective function, then
b−1 [{Φ}] = {∗}.
• ∀(A, B) ∈ Cs (S), ∀a ∈ A, ∀b ∈ B : b(a), b(b) ≺ b(t(A, B))
On (b6), (b7):
If C is a rooted extensional well-founded class, note that ∀(A, B) ∈ Cs (S), b[M umb (A, B)]
is a singleton in C, whenever M umb (A, B) is non-empty subclass of S.
Other versions of conditions (b6) and (b7) will appear in examples:
(b7)comp Means that (b7) holds and that, moreover, every pair s, s0 ∈ M umb (A, B)
are <-comparable.
This condition and (b6) entails: M umb (A, B) = {t(A, B)}.
Let M alb (A, B) the subclass of all s ∈ S such that A < s < B and that is ≺b -
minimal for that property (i.e., if z ≺b s, then A < z < B doesn’t hold). Since
M umb (A, B)⊆M alb (A, B) (because ≺ is irreflexive and (S, <) is an η∞ class), we define:
(b6)weak t(A, B) ∈ M alb (A, B);
(b7)strong if s, s0 ∈ M alb (A, B) are <-comparable (i.e., s = s0 or s < s0 or s0 < s), then
s = s0 .
If ≺ is a strict linear order relation in C, then: ∀x, y ∈ S, x vb y iff x ⊀b by. Thus
M umb (A, B) = M alb (A, B), ∀(A, B) ∈ Cs (A) and (b6) ⇔ (b6)weak , (b7) ⇔ (b7)strong .

85. N o as (strict) anchored SUR-algebra: Naturally, the SUR-algebra (N o, <

, −, ∗, t) (Chapter 2, Section 2.1), endowed with the birthday function b : N o On,
is a strictly anchored SUR-algebra, since:
• (On, ∈, ∅) is an well-founded extensional rooted class, that is derived from the standard
ZF-algebra On (see Remark 17.(a) and 18);
• By a careful reading of the results presented in the Section 3.2 of Chapter 1 and the fact
that (On, ∈) is a strictly linearly ordered class, we can see that are fulfilled all the axioms
(b1)–(b7) (and some other stronger versions). For that, note that in On: ≺=∈=<=(,
v=3 ⊆ =≤.
This anchored SUR-algebra has some distinctive additional properties:
• t : Cs (N o) → N o is a surjective function.
• (N o, <) is a strictly linearly ordered class (equivalently, since < is acyclic, < is a total
relation).
• It is anchored in (On, ∈, ∅) that is an well-ordered class.
• It satisfies (b6)weak , (b7)strong and (b7)comp .
• It does not satisfy (b5)strong since: {−1|1} = 0 and b(0) = 0 and b(±1) = 1, where
0 := ∗ = {∅|∅}, 1 := {∅|{0}}, −1 := {{0}|∅}.

Definition 86. The binary relations derived from an anchor: Each function
b : S → C (in particular, each anchor b : S → C) induces the following binary relations
in S.
Let x, y ∈ S:
• x ≡b y iff b(x) = b(y);
• x ≺b y iff b(x) ≺ b(y);
• x ∼b y iff b(x)≺ = b(y)≺ ;
• x vb y iff b(x) v b(y);
• x v0b y iff ∀s ∈ S, s ≺b x⇒s ≺b y.

Proposition 87. Let b : S → C be a function. Keeping the notation in the Definition

above, we have:
(a) ≡b and ∼b are equivalence relations in S and, whenever b satisfies the condition
(b2), every ≡b -equivalence class is a subset of S.
(b) vb and v0b are pre-order relations in S and ∼b = (vb ) ∩ (vb )op .
(c) ≡b ⊆ ∼b and, whenever b satisfies the condition (b1), they coincide iff ≺ is an
extensional relation in C.
(d) v0b =vb , whenever b satisfies condition (b1).
(e) ≺b is an well-founded relation in S, whenever b satisfies the condition (b2).
(f ) ≺b and vb are compatible with the equivalence relations ≡b and ∼b (e.g.: if x ≺b y
and y ∼b y 0 , then x ≺b y 0 .)
(g) x ≡b y ⇔ −x ≡b −y, whenever b satisfies condition (b4).
(h) x ∼b y ⇔ −x ∼b −y, whenever b satisfies the condition (b4).
(i) x ≺b y ⇔ −x ≺b −y, whenever b satisfies the condition (b4).
(j) root(S, ≺b ) = b−1 [{Φ}], whenever b satisfies the condition (b1).
(k) ∗ ∈ root(S, ≺b ), whenever b satisfies the conditions (b1) and (b3).
3
Since, for each ordinal α and set x, if x ∈ α then x is an ordinal.
(l) If b satisfies condition (b1) then: b satisfies (b5) iff for each (A, B) ∈ Cs (S) and
each s ∈ S such that ∀z ∈ A ∪ B, z ≺b s, then t(A, B) vb s.
Proof. Items (a), (b) follows directly from the definitions. Items (d) and (l) are conse-
quences of the surjectivity of the function b (anchor condition (b1)). Since −(−x) = x,
in the items (g), (h), (i) it is enough to verify the implications (⇒): they are direct
consequence of the anchor condition (b4). It remains to prove items (c), (e), (j) and (k).
(c) Clearly ≡b ⊆ ∼b .
Suppose that ≺ is an extensional relation in C and let x, x0 ∈ S, then: b(x) = b(x0 ) iff
∀c ∈ C, (c ≺ b(x)⇔c ≺ b(x0 )). Thus (x, x0 ) ∈≡b iff (x, x0 ) ∈∼b .
Conversely, if ≡b =∼b , and x, x0 ∈ S, then: ∀x, x0 ∈ S, b(x) = b(x0 ) iff ∀c ∈ C, (c ≺
b(x)⇔c ≺ b(x0 )). Since b is surjective, we get: ∀a, a0 ∈ C, a = a0 iff ∀c ∈ C, (c ≺ a⇔c ≺
a0 ), i.e. ≺ is an extensional relation in C.
(e)SLet x ∈ S. By definition of ≺b , for each x ∈ S we have x≺b := {y ∈ S : y ≺b
x} = {b−1 [{c}] : c ≺ b(x)}. Since ≺ is an well founded relation in C, the subclass
{c ∈ C : c ≺ b(x)} is a subset of C. Therefore, by anchor condition (b2), the subclass
of S given by x≺b is a subset of S.
Now let X be a non-empty subset of S. Then b[X] is a non-empty subset of C, thus we
can select c ∈ b[X] a ≺-minimal member. Consider any z ∈ b−1 [{c}] ∩ X, then clearly
z is a ≺b -minimal member of X.
(j) root(S, ≺b ) = b−1 [{Φ}]:
(⊇) Let x ∈ S be such that b(x) = Φ. Then ∀z ∈ S, z ≺b x iff b(z) ≺ b(x) = Φ. Thus
z ⊀b x, ∀z ∈ S, i.e. x is a ≺b -root.
(⊆) Let x ∈ S be such that ∀z ∈ S, z ⊀b x, then ∀z ∈ S, b(z) ⊀ b(x). Since b is
surjective: ∀c ∈ C, c ⊀ b(x). Thus b(x) ∈ root(C, ≺) = {Φ}, i.e. x ∈ b−1 [{Φ}].
(k) Since b(∗) = Φ (anchor condition (b3)), we have ∗ ∈ b−1 [{Φ}] and the result
follows from item (j) above.

Remark 88. A SUR-algebra S = (S, <, ∗, −, t) is a higher-order structure such that

the underlying first-order structure (S, <) is acyclic and a η∞ -relational structure (see
Remark 34). An anchor b : S → C on S induces another binary relation ≺b on S that
is well-founded (thus it is acyclic, in particular) and that, under convenient additional
hypothesis, can encode the anchor itself (see 116 and 117 in Section 3 below). First-
order structures of the form (S, <, ≺) were considered by P. Ehrlich in [Ehr01] and
by J. Hamkins in [Ham13], with the development of some interesting model-theoretic
aspects. Thus the concept of anchor in a SUR-algebra is a sort of balance: at one
hand it generalizes the ”birthday function”, a seminal ingredient for the development
of the theory of surreal numbers, and at the other hand is related to interesting first-
order structures that have been considered by respectable specialists in foundations of
mathematics.
Before we provide new examples of anchored SUR-algebra, we will establish some

simple results on more special kinds of anchors, to indicate the strength of the constrains
imposed by the anchor axioms.
Proposition 89. Let S be a SUR algebra and C = (C, ≺, Φ) be a rooted extensional

well-founded class (see Subsection 1.1.2, Chapter 1). Let / be a (well-founded) relation
induced by some anchor b : S → C, i.e. / =≺b , then there is only one such anchor b.
Proof. Let b : S → C and b0 : S → C be anchors such that ≺b =≺b0 and denote / this
binary relation in S. By Proposition 87.(e), / is a well-founded relation in S.
We will prove that ∀y ∈ S, b(y) = b0 (y) by /-induction. Suppose that b(x) = b0 (x) for
all x ∈ y / . Since b and b0 are surjective, we have:
b(y)≺ = {c ∈ C : c ≺ b(y)} = {b(x) ∈ C : x / y} = {b0 (x) ∈ C : x / y} = {c0 ∈ C : c ≺
IH
b0 (y)} = b0 (y)≺ .
Since ≺ is extensional, we have b(y) = b0 (y) and thus b = b0 .
Proposition 90. Let b : S → C be an anchored SUR-algebra such that:

(T) For each x ∈ S, there exists (A, B) ∈ Cs (S) such that x = t(A, B) and ∀a ∈ A, ∀b ∈
B, b(a), b(b) ≺ b(x). 4
Then for each SUR-algebra S 0 , there is at most one SUR-algebra morphism f : S → S 0 .
In particular: SU R − alg(S, S) = {idS } and then the structure of anchored SUR-
algebra (S, b) is rigid, i.e., idS is a the unique SUR-automorphism f : S → S such that
b ◦ f = b.
Proof. The proof is a simple ≺b -induction: given SUR-algebras morphisms f, g : S →

S 0 , suppose that f (z) = g(z) for all z ∈ x≺b . Take any (A, B) that is b-timely Conway-
cut representation of x. Since A ∪ B⊆x≺b , we have:
f (x) = f (t(A, B)) = t0 (f [A], f [B]) = t0 (g[A], g[B]) = g(t(A, B)) = g(x).
IH
Thus ∀y ∈ S, f (y) = g(y).

Remark 91.
• If (b5)strong holds, then the condition (T) in the Proposition above is equivalent to
require that t : Cs (S) → S be a surjective function.
4
Following [All87], section 2, chapter 4, page 125, (see also Fact 27, Chapter 1) a pair (A, B) as above
will be called a b-timely Conway cut representation of x. Note that, in particular, t : Cs (S) → S must
be surjective.
• The anchor b : N o → On described in 85 satisfies the conditions in the Propositions

89 and 90 above:
- (On, ∈ ∅) is an extensional rooted well founded-class;
- it satisfies condition (T) above (but does not satisfy (b5)strong ): by the Cuesta-Dutari
representation of a surreal number y (see [All87], pages 125, 129) we have y = {Ly |Ry },
where Ly = {x ∈ N o : x < y, b(x) ≺ b(y)} and Ry = {z ∈ N o : y < z, b(z) ≺ b(y)}.
• Soon we will provide anchors on the SUR-algebra SA that satisfies the conditions
in the Propositions 89 and 90 above.
• For the ”timely” anchored SUR-algebras S, i.e., those satisfying the condition
(T), we may wonder about the existence of some (unique) morphism with source S and
target a SUR-algebra S 0 . By (T), we could be tempted to define recursively a morphism
f : S → S 0 by f (t(A, B)) := t0 (f [A], f [B]), since A ∪ B⊆x≺b . Clearly, by the SUR-
algebra properties of S 0 , such rule f must preserves ∗ and −. But the main issue is that
when such rule can really define a function. We will re-address this topic in the future
(for the anchored SUR-algebra N o, see [All87], pages 125, 126).

In the sequel, we will provide (strict) anchors structures on the SUR-algebras SA

and ST (see Subsections 2.2.2 and 2.2.3, Chapter 2).
Proposition 92. Connecting SA and On:

s
We have a pair of functions (On SA), defined in the Subsection 2.2.2 of Chapter 2:
r
• The function s : On → SA, defined by recursion on the well-ordered proper class
(On, <), by s(α) := hs[α], ∅i, α ∈ On.
• The function r : SA → On, given by r(x) = min{α ∈ On : x ∈ SAα }, x ∈ SA.
Then:
(i) r ◦ s = idOn .
(ii) r is a strict anchor of SA, that satisfies (b5)strong , on the rooted extensional well-
founded class (On, ∈, ∅).
Moreover, ∀x ∈ SA,Sif x = t(A, B) then: r(x) = min⊆ {α ∈ On : ∀a ∈ Lx , ∀b ∈
Rx , r(a), r(b) ∈ α} = z∈Lx ∪Rx (r(z))+ .
Proof. In the items below, all the ”Claims” and ”Facts” mentioned are in the Sub-
subsection 2.2.2, Chapter 2. Item (i) was established in Claim 13. We will prove item
(ii):
(b1) r is surjective: since it has a section s : On → SA (i.e. r ◦ s = idOn ).
(b2) r is locally small (or has small fibers): since r−1 [{α}] = SAα \ SA(α) ⊆SAα is a
set, for each α ∈ On.
(b3) r(∗) = r(h∅, ∅i) = 0 = ∅, by Fact.(a)
(b4) r(−x) = r(x), ∀x ∈ SA, by Claim 14.(i).

(b5)strong For each (A, B) ∈ Cs (SA) and each α ∈ On, then:
(S) ∀a ∈ A, ∀b ∈ B, r(a), r(b) ≺ α (in On) iff r(t(A, B)) v α.
In On, we have the relations ≺=∈=< and v= ⊆ =≤. Thus the expression (S) above
can be described as:
(S’) ∀a ∈ A, ∀b ∈ B, r(a), r(b) < α iff r(t(A, B)) ≤ α.
The validity of (S’) is exactly the content of Claim 6. Equivalently, r(t(A, B)) =
min≤ {α ∈ On : ∀a ∈ A, ∀b ∈ B, r(a), r(b) < α}.
Let (A, B) ∈ Cs (SA) and M umr (A, B) = {u ∈ SA : A < u < B, ∀z ∈ SA(A < z <
B ⇒ u vr z)}.
Since ≺ in On is a strict linear order, M umr (A, B) = M alr (A, B) = {u ∈ SA : A <
u < B, ∀z ∈ SA(A < z < B ⇒ z ⊀r u)}. Thus (b6) ⇔ (b6)weak , (b7) ⇔ (b7)strong (see
Remark 84 above).
(b6) t(A, B) ∈ M umr (A, B):
A < t(A, B) < B, by Claim 11;
A < z < B ⇒ r(t(A, B)) ≤ r(z), by Claim 12.
(b7) If s, s0 ∈ M umr (A, B) are <-comparable (i.e., s = s0 or s < s0 or s0 < s), then
s = s0 .
Since ≺=∈=< is an extensional relation in On, r[M umr (A, B)] = {r(t(A, B)} (Remark
84). Thus, for all s, s0 ∈ M umr (A, B), r(s) = r(s0 ) and by Claim 7, s ≮ s0 and s0 ≮ s.
We finish the proof showingS that, for each x ∈ SA, r(x) = min⊆ {α ∈ On : ∀a ∈
+
Lx , ∀b ∈ Rx , r(a), r(b) ∈ α} = z∈Lx ∪Rx (r(z)) :
The first equality follows from the proof above, since in On, ≤=S ⊆ and <=∈. Let X :=
min⊆ {α ∈ On : ∀a ∈ Lx , ∀b ∈ Rx , r(a), r(b) ∈ α} and X 0 := z∈Lx ∪Rx (r(z))+ ∈ On.
Consider any z ∈ Lx ∪ Rx , then r(z) ∈ r(z)+ ⊆X 0 , i.e. X⊆X 0 . On the other hand, any
z ∈ Lx ∪ Rx , r(z) ∈ X, then r(z)+ ⊆X, thus X 0 ⊆X. Summing up, X = X 0 .

By a sequence of reasonings analogous to the above, but founded over the ”Facts”
and ”Claims” presented in the Subsection 2.2.3 of Chapter 2, we obtain the following:
Fact 93. Connecting ST and On:

s0
We have a pair of functions (On ST ), defined in the Subsection 2.2.3 of Chapter 2:
r0
0
• The function s : On → ST , defined by recursion on the well-ordered proper class
(On, <), by s0 (α) := hs0 [α], ∅i, α ∈ On.
• The function r0 : ST → On, given by r0 (x) = min{α ∈ On : x ∈ STα }, x ∈ SA.
Then:
(i) r0 ◦ s0 = idOn .
(ii) r0 is a strict anchor of ST , that satisfies (b5)strong , on the rooted extensional well-
founded class (On, ∈, ∅).

Moreover, ∀x ∈ ST ,Sif x = t(A, B) then: r0 (x) = min⊆ {α ∈ On : ∀a ∈ A, ∀b ∈
B, r0 (a), r0 (b) ∈ α} = z∈A∪B (r0 (z))+ .

Now, we will connect the initial SUR-algebra SA and the initial ZF-algebra V: since
V is a standard ZF-algebra (Subsection 1.2.2, Chapter 1), (V, ∈, ∅) is an well-founded
extensional rooted class.
Definition 94. (i) Let j ∗ : V → SA, defined by ∈-recursion by j ∗ (X) = t(j ∗ [X], ∅),
X ∈V.
(ii) Consider ≺r the well-founded relation induced by the above defined strict anchor
function r : SA → On, of SA on the rooted extensional well-founded class (On, ∈, ∅).
Since t : Cs (SA) → SA is a bijection (is the identity function), and r : SA → On
is a strict anchor, we can define a function b∗ : SA → V by ≺r -recursion: b∗ (x) =
b∗ [Lx ] ∪ b∗ [Rx ], where x = hLx , Rx i = t(Lx , Rx ).

Proposition 95.
(i) b∗ ◦ j ∗ = idV , thus b∗ is surjective and j ∗ is injective.
(ii) Let X, Y ∈ V . Then: X ∈ Y iff j ∗ (X) < j ∗ (Y ).
(iii) Let Y ∈ V and a ∈ SA. If a < j ∗ (Y ) and b∗ (a) ∈ b∗ (j ∗ (Y )) then a = j ∗ (X) for a
unique X ∈ Y .
Proof. All the ”Claims” mentioned below are in the Subsection 2.2.2, Chapter 2.
(i) By ∈-induction: b∗ (j ∗ (Y )) = ∗ b∗ (t(j ∗ [Y ], ∅)) = ∗ b∗ [j ∗ [Y ]] ∪ b∗ [∅] = {b∗ (j ∗ (X)) :
def. j def. b
X ∈Y}∪∅ = Y.
IH
(ii) If X ∈ Y , then j ∗ (X) ∈ j ∗ [Y ] and, since j ∗ (Y ) = hj ∗ [Y ], ∅i, then j ∗ (X) ∈ Lj ∗ (Y ) .
Thus, by Claims 4 and 9, j ∗ (X) < j ∗ (Y ). Conversely, if j ∗ (X) < j ∗ (Y ), then since
Rj ∗ (X) = ∅, we must have j ∗ (X) ∈ Lj ∗ (Y ) = j ∗ [Y ]. This means that there is X 0 ∈ Y such
that j ∗ (X) = j ∗ (X 0 ). By, item (i) above, j ∗ is injective, then X = X 0 ∈ Y , as we would
like.
(iii) If a < j ∗ (Y ) , then by Claims 4 and 9, a ∈ Lj ∗ (Y ) or j ∗ (Y ) ∈ Ra . In the latter case,
b∗ (j ∗ (Y )) ∈ b∗ [Ra ]⊆b∗ (a), and since ∈ is an irreflexive relation in V , this is incompatible
with the second hypothesis, b∗ (a) ∈ b∗ (j ∗ (Y )). Thus a ∈ Lj ∗ (Y ) = j ∗ [Y ], and exists
X ∈ Y such that a = j ∗ (X). Since j ∗ is injective, such X is uniquely determined.

Let us register the following useful:
Fact 96.
(i) For each x ∈ SA, b∗ (x) = b∗ [Lx ] ∪ b∗ [Rx ] = min⊆ {Y ∈ V : ∀a ∈ Lx , ∀b ∈

Rx , b∗ (a), b∗ (b) ∈ Y } = z∈Lx ∪Rx {b∗ (z)}.
S
(ii) By Claims 4 and 9 in the Subsection 2.2.2, Chapter 2: ∀x, y ∈ SA, x < y iff
x ∈ Ly (thus b∗ (x) ≺ b∗ (y)) or y ∈ Rx (thus b∗ (y) ≺ b∗ (x)). In particular:
• ∀x, y ∈ SA, x < y ⇒ b∗ (x) 6= b∗ (y).
• ∀(A, B) ∈ Cs (SA), b∗ [A] ∩ b∗ [B] = ∅.

Proposition 97.
(i) j ∗ extends s, through i. More precisely:
s j∗ i
(SA ← On) = (SA ← V ← On).
(ii) b∗ lifts r, through ρ. More precisely:
r b∗ ρ
(SA → On) = (SA → V → On).
Proof. (i) By ∈-induction. Let α ∈ On, then j ∗ (i(α)) = hj ∗ [i(α)], ∅i = t({j ∗ (Y ) : Y ∈

V, Y ∈ i(α)}, ∅) = t({j ∗ (i(β)) : β ∈ On, β ∈ α}, ∅) = t({r(β) : β ∈ On, β ∈ α}, ∅) =
IH
hr[α], ∅i = r(α).
(ii) By ≺r -induction. Let x = t(A, B), then, by Sthe Fact 96 just above:
∗ ∗
= ρ(b [A]∪b [B]) = ρ( z∈A∪B {b∗ (z)}) =5 z∈A∪B ρ({b∗ (z)}) =6
∗ ∗
S
ρ(b
S (x)) = ρ(b (t(A, B)))
∗ + +
S
z∈A∪B (ρ(b (z))) = z∈A∪B (r(z)) = r(t(A, B)) = r(x).
IH

Now, we are ready to state and prove a result analogous to Proposition 92:
Proposition 98. Connecting SA and V :
j∗
We have a pair of functions (V SA), defined in 94 above. Then:
b∗
∗ ∗
(i) b ◦ j = idV .
(ii) b∗ is an weakly7 strict anchor of SA, that satisfies (b5)strong , on the rooted extensional
well-founded class (V, ∈, ∅).
Proof. First of all, note that in (V, ∈, ∅), ≺=∈ and v= ⊆. Item (i) was proved in the
Proposition 95.(i). Now we will prove item (ii).
• From item (i), b∗ is surjective, i.e. (b1) holds.
• Let x ∈ V and denote ρ(x) = α ∈ On then, by Proposition 97.(ii):
b [{x}]⊆b∗ −1 [Vα+1 \ Vα ] = b∗ −1 [ρ−1 [{α}] = SAα \ SA(α)
∗ −1
thus b∗ −1 [{x}] is a set and (b2) holds.

5
See the Subsection 1.2.1
6
Idem.
7
I.e., it satisfies (b7) and, instead (b6), it satisfies (b6)weak , see Remark 84.
• The verifications of conditions (b3) and (b4) are straightforward:

(b3) b∗ (∗) = b∗ (t(∅, ∅)) = b∗ [∅] ∪ b∗ [∅] = ∅.
(b4) Let x ∈ SA, we prove by ≺r -induction that b∗ (−x) = b∗ (x):
Since t is bijective, let x = t(A, B), then ∀z ∈ A ∪ B, b∗ (z) ≺ b∗ (x) and:
b∗ (−x) = b∗ (−t(A, B)) = b∗ (t(−B, −A)) = b∗ [−B]∪b∗ [−A] = b∗ [B]∪b∗ [A] = b∗ (t(A, B)) =
IH
b∗ (x).
• Note that for each (A, B) ∈ Cs (SA), b∗ (t(A, B)) = b∗ [A] ∪ b∗ [B] = min⊆ {X ∈ V :
∀a ∈ A, ∀b ∈ B, b∗ (a), b∗ (b) ∈ X} (see Fact 96). Equivalently, for each X ∈ V , then:
∀a ∈ A, ∀b ∈ B, b∗ (a), b∗ (b) ∈ X iff b∗ (t(A, B)) = b∗ [A] ∪ b∗ [B]⊆X. This means that b∗
verifies the condition (b5)strong .
• The condition (b7) is satisfied:
Let (A, B) ∈ Cs (SA). Since ≺=∈ is an extensional relation in V , b∗ [M umb∗ (A, B)] has
at most one member (Remark 84). Thus, for all s, s0 ∈ M umb∗ (A, B), b∗ (s) = b∗ (s0 ) and,
by Fact 96, s ≮ s0 and s0 ≮ s.
• Let (A, B) ∈ Cs (SA) and let z ∈ SA such that z ≺b∗ t(A, B). Then b∗ (z) ∈ b∗ [A] ∪
b [B]. If b∗ (z) = b∗ (a), for some a ∈ A, then by Fact 96, it can’t hold A < z. Likewise,
∗
if b∗ (z) = b∗ (b), for some b ∈ B, then it can’t hold z < B. Thus t(A, B) ∈ M alb∗ (A, B)
and (b6)weak holds.

Note this anchor (SA, b∗ ) satisfies the conditions (T) in the Proposition 90 (but the
conclusion is not new, since we saw in Chapter 2 that SA is the initial object in the
category SUR).
In the sequel, we present characterizations/axiomatizations of the anchored SUR-
algebras (SA, r), (ST, r0 ) and (SA, b∗ ) in the same vein as the axiomatization of the
SUR-algebra (N o, b) (see Subsection 1.3.2, Chapter 1).
r
Theorem 99. An axiomatization of SA → On:
The anchored SUR-algebra (SA, r) is axiomatized by the following properties:
(rSA1) t : Cs (SA) → SA is bijective.
(rSA2) ∀u ∈ SA, S if u = t(A, B) then: r(u) = min⊆ {α ∈ On : ∀a ∈ A, ∀b ∈
B, r(a), r(b) ∈ α} = v∈A∪B (r(v))+ .
(rSA3) ∀u, v ∈ SA, denote u = t(Lu , Ru ) and v = t(Lv , Rv ), then are equivalent:
• u < v;
• u ∈ Lv or8 v ∈ Ru .
More precisely, if (S, b), is an anchored SUR-algebra, b : S → On, then are equiva-
lent:
(i) (S, b) satisfies the conditions (rSA1), (rSA2), (rSA3) above.
∼
=
(ii) There exists a unique isomorphism of anchored SUR-algebras u : (SA, r) → (S, b).
8
By (rSA2), this is a non-inclusive disjunction.
In particular:
• Any anchored SUR-algebra (S, b) that satisfies (rSA1), (rSA2), (rSA3) must be strict
and satisfies (b5)strong , since (SA, r) is an strictly anchored SUR-algebra satisfying (b5)strong .
• If (S, b) and (S 0 , b0 ) are anchored SUR-algebras that satisfies (rSA1), (rSA2), (rSA3),
∼
=
then there is a unique isomorphism of anchored SUR-algebras f : (S, b) → (S 0 , b0 ).
Proof. (ii)⇒(i) is clear from the results on (SA, r): see Subsection 2.2.2 and Proposition
92. We will show that (i)⇒(ii).
W
Denote Z := (Z, , s) the standard ZF-algebra On.
Let (S, b) be anchored SUR-algebra that satisfies (rSA1), (rSA2), (rSA3). For each
z ∈ Z, let S z := b−1 [z ≺ ]. Then clearly:
• u ∈ S z iff −u ∈ S z ;
• S Φ = ∅;
• z 6= Φ iff ∗ ∈ S z ;
• tz : Cs (S z , < ) → S s(z) is a bijection, by (rSA1), (rSA2);
• If y ≺ z, equivalently yεz or y < z, see Remark 17.(d) (in particular: ≺ is a transitive
relation), then y ≺ ⊆z ≺ and S y ⊆S z .
Thus, since z < s(z), S z is a partial SUR-algebra whenever z 6= Φ, where Cst (S z , <
) := t−1 z −1 s(z)
z [S ]⊆tz [S ] = Cs (S z , <) and tz : Cst (S z , <) → S z is the partial structure
map.SMoreover, 0 < y ≤ z 7→ S y ⊆S z is a increasing family of partial SUR-algebras and
S = z∈Z S z (a directed colimit of partial SUR-algebras, see Subsection 2.3.4, Chapter
2).
Let (S, b) and (S 0 , b0 ) be anchored SUR-algebras that satisfies (rSA1), (rSA2), (rSA3).
We will show that there is a unique SUR-algebra morphism f : S → S 0 such that
b = b0 ◦ f . Then, by uniqueness, the SUR-algebra morphism f 0 : S 0 → S such that
b0 = b ◦ f 0 is the inverse morphism of f , completing the proof.
Claim: For all x ∈ Z there is a unique function fx : S x → S 0 x that is a partial SUR-
algebra morphism (whenever x 6= Φ) and such that b0 (fx (u)) = b(u), for all u ∈ S x .
Proof: Suppose, by induction, that for all y ≺ x there is a unique fy : S y → S 0y
satisfying the conditions above. Note that if there is a map fx : S x → S 0 xSsatisfying the
conditions, then fy = (fx ) , ∀y ≺ x, by the uniqueness of fy . Since x≺ = yεx s(y)≺ and
S x = b−1 [x≺ ] = yεx S s(y) we consider two cases:
S
≺ ≺ x 0x y
S S S
S • x0 yis a limit: then x = yεx y and (fx : S → S ) = ( yεx fy : yεx S →
yεx S ) (it is the unique possibility, and it works).
• x is a successor: then x = s(z), zεx. Note that yεx S y = yεx,y6=z S s(y) . Let
S S
u ∈ S s(z) \ yεx S y , then b(u) = z and u = t(Lu , Ru ), with z = min≤ {w ∈ Z :

S
∀v ∈ Lu ∪ Ru , b(v) < z}. If there is fSx satisfies the S conditions, we must have fx (u)0 =
0 0
fx (t(Lu , Ru )) = t (fx [Lu ], fx [Ru ]) = t ( yεx fy [Lu ], yεx fy [Ru ]), z = min≤ {w ∈ Z : ∀v ∈
fy [Lu ∪ Ru ], b0 (v 0 ) < z}. Thus there is at most one map fx : Sx → S 0 x satisfying

S
yεx
the conditions. Now define fx (u) := t0 (fz [Lu ], fz [Ru ]), u ∈ Sx . Then this map satisfies
all the conditions required. Indeed if u < v ∈ S s(z) , then by (rSA3) u ∈ Lv or (exclusive)
v ∈ Ru , thus at most one member between u, v (they are distinct, since < is irreflexive),
belongs to b−1 [{z}]. Since L0fx (u) = fz [Lu ] and Rf0 x (u) = fz [Ru ], we obtain fx (u) <0 fx (v).
The other conditions on fx are even easier to prove.
By the Claim above, the family of functionsS{fx : S xS→ S 0 x ; x ∈SZ}, is compatible
(i.e., y ≤ x⇒fy = (fx ) ), thus (f : S → S 0 ) = ( y∈Z fy : y∈Z S y → y∈Z S 0 y ): it is the
unique possibility, and it works (i.e., f : S → S 0 is a SUR-algebra morphism such that
b0 ◦ f = b).

r0
Theorem 100. An axiomatization of ST → On:
The anchored SUR-algebra (SA, r) is axiomatized by the following properties:
(rST1) t : Cs (SA) → SA is bijective.
(rST2) ∀u ∈ ST , ifSu = t(A, B) then: r0 (u) = min⊆ {α ∈ On : ∀a ∈ A, ∀b ∈
B, r0 (a), r0 (b) ∈ α} = v∈A∪B (r0 (v))+ .
(rST3) ∀u, v ∈ ST , if u = t(A, B) and v = t(C, D) and denote α := max{r0 (u), r0 (v)}
then are equivalent:
• u < v.
• Exists n ∈ N, exists {w0 , · · · , wn+1 } where u = w0 , v = wn+1 and such that:
r0 (wi ) < α, whenever 1 ≤ i ≤ n; wj ∈ Lwj+1 or wj+1 ∈ Rwj , whenever 0 ≤ j ≤ n.
More precisely, if (S, b), is an anchored SUR-algebra, b : S → On, then are equiva-
lent:
(i) (S, b) satisfies the conditions (rST1), (rST2), (rST3) above.
∼
=
(ii) There exists a unique isomorphism of anchored SUR-algebras u : (ST, r0 ) → (S, b).
In particular:
• Any anchored SUR-algebra (S, b) that satisfies (rST1), (rST2), (rST3) must satisfies
(b5)strong , since (ST, r0 ) is an anchored SUR-algebra satisfying (b5)strong .
• If (S, b) and (S 0 , b0 ) are anchored SUR-algebras that satisfies (rST1), (rST2), (rST3),
∼
=
Proof. (ii)⇒(i) is clear from the results on (ST, r0 ): see Subsection 2.2.3 and Fact
93. The proof of (i)⇒(ii) follows closely the proof of the Theorem above: it can be
established an analogous:
Claim: For all x ∈ Z = On there is a unique function fx : S x → S 0 x that is a partial
SUR-algebra morphism (whenever x 6= Φ) and such that b0 (fx (u)) = b(u), for all u ∈ S x .
It is proof is established by induction in x ∈ Z = On, by considering separately the cases
x limit and x successor: in the latter case, we must to adapt the corresponding part of
the proof of Theorem 99, using the condition (rST3) instead (rSA3), but everything
works.
By the Claim above, the family of functions {f x

x : S →S S 0 x ; x ∈ Z},Sis compatible (i.e.,
y ≤ x⇒fy = (fx ) ), thus (f : S → S 0 ) = ( y∈Z fy : y∈Z S y → y∈Z S 0 y ): it is the
S
unique possibility, and it works (i.e., f : S → S 0 is a SUR-algebra morphism such that
b0 ◦ f = b).

b∗
Theorem 101. An axiomatization of SA → V :
The anchored SUR-algebra (SA, b∗ ) is axiomatized by the following properties:
(bSA1) t : Cs (SA) → SA is bijective.
(bSA2) ∀u ∈ SA, ifSu = t(A, B) then: b∗ (u) = min⊆ {x ∈ V : ∀a ∈ A, ∀b ∈
B, b∗ (a), b∗ (b) ∈ x} = v∈A∪B {b∗ (v)}.
(bSA3) ∀u, v ∈ SA, denote u = t(Lu , Ru ) and v = t(Lv , Rv ), then are equivalent:
• u < v;
• u ∈ Lv or9 v ∈ Ru .
More precisely, if (S, b), is an anchored SUR-algebra, b : S → V , then are equivalent:
(i) (S, b) satisfies the conditions (bSA1), (bSA2), (bSA3) above.
∼
=
(ii) There exists a unique isomorphism of anchored SUR-algebras u : (SA, b∗ ) → (S, b).
In particular:
• Any anchored SUR-algebra (S, b) that satisfies (bSA1), (bSA2), (bSA3) must be weakly10
strict and satisfies (b5)strong , since (SA, b∗ ) is an strictly anchored SUR-algebra satisfying
(b5)strong .
• If (S, b) and (S 0 , b0 ) are anchored SUR-algebras that satisfies (bSA1), (bSA2), (bSA3),
∼
=
Proof. (ii)⇒(i) is clear from the results on (SA, b∗ ): see Subsection 2.2.2 and Proposi-
tion 98. The proof of (i)⇒(ii) follows closely the proof of the Theorem 99 above.
W
Denote Z := (Z, , s) the standard ZF-algebra V .
Let (S, b) be anchored SUR-algebra that satisfies (bSA1), (bSA2), (bSA3). For each
z ∈ Z, let S z := b−1 [z ≺ ]. Then clearly:
• u ∈ S z iff −u ∈ S z ;
• S Φ = ∅;
• If z is a transitive set, then: z 6= Φ iff ∅ ∈ z iff ∗ ∈ S z ;
• If z is a non-empty transitive set then (S z , < , − , ∗, t ) is a partial SUR-algebra, where
and Cst (S z ) = t−1 [S z ] ∩ Cs (S z , <);
t : Cst → S z S
• Since V = α∈On Vα and each Vα is a transitive set, each S Vα+ is a partial SUR-algebra
and tVα : Cs (S Vα , < ) → S Vα+ is a bijection, by (bSA1), (bSA2) (note that b ◦ ρ provides
an anchor on S that satisfies the condition on Theorem 99);
• The map 0 < α ≤ β ∈ On 7→ S Vα ⊆S Vβ is a increasing family of partial SUR-algebras
9
By (bSA2), this is a non-inclusive disjunction.
10
I.e., it satisfies (b7) and, instead (b6), it satisfies (b6)weak , see Remark 84.
and S = α∈On S Vα+ (a directed colimit of partial SUR-algebras, see Subsection 2.3.4,
S
Chapter 2).
It can be established the following:
Claim: For all α ∈ On there is a unique function fα : S Vα → S 0 Vα that is a partial
SUR-algebra morphism (whenever α 6= Φ) and such that b0 (fα (u)) = b(u), for all
u ∈ S Vα .
It is proof is established by induction in α ∈ On, by considering separately the cases α
limit and α successor: in the latter case, we must to adapt the corresponding part of
the proof of Theorem 99, using the condition (bSA3) that coincides with (rSA3), but
everything works.
Vα 0 Vα
By the Claim above, the family of functions {fS
α : S →S ; α ∈ On},
S is compatible
(i.e., γ ≤ α⇒fγ = (fα ) ), thus (f : S → S ) = ( α∈On fα : α∈Z S → α∈Z S 0 α ): it is
0 α
S
the unique possibility, and it works (i.e., f : S → S 0 is a SUR-algebra morphism such
that b0 ◦ f = b).

We finish this Section with some general commentaries:

Proposition 102. Let (S, b, C) be an anchored SUR-algebra satisfying (b5)strong and
such that C is a rooted well-founded class underlying to a standard ZF-algebra. Then for
each (A, B) ∈ Cs (S), b(t(A, B)) = minv {c ∈ C : ∀a ∈ A, ∀b ∈ B, b(a), b(b) ≺ c} =
W
z∈A∪B s(b(z)).
Proof. Recall from Subsection 1.2.2, Chapter 1, that in a standard ZF-algebra C it

holds:
• d ≺ c iff s(d) ≤ c;
• d v c iff d ≤ c.
strong
0
W b(t(A, B)) = minv {c ∈ C : ∀a ∈ A, ∀b ∈ B, b(a), b(b) ≺ c} (by (b5)
Let x := )
and x := z∈A∪B s(b(z)).
Consider any z ∈ A ∪ B, then b(z) ≺ s(b(z)) ≤ x0 , thus b(z) ≺ x0 , i.e. x ≤ x0 . On the
other hand, for any z ∈ A ∪ B, b(z) ≺ x, then s(b(z)) ≤ x, thus x0 ≤ x. Summing up,
x = x0 .
Remark 103.
(i) There are many possible notions of morphism between anchored SUR-algebras over
different rooted well-founded classes, (S, b, C) → (S 0 , b0 , C 0 ), that determine categories.
The simplest one is just given just by ”commutative squares”, i.e. pairs (f, v) where
f : S → S 0 is a SUR-algebra morphism, v : C → C 0 is a morphism of rooted well founded
classes (i.e., it preserves the root and the binary relation) and v ◦ b = b0 ◦ f . Note that
if x ≺b y in S, then f (x) ≺b0 f (y) in S 0 . With the obvious notions of identity and
composition, these data provides a category: ancSU R−alg.
Some examples of arrows: (can, idOn ) : (SA, r, On) → (ST, r0 , On) and (idSA , ρ) :
(SA, b, V ) → (SA, r, On).
(ii) If f : S → S 0 is a SUR-algebra morphism with small fibers and b0 : S 0 → C 0 is an
anchor in S 0 , then consider C the substructure of (C 0 , ≺0 , Φ0 ) where C = b0 [f [S]]. Note
that b0 (f (∗)) = b0 (∗0 ) = Φ0 . Suppose that the inclusion i : C ,→ C 0 reflects minimal
elements of every non-empty subclass K⊆C. Then C is a rooted well-founded class,
i : C ,→ C 0 is a morphism of rooted well-founded class, b := b0 ◦ f : S → C is an anchor
in S and (f, i) : (S, b, C) → (S 0 , b0 , C 0 ) is a morphism of anchored SUR-algebras.
As an example consider: (can, idOn ) : (SA, r, On) → (ST, r0 , On).
(iii) If v : C → C 0 is a surjective map with small fibers such that v(Φ) = Φ0 and c ≺ d
in C⇔v(c) ≺0 v(d) in C 0 and b : S → C is an anchor in S, then b0 := v ◦ b : S → C 0 is an
anchor in S and (idS , v) : (S, b, C) → (S, b0 , C 0 ) is a morphism of anchored SUR-algebras.
(iv) Another notion of morphism –that is related to item (iii) above– is given by pairs
(f, v) where:
• f : S → S 0 is a SUR-algebra morphism;
• v : C → C 0 is such that: v(Φ) = Φ0 , c ≺ d in C⇔v(c) ≺0 v(d) in C 0 and v is surjective;
• ∀s ∈ S, b0 (f (s)) v0 v(b(s)) (respect. and/or v(b(s)) v0 b0 (f (s))).
As an example consider: (ρ∗ , idOn ) : (SA, r, On) → (N o, b, On), see Proposition
105.(i) in the next Section.
(v) Another interesting possibilities to define morphisms (f, v) is when:
• we just consider morphisms v : C → C 0 between rooted well founded classes that are
underlying to standard ZF-algebras: example, i : On ,→ V ;
or
• we just consider morphisms v : C → C 0 of rooted well-founded classes that are under-
lying to ZF-algebra morphisms of standard ZF-algebras: example ρ : V → On.
Concerning the second possibility: the anchor b∗ : SA → V is the ”best anchor on SA
that satisfies (b5)strong ”, since there is a unique ZF-algebra morphism u : V → C, where
C is a standard ZF-algebra and given b : SA → C an anchor that satisfies (b5)strong
then, by Proposition 102, u ◦ b∗ = b.

3.2 The main diagram
In this short Section we register the relations between, in one hand the SUR-algebras
SA, N o and at the other hand, the ZF-algebras V, On: these will be useful in the next
Chapter. The following diagram of classes and functions summarizes these relations:
3.2. THE MAIN DIAGRAM 77
j∗
V
-
SA
∗
6 b
ρ i ρ∗
? b ?

On - No
j
Where:
• The arrow ρ : V → On is the universal ZF-algebra morphism: the rank function
(see Proposition 12).
• The arrow i : On ,→ V is the inclusion function.
• The arrow b : N o → On was described in the Subsection 1.3.2, Chapter 1.
• The arrow j : On → N o was described in the Subsection 1.3.3, Chapter 1.
• ρ∗ : SA → N o is universal SUR-algebra arrow (see Section 2.4, Chapter 2).
• s : On → SA was defined by ∈-recursion in 92.
• r : SA → V was defined in 92.
• j ∗ : V → SA was defined by ∈-recursion in 94.(i).
• b∗ : SA → V was defined by ≺r -recursion in 94.(ii).
Remark 104. Concerning the maps in the diagram above:

• The top → down arrows ρ : V → On and ρ∗ : SA → N o are ”rank” arrows in the
category of ZF-algebras and of SUR-algebras.
j j∗
• The pairs of horizontal arrows On N o and V SA are pairs of heteromorphisms
b b∗
–see [Ell07]– i.e. they are pairs of ”chimera”-arrows (they have head in a category and tail
in another category): the right → left component of the pairs are anchors structures from
the SUR-algebras N o, SA to (standard) ZF-algebras On, V ; the left → right component
of the pairs are ”inclusions” of the ZF-algebras On, V into the SUR-algebras N o, SA.
• Some extra object, ST and extra arrows (between ST and SA, N o, On) can be
added to this diagram, but we will not make use of these extra information.

Proposition 105. We have the following relations between the maps in the diagram
above:
(a) ρ ◦ i = idOn
(b) b ◦ j = idN o
(c) r ◦ s = idOn
(d) b∗ ◦ j ∗ = idV
(e) s = j ∗ ◦ i
(f ) r = ρ ◦ b∗
(g) j = ρ∗ ◦ s
(h) ρ∗ ◦ j ∗ = j ◦ ρ
(i) b ◦ ρ∗ ≤ r
Proof. Item (a) is well known and was already mentioned in the Remark 14.(ii), Chapter
1. Item (b) was established in the Subsection 1.3.3, Chapter 1. Item (c) was established
in the Proposition 92.(i). Item (d) was established in the Proposition 95.(i). Items (e)
and (f) were established in the Proposition 97.
(g) Let us recall the definitions: j : On → N o is defined by ∈-recursion on ordinal
numbers α ∈ On, j(α) = {j[α]|∅} = t0 (j[α], ∅) and j ∗ : V → SA is defined by
notation
∈-recursion on x ∈ V , j ∗ (x) = hj ∗ [x], ∅i = t(j[x], ∅).
notation
By items (e) and (f), we must to establish the equality j = ρ∗ ◦ j ∗ ◦ i. This follows
from ∈-induction on ordinal numbers: let α ∈ On, then j ∗ (i(α)) = h{j ∗ (x) : x ∈ V, x ∈
i(α)}, ∅i = h{j ∗ (i(β)) : β ∈ On, β ∈ α}, ∅i = t({j ∗ (i(β)) : β ∈ On, β ∈ α}, ∅), since
every member of an ordinal number is an ordinal and i preserves and reflects the binary
relations = and ∈. Since ρ∗ is a SUR-algebra morphism:
ρ∗ (j ∗ (i(α))) = ρ∗ (t({j ∗ (i(β)) : β ∈ On, β ∈ α}, ∅)) = t0 (ρ∗ [j ∗ [i[{β ∈ On : β ∈ α}]]], ∅) =
IH
t0 (j[{β ∈ On : β ∈ α}], ∅) = t0 (j[α], ∅) = j(α).
(h) We will establish the equality ρ∗ ◦ j ∗ (x) = j ◦ ρ(x) by ∈-induction on x ∈ V .
Unraveling both sides of the equation we obtain:
ρ∗ ◦j ∗ (x) = ρ∗ (t(j ∗ [x], ∅)) = t0 (ρ∗ [j ∗ [x]], ρ∗ [∅]) = t0 (ρ∗ [j ∗ [x]], ∅) = t0 (j[ρ[x]], ∅) = t0 ({j(ρ(y)) :
IH
y ∈ x}, ∅).
j ◦ ρ(x) = t0 (j[ρ(x)], ∅) = t0 (j[ y∈x ρ(y)+ ], ∅) = t0 ({j(z) : ∃y ∈ x, z ∈ ρ(y)+ }, ∅).
S
Note that {j(ρ(y)) : y ∈ x}⊆{j(z) : ∃y ∈ x, z ∈ ρ(y)+ }. On the other hand, if z ∈ ρ(y)+ ,
then j(z) = j(ρ(y)) or j(z) ∈ j[ρ(y)] (and j(z) < t0 (j[ρ(y)], ∅) = j(ρ(y))). Summing up,
the sets of surreal numbers {j(ρ(y)) : y ∈ x} and {j(z) : ∃y ∈ x, z ∈ ρ(y)+ } are mutually
cofinal, and by Fact 27.(a), t0 ({j(ρ(y)) : y ∈ x}, ∅) = t0 ({j(z) : ∃y ∈ x, z ∈ ρ(y)+ }, ∅),
completing the proof.
S (i) We will prove by ≺r induction on x ∈ SA. Write x = t(Lx , Rx ). Then r(x) =

{r(z)+ : z ∈ Lx ∪ Rx }.
• If {r(z)+ : z ∈ Lx ∪ Rx } does not has maximum. Then ∀w ∈ Lx ∪ Rx , r(x) > r( w)+ >
r(w) ≥ b ◦ ρ∗ (w), by the induction hypothesis. Thus b ◦ ρ∗ (x) = b ◦ ρ∗ (t(Lx , Rx )) =
b(t(ρ∗ [Lx ], ρ∗ [Rx ])) ≤ r(x), by the property (b5).
• If {r(z)+ : z ∈ Lx ∪ Rx } has maximum r(z)+ . Then ∀w ∈ Lx ∪ Rx , r(x) = r(z)+ ≥
3.2. THE MAIN DIAGRAM 79
r( w)+ > r(w) ≥ b◦ρ∗ (w), by the induction hypothesis. Thus b◦ρ∗ (x) = b◦ρ∗ (t(Lx , Rx )) =
b(t(ρ∗ [Lx ], ρ∗ [Rx ])) ≤ r(x) by property (b5).

Remark 106.
1. The free SUR-algebra SA is an expansion of the free ZF-algebra V , i.e. the class of
all sets, via the map j ∗ : V → SA, in an analogous fashion that the SUR-algebra
N o is an expansion of the ZF-algebra On.
2. The SUR-algebra SA is ranked on the linear SUR-algebra N o of all surreal numbers

in a analogous fashion that the ZF-algebra V is ranked on the well-ordered ZF-
algebra On of all ordinal numbers, expanding the traditional set-theoretical link
ρ ρ∗
V → On to this new setting SA → N o.
j j∗
3. We saw above that the pairs of horizontal arrows On N o and V SA are pairs
b b∗
of heteromorphisms, connecting ZF-algebras and SUR-algebras.
4. As we already have mentioned in Remark 14.(ii), the inclusion map i : On ,→ V

is a section of the (unique) ZF-algebra morphism ρ : V → On (i.e. ρ ◦ i = idOn ),
that preserves arbitrary suprema (= reunions), preserves and reflects the binary
relations = and ε (= ∈) and such that ∀x ∈ V, ∀β ∈ On, x ∈ i(β) iff x = i(α) for
some and unique α ∈ On. But i : On ,→ V does not preserves successors. In fact,
as a consequence of the universal property of V , there is no ZF -algebra morphisms
h : On → V .
5. Recall that, by Corollary 82, Chapter 2 there is no SUR-algebra morphism h :

N o → SA, but we may wonder if there is an arrow i∗ : N o → SA that is an
”elementary” section of the (unique) SUR-algebra morphism ρ∗ : SA → N o, i.e.,
ρ∗ ◦ i∗ = idN o and i∗ preserving and reflecting every structure in sight (=, <, ∗, −)
except the t-maps. In fact, is not even clear if ρ∗ : SA → N o is surjective or not.
The way that SA (respect. ST ) lie above N o, seems to be a (non-trivial) model-
theoretic question that we intend to address in a future work, it seems related to
some parts of the work [Ham13].
j∗
V
-
SA
∗
b
ρ ρ∗
? b ?

On - No
j

3.3 The Hereditary Positive members of an equipped

SUR-algebra
Given an anchored SUR-algebra (S, b, C) is natural to consider subclasses P ⊆S such

that the restrictions of the acyclic relations < and ≺b coincide11 . Motivated by the well-
known natural ”inclusion” the class On of all ordinal numbers into the class N o of all
surreal numbers and by our proofs that the (surjective) anchor mappings r : SA → On,
b∗ : SA → V and r : ST → On have, in fact, ”well-behaved” section mappings, we start
in this Section the development of a generalization: ”The Hereditary Positive members
of an equipped SUR-algebra”.
Two questions arise immediately:
• On the terminology: what is the meaning of being ”positive” in a SUR-algebra? what
are the ”hereditary positives” in a SUR-algebra?
• Why (and how) the ”hereditary positives” of (S, b, C) are related to our original mo-
tivation: subclasses of P ⊆S such that <P = (≺b )P ?
Obviously, in the strict totally ordered abelian group N o there is a natural notion of
positive members. On other hand, since the relation < is not total in SA (respect. ST )
and not every member of SA (respect. ST ) are <-related12 to ∗ = 0 = h∅, ∅i, the notion
of ”being positive” requires a fresh conception: u ∈ SA (respect. u ∈ ST ) is positive
–notation: u ∈ P os(SA)– when there is a simple way to ensure that u < 0 do not occur:
we just set Ru = ∅.
Concerning SA, we register the following results (that can be easily derived from the
definitions and Proposition 97):
11
In particular, <P = (≺b )P is an well-founded relation in P ⊆S, whenever the inclusion P ,→ S
reflects minimal elements.
12
Note that hA, Bi is <-related with 0 iff −hA, Bi is <-related with 0.
3.3. THE HEREDITARY POSITIVE MEMBERS OF AN EQUIPPED SUR-ALGEBRA81
Fact 107. Let u, v ∈ SA. Then:

(a) If u ≺b∗ v, then u ≺r v.
(b) If u < v ∈ SA, then u ≺b∗ v ⇔ u ≺r v.
(c) If Ru = Rv = ∅, then u < v =⇒ (u ≺b∗ v ⇔ u ≺r v).

Having some idea of what should be ”the positives” in the main SUR-algebras of this
work, we turn our attention to the ”hereditary positives”.
108. Concerning N o:
• Every ordinal j(α) in N o is positive, i.e. j(α) ≥ 0, and is determined by the previous
ordinals in N o (j(α) = {{j(β) : β ∈ α}|∅}.
• Since b ◦ j = idN o and β < α ∈ On iff j(β) < j(α) ∈ N o, then <j[On] = (≺b )j[On] .

Note that in SA, since b∗ ◦ j ∗ = idV and y ∈ x ∈ V iff j ∗ (y) < j ∗ (x) ∈ SA, then
<j ∗ [V ] = (≺b∗ )j ∗ [V ] . The following result, concerning SA, is analogous to 108 above:
Proposition 109. For each u ∈ SA, are equivalent:
(a) u ∈ j ∗ [V ]⊆SA.
(b) For each finite sequence z0 , · · · , zn ∈ SA such that z0 = u, zi+1 < zi , zi+1 ≺b∗ zi ,
i < n, then zj ∈ P os(SA), j ≤ n.
(c) For each finite sequence z0 , · · · , zn ∈ SA such that z0 = u, zi+1 < zi , zi+1 ≺r zi ,
i < n, then zj ∈ P os(SA), j ≤ n.
Proof. (a)⇔(b) follows by ≺b∗ -induction and (b)⇔(c) follows from Fact 107 above.

Having experimented in particular cases the concepts of ”positive” (Pos) and ”hered-
itary positive” (HP), we pass now to a more general situation:
Definition 110.
(a) Let S be a SUR-algebra, define P os(S) := {u ∈ S : u = t(A, ∅), for some subset
A⊆S}.
(b) Let (S, b, C) be anchored SUR-algebra, define HP (S, b, C) as the range of the function
HP : C → Ps (S) defined by ≺-recursion:
S
HP (c) := {u ∈ S : b(u) = c, u = t( d∈c≺ HP (d), ∅)}, c ∈ C.

Note that, for each subset A⊆N o: (i) if there is a ∈ A such that a ≥ 0, then {A|∅} >
0; (ii) otherwise, a < 0 for all a ∈ A and {A|∅} = 0. Thus P os(N o)⊆{u ∈ N o : u ≥ 0}.
111. Let (S, b, C) be anchored SUR-algebra.

(a) By induction, HP (c)⊆P os(S) and it is a singleton whenever it is non-empty, and
its unique member belongs to b−1 [{c}], in particular HP (c) is a subset of S and the
codomain of the function HP is adequate.
(b) If d ≺ c ∈ C, then ∀v ∈ HP (c)∀w ∈ HPS(d) we have w < v. Indeed: if v ∈ HP (c)
and w ∈ HP (d), since d ≺ c, we have w < t( d∈c≺ HP (d), ∅) = v.
(c) Suppose that, for each c ∈ C, HP (c) is a singleton, and denote p(c) the unique
member of HP (c). Then we obtain a function pS : C → S such that b ◦ p = idC .
(d) Note that ∗ = t(∅, ∅) ∈ {u ∈ S : b(u) = Φ, t( d∈Φ≺ HP (d), ∅)}, thus p(Φ) = ∗ ∈ S.

Proposition 112. Let (S, b, C) be anchored SUR-algebra.

(a) Suppose that, HP (c) = {p(c)}, for each c ∈ C. Let c, d ∈ C and consider:
(p1) p(d) ≺b p(c) in S
(p2) d ≺ c in C
(p3) p(d) < p(c) in S
Then (p1)⇔(p2)⇒(p3)
(b) Suppose that (S, b, C) be an anchored SUR-algebra satisfying the additional con-
dition below:
(L) (S, b, C) be an anchored SUR-algebra such that C = (C, ≺, Φ) is a rooted well-
founded class where ≺ is a strict linear order.
Then the items (p1), (p2), (p3) are equivalent.
(c) Suppose that (S, b, C) be an anchored SUR-algebra satisfying the additional con-
dition below:
(S5) (S, b, C) be an anchored SUR-algebra satisfying (b5)strong and such that C = (C, ≺
, Φ) is a rooted well-founded class underlying to a standard ZF-algebra.
Then HP (c) = {p(c)}, for each c ∈ C.
Proof.
(a)
(p1)⇔(p2): By definition of HP (d), b(p(d)) = d, ∀d ∈ C. Thus, the equivalence follows
from the definition of ≺b . S S
(p2)⇒(p3): If d ≺ c, then p(d) ∈ d∈c≺ thus p(d) < t( d∈c≺ HP (d), ∅) = p(c).
(b) Note that when (C, ≺) is a strictly linearly ordered class then (p3)⇒(p2): if c ⊀ d,
then c = d or d ≺ c thus p(c) = p(d) or p(d) < p(c) (because (p2)⇒(p3)) and, in any
case, p(c) ≮ p(d), since < is an acyclic relation in S.
(c) Let (S, b, C) be an anchored SUR-algebra satisfying (b5)strong and such that C is
a rooted well-founded class underlying to a standard ZF-algebra.
We will need the following characterization (see PropositionW102): let (A, B) ∈ Cs (S),
b(t(A, B)) = minv {c ∈ C : ∀a ∈ A, ∀b ∈ B, b(a), b(b) ≺ c} = z∈A∪B s(b(z)).
S
Now we will verify that c = b(t( d∈c≺b HP (d), ∅)), ∀c ∈ C. Let c ∈ C and as-
S that HP (d) = {p(d)}, for all d ≺ c. ThenW d ≺ c ⇒ d =
sume, by induction,
b(p(d)) and b(t( d∈c≺ HP (d), ∅)) = b(t({p(d) : d ≺ c}, ∅)) = z∈p[c≺ ] s(b(z)) =
W W IH Claim
d∈c≺ s(b(p(d))) =
IH d∈c≺ s(d) = c, where the last equality follows since C is obtained
from a standard ZF-algebra.

Alternatively, we can:
• Define first a map j : C → S, by ≺-recursion: j(c) = t(j[c≺ ], ∅), c ∈ C;
• After that, consider the subclass of S given by range(j) = j[C]⊆S.
Both approaches, i.e. beginning from HP : C → Ps (S) or from j : C → S, are
related:
Proposition 113. The following items are equivalent:
(a) b ◦ j = idC ;
(b) HP (c) 6= ∅, ∀c ∈ C;
(c) HP (c) = {p(c)}, ∀c ∈ C;
(d) j(c) = p(c), ∀c ∈ C.
Proof. We already have observed that (b)⇔(c).

(d)⇒(c): Since HP (c) has at most one member. S
(c)⇒(d): Since {p(c)} = HP (c) = {u ∈ S : b(u) = c, u = t( d∈c≺ HP (d), ∅)} =
{t({p(d) : d ∈ c≺ }, ∅)} ∩ b−1 [{c}], we have p(c) = t({p(d) : d ∈ c≺ }, ∅) ∈ b−1 [{c}]. Then,
by ≺-induction, we obtain j(c) = p(c), ∀c ∈ C.
(d)⇒(a): As above, j(c) = p(c) ∈ b−1 [{c}], thus b(j(c)) = c, ∀c ∈ C.
(a)⇒(d): If b(j(c)) = c, then j(c) ∈ b−1 [{c}], ∀c ∈ C. By induction, suppose that
j(d) = p(d), ∀d S ∈ c≺ ⊆C. Then HP (d) = {j(d)}, ∀d ∈ c≺ ⊆C and HP (c) = {u ∈ S :
b(u) = c, u = t( d∈c≺ HP (d), ∅)} = {t({p(d) : d ∈ c≺ }, ∅)} ∩ b−1 [{c}] = {t({j(d) : d ∈
IH
c≺ }, ∅)} ∩ b−1 [{c}] = {j(c)}, since j(c) ∈ b−1 [{c}]. Thus p(c) = j(c) and the result
follows by induction.

Combining the results above we obtain the:
Corollary 114. For anchored SUR-algebras (S, b, C) such that b ◦ j = idC and (j(d) <
j(c) in S ⇒ d ≺ c in C):
(a) The items below are equivalent:
• j(d) ≺b j(c) in S
• d ≺ c in C
• j(d) < j(c) in S
(b) For each x ∈ S are equivalent:

• x is hereditary positive (i.e. x ∈ HP ((S, b, C)))
• x ∈ range(j)
(c) For the subclass HP = HP ((S, b, C))⊆S, we have (C, ≺, Φ) ∼
= (HP, (≺b ) , ∗) =
j
(HP, < , ∗).

Remark 115. The special cases: Note that the anchored SUR-algebras (N o, b, On),
(ST, r, On), (SA, r, On), (SA, b∗ , V ) satisfy both the additional conditions in the Corol-
lary above:
• (N o, b, On) satisfies condition (L) in Proposition 112.(b), thus (j(d) < j(c) in S ⇒
d ≺ c in C). Moreover, b ◦ j = IdOn (see Section 3, Chapter 1).
• (SA, b∗ , V ) satisfies condition (S5) in Proposition 112.(c), b∗ ◦ j ∗ = IdV . Moreover,
(j ∗ (d) < j ∗ (c) in SA ⇒ d ≺ c in V ).
• (ST, r, On) and (SA, r, On) satisfy both the conditions (L) and (S5) in Proposition
112.(b),(c), thus they satisfy both the additional conditions in the Corollary above.
This ensures that our main anchored SUR-algebras have nice notions of hereditary
positive subclass: see items (a), (b), (c) in the Corollary above.

116. The equipped SUR-algebra induced by a anchor:

Recall that, from Definition 86, each anchor b : S → C) induces five binary relations
in S: ≺b , ≡b , ∼b , vb , v0b . According Proposition 87, these relations satisfy many inter-
esting properties. The arrangement is particularly simple whenever ≺ is an extensional
relation.
We can consider, instead of the anchor b on S, equip S with a derived pair of binary
relations (S, ≺b , ≡b ). Note that the anchor b : S → C can be essentially recovered from
these data, since S/ ≡b ∼ = C. When ≺ is extensional, then the equivalence relation ≡b
is determined by the well-founded relation ≺b and we can consider just the structure
(S, ≺b ).
It can be defined by ≺b -recursionSa function HP ∗ : S → Ps (S):
HP ∗ (u) := {u0 ∈ S : u0 ≡b u, u0 = t( w∈u≺b HP ∗ (w), ∅)}.
Note that:
• u0 ≡b u1 ⇒ u≺ ≺b ∗ ∗
0 = u1 and HP (u0 ) = HP (u1 );
b
∗
• HP (u) = HP (b(u)), ∀u ∈ S (by ≺b -induction);
• HP ∗ (u) has at most one member.
In the process of associated a subclass of ”hereditary positive members” for some
anchored SUR-algebras (S, b) we consider naturally a structure given by class endowed
by two binary and acyclic relations (S, <, ≺b ). This seems related to the notions of
s-hierarchical structures considered by P. Ehrlich in [Ehr01].

To complete the ”analysis synthesis” process, we consider:
117. The anchor induced by an abstract equipped SUR-algebra:

We can consider equipped SUR-algebras (S, , ≈) where ≈ is an equivalence relation
on S and is an well-founded relation on S satisfying some additional conditions (see
Proposition 87):
(1) is compatible with ≈: i.e. u ≈ u0 , v ≈ v 0 , u v⇒u0 v 0 ;
(2) Every ≈-equivalence class is a subset of S (in particular, there exists the quotient
(proper) class S/ ≈:= {[u]≈ : u ∈ S};
(3) root(S, ) = [∗] = {u ∈ S : u ≈ ∗};
(4) −u ≈ u, ∀u ∈ S. (5) For each (A, B) ∈ Cs (S)⊆S and each s ∈ S such that
∀z ∈ A ∪ B, z s, then t(A, B) v s, where v is the pre-order relation on S derived from
.
define the notion by -recursion: HP ∗ : S → Ps (S), HP ∗ (u) = {u0 ∈
As above, we S
S : u ≈ u, u = t( w∈u HP ∗ (w), ∅)}.
0 0
Note that: u0 ≈ u1 ⇒u ∗ ∗

0 = u1 and HP (u0 ) = HP (u1 )
Set C := S/ ≈. In C define, Φ := [∗] and [u] ≺ [v] iff u v

Claim 1: (C, ≺, Φ) is a rooted well founded class.
Indeed: S
[v]≺ = u∈v [u] is a subset of C and, if A⊆C is a non-empty subclass, then Ǎ :=
[u]∈A [u]⊆S is a non-empty class. Selecting any v ∈ Ǎ a -minimal member of Ǎ, then
S
[v] ∈ A is a ≺-minimal member of A. By the condition (3) above, Φ is the only root of
(C, ≺).
Claim 2: q : S → C, u 7→ [u] is an anchor in S.
Indeed: (b1) obviously holds and (b2)–(b5) are obtained from (2)–(5) above.

118. Abstract Hereditary Positives:

We saw in Corollary 114.(c), that for ”well-behaved” anchored SUR-algebras (S, b, C),
the subclass HP = HP ((S, b, C))⊆S, provides a perfect encoding of C, since (C, ≺, Φ) ∼
=
(HP, (≺b ) , ∗) = (HP, < , ∗).
Alternatively, from 116 and 117 above, for an ”well-behaved” equipped SUR-algebra
(S, , ≈), we define the notion of abstract hereditary positive subclass by -recursion:
HP : S → Ps (S), HP (u) = {u ∈ S : u ≈ u, u = t( w∈u HP ∗ (w), ∅)}. Note that:
∗ ∗ 0 0 0
S
u0 ≈ u1 ⇒u ∗ ∗
0 = u1 and HP (u0 ) = HP (u1 ).
Distinct well-behaved structures (, ≈) on S induces distinct subclasses HP ∗ (S, , ≈

)⊆S, however (HP ∗ , , ∗) = (HP ∗ , < , ∗), and the possible difference occurs only on
the particular subclass, since they share the same root ∗ and the same binary relation
¡. The different maps HP ∗ : S → Ps (S) are just different retractions of the distinct
inclusions HP ∗ (S, , ≈) ,→ S.
We can also considerSjust a (nice) well-founded relation on S and:
• define HP 0 (S, ) = {K ∈ Ps (S) : iK : K ,→ S reflects -minimal members and
< ∩(K × K) = ∩ (K × K)};
• denote HP 0 := HP 0 (S, ).
Note that ∗ ∈ HP 0 , i : HP 0 ,→ S reflects -minimal members and < ∩(HP 0 × HP 0 ) =
∩ (HP 0 × HP 0 ) is a rooted well-founded class (proper or improper).
It is natural ask if there is a maximal (or even a largest) subclass H⊆S, such that
< ∩(H × H) coincides with ∩ (H × H), for some well-behaved structure (, ≈) on S.
Such kind of extremal question is related to Remark 103.(v).

3.4 Spaces of Signs

In the previous Section we have introduced the (recursively-defined) subclass of hered-
itary positive members of an anchored SUR-algebra b : S → C, HP (S, b), that under
convenient hypothesis encodes C (e.g. HP (SA, b∗ ) ∼ = V, HP (N o, b) ∼
= On). Combining
the work developed in the previous section with the work that we will present in this short
Section, we establish relations (in both directions) between certain classes of equipped
SUR-algebras and certain classes of equipped standard ZF-algebras, that ”explains” and
j j∗
”expands” the relations On N o and V SA.
b b∗
Motivated by the construction of N o as a space of signs builded over the class On

as a certain class of functions with domain ∈ On and codomain as a set of ordered
pairs with first component13 ∈ {−, +} (see Subsection 1.3.1.3, Chapter 1), we expand
the notion of ”space of signs”. Roughly speaking, given an standard ZF-algebra Z
(e.g., On, V ) and an ”operation on binary relations” op (e.g., ”identity”, ”transitive
closure”) we can build a corresponding SUR-algebra space of signs Sig(Z, op) (e.g.,
Sig(On, id) ∼ = SA, Sig(On, trcl) ∼= ST, Sig(V, id) ∼
= SA) and every SUR-algebra space
of signs is anchored on its underlying standard ZF-algebra, the domain of a sign function
determining the anchor d : Sig(Z, op) → Z, f 7→ d(f ) = domain(f ).
r
119. The space of signs of SA → On:
13
× Ps ({−, +}α ), such
S
Given β ∈ On, F : β → {−, +} can be identified with f : β → α∈β {−, +}
that f (α) = (F (α), F α ), α ∈ β.
3.4. SPACES OF SIGNS 87
S
We define, by ∈-recursion in On a subset Sig(On, id)(β)⊆{f :Sβ → α∈β {−, +} ×
Ps (Sig(On, id)(α))}
S and a (small) partial SUR-algebra structure on α∈β + Sig(On, id)(α),
β
S ∈ On, (or α∈β Sig(On, id)(α), β > 0) in such a way that Sig(On, id) :=
α∈On Sig(On, id)(α) (note that this reunion is disjoint) is endowed with a natural
structure of anchored SUR-algebra, with the obvious anchor d : Sig(On, id) → On,
f 7→ d(f ) = domain(f ) ∈ On. This construction (Sig(On, id), d) will be a representa-
tion of (SA, r) since it satisfies the conditions of the characterization theorem 99. The
steps of the Sig(On, id) construction are (of course) very similar to the SA construction
presented in the Subsection 2.2.2, Chapter 2, the only (irrelevant) difference is that the
emphasis is on the ”new” members of Sig(On, id), Sig(On, id)(β), instead of ”made”
members of SA, SAβ , β ∈ On. Thus we will just present sketches of the proofs.
To keep the notation simple, we will just denote Sig(On, id)(β) by Sig(β), in this
item.
SWe define Sig(β) and <β , β ∈ On, by recursion, where <β is a binary acyclic relation
in γ∈β + Sig(γ).
S
Sig(β)⊆{f : β → α∈β {−, +} × Ps (Sig(α))} is the subclass (that is a subset, by
induction) satisfying the following conditions:
(s1) If α ∈ β, then f (α) ∈ {−, +} × Ps (Sig(α));
(s2) (L(f ), R(f )) ∈ Cs (Sig (β) ) \ Cs (Sig (γ) ), ∀γ ∈ β, where L(f ) := α∈β {π2 f (α) :
S
π1 f (α) = +}, R(f ) := α∈β {π2 f (α) : π1 f (α) = −} and Cs (Sig (δ) ) := {(A, B) : A, B ∈
S
Ps ( α∈δ Sig(α)) : A <(δ) B}, <(δ) := α∈δ <α , δ ∈ On.

S S
Note that: S
• Sig(0) = {f : 0 → α∈0 {−, +} × Ps (Sig(α))} = { the unique function f : ∅ → ∅} =
{∗}. S S
• f : β → α∈β {−, +} × Ps (Sig(α)) ∈ Sig(β) iff −f : β → α∈β {−, +} × Ps (Sig(α)) ∈
Sig(β), where {π1 (−f (α))} = {+, −} \ {π1 (f (α))} , α ∈ β.
• For each β ∈ On and f ∈ Sig(β), define [f ] := {α ∈ β : π1 f (α) = ∈ {−, +}}, then:
[f ]− ∩ [f ]+ = ∅ (by s1) and L(f ) ∩ R(f ) = ∅ (otherwise ∃α ∈ [f ]− , ∃α0 ∈ [f ]+ , ∃g ∈
π2 f (α) ∩ π2 f (α0 )⊆Sig(α) ∩ Sig(α0 ) = ∅).
Define <β :=<(β) ∪{(g, f ) : f ∈ Sig(β), g ∈ L(f )} ∪ {(f, h)S: f ∈ Sig(β), h ∈ R(f )}.
We can check, by induction, that <β is an acyclic relation in γ∈β + Sig(γ).
.
S S
Let Sig := β∈On Sig(β) and <:= β∈On <β . Define Cs (Sig) := {(A, B) : A, B ∈
Ps ( α∈On Sig(α)) : A < B}, then Cs (Sig) = β∈On Cs (Sig (β) ).
S S
We can define − : Sig → Sig, f ∈ Sig(β) 7→ −f ∈ Sig(β). Define d : Sig → On,

f →7 d(f ) = domain(f ). Since d−1 [{β}] = Sig(β), d has small fibers. It is clear that
g < h iff (d(g) ∈ d(h) and g ∈ L(h)) or (d(h) ∈ d(g) and h ∈ R(g)).
Define uβ : γ∈β + Sig(γ) → Cs (Sig (β) ), f 7→ (L(f ), R(f )) . By induction, uβ is a
S
bijection. Denote tβ = (uβ )−1 : Cs (Sig (β) ) → γ∈β + Sig(γ), β ∈ On. The gluing of this
S
increasing and compatible family of bijections is the bijection t : Cs (Sig) → Sig.

It is a routine checking that Sig = (Sig, <, −, ∗, t) is a SUR-algebra.
S Let f ∈ Sig(β).SBy (s2) β is the least ordinal > α, ∀α ∈ [f ]− ∪ [f ]+ , thus d(f ) = β =

{α+ : α ∈ β} = {d(g)+ : g ∈ L(f ) ∪ R(f )}. The map s : On → Sig, s(β) ∈ Sig(β),
π1 (s(β)(α)) = +, ∀α ∈ β, defines a section of d, i.e. d ◦ s = idOn , thus d is surjective.
Then d : Sig → On is an anchor on Sig that satisfies (b5)strong .
Summing up, the anchored SUR-algebra (Sig, d) satisfies the conditions (rSA1),
(rSA2), (rSA3) in the characterization Theorem 99. Thus there is a unique isomorphism
∼
=
of anchored SUR-algebras (SA, r) → (Sig, d).

r0
120. The space of signs of ST → On:
S we can define, by ∈-recursion in On a subset

In the same vein of 119 above,
Sig(On, trcl)(β)⊆{fS: β → α∈β {−, +} × Ps (Sig(On, trcl)(α))} and a partial SUR-
algebra
S structure on α∈β + Sig(On, trcl)(α), β ∈ On, in such a way that Sig(On, trcl) :=
α∈On Sig(On, trcl)(α) is endowed with a natural structure of anchored SUR-algebra,
with the obvious anchor d : Sig(On, trcl) → On, f 7→ d(f ) = domain(f ) ∈ On.
To keep the notation simple, we will just denote Sig(On, trcl)(β) by Sig(β), in this
item.
SWe define Sig(β) and <β , β ∈ On, by recursion, where <β is a binary acyclic relation
in γ∈β + Sig(γ).
S
Sig(β)⊆{f : β → α∈β {−, +} × Ps (Sig(α))} is the subclass (that is a subset, by
(s1) If α ∈ β, then f (α) ∈ {−, +} × Ps (Sig(α));
(s2) (L(f ), R(f )) ∈ Cs (Sig (β) ) \ Cs (Sig (γ) ), ∀γ ∈ β, where L(f ) := α∈β {π2 f (α) :
S
π1 f (α) = +}, R(f ) := α∈β {π2 f (α) : π1 f (α) = −} and Cs (Sig (β) ) = {(A, B) : A, B ∈
S
Ps ( α∈β Sig(α)) : A <(β) B}, <(β) = α∈β <α .

S S
Define <β as the transitive closure of the relation <(β) ∪{(g, f ) : f ∈ Sig(β), g ∈
L(f )} ∪ {(f, h) : f ∈ Sig(β), h ∈SR(f )}. We can check, byS induction, that <β is an
acyclic and transitive relation in γ∈β + Sig(γ). Then <:= β∈On <β is an acyclic and
.
S
transitive relation in Sig := β∈On Sig(β).
The definitions of ∗, −, t, d, s are similiar to the Sig SA case. Then Sig = (Sig, <
, −, ∗, t) is a SUR-algebra such that t : Cs (Sig) → Sig is bijective and d : Sig → On
is an anchorSon Sig that satisfies (b5)strong , since for each β ∈ On and f ∈ Sig(β),
d(f ) = β = {α : α ∈ β} = {d(g)+ : g ∈ L(f ) ∪ R(f )}.
+
S
Summing up, the anchored SUR-algebra (Sig, d) satisfies the conditions (rST1),
(rST2), (rST3) in the characterization Theorem 100. Thus there is a unique isomor-
∼
=
phism of anchored SUR-algebras (ST, r0 ) → (Sig, d).

b∗
121. The space of signs of SA → V :
S
We define, by recursion on y ∈ V , a subset Sig(V, id)(y)⊆{f : y → z∈y {−, +} ×
Ps (Sig(V, id)(z))}, and we admit that we have already defined Sig(V, id)(z), ∀z ∈ Vρ(y)
and S Vρ(y) := z∈Vρ(y) Sig(V, id)(z) is a (small) partial SUR-algebra whenever ρ(y) > 0.
S
S S
Since V = α∈On Vα (Vα is a transitive set), the class Sig(V, id) := y∈V Sig(V, id)(y)
(note that this reunion is disjoint) is endowed with a natural structure of anchored SUR-
algebra, with the obvious anchor d : Sig(V, id) → V , f 7→ d(f ) = domain(f ) ∈ V .
This construction (Sig(V, id), d) will be a representation of (SA, b∗ ) since it satisfies the
conditions of the characterization theorem 101. The steps of the Sig(V, id) construction
are (of course) very similar to the SA construction presented in the Subsection 2.2.2,
Chapter 2, the only (irrelevant) difference is that the emphasis is on the ”new” members
of Sig(V, id), Sig(V, id)(y), instead of ”made” members of SA, SAρ(y) , y ∈ V .
We will just present sketches the constructions and results. To keep the notation
simple, we will just denote Sig(V, id)(y) by Sig(y), y ∈ V , in this item.
S
Let Sig(y)⊆{f : y → z∈y {−, +} × Ps (Sig(z))} be the subclass (that is a subset, by
(s1) If x ∈ y, then f (x) ∈ {−, +} × Ps (Sig(x));
(s2) (L(f ), R(f )) ∈ Cs (Sig (y) )\Cs (Sig (z) ), ∀z ∈ Ps (y)\{y}, where L(f ) := x∈y {π2 f (x) :
S
π1 f (x) = +}, R(f ) := x∈y {π2 f (x) : π1 f (x) = −} and Cs (Sig (y) ) := {(A, B) : A, B ∈
S
Ps ( x∈y Sig(x)) : A <(y) B}, <(y) := z∈y <z , y ∈ V .

S S
Note that:
• S V∅ = ∅. S
• Sig(∅) = {f : ∅ → z∈V0 {−, +} × Ps (Sig(z))} = { the unique function f : ∅ → ∅} =
{∗}. S S
• f : y → z∈Vρ(y) {−, +} × Ps (Sig(z)) ∈ Sig(y) iff −f : y → z∈Vρ(y) {−, +} ×
Ps (Sig(z)) ∈ Sig(y), where {π1 (−f (x))} = {+, −} \ {π1 (f (x))} , x ∈ y.
• For each y ∈ V and f ∈ Sig(y), define [f ] := {x ∈ y : π1 f (x) = ∈ {−, +}}, then:
[f ]− ∩ [f ]+ = ∅ (by s1) and L(f ) ∩ R(f ) = ∅ (otherwise ∃x ∈ [f ]− , ∃x0 ∈ [f ]+ , ∃g ∈
π2 f (x) ∩ π2 f (x0 )⊆Sig(x) ∩ Sig(x0 ) = ∅).
If β 6= ∅, S Vβ := z∈Vβ Sig(z) is a partial SUR-algebra:
S
• <β , β ∈ On, is a binary acyclic relation on S Vβ+ given by <β :=<(Vβ ) ∪ ρ(y)=β∈On {(g, f ) :
S
f ∈ Sig(y), g ∈ L(f )} ∪ {(f, h) : f ∈ Sig(y), h ∈ R(f )}.

• Define uβ : S Vβ+ → Cs (Sig (Vβ ) ), f 7→ (L(f ), R(f )) . By induction, uβ is a bijection.
Denote tβ = (uβ )−1 : Cs (Sig (Vβ ) ) → S Vβ+ , β ∈ On.
.
Let Sig := y∈V Sig(y) = β∈On S Vβ and <:= β∈On <β . Define Cs (Sig) :=
S S S
{(A, B) : A, B ∈ Ps ( y∈V Sig(y)) : A < B}, then Cs (Sig) = y∈V Cs (Sig (y) ). The
S S
gluing of the increasing and compatible family of bijections {tβ : β ∈ On} is the bijec-
tion t : Cs (Sig) → Sig. We can define − : Sig → Sig, f ∈ Sig(y) 7→ −f ∈ Sig(y).
Define d : Sig → On, f 7→ d(f ) = domain(f ). Since d−1 (β) = Sig(β), d has small fibers.
It is clear that g < h iff (d(g) ∈ d(h) and g ∈ L(h)) or (d(h) ∈ d(g) and h ∈ R(g)).
It is a routine checking that Sig = (Sig, <, −, ∗, t) is a SUR-algebra.
Let f ∈ Sig(y), y ∈SV . By (s2) y is the ⊆-least
S w ∈ V such that ∀x ∈ [f ]− ∪ [f ]+ , x ∈
w, thus d(f ) = y = {{x} : x ∈ y} = {{d(g)} : g ∈ L(f ) ∪ R(f )}. The map
s : V → Sig, s(y) ∈ Sig(y), π1 (s(y)(x)) = +, ∀x ∈ y, defines a section of d, i.e.
d ◦ s = idOn , thus d is surjective. Then d : Sig → V is an anchor on Sig that satisfies
(b5)strong .
Summing up, the anchored SUR-algebra (Sig, d) satisfies the conditions (bSA1),
(bSA2), (bSA3) in the characterization Theorem 101. Thus there is a unique isomor-
∼
=
phism of anchored SUR-algebras (SA, b∗ ) → (Sig, d).

We finish this short Section with some commentaries that indicates the possible
expansions of the constructions above.
Remark 122.
We have defined above anchored SUR-algebras ”space of signs” Sig(Z, op), where
Z(= On, V ) is a standard ZF-algebra and op(= id, trcl) is an ”operation” on relations,
such that t : Cs (Sig(Z, op)) → Sig(Z, op) is a bijection. Obviously such constructions
cannot represent our original motivation: the anchored SUR-algebra (N o, b) is such that
t : Cs (N o) → N o is surjective, but its fibers are proper classes.
The process of obtaining an anchored SUR-algebras ”space of signs” can be slightly
generalized adding another components to the construction: a initial step (a convenient
partial SUR-algebra) and a coherent specification of equivalence relations: the N o situ-
ation is the motivation to consider coherent equivalence relations (see Fact 27).
If Z is a standard ZF-algebra such that s(z) ≤ s(Φ) iff z = Φ, then we define a ”space
of signs specification” as a 6-upla σ = (Z, L, v, op, I, ∼), where:
• I is partial SUR-algebra such that t : Cst (I) → I is surjective;
• L⊆Z is a well-ordered subclass that contains the root Φ ∈ Z and such that v : L → Z
satisfies v(Φ) = s(Φ), λ0 < λ ∈ L⇒v(λ0 )εv(λ) and ∀z ∈ Z, ∃λ ∈ L, zεv(λ). Denote
rv (z) = λ ∈ L iff zεv(λ) but z 6 εv(λ0 ), ∀λ0 < λ.
By a convenient cumulative construction by ε-recursion (sketched below) we can
obtain an anchored SUR-algebra Sig(σ) –that from now on we just denote simply by
Sig – that expands the partial SUR-algebra I, such that t : Cs (Sig) → Sig is surjective
and the domain function determines the anchor d : Sig(σ) → Z, f 7→ d(f ) = domain(f ):
• Sig(Φ) = I. S
• For each y ∈ Z \ {Φ}, Sig(y)⊆{f : y → zεy {−, +} × Ps (Sig(z))}.
• Define Σ(λ) = Cs ( zεv(λ) Sig(z)) = {(A, B) : A, B ∈ Ps ( zεv(λ) Sig(z)), A <(λ) B},
S S
where <(λ) = op( λ0 ελ <λ0 ). Note that Σ(Φ) = Cs (I).

S
• ∼l ⊆Σ(λ) × Σ(λ) is an equivalence relation compatible with <(λ) (see. for instance, Fact
27) and ∼Φ is such that Σ(Φ) / ∼Φ ∼ = I.
0
• A coherent family of injective maps λ
S jz0 : Sig(z ) ∼ rv (z 0 ) ≤ λ with disjoint
Σ(λ) / ∼λ , S
images that covers Σ / ∼λ (thus rv (z)=λ Sig(z) = Σ / ∼λ \ rv (z0 )<λ jzλ0 [Sig(z 0 )] is a
(λ) (λ)
class of ”new members” at level λ and Σ(λ) / ∼λ represents the ”made members” at level
λ).
• Given a ”pre-number” (A, B) ∈ Σ(λ) = Cs ( zεv(λ) Sig(z)) define t(A, B) as the unique
S
0
function f ∈ Sig(z
S ) such that jzλ0 (f ) = [(A, B)]∼λ .
• Define Sig = z∈Z Sig(z) ∼ = colimλ∈L Σ(λ) .
It can be defined a notion of morphism of space of signs specification (a morphism
of rooted well founded classes that is compatible with additional structure): this will
induce an anchored SUR-algebra morphism (see Remark 103).

Chapter 4
Set Theory in the free Surreal

Algebra
The initial motivation for this work is to obtain an ”algebraic set theory for surreals”
along the lines of the Algebraic Set Theory: to establish abstract and general links
between the class of all surreal numbers and a universe of ”surreal sets” similar to (but
expanding it) the (ZF-algebra) relations between the classes On and V , giving the first
steps towards a certain kind of (alternative) ”relative set theory” (see [Fre16] for another
presentation of this general notion).
Noting that:
(i) the (injective) map j : On → N o, is a kind of ”homomorphism”, partially encoding
the ordinal arithmetic;
(ii) the traditional set-theoretic constructions (in V ) keep some relation with itsS(ordinal)
complexity
S (e.g., x ∈ y → ρ(x) < ρ(y), ρ({x}) = ρ(P (x)) = ρ(x) + 1, ρ( i∈I xi ) =
i∈I ρ(xi ));
then we wonder about the possibility of this new structured domain SA determines a
category, by the encoding of arrows (and composition) as objects of SA, in an analogous
fashion to the way that the class V of all sets determines a category, i.e. by the encoding
of some notion of ”function” as certain surreal set (i.e. an objects of SA); testing, in
particular, the degree of compatibility of such constructions with the map j ∗ : V → SA
and examining if this new expanded domain could encode homomorphically the cardinal
arithmetic.
Remark that the usual set/class theories (as ZFC or NBG) have the power of ”encode”
(syntactically) its Model Theory: constructions of models of set theory by the Cohen
forcing method or through boolean valued models method are done by a convenient
encoding of the fundamental binary relations ∈ and =.
Thus, for us, there are three main requirements for a theory deserves be named a
”Set Theory”:
93
94 CHAPTER 4. SET THEORY IN THE FREE SURREAL ALGEBRA
(i) have the potential to define arrows (category) as entities of the theory, through a
fundamental binary relation;
(ii) be the ”derived set theory” of a free object in a category (like in ZF-algebra setting);
(iii) its ”internal” category is a topos-like category.
We develop in this Chapter the first steps of a certain kind of set theory based (or
ranked) on surreal numbers, that expands the relation between V and On.
In fact, we work out a ”positive set theory” on SA ranked on N o, that expands the
ZF-algebra relationship V → On through the ”positive” map j + : V → SA, given by
j + (X) = j ∗ (X) = hj + [X], ∅i, X ∈ V . Thus the free/initial SUR-algebra SA supports, in
many senses, an expansion of the free/initial ZF-algebra V and its underlying set theory.
To accomplish this, we use the functions j ∗ and b∗ that establishes connections between V
and SA. Under logical and set-theoretical perspective, the map j + : V → SA preserves
and reflects many constructions. On the category-theoretic perspective, the map j +
defines a full, faithful and logical functor j + : Set → Cat+ (SA), from the topos Set
associated to V into the topos Cat+ (SA) associated to SA.
4.1 Set Theory in the Surreal Algebra SA

The aim of this Section is the introduction of certain set-theoretic-like structure on
the class SA. To accomplish this we will use the functions j ∗ and b∗ that establishes
connections between V and SA.
The requirements for the set-theoretic operations are the following: for the ”stan-
dard sets”, i.e., the SA members in the copy j ∗ [V ] of the set-theoretic universe V , the
operations are essentially the same.
All the proofs in this section are straightforward verifications and will be frequently
omitted.
4.1.1 The maps (·)p and (·)n
In order to study the maps b∗ and j ∗ we need to introduce two functions of great impor-
tance for the constructions in SA
Definition 123.If A ∈ Ps (SA), we define:

• Ap := hA, ∅i ∈ SA;
• An := h∅, Ai ∈ SA;
We have then two bijective maps that associates subsets of SA to elements of SA:
• (·)p : Ps (SA) → P os(SA);
• (·)n : Ps (SA) → N eg(SA).
4.1. SET THEORY IN THE SURREAL ALGEBRA SA 95
Remark 124. If x ∈ V we can consider j + (x) := hj + [x], ∅i and j − (x) := h∅, j − [x]i, then
j + (x) = j ∗ (x) = −j − (x) (by induction on ρ(x)).

Definition 125. Let u = hLu , Ru i, v = hLv , Rv i ∈ SA. Define:
(a) u ≺ v iff b∗ (u) ∈ b∗ (v).
(b) l(u, v) iff u < v and u ≺ v;
(c) r(u, v) iff u < v and v ≺ u.
(d) uεv iff l(u, v), whenever v ∈ P os(SA).

Remark 126.
(a) By Proposition 87.(c) and 98 ≺ is a well founded relation in SA. We also have
that u < v implies u ≺ v or v ≺ u (by Fact 96).
(b) By Proposition 95 and its proof:
(i) b∗ ◦ j ∗ = idV , thus b∗ is surjective and j ∗ is injective.
(ii) Let X, Y ∈ V . Then: X ∈ Y iff j ∗ (X) < j ∗ (Y ).
(iii) Let Y ∈ V and a ∈ SA. If a < j ∗ (Y ) and b∗ (a) ∈ b∗ (j ∗ (Y )) then a = j ∗ (X) for a
unique X ∈ Y .
(c) If v ∈ SA, then {u ∈ SA : l(u, v)}, {w ∈ SA : r(v, w)} ∈ Ps (SA).
(d) If w = hW, ∅i ∈ P os(SA), then wε = {u ∈ SA : l(u, w)} = {u ∈ SA : u ∈ Lw } =
{u ∈ SA : u ∈ W } ∈ Ps (SA)

4.1.2 Inclusion, union, intersection and power
For the definitions of inclusion, union and order we have definitions, for all SA members,
that are compatible with the maps j + and j − .
Definition 127.Let u, v ∈ SA. We define the inclusion ⊆∗ as the following relation:
u ⊆∗ v iff Lu ⊆ Lv and Ru ⊇ Rv

Definition 128.Let {ui }i∈I an indexed non-empty set of elements of SA.
\∗ \ [
1. ui := h Lui , Rui i
i∈I i∈I i∈I
[∗ [ \
2. ui := h Lui , Rui i
i∈I i∈I i∈I
S∗
(when I = ∅ define ∅ = h∅, ∅i)
The next result follows directly from the correspondent properties of ⊆:

Proposition 129.⊆∗ is an order relation in SA.

In respect of the maps j + and ρ∗ : SA → N o we have the following result:

Proposition 130.Let X, Y ∈ V and u, v ∈ SA
1. X ⊆ Y iff j + (X) ⊆∗ j + (Y )
2. u ⊆∗ v implies ρ∗ (u) 6 ρ∗ (v)
With this definitions we have interesting properties, similar to the correspondent ones
we have in set theory:
Proposition 131.Let ui ∈ SA for i ∈ I 6= ∅ and let v ∈ SA. Then:
[∗
1. ∀i ∈ I, ui ⊆∗ v ⇐⇒ ui ⊆∗ v
i∈I
\∗
2. ∀i ∈ I, ui ⊇∗ v ⇐⇒ ui ⊇∗ v
i∈I
Another interesting fact about j ∗ and our definitions of union and intersection is the
following result:
Proposition 132.If {Ai }i∈I is an indexed family of sets in V we have:
[ [∗
1. j + ( Ai ) = j + (Ai )
i∈I i∈I
\ \∗
2. j + ( Ai ) = j + (Ai )
i∈I i∈I
4.1. SET THEORY IN THE SURREAL ALGEBRA SA 97

Definition 133.Let u = hU, ∅i ∈ P os(SA). We define P + (u) := h{hz, ∅i : z⊆U }, ∅i

The relationship between (positive) powers and unions in SA extends the relationship
between powers and unions in V .
Proposition 134.For each C, U ∈ Ps (P os(SA)):
∗
C⊆∗ U p iff C p ⊆∗ P + (U p )
S
4.1.3 Singletons, doubletons and ordered pairs
In order to define the notion ordered pairs in SA we need suitable notions of singletons
and doubletons:
Definition 135.Let u, v, w ∈ SA.
• {u}+ := h{u}, ∅i
• {u, v}+ := h{u, v}, ∅i
• (u, v)+ := {{u}+ , {u, v}+ }+
• (u, v, w)+ := ((u, v)+ , w)+
Proposition 136.(u, v)+ = (u0 , v 0 )+ ∈ SA iff u = u0 and v = v 0 iff (u, v) = (u0 , v 0 ) ∈ V
All these constructions are compatible with the set theoretic notions induced by j ∗ .
Proposition 137.Let x, y ∈ V .
1. j + (∅) = ∅+
2. j + ({x}) = {j + (x)}+
3. j + ((x, y)) = (j + (x), j + (y))+
Proof.
1. j + (z) = ∅+ iff z = ∅
2. j + (z) = {r}+ iff ∃!x ∈ V, z = {x}, j + (x) = r
3. j + (z) = (r, s)+ iff ∃!x, y ∈ V, z = (x, y), j + (x) = r, j + (y) = s

4.1.4 Cartesian product, relations and functions
Having the notion of (positive) ordered pair in SA, we are able to define (positive)
cartesian product, relations and funcions in SA.
Definition 138.
1. If w ∈ P os(SA), dom+ (w) := h{a : ∃b, l((a, b)+ , w)}, ∅i ∈ P os(SA) and
ran+ (w) := h{b : ∃a, l((a, b)+ , W )}, ∅i ∈ P os(SA).
2. A positive (small) relation r is a positive SA member whose ε-members are positive
ordered pairs. More precisely, r = Rp = hR, ∅i, where R = h{(ai , bi )+ : i ∈ I}, ∅i ∈
P os(SA).
3. Positive cartesian product: if u, v ∈ P os(SA),
(u ×+ v) := h{(a, b)+ : l(a, u), l(b, v)}, ∅i ∈ P os(SA).
4. Positive composition of positive relations:
Rp ◦+ S p = h{(a, c)+ : ∃b, l((a, b)+ , S p ), l((b, c)+ , Rp )}, ∅i ∈ P os(SA).
5. Positive identity relations: if v ∈ P os(SA), define ∆+
v = h{(u, u)
+
: uεv}, ∅i ∈
P os(SA).
6. If u, v ∈ P os(SA), then a positive functional relation from u into v is a positive
relation r such that dom+ (r) = u, ran+ (r)⊆∗ v and such that (a, b)+ , (a, b0 )+ ∈
L r ⇒ b = b0 .

Proposition 139.
(a) Let u = hU, ∅i, v = hV, ∅i ∈ P os(SA), then: (a, b)+ ε u ×+ v iff (a, b) ∈ Lu × Lv =
U × V . Thus
(b) If r ∈ P os(SA) is a positive relation, then: r⊆∗ u ×+ v, where u = dom+ (r), v =
ran+ (r) and R := {(a, b) : (a, b)+ ∈ Lr }⊆Lu × Lv , Lu ⊆ dom(R) and Lv ⊆ ran(R).
(c) If u, v ∈ P os(SA) and r ∈ P os(SA) is a positive functional relation from u into
v then {(a, b) : (a, b)+ ∈ Lr } is a functional relation from Lu into Lv .
(d) If x, y ∈ V , then j + (x, y) = (j + (x), j + (y))+ .
(e) If X, Y ∈ V , then j + (X × Y ) = j + (X) ×+ j + (Y ).
(f ) If R, S are small relations in V , then j + (R ◦ S) = j + (R) ◦+ j + (S).
(g) If X ∈ V , then j + (∆X ) = ∆+
j + (X) .
(h) If X, Y ∈ V , then r⊆∗ j + (X) ×+ j + (Y ) is a positive functional relation from

j + (X) into j + (Y ) iff r = j + (R) for a unique R⊆X × Y , a functional relation from X
into Y .
J∗
4.2. LOGICAL ANALYSIS OF THE FUNCTIONS V SA 99
B∗
Proof. We will just sketch some proofs.

Item (e): j + (X × Y ) = hj + [X × Y ], ∅i = h{(j + (a), j + (b))+ : (a, b) ∈ X × Y }, ∅i =
h{(j + (a), j + (b))+ : (a, b) ∈ X × Y }, ∅i = h{(j + (a), j + (b))+ : (j + (a), j + (b)) ∈ j + [X] ×
j + [Y ]}, ∅i = j + (X) ×+ j + (Y )
Item (g): j + (∆X ) = h{j + ((x, x)) : x ∈ X}, ∅i = h{(j + (x), j + (x))+ : x ∈ X}, ∅i =
∆+
j + (X)
j∗
4.2 Logical analysis of the functions V SA
b∗
Since j ∗ (x) = hj ∗ [x], ∅i ∈ P os(SA), x ∈ V , we have the injective (class) function j + :

V → P os(SA). We saw in the previous Section that j + is ”elementary”, i.e., it preserves
and reflects many set-theoretic constructions: ∅, singletons, (ordered) pairs, cartesian
products, relations, functions, composition of relations/functions, identities, ...
On the other hand, recall that b∗ (u) = b∗ [Lu ] ∪ b∗ [Ru ], for each u = hLu , Ru i in SA.
We will denote b+ = b∗ : P os(SA) → V . b+ preserves some set theoretical constructions:
140. b∗ : SA → V and set-theoretical constructions:
(a)
• 0 = h∅, ∅i ∈ P os(SA) and b+ (0) = b∗ (h∅, ∅i) = b∗ [∅] ∪ b∗ [∅] = ∅
• If x ∈ P os(SA), then {x, y}+ ∈ P os(SA) and b+ ({x, y}+ ) = b∗ (h{x, y}, ∅i) =
b [{x, y}] ∪ b∗ [∅] = {b∗ (x), b∗ (y)} = {b+ (x), b+ (y)}
∗
• If x, y ∈ P os(SA), then (x, y)+ ∈ P os(SA) and b+ ((x, y)+ ) = b+ ({{x}+ , {x, y}+ }+ ) =
{b ({x}+ ), b∗ ({x, y}+ )} ∪ b∗ [∅] = {{b∗ (x)}, {b∗ (x), b∗ (y)}} = (b+ (x), b+ (y))
∗
(b) If u, v ∈ P os(SA)
• if u⊆∗ v, then b+ (u)⊆b+ (v)
• b+ (dom+ (u)) = dom(b+ (u))
• b+ (ran+ (u)) = ran(b+ (u))
• b+ (u ×+ v) = b+ (u) × b+ (v)
(c) If r is a positive relation in SA, r⊆∗ u ×+ v. Then
• b+ (r)⊆b+ (u) × b+ (v) is a relation
• b+ (r ◦+ s)⊆b+ (r) ◦ b+ (s)

141.
(I) In order to study more closely the relations between V and SA we recall the
Definition 125: let u = hLu , Ru i, v = hLv , Rv i ∈ SA:
• u ≺ v iff b∗ (u) ∈ b∗ (v).
• l(u, v) iff u < v and u ≺ v;
• r(u, v) iff u < v and v ≺ u.
• uεv iff l(u, v), whenever v ∈ P os(SA).
We also define two languages:
• LSur = {l(−, −), r(−, −)}
• L+
Sur = {l(−, −)}
(II) We have a “left” interpretation (−)] : F or(LZF ) → F or(LSur ) that is defined

by:
• atomic: (v1 ∈ v2 ) 7→ l(v1 , v2 );
• ¬, ∧, ∨, →, ∃, ∀ (and =): defined in the obvious way.
We also have a “Reversing” interpretation: (−)[ : F orm(L+
Sur ) → F orm(LZF ):
• atomic: l(v1 , v2 ) 7→ (v1 ∈ v2 );
• ¬, ∧, ∨, →, ∃, ∀ (and =): defined in the obvious way.
(III) The (positive) set theory in SA ”extends” the set theory in V in the following
sense:
”Axioms” (= properties of SA, P os(SA))
Extensionality:
Ext∗ (s, s0 ) := ∀u∀v((l(u, s) ↔ l(u, s0 )) ∧ (r(s, v) ↔ r(s0 , v)) → s = s0
Ext+ (s, s0 ) := ∀u(l(u, s) ↔ l(u, s0 )) → s = s0
SA ∀s, ∀s0 Ext∗ (s, s0 )
SA ∀pos s, ∀pos s0 Ext+ (s, s0 )
SA ∀x∀x0 (ExtZF (x, x0 ))] ↔ Ext+ (j + (x), j + (x0 ))
V ∀pos s∀pos s0 ((Ext+ (s, s0 ))[ → ExtZF (b+ (s), b+ (s0 ))
(IV) We have convenient (positive) versions of [emptyset], [comprehension], [regu-

larity],[pair], [reunion], [parts],..
4.3. CAT + (SA) AND THE FUNCTOR INDUCED BY J + : V → SA 101
(V) Metatheorem: V |= ψ(ȳ|ā) iff SA |= ψ ] (ȳ|j(ā))

Remark 142.
As in the algebraic set theory (see Section 2, Chapter 1 or [JM95]), we can try to
describe the ”surreal” set theory as the theory of derived from its the initial object SA.
Thus another possibility is consider a more comprehensive fragment of the language
of the initial (equipped) SUR-algebra SA, given by <, −, ∗, ≺, where ≺=≺b∗ 1
We can define a ternary incidence relation in SA: for each a, u, b ∈ SA, m(a, u, b)
iff a ∈ Lu and b ∈ Ru (iff l(a, u) and r(u, b)). Thus m(a, u, b)⇒a < b, a ≺ u, b ≺ u.
This seems to have good potential (that we intend to explore in the future). For
instance:
• With this new language we can formulate a new and general ”extensionality prin-
ciple” for general members of SA:
ExtSA (u, v) : (∀a∀b(m(a, u, b)⇔m(a, v, b)) ⇔ u = v), for each u, v ∈ SA.
The ”extensionality principle” is the formula ∀u∀v(ExtSA (u, v)).
This encodes u = v iff Lu = Lv and Ru = Rv .
• We have the empty surreal set axiom/property:
Empty SA (u) : (∃u∀a∀b(¬m(a, u, b))
The ”empty surreal set axiom/property” is the formula ∃u(Empty SA (u)).
This encodes the SA member ∗ = h∅, ∅i.

4.3 Cat+(SA) and the functor induced by j + : V → SA

In this Section, we describe a category of positive SA members and positive arrows,
Cat+ (SA), and establish functorial relationship between this category and the category
Cat(V ) = Set of of all sets and functions: this turns out to be induced by the ”extension”
map j + : V → SA. The main categorial-theoretic references for this Section are [BW85]
and [Bor94].
Definition 143. We denote Cat+ (SA) the category whose objects are all the positive
objects of SA, i.e. Obj(Cat+ (SA)) = P os(SA) and such that, for each u, v ∈ P os(SA),
Cat+ (SA)(u, v) = {(u, φ, v)+ : φ : u →+ v is a positive function from u into v}2 . Since
1
Since b∗ : SA → V is the ((b5)strong ) anchor from the initial SUR-algebra SA on the initial
(standard) ZF-algebra V .
2
Notation: (a, b, c)+ := ((a, b)+ , c+ )+ .
the composition of positive functions is a associative and the identity positive function
id+ +
u : u → u is an identity for that composition, this indeed turns out to be a category.
Note that this category Cat+ (SA) is large (i.e. its class of objects and class of arrows
are proper classes) but is locally small: if A, B ∈ Ps (SA), let α the least ordinal such
that A, B⊆SA(α) , then Ap = hA, ∅i, B p = hB, ∅i ∈ P osα (SA) = P os(SA) ∩ SAα and
then Cat+ (SA)(Ap , B p )⊆P osα+9 (SA)⊆SAα+9 , thus it is a set.
144. The functor induced by j + : V → P os(SA):

By the properties established in Section 1 (see Proposition 139), the map j + : V →
P os(SA) determines a functor J + = (J0+ , J1+ ) : Cat(V ) → Cat+ (SA), given by:
• If X ∈ V , then J0+ (X) := j + (X) ∈ P os(SA);
• If f = (X, F, Y ) : X → Y is a function from X into Y , then J1+ (f ) = (j + (X), Φ, j + (Y )) :
J0+ (X) →+ J0+ (Y ) is a positive function from j + (X) into j + (Y ) determined by the posi-
tive functional relation Φ from j + (X) into j + (Y ): Φ = h{(j + (a), j + (b))+ : (a, b) ∈ F }, ∅i.
Note that LΦ is a set.
By the Proposition 139, the functor J + : Cat(V ) → Cat+ (SA) is injective on objects,
full and faithful (thus it reflects isomorphisms).

The functor J + : Cat(V ) → Cat+ (SA) will help us to establish categorial construc-
tions and properties of Cat+ (SA). Below we construct categorial products in Cat+ (SA).
Proposition 145. Let D : I → Cat+ (SA) be a diagram (= functor) over a small

discrete category I (≡ a set I). Then D has admits a cone limit in Cat+ (SA). I.e.,
D
given a function i ∈ I 7→ Q
D(i) = ui = hUi , ∅i ∈ P os(SA), then we can define
Q+ an object
+ +
u ∈ P os(SA), denoted by i∈I ui and a family of positive functions πk : i∈I ui →+ uk ,
k ∈ I, such that for each w = hW, ∅i ∈ P os(SA) and a family of positive
Q+ functions
+ + +
φk : w → uk , k ∈ I, there is a unique positive function φ : w → i∈I ui such that
+ +
πk ◦ φ = φk , ∀k ∈ I.
Proof. We start with a set-theoretical construction in P os(SA). Define +

Q
i∈I ui :=
hP, ∅i, where:
P := {σ | σ = (j + (I), Σ, + +
is a positive function j + (I) →+ +
S S
i∈I ui ) i∈I ui such that
+
σ(j (i)) ∈ Ui , i ∈ I}.
Note that:
• Σ⊆∗ j + (I) ×+ ( +
S
i∈I ui );
• Lj + (I) = j + [I] = {j + (i) : i ∈ I} and LS+ ui = i∈I Ui ;
S
i∈I
• (a, b)+ ∈ LΣ ⇒ (a, b) ∈ j + [I] × i∈I

S
∼ Q Ui ;
• P is a set: indeed P = L + ui = j + (i)∈j + [I] Ui .
Q
i∈I
For each k ∈ I, define πk+ = ( + +

Q+
+ +
uk by πk+ (σ) =
Q
i∈I ui , Πk , uk ) : i∈I ui →
σ(j + (k)) ∈ Uk = Luk , σ ∈ P . This defines a positive function from +
Q
i∈I ui into
uk . Note that, through the bijection P = L Q +
∼ Q
= j + (i)∈j + [I] Ui , we correspond
i∈I ui
+
Q+ +
Q
πk : i∈I ui → uk with projk : j + (i)∈j + [I] Ui → Uk .
Now let w = hW, ∅i ∈ P os(SA) and a family of positive functions {φ+ k Q: w →+ uk :
+ +
k ∈ I}. We have to prove that there is a unique positive function φ : w → i∈I ui such
+ + +
that πk ◦ φ = φk , ∀k ∈ I.
Candidate and uniqueness: Suppose that φ : w →+ +
Q
i∈I ui is a positive function
+ + +
such that πk ◦ φ = φk , ∀k ∈ I.
Then for each x ∈ W = Lw , φ+ + +
k (x) = (πk ◦ φ)(x) = πk+ (φ(x)) = (φ(x))(j + (k)), ∀k ∈ I.
I.e., since φ(x) is a positive function j + (I) →+ +
S
i∈I ui , φ(x) is uniquely determined by
the commutativity condition.
Existence: Consider the family of functions fk : W → Uk , k ∈ I, Q such that for
+
each x ∈ W = Lw , fk (x) := φk (x) ∈ Luk = Uk . Denote f : W → j + (i)∈j + [I] Ui
+ +
the unique function such that (f (x))(j (k)) = fk (x) = φk (x), k ∈ I, x ∈ W . Define
φ
φ : w →+ +
Q + +
S+ +
i∈I ui by: x ∈ W = Lw 7→ φ(x) : j (I) → i∈I ui , where j (k) ∈
φ(x)
j + [I] = Lj + (I) →
7 (φ(x))(j + (k)) := (f (x))(j + (k)) = fk (x) = φ+
k (x) ∈ Uk = Luk . Then
+ +
πk ◦ φ = φk , ∀k ∈ I.

We saw in Section 1 that j + (A×B) = j + (A)×+ j + (B): this suggests that the functor
J + : Cat(V ) → Cat+ (SA) preserves arbitrary products in general. Instead of proceed
into long and direct calculations (in the same vein of the previous result), we establish
below general results that will provide to us a full description of the category Cat+ (SA).
We begin with the following:
146. The forgetful functor Λ : Cat+ (SA) → Cat(V ):

By the properties established in Section 1 (see Proposition 139), we can define two
maps, Λ0 , Λ1 , between proper classes, given by:
• If u = U p = hU, ∅i ∈ P os(SA), then Λ0 (u) := Lu = U ∈ V ;
• If φ = (u, Φ, w)+ : u →+ w is a positive function from u into v, then Λ1 (φ) = (U, F, W ) :
Λ0 (u) → Λ0 (w) is a function from U into W determined by the functional relation F
from U into W : F = {(a, b) : (a, b)+ ∈ LΦ } = {(a, b) : (a, b)+ εΦ}. Note that F is a set.
It is easy to see that this data determines a functor Λ = (Λ0 , Λ1 ) : Cat+ (SA) →
Cat(V ). Moreover Λ is injective on objects, full and faithful (thus it reflects isomor-
phisms).
147. Connecting J + and Λ:

J+
We present two ”natural” ways to connect Cat(V ) Cat+ (SA):
Λ
+
(a) For each x ∈ V , denote ηx : x → Λ(J (x)) the function given by: y ∈ x 7→
j (y) ∈ j + [x] = Λ(J + (x)). Since j + is injective, ηx is a bijection. We will denote
+
η := (ηx )x∈V .
(b) For each u ∈ P os(SA), denote θu : u →+ J + (Λ(u)) the positive function given by:
vεu 7→ j + (v)εj + (U ) = J + (Λ(u)). Since j + is injective and vεu = hU, ∅i⇔v ∈ Lu = U ,
then θu is a positive bijection3 . We will denote τu = (θu )−1 : J + (Λ(u)) →+ u, the inverse
bijection. We will denote τ := (τu )u∈P os(SA) .
Note that, in particular, every object in Cat+ (SA) is positively isomorphic to a standard
set positive object: i.e. for each u ∈ P os(SA) there are x ∈ V and a positive isomorphism
=+
∼
J + (x) → u.

Now we are ready to state and prove the:
Theorem 148. Consider the following data: (J + , Λ, η, τ ). Then:

∼
=
(a) η := (ηx )x∈V determines an invertible natural transformation η : IdCat(V ) →
Λ ◦ J +;
∼
=
τ := (τu )u∈P os(SA) determines an invertible natural transformation τ : J + ◦Λ → IdCat+ (SA) .
(b) (J + , Λ, η, τ ) is an equivalence of adjunction with unity η and co-unity τ .
Proof.
We will just proof that:
(i) For each x ∈ V , the function ηx : x → Λ(J + (x)) is a universal function from the
objects x into the functor Λ, i.e., for each u ∈ P os(SA) and each function f : x → Λ(u),
there is a unique positive function f ] : J + (x) →+ u such that f = Λ(f ] ) ◦ ηx ;
(ii) For each u ∈ P os(SA), τu = (idΛ(u) )] : J + (Λ(u)) →+ u.
Then, by well-known category-theoretic results:
• η and τ are (invertible) natural transformations;
• (J + , Λ, η, τ ) is an adjunction with unity η and co-unity τ .
Thus we obtain items (a) and (b).
3
Clearly, the positive bijections coincides with the isomorphisms in the category Cat+ (SA).
(i) Let x ∈ V , u ∈ P os(SA) and f : x → Λ(u) be a function:

Candidate and uniqueness: Suppose that φ = (J + (x), Φ, u) : J + (x) →+ u is a positive
function such that f = Λ(φ) ◦ ηx . Then, for each y ∈ x, (j + (y), b)+ εΦ iff b = φ(j + (y)) =
(Λ(φ))(j + (y)) = (Λ(φ) ◦ ηx )(y) = f (y). Thus the positive functional relation from J + (x)
into u, Φ, is uniquely determined by the condition.
Existence: Define f ] := (J + (x), Φ, u) where Φ := h{(j + (y), f (y))+ : y ∈ x}, ∅i. Then Φ
is a positive functional relation such that dom+ (Φ) = j + (x) and ran+ (Φ)⊆∗ u. If y ∈ x,
then f ] (j + (y)) = f (y), thus f = Λ(f ] ) ◦ ηx , establishing (i).
(ii) Let u ∈ P os(SA):
Then by the proof of (i) just above, (idΛ(u) )] = (J + (Λ(u)), Φ, u), where Φ = h{(j + (v), idLu (v))+ :
v ∈ Lu }, ∅i = h{(j + (v), v)+ : v ∈ Lu }, ∅i: this is precisely the positive functional rela-
tion from J + (Λ(u)) into u that determines τu , see 147.(b) above, thus τu = (idΛ(u) )] ,
establishing (ii).

Corollary 149. As a consequence of the previously established equivalence of categories

J+
Cat(V ) Cat+ (SA) we obtain:
Λ
• The functors J + : Cat(V ) → Cat+ (SA) and Λ : Cat+ (SA) → Cat(V ), preserve
and reflect limits and colimits.
• Since Set = Cat(V ) is a (boolean) Grothendieck topos (in particular, it is a complete
and cocomplete category), then Cat+ (SA) is a (boolean) Grothendieck topos.
• Since {1} is a generator subclass in the category Cat(V ), then {J + (1)} is a genera-
tor subclass in the category Cat+ (SA), i.e. for each pair of distinct parallel positive func-
tions φ, ψ : u →+ v, there is a positive function θ : J + (1) →+ u such that φ ◦+ θ 6= ψ ◦+ θ.
Thus, as in Set, a positive arrow in Cat+ (SA) is an monomorphism iff it is a positive
injection.
• Since {2} is a cogenerator subclass in the category Cat(V ), then {J + (2)} is a
cogenerator subclass in the category Cat+ (SA), i.e. for each pair of distinct parallel
positive functions φ, ψ : u →+ v, there is a positive function θ : v →+ J + (2) such that
θ ◦+ φ 6= θ ◦+ ψ. Thus, as in Set, a positive arrow in Cat+ (SA) is an epimorphism iff
it is a positive surjection.
• The arithmetic of cardinalities (i.e., of the equivalence classes of objects under
isomorphisms) in Cat+ (SA) coincides with the usual arithmetic of cardinalities is Set.
• Since T : 1 → 2, T (0) = 1 is the (unique up to unique isomorphism) subobject clas-
sifier in Cat(V ), then J + (T ) : J + (1) →+ J + (2) is the subobject classifier in Cat+ (SA).
• J + and λ are logical functors (i.e. they preserves finite limits, exponentiation and
subobject classifiers).
We finish this Chapter with the following:

Remark 150.
The results in Theorem 148 and Corollary 149 above suggest that consider only the
positive members of SA and develop set-theoretical notions and define a category through
the definition of some notion of ”function” as certain members of P os(SA), is ”too good”
to create new phenomena. Obviously the same situation is obtained when we depart from
the class N eg(SA) of all negative members of SA and the negative functions between
J− J+
them (Cat− (SA) ' Cat(V ) ' Cat+ (SA)). Thus we need more general classes of objects
and functions as certain members of SA to obtain proper expansions of Set = Cat(V ).
The inclusion relation in SA defined by hA, Bi⊆∗ hA0 , B 0 i iff A⊆A0 and B ⊇ B 0
suggests consider a new category, denoted Cat(SA), whose objects are all members of SA
f
and whose arrows are pairs of ordinary functions – f = (fl , fr ) : u →∗ v, where Lu →l Lv
fr
and Ru ← Rv , u, v ∈ SA – endowed with coordinatewise composition and identities: thus
if hA, Bi⊆∗ hA0 , B 0 i, then there is a ”inclusion” morphism i = (il , ir ) : hA, Bi →∗ hA0 , B 0 i.
In this way, obviously Cat(SA) Set × Setop is identified with a full subcategory of
the product category. Under the categorial perspective, this eventually provides a nice
”set-theory”, but it lacks a stronger connection with the global arrangement of members
of SA as the free SUR-algebra.
A more natural, and potentially useful approach, is try to develop natural set-
theoretical definitions dealing directly with the fundamental ”incidence” relation between
members of SA: this is a ternary relation in SA given by m(a, u, b) iff a ∈ Lu and b ∈ Ru
(iff l(a, u) and r(u, b)), a, u, b ∈ SA. This seems to keep good potential: for instance, we
can consider general immediate successor and immediate predecessor of any member of
SA: if u = hLu , Ru i ∈ SA, then suc(u) := hLu ∪ {u}, Ru i and pred(u) := hLu , {u} ∪ Ru i.
Of course, there is no problem in the form of the (faithful) encoding of sets (x ∈ V ) as
”standard members” of SA, i.e as hereditary positive members of SA (J(x) = h{J(y) :
y ∈ x}, ∅i), the point is how to present a more general but still (set-theoretical) notion of
”function” between general members of SA that faithfully encodes the ”standard func-
tions” between standard members, but possibly allows non-standard functions between
some standard members of SA.
The task of define a category, Cat(SA), in this expanded setting (i.e. whose class of
objects is SA) through the definition of some notion of ”function” as certain members
of SA, and still obtaining nice categorial-theoretical properties (topos?), seems to be
an interesting challenge that we will intend to address in the future. We must also be
able to define a functor J : Cat(V ) → Cat(SA) that is injective on objects and arrows
(to represent the expansion process), that will preserves/reflect nice categorial notions
(logical functor?) but that is not an equivalence of categories.

Chapter 5
Conclusions and future works
The present thesis is essentially a collection of elementary results where we develop, from
scratch, a new (we hope!) and complementary aspect of the Surreal Number Theory.
There is much work to be done: it is clear for us that we just gave the first steps in
the Surreal Algebras Theory and in Set theory based on Surreal Algebras.
In the sequel, we briefly present a (non-exhaustive) list of questions that have occurred
to us during the elaboration of this thesis, that we can not be able to deal in the present
work by lack of time and/or of skills, but that we intend to address in the future.
Questions on Chapter 2:
We have described some general constructions in categories of partial SUR-algebra
(with at least 2 kinds of morphisms): initial object, non-empty products, substructures
and some kinds of directed inductive (co)limits. There are other general constructions
available in these categories like quotients and coproducts? A preliminary analysis was
made and indicates that the characterizations of the conditions where such constructions
exists is a non trivial task.
A specific construction like the (functor) cut surreal for SUR-algebras and its partial
version turns out to be very useful to the development of the results of the (partial) SUR-
algebra theory: the situation is, in some sense, parallel to the specific construction of
rings of fractions construction in Commutative Algebra and Algebraic Geometry. There
are other natural and nice specific constructions of (partial) SUR-algebra that, at least,
provide new classes of examples?
We have provided, by categorial methods, some universal results that characterizes
the SUR-algebras SA and ST , and also some relative versions with base (”urelements”)
SA(I), ST (I 0 ) where I, I 0 are partial SUR-algebra satisfying a few constraints. There is
an analog result satisfied by the SUR-algebra N o? There are some natural expansions
109
110 CHAPTER 5. CONCLUSIONS AND FUTURE WORKS
of N o by convenient I 00 are partial SUR-algebra, N o(I 00 ), that also satisfies a universal

property that characterizes it up to a unique isomorphism?
We have provided axiomatizations of SA (respec. ST ) as an anchored SUR-algebra
onto On and V (respec. onto On), in analogous fashion to the axiomatization of N o
through its ”birthday” function b : N o On. It is possible provide anchors, onto
convenient rooted well-founded classes, of the relative versions SA(I), ST (I 0 ) such that
they provide axiomatizations of these relative constructions?
We have seen that the canonical SUR-algebra morphism SA → ST is injective, in
particular it has small fibers. On the other hand, the canonical ZF-algebra morphism
ρ : V → On (the rank function) has small non-empty fibers, in particular it is surjective
and On ,→ V is a section of ρ. It is natural to ask if the canonical SUR-algebra morphism
SA → N o share some of these properties of ρ: it has small fibers? are all its fibers
non-empty (= surjectivity)? it has a section (N o, <) → (SA, <)? the space of signs
representation of SA and N o can aid us to answer some of these questions? The same
kind of questions can also be posed for the canonical SUR-algebra morphism ST → N o.
In the process of associated a subclass of ”hereditary positive members” for some
anchored SUR-algebras (S, b) we consider naturally a structure given by class endowed
by two binary and acyclic relations (S, <, ≺b ). This seems related to the notions of
s-hierarchical structures considered by P. Ehrlich in [Ehr01]: they are some algebraic
structures (group, field, etc) defined over a lexicographically ordered binary tree, (S, <
, ≺s ). Can be illuminating to establish (and explore) a precise relation between both
notions.
There is some clues obtained from some of our proofs (see also [All87] and [Ehr01])
that could be useful expand/adapt the concept of anchor for partial SUR-algebras and
its relatives morphisms.
We have established links, in both directions, between SUR-algebras and ZF-algebras:
we have anchored some SUR-algebras S onto standard ZF-algebras (its hereditary pos-
itive part, HP (S)) and for some standard ZF-algebras Z we have constructed a SUR-
j
algebra (its space of signs, Sig(Z)). Moreover, the pair of functions On N o and
b
j∗
V SA are ”chimera”-morphisms: they have head in a category and tail in another
b∗
category. Notions of morphisms like that are considered in [Ell07], under the name of
”heteromorphisms”: ideally they occur when it is available of a pair of adjoint func-
tors between the categories in sight. It is natural question determine subcategories of
(anchored) SUR-algebras and ZF-algebras such that the mappings S 7→ HP (S) and
Z 7→ Sig(Z) can be extended to adjoint pair of functors - this could be useful for obtain
111
general results of transfer/preservation of properties from one category to another. More-

over, it will be interesting to study the relations of (relatively) free SUR-algebras and
(relatively) free ZF-algebras whenever they belong to such nice subcategories of SU R
and ZF .
We saw that the free/initial SUR-algebra SA is, in many senses, an expansion of
the free/initial ZF-algebra V ans its underlying set theory. Relative constructions are
available for SUR-algebras and for ZF-algebras (see [JM95]). In particular, it can be
interesting examine possible natural expansions of set theories:
(i) with urelements B, V (B), to some convenient relatively free SUR-algebra SA(B̂);
(ii) obtained from the free transitive SUR-algebra ST → N o
We have developed a ”positive set theory” on SA ranked on N o, that expands the
(ZF-algebra) relation V → On through the ”positive” map j + : V → SA, j + (X) =
hj + [X], ∅i, X ∈ V . Naturally we can obtain a ”mirrored” set theory in SA, developed
from the ”negative” map j − : V → SA, j − (X) = h∅, j − [X]i, X ∈ V . There are other
natural and interesting ”mixed” set theories available from the free SUR-algebra support
SA?
A combination of the tree lines of research above mentioned can be a interesting
(”second-order”) task: it will be a line of development of general relative set theories
that are base independent.
Unexplored possibilities:
There exists at least two major aspects of the theory of SUR-algebras that we have
not addressed in this work:
• the analysis of its model-theoretic aspects;
• the consideration of possible applications of SUR-algebras into ”traditional” set/class
theory, to answer specific questions on ZFC/NBG theories.
It is worthy to note that two lines of research can present interesting cross feedings,
as the considerations below will indicate.
First of all, we recall that:
- rational number line (Q, <) is a (or ”the”) countable dense totally ordered set without
endpoints;
- a dense totally ordered set without endpoints is a ηα -set if and only if it is ℵα -saturated
structure, α ∈ On;
- the the surreal number line, (N o, <), is for proper class linear orders what the ratio-
nal number line (Q, <) is for the countable linear orders. In fact, (N o, <) is a proper
class Fraı̈ssé limit of the class of all finite linear orders. The surreal numbers are set-
homogeneous and universal for all proper class linear orders.

- the relational structure (S, <) underlying a SUR-algebra S is acyclic and η∞ , a natural
generalization of the properties above mentioned.
We consider below two remarkable instances of model-theoretic properties applied to
set theory that, we believe, could be related to our setting:
(I) J. Hamkins have defined in [Ham13] the notion of ”hypnagogic digraph”, (Hg, *),
an acyclic digraph graded on (N o, <)1 . The hypnagogic digraph is a proper-class ana-
logue the countable random Q-graded digraph: it is the Fraı̈ssé limit of the class of all
finite N o-graded digraphs. It is simply the On-saturated N o-graded class digraph, mak-
ing it set-homogeneous and universal for all class acyclic digraphs.
Hamkins have applied this structure, and some relativized versions, to prove interesting
results concerning models of ZF set theory. For instance:
• every countable model of set theory (M, ∈M ), is isomorphic to a submodel of its own
constructible universe (LM , ∈M );
• the class of countable models of ZFC is linearly pre-ordered by the elementary embed-
ding relation.
As a part of a program of model theoretic studies of (relatively free) SUR-algebras,
seems natural to determine (and explore) a precise relation between the N o-ranked
relational classes (Hg, *) and (SA, <) (or (ST, <)). And what about the relativized
versions of Hg and SA (or ST )? This kind of question is very natural as part of an
interesting general investigating program relating Model Theory and Category Theory:
in one hand we have the model-theoretic universality (from inside or above) of Hg
and, on the other hand, we have the category-theoretic universality of (relatively) free
constructions (to outside or below) of SA and ST .
Can we construct new models of ZF(C) by establishing relations
[Cat(SA)] [Hamkins digraph models] (and/or some variants)
in a way in some sense analogous to the relation:
[sheaves over boolean algebras] [Cohen f orcing models]?
(II) J. Hirschfeld have provided in [Hir75] a list of axioms - that include axioms for
∈-acyclicness and for ∈-density - that describes the model companion of ZF set theory.
He emphasizes in the page 369:
”...This model companion, however, resembles more a theory of order (Theorem 3) than a set
theory, and therefore, while supplying an interesting example for model theory it does not shed
any new light on set theory...”
We can wonder about the possible relations of our SUR-set theories and model the-
oretic (Robinson) forcing(s). This is a natural question since the models of the model
companion of ZF have a ”nice” relational structure and the model theoretic forcing can
1
I.e., it is given a ”rank” function v : Hg → N o such that: for each x, y ∈ Hg, if x * y, then
v(x) < v(y).
113
provide the description of model companion/completion of a first order theory. Consid-

erations involving large infinitary languages are also been in sight, since SUR-algebras
are η∞ acyclic relational classes.
Bibliography
[All87] Norman L. Alling, Foundations of analysis over surreal number fields, North-Holland Math-
ematics Studies, vol. 141, North-Holland Publishing Co., Amsterdam, 1987. Notas de
Matemática [Mathematical Notes], 117.
[Bor94] Francis Borceux, Handbook of Categorical Algebra vol.3: Categories of Sheaves, Encyclopedia
of Mathematics and its Applications, vol. 52, Cambridge University Press, Cambridge, 1994.
[BW85] M. Barr and C. Wells, Toposes, Triples and Theories, Grundlehren der Mathematischen
Wissenschaften, 278, Springer-Verlag, Berlin, 1985.
[Acz88] Peter Aczel, Non-well-founded sets, CSLI Lecture Notes, vol. 14, Stanford University, Center
for the Study of Language and Information, Stanford, CA, 1988. With a foreword by Jon
Barwise [K. Jon Barwise].
[Con01] J. H. Conway, On numbers and games, 2nd ed., A K Peters, Ltd., Natick, MA, 2001.
[Gon86] Harry Gonshor, An introduction to the theory of surreal numbers, London Mathematical So-
ciety Lecture Note Series, vol. 110, Cambridge University Press, Cambridge, 1986.
[JM95] A. Joyal and I. Moerdijk, Algebraic set theory, London Mathematical Society Lecture Note
Series, vol. 220, Cambridge University Press, Cambridge, 1995.
[Jec03] Thomas Jech, Set Theory, Springer Monographs in Mathematics, Springer-Verlag, Berlin,
2003.
[Kun13] Kenneth Kunen, Set Theory, Studies in Logic, vol. 34, College Publications, London, 2013.
[All62] Norman L. Alling, On the existence of real-closed fields that are ηα -sets of power ℵα , Trans.
Amer. Math. Soc. 103 (1962), 341–352. MR0146089 (26 #3615)
[AE86a] Norman L. Alling and Philip Ehrlich, An abstract characterization of a full class of surreal
numbers, C. R. Math. Rep. Acad. Sci. Canada 8 (1986), no. 5, 303–308. MR859431 (87j:04008)
[AE86b] , An alternative construction of Conway’s surreal numbers, C. R. Math. Rep. Acad.
Sci. Canada 8 (1986), no. 4, 241–246. MR850107 (87j:04007)
[vdDE01] Lou van den Dries and Philip Ehrlich, Fields of surreal numbers and exponentiation, Fund.
Math. 167 (2001), no. 2, 173–188.
[Ehr88] Philip Ehrlich, An alternative construction of Conway’s ordered field No, Algebra Universalis
25 (1988), no. 1, 7–16, DOI 10.1007/BF01229956. MR934997 (89d:04004a)
[Ehr94] , All numbers great and small, Real numbers, generalizations of the reals, and theo-
ries of continua, Synthese Lib., vol. 242, Kluwer Acad. Publ., Dordrecht, 1994, pp. 239–258.
MR1340465
115
116 BIBLIOGRAPHY
[Ehr01] , Number systems with simplicity hierarchies: a generalization of Conway’s theory of

surreal numbers, J. Symbolic Logic 66 (2001), no. 3, 1231–1258.
[Ehr11] , Conway names, the simplicity hierarchy and the surreal number tree, J. Log. Anal.
3 (2011), Paper 1, 26, DOI 10.4115/jla.2011.3.11. MR2769328 (2012i:06002)
[Ehr12] , The absolute arithmetic continuum and the unification of all numbers great and small,
Bull. Symbolic Logic 18 (2012), no. 1, 1–45. MR2798267
[Ell07] David Ellerman, Adjoint Functors and Heteromorphisms, ArXiv.org 0704.2207 (2007), 28.
[Fre16] Rodrigo de Alvarenga Freire, Grasping Sets Through Ordinals: On a Weak Form of the Con-
structibility Axiom, South American Journal of Logic 2 (2016), no. 2, 347–359.
[Ham13] Joel David Hamkins, Every countable model of set theory embeds into its own constructible
universe, J. Math. Log. 13 (2013), no. 2, 1350006, 27.
[Hir75] Joram Hirschfeld, The Model Companion of ZF, Proceedings of the American Mathematical
Society 50 (1975), no. 1, 369–374.
[KM15] Salma Kuhlmann and Mickaël Matusinski, The exponential-logarithmic equivalence classes of
surreal numbers, Order 32 (2015), no. 1, 53–68.
[LE09] E. G. K. López-Escobar, Logic and mathematics: propositional calculus with only three prim-
itive terms, The many sides of logic, Stud. Log. (Lond.), vol. 21, Coll. Publ., London, 2009,
pp. 153–170.
[Lur98] Jacob Lurie, The effective content of surreal algebra, J. Symbolic Logic 63 (1998), no. 2,
337–371.
[Mac39] Saunders MacLane, The universality of formal power series fields, Bull. Amer. Math. Soc. 45
(1939), 888–890.
[Sco69] Dana Scott, On completing ordered fields, Applications of Model Theory to Algebra, Analysis,
and Probability (Internat. Sympos., Pasadena, Calif., 1967), Holt, Rinehart and Winston,
New York, 1969, pp. 274–278.

Tese Dimi 2018 09 14

Uploaded by

Copyright:

Available Formats

Tese Dimi 2018 09 14

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tese Dimi 2018 09 14

Uploaded by

Copyright:

Available Formats

An algebraic framework to

a theory of sets based on

Dimi Rocha Rangel

Programa: Doutorado em Matemática do IME-USP

Durante o desenvolvimento deste trabalho o autor recebeu auxı́lio financeiro da CAPES

Esta versão da tese contém as correções e alterações sugeridas

• Prof. Dr. Hugo Luiz Mariano (Presidente) - IME-USP

RANGEL D.R. Um contexto algébrico para uma teoria de conjuntos baseada

Palavras-chave: Números surreais; Teoria Algébrica dos Conjuntos; SUR-álgebras.

RANGEL D.R. An algebraic framework to a theory of sets based on the sur-

Keywords: Surreal numbers; Algebraic Set Theory; SUR-algebras.

2 Introducing Surreal Algebras 27

2.2.4 The cut surreal algebra . . . . . . . . . . . . . . . . . . . . . . . . 39

3 SUR-algebras and ZF-algebras 61

4 Set Theory in the free Surreal Algebra 93

4.3 Cat+ (SA) and the functor induced by j + : V → SA . . . . . . . . . . . . 101

5 Conclusions and future works 109

(II) Surreal Numbers and Set Theory

Chronology of the work:

Overview of this thesis:

1.1 Set theoretic backgrounds

A pre-order is a reflexive and transitive relation. A partial order is a antisymmetric

3. Induced binary relations:

1.1.2 Well founded relations

is an well-founded relation, since for each x, y ∈ X, x< ⊆y < iff x ≤ y.

1.1.3 Cuts and densities

1.2 A glance on Algebraic Set Theory

1.2.1 Set theory and ZF-algebras

improper classes are the ”small” objects6 .

8. Small functions in NBG:

9. ZF-algebras and its morphisms:

It is clear that, between the inclusion maps On ,→ T r, On ,→ V, T r ,→ V, only the

1.2.2 Standard ZF-algebras

(d2) Consider the following conditions on z ∈ A:

18. Examples and counter-examples:

1.3 On Surreal Numbers

1.3.1.1 The Conway’s construction formalized in NBG

(i) G0 = {h∅, ∅i}

x 6 y iff no xL satisfies xL > y and no y R satisfies x > y R ,

Proposition 19.With this operations, N o is a real-closed Field. In addition, every (set)

1.3.1.2 The Cuesta-Dutari cuts construction

χ(T ) = (T ∪ CD(T ), <0 ),

with <0 defined as follows:

(i) If x, y ∈ T , then the order is the same as in T ;

(ii) If x = (L, R), y = (L0 , R0 ) ∈ CD(T ), then x <0 y if L ( L0 ;

(iii) If x ∈ T and y = (L, R) ∈ CD(T ), then x <0 y if x ∈ L or y <0 x if x ∈ R.

And finally we have [

1.3.1.3 The binary tree construction or the space of signs construction

Consider the class Σ = {f : α → {−, +} : α ∈ On}. We can define in this class an

1.3.2 The axiomatic approach

(ii) b : S → On is a surjective function;

Remark 21. Let S = (S, <, b) be a full class of surreal numbers.

As mentioned in section 3 of chapter 4 in [All87], by results proven in Conway’s book

Remark 26. Let S = (S, <, b) be a full class of surreal numbers.

Definition 28.j(α) = {j[α]|∅}, α ∈ On.

Proposition 29.[[Gon86], p. 41] b ◦ j = idOn

Proposition 30.α < β iff j(α) < j(β), ∀α, β ∈ On.

˙ 0, 1) has an isomorphic copy in N o given by

Introducing Surreal Algebras

• <α =<SAα ×SAα

• Cst (SAα ) = Cs (SA(α) ) and tα = t Cs (SA(α) )