Anda di halaman 1dari 141

Principia, Vol.

13 (3) on-line

an international journal of



This issue is also available in CD-ROM (ask yours here - see

VOLUME 13 conditions ) and hard copies (ask yours here - see conditions ).
Principia's editorial information (PDF file)

Papers (PDF files):

© NEL -
Epistemology Quantifiers and the Foundations of Quasi-Set Theory
and Logic Jonas R. Becker Arenhart & Décio Krause
Research Group.
Federal Models of Reduction
University of
Santa Catarina, Otávio Bueno
All rights Meeting Hintikka's Challenge to Paraconsistentism
reserved. Walter Carnielli

A Paraconsistentist Approach to Chisholm's Paradox

Marcelo Esteban Coniglio & Newton Marques Peron

Fregean One-to-one Correspondence and Numbers as

Object Properties
Boris Grozdanoff

A Refined Geometry of Logic

David Miller

El Significado de la Negación Paraconsistente

Gladys Palau & Cecilia Duran

O Naturalismo como Atitude: Mach em Disputa com a

Antonio Augusto Passos Videira

file:///C|/Cella/Rascunhos/olvol133.html (1 de 2) [06/08/2010 16:45:20]

Principia, Vol.13 (3) on-line

file:///C|/Cella/Rascunhos/olvol133.html (2 de 2) [06/08/2010 16:45:20]

revista internacional de epistemologia

NEL – Núcleo de Epistemologia e Lógica, UFSC

Reitor Alvaro Toubes Prata
Vice-Reitor Carlos Alberto Justo da Silva


Coordenador Cezar Augusto Mortari
Sub-Coordenador Luiz Henrique de Araújo Dutra

Os artigos publicados em Principia são indexados por Philosopher’s Index.

Papers published in Principia are indexed by the Philosopher’s Index.

Principia está disponível em edição impressa (ISSN 1414-4247),

em edição on-line:∼principi (ISSN 1808-1711),
e em edição em CD-ROM (ISSN 1808-1525).

Principia is available in hard copies (ISSN 1414-4247),

on-line:∼principi (ISSN 1808-1711),
and in CD-ROM edition (ISSN 1808-1525).

(Catalogação na fonte pela Biblioteca Universitária da

Universidade Federal de Santa Catarina)

Principia : revista internacional de epistemologia /

Universidade Federal de Santa Catarina. Núcleo
de Epistemologia e Lógica. – v. 1, n. 1 (jun.
1997)-. --Florianópolis : Editora da UFSC,
1997- v. ; 26 cm

ISSN 1414-4247

I. Universidade Federal de Santa Catarina. Núcleo

de Epistemologia e Lógica.
revista internacional de epistemologia

Special Issue
in honor of


on the occasion of his 80th birthday

(Part II)

editor of the special issue


ISSN 1414-4247

Principia | Florianópolis | v. 13 | n. 3 | p. 251–384 | dez. 2009

revista internacional de epistemologia

Editor Responsável/Editor:
Luiz Henrique de Araújo Dutra
Editor Assistente/Associate Editor:
Cezar Augusto Mortari

Comissão Editorial/Editorial Board:

Sara Albieri Gustavo Andrés Caponi

Alberto Oscar Cupani Luiz Henrique de Araújo Dutra
Marco Antônio Franciotti Cezar Augusto Mortari

Corpo Editorial/Advisory Editorial Board:

Harvey Brown (Wolfson College, Oxford, UK)
Mario Bunge (McGill University, Canada)
Oswaldo Chateaubriand Filho (PUC-Rio de Janeiro, Brazil)
Newton C. A. da Costa (University of São Paulo, Brazil)
Dagfinn Føllesdal (University of Oslo, Norway)
Bas C. van Fraassen (Princeton University, USA)
Jean Gayon (University of Paris I–Panthéon Sorbonne, France)
Michel Ghins (Catholic University of Louvain, Belgium)
Alvin I. Goldman (Rutgers University, USA)
Gilles-Gaston Granger (Collège de France, France)
Susan Haack (University of Miami, USA)
Ted Honderich (University of London, UK)
Hugh Lacey (Swarthmore College, USA)
João Paulo G. Monteiro (University of Lisbon, Portugal)
Oscar Nudler (Bariloche Foundation, Argentina)
Ezequiel de Olaso (†)(San Andrés University, Argentina)
Michel Paty (University of Paris 7–Denis Diderot, France)
Eduardo Rabossi (†)(University of Buenos Aires, Argentina)
Roberto Torretti (University of Puerto Rico, USA)
Daniel Vanderveken (University of Trois Rivières, Canada)
Andrew Woodfield (University of Bristol, UK)
revista internacional de epistemologia
















ISSN 1414-4247

Principia | Florianópolis | v. 13 | n. 3 | p. 251–384 | dez. 2009

Núcleo de Epistemologia e Lógica
Universidade Federal de Santa Catarina∼nel
(48) 3721-8612 fax: (48) 3721-9751

Principia é uma revista de filosofia que publica artigos em epistemologia, filosofia

da ciência e domínios conexos, aceitando textos em português, espanhol, inglês e

Principia is a philosophical journal that publishes papers on epistemology, philosophy

of science and related areas. It accepts texts in Portuguese, Spanish, English and French.

Correspondência editorial, assinaturas e números publicados:

Editorial Correspondence, subscriptions and back issues:
Revista Principia/Editor Responsável
Núcleo de Epistemologia e Lógica/CFH/UFSC
Cx. Postal 476
88010-970 Florianópolis, SC, Brasil

Outras informações, inclusive aos autores:

Further information, including to contributors:∼principi

Ou enviar mensagem a: /Or email:

Editoração Eletrônica, Publicação e Distribuição

NEL – Núcleo de Epistemologia e Lógica, UFSC
Federal University of Santa Catarina

Abstract. In this paper we discuss some questions proposed by Prof. Newton da Costa on
the foundations of quasi-set theory. His main doubts concern the possibility of a reasonable
semantical understanding of the theory, mainly due to the fact that identity and difference
do not apply to some entities of the theory’s intended domain of discourse. According to
him, the quantifiers employed in the theory, when understood in the usual way, rely on the
assumption that identity applies to all entities in the domain of discourse. Inspired by his
provocation, we suggest that, using some ideas presented by da Costa himself in his seminars
at UFSC (the Federal University of Santa Catarina) and by one of us (DK) in some papers,
these difficulties can be overcome both on a formal level and on an informal level, showing
how quantification over items for which identity does not make sense can be understood
without presupposing a semantics based on a ‘classical’ set theory.

Keywords: Quasi-sets, semantic of non-classical systems, logic of quantum mechanics.

Dedicated to Prof. Newton da Costa for his 80th birthday

and for his inspiration in so many philosophical questions.

1. Introduction

One of the challenges that appear in the development of non-classical logics is re-
lated to the semantics of such systems. Generally speaking, it is recognized that
a semantics conduced according to the standard methods employ as metalanguage
some one of the ‘classical’ set theories, such as ZF (the Zermelo-Fraenkel set theory,
generally encompassing the axiom of choice). Such a semantics comprises some im-
mediate difficulties on the foundational side so that, from a philosophical point of
view, it would seems to be not completely absent of a priori commitments (given by
such meta-framework). Saying in brief, since standard ZF is itself based on classi-
cal logic, it seems to be already committed to the very principles of this logic, the
same ones whose avoidance was the main driving force behind the proposal of a
non-classical system.1 Thus, instead of liberating us from the classical principles,
a semantics to a non-classical system, if grounded in ZF (and the standard mathe-
matical frameworks in general), commits us again to classical logic, viz., that one

Principia 13(3): 251–68 (2009).

Published by NEL — Epistemology and Logic Research Group, Federal University of Santa Catarina (UFSC), Brazil.
252 Jonas R. Becker Arenhart & Décio Krause

assumed in the metalanguage. Alternatively, in a homier way, as put by da Costa,

Béziau, and Bueno, “[o]ne reintroduces, so to speak, by the backdoors, exactly what
was intended to be left on the entrance!” (da Costa et al. 1995, 44).2
A specific case in which this problem can be clearly seen is related to the consid-
eration of non-reflexive logics (see da Costa & Krause 1994, 1997; French & Krause
2006, chaps.7 and 8).3 One of the main motivations for the use of these systems
come from the supposition that some items in the intended domain of discourse lack
identity conditions. For instance, it is sometimes assumed that non-relativistic quan-
tum mechanics can be seen as providing us a world with such kind of entities. If
we assume a metaphysical hypothesis like this one, it seems that, at least from a
philosophical point of view, we should have an adequate non-classical logic to deal
with these entities (‘entities without identity’, which may be indistinguishable but
without turning to be the same object–identical). Based on such motivations, a par-
ticular type of non-reflexive set theory was developed, strongly enough to encompass
a mathematics suitable for physics, termed quasi-set theory. This theory aims at to
pursue seriously the motto inherited from H. Post (Post 1963), namely, that the indis-
cernibility (or indistinguishability) of quantum objects should be attributed to these
entities right from the start, so as in recognizing (following Schrödinger) that neither
identity nor difference would make sense to some entities (namely, the ‘elementary
particles’).4 Usually, the way of dealing with indiscernible things within ‘standard’
frameworks is by assuming certain symmetry conditions, which mask the logical fact
that every entity that obeys classical logic is, in a certain sense, an individual; for a
detailed discussion, see French & Krause 2006. These collections, called quasi-sets,
are so that, despite the impossibility of identification of its elements, they do have a
cardinal number, which intuitively speaking expresses how many elements there is
in the considered quasi-set.5 Furthermore, the theory encompasses a weaker rela-
tion of indistinguishability that holds among items of the same type, which enables
us to express that elements may be indistinguishable (indiscernible), although not
identical (see the following sections for a sketch of quasi-set theory).
In his seminars at UFSC and through particular conversations, da Costa has ex-
pressed doubts about the possibility of explaining how quantifiers are used in quasi-
set theory for, in principle, quantification would make sense only when the concept
of identity can be applied to the considered objects. According to da Costa, a non-
classical logic, in order to be a legitimate rival of classical logic, would encompass
a semantics which conforms to its basic assumptions, that is, a semantics in which
the main motivations of the considered logic are maintained (see da Costa et al.
1995). So, he claims, if one cannot make sense of quantifiers without presupposing
identity, then one cannot claim that a logic without identity is a suitable system of
logic for dealing, for instance, with indistinguishable elementary particles, typical of
non-relativistic quantum mechanics.

Principia 13(3): 251–68 (2009).

Quantifiers and the Foundations of Quasi-set Theory 253

In what concerns quasi-set theory, since its language involves quantification,

then, in accordance with such ideas, in order to make statements which intuitively
mean, say, ‘every element of the quasi-set x is so and so’, or ‘there is an element in
x which is so and so’, we should be able to explain how do we grasp the meaning of
such sentences without presupposing that identity is valid for the items under quan-
tification. In particular, if we want to show that we can really do that, we must do
that without implicitly assuming that our metalanguage is a classical set theory like
ZF, even though it is kept on an informal level only. The problem, then, amounts in
finding an explanation of the semantical basis for the quasi-set theory’s underlying
logic. Certainly, from these discussions it follows that the underlying logic of Q is
not classical logic, whose semantics is a classical one, and it must be some kind of
non-reflexive logic, in order to preserve our motivations.
For some, it would seem that a classical metalanguage is necessary to make
intelligible the usual reading of the quantifiers, for according to this view, such a
reading is not possible without an overall validity of the concept of identity. They
could argue that when we say, for example, ‘there is some x such that x is P’, this
should be taken to mean ‘there is a set that is the extension of P, to which the
denotation of x belongs’, leading us back to set theory. If this kind of reasoning
was correct, then any reasonable meaning of the quantifiers would be tied inevitably
to identity and, then, by employing quantifiers in quasi-set theory, we would re-
introduce identity for the items under discussion through the metalanguage, causing
us to have to abandon the very project we began with. To deal properly with these
questions, we should state adequately the underlying logic and the metalanguage
we are using to study quasi-set theory. Our aim here is to show that a reasonable
understanding of the quantifiers exist, so that we can avoid the risk of re-introducing
identity for every object in the domain of the discourse.
Our procedure will be based in the classical case, really, it will be a generaliza-
tion of it. So, let us take a brief look at those parts of the usual approach which
are relevant for our purposes. Consider, for example, quantification in ZF. How do
we make clear and mathematically precise our intuitive understanding of the quan-
tifiers used by the underlying logic in this case? We rely, in general, on the Tarskian
semantics for first order languages, assuming that the underlying logic of ZF is (say)
classical first order logic with identity. This semantics, it is claimed, gives us a math-
ematization of the intuitive meaning of the quantifiers. First-order logic, besides its
axioms, has its specific semantics, and both contribute to make clear the meaning of
the logical constants, including the quantifiers.
Now, Tarskian semantics is generally stated using informal set theory, and if
pressed to make our assumptions clear, we say that we are relying on ZF, the same ZF
which is being taken here as our case study, whose underlying logic is (by hypothe-
sis) exactly the first-order logic whose semantical understanding is in question! This

Principia 13(3): 251–68 (2009).

254 Jonas R. Becker Arenhart & Décio Krause

is not a case of circularity, as it becomes clear when we make the simple distinction
between the meta-theory in which we work and the object theory we are interested
in. In this case, both are taken to be ZF. This case is an example of the ‘demand
of coherence’ imposed generally on non-classical logics, as stated above, that the
metalanguage should not commit us with principles deviant from those granted by
the very logic under study. So, ZF is used to state and clarify the meaning of logical
constants for the system of logic which is its own underlying logic, as we shall see
below. One possible way to go out of such an assumption, if there is one, seems to
suppose that the syntax of the formal system alone is able to clarify the meaning of
the terms of the language through the axioms and rules of inference (in a strict for-
mal way, as suggested for instance by Haskell B. Curry (Curry 1951). But then, one
must face the problem that the same set of axioms can be used to ground systems of
logic that could be different when semantic considerations entered the discussion,
that is, some systems of logic may differ only in the semantics presented, having the
same formal apparatus when viewed from syntactic point of view alone. We will not
follow this ‘syntax only’ strategy here. On the contrary, we shall be concerned with
semantical issues, for they are of course in the core of the discussions regarding the
ontological commitments of theories.
Thus, our strategy will be the same one used in classical logic. The underlying
logic of Q will have its semantic based in quasi-set theory itself, giving the adequate
meaning to its logical constants. We first recover very briefly how this is done for
classical first order logic, roughly speaking, by mapping the language in set theory
and then with this theory (ZF) as the background language, it is possible to design
a formal semantics for the language of first order logic. Obviously, our claim is that
the choice of the metalanguage always carries some commitments that drive the
semantics we are considering. This shows that we can in fact study the axiomatics
of first-order ZF within ZF proper, although we cannot prove that there is a model
of this theory within ZF (due to the second Gödel’s incompleteness theorem). In our
specific case, we show how to do such a thing for the language of the underlying
logic of the object theory Q, this time translating it to our metatheory Q, where
the adequate semantic meaning of the logical symbols can be done yet, we insist,
we cannot prove that there is a model of Q within Q (supposed consistent). Any
charge of circularity here would be rightly directed to the classical case, for the
procedure is really a generalization of that case, and we take as an easy matter to
avoid the problem of an adequate distinction between the object language and the

2. The ‘classical’ case

To make our paper self-contained, we recall briefly in this section the procedure de-
scribed in Ebbinghaus et al. 1994, 113–4). There, it is shown how a system of first
Principia 13(3): 251–68 (2009).
Quantifiers and the Foundations of Quasi-set Theory 255

order classical logic can be developed inside ZF, so that both its semantics and syn-
tax are made using a classical metalanguage (namely, that of ZF proper). Besides
an enumerable quantity of variables x 1 , x 2 , . . . , x n . . ., we take as logical symbols
the connectives ¬ and →, the universal quantifier ∀, and the equality sign =. For
simplicity we restrict ourselves to a language L with predicate symbols P 1, P 2, . . . as
non-logical constants only, where P n denotes an n-ary predicate symbol, and differ-
ent predicates of the same weight can be distinguished by a sub-index, but we won’t
consider them here.
We begin by showing how the symbols of the language can be seen as certain
sets. The variables are identified with the elements of ω, vis., the set of natural
numbers within ZF, following von Neumann’s standard definition. So, we assume,
for example, that the variable x n is identified with the natural number n. The other
logical symbols are defined as follows: ¬ is associated to the ordered pair 〈0, 0〉, →
to the pair 〈0, 1〉, ∀ is identified with 〈0, 2〉, while = with 〈0, 3〉. For the non logical
symbols, we represent each predicate P n with n 6= 0 by the pair 〈1, n〉.
Now, let us consider the case of atomic formulas. In our language, they come in
two versions: in the first case, they are formulas of the form x n = x m , which can be
seen as ordered triples 〈n, =, m〉, with n, m ∈ ω. All the atomic formulas of this type
constitute the set At = which by definition is the set ω × {=}× ω, where ‘×’ denotes
the cartesian product of sets. The second case is constituted by formulas of the form
P n x i0 . . . x in−1 , which are taken to be pairs such as 〈P n , f 〉, where f is a function from
n (which is a set) to ω, so that, for example, if f is such that f (0) = 2 and f (1) = 6,
then 〈P 2 , f 〉 is the formula P 2 x 2 x 6 . The set of all such formulas, At P is by definition
the set {〈P n , f 〉 : n ∈ ω and f : n 7→ ω}.
Thus, the set F of formulas of L is the smallest set satisfying the following con-

1. At = ∪ At P ∈ F;

2. if α ∈ F, then 〈¬, α〉 ∈ F;

3. if α, β ∈ F, then 〈α, →, β〉 ∈ F;

4. if n ∈ ω and α ∈ F, then 〈∀, n, α〉 ∈ F.

Now, following the usual methods (Ebbinghaus et al., loc.cit.), one can develop
a formal semantics for this language. Since this is supposed to be well known, we
will not recall the details here. Our point is to emphasize the general technique, for
later we want to enlighten the similarities between this ‘classical’ case and quasi-set
theory. What really matters is that, since the theory used to establish this standard
semantics is classical set theory, there are some features of the semantics due to this
assumption that cannot be overlooked from a philosophical and foundational point

Principia 13(3): 251–68 (2009).

256 Jonas R. Becker Arenhart & Décio Krause

of view. The crucial assumption, in this case, at least in what concerns our purposes,
is that identity makes sense to every entity in the domain of quantification.

3. Outlining quasi-set theory

In this section, we give an outline of the most basic notions of quasi-set theory
that play an important role in the discussion that follows (for further details, see
French & Krause 2006, chap.7). This is not the place to present all the postulates
and definitions of the theory, so we just revise the main ideas and results which are
important for what follows. Intuitively speaking, a quasi-set is a collection of objects
such that some of them may be indistinguishable without turning out to be identical.
Of course this is not a strict ‘definition’ of a quasi-set, acting more or less as Cantor’s
‘definition’ of a set as “any collection into a whole M of definite and separate [that
is, distinguishable] objects m of our intuition or our thought" (see the discussion
in French & Krause 2006, §6.4), serving just to provide an intuitive account of the
The quasi-set theory Q has in its main motivations some considerations taken
from quantum physics, mainly in considering Schrödinger’s idea that the concept of
identity do not make sense when applied to elementary particles (Schrödinger 1952,
17–8; as remarked earlier, he considered just non-relativistic quantum mechanics).
Another motivation is (in our opinion) the need, stemming from philosophical wor-
ries, of dealing with collections of absolutely indistinguishable items that need not
be the same ones.6 Of course, viewed from a formal point of view, Q can also be de-
veloped independently of any intended interpretation, but here we shall always keep
in mind this ‘quantum’ motivation since, after all, it is the intended interpretation
that has originated the problem of the development of the theory.
The first point is to guarantee that identity and indistinguishability (or indiscerni-
bility) will nor collapse into one another when the theory is formally developed. We
assume that identity (symbolized by ‘=’) is not a primitive relation, but the theory
has a weaker concept of indistinguishability (symbolized by ‘≡’) instead. This is just
an equivalence relation and holds among all objects of the considered domain. If the
domain is divided up into objects of two kinds, the m-objects (standing for ‘micro-
objects’) and M -objects (for ‘macro-objects’) and quasi-sets of them (and probably
having other quasi-sets as elements as well), then identity (defined as so that it has
all the properties of standard identity of ZF) can be defined for M -objects and quasi-
sets having no m-objects in their transitive closure (this concept is like the standard
one). Thus, if we take just the part of theory obtained by ruling out the m-objects
and collections (quasi-sets) having them in their transitive closure, we get a copy of
ZFU (ZF with Urelemente); if we further eliminate the M -objects, we get a copy of
the ‘pure’ ZF.

Principia 13(3): 251–68 (2009).

Quantifiers and the Foundations of Quasi-set Theory 257

Technically, expressions like x = y are not always well formed, for they are
not formulas when either x or y denote m-objects. We express that by saying that
the concept of identity does not make sense for all objects (it should be understood
that this is just a way of speech). The objects (the m-objects) to which the defined
concept of identity does not apply are termed non-individuals for historical reasons
(see French & Krause 2006). As a result (from the axioms of the theory), we can
form collections of m-objects which have no identity–in this sense; these collections
may have a cardinal (termed its ‘quasi-cardinal’) but not an associated ordinal. Thus,
the concept of ordinal and of cardinal are independent, as in some formulations of
ZF proper. So, informally speaking, a quasi-set of m-objects is such that its elements
cannot be identified by names, counted, ordered, although there is a sense in saying
that these collections have a cardinal (which cannot be defined by means of ordinals,
as usual—but see below).
It is important to remark that, when Q is used in connection with quantum
physics, the m-objects are thought of as representing quantum entities (henceforth,
q-objects), but they are not necessarily ‘particles’ in the standard sense—associated
with classical physics or even with orthodox quantum mechanics; waves, field exci-
tations (the ‘particles’ in quantum field theory—QFT) or perhaps even strings can be
taken as possible interpretations of the m-objects, although this needs qualification.
Generally speaking, whatever ‘objects’ sharing the property of being indistinguish-
able can also be values of the variables of Q (see Falkenburg 2007, chap.6 for a
survey of the various different meanings that the word ‘particle’ has acquired in con-
nection with quantum physics).
Another important feature of Q is that standard mathematics can be developed
using its resources, for the theory is conceived in such a way that ZFU (and hence
also ZF, perhaps with the axiom of choice, ZFC) is a subtheory of Q. In other words,
the theory is constructed so that it extends standard Zermelo-Fraenkel with Urele-
mente (ZFU); thus standard sets (of ZFU) can be viewed as particular qsets (that is,
there are qsets that have all the properties of the sets of ZFU, and the objects in Q
corresponding to the Urelemente of ZFU are identified with the M -atoms of Q). The
‘sets’ in Q will be called Q-sets, or just sets for short. To make the distinction, the
language of Q encompasses a unary predicate Z such that Z(x) says that x is a set.
It is also possible to show that there is a translation from the language of ZFU into
the language of Q so that the translations of the postulates of ZFU are theorems of
Q; thus, there is a ‘copy’ of ZFU in Q, and we refer to it as the ‘classical’ part of Q.
In this copy, all the usual mathematical concepts can be stated, as for instance, the
concept of ordinal (for the Q-sets). This ‘classical part’ of Q plays an important role
in the formal developments of the next sections.
Furthermore, it should be recalled that the theory is constructed so that the rela-
tion of indiscernibility, when applied to M -atoms or Q-sets, collapses into standard

Principia 13(3): 251–68 (2009).

258 Jonas R. Becker Arenhart & Décio Krause

identity (of ZFU). The Q-sets are qsets whose transitive closure (defined as usual)
does not contain m-atoms (in other words, they are ‘constructed in the classical part
of the theory—see Fig. 1).

B On 

B B B 
B pure qsets B copies of ZFU-sets B copies of ZF-sets 
B B B 
B (sets with M -atoms) B (‘pure’ sets)
B (here, no identity) B B 
B x≡y B a B 
B B c B 
'$ b
B &%
B B B 
B (here we can speak B x = y)
∗ '$ B
∗ ∗ '$
B B 
B B d B 
B &% B f B 
Indiscernible qsets ∗ ∗ e

B B B 
B B &% B 
B &% B B 
B m-atoms M -atoms
B B B 
B B ;

Figure 1: The Quasi-Set Universe Q: On is the class of ordinals, defined in the

classical part of the theory.

In order to distinguish between Q-sets and qsets that may have m-atoms in their
transitive closure, we write (in the metalanguage) {x : ϕ(x)} for the former and
[x : ϕ(x)] for the latter. In Q, we term ‘pure’ those qsets that have only m-objects
as elements (although these elements may be not always indistinguishable from one
another, that is, the theory is consistent with the assumption of the existence of
different kinds of m-atoms–that is, not all of them must be indiscernible from one
another), and to them it is assumed that the usual notion of identity cannot be
applied (that is, let us recall, x = y, as well as its negation, x 6= y, are not well
formed formulas if either x or y stand for m-objects). Notwithstanding, the primitive
relation ≡ applies to them, and it has the properties of an equivalence relation.
The concept of extensional identity, as said above, is a defined notion, and it has
the properties of standard identity of ZFU. More precisely, we write x = E y (read ‘x
and y are extensionally identical’) iff they are both qsets having the same elements
(that is, ∀z(z ∈ x ↔ z ∈ y)) or they are both M -atoms and belong to the same qsets
(that is, ∀z(x ∈ z ↔ y ∈ z)). From now on, we shall not bother to always write = E ,
using simply the symbol “=” for the extensional equality, as we have done above.
Since m-atoms are to stand for entities which cannot be labeled, for they do not
enter in the relation of identity, it is not possible in general to attribute an ordinal

Principia 13(3): 251–68 (2009).

Quantifiers and the Foundations of Quasi-set Theory 259

to collections whose elements are denoted by m-atoms. As a consequence, for these

collections it is not possible to define the notion of cardinal number in the usual
way, that is, through ordinals.7 In the version of the theory we shall be considering,
to remedy this situation, we admit also a primitive concept of quasi-cardinal which
intuitively stands for the ‘quantity’ of objects in a collection.8 The axioms for this
notion grant that certain quasi-sets x (in particular, those whose elements are m-
objects) may have a quasi-cardinal, written qc(x), even when it is not possible to
attribute an ordinal to them.
To link the relation of indistinguishability with qsets, the theory also encom-
passes an ‘axiom of weak extensionality’, which states (informally speaking) that
those quasi-sets that have the same quantity (expressed by means of quasi-cardinals)
of elements of the same sort (in the sense that they belong to the same equivalence
class of indistinguishable objects) are indistinguishable by their own. One of the
interesting consequences of this axiom is related to the quasi-set version of the non
observability of permutations, which is one of the most basic facts regarding indistin-
guishable quanta (for a discussion on this point, see French & Rickles 2003). In brief,
remember that in standard set theories, if w ∈ x, then of course (x − {w}) ∪ {z} = x
iff z = w. That is, we can ‘exchange’ (without modifying the original arrangement)
two elements iff they are the same elements, by force of the axiom of extensionality.
In contrast, in Q we can prove the following theorem, where [[z]] (and similarly
[[w]]) stand for a quasi-set with quasi-cardinal 1 whose only element is indistin-
guishable from z (respectively, from w) –the reader shouldn’t think that this element
is identical to either z or w, for the relation of equality doesn’t apply to these items;
the set theoretical operations can be understood according to their usual definitions):
Theorem: [Unobservability of Permutations] Let x be a finite quasi-set such that x
does not contain all indistinguishable from z, where z is an m-atom such that z ∈ x.
/ x, then there exists [[w]] such that
If w ≡ z and w ∈

(x − [[z]]) ∪ [[w]] ≡ x.

The theorem works to the effect that, supposing that x has n elements, then
if we ‘exchange’ their elements z by corresponding indistinguishable elements w
(set theoretically, this means performing the operation (x − [[z]]) ∪ [[w]]), then
the resulting quasi-set remains indistinguishable from the one we started with. In
a certain sense, it does not matter whether we are dealing with x or with (x −
[[z]]) ∪ [[w]]. So, within Q, we can express that ‘permutations are not observable’,
without necessarily introducing symmetry postulates, and in particular to derive ‘in
a natural way’ the quantum statistics (see Krause et al. 1999; French & Krause 2006,
chap.7). Further applications to the foundations of quantum mechanics can be seen
in Domenech et al. 2008.

Principia 13(3): 251–68 (2009).

260 Jonas R. Becker Arenhart & Décio Krause

4. Formal semantics: the case of Q

From now on, we shall be working in the classical part of Q (that is, we use it as
our metatheory), which was shown in the above section to be that part of Q in
which classical mathematics can be developed, and which deals with what we call
the ‘classical entities’ of the theory. Our goal is to show how the syntax of first-
order theories can be developed in Q, and in particular, the syntactical part of Q
itself, (formulas and symbols should be understood as classical items) when Q is
also taken as the object language. We will not go into the details when they can be
recovered from our previous discussion for the classical case.
Once again, the individual variables of the theory are to be represented by the
set ω, which exists in the classical part of Q. For the case of the logical symbols, as
before ¬ is the pair 〈0, 0〉, → is 〈0, 1〉, and ∀ is 〈0, 3〉. We can go on and define the
non-logical vocabulary for first-order theories, just by considering how predicates
can be taken to be certain quasi-sets as we did before in the case of classical logic,
taking them to be some specific sets. Our goal now is to consider the particular case
of the non-logical vocabulary of Q itself, considered as an object theory.
For m, M and Z, which are unary predicate symbols of Q, we define respectively
the quasi-sets 〈1, 1〉, 〈2, 1〉 and 〈3, 1〉. Here, the second number indicate the rank of
the symbol, while the first one indicates the particular enumeration of the symbol.
This is the general reasoning behind the attribution. For the binary relations ≡ and
∈ we define 〈1, 2〉 and 〈2, 2〉, respectively. Finally, for the functional symbol qc we
attribute 〈4, 1〉.
As before, the quasi-sets of formulas can now be defined but, since now we have
a functional symbol qc as an additional term along with variables, we must say what
we mean by a quasi-set of terms. As said above, the variables are the elements of ω.
For the terms of the form qc(x n ), we take the quasi-set [〈qc, n〉 : n ∈ ω]. The terms
are the elements of the q-set T defined as ω ∪ [〈qc, n〉 : n ∈ ω]. For easy reference,
we denote an arbitrary element of T by t n .
Thus we have in our particular case atomic formulas of the type t n ≡ t m , which
as before are seen as triples, but now with ≡ replacing =, that is, 〈t n , ≡, t m 〉. The
atomic formulas of this type constitute the q-set At ≡ , which by definition is the quasi-
set T ×{≡}× T. Formulas of the form mt n , Z t n , M t n , and t n ∈ t m are particular cases
of formulas of the general form P n x i0 . . . x in−1 . As before, each of them is taken to
be a pair 〈P n , f 〉, where f is a quasi-function from n (the rank of the symbol in
question) to T, so that, for example, if f is such that f (0) = 〈qc, n〉 and f (1) = 2,
then 〈∈, f 〉 is the formula qc(x n ) ∈ x 2 . The quasi-set of all such formulas, At P is by
definition [〈P n , f 〉 : n ∈ ω and f : n 7→ T].
With these details explicated, the quasi-set F of formulas of quasi-set theory turns
to be the smallest quasi-set satisfying the conditions:

Principia 13(3): 251–68 (2009).

Quantifiers and the Foundations of Quasi-set Theory 261

1. At ≡
At P ∈ F;

2. if α ∈ F, then 〈¬, α〉 ∈ F;

3. if α, β ∈ F, then 〈α, →, β〉 ∈ F;

4. if n ∈ ω and α ∈ F, then 〈∀, n, α〉 ∈ F.

5. The underlying logic of Q and its semantics

Let us call the first-order underlying logic of Q ‘the logic of indiscernibility’, LQ . Its
language has the usual logical connectives ¬, ∧, ∨ and → (one needs to assume only
¬ and → as primitive, the other ones being defined as usual), a denumerable set of
individual variables x 1 , x 2 , . . ., and a binary predicate symbol ≡. Specific non-logical
symbols can be added, as for instance, in the case of Q, the unary predicate symbols
m, M and Z, the binary predicate symbol ∈ and the unary functional symbol qc.
Terms and formulas are defined as usual.
The postulates of LQ are the following ones:

1. A complete set of postulates for first-order classical logic without identity.

2. The postulates of indiscernibility, namely:

(a) ∀x(x ≡ x)
(b) ∀x∀ y(x ≡ y → y ≡ x)
(c) ∀x∀ y∀z(x ≡ y ∧ y ≡ z → x ≡ z)

It is clear that any equivalence relation models ≡, in particular, standard identity

= models it. But the most interesting case is when we can speak of indiscernible
but not identical things, and this can be done within standard systems (such as
ZFC) only if we restrict the discourse to a particular structure or consider just a
few primitive predicates and relations so that the considered structure has other
automorphisms than the identity function. But since in ZFC we can always extend a
structure to an structure having only the identity mapping as its automorphism, the
talk of indiscernible things in ZFC turns to be quite artificial. The better way, to do
it, in our opinion, is to employ an adequate mathematical theory, such as Q.
Let’s recover for one moment what we are pursuing with the constructions just
presented. The first point is that, since the very presentation of the syntax of the
language of LQ involves some set theoretical apparatus, like a family of variables
and the use of recursion on sets of formulas, we don’t need to commit ourselves
to classical set theoretical apparatus and all that comes with it. We can confine

Principia 13(3): 251–68 (2009).

262 Jonas R. Becker Arenhart & Décio Krause

ourselves to a classical part of Q and grant that we are not taking everything to be
classical sets, in particular, not everything in the intended range of quantification
must be an individual, when the intended interpretation of the symbols come into
The situation is analogous to the formulation of classical set theory. One starts
with some sets of symbols and, by applying the usual procedures, we can develop
a formal theory of set theory. So, if one wants to argue that quasi-set theory neces-
sitates a classical theory to be formulated, for the very syntax presupposes classical
concepts, we can respond to this objection by showing that it indeed needs some
classical concepts, but these can come from a quasi-set theory which is being in this
case used as metalanguage.
The second and most important point of interest for our purposes is that since Q
has a first-order underlying logic LQ , the foregoing construction helps us in avoiding
falling in the trap mentioned in the beginning of our paper. There we mentioned that
a non-classical logic becomes committed, through the metalanguage, with the very
classical principles it meant to avoid, if its semantics is developed in ZFC or other
classical set theory. Now we recall that the underlying logic of quasi-set theory must
not be simply classical first-order logic without identity, for the semantics for these
logics are made in ZFC, as we have exemplified above with the case of an equivalence
relation modeling ≡. So, despite the fact that the axioms were the same as those
presented for first-order classical logic, the semantics of LQ is based in quasi-set
theory, allowing for non-individuals in the range of quantification. This kind of
semantics is a generalization of Tarskian semantics, and it is in fact the specific case
presented in French & Krause 2006, chap.8.
To insist a little more on this topic, the underlying logic of Q, when considered
from the syntactical point of view, is a first order language without identity, whose
axioms are the same as those used for classical logic (see the presentation of LQ
above). But the intended semantics for this logics differs from the specific one for
classical logic. When providing the semantics for LQ , we allow that not only sets can
be used in the domain of quantification, but also quasi-sets, so that non-individuals
also may appear there. It is a matter of generalization of the usual case, having the
classical semantics as a limit case.
Again, to avoid confusion, it is important to distinguish between the theory we
are studying and the theory we are using to develop these matters. We use quasi-
set theory to clarify the semantics of a first-order logic LQ . By the nature of its
semantics, LQ is a non-classical logic, more specifically, it is a kind of non-reflexive
logic, since it permits that objects for which neither identity nor difference applies
appear in the domain of quantification, and it is exactly the logic underlying the
formal system of quasi-set theory Q which we investigate, the object theory. So,
through these devices we can show how to develop the underlying logic of Q without

Principia 13(3): 251–68 (2009).

Quantifiers and the Foundations of Quasi-set Theory 263

having to re-introduce individuality for every item in the domain of quantification,

and granting that the first-order theories that employ this logic as underlying logic
can ‘talk’ about non-individual in a (supposed) philosophically satisfying way (see
Krause 2008).
The discussion we have advanced relies on the fact that a system of logic can
be said to acquire its characterization not only through its syntax, but also through
its semantics (by the way, this is Newton da Costa’s position too). As we said, one
should not assume from the outset that the reading of the formulas is based on
classical set-theoretical semantics, which would commit us with a classical ontology,
but one should choose the adequate framework for the semantical developments,
namely, Q.
The point that a classical semantics for these logics would commit us with a kind
of classical ontology is another thesis of philosophical relevance. In general, a system
of logic (of first or higher order, or set theories) carries with it some metaphysical
implications. Classical logic, in particular, takes it to be the case that the world
is composed of atomic facts, in which, for example, laws like the excluded middle
holds (this resembles Russell and Wittgenstein’s logical atomism) and, in particular,
what is relevant for our discussion, it takes identity to be everywhere applicable.
These features are also encapsulated in the framework in which the semantics is
formulated, so, as we would say it, it describes a world of individuals. When we want
to discuss the problem of indiscernible particles in quantum physics, for example,
and hold that they are not individuals, failing to satisfy the axioms for identity of
classical logic, then these commitments should be avoided, and our logic must cope
with our philosophical position, as we hold that quasi-set theory may do that.
In fact, using Q as the ‘background’ theory in Quine’s sense (Quine 1969), we
can commit ourselves with non-individuals in such a way that we can propose the
negation of one of Quine’s most famous mottos, namely, that there are no entities
without identity, for we can, with the resources of Q, argue that non-individuals
can be values of variables as well, so we can state our maxim: there are entities
without identity (for details, see Krause 2008). This in particular shows that da
Costa’s generalization of Quine’s dictum that to be is to be the value of a variable,
by adding (contrary to Quine), “of a certain language and subjected to a certain
underlying logic”, we should also add and modulo a given background theory used as
its metalanguage for, as we have seen, depending on this theory (for instance, taking
Q), we can be committed to non-individuals in our ontology. Of course this point
should be investigated further.

Principia 13(3): 251–68 (2009).

264 Jonas R. Becker Arenhart & Décio Krause

6. Further Philosophical Discussion

An additional relevant point in the philosophical discussion of semantics for formal

languages is the extent to which we can say that a Tarski-style like semantics really
elucidates the meaning of the logical terms of the considered language. We are
aware of the difficulties involved in such discussions, and do not aim at to dispute
the fact that such a semantics serves mainly as a mathematical procedure to make
rigorous and sharp some intuitive understanding of the assumed concepts. This
intuitive understanding guides the definitions adopted in the semantics, as it may
be seen from a valuation semantics for the classical propositional calculus. The fact
is that it is of course debatable if the formal semantics really captures the relevant
aspects of the informal use of the considered concepts.
So, one may still be not satisfied with the technical developments of the last
section and the use we have made of it to justify the primitive concepts of the theory
Q. Sure, one may still ask for the explication of the intuitive basis underlying those
developments. The dissatisfaction perhaps can be expressed in the following lines:
“It is all right to proceed that way for classical logic, for the formal semantics is
only a mean to write in a mathematical fashion some of our informal understanding
of the logical terms. In the case of quasi-set theory, one ought to provide also an
informal understanding, otherwise not much is really gained through the proposed
formal semantics.”
We think that these objections can be overcome, in some sense, since the axioms
of quasi-set theory can also be seen as written to cope with some intuitive features
of our use of the language in contexts in which we can take identity as a meaning-
less concept. We think that in some contexts it is possible to consider identity as
a derived notion, and not a primitive one, and in these contexts it may be seen as
‘parasitical’ one in relation to other more basic notions such as indistinguishability
and cardinality. Really, identity can be understood in terms of these last ones, and
thus, taken as a convenient fiction useful for practical purposes. From an epistemo-
logical point of view, Hume may be read as a precursor of this view, obviously not
understanding identity as relying on cardinality. Really, if we are free to suggest that
those who have questioned the intuitive concept of identity, or of its applicability,
then David Hume can in fact be regarded as a forerunner of non-reflexive logics.
Obviously without any commitment to logical systems as we understand them today,
Hume has questioned the applicability of identity in the re-identification processes,
as it is clear in his Treatise (Hume 1985). For an analysis of this idea and of its
possible relationships with Schrödinger, see Becker & Krause 2006.
We take it to be the case that, for practical purposes, and not only in contexts
where indiscernible particles are concerned, identity does not need to be taken as a
primitive concept, for it is enough to assume a weaker notion of indistinguishability

Principia 13(3): 251–68 (2009).

Quantifiers and the Foundations of Quasi-set Theory 265

(as in quasi-set theory, where it is, as we have seen, a congruence relation). With
indistinguishability, one obtains identity as a derived notion, which we can apply
when dealing with everyday objects, by taking its negation, discernibility, to be un-
derstood as difference. That is, for usual things, we take that discernibility acts as
a fake individuation principle, and we make no claim that it neither really individ-
uates only one object, nor that it ascribes a kind of primitive thisness to it. That is,
we do not go beyond those assumption that are convenient for practical purposes.
So, language works as usual, with names and quantification, for these items, and
everything works as if they were individuals (some of these ideas are a modification
of some proposals that were first advanced by Teller 1998).
Secondly, to understand the quantifiers in an informal sense, when they are ap-
plied to the intuitive aggregates of non-individuals, we don’t need to use identity, for
we can resort to a distinguishing feature of such aggregates: the cardinal number.
When we say something such as ‘all items are so and so’, we can make a paraphrase
and consider it to mean ‘All m of them are such and such’, the m belonging to a
suitable qset. Difference (the negation of identity), at least within these contexts
involving indistinguishable items, can be derived from the cardinality talk. Taking
the cardinals as having primacy, difference is then a derived by-product, for we may
assume that, since there are ‘more than one item’, they may be thought as being
distinct, although this kind of talk might be not expressed in the object language.
Anyway, the ‘cardinal talk’ does not imply difference. Just for a simple example,
let us consider a collection with two indistinguishable members (for instance, two
bosons in the same state). We can say (in the metalanguage) that they are different,
not because we present some distinctive feature (they really do not present any),
but because they are two, and the difference should be taken not as the negation of
usual identity, but just as an abbreviation of speech.
Our claim is that one can consistently maintain that all these features are cap-
tured in our previous discussion on the semantics of the underlying logic of Q. This
is one of the possible ways to give and intuitive semantics to quantifiers when we
consider that we are dealing with aggregates that might have a quasi-set theoretical
behavior. Our tentative proposal of basing the intuitive understanding of quantifiers
and even the usual reference mechanisms for classical objects, is a first approach on
the subject, but we think there is still much to be said in favor of such an approach,
and we shall work on the details in future papers.

7. Conclusion

In this paper we were concerned with a semantics for a particular kind of non-
reflexive logic. Our suggestion was that we can employ quasi-set theory to help us

Principia 13(3): 251–68 (2009).

266 Jonas R. Becker Arenhart & Décio Krause

to understand quantification over non-individuals, following a generalization of the

usual procedure employed for classical logic semantics. The domain of quantification
can be a quasi-set, and the usual definitions used for classical logic can be extended
to the case under consideration.
From an informal point of view, we related the issue to a suggestion that non-
individuals may be metaphysically prior to individuals, and that identity (and dif-
ference) may be seen as ‘derived’ from indistinguishability and cardinality. Working
on this possibility, one can see that quantification and naming can be recovered
for what we have called the ‘standard (or usual) objects’, and applies for items in
general. This is a first approach to the subject, but we think that it shows that non-
reflexive logics, in particular quasi-set theory, may be relevant in this context for it
seems to be on a par with the specific type of certain metaphysical claims (specifi-
cally, a metaphysics of non-individuals), helping us to articulate rigorously a general
view on such particulars.9

Becker, J. & Krause, D. 2006. Hume, Schrödinger e a individuação de objetos físicos. Revista
Eletronica Informação e Cognição 5(2): 59–71.
Curry, H. B. 1951. Outlines of a Formalist Philosophy of Mathematics. Amsterdam: North-
Holland Publishing Company.
da Costa, N. C. A.; Béziau, J. Y.; Bueno, O. 1995. What is semantics? A brief note on a huge
question. Sorites–Electronic Quarterly of Analytical Philosophy 3: 43–7.
da Costa, N. C. A. & Krause, D. 1994. Schrödinger Logics. Studia Logica 53(4): 533–50.
———. 1997. An intensional Schrödinger logic. Notre Dame Journal of Formal Logic 38(2):
Devlin, K. 1993. The Joy of Sets: Fundamentals of Contemporary Set Theory. New York:
Domenech, G. & Holik, F. 2007. A Discussion on Particle Number and Quantum Indistin-
guishability. Foundations of Physics 37(6): 855–78.
Domenech, G.; Holik, F.; Krause, D. 2008. Q-spaces and the foundations of quantum me-
chanics. Foundations of Physics 38(11): 969–94.
Ebbinghaus, H. D.; Flum, J.; Thomas, W. 1994. Mathematical Logic. New York: Springer-
Falkenburg, B. 2007. Particle Metaphysics: A Critical Account of Subatomic Reality. Berlin:
French, S. & Krause, D. 2006. Identity in Physics. A historical, philosophical, and formal anal-
ysis. Oxford: Oxford Un. Press.
French, S. & Rickles, D. P. 2003. Understanding Permutation Symmetry. In K. Brading and E.
Castellani (eds.) Symmetries in Physics: New Reflections. Cambridge: Cambridge Univer-
sity Press, 212–38.
Hume, D. 1985. Treatise of Human Nature. 2nd ed. L. A. Selby-Bigge. Oxford: Oxford Un.

Principia 13(3): 251–68 (2009).

Quantifiers and the Foundations of Quasi-set Theory 267

Jammer, M. 1974. The Philosophy of Quantum Mechanics. New York: John Wiley & Sons.
Krause, D. 2008. Nota sobre o comprometimento ontológico com não-indivíduos. In R. A.
Martins, C. S. Silva, J. M. H. Ferreira e L. A. P. Martins (eds.) Filosofia e História da Ciência
do Cone Sul – Seleção de Trabalhos do 5o Encontro. Campinas: Associação de Filosofia e
História da Ciência do Cone Sul, 125–32.
Krause, D.; Sant’Anna, A. S.; Volkov, A. G. 1999. Quasi-set theory for bosons and fermions:
quantum distributions. Foundations of Physics Letters 12(1): 51–66.
Post, H. 1963. Individuality and Physics. The Listener, 10 october 1963, pp. 534–7. (reprinted
in Vedanta for East and West 132, 1973, pp.14–22.)
Quine, W. V. 1969. Ontological Relativity and Other Essays. New York: Columbia Un. Press.
Schrödinger, E. 1952. Science and Humanism. Cambridge: Cambridge Un. Press.
Teller, P. 1998. Quantum Mechanics and Haecceities. In E. Castellani (ed.) Interpreting Bodies:
Classical and Quantum Objects in Modern Physics. Princeton: Princeton University Press,


Programa de Pós-Graduação em Filosofia
Universidade Federal de Santa Catarina
Florianópolis, SC

Resumo. Neste artigo discutimos algumas questões propostas por Newton da Costa rela-
cionadas aos fundamentos da teoria de quase-conjuntos. Seus questionamentos aqui consi-
derados tratam da possibilidade de uma compreenão semântica da teoria, principalmente
devido ao fato de que identidade e diferença podem não ser aplicáveis para algumas das
entidades no domínio pretendido da teoria. De acordo com ele, o modo usual de se com-
preender os quantificadores utilizados na teoria depende da hipótese de que a identidade
deve valer para todas as entidades no domínio de discurso. Inspirados pelas suas questões,
sugerimos que essas dificuldades podem ser superadas tanto em um nível formal quanto em
um nível informal, mostrando como a quantificação sobre itens para os quais a identidade
não faz sentido pode ser entendida sem pressupor uma semântica baseada em uma teoria
de ‘clássica’ de conjuntos.

Palavras-chave: Quase-conjuntos, semântica para lógicas não clássicas, lógica da mecânica


Of course there are several distinct formulations of ZF, say in first-order logic or higher-
order logic, and even taken one of these approaches, one has several alternatives.
Interesting that the physicist and philosopher C. F. von Waizsäcker expressed similar opin-
ion by means of his ‘principle of semantic consistency’, according to which “the rules by we

Principia 13(3): 251–68 (2009).

268 Jonas R. Becker Arenhart & Décio Krause

describe and guide our measurement, defining the semantics of the formalism of a theory,
must be in accordance with the laws of the theory” (cited by Jammer 1974, 156).
We call non-reflexive those logics which deviate from the ‘traditional theory of identity’,
in particular, by violating the principle of identity in some of its forms. The paradigmatic
case is related to the first-order reflexive law of identity, namely, ∀x(x = x), also called the
principle of identity of the first-order classical logic.
Although Schrödinger restricted his discussions to the particles of non-relativistic quantum
mechanics, we can extend the discussion to the ‘particles’ of other quantum theories as well;
for an account on such distinctions, see Falkenburg 2007, chap.6.
It should be remembered that according to the standard definition, a cardinal is a particular
ordinal, hence it makes sense only under an identification of the considered elements.
This is of course a way of speech.
We just recall that an ordinal is a transitive set which is well-ordered by the membership
relation, and that a cardinal is an ordinal α such that for no β < α there does not exist a
bijection from β to α. For details of these concepts, see Devlin 1993.
The notion of quasi-cardinal can be defined for finite quasi-sets; see Domenech & Holik
The second author was supported by an individual research grant from The National Coun-
cil for Scientific and Technological Development (CNPq), process #304540/2006-4.

Principia 13(3): 251–68 (2009).

University of Miami

Abstract. In this paper, I examine three models of reduction. The first, and the most restric-
tive, is the model developed by Ernest Nagel as part of the logical empiricist program. The
second, articulated by Jerry Fodor, is significantly broader, but it seems unable to make sense
of a salient feature of scientific practice. The third, and the most lenient, model is developed
within Newton da Costa and Steven French’s partial structures approach. I argue that the
third model preserves the benefits of Fodor’s proposal, and it is still able to accommodate
relevant aspects of scientific practice. In particular, it offers a conception of reduction with-
out reductionism, and an account of the relation between reduced and reducing theories via
unsharp mappings and partial structures—even in the presence of incomplete information.

Keywords: Reduction, models, special sciences, partial structures, partial mappings.

1. Introduction

The concept of reduction has played a significant role in early conceptions of sci-
ence, particularly those articulated by logical empiricists, such as Rudolf Carnap
(1928/1967) and Ernest Nagel (1961). It also played an important role in philo-
sophical reconstructions of mathematics in the hands of Gottlob Frege (1893, 1903),
with the reduction of arithmetic to second-order logic plus definitions. Part of the
difficulty of these early approaches to reduction is the extremely tight connection
they require between the domains under study: the reducing and the reduced.
In response to this sort of concern, Jerry Fodor has offered a very interesting, and
much less restrictive, model of reduction, with the additional bonus of preserving
the autonomy of the special sciences (see Fodor 1974 and 1997). This is clearly the
right move to make. But Fodor’s proposal faces a difficulty to accommodate certain
aspects of scientific practice, in particular given the use of quantum mechanics in
chemistry. In this paper, building on Fodor’s approach, I offer an alternative, also
less restrictive, model of reduction. In order to do that, I explore the resources of
Newton da Costa and Steven French’s partial structures approach (see da Costa and
French 2003), and examine in which way we can still consider a place for reduction
in contemporary science.

Principia 13(3): 269–82 (2009).

Published by NEL — Epistemology and Logic Research Group, Federal University of Santa Catarina (UFSC), Brazil.
270 Otávio Bueno

2. Two models of reduction

To set out the scene, I will start by considering two models of reduction: Nagel’s
reduction and Fodor’s reduction.2 Nagel’s reduction is obtained via derivability: one
is supposed to establish a derivation relation between reduced and reducing theories.
The proposal is inspired in Frege’s approach to the logical foundations of arithmetic,
and it extends the approach to the empirical sciences. Similarly to Frege’s, Nagel’s
proposal demands a very strict relation between the theories involved. In contrast,
Fodor’s reduction is obtained via the introduction of disjunctive properties. The
proposal offers a significantly less restrictive connection between the reducing and
reduced theories. And, in particular, it allows for exceptions in the laws of the
special sciences—a much more promising strategy. I will critically examine each of
these models of reduction in turn.

2.1. Nagel’s reduction

Nagel conceptualizes the reduction between two theories in terms of one theory
explaining the other, where the model of explanation that he adopts is the deductive-
nomological account. He emphasizes the point:
Reduction [. . . ] is the explanation of a theory or a set of experimental laws
established in one area of inquiry, by a theory usually though not invariably
formulated for some other domain. (Nagel 1961: 338.)

Given the account of explanation that is used—according to which an explanation is

ultimately a deductively valid argument—the relation between the reducing and the
reduced theory becomes one of derivation. To reduce a theory T to a theory T 0 , we
should be able to derive the principles or laws that characterize T from those of T 0 .
However, Nagel identifies a difficulty that emerges immediately given the re-
quirement of logical derivability:
If the laws of the secondary science [that is, the reduced theory] contain
terms that do not occur in the theoretical assumptions of the primary disci-
pline [that is, the reducing theory] [. . . ] the logical derivation of the former
from the latter is prima facie impossible. (Nagel 1961: 352.)

Clearly, if the reduced and reducing theories don’t have a shared vocabulary, no
logical derivation between them could be established.
In order to address this difficulty, Nagel introduces two “necessary formal condi-
tions” that need to be in place so that the reduction can occur. (a) The first is the
existence of bridge laws, which are used to link the vocabularies of the reduced and
the reducing theories.

Principia 13(3): 269–82 (2009).

Models of Reduction 271

Assumptions of some kind must be introduced which postulate suitable re-

lations between whatever is signified by ‘A’ [i.e. the offending term in the
reduced theory] and traits represented by theoretical terms already present
in the primary [reducing] science. (Nagel 1961: 353.)

The assumptions Nagel refers to above are precisely the bridge laws. (b) The second
required formal condition is derivability. In order to be able to reduce one theory to
another, on Nagel’s model, one needs to be able to derive the laws of the reduced
theory from those of the reducing one. As Nagel points out:
With the help of these additional assumptions (the bridge laws], all the
laws of the secondary science, including those containing the term ‘A’, must
be logically derivable from the theoretical premises and their associated co-
ordinating definitions in the primary discipline. (Nagel 1961: 353–4.)

If no such derivability can be established between the theories in question, there is

no reduction between them.
Given the crucial role played by bridge laws in Nagel’s reduction, various issues
surface (see Batterman 2002: 63). What exactly is the status of the bridge laws?
That is, how should these laws be understood? Do they emerge from some sort of
linguistic analysis? Or are they empirical discoveries? If bridge laws are empiri-
cal statements, what kind of necessity do they display? In particular, if bridge laws
establish identity relations among suitable expressions of the reducing and the re-
duced theories, are such identity relations synthetic or analytic? The development
of Nagel’s conception presupposes answers to all of these questions.
Let me focus on the last question. It seems clear that, in order for Nagel’s ap-
proach to work, the bridge laws need to establish some kind of synthetic identity
relation (Batterman 2002: 63). Consider the reduction of the theory of physical
optics to that of electromagnetic radiation. The reduction is implemented by identi-
fying two groups of entities: light waves and electromagnetic radiation. As Lawrence
Sklar points out:
The place of correlatory laws [i.e. bridge laws] is taken by empirically estab-
lished identifications of two classes of entities. Light waves are not correlated
with electromagnetic waves, for they are electromagnetic waves. (Sklar
1967: 120.)

It is, therefore, crucial to establish, empirically, the identity of the relevant objects.
What Nagel needs, before Saul Kripke, are necessary a posteriori identity statements.
It turns out, however, that Nagel’s proposal faces several difficulties. (1) Strictly
speaking, we hardly ever have logical derivations in scientific practice. Thus, we
hardly ever find actual cases of reduction in Nagel’s sense. The paradigmatic ex-
ample that Nagel offers of reduction is the reduction of the Galilean laws of falling

Principia 13(3): 269–82 (2009).

272 Otávio Bueno

bodies to Newtonian mechanics. But not even this case actually works. At best, we
can derive from Newtonian mechanics approximations to the laws of the Galilean
theory. But these approximations are, in fact, incompatible with the actual laws.
Thus, no logical derivations are actually established between the two theories (see
Sklar 1967: 110–1, and Batterman 2002: 62).
(2) Even if we could derive the reduced theory from the reducing one, the latter
typically corrects the former (Feyerabend 1962). Consider, for instance, the con-
cept of temperature from classical thermodynamics. This is a nonstatistical con-
cept, characterized in terms of Carnot cycles, and it is related to the nonstatistical
second law of thermodynamics. Consider, now, the alleged reduction of classical
thermodynamics to statistical mechanics. The “reduction” doesn’t identify nonsta-
tistical items in the reducing theory (statistical mechanics) with the nonstatistical
concept of temperature as the latter is found in the reduced theory (thermody-
namics). In fact, given that, as opposed to statistical mechanics, thermodynam-
ics is not a statistical theory, it is unclear how we could even find a bridge law
that captures the concept of temperature and the function it plays in thermodynam-
ics. That is, it is unclear how the alleged reduction could ever be established (see
Batterman 2002: 64).
Perhaps the Nagelian could take the point. She could incorporate into the char-
acterization of reduction the idea that the reduced theory ends up being corrected by
the reducing theory. That is, the outcome of the reduction process may well be a cor-
rected version of the reduced theory. In fact, several physicists now would agree that
the concept of temperature in classical thermodynamics and other concepts in that
theory, such as entropy, have to be changed in light of the “reduction” to statistical
mechanics. Physicists even speak of the theory of statistical thermodynamics. Thus,
reduction can play an important role in the process of correction and refinement of
scientific theories.
Schaffner (1976) explicitly incorporates into a Nagelian framework the idea of a
“corrected reduced theory” (Shaffner 1976: 618). A theory T reduces a theory T 0 if,
and only if, there is a corrected version of T 0 , call it T 0∗ , such that:

(a) the primitive terms of T 0∗ are associated via bridge laws with various terms
of T ;
(b) T 0∗ is derivable from T when it is supplemented with such bridge laws;
(c) T 0∗ corrects T 0 in that it makes more accurate predictions than T 0 does;
(d) T 0 is explained by T in that T 0 and T 0∗ are strongly analogous to one another,
and T indicates why T 0 works as well as it does in its domain of validity.

Schaffner’s proposal is very reasonable, given the Nagelian framework it as-

sumes. But perhaps rather than theory reduction, what Schaffner offers is an ac-

Principia 13(3): 269–82 (2009).

Models of Reduction 273

count of a different concept: what we are dealing with here is a process of theory
refinement (see also Batterman 2002: 64–5).
(3) Even the revised Nagelian approach does not quite work due to the phe-
nomenon of multiple realization (Fodor 1974 and 1997; see Batterman 2002: 65-
68). The central idea is that “upper level” properties (such as psychological states)
are potentially realized in a huge variety of heterogeneous lower level physical states.
This multiple realizability would make it impossible to establish the bridge laws that,
on the Nagelian approach, are required for the derivation of the laws of the reduced
theory (in the special sciences) from the laws of the reducing theory (e.g. physics).
The central point is the heterogeneity of the lower level realizers of the properties of
the upper level special sciences. The heterogeneity makes it impossible to identify
the relevant items in the bridge laws. (Recall that the latter requires a necessary a
posteriori identity relation.)
Given these difficulties with the Nagelian approach to reduction, a different,
more lenient, model is in order. Fodor offers this model.

2.2. Fodor’s reduction

Fodor’s approach to reduction is shaped, in crucial ways, by the multiple realizability
problem just mentioned. In his own words, here is how he presents the problem:
The very existence of the special sciences testifies to reliable macrolevel reg-
ularities that are realized by mechanisms whose physical substance is quite
typically heterogeneous. [. . . ] Damn near everything we know about the
world suggests that unimaginably complicated to-ings and fro-ings of bits
and pieces at the extreme microlevel manage somehow to converge on sta-
ble macrolevel properties. On the other hand, the ‘somehow’ really is entirely
mysterious [. . . ]. [Some do not] see why there should be (how there could
be) [macrolevel regularities] unless, at a minimum, macrolevel kinds are
homogeneous in respect of their microlevel constitution. Which, however,
functionalists in psychology, biology, geology, and elsewhere, keep claiming
that they typically aren’t. (Fodor 1997: 160–1.)

Given the multiple realizability phenomenon, it comes as no surprise that Fodor

is a not a (classical) reductionist, and certainly not of a Nagelian sort. Instead, he
offers an alternative, less stringent account of the relation of the special sciences to
physics, and defends a corresponding weaker claim: token physicalism (Fodor 1974).
But what exactly is token physicalism? As he notes:
If the bridge laws express event identities, and if every event that falls under
the proper laws of a special science falls under a bridge law, we get the truth
of a doctrine that I shall call ‘token physicalism’. Token physicalism is simply
the claim that all the events that the sciences talk about are physical events.
(Fodor 1974: 100.)

Principia 13(3): 269–82 (2009).

274 Otávio Bueno

In other words, the crucial feature of token physicalism is the idea that, ultimately,
the events described by the sciences are physical.
It is important, however, to distinguish token physicalism from type physicalism.
According to type physicalism, “every property mentioned in the laws of any sci-
ence is a physical property” (Fodor 1974: 100). Token physicalism is weaker than
type physicalism, given that the former does not entail, but is entailed by, the latter.
After all, “the contingent identity of a pair of events presumably does not guar-
antee the identity of the properties whose instantiation constitutes the events; not
even when the event identity is nomologically necessary” (Fodor 1974: 100). Type
physicalism, though, does entail token physicalism. For “if an event is simply the in-
stantiation of a property [. . . ], two events will be identical when they consist of the
instantiation of the same property by the same individual at the same time” (Fodor
1974: 100). Fodor is only committed to the weaker, and more plausible, form of
What is the connection between token physicalism and reductionism (or reduc-
tivism)? Here is Fodor’s answer:
Reductivism [or reductionism] is the conjunction of token physicalism with
the assumption that there are natural kind predicates in an ideally com-
pleted physics which correspond to each natural kind predicate in any ide-
ally completed special science. (Fodor 1974: 100.)

Note the sudden introduction of (i) natural kinds, and (ii) the requirement that
there be a correspondence between natural kind predicates in a completed physics
with each natural kind predicate in the completed special sciences. But what exactly
would count as an “ideally completed physics” and an “ideally completed special
science”? Can we really make sense of what such a science would be like? Neither
natural kinds nor the idea of an “ideally completed science” are found in Nagel’s
conception of reduction.
Given this conception of reductionism, it is not surprising that Fodor wants to
move away from it. As he emphasizes:
It will be one of my morals that the truth of reductivism [reductionism] can-
not be inferred from the assumption that token physicalism is true. Reduc-
tivism is a sufficient, but not a necessary, condition for token physicalism.
(Fodor 1974: 100-1.)

Instead of a strict reductionist picture, Fodor offers a different model of reduction

altogether. The diagram below illustrates the central features of Fodor’s reduction
(see Fodor 1974: 109).

Principia 13(3): 269–82 (2009).

Models of Reduction 275

(a) Law of special science: S1 x - S y

(b) Bridge statements:  @Q A@ HH
 @QQ A @ H
(c) Disjunctive predicates
 @ Q A @ H
 @ Q A @ H
of reducing science: P1 x ∨ P2 x ∨ . . . ∨ Pn x, P 0 x P1∗ y ∨ P2∗ y ∨ . . . ∨ Pm∗ y
6 6 6

(d) Laws of reducing science

Figure 1: Fodor’s reduction

Fodor’s proposal has, in outline, four central aspects:

(i) The proposal allows us to see how the laws of the special sciences can reason-
ably have exceptions—a well-known fact about these sciences. Any S1 event
of the type P 0 —for which there is no corresponding S2 event—will be an ex-
ception to the law S1 x → S2 x.
(ii) Fodor’s proposal does not require that bridge statements be laws. Given the
multiple realizability phenomenon bridge statements will inevitably connect
disjunctive predicates. Fodor doubts that disjunctive predicates can be taken
to be kind predicates (Fodor 1974: 109), but given that he also takes laws to
be connections among natural kinds (Fodor 1974: 101–2), it follows that, on
his view, bridge statements are not laws.
(iii) Fodor’s proposal is weaker than reductionism (in Fodor’s sense). After all, the
proposal “does not imply a correspondence between the kind predicates of the
reduced and the reducing science” (Fodor 1974: 110).
(iv) Finally, Fodor’s proposal implies physicalism. After all, his proposal satisfies
“the same assumption that makes standard reductivism [reductionism] physi-
calistic (namely, that bridge statements express true token identities)” (Fodor
1974: 110).

Although in many ways Fodor’s proposal offers an interesting alternative to

Nagel’s account, it is unclear that the proposal can make sense of a central feature
of scientific practice. To illustrate this point, I will discuss an example that explores
the connection between chemistry and quantum mechanics, and which raises a diffi-
culty for Fodor’s proposal. I will then indicate a way of accommodating the difficulty,
by offering a third model of reduction via unsharp mappings and partial structures,
which still preserves the central features of Fodor’s account.

Principia 13(3): 269–82 (2009).

276 Otávio Bueno

3. A third model of reduction: unsharp and partial mappings

In order to motivate the third model of reduction, I will consider an aspect of the re-
lation between chemistry and non-relativist quantum mechanics (QM) as it emerges
in the context of actual chemical practice. One of the most unexpected features of
QM is the way in which, at least on a significant interpretation of the theory, the
concept of identity cannot be applied to elementary particles.3 According to Erwin
Schrödinger, one of the main proponents of this view:
As our mental eye penetrates into smaller and smaller distances and shorter
and shorter times, we find nature behaving so entirely differently from what
we observe in visible and palpable bodies of our surroundings that no model
shaped after our large-scale experiences can ever be true. (Schrödinger
1952: 17–8.)
In particular, as Schrödinger insists (1952: 17–8), one of the most significant dif-
ferences between the macroscopic and the microscopic worlds is that the concept of
identity lacks any sense if it is applied to elementary particles.
Now, if quantum particles lack identity conditions, it is unclear how exactly the
laws of quantum mechanics can be used to explain how “larger” objects that are
studied in chemistry (such as molecules, inorganic crystals etc.) do have identity
conditions. Moreover, if quantum particles lack identity conditions, it is unclear
how we will be able to express a bridge statement that connects chemical predicates
(such as, “x is a crystalline inorganic solid”) with corresponding quantum mechan-
ical predicates involving quantum particles (such as “x is the electron that. . . ”).
After all, on Schrödinger’s interpretation, “the electron” is not a meaningful expres-
sion. And if such bridge statements are not available, Fodor’s proposal cannot get
off the ground, given that, ultimately, it depends on the identification of items in the
reducing science to items in the special science. Given the presumed lack of identity
conditions for elementary particles, how do chemists manage to do their own work?
As is well known, we can divide crystalline inorganic solids into three groups:
metals, semiconductors, and insulators. All of them have bands, that is, nearly con-
tinuous energy levels. The central different is that while metals have a partially filled
band, semiconductors have a completely filled band (called the “valence band”),
which is separated from the (mostly empty) conduction band by a bandgap E g . In-
sulators, in turn, have the same electronic structure as semiconductors, except that
they have a larger bandgap E g . As a result, insulators are poor conductors, since
they demand more energy to shift electrons from one band to another. Quantum
dots are semiconductors with all three dimensions in the 1-10 nanometer (nm) size
range (see Murphy and Coffer 2002).
Let us now suppose that a semiconductor is irradiated with light of energy hv >
E g . In this case, Murphy and Coffer note:

Principia 13(3): 269–82 (2009).

Models of Reduction 277

An electron will be promoted from the valence band to the conduction band,
leaving a positively charged “hole” behind. This hole can be thought of as
the absence of an electron and acts as a particle with its own effective mass
and charge in the solid. (Murphy and Coffer 2002: 17A.)

And suppose that we are interested in calculating the spatial separation of the elec-
tron and its hole (which is called an “exciton”). We can carry out the calculation by
invoking a modified Bohr model.4 As Murphy and Coffer indicate:
For typical semiconductor dielectric constants, the calculation suggests that
the electron-hole pair spatial separation is ~1-10 nm for most semiconduc-
tors. In this size range, when the exciton is created, the physical dimensions
of the particle confine the exciton in a manner similar to the particle-in-a-
box problem of physical chemistry. Therefore, the quantum effects such as
quantization of energy levels can be observed in principle. (Murphy and
Coffer 2002: 17A.)

If we note the way in which Murphy and Coffer refer to electrons in this context,
it becomes clear what is needed in order to refer to electrons despite their assumed
lack of identity conditions. What we need is a broader notion of mapping: an un-
sharp mapping from the relevant models of the reducing science (QM) to the relevant
models of the special science (chemistry), which does not require that the things that
are being mapped have well-defined identity conditions. The mapping is unsharp in
that instead of being defined for each individual element of its domain, it is defined
for equivalence classes of elements of the domain (see Bueno [2006]). With such
an unsharp mapping in place, we can make sense of the way in which chemists can
do their work even if they are dealing with elementary particles that lack identity
Note, once again, the way in which Murphy and Coffer describe the situation:
An electron will be promoted from the valence band to the conduction band
[. . . ]. This hole can be thought of as the absence of an electron and acts as
a particle [. . . ]. (Murphy and Coffer 2002: 17A; italics added.)

Clearly, identity conditions for electrons are never asserted here, nor is it even pre-
supposed that electrons have identity conditions. All that is required is that we pick
out an electron from a whole bunch of indistinguishable electrons—any one of them
will do. Instead of a precise way of uniquely picking out a particular electron, all we
need is an unsharp mapping from equivalence classes of electrons to the Bohr model,
and from that model to the models that describe the chemical phenomena under
consideration. In other words, what we need are unsharp mappings between the
models of the reducing theory (QM) and the models of the special science (chem-

Principia 13(3): 269–82 (2009).

278 Otávio Bueno

Suitably interpreted in terms of unsharp mappings, Fodor’s broader model of

reduction offers an interesting account of the relation between the special sciences
and the reducing science. But we need not assume the metaphysics of natural kinds
and laws of nature to appropriate the structure of Fodor’s model. In fact, those who
favor an empiricist outlook—and I am certainly among them—are likely to remain
agnostic about the existence of such items (see van Fraassen [1989]). So just the
overall structure of Fodor?s model needs to be adopted to formulate the unsharp
mappings approach. The central move is to introduce unsharp mappings to link
the models of the reducing and the reduced theories—particularly in those cases
in which the reducing science involve elementary particles, and hence we face the
difficulty of being unable to identify the events in question, given the presumed lack
of well-defined identity conditions for such particles.
Clearly, there are cases in which the special science and the reducing science do
not depend directly on QM, and hence the unsharp mappings need not be invoked.
(The latter will be used, though, when we get to quantum mechanical descriptions.)
However, even in the cases that do not depend on QM, we will need to accommodate
the ubiquitous fact that we do not have complete information about the domains un-
der consideration, including those described by our best theories (see da Costa and
French [2003]). In this case, in order to develop a concept of “reduction” that is sen-
sitive to the informational limitations that characterize scientific practice, we need to
offer connections between the models of the reducing and the reduced theories that
take into account what is still unknown about the relevant domains. Partial struc-
tures (structures that represent the partiality of information about such domains)
and partial mappings (transformations that connect partial structures, preserving the
partiality of information) can then be introduced (see da Costa and French [2003]).
These weaker mappings between partial structures involve transformations, such as
partial isomorphism and partial homomorphism, which can be applied even in con-
texts of partial information (see Bueno, French, and Ladyman [2002]).
But what, exactly, is a partial structure?5 A partial structure is an ordered pair
〈D, R i 〉i∈I , where D is a non-empty set, and (R i )i∈I is a family of partial relations
defined over D. In turn, an n-place partial relation R over D is a triple 〈R1 , R2 , R3 〉,
where R1 , R2 , and R3 are mutually disjoint sets, with R1 ∪R2 ∪R3 = D n , and such that:
R1 is the set of n-tuples that (we know that) belong to R; R2 is the set of n-tuples
that (we know that) do not belong to R, and R3 is the set of n-tuples for which it is
not known (or, for reasons of simplification, it is ignored that it is known) whether
they belong or not to R. (Notice that if R3 is empty, R is a usual n-place relation that
can be identified with R1 .)
As noted, partial structures and partial relations offer a way of representing the
partiality of information about a given domain of inquiry. As an illustration, suppose
that this domain is the physics of particles, and let us denote it by ∆. In order to

Principia 13(3): 269–82 (2009).

Models of Reduction 279

study this domain, researchers formulate a conceptual framework that helps them
systematize and interpret the information they obtain about ∆. This domain can be
represented by a set D of objects (which includes real objects, such as configurations
in a Wilson chamber and spectral lines, and “ideal” objects, such as quarks). D is
studied by the examination of the relations that hold among its elements. However,
it often happens that, given a relation R defined over D, we just do not know whether
R relates all objects of D (or n-tuples thereof). Moreover, we often need to ignore
some of the relations that are known to hold among objects of D in order to study
other relations in that domain in a tractable way. This is part of the incompleteness
and partiality of our information about ∆, and is formally accommodated by partial
relations and partial structures.
In terms of these structures, appropriate characterizations of partial isomorphism
and partial homomorphism can then be offered (see Bueno, French and Ladyman,
2002). Given that the latter notions are more open-ended than the standard ones,
they accommodate better the partiality of structures found in scientific practice. Let
S = 〈D, R i 〉i∈I and S 0 = 〈D0 , R0i∈I be partial structures. So, each R i is a partial relation
of the form 〈R1 , R2 , R3 〉, and each R0i a partial relation of the form 〈R01 , R02 , R03 〉.6
We say that a partial function7 f : D → D0 is a partial isomorphism between S
and S 0 if (i) f is bijective, and (ii) for every x and y ∈ D, R1 x y ↔ R01 f (x) f ( y)
and R2 x y ↔ R02 f (x) f ( y). So, when R3 and R03 are empty (that is, when we are
considering total structures), we have the standard notion of isomorphism.
Moreover, we say that a partial function f : D → D0 is a partial homomor-
phism from S to S 0 if for every x and every y in D, R1 x y → R01 f (x) f ( y) and
R2 x y → R02 f (x) f ( y). Again, if R3 and R03 are empty, we obtain the standard no-
tion of homomorphism as a particular case.
Note two crucial differences between partial isomorphism and partial homomor-
phism. First, a partial homomorphism does not require that the domains D and D0
of the partial structures under study have the same cardinality. Second, a partial
homomorphism does not map the relation R0i into a corresponding relation R i (it is
a one-way mapping only). Clearly a partial homomorphism establishes a much less
strict relation between partial structures.
Moreover, partial isomorphism and partial homomorphism offer mappings
among partial structures that are less tight than their corresponding full counter-
parts—isomorphism and homomorphism. Partial mappings, as transformations that
connect different partial structures that can be used in scientific practice, allow for
the transferring of information from one domain into another—even when the in-
formation in question is incomplete.
We can then use partial structures and partial mappings to represent the mod-
els of the reducing and the reduced theories. This offers an additional source of
openness for Fodor’s account of reduction. The partial mappings make possible to

Principia 13(3): 269–82 (2009).

280 Otávio Bueno

connect different scientific domains with no requirement that they be structurally

the same, since only some structure (the relations that are known to hold in the
relevant domains) is transferred from one domain to another. And depending on the
type of mapping that is used, different aspects of the partial structures in question
are preserved. For instance, with a partial isomorphism, we are able to preserve in
full certain parts of the structures under study, as long as the structures under con-
sideration have the same cardinality. But clearly the last condition cannot always be
met. In this case, a partial homomorphism between the structures will be more ap-
propriate. In none of these cases, however, is it required that the objects in question
have well-defined identity conditions, given that a formulation of partial structures
can be given in a set theory that does not require that every object has well defined
identity conditions (see Bueno 2000).8
The combination of unsharp mappings and partial mappings also allows us to
accommodate the difficulties that plagued Nagel’s proposal. (a) Clearly, there is no
requirement of logical derivability of the special sciences from the reducing science.
Instead, unsharp mappings and partial mappings are established between the rele-
vant models. (b) It is perfectly viable to incorporate corrected versions of reduced
theories by establishing new unsharp mappings and partial mappings between old
and new theories. (c) Similarly to what goes on in Fodor’s approach the multiple re-
alizability phenomenon is accommodated with the introduction of disjunctive pred-
icates, which, in turn, are related in the models of the reducing science via suitable
unsharp mappings and partial mappings.
Note that this partial structure approach to “reduction”, similarly to Fodor’s, does
not endorse reductionism. There is no place for natural kinds in it, nor is there any
commitment to fundamental laws of nature. The autonomy of the special sciences is
preserved throughout. However, the proposed framework offers a certain conception
of the relation between the various sciences, keeping along the way the inevitable
pluralism among the different domains of scientific inquiry. It seems to me that
weaker forms of “reduction”, such as the one suggested here, may play a role in
contemporary scientific research. Anything significantly stronger seems to fly on the
face of actual scientific activity.

4. Conclusion

If the considerations above are near the mark, just as with Fodor’s own account, the
use of unsharp mappings and partial mappings provides us with a minimal, weak
conception of “reduction” without reductionism. And, by invoking these mappings,
the partial structure proposal just sketched has the added bonus of being a little
closer to scientific practice.

Principia 13(3): 269–82 (2009).

Models of Reduction 281

Batterman, R. 2002. The Devil in the Details: Asymptotic Reasoning in Explanation, Reduction,
and Emergence. New York: Oxford University Press.
Bueno, O. 2000. Quasi-Truth in Quasi-Set Theory. Synthese 125: 33–53.
———. 2006. Representation at the Nanoscale. Philosophy of Science 73: 617–28.
Bueno, O.; French, S.; Ladyman, J. 2002. On Representing the Relationship between the
Mathematical and the Empirical. Philosophy of Science 69: 497–518.
Carnap, R. 1928/1967. The Logical Structure of the World and Pseudoproblems in Philosophy.
(Translated by Rolf A. George.) Chicago: Open Court.
Cohen, R. S.; Hooker, C. A.; Michalos, A. C.; van Evra, J. W. (eds.) 1976. PSA 1974: Pro-
ceedings of the 1974 Biennial Meeting of the Philosophy of Science Association. Boston: D.
Reidel Publishing Company.
da Costa, N. C. A. & French, S. 2003. Science and Partial Truth. New York: Oxford University
Feigl, H. & Maxwell, G. (eds.) 1962. Minnesota Studies in the Philosophy of Science, Volume
3. Minneapolis: University of Minnesota Press.
Feyerabend, P. 1962. Explanation, Reduction, and Empiricism. In Feigl and Maxwell (eds.)
1962, pp. 28–97.
Fodor, J. 1974. Special Sciences (Or: The Disunity of Science as a Working Hypothesis).
Synthese 28: 97–115.
Fodor, J. 1997. Special Sciences: Still Autonomous After All These Years. Philosophical Per-
spectives 11: 149–63.
Frege, G. 1893/1903. Grundgesetze der Arithmetik, 2 volumes. Jena: Pohle. (Reprinted by
Georg Olms, Hildesheim, 1962.)
French, S. & Krause, D. 2006. Identity in Physics. Oxford: Clarendon Press.
Murphy, C. & Coffer, J. 2002. Quantum Dots: A Primer. Applied Spectroscopy 56: 16A–27A.
Nagel, E. 1961 The Structure of Science. London: Routledge and Kegan Paul.
Schaffner, K. 1976. Reductionism in Biology: Prospects and Problems. In Cohen et al. (eds.)
1976, pp. 613–32.
Schrödinger, E. 1952. Science and Humanism. Cambridge: Cambridge University Press.
Sklar, L. 1967. Types of Inter-Theoretic Reduction. British Journal for the Philosophy of Science
18, pp. 109–24.
van Fraassen, B. C. 1989. Laws and Symmetry. Oxford: Clarendon Press.

Department of Philosophy
University of Miami
Coral Gables, FL 33124-4670, USA

Resumo. Neste artigo, examino três modelos de redução. O primeiro, e mais restritivo,
foi desenvolvido por Ernest Nagel como parte do programa empirista lógico. O segundo,
articulado por Jerry Fodor, embora sendo significativamente mais amplo, é incapaz de dar
sentido a uma característica saliente da prática científica. O terceiro modelo, que é também

Principia 13(3): 269–82 (2009).

282 Otávio Bueno

o mais condescendente, é desenvolvido empregando-se a abordagem baseada em estruturas

parciais proposta por Newton da Costa e Steven French. Argumento que este terceiro modelo
preserva os benefícios da proposta de Fodor e é ainda capaz de acomodar aspectos relevantes
da prática científica. Em particular, ele oferece uma concepção de redução sem reducionismo,
e descreve a relação entre a teoria que é reduzida e a que reduz por meio de mapeamentos
não precisos e estruturas parciais — mesmo na presença de informação incompleta.

Palavras-chave: Redução, modelos, ciências especiais, estruturas parciais, mapeamentos


It is a pleasure to dedicate this paper to Professor Newton da Costa, on the occasion of his
80 th birthday. Since my first publication—a joint paper with Professor Newton (as we call
him in Portuguese)—so much of my work has been shaped and nurtured by his insightful
and illuminating conception of the foundations of science, by his generous mind, and his
critical acumen. His friendship, continuous help, and the collaboration with him mean the
world to me. This paper, which uses the partial structures approach, is of course no exception
among all of those works that rely on his sophisticated conception. I like to think that he will
enjoy the paper, given that he is definitely not a reductionist—even though I am sure he will
not approve of my loose use of the concept of model! (I promise I will not do that again
when I write my paper celebrating his 90 th birthday!)
I follow here the very lucid discussion offered by Robert Batterman (see Batterman 2002:
For a systematic and insightful discussion of this issue, see French and Krause 2006.
According to this model:
r = εh2 /πm t e2
where r is the radius of the sphere defined by the three-dimensional separation of the
electron-hole pair; ε is the dielectric constant of the semiconductor; m t is the reduced mass
of the electron-hole pair; h is Planck’s constant, and e is the charge of the electron (see
Murphy and Coffer 2002: 17A).
I follow here da Costa and French [2003].
For simplicity, I will take the partial relations in the definitions that follow to be two-place
relations. The definitions, of course, hold for any n-place relations.
A partial function is a function that is not defined for every object in its domain.
The set theory in question is quasi-set theory; for a thorough discussion, see French and
Krause 2006.

Principia 13(3): 269–82 (2009).

State University of Campinas — UNICAMP

Abstract. Jaakko Hintikka, in a series of talks in Brazil in 2008, defended that IF (“in-
dependence-friendly”) logic and paraconsistent logic are, in a sense, very similar. Having
sketched the proposal of a new paraconsistent system, he maintains that several achieve-
ments of IF logic could be reproducible in paraconsistent logic. One of the major difficulties,
left as a challenge, would be to formulate some truth conditions for this new paraconsistent
first-order language in order to make IF logic and paraconsistent logic more inter-related.
My proposal is that this would demand an innovative game-theoretical semantic approach
to paraconsistentism, but also that the syntax of the paraconsistent “Logics of Formal In-
consistency” can model the internal logic of Socratic elenchi. I aim to discuss these, and
other points posed by Hintikka, as challenges and opportunities for paraconsistentism, para-
consistent logics and IF logics, as well as to raise some criticisms on Hintikka’s view about

Keywords: Independence friendly logic, paraconsistency, Socratic elenchi.

1. What IF logic is not

Independence friendly (IF) logic, as Jaakko Hintiikka defends in several places and
in Hintikka 2010, is not any alternative or “nonclassical” logic, but replaces first-
order logic by fixing some of its flaws. One of these flaws, according to Hintikka,
is that usual quantification theory is unable to express dependence or independence
from a variable x bounded by a quantifier Q(x) with respect to another variable
y bounded by another quantifier Q0 ( y). A satisfactory logical syntax ought to be
able to express all such logically possible patterns of dependence and independence
between variables, but usual quantification does not, since it just specifies that the
scope of one quantifier would be included in the scope of the other. Appealing to
generalized quantifiers is not better, as the entire theory of quantifiers, generalized
from the classics, suffers from analogous deficiency.
The key is to understand how nested quantifiers are connected or dependent. If,
for instance, quantifiers Q 1 x and Q 2 y are nested, as in:

(1) (Q 1 x)(Q 2 y)A(x, y)

Principia 13(3): 283–97 (2009).

Published by NEL — Epistemology and Logic Research Group, Federal University of Santa Catarina (UFSC), Brazil.
284 Walter Carnielli

the intended meaning is that the scope of Q 2 y is included in the scope of Q 1 x;

inclusion, however, is just a particular kind of relation (in this case, antisymmetric
and transitive, and not necessarily reflexive). But, as much as in relational models
of modal logic suggest, why should we be bound to just this relation? Why no
relation at all? IF logic aspires to remedy this frailty, but setting quantifiers free.
This independency is syntactically expressed by means of a slash notation, Q 2 y/Q 1 x,
meaning that the quantifier Q 2 y occurring within the scope of Q 1 x is not necessarily
bound by just an “inclusion” (linear) relation.
Henkin (1961) provided the first syntactic and semantic analysis for independent
quantifiers; for instance
∀x∃ y(∃z/∀x)A(x, y, z).
The slash means that ∃z is outside the scope of ∀x, which appears a previous
quantifier (and also outside the scope of ∃ y). Henkin already noted that Skolem
functions could express quantifier unboundedness, for instance in the case of this
example, as
∃ f ∃g∀xA(x, f (x), g(z)).
As Cook and Shapiro recognize in (1998), IF first-order logic can indeed be seen
as a natural extension of first-order logic (not necessarily second-order logic) with
enormous expressive resources. Indeed, some important metatheoretical results in-
dependent of traditional first-order logic such as the axiom of choice and König’s
lemma are logical truths of IF logic. Expressing other concepts like infinity and
equicardinality, which need higher-order logics, is also possible in IF logic.
Hintikka’s observation that games can be assigned to formulas was somehow
preceded by Paul Lorenzen in the sixties, but perhaps the most significant finding
by Hintikka was that one can read Skolem functions as winning strategies in such
games. Starting from game-theoretical semantics (GTS) for traditional first-order
logic (with the usual “verifiers” and “falsifiers”) a semantics for IF logic can be ob-
tained by allowing games with imperfect information. Since falsity is defined as the
existence of a winning strategy for the falsifier, not all IF first-order sentences are
either true or false, and the law of excluded middle is not validated in the GTS for IF
logic. In this respect we already attain an obvious connection between IF logic and
paracomplete logics, the dual of paraconsistent logics (cf. Loparić & da Costa 1984).
Paracomplete logics do not validate the law of excluded middle, and it sounds amaz-
ing that IF logicians have overlooked this 25-year-old idea.
Although Hintikka seemed to believe that no compositional semantics could be
given to IF logic (cf. Hintikka 1996), Hodges (in 1997a; see also Hodges 1997b)
showed that games can be replaced by compositional semantics, in which a formula
is interpreted by a set of sets of assignments. Such sets are called teams by Abramsky
& Vänänen 2008, where it is shown that Hodges’ construction is not ad hoc, but is just

Principia 13(3): 283–97 (2009).

Meeting Hintikka’s Challenge to Paraconsistentism 285

a particular case of a more general algebraic semantics connected — not surprisingly

— to intuitionistic implication. We shall return to this point later on below.

2. IF logic defining new paraconsistent logics

Hintikka claims that IF logic suggests several questions concerning paraconsistent

logic, although they have a completely different initial motivation. Taking into ac-
count the failure of tertium non datur in IF logic and the failure of the Principle of
Pseudo-Scotus or Principle of Ex Contradictione Sequitur Quodlibet in paraconsistent
logic, the independence-friendly and the paraconsistent programs can be compared
up to duality. Hintikka proposes that, if we translate the role of basic truth-values in
IF logic (recalling that not all first-order sentences in IF logic are either true or false,
and thus some can be seen as indefinite) we will gain a new first-order paraconsis-
tent logic:

old truth-values translated truth-values

true or indefinite true
false or indefinite false
true true but not false
false false but not true
indefinite true and false

To honor the town of Paraty, where the meeting that inspired his paper was
held, Hintikka dubbed this new logic paratyconsistent logic, which is obtained as a
semantical translation of IF logic. Several questions are prompted by this new logic:

(1) How to define the corresponding syntactical translation from IF logic which
would axiomatize paratyconsistent logic?

This enterprise is in principle not unfeasible, since the true sentences of paraty-
consistent logic are obtained from the “not false” sentences in IF logic, which are
known to make a recursively enumerable collection. However, it is doubtful that
such a recursive axiomatization would grant completeness to paratyconsistent logic
with respect to the intended semantics, since IF logic itself is not characterizable by
any recursive axiomatization.
But even before tackling the syntactical side of this new paraconsistent logic, its
semantics have problems waiting to be clarified. As in any paraconsistent logic, in-
terpreting negation is challenging. A quotation in Hintikka 2010, section 7, adduces
a particular point:

Principia 13(3): 283–97 (2009).

286 Walter Carnielli

Especially basic are questions concerning negation. What is the natural

treatment of negation in paratyconsistent logic? Should the negation that
is in fact used in paraconsistent logics be interpreted as the contradictory
negation or as some kind of stronger (dual) negation? Some of the prob-
lems listed above can only be solved by introducing a second negation into
one’s logic. How can this be done in the context of paraconsistent logic?

This difficulty, however, has been already faced by paraconsistent logic with a
relatively high degree of success: the possible-translations semantics, introduced al-
most two decades ago in Carnielli 1990 (re-worked in Marcos 1999 and Carnielli
Possible-translations semantics make extensive use of a translation between a
logic L1 and a logic L2: a translation is a mapping ∗ : L1 / L2 such that, for
every set Γ ∪ {α} of L1-formulas,

Γ `L1 α implies Γ∗ `L2 α∗ .

Here, α∗ denotes ∗(α) and Γ∗ stands for {γ∗ : γ ∈ Γ}. If ‘implies’ is replaced by ‘iff’
in the definition above, then ∗ is called a conservative translation. A general account
on translations and their generalizations (transfers and contextual translations) is
done in Carnielli, Coniglio & D’ Ottaviano 2009. It should be remarked that it is this
idea of translation, albeit in a less formal way, which underlies the shift from IF logic
to paratyconsistent logic explained above.
Now, possible-translations semantics deal in a very natural way with the question
of handling two negations (or, in general, several connectives of a similar kind) in
a same logic. Consider, for instance, the following three-valued matrices, where T
and t are the designated values:

∧ T t F ∨ T t F
T t t F T t t t
t t t F t t t t
F F F F F t t F

→ T t F ¬ 1 ¬ 2 ◦1 ◦2
T t t F T F F t F
t t t F t F t F F
F t t t F T t t F

Such tables give a precise semantical interpretation for the logic mbC, a simple
and decent logic of the family of the “Logics of Formal Inconsistency” studied in
Carnielli, Coniglio & Marcos 2007. The truth-value t may be interpreted as ‘true by

Principia 13(3): 283–97 (2009).

Meeting Hintikka’s Challenge to Paraconsistentism 287

default’, or ‘true by lack of evidence to the contrary’, and T and F are the familiar
‘true’ and ‘false’ truth-values. As it can be seen by inspecting the tables, the matrices
for conjunction, disjunction and implication are free from the value T , so the com-
pound sentences are always either frankly false (i.e., receive the truth-value F ) or
‘true by default’.
The novelty (with respect to traditional semantics) is that there are two distinct
interpretations for the negation connective ¬ and for the consistency operator ◦.
This is understandable from the idea that propositions can be valuated in multiple
scenarios. We may think that propositions are either plain false, or are evaluated as t
or T . The first situation concerns a true-by-default proposition, which is treated as a
false proposition by the negation ¬2 (since ¬2 t = t) and as true by ¬1 (since ¬1 t =
F ). On what concerns the consistency operator ◦, the first interpretation ◦1 only
considers as consistent the ‘classical’ values T and F , while ◦2 considers that nothing
is consistent (i.e., ◦2 assigns F to every truth-value) . This can be viewed as two
“logical worlds”, one endowed with ∧, ∨, →, ¬1 and ◦1 and other with with ∧, ∨, →
, ¬2 and ◦2 . However, these two “logical worlds” engender combined valuations,
according to which three-valued interpretation is assigned to the connectives by the
This collection of truth-tables, which we call M0 , will be used to give the de-
sired semantics for mbC. Now, considering the algebra ForM of formulas generated
by P over the signature of M0 , let’s define the set TR0 consisting of all functions
∗ : For ◦ / For 0 subjected to the following clauses:

(t r0) p∗ = p, if p ∈ P ;
(t r1) (α#β)∗ = (α∗ #β ∗ ), for all # ∈ {∧, ∨, →};
(t r2) (¬α)∗ ∈ {¬1 α∗ , ¬2 α∗ };
(t r3) (◦α)∗ ∈ {◦1 α∗ , ◦2 α∗ , ◦1 (¬α)∗ }.

We say the pair PT0 = 〈M0 , TR0 〉 is a possible-translations semantical structure for
mbC. If 0M denotes the consequence relation in M0 , and Γ∪{α} is a set of formulas
of mbC, the associated PT-consequence relation, |=PT0 , is defined as:
Γ |=PT0 α iff Γ∗ 0M α∗ for all translations ∗ in TR0 .
We will call a possible translation of a formula α any image of it through some
function in TR0 .
Not only mbC is sound and complete with respect to such possible-translations
semantics, but all logics Cn in the well-known da Costa’s hierarchy can be char-
acterized by such semantics in terms of copies of a three-valued logic (L F I1, or
equivalently, J3, see Carnielli 2000).

Principia 13(3): 283–97 (2009).

288 Walter Carnielli

Although specially devised to deal with paraconsistency, possible translations

semantics can be also applied to several other logical systems; but in connection
with paraconsisent logics it offers a solution to the difficult problem of algebraizing
the logics Cn by means of a new concept of algebraization, as argued in Bueno-Soler
& Carnielli 2005.
In (2010), Hintikka claims that a perfect correlation between IF logic and para-
consistency would only be achieved if paraconsistent logicans were prepared to give
up compositionality:
. . . the semantics of IF first-order logic is unavoidably noncompositional.
This non-compositionality is due to the fact that the force of a quantifier is
affected by its dependence and independence of the quantifiers in whose
scope it occurs. (Compositionality is equivalent with semantical context-
independence.) Now to the best of my knowledge non-compositional se-
mantics for paraconsistent logic has not been seriously contemplated. Hence
something has to be fundamentally changed if in semantical treatments of
paraconsistent logic paraconsistent logicians want to use the bridge pro-
vided by paratyconsistent logic for the purpose of reaching a self-applicable
truth condition and for the wider purpose of enabling a paraconsistent lan-
guage to serve as its own metalanguage. In a general theoretical perspective,
having to give up compositionality is likely to be the deepest methodologi-
cal change needed for paraconsistent logicans to make use of the kinship of
paraconsistent logic with IF logic.

This requirement for “the deepest methodological change needed for paraconsis-
tent logicans to make use of the kinship of paraconsistent logic with IF logic” seems
strange in view of the above mentioned compositional semantics given by Hodges
to IF logic. The challenge reduces, perhaps, to whether the possible-translations se-
mantics is compositional. But considering that the “Principle of Compositionality”
(or “Frege’s Principle”) asserts that the meaning of a complex expression is fully de-
termined by the meanings of its constituents and by its structure, it is unequivocal
that possible-translations semantics are compositional. The only caveat is that the
meaning of the constituent expressions is also determined by the translations, which
are part of the rules used to combine constituent expressions. So, this challenge
seems to be easily dissolved by no-show: on the one hand, the semantics of IF first-
order logic does not need to be unavoidably non-compositional, and on the other
hand the semantics of paraconsistent is indeed compositional.

3. Interrogative games for paraconsistent and IF logics

Hintikka criticizes paraconsistent logic by its “weak spot”: its interpretation. And
blow his own horn on proudly speaking on the impact of IF logic as “the best thing

Principia 13(3): 283–97 (2009).

Meeting Hintikka’s Challenge to Paraconsistentism 289

that could have happened to paraconsistent logic”.

But what on earth can it mean for a proposition to be true and false? If Cain
tells in a cross-examination that it is both true and false that he killed Abel,
he might be cited for contempt of the court. And it does not seem to be
any less strange to say that a proposition and its negation can both be true.
The existing discussions of paraconsistent logics do not yield a satisfactory
unique account of the concrete down-to-earth meaning of self-contradictory
but yet not disprovable propositions. And it would seem that it will be
equally difficult to give a fully operationalized account of how a proposition
and its negation can both be true.

There is a big distinction, however, between “a proposition to be true and false”

(a situation which never concerned paraconsistency) and another situation where a
proposition and its negation can both be true. Hintikka conflates the two situations,
claiming that the problem of interpreting propositions of paraconsistent logic that
are both true and false corresponds to the problem of interpreting the propositions
in an IF language that have the truth-value “indefinite”. But this “indefinite” value of
IF logic, if it really means that a proposition is both true and false, does not appeal
to paraconsistency.
He is basically right when he talks about game theoretical semantics. Not, as
he puts it, because “the game theoretical semantics for IF logic seem to provide
paraconsistent logic with the concrete semantical interpretation it has been missing”,
but due to the fact that we can give a dual game theoretical semantics to certain
paraconsistent logics with respect to IF logic.
The semantical games that interpret IF logics emerge from the intuition that the
truth of a sentence S means the existence of a winning strategy for the “verifier”
in a semantical game G(S) related to S, so to try to determine the truth of S is to
determine whether such a winning strategy exists or not.
Such games are questioning or interrogative games, deeply distinct from seman-
tical games that see the notion of truth as equivalent to the notion of winning a
game. Interrogative games are knowledge-seeking procedures, according to Hin-
tikka 2004, ch. 13, and are central on Aristotle’s theory and practice of scientific and
philosophical argumentation. Aristotle was committed to two-person interrogative
games (not dissimilar to the Socratic elenchus) in his early proposal for a scientific
method, not only in the Topics, but in both the Analytics. Dialectical games are at the
origin of logic (cf. Hintikka 2007, cf. also Kapp 1942): Aristotle became the founder
of deductive logic by just studying predictable answers.
I completely agree with Hintikka (cf. Hintikka 2010) that the problem of coping
with contradictory information belongs to interrogative games (not to semantical
games or to the formal games of theorem-proving). But he says more:

Principia 13(3): 283–97 (2009).

290 Walter Carnielli

A different question which is not even controversial is that the logic of inter-
rogative games should be paraconsistent in the sense of admitting situations
where one’s prima facie information involves contradictions.

But how could contradictions be dealt with by means of strategic rules? As

much as games that interpret IF logics emerge from the intuition that the truth of a
sentence S means the existence of a winning strategy for the “verifier” in a game G(S)
related to S, the games that would interpret paraconsistent logics would amount to
the existence of a non-losing strategy for the “falsifier”,1 as in the famous Tic-Tac-
Toe game: any of both players cannot lose if playing strategically, but cannot win
either if the adversary is also playing strategically. Of course, every game ends in
a tie if both players are playing strategically. So, to determine (from the game-
theoretical viewpoint) the truth of S in a paraconsistent logic (in particular, the
“paratyconsistent”) is to determine whether such a non-losing strategy exists or not.
This does not exclude, I believe, the interest on deductive rules of inference.

4. Paraconsistency and the elenchus

In the dialectical games, as exemplified in dialogues of Plato with Socrates, a pro-

ponent of a thesis A has to reply to a series of questions from an opponent whose
task is to show, by driving the proponent to an elenchus, that his thesis A is part of
an inconsistent set of premises. The interpretation that the opponent (a role usually
undertaken by Socrates), when conducting the dialogue to an elenchus, is actually
arguing for the truth of ¬A is convincingly rejected by Castelnérac & Marion:
The point of dialectical games is therefore that the proponent of a thesis
A has to pass a test of consistency: A has to fit into a set of beliefs Γ held
by its proponent in such a way that the addition of A to Γ does not render
it inconsistent. If and when the opponent unveils an inconsistency, it is
shameful for the proponent.

They cite, as an example, Charmides and Critias with their high opinions on
themselves, and Critias becoming silent at Charmides (169e) because of feeling
ashamed of having fallen upon an inconsistency.
The dialogical games are a kind of modern hair of dialectical games, which inher-
its a dialogue ability to provide a semantics for several logical systems. A dialogical
formulation of paraconsistency was proposed in Rahman & Carnielli 2000 which
permits the building of paraconsistent and paracomplete systems (systems reject-
ing tertium non-datur) for propositional and first-order logic, with their correspond-
ing tableaux (see remarks, criticisms and proposals for furthering in Van Bendegem

Principia 13(3): 283–97 (2009).

Meeting Hintikka’s Challenge to Paraconsistentism 291

Dialectical games and dialogical games can be seem to be symmetrically oppo-

site, as defended in Castelnérac & Marion. In dialectical games the Proponent must
first commit himself to a thesis A, to which the Opponent adduces, with the Propo-
nent’s agreement, a new hypothesis B which leads to an absurdum. In dialogical
logic, the situation is dually symmetrical: the Opponent is first committed to a thesis
A. Although this may add force to the above proposal of a “non-losing strategy” for
games that would interpret paraconsistent logics, I cannot completely discern the
relevance of such dual symmetry. For the purpose of this discussion it is important to
show how paraconsistent logics can indeed establish that the Opponent argues for
invalidity in dialectical games, as argued in Castelnérac & Marion.
In fact, if the Proponent commits himself to further theses (say, in conjunct B,
for simplicity) and at a final step of the dialogue the Opponent is able to show the
inconsistency of the Γ = {A, B}, and then inferring A from B are, as a matter of fact,
two distinct inferential moves in paraconsistent logics.
By borrowing the interpretation of the elenchus in Castelnérac & Marion:

1. The Proponent assert a thesis, A, which the Opponent (Socrates) considers

false and targets for refutation.

2. Socrates brings further premises into agreement, say collectively referred to

as B, which Socrates intends to use.

3. Socrates then argues (and the Proponent agrees), that B entails ¬A.

4. Socrates then claims that he has shown that ¬A is true and A false.

The main criticism against this interpretation by Vlastos (1994, 11), in short,
is (using logic notions not available to the Greeks, but in a benevolent sense) that
if Γ = {A, B} is inconsistent, then A, B ` ⊥, where ⊥ is the symbol for absurdity
or contradiction (falsum). This is in line with the fact that in usual logic systems
inconsistency and contradictions are indistinguishable, but if A, B ` ⊥ then B ` ¬A,
as the intended interpretation maintains; however, it also holds that A ` ¬B. This is
clearly incoherent with any claim of having shown A to be false and ¬A true.
It is not difficult to be convinced, indeed, that the Socratic method does not con-
cern justifying truth claims, but rather testing for logical coherence or consistency.
In favor of Vlastos program, it has to be said that in Vlastos 1982, 711, where com-
menting on the difference of what he calls “indirect elenchus” versus the “standard
elenchus” by the Socrates of Plato’s earlier dialogues, he does not only point out
that the former mode of argument is a potent instrument for exposing inconsistency
within the interlocutor’s beliefs, but at page 714 of Vlastos explicitly say:

Principia 13(3): 283–97 (2009).

292 Walter Carnielli

Success in elenctic argument need not show that one’s own beliefs are con-
sistent; it may show only that the opponent’s efforts to probe their inconsis-
tencies have been blocked by one’s superior dialectical skill. Socrates could
hardly be unaware of these hazards.

But if this is Socrates’ main instrument of philosophical investigation, why should

we not devote more attention to the minimal logic which supports Socratic reason-
ing, instead of committing Socrates to all the formidable arsenal of full “classical”
The logic mbC, briefly explained at Section 2, permits us to express the main
point underlying the elenchus starting from the consistency operator ◦. Indeed:

1. The proponent asserts a thesis, A

2. Socrates procures further premises, say B

3. Socrates then argues (with the Proponent’s agreement), that {A, B} is incon-
sistent; in our meticulous logic mbC, A, B `mbC ¬ ◦ (A ∧ B), but from this it
does not follow neither ¬A nor ¬B ;

4. Socrates has then driven the Opponent to an elenchus without any commit-
ment to ¬A being true, and far less to ¬B being true.

An important point is that, in (3), the fact that A, B `mbC ¬◦(A∧B) does not entail
in mbC neither ¬A nor ¬B can be proven through the possible-translation semantics
for mbC.
Although the consistency operator ◦ of mbC is very elementary, further prop-
erties of consistency(e.g., propagation of consistency through disjunction) can be
added by just attaching new axioms; details can be found in Carnielli, Coniglio &
Marcos 2007 and in Carnielli & Marcos 2002. The logic Ci, another member of the
family of the “Logics of Formal Inconsistency”, is obtained from mbC by adding (cf.
Theorem 102 in Carnielli, Coniglio & Marcos 2007) the following axiom schemas:
(ci) ¬◦α → (α ∧ ¬α),
(cf) ¬¬α → α.
It is to be noted that axiom (ci) establishes that the negation of consistency
implies contradiction, while axiom (cf) is just the reduction of negations.
Now one may define an inconsistency connective • by setting •α := ¬◦α. It can be
shown that ¬◦α and (α∧¬α) are equivalent in Ci; moreover, Ci has some (restricted)
forms of contraposition which were missing in mbC, such as α → ◦β ` ¬ ◦ β → ¬α.
But we can go a bit further. The logic Cio, another logic of formal inconsistency,
is obtained (see Carnielli, Coniglio & Marcos 2007, 71) by adding the following
axiom schemas to Ci:

Principia 13(3): 283–97 (2009).

Meeting Hintikka’s Challenge to Paraconsistentism 293

(co1) (◦α ∨ ◦β) → ◦(α ∧ β);

(co2) (◦α ∨ ◦β) → ◦(α ∨ β);
(co3) (◦α ∨ ◦β) → ◦(α → β).

Now, in Cio the law of “spreading consistency”, ◦α → ◦(α ∧ β), holds well;
together with the above contraposition law of Ci one thus easily derives:

`C io ¬ ◦ (α ∧ β) → ¬ ◦ α

which, from the elenchus

A, B `C io ¬ ◦ (A ∧ B)


A, B `C io ¬ ◦ A

but since

¬ ◦ A `C io ¬A

we in fact obtain

A, B `C io ¬A,

of course, at the cost of assuming a lot of logical laws. It is worth noting that all
such laws are classically valid, but it is clearly excessive to assume all of them in the
elenchic derivation. What seems to be conspicuous is that we do not need to assume
all the machinery of “classical” logic in order to logically express the elenchus. But it
is really relevant to make clear that we are not supposing any extravagant variety of
logic, but just a careful fragment.
The fact that this logical fragment is able to express elenchic derivations, by itself
raises serious doubts on an indulgently accepted tradition: ‘classical” logic does not
seem to coincide with the logic of the classics!
Of course, we also obtain in our analysis the undesired A, B `C io ¬B, which only
shows how excessive the extra laws are. It is also to be remarked that what one
derives is ¬A for a paraconsistent negation ¬. This is not classical negation yet;
in order to derive ∼ A , where ∼ has all properties of classical negation, we need a
further assumption on the degree of consistency (or ‘entrenchment’, as in Castelnérac
& Marion) of B. Only in a situation where one can secure this information on B such
a derivation is possible. In our setting, this means precisely assuming or granting
◦B. The fact that this additional information is necessarily extra-logic is argued by
Hintikka himself in Hintikka 2010:

Principia 13(3): 283–97 (2009).

294 Walter Carnielli

No logic in a genuine sense of the term can provide strategies for deciding
which putative information to reject or not to reject in a context of actual
material inquiry.

But we are left with a conceptual problem: in the “Logics of Formal Inconsis-
tency’, as we have seen, contradiction needs not to be identified with inconsistency,
and neither is consistency necessarily identifiable with non-contradiction (though
they may coincide). When not coincidental, what does consistency (represented by
the operator ◦) really mean?
Well, many attributes can be associated with joint denial without being repre-
sented by contradictory pairs. In a procedure, for instance, “duration” and “end”
are contrary ideas –they cannot hold together, but a procedure may has not been
started yet, and not finished. So they are contrary, but not contradictory. Or ‘all
courage is endurance of the soul’ and ‘some kind of courage is not endurance of
the soul’ which, again, cannot be both true but can be both false. To suppose that
the notion of contrary subsumes the notion of contradictoriness seems to be in line,
for instance, with Anton 1987; in this way contradictory attributes are just a subset
of contrary attributes, and “inconsistency” may apply to contrary attributes as well,
not only to contradictory ones. When driven to an elenchus, the proponent may be
contemplating an inconsistent set of premises in the sense of contrary premises, a
situation which is not acceptable, but which is not identified as a contradiction.
I do not mean that inconsistency is reducible to contrariness, just that it en-
compasses contradictoriness. Back to our logical discussion, assuming ◦(A ∧ B) may
mean, for instance, to assume that A and B are not contraries.
The are lots of misunderstanding, possibly by lack of attention to the relevant lit-
erature. In Pietarinen 2002, for instance, when referring to the well-known Jaskow-
ski’s problem A. Pietarinen, perhaps unpremeditatedly, says that: “This problem [of
finding a logic deserving the label name of paraconsistent] has proved elusive thus
far, as the received paraconsistent logics fall short of having a satisfactory semantic
Forms of dialectical (and dialogical) games are, of course, played today, and
they can solve informational conflicts and provide new information. In a criminal
investigation, when interviewing suspects, a detective will only know that one of
them is lying if an inconsistency (what we may call a tripartite elenchus) arises.
The reason Hintikka in (2010: 12) tags such useful information as an “illusion” is
disconcerting, for the least:
Perhaps the illusion that paraconsistent logics could tell us something about
the rejection of information is due to an uncertainty about the meaning of
negation in paraconsistent logic.

Principia 13(3): 283–97 (2009).

Meeting Hintikka’s Challenge to Paraconsistentism 295

5. How far can we go? Mind the gap!

The rapprochement between paraconsistency and independence-friendly modes of

thinking is due to Hintikka’s perspicuity, but the issue raises many more questions
than he is perhaps willing to admit. It may be true (see quotation above) that the
impact of independence-friendly logic is one of the best things that could have hap-
pened to paraconsistent logic, but the reciprocal is also worthy of admission.
The paraconsistency program seems to be broader than the independence-friend-
ly movement: certain dialogical games have a paraconsistent character, as shown in
Rahman & Carnielli 2000, and certain dual IF logic, as the ‘paratyconsistent’ logic
sketched by Hintikka, will be paraconsistent.
The insistence of non-compositional semantics as a “semantical solution” for
paraconsistency does not seem to be very fruitful, as argued at Section 2. Game
theoretical semantics (emphasizing non-losing strategies, rather than winning strate-
gies) can be given to paraconsistent logics, or at least to ‘paratyconsistent’ logic, but
paraconsistent logics have other semantics available, as the possible-translations se-
If the internal logic of interrogative games (so important in Hintikka’s view on
Aristotle’s theory) should be paraconsistent as he admits, the analogous roles be-
tween the verifier and falsifier in game semantics, and the Proponent and the Op-
ponent in dialectical games, does not seem to give any definite reason to dismiss
“the formal games of theorem-proving” (Hintikka 2010: 12). On the contrary, the
theorem-proving machinery of the Logics of Formal Inconsistency, which represent
a good majority of paraconsistent logics, permits accurate expression of the internal
logic of an elenchus.
A consequence of narrowing the gap between paraconsistent and independence-
friendly logics is that difficult questions are posed for both sides, and the conciliation
also deeply defies theorists of IF logic as well as of paraconsistency. Borrowing from
Hintikka: what else could logicians hope for?

Abramsky, S. & Vänänen, J. 2009. From IF to BI: a tale of dependence and separation. Syn-
these 167(2): 207–30.
Anton, J. P. 1987. Aristotle’s Theory of Contrariety. Lanham: University Press of America.
Bueno-Soler, J. & Carnielli, W. A. 2005. Possible-translations algebraization for paraconsis-
tent logics. Bulletin of the Section of Logic 34(2): 77–92.
Cameron, P. & and Hodges, W. 2001. Some Combinatorics of Imperfect Information. Journal
of Symbolic Logic 66(2): 673–84.

Principia 13(3): 283–97 (2009).

296 Walter Carnielli

Carnielli, W. A. 1990. Many-valued logics and plausible reasoning. Proceedings of the Twen-
tieth International Symposium on Multiple-Valued Logic. Charlotte, NC: IEEE Computer
Society, 328–35.
———. 2000. Possible-translations semantics for paraconsistent logics. In Batens, D. et al.
(eds.) Frontiers of Paraconsistent Logic: Proceedings of the I World Congress on Paracon-
sistency, Logic and Computation Series. Baldock: Research Studies Press, King’s College
Publications, 149–63.
Carnielli, W. A. & Marcos, J. 2002. A Taxonomy of C-Systems. In W. A. Carnielli; M. E.
Coniglio; I. M. L. D’Ottaviano (eds.) Paraconsistency: the Logical Way to the Inconsistent.
(Lecture Notes in Pure and Applied Mathematics, v. 228.) New York: Marcel Dekker,
1–94. Preprint available at CLE e-Prints 1(5).
Carnielli, W. A.; Coniglio, M. E.; Marcos, J. 2007. Logics of formal inconsistency. In D. Gabbay
and F. Guenthner (eds.) Handbook of Philosophical Logic, v. 14, 2nd. ed. Springer, 15–107.
Preprint available at CLE e-Prints 5(1).,n_1,2005.html
Carnielli, W. A.; Coniglio, M. E.; D’Ottaviano, I. M. L. 2009. New dimensions on translations
between logics. Logica Universalis 3(1): 1–18.
Castelnérac, B. & Marion, M. (to appear) Arguing for inconsistency: dialectical games in the
academy. In G. Primiero and S. Rahman (eds.) Acts of Knowledge: History, Philosophy and
Logic. London: College Publications.
Cook, R. & Shapiro, S. 1998. Hintikka’s Revolution: “The Principles of Mathematics Revis-
ited.” Brit. J. Phil. Sci 49: 309–16.
Hintikka, J. 1996. Principles of Mathematics Revisited. Cambridge: Cambridge University
———. 2007. Socratic Epistemology: Explorations of Knowledge-Seeking by Questioning.
New York: Cambridge University Press.
———. 2004. Analyses of Aristotle. Dordrecht: Kluwer Academic Publishers.
———. 2010. IF Logic Meets Paraconsistent Logic. In W. A. Carnielli, M. E. Coniglio, and
I. M. L. D’Ottaviano (eds.) The Many Sides of Logic. London: College Publications, 3–13.
Hintikka, J.; Halonen, I.; Mutanen, A. 1999. Interrogative logic as a general theory of rea-
soning. In R. Johnson and J. Woods (eds.) Handbook of Applied Logic. Dordrecht: Kluwer
Academic Publishers.
Hintikka, J. & Sandu, G. 1996. Game-theoretical semantics. In Johan van Benthem and Alice
ter Meulen (eds.) Handbook of Logic and Language. Amsterdam: Elsevier.
Hodges, W. 1997a. Compositional semantics for a language of imperfect information. Logic
Journal of the IGPL 5: 539–63.
———. 1997b. Some strange quantifers. In G. R. J. Mycielski and A. Salomaa (eds.) Struc-
tures in Logic and Computer Science. Vol. 1261 of Lecture Notes in Computer Science.
Springer, 51–65.
Kapp, E. 1942. Greek Foundations of Traditional Logic. New York: Columbia University Press.
Loparić, A. & da Costa, N. C. A. 1984. Paraconsistency, paracompleteness and valuations.
Logique et Analyse 106: 119–31.
Marcos, J. 1999. Semânticas de Traduções Possíveis (Possible-Translations Semantics, in
Portuguese). Master’s thesis, State University of Campinas (Unicamp), Campinas.

Principia 13(3): 283–97 (2009).

Meeting Hintikka’s Challenge to Paraconsistentism 297
Pietarinen, A. 2002. Some Useful Social Metaphors in Logic. In H. R. Arabnia and Y. Mun
(eds.) Proceedings of the International Conference on Artificial Intelligence. CSREA Press,
Rahman, S. & Carnielli, W. A. 2000. The dialogical approach to paraconsistency. Synthese
125(1–2): 201–32.
Van Bendegem, J. P. 2001. Paraconsistency and dialogue logic: critical examination and fur-
ther explorations. Synthese 127(1–2): 35–55.
Vlastos, G. 1982. The Socratic Elenchus. The Journal of Philosophy 79(11): 711–14.
———. 1994. Socratic Studies. Cambridge: Cambridge University Press

Centre for Logic, Epistemology and the History of Science – CLE
and Department of Philosophy
State University of Campinas – UNICAMP
P.O. Box 6133
13083-970 Campinas, SP

Resumo. Em uma série de seminários e conferências no Brasil em 2008, Jaakko Hintikka,

em uma série de palestras no Brasil em 2008, defendeu que a “IF-lógica” (“independence
friendly logic”) e a lógica paraconsistente são, em certo sentido, bastante similares. A par-
tir do esboço de um novo sistema paraconsistente, ele afirma que várias potencialidades da
IF-lógica podem ser reproduzidas na lógica paraconsistente. Uma das grandes dificuldades,
deixada como um desafio, seria a formulação de condições de verdade para esta nova lin-
guagem paraconsistente de primeira ordem, com vistas a garantir uma maior inter-relação
entre esta e a IF-lógica. Argumento que tais condições de verdade não somente sugerem
uma abordagem inovadora em semântica de jogos para a paraconsistência, mas também
que a partir deste ponto de vista as “Lógicas da Inconsistência Formal” podem modelar a
lógica interna dos elencos socráticos. Pretendo discutir estes e outros pontos levantados por
Hintikka, tomando-os como desafios e oportunidades para o paraconsistentismo e para a
IF-lógica, bem como levantar algumas críticas sobre a visão de Hintikka sobre a paraconsis-

Palavras-chave: IF-lógica, paraconsistência, elencos socráticos.

There are some evidences that Hintikka would agree to this suggestion, which I proposed
during his seminars at UNICAMP in 2008.

Principia 13(3): 283–97 (2009).

State University of Campinas

Abstract. The Logics of Deontic (In)Consistency (LDI’s) can be considered as the deontic
counterpart of the paraconsistent logics known as Logics of Formal (In)Consistency. This
paper introduces and studies new LDI’s and other paraconsistent deontic logics with different
properties: systems tolerant to contradictory obligations; systems in which contradictory
obligations trivialize; and a bimodal paraconsistent deontic logic combining the features of
previous systems. These logics are used to analyze the well-known Chisholm’s paradox,
taking profit of the fact that, besides contradictory obligations do not trivialize in LDI’s,
several logical dependencies of classical logic are blocked in the context of LDI’s, allowing to
dissolve the paradox.

Keywords: Paraconsistent logic, deontic logic, deontic paradoxes, chisholm’s paradox, moral
dilemmas, logics of formal inconsistency.


The phenomenon of inconsistency has catastrophic consequences in theories based

on classical logic. There is a simple reason to this: in the context of classical logic, a
contradiction like α and ¬α trivializes the whole theory, sinc everything is logically
derived from a contradiction. Paraconsistent logics, developed independently by da
Costa 1963 and Jaśkowski 1948, are logics tolerant to contradictions, and so they
admit non-trivial inconsistent (closed) theories.
The Logics of Formal (In)Consistency (LFI’s), introduced in Carnielli & Marcos
2002 and also studied in Carnielli, Coniglio & Marcos 2007, are a special class of
paraconsistent logics in which there is a formal distinction between notions such as
contradiction, inconsistency and triviality. This is attained by the use in the formal
language of connectives for consistency and inconsistency.
A phenomenon analogous to inconsistency takes place with deontic logics, in the
context of moral dilemmas (see, for instance, McConnell 2006). If O denotes the
deontic “obligation” operator, a moral dilemma is a situation of conflicting or con-
tradictory obligations of the form Oα and O¬α. Note that this is not a particular
case of contradictoriness like Oα and ¬Oα, but a different (however related) situ-
ation. In the presence of conflicting obligations, classical deontic logics trivialize,

Principia 13(3): 299–326 (2009).

Published by NEL — Epistemology and Logic Research Group, Federal University of Santa Catarina (UFSC), Brazil.
300 Marcelo Esteban Coniglio & Newton Marques Peron

in the same manner as classical logic trivializes when a contradiction occurs. This
similarity justify a paraconsistentist approach to deontic paradoxes.
The first paraconsistent deontic system, called C1D , was introduced in da Costa &
Carnielli 1986 with the aim of analyzing moral dilemmas. It is a deontic extension of
da Costa’s logic C1 , being also suitable to the study of deontic paradoxes in general.
The Logics of Deontic (In)Consistency (LDI’s) were firstly introduced in Coniglio
2007, by considering two deontic systems, DmbC and DLFI1, based on two different
LFI’s. It was shown that these systems are appropriate to analyze the well-known
deontic paradox of Chisholm introduced in Chisholm 1963.
In Peron & Coniglio 2008 two new LDI’s were introduced, SDmbC and BDmbC,
enlarging the possibilities of analysis of Chisholm’s paradox, and in Peron 2009 it
was presented the distinction between LDI’s and deontic LFI’s, among other results.
The present paper extends our previous discussions on the subject, in particular
the analysis of Chisholm’s paradox in this context. Moreover, two new paraconsis-
tent deontic systems are introduced, which can be considered, in some sense, as
minimal: DPI, a LDI which is not a LFI, and SDPI, both a LDI and a LFI such that
conflicting obligations trivialize, provided that the involved sentence is deontically
consistent. Concerning Chisholm’s paradox, it is concluded here that the violation of
the Principle of Deontic Consistency is not the unique source of problems in deon-
tic paradoxes based on contrary-to-duty obligations. Indeed, the rigidity of classical
logic, in which too much sentences are logically identified, originates unnecessary
logical dependencies among the premises of a given information set, and then sev-
eral plausible formalizations of the given premises must be discarded. The use of
a richer language based on a more flexible logic allows more translations from the
natural language to the formal one, and so the paradox can vanish. If, in addition,
the Principle of Deontic Consistency is weakened, there are still more possibilities to
avoid the paradox.
The organization of the paper is as follows: in Section 1 the basic notions con-
cerning paraconsistency in general and LFI’s in particular are briefly recalled. In
Section 2 it is firstly analyzed the Standard Deontic Logic SDL and then the LDI’s
are introduced, describing succinctly the main systems to be studied in the paper.
Section 3 is devoted to the analysis of Chisholm’s paradox under the perspective of
some systems, namely SDL, DmbC, SDmbC, DPI, SDPI and BDmbC. Finally, Sec-
tion 4 concludes with a brief evaluation of what was done in this paper.

1. Paraconsistency and Formal (In)Consistency

Da Costa’s approach to paraconsistency is based on the idea that certain sentences

of the object language can be marked as being “classical” or “well-behaved” (cf. da

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 301

Costa 1963). More precisely, in a logical context in which contradictions are possible,
in the sense that a contradiction does not necessarily trivialize, it is possible to define
or to assume as hypothesis inside a formal deduction that an specific sentence, say
α, has classical properties. The classical (or well) behavior of α is another sentence,
¬(α ∧ ¬α), usually denoted by α◦ . In order to simplify the exposition we limit
ourselves to the logic C1 , the first logic in the decreasing hierarchy of paraconsistent
logics (Cn )n≥1 . Being so, in general

α, ¬α 6` β

for some β, but always

α◦ , α, ¬α ` β
for every β (here, ‘`’ is the symbol for the consequence relation, and so ‘6`’ means
“does not follow”). That is, contradictions do not necessarily trivialize, but contra-
dictions of classically (or well) behaved sentences always trivialize (or “explode”).
There is another way of interpreting da Costa’s approach to paraconsistency:
a contradiction is not enough to trivialize, but a contradiction of a contradiction
α, ¬α 6` β
for some β, but always
(α ∧ ¬α), ¬(α ∧ ¬α) ` β
for every β. In other words, contradictions are not logically trivial, but a contradic-
tion involving a contradiction is!
It can be observed that the statement that α is well-behaved, namely α◦ , is ex-
pressed in terms of the other connectives of the object language. Thus, if (α ∧ ¬α)
can be read as “α is contradictory” then α◦ says that “α is not contradictory”. Being
so, non-contradictoriness is equivalent to well-behaviorness, or classicality, or con-
sistency. But, are the notions of non-contradiction and consistency (or, dually, the
notions of contradiction and inconsistency) necessarily the same?
This is the standpoint of Logics of Formal (In)Consistency (in short, LFI’s), intro-
duced in Carnielli & Marcos 2002 with the aim of generalizing da Costa’s C-systems.
In this approach the notion of well-behavior or consistency of a sentence α is repre-
sented by the sentence ◦α, where ◦ is a primitive connective of the object language
(and, different to the logics Cn , in general this connective cannot be expressed in
terms of the others). Dually, it is possible to consider a inconsistency connective •
as a primitive, such that the interdefinability laws ◦α ≡ ¬•α and •α ≡ ¬◦α hold in
In term of consistency, LFI’s has the following property (compare with C1 above):

α, ¬α 6` β

Principia 13(3): 299–326 (2009).

302 Marcelo Esteban Coniglio & Newton Marques Peron

for some β, but always

◦α, α, ¬α ` β
for every β. But, different to da Costa’s systems, the consistency operator ◦ is primi-
tive and is not necessarily identical to non-contradiction: in the weaker LFI’s, ◦α and
¬(α ∧ ¬α) are not equivalent. Moreover, in general

◦α ` ¬(α ∧ ¬α) but ¬(α ∧ ¬α) 6` ◦α;

¬◦α ` (α ∧ ¬α) but (α ∧ ¬α) 6` ¬◦α.

Thus, in the context of LFI’s the notions of consistency and non-contradiction, as
well as the notions of inconsistency and contradiction, can be separated. The steps
through the definition of LFI’s are briefly outlined below (the interested reader can
consult Carnielli & Marcos 2002 and Carnielli, Coniglio & Marcos 2007).
Let L be a (propositional) logic having a negation connective ¬. A theory Γ of
L (that is, a set Γ of sentences of L) is contradictory if Γ ` α and Γ ` ¬α for some
sentence α. Obviously, Γ is not contradictory if and only if, for every α, either Γ 6` α
or Γ 6` ¬α. A theory Γ is trivial if Γ ` α for every α, and is non-trivial otherwise.
If Γ, α, ¬α ` β for every α and β the theory Γ is said to be explosive. A logic L is
non-trivial if it has a non-trivial theory, and it is explosive if every theory is explosive
(equivalently, if α, ¬α ` β for every α and β). A logic L is paraconsistent if it is
non-trivial and non-explosive.
Let (p) be a set of sentences depending exactly on the propositional letter p
such that

(a) (α), α 0 β

(b) (α0 ), ¬α0 0 β 0

for some α, α0 , β and β 0 . A logic L is gently explosive (w.r.t. (p)) if

(c) (α), α, ¬α ` β

for every α and β. Given a sentence α, the set (α) expresses the consistence of
α relative to logic L (and negation ¬). Whenever the set (p) is a singleton, the
unique element of it will be denoted by ◦p and then the sentence ◦α denotes the
consistency of α, where ◦ is called a consistency operator. By the very definition of
consistency operator, ◦α, α, ¬α ` β for every α and β.

Definition 1.1. A Logic of Formal Inconsistency (LFI) is a paraconsistent logic having

a set (p) satisfying properties (a), (b) and (c) as above. „

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 303

It is immediate to see that logic C1 is a LFI with ◦α = α◦ = ¬(α ∧ ¬α). Moreover,

every logic Cn (for n ≥ 1) is a LFI with an appropriate definition of the consistency
operator ◦n . In fact, consider the following recursive definition: α1 = α◦ , and αn+1 =
(αn )◦ (for n ≥ 1). Finally, let α(1) = α◦ and α(n+1) = α(n) ∧ αn+1 (for n ≥ 1). Then,
consistency in logic Cn is given by the operator ◦n p = p(n) or, equivalently, by the set
n (p) = {p1 , p2 , . . . , p n } (for n ≥ 1). It is interesting to observe that the logic C Lim
(cf. Carnielli & Marcos 1999), the deductive limit of the hierarchy (Cn )n≥1 , is a LFI
in which consistency is expressed by the infinite set (p) = {p1 , p2 , p3 , . . .}. That
is, there are LFI’s in which consistency is expressed by an infinite set of sentences
instead of a single consistency operator.
The simplest LFI introduced in Carnielli & Marcos 2002 is called mbC, defined
as follows:

Definition 1.2. The logic mbC is defined over the language ◦, ¬, ∧, ∨, → by means
of the following:
Axiom schemas:

(Ax1) α → (β → α)

(Ax2) (α → β) → (α → (β → γ)) → (α → γ))

(Ax3) α → (β → (α ∧ β))

(Ax4) (α ∧ β) → α

(Ax5) (α ∧ β) → β

(Ax6) α → (α ∨ β)

(Ax7) β → (α ∨ β)

(Ax8) (α → γ) → ((β → γ) → ((α ∨ β) → γ))

(Ax9) α ∨ (α → β)

(Ax10) α ∨ ¬α

(bc1) ◦α → (α → (¬α → β))

Inference rule:
α α→β
β „

Principia 13(3): 299–326 (2009).

304 Marcelo Esteban Coniglio & Newton Marques Peron

From now on, the set of sentences defined over the basic language ¬, ∧, ∨, → will
be denoted by For, and the set of sentences of mbC (which is obtained by adding
the unary connective ◦) will be denoted by For ◦ .
It is worth noting that the system defined over For by means of (Ax1)–(Ax9)
and (MP) is the so-called Positive Classical Logic, denoted by PCL. Classical Logic
CL defined over For is obtained from PCL by adding (Ax10) and the “explosion law”

(exp) α → (¬α → β)

Of course a version of CL over For ◦ , called eCL, can be obtained from mbC by
adding the axiom schema ◦α; in this sense CL is a deductive extension of mbC in
which every sentence is consistent (and so (bc1) collapses to (exp)). On the other
hand, ⊥ =d e f ◦α∧(α∧¬α) is a bottom sentence in mbC, that is, ⊥ `mbC β for every
β and so ∼α =d e f (α → ⊥) is a classical negation in mbC. Therefore mbC defined
over the set of sentences For ◦∼ generated by ∼, ◦, ¬, ∧, ∨, → (that is, with classical
negation ∼ as a primitive) can be seen as the conservative extension of CL (now
defined over the language ∼, ∧, ∨, →) obtained by adding axiom schemas (Ax10),
(bc1) and ∼α ↔ (α → ⊥) where, as usual, (α ↔ β) stands for (α → β) ∧ (β → α).
The logic mbC (as most of the LFI’s presented in Carnielli & Marcos 2002 and
Carnielli, Coniglio & Marcos 2007) does not satisfy the replacement property, namely:

if α a` β then ϕ(α) a` ϕ(β)

for every α, β, ϕ (here, α a` β means that α and β are interderivable, that is,
α ` β and β ` α). For instance, (p ∧ q) is interderivable with (q ∧ p), but ¬(p ∧
q) is not interderivable with ¬(q ∧ p); analogously, ◦(p ∧ q) is not interderivable
with ◦(q ∧ p). This means that ¬ and ◦ are non-truth-functional connectives and
so semantics is not a trivial issue for mbC (nor for the other LFI’s with non-truth-
functional connectives). In Carnielli, Coniglio & Marcos 2007 the following two-
valued semantics was proposed for mbC:

Definition 1.3. Let 2 = {0, 1} be a set of truth-values, where 1 denotes “true” and 0
denotes “false”. A mbC-valuation is a mapping v : For ◦ −→ 2 satisfying the following

(v1) v(α ∧ β) = 1 iff v(α) = v(β) = 1,

(v2) v(α ∨ β) = 0 iff v(α) = v(β) = 0,
(v3) v(α → β) = 0 iff v(α) = 1 and v(β) = 0,
(v4) v(¬α) = 1 implies v(α) = 0,
(v5) v(◦α) = 1 implies v(α) = 0 or v(¬α) = 0. „

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 305

As usual, we define Γ mbC α iff v(α) = 1 for every mbC-valuation v such that
v(Γ) ⊆ {1}. Thus, the logic mbC is sound and complete for its valuation semantics,
cf. Carnielli, Coniglio & Marcos 2007:

Theorem 1.4. Let Γ ∪ {α} be a set of sentences in For ◦ . Then, Γ `mbC α if and
only if Γ mbC α.

2. Logics of Deontic (In)Consistency

2.1. Standard Deontic Logic and Paraconsistency

Deontic logics study notions such as “obligation”, “permission”, “prohibition” and
similar concepts. Deontic logics were strongly influenced by notions from modal
logic. Despite the analogy between modal and deontic concepts can be traced to the
XIV century (cf. McNamara 2006), it can be argued that the symbolic and mathemat-
ical approach to deontic logics was inaugurated in the celebrated paper (von Wright
1951). In that article, von Wright distinguishes three kinds of modalities: alethic,
epistemic and deontic. The first ones deal with concepts such as “necessary” and
“possible”; the second ones concern notions as “verifiable” and “falsifiable”, whereas
the latter are related to notions such as “obligatory” and “permitted”.
There exist several systems formalizing deontic logic (see for instance Prakken
& Sergot 1994). The basic system is called SDL – Standard Deontic Logic. The idea
behind this logic is to extend the basic modal system, in which the usual box operator
ƒ for necessitation is substituted by O (meaning “it is obligatory that”) by adding the
restriction that conflicting obligations are not allowed. This restriction is formulated
by means of axiom (O-E) below. The following axiomatization of SDL presented in
Coniglio 2007 is appropriate to our purposes.

Definition 2.1. The logic SDL is defined over the language O, ¬, ∧, ∨, → by means
of the following:
Axiom schemas:
(Ax1)–(Ax10) from mbC (recall Definition 1.2) plus

(exp) (α → (¬α → β))

(O-K) O(α → β) → (Oα → Oβ)

(O-E) Ofα → fα where fα =d e f (α ∧ ¬α)

Inference rules:

Principia 13(3): 299–326 (2009).

306 Marcelo Esteban Coniglio & Newton Marques Peron

α α→β

` Oα „

From now on, the set of sentences generated by the language O, ¬, ∧, ∨, → will
be denoted by For O .
The usual axiomatizations of SDL depart from the set of axioms formed by the
instances in the deontic language of the classical tautologies (arguably, this is a lazy
and inelegant way to present an axiomatic extension of classical logic) and then
consider, instead of (O-E), the axiom schema

(D) Oα → ¬O¬α
(usually called the Principle of Deontic Consistency) and the inference rules (MP)
and (O-Nec). It is useful and convenient to consider the derived deontic operator
Pα =d e f ¬O¬α (meaning “α is permitted” or “α is permissible”). Thus, (D) can be
interpreted as follows: whatever is obligatory is permitted. Since classically ¬β is
equivalent to β → fα (for any α and β) then (D) can be reformulated as:

(D*) ¬(Oα ∧ O¬α)

stating that conflicting obligations are always false. Moreover, it can be proved that
(O-E), (D) and (D*) are equivalent in the presence of classical logic plus (O-K),
(O-Nec) and (MP).
An interesting alternative axiomatization of SDL is proposed in Chellas 1980 by
substituting (O-K) and (O-Nec) by the following:

Oα → Oβ
(OC) (Oα ∧ Oβ) → O(α ∧ β)

(ON) O>

(OD) ¬O⊥

where > can be taken as a classical tautology, for instance α ∨ ¬α. Indeed, (OD) is
an alternative formulation of (D*), while (ON) and (O-Nec) are closely related. By
its turn, (O-K) can be derived from (ROM) and (OC).
Concerning semantics, and being deontic logics particular cases of modal logics,
the well-known Kripke semantics of possible-worlds offer an appropriate framework.

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 307

Definition 2.2. A (generalized) Kripke struture is a triple 〈W, R, {vw }w∈W 〉 where:

1. W is a non-empty set (of possible-worlds);

2. R ⊆ W × W is a relation (of accessibility) between worlds;

3. {vw }w∈W is a family of mappings vw : For O → 2 such that:

(k1) vw (α ∧ β) = 1 iff vw (α) = vw (β) = 1,

(k2) vw (α ∨ β) = 0 iff vw (α) = vw (β) = 0,
(k3) vw (α → β) = 0 iff vw (α) = 1 and vw (β) = 0,
(k4) vw (¬α) = 1 iff vw (α) = 0,
(k5) vw (Oα) = 1 iff vw 0 (α) = 1 for every w 0 in W such that wRw 0 . „

Let M = 〈W, R, {vw }w∈W 〉 be a a Kripke structure, w ∈ W and α a sentence. We

say that α is satisfied by M in w, denoted by M , w  α, if vw (α) = 1; α is satisfiable
in M if M , w  α for some w ∈ W , and α is true in M , denoted by M  α, if M , w  α
for every w ∈ W . Given a class K of Kripke structures, a sentence α is valid (w.r.t.
K ) if, for every M ∈ K , M  α; and α is satisfiable (w.r.t. K ) if it is satisfiable for
some M ∈ K . Finally, the consequence relation K is defined as follows: Γ K α iff
there exists a finite set {α1 , . . . , αn } ⊆ Γ such that (α1 ∧ . . . ∧ αn ) → α is valid (w.r.t.
K ).

Theorem 2.3. Let KSDL be the class of Kripke strutures such that the accessibility
relation R is serial, that is: for every w ∈ W there exists w 0 ∈ W such that wRw 0 .
Then SDL is sound and complete for KSDL , that is: Γ `SDL α iff Γ K α.
Now it is convenient to establish a parallel between some metalogical notions
from classical and deontic logics. Recall that in classical logic a contradictory theory
is trivial, and vice-versa. On the other hand, as observed in da Costa & Carnielli
1986, if a theory Γ derives conflicting obligations Oα and O¬α in SDL then Γ is
deontically trivial, that is, Γ ` Oβ for every β (and, a posteriori, it is trivial: Γ ` β
for every β). The converse also holds, and so deontic contradictoriness is equivalent
to deontic triviality, transposing to the deontic context a situation from classical logic.
In the previous section it was shown that LFI’s incorporate metalogical notions
such as consistency into the object language. This suggest us that some axioms as
(ON) and (OD) can be treated as metalogical notions which can be internalized
in the object language. Adapting to the modal context the definitions presented in
Section 1, the following definitions are proposed:

Principia 13(3): 299–326 (2009).

308 Marcelo Esteban Coniglio & Newton Marques Peron

Definition 2.4. Let L be a logic defined over the set For O , and let Γ be a theory of
L. Then, Γ is said to be deontically contradictory or deontically conflicting (w.r.t. O) if
Γ ` Oα and Γ ` O¬α for some sentence α. The theory Γ is deontically trivial if Γ `
Oα for every α. If Γ, Oα, O¬α ` Oβ for every α and β then Γ is said to be deontically
explosive. The logic L is deontically non-trivial if it has a deontically non-trivial theory,
and it is deontically explosive if every theory is deontically explosive (equivalently, if
Oα, O¬α ` Oβ for every α and β). The logic L is deontically paraconsistent if it is
deontically non-trivial and deontically non-explosive. „

It is worth noting that a deontic contradiction is of the form Oα ∧ O¬α instead

of Oα ∧ ¬Oα. Continuing with the analogy with LFI’s, let Œ(p) be a set of sentences
in For O depending exclusively on the propositional letter p such that, for some
sentences α, α0 , β and β 0 ,

(Da) Œ(α), Oα 1 Oβ
(Db) Œ(α0 ), O¬α0 1 Oβ 0

A logic L is deontically gently explosive (w.r.t. Œ(α)) if

(Dc) Œ(α), Oα, O¬α ` Oβ)

for every α and β. Given a sentence α, the set Œ(α) expresses the deontic consistence
of α relative to logic L (and negation ¬ and obligation O). Whenever the set Œ(p) is
a singleton, their unique element will be denoted by Œp and then the sentence Œα
denotes the deontic consistency of α, where Œ is called a deontic consistency operator.
The following definition is the deontic version of Definition 1.1

Definition 2.5. A Logic of Deontic Inconsistency (LDI) w.r.t. O and ¬ is a deontically

paraconsistent logic (w.r.t. O and ¬) having a set Œ(p) satisfying properties (Da),
(Db) and (Dc) as above. „

It should be noticed that a deontically paraconsistent logic (w.r.t. a negation ¬)

is necessarily paraconsistent (w.r.t. the same negation ¬). Indeed, by (exp), (O-
Nec) and (O-K), an explosive deontic logic is, a posteriori, deontically explosive.
Therefore, a deontically paraconsistent logic is, by force, a paraconsistent deontic
On the other hand, it is possible to define paraconsistent deontic logics, that is,
to enrich a given paraconsistent logic with a deontic operator, while preserving the
property that conflicting obligations trivialize. Thus, a paraconsistent deontic logic
is not necessarily a deontically paraconsistent logic. The analysis of systems of both
types is the topic of the next section.

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 309

2.2. Paraconsistent Deontic Logics and Deontically Paraconsistent

The first paraconsistent deontic system proposed in the literature, called C1D , was
introduced in da Costa & Carnielli 1986 and it was also studied in Puga, da Costa &
Carnielli 1988. As suggested by its name, it is a deontic extension of C1 , in which
some interesting relations between the consistency operator ·◦ and the deontic oper-
ator O arise, as we shall see below.
Related to the approach mentioned above, in Cruz 2005 it was studied a deon-
tic extension of C1 but, instead of taking (O-E), a stronger version was considered:
Oα → α (observe that (O-E) is a particular case, with α substituted by fα ). Another
difference with the previous works is that the deontic operator O cannot be iterated,
and so sentences such as OOα and O(α → Oβ) are not allowed. A paraconsistent
dyadic modal logic was also considered, extending C1 with a dyadic deontic oper-
ator Oβ α meaning “α is obligatory in context (or under condition) β”. But, again,
an excessively strong axiom was considered: Oβ α → (α ∧ β). Being so, some deon-
tic paradoxes as Chisholm’s paradox (see Section 3 below) cannot be satisfactorily
treated in this framework.
In the rest of this section some paraconsistent deontic system will be analyzed.

2.2.1. The logic DmbC

In Coniglio 2007 a deontic dimension was added to mbC, a minimal LFI (recall
Definition 1.2), by appropriately adapting the axiomatization of SDL presented in
Definition 2.1. The resulting system, called DmbC, has the consistency operator ◦ as
primitive (since it is based on mbC) but no interaction between ◦ and O is required.

Definition 2.6. The logic DmbC is defined over the language O, ◦, ¬, ∧, ∨, → by add-
ing to mbC (recall Definition 1.2) the following:
Axiom schemas:

(O-K) O(α → β) → (Oα → Oβ)

(O-E)◦ O⊥α → ⊥α where ⊥α =d e f (α ∧ ¬α) ∧ ◦α

Inference rule:

` Oα „

Alternatively, DmbC can be defined from the rules and axioms of SDL over For ◦O
(the set of sentences generated by the language O, ◦, ¬, ∧, ∨, →) by replacing (exp)

Principia 13(3): 299–326 (2009).

310 Marcelo Esteban Coniglio & Newton Marques Peron

by (bc1) and (O-E) by (O-E)◦ , respectively. It can be proved that DmbC is both
a LFI and a LDI. Thus, it has explicit operators for both consistency and deontic
consistency. This fact will be proved in Theorem 2.8 below. The first step is to
define a Kripke semantics for DmbC. This is obtained by modifying appropriately
Definition 2.2 as follows:

Definition 2.7. A Kripke struture for DmbC is a triple 〈W, R, {vw }w∈W 〉 where:

1. W is a non-empty set (of possible-worlds);

2. R ⊆ W × W is a relation (of accessibility) between worlds;

3. {vw }w∈W is a family of mappings vw : For ◦O → 2 such that:

(kc1) vw (α ∧ β) = 1 iff vw (α) = vw (β) = 1,

(kc2) vw (α ∨ β) = 0 iff vw (α) = vw (β) = 0,
(kc3) vw (α → β) = 0 iff vw (α) = 1 and vw (β) = 0,
(kc4) vw (¬α) = 0 implies vw (α) = 1,
(kc5) vw (α) = vw (¬α) = 1 implies vw (◦α) = 0,
(kc6) vw (Oα) = 1 iff vw 0 (α) = 1 for every w 0 in W such that wRw 0 . „

Soundness and completeness of DmbC w.r.t. the semantics above was proved in
Coniglio 2007. As a matter of fact, this result, as well as a similar result for system
SDmbC to be defined below, are particular cases of a general completeness theorem
stated in Bueno-Soler 2008 for logics based on LFI’s enjoying the so-called G k,l,m,n
axiom (cf. Carnielli & Pizzi 2008).

Theorem 2.8. DmbC is both a LFI and a LDI.

Proof. Consider a Kripke model with W = {w}, wRw and vw such that vw (p) =
vw (¬p) = 1 and vw (q) = 0 for propositional letters p 6= q. Then it is easy to see that
p, ¬p 1DmbC q and O p, O¬p 1DmbC Oq, and so DmbC is neither explosive nor
deontically explosive. In particular, DmbC is neither trivial nor deontically trivial.
Using axiom (bc1) it follows that DmbC is a LFI in which the consistency of α is
expressed by ◦α. Finally, let Œα =d e f O◦α. By (O-Nec) applied to (bc1), followed
by (O-K) and (MP) it follows that DmbC is a LDI in which the deontic consistency
of α is expressed by Œα. „
It is easy to see that by adding to DmbC the axiom schema ◦α it is obtained
a version of SDL over For ◦O , denoted by eSDL. In fact, as proved above for mbC,

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 311

axiom (exp) follows in that axiomatic extension of DmbC. On the other hand, by
(O-Nec), O◦α is a theorem for every α and so (O-E) follows.
As observed in Carnielli, Coniglio & Marcos 2007, the inconsistency operator •
can be defined in mbC as •α =d e f ∼◦α in terms of the classical negation ∼, where
∼α =d e f (α → ⊥) (the expected definition •α =de f ¬◦α in terms of the paracon-
sistent negation ¬ just works in stronger systems). Inspired by this, it is possible to
define in DmbC an operator for deontic inconsistency as follows:

‚α =d e f O∼◦α.

Since ◦α, ∼◦α ` ⊥ then O◦α, O∼◦α ` O⊥ and so Œα, ‚α ` ⊥, by (O-E)◦ . This
means that ‚α ` ∼Œα and Œα ` ∼‚α, but the converses are not true (as it can be
easily proved by using Kripke structures). Thus, the analogy between • and ‚ is not
complete: •α ≡ ∼◦α and ◦α ≡ ∼ • α, but neither ‚α ≡ ∼Œα nor Œα ≡ ∼‚α.

2.2.2. The logic DPI

Observe that deontic consistency is a derived connective in DmbC (as in some other
systems to be defined below). A natural question is: it would be possible to define the
deontic counterpart of mbC, in the sense of being a LDI such that deontic paracon-
sistency is expressed by a primitive connective? This logic would be non-explosive
but without having a consistency connective. In other words, we are looking for a
minimal LDI which is not a LFI.
At this point, we must recall that mbC, the minimal LFI, has just one axiom
for the consistency operator ◦, namely (bc1). By removing that axiom (and by
removing ◦ from the language) it is recovered the interesting paraconsistent logic PI
introduced in Batens 1980.

Definition 2.9. The logic PI is defined over the basic language ¬, ∧, ∨, → as follows:
Axiom schemas: Axioms (Ax1)-(Ax10) (recall Definition 1.2)
Inference rule: (MP) (recall Definition 1.2) „

The logic PI can be transformed in a LDI by adding the modality O, a connective

for deontic consistency and appropriate rules and axioms, as follows.

Definition 2.10. The logic DPI is defined over the language O, Œ, ¬, ∧, ∨, → by add-
ing to PI the following:
Axiom schemas:

(O-K) O(α → β) → (Oα → Oβ)

Principia 13(3): 299–326 (2009).

312 Marcelo Esteban Coniglio & Newton Marques Peron

(Dbc1) Œα → (Oα → (O¬α → Oβ))

Inference rules:
α α→β

` Oα „

Concerning semantics, DPI can be characterized by Kripke structures as those of

DmbC, but now R is not necessarily serial, and the valuations must satisfy clauses
(kc1)-(kc4) and (kc6) of Definition 2.7, plus the following:

(kc5.1) vw (Œα) = vw (Oα) = vw (O¬α) = 1 implies vw (Oβ) = 1 for every β.

Using the usual techniques, it can be proved that DPI is sound and complete
w.r.t. its Kripke semantics.
As observed in Carnielli & Marcos 2002, the logic PI is not a LFI and the same
applies to DPI. On the other hand, by using Kripke structures it is easy to see that DPI
is deontically paraconsistent, being obviously a LDI. So, the following result follows:

Theorem 2.11. DPI is a LDI but it is not a LFI.

The semantics for DPI deserves some comments. Firstly, it should be observed
that it is possible to have a trivial valuation vt rue for PI such that vt rue (α) = 1 for
every α ∈ For. This is the semantical counterpart of the fact that PI is not finitely
trivializable, and so there is no bottom formula, that is, a sentence ⊥ such that
⊥ `PI α for every α. Being so, it would be expected that its deontic extension DPI
should maintain this feature and so it would be possible to satisfy Oα for every α.
This is why the accessibility relation of its Kripke frames is not necessarily serial:
it is possible to have an isolated world w in which vw (α) = 1 (and, in particular,
vw (Oα) = 1) for every α ∈ For O .
The lack of a bottom formula (and so the absence of a consistency operator)
prevents the definition of a strong (that is, a explosive) negation ∼ in both PI and
DPI. Thus, the logic DPI is a relatively weak logic system.

2.2.3. The logic SDPI

From the last analysis, it seems that the logic DPI could be strengthened. A stronger
deontic extension of PI is possible, closer to SDL, by allowing logical explosion from
conflicting, deontically consistent obligations; in particular, this would allow the
definition of a bottom formula ⊥. It is enough to slightly modify DPI as follows:

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 313

Definition 2.12. The logic SDPI is defined over the language O, Œ, ¬, ∧, ∨, → by

substituting in DPI axiom (Dbc1) by the following:

(SDbc1) Œα → (Oα → (O¬α → β))


The Kripke semantics for SDPI consists of the Kripke structures of DPI satisfying
the following clause:

(kc5.2) vw (Oα) = vw (O¬α) = 1 implies vw (Œα) = 0.

Note that (kc5.1) is a consequence of (kc5.2). It is straightforward to prove that

SDPI is sound and complete w.r.t. its Kripke semantics.
It should also be noted that SDPI admits the definition of a bottom formula:
indeed, ⊥α =d e f (Oα ∧ O¬α) ∧ Œα is such that ⊥α `SDPI β for every α and β.
However, it is still possible to define a Kripke structure for SDPI with an isolated
world w such that vw (Oα) = 1 for every α (and, in particular, with vw (O⊥α ) = 1).
This is why the accessibility relation is not required to be serial for SDPI, despite
having a bottom formula definable in it.
By using any bottom ⊥, it is possible to define a strong (that is, explosive) nega-
tion in SDPI as ∼α =d e f (α → ⊥). Note that

O∼α `SDPI O¬α but O¬α 0SDPI O∼α.

Moreover (and this is a general fact of paraconsistent logics having a bottom for-
mula), SDPI is a LFI such that the consistency operator is defined à la da Costa as
follows: ◦α =d e f (α ∧ ¬α) → ⊥. That is, ◦α = ∼(α ∧ ¬α).

Theorem 2.13. SDPI is both a LFI and a LDI. Moreover, conflicting obligations
involving a deontically consistent sentence trivialize.

This shows that this system has a reasonable expressive power, despite being
based on a LFI weaker than mbC. Arguably, SDPI could be considered as the mini-
mal paraconsistent version of SDL or, more precisely, of eSDL∗ , the version of SDL
over For OŒ obtained from SDPI by adding the axiom schema Œα. As expected, the
accessibility relation of any Kripke structure for eSDL∗ must be serial.
To finalize, it can be observed that while DmbC and SDPI are both LFI’s and
LDI’s, they are different in nature. While the former has the consistency operator ◦
as a primitive and the deontic consistency operator is derived, the latter follows the
opposite way: the deontic operator Πis primitive, and the consistency operator is

Principia 13(3): 299–326 (2009).

314 Marcelo Esteban Coniglio & Newton Marques Peron

2.2.4. The logic C1D

Now we shall briefly analyze C1D . As mentioned above, C1D is obtained from C1
by adding some deontic axioms. Recall that C1 is defined over the basic language
For (where α◦ =d e f ¬(α ∧ ¬α)) by means of rule (MP), axioms (Ax1)-(Ax10) (see
Definition 1.2) and:

(bC1) α◦ → (α → (¬α → β))

(ca1) (α◦ ∧ β ◦ ) → (α ∧ β)◦

(ca2) (α◦ ∧ β ◦ ) → (α ∨ β)◦

(ca3) (α◦ ∧ β ◦ ) → (α → β)◦

The last three axioms state that consistency is “propagated” through the connec-
tives (it can be proved that α◦ → (¬α)◦ ). Different to mbC, the classical negation is
defined as ∼α =d e f ¬α ∧ α◦ .

Definition 2.14. The logic C1D is defined over language For O by adding to C1 the
Axiom schemas:

(O-K) O(α → β) → (Oα → Oβ)

(O-D) Oα → ∼O∼α

(ca4) α◦ → (Oα)◦

Inference rule:

` Oα „

Despite being formally analogous to axiom (D) of SDL, axiom (O-D) is equiva-
lent to (O-E)◦ of DmbC, at it can be easily proved. Thus, the conception of C1D is
close to that of DmbC. It could be said that the modal extension of C1 producing C1D
is in essence the same modal extension of mbC producing DmbC: the only additional
modal axiom is (ca4), concerning the preservation of consistency through the op-
erator O. But this makes sense, because the preservation of consistency through
the connectives is a general principle of C1 . By defining deontic consistency as
Œα =d e f O(α◦ ), it can be proved the following:

Theorem 2.15. C1 is both a LFI and a LDI.

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 315

2.2.5. The logic SDmbC

The next step is the presentation of deontic systems which are LFI’s but are not LDI’s.
This is the case of SDmbC and BDmbC, introduced in (Peron & Coniglio 2008). As
we shall see, the former is a paraconsistent logic deontically explosive; the latter is
a bimodal deontic system, being deontically explosive with respect to one modality,
and deontically paraconsistent with respect to the other.

Definition 2.16. The logic SDmbC is defined over For ◦O by replacing in DmbC
(recall Definition 2.6) axiom (O-E)◦ by

(O-E)∗ Ofα → ⊥α where fα = (α ∧ ¬α) and ⊥α = (α ∧ ¬α) ∧ ◦α


It is worth noting that SDmbC is a paraconsistent deontic logic, but it is not a

deontically paraconsistent logic (cf. Theorem 2.17 below). As in the case of DmbC,
if the axiom schema ◦α is added to SDmbC then it is obtained eSDL, the version of
SDL over For ◦O . This situation is analogous to the relation between mbC and eCL,
the version of CL over For ◦ , described in Section 1.
With respects to the semantics for SDmbC, it is enough to extend the Kripke
semantics for DmbC (cf. Definition 2.7) by adding to the valuation mappings the
following clause:

(kc7) vw (O¬α) = 1 implies vw 0 (α) = 0 for every w 0 in W such that wRw 0 .

Clause (kc7) expresses that it is impossible to have conflicting obligations in a

world w. As shown in (Peron 2009), the following results hold:

Theorem 2.17.
(i) SDmbC is a LFI.
(ii) SDmbC is deontically explosive and so it is not a LDI.

Theorem 2.18. SDmbC is sound and complete with respect to its Kripke semantics.

Another interesting feature of SDmbC is that the strong negation ∼ and the
paraconsistent negation ¬ collapse within the scope of the deontic operator.

Theorem 2.19. O¬α a`SDmbc O∼α.

Principia 13(3): 299–326 (2009).

316 Marcelo Esteban Coniglio & Newton Marques Peron

2.2.6. The logic BDmbC

The systems DmbC and SDmbC can be interesting in the analysis of deontic para-
doxes, as we shall see in the next section. However, a logic combining both sys-
tems can enrich this analysis. Such a logic, called BDmbC, is obtained by combining
DmbC and SDmbC, in a way that one deontic operator behaves as in DmbC, whereas
the other behaves as in SDmbC.
The idea of considering paraconsistent bimodal deontic logics is not new. In
Puga, da Costa & Carnielli 1988 a bimodal system defined over C1 called C1 having
a deontic operator O and an alethic operator ƒ was presented. Some interactions
between both operators where forced, namely Oα → ◊α (where, as usual, ◊α =de f
¬ƒ¬α is the “possibility” operator) and ƒα → Oα. This principles are based on
ideas by Kant and Hintikka, respectively. Since ƒα → α is an axiom of C1 then this
logic is both a LFI and a LDI with respect to both modalities, where consistency and
alethic consistency of α are expressed by the same sentence, namely α◦ ; on the other
hand, the deontic consistency of α is expressed by O(α◦ ).
Our proposal, however, is different, by several reasons. On the one hand, both
modalities of BDmbC are deontic, in contrast to C1 in which one modality is alethic.
On the other hand, BDmbC is deontically explosive with respect to one modality and
deontically paraconsistent with respect to the other, in contrast to C1 in which both
modalities are not explosive, as mentioned above.

Definition 2.20. The logic BDmbC – Bimodal Deontic mbC – is defined in the set
For ◦OO of sentences generated by O, O, ◦, ¬, ∧, ∨, → by adding to mbC the follow-
Axiom schemas:
(O-K) O(α → β) → (Oα → Oβ)

(O-K) O(α → β) → (Oα → Oβ)

(O-E)◦ O⊥α → ⊥α where ⊥α =d e f (α ∧ ¬α) ∧ ◦α

(O-E) Ofα → ⊥α where fα =d e f α ∧ ¬α

(BA) Oα → Oα

Inference rules:

` Oα

` Oα
Principia 13(3): 299–326 (2009).
A Paraconsistentist Approach to Chisholm’s Paradox 317

The operator O can be interpreted as “classically obligatory”; this notion is ex-

pressed in (O-E). On the other hand, O is interpreted as “weakly obligatory”, as
expressed in (O-E)◦ . However, an additional interaction between both operators is
intended: namely, it is convenient that O be “stronger” or “stricter” than O, which
justifies the inclusion of (BA). This axiom was inspired by the analogous one pre-
sented in the bimodal system KT€ , cf. Carnielli & Pizzi 2008.
The logic BDmbC suggest an additional analogy between LFI’s and LDI’s. Just
as mbC is ¬-paraconsistent but it is not ∼-paraconsistent, the theorem below shows
that BDmbC is O-paraconsistent but it is not O-paraconsistent. The details of the
proof are left to the reader.

Theorem 2.21.
(i) BDmbC is a LFI.
(ii) BDmbC is a LDI with respect to O.
(iii) BDmbC is deontically explosive with respect to O and so it is not a LDI with
respect to O.

As expected, the semantics for BDmbC is given by Kripke structures having two
accessibility relations, one for each modality. In details:

Definition 2.22. A Kripke structure for BDmbC is a tuple 〈W, R, R, {vw }w∈W 〉 where:

1. W is a non-empty set (of possible-worlds);

2. R ⊆ W × W and R ⊆ W × W are relations (of accessibility) between worlds

which are serial;

3. R ⊆ R;

4. {vw }w∈W is a family of mappings vw : For ◦OO → 2 satisfying clauses (kc1-

kc6) of Definition 2.7 plus the following clauses:

(kc6.1) vw (Oα) = 1 iff vw 0 (α) = 1 for every w 0 in W such that wRw 0 ,

(kc7.1) vw (O¬α) = 1 implies vw 0 (α) = 0 for every w 0 in W such that wRw 0 . „

The proof of the completeness of BDmbC with respect to its Kripke semantics
can be found in Peron 2009. Since in BDmbC the operator Oα behaves as in DmbC,
as long as O has the same behavior as the deontic operator of SDmbC, it is easy to
prove (by semantical means) the following result:

Principia 13(3): 299–326 (2009).

318 Marcelo Esteban Coniglio & Newton Marques Peron

Theorem 2.23. In BDmbC:

O¬α a`BDmbC O∼α holds,
but O¬α a`BDmbC O∼α does not hold.

Of course it is possible to consider a plethora of paraconsistent deontic logics

and LDI’s by taking different LFI’s and/or with combinations of (O-E) and (O-E)◦
(as it was done with BDmbC). In particular, in (Coniglio 2007) the system DLFI1
based on the logic LFI1 was considered. The latter is a LFI enjoying nice features, for
instance some axioms propagating the inconsistency operator •. Another interesting
point about LFI1 is that it has a simple 3-valued matrix semantics. The deontic
system DLFI1 is both a LFI and a LDI such that the deontic inconsistency operator ‚
satisfies the following: ‚α ↔ (Oα ∧ O¬α). Moreover, ‚ propagates under certain
circumstances, which is an useful tool in the analysis of deontic paradoxes. The
property of propagation of inconsistency resembles the propagation of consistency
enjoyed by C1D . However, the underlying paraconsistent logics C1 and LFI1 are quite
different: while the latter is a 3-valued logic, the former cannot be characterized by
finite matrices. Moreover, the propagation properties are not the same.
A previous and related approach to the analysis of modal paradoxes by using
modal LFI’s can be found in Costa-Leite 2003, where a paraconsistent alethic system
called Ci T was proposed in order to analyze Fitch’s paradox of knowability. The
system Ci T is is a modal extension of Ci, a LFI introduced in Carnielli & Marcos
Instead of defining more and more paraconsistent deontic systems and LDI’s,
the next section will analyze a classical deontic paradox, the so-called Chisholm’s
paradox, in the context of some of the system considered in this section.

3. Chisholm’s Paradox

In general, a deontic paradox consists of a set of sentences in natural language, in-

tuitively consistent and without logical dependencies such that, when formalized in
a deontic language, it is logically trivial or it has logical dependencies. This situation
occurs more frequently when the given sentences refer to norms, laws and moral
principles: in this case contradictions can arise. Thus, most of such paradoxes are
originated when the given set of premises is formalized in SDL as a set Γ of logically
independent sentences, but conflicting obligations such as Oα and O¬α are derived
from Γ within SDL. As pointed out in (McConnell 2006), the problem is related to
the difficulties of SDL for dealing with contrary-to-duty obligations.
In this section, the well-known Chisholm’s deontic paradox will be analyzed un-
der the light of some LDI’s discussed in the previous section. The aim is to show to
what extent the paraconsistentist approach to deontic paradoxes can be fruitful.

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 319

One of the first deontic paradoxes was proposed in Chisholm 1963. The follow-
ing formulation of it was given in Åqvist 2002.
Consider the set C formed by the following sentences:

(1) It ought to be that John does not impregnate Suzy Mae.

(2) Not-impregnating Suzy Mae commits John to not marrying her.

(3) Impregnating Suzy Mae commits John to marry her.

(4) John impregnates Suzy Mae.

Let A and B be propositional letters representing the sentences “John impregnates

Suzy Mae” and “John marries Suzy Mae”, respectively. It is possible to formulate the
set (1)–(4) above in a deontic language in several ways. In order to simplify the
presentation, the following notions will be useful.

Definition 3.1. Consider in Dmbc the following derived deontic operators:

(i) F1 α =d e f O¬α (prima-facie prohibition)
(ii) F2 =d e f O∼α (strong prohibition) „

The name of such operators is inspired by Ross 1930.

Returning to the set C , it is clear that (1) admits two formulations in Dmbc:
F1 A and F2 A. By its turn, (2) has three interpretations: ¬A → F1 B, ¬A → F2 B and
O(¬A → ¬B). Analogously, sentence (3) can be formalized as A → OB or O(A → B).
Being so, there exist 12 possibilities to formalize the set C , which are conveniently
arranged below:

Γ1.1 = {F1 A, ¬A → F1 B, A → OB, A}

Γ1.2 = {F1 A, ¬A → F2 B, A → OB, A}

Γ1.3 = {F2 A, ¬A → F1 B, A → OB, A}

Γ1.4 = {F2 A, ¬A → F2 B, A → OB, A}

Γ2.1 = {F1 A, O(¬A → ¬B), A → OB, A}

Γ2.2 = {F2 A, O(¬A → ¬B), A → OB, A}

Γ3.1 = {F1 A, O(¬A → ¬B), O(A → B), A}

Γ3.2 = {F2 A, O(¬A → ¬B), O(A → B), A}

Γ4.1 = {F1 A, ¬A → F1 B, O(A → B), A}

Principia 13(3): 299–326 (2009).

320 Marcelo Esteban Coniglio & Newton Marques Peron

Γ4.2 = {F1 A, ¬A → F2 B, O(A → B), A}

Γ4.3 = {F2 A, ¬A → F1 B, O(A → B), A}

Γ4.4 = {F2 A, ¬A → F2 B, O(A → B), A}

Clearly some of the sets above are not appropriate because some logical depen-
dencies are present. This is exactly the paradoxical aspect of the set C , because
it is consistent and the sentences are logically independent. Now we will analyze
Chisholm’s paradox in some of the deontic systems defined above.

3.1. Analysis in SDL

Given that in SDL just one ‘prohibition’ operator is definable in SDL (because just
classical negation is present) then the unique candidates to formalize C in SDL are
Γ1.4 , Γ2.2 , Γ3.2 and Γ4.4 . But the following laws hold in SDL:

`SDL α → (¬α → Oβ) and `SDL O¬α → O(α → β).

Then, (2) and (4) are logically dependent in Γ1.4 ; (1) and (3) are logically dependent
in Γ3.2 ; by its turn, (1) and (3), on the one hand, and (2) and (4), on the other, are
logically dependent in Γ4.4 .
So, the unique alternative in SDL is Γ2.2 . But then OB follows by (MP) between
(3) and (4). On the other hand, `SDL O(¬A → ¬B) → (O¬A → O¬B) and so,
by (MP) two times it follows O¬B. That is, the set Γ2.2 is logically trivial in SDL,
arriving so to a paradox.

3.2. Analysis in DmbC

Moving now to DmbC, there are much more alternatives for choosing a set Γi. j ,
because there are less logical interdependencies.
By taking appropriate Kripke structures, it is easy to prove that

2DmbC α → (¬α → O¬β) and 2DmbC α → (¬α → O∼β).

Moreover, by using Kripke structures it is easy to see that the sentences in Γ1.1 , Γ1.2 ,
Γ1.3 and Γ1.4 are logically independent in DmbC. The independence of the premises
in Γ2.1 , Γ2.2 , Γ3.1 , Γ4.1 and Γ4.2 is obtained by analogous arguments. By (O-K) and
(MP), OB is derived from Γ1. j (for 1 ≤ j ≤ 4) and also from Γ2.1 and Γ2.2 . By a
similar argument, O¬B is derived from Γi. j (for i. j = 2.1, 2.2, 3.1). This means that
Γ2.1 and Γ2.2 are deontically contradictory.

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 321

Concerning the set Γ3.2 , note that `DmbC ∼A → (A → B) and then, by (O-Nec)
and (O-K) there is a dependence between (1) and (3). Then, this set cannot be used
for the analysis. By the same reason the sets Γ4.3 and Γ4.4 must be discarded.
Altogether, there are nine sets logically independent in DmbC formalizing C and
so being potentially useful for the analysis of Chisholm’s paradox, in contrast to SDL
which have just one. Remarkably, none of them are logically trivial and so the nine
sets are relevant for the analysis. It is easy to prove by using Kripke structures that
Γi. j does not deduce O¬B for i. j = 1.1, 1.2, 1.3, 1.4, 4.1, 4.2. By its turn, Γi. j does
not deduce OB for i. j = 3.1, 4.1, 4.2. Thus, all these sets do not infer conflicting
obligations. Finally, Γi. j deduce conflicting obligations for i. j = 2.1, 2.2, but there is
no deontic explosion (and so there is no logical explosion).
Summarizing, in DmbC we have the following scenarios for Chisholm’s paradox:

Γi. j `DmbC OB but Γi. j 0DmbC O¬B for i. j = 1.1, 1.2, 1.3, 1.4

Γ3.1 `DmbC O¬B but Γ3.1 0DmbC OB

Γi. j `DmbC OB, O¬B without trivializing deontically, for i. j = 2.1, 2.2

Γi. j 0DmbC OB and Γi. j 0DmbC O¬B for i. j = 4.1, 4.2

That is, DmbC has enough expressive power to encompass all the possible situa-
tions, namely:

• just OB is derived;

• just O¬B is derived;

• both OB and O¬B are derived, without having deontic explosion (and so,
without trivializing);

• neither OB nor O¬B are derived.

3.3. Analysis in SDmbC

As happens in SDL, in the logic SDmbC the deontic operators F1 and F2 collapse; let
F be the unique prohibition operator in SDmbC. Thus, there are just four possibilities
in SDmbC formalizing C , namely:

Γ0.1 = {FA, ¬A → FB, A → OB, A}

Γ0.2 = {FA, O(¬A → ¬B), A → OB, A}

Γ0.3 = {FA, O(¬A → ¬B), O(A → B), A}

Principia 13(3): 299–326 (2009).

322 Marcelo Esteban Coniglio & Newton Marques Peron

Γ0.4 = {FA, ¬A → FB, O(A → B), A}

It is easy to see that in SDmbC it holds:

2SDmbC α → (¬α → Oβ) but SDmbC O¬α → O(α → β)

and so Γ0.3 and Γ0.4 are discarded because there are logical dependencies among
the premises. On the other hand, Γ0.2 trivializes in SDmbC, because this system
does not accept conflicting obligations. Thus, the only admissible set is Γ0.1 which
satisfies the following:

Γ0.1 `SDmbC OB but Γ0.1 0SDmbC O¬B.

That is, Chisholm’s paradox can again be avoided, producing the expected result:
John has to marry Suzy Mae.

3.4. Analysis in DPI and in SDPI

Since DPI admits just one negation, namely the paraconsistent one, just the sets Γ0. j
(for j = 1, . . . , 4) of SDmbC are obtained, with Fα = O¬α. As expected, Γ0.1 just
derives OB; Γ0.3 just derives O¬B; Γ0.2 derives both OB and O¬B without deontic
explosion (and so without trivializing); and Γ0.4 neither derives OB nor derives O¬B.
Concerning SDPI, the same results obtained for DmbC hold in this logic, mutatis

3.5. Analysis in BDmbC

The logic BDmbC validates all the inferences in DmbC about O, as well as the in-
ferences of SDmbC about O. Moreover, three prohibition operators can be defined:
O¬α, O∼α and O¬α. Since two obligation operators are available, namely Oα and
Oα, there are much more alternatives to formalize the set C . Being an extension
of DmbC, all the possible situations for the paradox described in Subsection 3.2
can also be obtained in BDmbC. However, since there are two kinds of obligations
in BDmbC, there are strictly more possible situations than in DmbC. Namely, the
following obligations concerning B are obtained from the different sets of premises:

• just OB is derived;

• just OB (and so OB) is derived;

• just O¬B is derived;

• just O¬B (and so O¬B) is derived;

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 323

• OB and O¬B are derived, without having deontic explosion;

• OB (and so OB) and O¬B are derived, without having deontic explosion;

• neither OB (nor OB) nor O¬B (nor O¬B) are derived;

• everything is derived (logical explosion).

3.6. Summarizing
To summarize, all the plausible answers to the question “Should John marry Suzy
Mae?” are obtained in DmbC, DPI and SDPI. In the case of BDmbC, there are still
more variants for the answers (even the logical explosion can be attained), as shown
above. On the other hand, SDmbC produces just one answer, the intuitively expected
one. In contrast, SDL just produces logical explosion (or logical dependencies).
The main point of our analysis lies in the fact that Chisholm’s paradox is not ex-
clusively based on the violation of the Principle of Deontic Consistency as presented
frequently in the literature (this position is defended, for instance, in (Prakken &
Sergot 1994)). The analysis above suggests that the logical dependence between
the premises in C (dependency arising because of the use of classical logic) plays
an important role. Thus, in SDmbC the principle of deontic explosion is maintained
while rejecting the principle of explosion, and then it is possible to avoid the para-
dox. That is, by eliminating some logical dependencies typical of classical logic, the
paradox vanishes. In the case of DmbC, DPI and SDPI, both principles of explo-
sion are weakened and so there are more possibilities to avoid the paradox. Finally,
BDmbC combines DmbC and SDmbC, obtaining still more solutions to the paradox
by combination of its deontic operators.

4. Final remarks

Our analysis departs from the transposition of the approach to paraconsistency of

LFI’s to the deontic context. This allows to consider notions such as deontic explosion
and deontic paraconsistency.
The Logics of Deontic (In)Consistency seem to encompass several paraconsistent
deontic systems already proposed in the literature, suggesting the definition of a tax-
onomy of such systems, in a similar way as the proposal for LFI’s started in Carnielli
& Marcos 2002.
The approach to Chisholm’s paradox presented here is based on the generality
of LDI’s, and so it is possible to obtain solutions to the paradox based on the lack
of logical dependencies and/or the elimination of the principle of deontic explosion.
The distinction between the two sources of paradox can be useful to analyze other

Principia 13(3): 299–326 (2009).

324 Marcelo Esteban Coniglio & Newton Marques Peron

deontic paradoxes, as those presented in Prakken & Sergot 1994, Prakken & Sergot
1997 and Carmo & Jones 1997.
Clearly, the research on paraconsistent deontic logic and deontic paradoxes has
several possibilities of further development. We hope that our discussion can con-
tribute to some extent to the analysis of the interesting question of deontic paradoxes
and related areas.1

Åqvist, L. 2002. Deontic logic. In D. Gabbay and F. Guenthner (eds.) Handbook of Philosoph-
ical Logic (2nd. edition), volume 8, 147–264. Kluwer Academic Publishers.
Batens, D. 1980. Paraconsistent extensional propositional logics. Logique et Analyse 90-91:
Bueno-Soler, J. 2008. Possible-translations semantics for cathodic modal logics. CLE e-Prints
URL =,n_6,2008.html.
Carmo, J. & Jones, A. 1997. Dyadic deontic logic and contrary-to-duty obligations. In D. Nute
(ed.) Defeasible Deontic Logic, volume 263 of Synthese Library, 317–44. Kluwer Publishing
Carnielli, W. A.; Coniglio, M.; Marcos, J. 2007. Logics of Formal Inconsistency. In D. Gabbay
and F. Guenthner (eds.) Handbook of Philosophical Logic (2nd. edition), volume 14, 15–
107. Springer.
Carnielli, W. A. & Marcos, J. 1999. Limits for paraconsistency calculi. Notre Dame Journal of
Formal Logic 40(3): 375–90.
———. 2002. A Taxonomy of C-systems. In: Carnielli, W. A., Coniglio, M. E. and D’Ottaviano,
I. M. L. (eds.) Paraconsistency: The Logical Way to the Inconsistent, pp. 1–94, Marcel
Dekker. Preliminary version available at CLE e-Prints 1(5), 2001.
Carnielli, W. A. & Pizzi, C. 2008. Modalities and Multimodalities, volume 12 of Logic, Episte-
mology, and the Unity of Science. Springer Verlag.
Chellas, B. 1980. Modal Logic: an introduction. Cambridge: Cambridge University Press.
Chisholm, R. 1963. Contrary-to-duty imperatives and Deontic Logic. Analysis 24: 33–6.
Coniglio, M. E. 2007. Logics of Deontic Inconsistency. CLE e-Prints 7(4).
URL =,n_4,2007.html.
Costa-Leite, A. 2003. Paraconsistência, modalidade e cognoscibilidade (Paraconsistency, Modal-
ity and Cognoscibility, in Portuguese). Master’s thesis, Institute of Philosophy and Human
sciences, State University of Campinas (UNICAMP), Campinas.
Creswell, M. J. & Hughes, G. E. 1996. A New Introduction to Modal Logic. London: Houtledge.
Cruz, A. M. P. 2005. Lógica Deôntica Paraconsistente: Paradoxos e Dilemas (Paraconsistent
Deontic Logic: Paradoxes and Dilemmas, in Portuguese). Natal: Editora da UFRN, Brazil.
da Costa, N. C. A. 1963. Sistemas Formais Inconsistentes (Inconsistent Formal Systems, in
Portuguese). Habilitation Thesis. Republished by Editora UFPR, Curitiba, 1993.
———. 1974. On the theory of inconsistent formal systems. Notre Dame Journal of Formal
Logic XV: 497–510.

Principia 13(3): 299–326 (2009).

A Paraconsistentist Approach to Chisholm’s Paradox 325

da Costa, N. C. A. & Carnielli, W. A. 1986. Paraconsistent Deontic Logic. Philosophia 16(3/4):

Jaśkowski, S. 1948. Rachunek zdan dla systemów dedukcyjnych sprzecznych. Societatis Sci-
entiarun Torunesis, Sectio A I: 55–77. Translated as: “Propositional Calculus for Contra-
dictory Deductive Systems”, Studia Logica 24: 143–157, 1969.
McConnell, T. 2006. Moral Dilemmas. In E. N. Zalta (ed.), The Stanford Encyclopedia of Phi-
McNamara, P. 2006. Deontic Logic. In E. N. Zalta (ed.) The Stanford Encyclopedia of Philoso-
Peron, N. M. 2009. Lógicas da Inconsistência Deôntica (Logics of Deontic Inconsistency, in Por-
tuguese). Master’s thesis, Institute of Philosophy and Human sciences, State University
of Campinas (UNICAMP), Campinas.
Peron, N. M. & Coniglio, M.E. 2008. Logics of Deontic Inconsistencies and Paradoxes. CLE
e-Prints 8(6).
URL =,n_6,2008.html.
Prakken, H. & Sergot, M. 1994. Contrary-to-duty imperatives, defeasibility and violability. In:
Jones, A. and Sergot, M. (eds.), Proceedings of the 2nd International Workshop on Deontic
Logic in Computer Science (DEON94), volume 1/94 of CompLex, pp. 296–318, Tano A.S.,
———. 1997. Dyadic deontic logic and contrary-to-duty obligations. In: Nute, D. (ed.), De-
feasible Deontic Logic, volume 263 of Synthese Library, pp. 223–62. Kluwer Publishing
Puga, L.Z., da Costa, N. C. A. & Carnielli, W. A. 1988. Kantian and Non-Kantian Logics.
Logique & Analyse 31(121/122): 3–9.
Ross, D. 1930. The Right and the Good. Oxford: Oxford University Press.
von Wright, G. 1951. Deontic Logic. Mind 60: 1–15. Republished in: Logical Studies (by G.
H. von Wright). Londres: Routledge and Kegan Paul, 1967, 58–74.


Department of Philosophy
Centre for Logic, Epistemology and the History of Science (CLE)
State University of Campinas (UNICAMP)
P.O. Box 6110
13081-970 Campinas, SP, Brazil and

Resumo. As Lógicas da (In)Consistência Deôntica (LDI’s) podem ser consideradas como

sendo a contraparte deôntica das lógicas paraconsistentes chamadas de Lógicas da (In)Con-
sistência Formal. Neste artigo são introduzidas e estudadas novas LDI’s e outras lógicas deôn-
ticas paraconsistentes satisfazendo diferentes propriedades: sistemas tolerantes a obrigações
contraditórias; sistemas em que as obrigações contraditórias produzem trivialização; e uma
lógica deôntica paraconsistente bimodal que combina as características de sistemas previ-
amente introduzidos. Estas lógicas são utilizadas para analisar o conhecido paradoxo de

Principia 13(3): 299–326 (2009).

326 Marcelo Esteban Coniglio & Newton Marques Peron

Chisholm aproveitando-se do fato de que, além que as obrigações contraditórias não triviali-
zam nas LDI’s, varias das dependências lógicas da lógica clássica são bloqueadas no contexto
das LDI’s, permitindo assim dissolver o paradoxo.

Palavras-chave: Lógica paraconsistente, lógica deôntica, paradoxos deônticos, paradoxo de

Chisholm, dilemas morais, lógicas da inconsistência formal.

The first author was supported by an individual research grant from The National Council
for Scientific and Technological Development (CNPq), process #304090/2008-5.

Principia 13(3): 299–326 (2009).

University of Oxford

Abstract. The paper critically examines an unpopular line of Frege’s view on numbers in
the Foundations of Arithmetic. According to this view, which analyzes numbers in terms of
properties and not in terms of extensions, numbers are properties of concepts vs. properties
of objects. The latter view is held by Mill and is famously criticized in the Foundations.
I argue that on the property account numbers cannot only be properties of concepts but
they also have to be properties of objects. My main argument rests on purely metaphysical
grounds. It stems from the motivation that were numbers only properties of concepts we
would not have been able to explain mathematical truths about the physical world or those
truths would have been miraculous. On pains that we do have mathematical truths about
the physical world that are not miraculous we cannot agree with Frege’s property line about
the metaphysical nature of numbers.

Keywords: Numbers, properties, one-to-one correspondence, Frege.

Frege’s view of the nature of numbers is perhaps the most influential one in 20 th cen-
tury philosophy of mathematics, both in its richness and in the difficulties it raises
(Frege 1884). In metaphysics it is widely accepted that Frege has refuted Mill’s view
on numbers as properties of objects as an untenable one (cf. Mill 1874). Attempts
to defend reformulated versions of Mill’s view are extremely rare: Philip Kitcher’s
recent one is perhaps the most notable (Kitcher 1998). And yet even Kitcher concen-
trates exclusively on the epistemic side of Mill’s view. This side is all too often taken
together with, and throws shadow over, Mill’s metaphysical claim about the numbers
as properties. Among the metaphysical attempts is the Lambros’ modest view that
“there are no compelling reasons why number words cannot be thought of as predi-
cates of objects” (Lambros 1976, 381). Lambros’ arguments are logically interesting
and I take them as appealing, but his main aim is to vindicate the predicative use of
number words, and where he draws metaphysical conclusions (Lambros 1976, 384–
5) they stem from a prevailing logical motivation rather than from a metaphysical
one per se. In this brief paper I will offer metaphysical reasons for accepting within
the property line view of the nature of numbers that, Millean or not, a conception
of numbers as properties of objects is not merely an acceptable but a necessary one.

Principia 13(3): 327–38 (2009).

Published by NEL — Epistemology and Logic Research Group, Federal University of Santa Catarina (UFSC), Brazil.
328 Boris Grozdanoff

This perhaps unexpected thesis is suggested as one that does not contradict but that
actually complements the reading of Frege’s view of numbers in The Foundations of
Arithmetic, according to which numbers are not simply self-subsistent objects but
also properties of concepts. The paper certainly does not attempt the difficult task
of developing this account; rather, it shows metaphysical reasons why it needs to be
Famously, Frege argued that (1) numbers are self-subsistent objects that fall as
extensions under concepts about equinumerosity (in paragraphs such as #57 and
#69–#74 in the Foundations) but he also argued that (2) numbers are properties
of concepts (in paragraphs such as #46). It is more or less uncontroversial that the
leading view is (1) and yet (2) is what mainly stays behind Frege’s criticism of Mill
(in paragraphs as #8, #21–#25). Also, it is (2) which captures best our numerically-
predicative intuition for expressions like “The apples in the basket are five” (P). The
fact that we could, following Frege, translate such expressions into identities like
“The number of apples in the basket = five” (Pi ) does not annihilate the previous
fact that predicative expressions like P are well-formed and more often used than
their Pi -type translations as identities. Also, and most importantly for the purposes of
science, it is P-type expressions, and not Pi -type expressions, that capture the core of
the applied mathematical language. The claim in The Foundations that Frege would
not give up is the ontological thesis (1) and it is controversial whether he actually
held (2), which is maintained simultaneously in the book, after he came up with
(1).1 What is also controversial is whether both claims could actually be maintained
together. My purpose is not to discuss these controversies but to examine the prop-
erty line of Frege’s theses and especially in connection with applied mathematical
expressions like the those in #70.
Let us assume that Frege’s reading that numbers are not properties of objects
is correct and that, were they properties of anything at all, numbers would have
been solely and only properties of concepts (as he is explicit in #46). In accordance
with the context principle the main function of concepts within propositions is to
say something meaningful and to contribute to propositional truths. To take Frege’s
well known example, let us imagine that there are (only) 52 cards on the table,
in their original unopened pack (let us denote this state of affairs “S”). We cannot
just ask the question “How many? (Q)” for we lack the sortal concept that would
determine the number of what we are inquiring about: it might be the number of
packs, the number of cards, the number of spots on the table, the number of spades,
the number of whatever. So let us supply a sortal concept, say “cards”, and ask
our question, which takes the form of “How many cards are there on the table?”
Now the question is correctly formulated and points to a specific, sorted state of
affairs. The main semantic use of this and all similar propositions is to reach for
their truth or falsity, among other things and through their being meaningful in the

Principia 13(3): 327–38 (2009).

Fregean One-to-one Correspondence and Numbers as Object Properties 329

first place. The answer, given the concrete state of affairs, is “There are 52 cards
on the table” (P). Is this a correct answer? It seems so. If we deny that P is
meaningful we would harm the conception of sortal concept as carriers of numbers
as properties. If we deny that it is true we would again harm the same conception
as semantically useless, which would contrast with the striking intuitiveness and
precision of mathematical truths. It seems then that we are compelled to accept that
P is both meaningful and true. The problem, however, with this is that P attests not
merely a fact about the state of affairs S but attests a numerical fact about S. That is,
it says not simply that there are cards on the table but that there are fifty two cards
on the table. Did something change in S between the reformulation of Q into P? It
obviously did not. Were there 52 cards on the table before we have even bothered
to ask the question? There certainly were. Are there 97 cards on the table given S?
There certainly are not. If there were a Q answer to the Q question SQ would have
stayed exactly the same as S P , whatever the semantic value of the Q answer. This
and similar counterfactual differences in S are sufficiently interesting in order to be
accounted for semantically, and numerically capturing the difference seems to be the
most suitable and the most exhaustive way. Thus, there seem to be “reasons” in S
that on the one hand contribute to the meaning of P and on the other hand, that
sufficiently strongly participate in the truth of P. We can distinguish between two
different problems for Frege here:

1. The “no change” problem

2. The “participation” problem

The no change problem claims that whatever sortal question we formulate and ask
S remains the same on the set of all possible reading of S and especially on a numer-
ical reading, the one that seems to be telling something precisely true about S after
the sortal question is asked. We might well have chosen any of the other possible
questions “How many decks are on the table?”, “How many packs are there on the
table?”, “How many Spades are there on the table?” and for every single one of
them the state of affairs S that would participate in the delivery of the true answer-
proposition would remain physically unchanged. The answers would contain, of
course, different unions of sortal concepts and numbers (one deck, one pack, thir-
teen Spades) but what is there on the table that makes those at least partially true
does not change on any reasonable understanding of “change”. Frege sees and in a
convincing manner the numbers coming from our conceptually sorted expressions.
And we certainly do need sortal concepts to identify what we are inquiring about.
But this pertains more to our knowledge of what it is a number of than to the nature
of a number of per se: the first one is a cognitive attitude and the second one is a

Principia 13(3): 327–38 (2009).

330 Boris Grozdanoff

To inquire about the very nature of numbers is not to inquire about our knowl-
edge of this nature: it is rather to ask a metaphysical question. Therefore, what
Frege manages to do in a most brilliant way is to give the epistemic-semantic side of
the problem. Where he commits a categorical mistake is when he reformulates the
epistemic part as a metaphysically exhaustive one. Even if it is true that numbers are
properties of the sortal concepts that serve to identify them, from this it does not fol-
low that numbers are the only properties of sortal concepts. This is highly non-trivial
and is seen best in “applied cases”, typical for natural sciences. This leads us to the
second problem, the “participation problem”.
Whatever numbers are, how could we possibly have numerically true proposi-
tions about, say, the physical world, without the physical world is being numerically
responsible for these truths? If this were indeed the case (and it would have been
completely arbitrary whether there are 52 cards on the table or not, one electron
in the hydrogen atom or not) the very truth of the propositions would not have
exceeded a value beyond an arbitrary one. There are at least two reasons that ratio-
nally stop us from accepting this. The first one is the mind-boggling strength of pure
mathematical truths and the second one, which ties it with things like the external
world, is what Wigner famously called “unreasonable effectiveness of (pure) math-
ematics in naturals science” (Wigner 1960: 222). In cases of sufficiently successful
physical theories, like the Standard model of Quantum Mechanics and General The-
ory of Relativity, the mathematical success of those theories is difficult to even com-
prehend properly. The parallelism between the pure truths of mathematics and the
applied truths of our best scientific theories could not simply be a result of an arbi-
trary choice of a particular set of special concepts. There must be something in what
the applied mathematical theories are actually about in order for this parallelism to
take place and not some other. And the most simple metaphysical answer seems
to be that numbers are both properties of sortal concepts and things those concepts
(within propositions) are about. On this scenario we both keep our numerically-
predicative intuitions for typical applied mathematical expressions and have at our
disposal an account on which the success of pure mathematics in natural sciences
is not miraculous anymore and is reduced to spelling out the proper coordination
between one and the same properties, being possessed by different metaphysical
For Frege it is not sufficient that numbers are properties of concepts. He needs
to establish a relation between that concepts that would preserve their numerosity
in a meaningful way and this relation should be logical too. His insight is the notion
of equinumerosity. Its function is crucial: within a principle (Hume’s Principle) it
delivers nothing less than the identity of numbers:

For any concepts, F and G, the number belonging to F is identical to the

Principia 13(3): 327–38 (2009).

Fregean One-to-one Correspondence and Numbers as Object Properties 331

number belonging to G iff F and G are equinumerous. (#63, Hume’s Prin-


Equinumerosity is defined through an appeal to the extensions of the concepts that

bear numbers as properties: two concepts, F and G are equinumerous when there
is a one-to-one correspondence between the objects that fall under F and G. In #70
Frege writes:
If a waiter wishes to be certain of laying as many knives on a table as plates,
he has no need to count either of them; all he has to do is to lay immedi-
ately to the right of every plate a knife, taking care that every knife on the
table lies immediately to the right of the plate. Plates and knives are thus
correlated one to one by the very means used, viz. The identical spatial

If numbers are taken to be only properties of concepts, then the relation between
them, equinumerosity, has to be a relation only between concepts too. Yet, obvi-
ously, numerical truths seem to hold not just between logical entities but for, fol-
lowing Frege, things concepts are about, and therefore Frege needs to go beyond
mere conceptual equinumerosity. In fact, he goes in a way even further. Conceptual
equinumerosity is arrived at through the very notion of conceptual extension. We are
presented with the following metaphysical picture: numbers receive their identity if
and only if the concepts that bear them (while nothing else but concepts is able to
bear them) are equinumerous. The concepts are equinumerous if and only if there
is a one-to-one correspondence between the objects which fall (as their extensions)
under them. The one-to-one correspondence is available, in Frege’s example with
knives and plates, through a “one” to “one” “correlation” which, in this particular
case, is made possible through a sort of “spatial relationship”. Besides the notion of
equality (of a numerical sort, nevertheless) the notion of being numerised is what is
precious in the notion of equinumerosity. For in order to be capable to participate
in a relation of equi-numerosity the relata must be able to be of such a kind that
something in them metaphysically allows an equality with respect to numerosity to
be evaluated in the first place. For it is equality with respect to numerosity that deliv-
ers the crucial phenomenon of equinumerosity. The metaphysical sequence, which
Frege depicts, must therefore demonstrate how the very notion of numerosity, the
one which is at the heart of equality within the central notion of equi-numerosity,
emerges in the first place. A critical inquiry must then investigate the metaphysical
transfer from the source of the notion, whatever this turns out to be, to the notion
of equinumerosity.
The first transfer is the transition from the notion of one-to-one correspondence
to that of equinumerosity. This is more or less unproblematic: being one is the
same as being of the number one and thus as being of a certain number (in this

Principia 13(3): 327–38 (2009).

332 Boris Grozdanoff

case one). The notion of numerosity is clearly preserved so the numerical transition
faces no problem. In the above metaphysical setting, however, it is not at all obvious
why numerosity should not be comprehended and rationally approached in terms of
properties. For it is clear that it is not an object that is allegedly being carried from
the source of whatever stays behind the notion of numerosity as it participates in
Frege’s central notion of equinumerosity. A property would do the job much more
naturally. The numerosity meant here is certainly not merely a linguistic one: it is
not a rudimentary conceptual discharge following purely linguistic recognition of the
concept of numerosity within the higher complex order concept of equinumerosity.
The numerosity inquired about here is the one that makes possible the very meta-
physical ability of being capable of being a one in the first place. For in only such a
way an entity could be related to another entity, whatever their metaphysical status
might be, within the one-to-one correspondence relation, the relation that delivers
There is another level of transition too, however, which is of a different meta-
physical order: in The Foundations equinumerosity is a relation between concepts,
but the relation of one-to-one correspondence is between objects that fall under con-
cepts. The metaphysical distinction between concepts and objects is stressed in a
quite straightforward way: we are supposed to reach equinumerosity of concepts
from the numerosity stance of objects. This metaphysical level of transition is not so
clear as the purely numerical one. But as far as Frege is concerned, concepts being
objects is not so problematic either. What is more interesting is the next step: the
metaphysical level of numerosity that is prior to one-to-one correspondence. This
level should be a numerical source level for one-to-one correspondence and thus
would be of utmost metaphysical interest. In #70 Frege is quite clear about this
level: it is the level of the physical entities knives and plates and not merely the level
of the concepts “knives” and “plates”. For the waiter is not putting the concept “knife”
spatially next to the concept “plate” following some sort of a “spatial relationship”;
such relationship could not exist among concepts. The waiter has not to count real
physical knives and plates and again not their concepts. Thus the entities that have
to be in a one-to-one correspondence are from the level inhabited by the physical
objects “knives” and “plates”. The last sentence in the quotation from #70 is that
“Plates and knives are thus correlated one by one (my italics)”. The metaphysical
level is defined as clearly as it could be. It is plain that the things to be one-to-one
corresponded are classical physical entities.2 Yet, the notion of numerosity exists
quite alive here as well: the notion of “one” is a numerically full-blooded notion.
And it is metaphysically crucial. For it delivers what is to be preserved through the
transitions until it reaches equinumerosity: numerosity itself. Knives and plates are
not just “spatially ordered” but they are also numerically individuated, that is, they
are of some number and this number in every case of an individual entity as “knife”

Principia 13(3): 327–38 (2009).

Fregean One-to-one Correspondence and Numbers as Object Properties 333

and “plate” is suspiciously difficult to distinguish from the natural number one. Only
in virtue of this could a knife and a plate be one-to-one, that is, due to their being
numerically individuated in the most primitive metaphysical way, numerically or-
dered with respect to each other in some sort of way. And the number one seems to
stay in relation to the entities as a property of theirs in no lesser way as it stays as
a property of a Fregean concept. No one-to-one correspondence is conceivable with
no oneness being present. And spatial relationship, being a geometrical relation-
ship, might deliver a numerical individuation, that is, it might individuate an entity
in such a way that it becomes rationally approachable with respect to its metaphys-
ical status as one entity. But it is only in virtue of the entity being one entity that it
could participate in a one-to-one correspondence relation and not merely in virtue
of its being 3 centimetres on the right of some other entity.
Thus, a notion of numerosity seems to be present in the bottom Fregean meta-
physical level as well. And here the problem rises its ugly head, for this being the
last metaphysical level in the story at hand, there is no other possible metaphysi-
cal source to take numerosity from. And if numerosity is present in all metaphysical
levels this means that we simply track a property of the inhabitants of all those meta-
physical levels, a property they all share as common. Yet, as it was stressed above,
not all of those inhabitants are concepts. Knives and plates are physical entities and
such entities are of metaphysically different category. This being the case, however,
numbers could not be just properties of concepts for they appear to be as much prop-
erties of physical (as any other objects of course too: and primarily abstract objects)
entities as properties of concepts. Freges’s metaphysical theory about numbers is in
this aspect incomplete and in its incompleteness it fails. In order for a one-to-one
correspondence to be possible there have to be “ones” beforehand available. The
“ones” do not somehow emerge in virtue of correspondence and so do not receive
their numerical individuation by merely being put one-to-one. The “ones” make pos-
sible the one-to-one-correspondence in the first place. There does not seem to be a
meaningful way of a talk about one-to-one correspondence unless the elements put
in such a correspondence are independent of the relation of numerical individua-
tion; the one that comes along with, perhaps, their standard property individuation.
Therefore, conceptual one-to-one correspondence could not serve as a way to de-
fine numerosity, the heavy burden Frege puts on it in the Foundations of Arithmetic.
Because numerosity is already at hand, the conceptual one-to-one correspondence
simply inputs already numerically individuated metaphysical elements. The effect
of this is far reaching – no definition of numerosity could succeed if based on the
exposed as non-primitive notion of conceptual one-to-one correspondence.
Equinumerosity delivers no other kind of identity but the numerical identity of
concepts: two concepts are equinumerous if and only if it is possible to coordinate
the objects that “fall” under them in such a way so that they are related in the

Principia 13(3): 327–38 (2009).

334 Boris Grozdanoff

alleged “one-to-one correspondence”. In #70 Frege illustrates this coordination with

the spatial example of knives and plates. In such physical cases it is clear that in
order to have one-to-one coordination we need not merely a linguistic and logical
distinction between a knife and a plate. For the waiter physically distinguishes one
from the other by doing a sort of physical coordination: putting one next to the other.
The fact that I can only approach the physical domain through concept does not
annihilate its categorically distinct properties. Certainly, the waiter needs to have a
rational grasp of the difference but without the physical counterpart of the rational
difference between a knife and a plate he would have been unable to physically
coordinate them next to each other. Yet, an objection might have it, we do not really
need a physical coordination at all in order to have a one-to-one coordination. All
we need is the rational one. Appealing as this is, this depends on the given situation.
For some situations we really do not need it, and for some we do; for example, in
situations when we do not have a rational grasp of what is to be coordinated with
what we seem to be unable to pass without it.
To illustrate, imagine two baskets full with things that the coordinating agent is
not aware of. And she is asked to put one entity from the basket A next to an entity
from a basket B on a table but without looking into the baskets. If the entities have
random sticky sides they will form a sort of aggregate, to use Mill’s expression, but
the agent would not be able to predict what aggregate would be formed. For the
aggregation would be, besides the initial physical conditions in the basket, formed
entirely due to her putting her hands in the basket and changing the geometrical
configuration of the sticky sides while trying to “take a physical object”. Now, in this
and in similar cases, it is obvious that at the time of the coordination (let’s furnish
the example by her not seeing and not feeling the true configuration of the aggregate
even during her taking it out of the basket and even when putting it on the table:
we can conceive of suitable weight and swinging behaviour of the sticky entities,
perhaps even electronically adjusted in real time, such that she would not know
how many of them would form an aggregate; she would only learn the randomly
occurred aggregation that would form an entity to be one-to-one coordinated when
she looks at it on the table) the agent has a rational grasp neither of the conditions
of the individuation of the objects nor of the exact aggregates to be coordinated
with each other. In such an occasion we would have a physical coordination of one-
to-one aggregates from basket A and basket B but we would not have the rational
coordination. In the best scenario we would have some sort of a false perceptual
feeling which even if leading to a rational belief (which is not necessary at all: the
agent might be aware that what she feels is in most cases not the actual aggregate
formed; or we could just leave her to think what she wants and in this case she would
end up having a false rational coordination about the objects to be coordinated) that
differs from the physical coordination on the table; the agent might think that there

Principia 13(3): 327–38 (2009).

Fregean One-to-one Correspondence and Numbers as Object Properties 335

are two by two coordinated entities while there are actually not and there is just
one A entity on the one side put in one-to-two correspondence with two B entities.
This case might be reached for when an aggregate put on the table sticks to another
already on the table and she is not aware of this: given that the most intuitive
conditions of individuations of objects of coordination are not the sticky elements of
the aggregates but the aggregates themselves (we could complicate the experiment
even more by eliminating spatial individuation of separate elements at all by making
them completely lose the element identity they have from the time in the basket but
this does not seem necessary for the point).
An immediate objection might argue that the basket situation shows nothing
above a merely epistemic problem and thus it is not a problem for metaphysics.
Such objection, nevertheless, fails. For obviously there is an epistemic component in
the thought experiment and at that one that does not look comfortable for Frege’s a
priori attitude. But the metaphysical point is the central one: how could the one-to-
one correspondence be numerically independent from the properties of the objects
if it is a correct correspondence? How do we know that the concept “knife” and
the concept “plate” are equinumerous in Frege’s illustration from #70 if there is one
knife on the table and 5 plates? Obviously there is no one-to-one correspondence
between the knives and the plates. But how do we know that? Besides the epistemic
story, which in the case of properties of physical objects would be one or another
version of the contemporary empiricism, such as the one recently being offered by
Philip Kitcher and particularly for mathematical cases, there should be something in
the domain of observation that allows for the correct participation of the objects in
the rational one-to-one correspondence. And this is metaphysical enough.
Another objection might stem from the power of rationality. What if the basket
case is not possible without concepts? What if the agent could not and actually does
not coordinate the A aggregates and the B aggregates if she does not have the rele-
vant concept tied in proper propositional relations? This objection does not manage
to deliver the goods too. We can imagine an inanimate machine, not even a robot,
which by random repetition performs the same coordination. It has neither concepts
nor what we would call “rationality”. We would have the same physical result. In
this case the question about the relevance of the entity individuation of the randomly
coordinated aggregates would matter, of course, not for the machine, which is not a
rational agent, but for any rational agent, who is interested in the problem. Is the
result of the machine coordination relevant for such cases? Well, when entertained
by a rational agent who attempts to define a one-to-one correspondence between
the As and Bs this is a most relevant result. And an agent who wishes to clarify the
subject sufficiently well could not allow herself to ignore such cases. We should be
philosophically interested about such cases as if they were attempted as one-to-one
coordinated. Merely ignoring the case as irrelevant would not help much.

Principia 13(3): 327–38 (2009).

336 Boris Grozdanoff

Similar illustrations show that the physical portion of the coordination is not
merely dispensable with but necessary for Frege’s extension conception. For all sci-
entific knowledge which deals with the physical world needs physical objects, the
same ones which in virtue of being an extension of the concepts to be equinumer-
ous, need to be equinumerous as well. And in failed coordinating cases like the
above basket case, even if we form propositions about the coordination it would be
a coordination that would not be true by not being isomorphic to the primitive struc-
ture of the natural numbers. For applied mathematical cases, the physical portion of
the coordination, that is, to use Frege’s wording, the physical portion of the one-to-
one correspondence, is a necessary part which is not linguistic. And for successful
cases like the one of knives and plates, we obviously need something in the physical
world that would allow us to get the rational coordination right too. This something
has the heavy responsibility to deliver the physical conditions of the possibility to
physically individuate an entity that could physically and rationally participate in a
one-to-one correspondence. The best candidate for such “something” is most suit-
ably captured in terms of the familiar numerical properties, not so different from the
ones put forward by Mill in The System of Logic and criticized somewhat hasty by
Frege in The Foundations.
The extension portion of Frege’s famous view on the nature of numbers proved
to be untenable in view of the Russell’s paradox or, in the best possible scenario,
proved to be in a need of heavy modifications in order to become vital. The property
portion is a separate case. Thoughts in terms of properties are intimately connected
with quantitative expressions and numerical ones in particular. On the suggested
above account the property portion of Frege’s view is accepted and complemented,
unexpectedly, with a neo-Millean sort of view according to which numbers are also
properties of objects, be they classical physical entities or more exotic objects like
abstract ones. The truths of applied mathematics thus need not be miracles any-
more. In truthful natural scientific cases we will have a successful match between a
set of conceptual properties SC and a set of object properties SO . The mapping is a
task for the philosophy of science but by no means a miraculous one. History of Phi-
losophy of Science shows that from Reichenbach’s famous principles of coordination
to Michael Friedman’s contemporary account of constitutive a priori principles there
are non-miraculous positions struggling to spell out the coordination. Together with
prevailing naturalist doctrines like the neo-Millean one of Philip Kitcher of applied
mathematical truths they shape a significantly rich body of philosophical choices,
which could now be metaphysically shown to be internally coherent given the now
common metaphysical account of mathematical properties as belonging both to the
abstract and the physical domains.3

Principia 13(3): 327–38 (2009).

Fregean One-to-one Correspondence and Numbers as Object Properties 337

Frege, G. 1884. Die Grundlagen der Arithmetik. (References to the paragraphs of this work
are to the English translation, The Foundations of Arithmetic, translated by J. L. Austin,
2nd rev. edition, Illinois: Northwestern University Press, 1980.)
Kitcher, P. 1998. Mathematics and the Naturalist Tradition. In J. Skorupski (ed.) The Cam-
bridge Companion to Mill. New York: Cambrigde University Press.
Lambros, C. 1976. Are Numbers Properties of Objects. Philosophical Studies 29(6): 381–9.
Mendelsohn, R. 1989. On an Alleged problem for Frege’s account of number. Philosophical
Studies 56: 193–7.
Mill, J. S. 1874. A System of Logic. New York: Harper & Brothers.
Minogue, B. 1977. Numbers, Properties, and Frege. Philosophical Studies 31: 423–7.
Wigner, E. 1967. The unreasonable effectiveness of mathematics in the natural sciences. In
his Symmetries and Reflections. Bloomington: Indiana University Press, p. 222–37.

Marie Curie Research Fellow
Philosophy Department
Wolfson College
University of Oxford, UK

Resumo. Este artigo examina criticamente uma linha impopular da concepção fregeana
de número nos Fundamentos da aritmética. Segundo essa concepção, que analisa números
em termos de propriedades e não em termos de extensões, números são propriedades de
conceitos vs. propriedades de objetos. Essa última concepção é sustentada por Mill e cele-
bremente criticada nos Fundamentos. Sustento que na explicação de propriedade os números
não podem somente ser propriedades de conceitos mas também têm que ser propriedades de
objetos. Meu argumento principal baseia-se em razões puramente metafísicas. Origina-se da
motivação de que, se números fossem somente propriedades de conceitos, ou não seríamos
capazes de explicar verdades matemáticas sobre o mundo físico ou essas verdades seriam
milagrosas. Considerando que de fato temos verdades matemáticas sobre o mundo físico
que não são milagrosas, não podemos concordar com a concepção de propriedade de Frege
acerca da natureza metafísica dos números.

Palavras-chave: Números, propriedades, correspondência biunívoca, Frege.

For a discussion on the controversy see Minogue 1977, 423–7 and Mendelsohn 1989, 193–

Principia 13(3): 327–38 (2009).

338 Boris Grozdanoff

A knife is metaphysically possible only as one knife. The very term “knife”, when it refers
to the real physical object “knife”, is meaningful only if it denotes something numerically
individuated. An entity comes with its numerical individuation.
This paper has been written while I was a Marie Curie IEF Fellow at the University of
Oxford and all the help of the EC is gratefully acknowledged.

Principia 13(3): 327–38 (2009).

University of Warwick

Abstract. In order to measure the degree of dissimilarity between elements of a Boolean

algebra, the author’s (1984) proposed to use pseudometrics satisfying generalizations of
the usual axioms for identity. The proposal is extended, as far as is feasible, from Boolean
algebras (algebras of propositions) to Brouwerian algebras (algebras of deductive theories).
The relation between Boolean and Brouwerian geometries of logic turns out to resemble in a
curious way the relation between Euclidean and non-Euclidean geometries of physical space.
The paper ends with a brief consideration of the problem of the metrization of the algebra
of theories.

Keywords: Boolean algebra, Brouwerian algebra, metric, measure, partial identity.

0. Background

In (1984) I proposed an axiom system A for a function d on a poset L that in a

rather straightforward manner generalizes and strengthens the usual elementary
axioms for identity. The value of d(a, c) can be understood as one way, by no means
the only one, of grading the degree of dissimilarity or diversity of the elements a
and c of L. The function d is not required to be real-valued, but that is its most
natural interpretation, in which case, as is easily demonstrated, it satisfies the usual
requirements of a pseudometric operation. The system A was applied to an arbitrary
Boolean algebra L, for which it was proved to be equivalent, given some appropriate
definitions, to two other axiomatic systems: B, a different and more familiar set of
axioms for a pseudometric operation d; and C, the standard system of axioms for an
unnormalized measure or positive valuation function µ. As was remarked, the close
relation between the systems B and C in Boolean algebras (in which every strictly
positive unnormalized measure is also a positive isotone valuation) is well known.
The paper also drew attention to a number of variations and generalizations, some
of them first mentioned in Miller (1979). Several of the derivations, for example, but
not all of them, can be carried out when L is assumed not to be a Boolean algebra,
but only a lattice. It was noted that if L is a lattice, the system B, together with the
assumption that the pseudometric d is a metric (that is, d(a, c) = 0 only if a = c),
compels L to be modular (Birkhoff 1967, Theorem X.2). It was observed also that
if one of the axioms in B is strengthened from an inequality to an identity, then L

Principia 13(3): 339–56 (2009).

c Copyright D. W. Miller 2009
Published by NEL — Epistemology and Logic Research Group, Federal University of Santa Catarina (UFSC), Brazil.
340 David Miller

is forced to be distributive. The derivations, furthermore, were shown to be valid

when the range of d is not the real line but an arbitrary partially ordered abelian
semigroup with unit and some weak cancellation properties.
The Boolean operation 4 of symmetric difference plays an important role in many
of the derivations in the systems A and B, and in the definition in the system C of
the pseudometric d in terms of the measure µ. Yet no thought was given in (1984)
to whether similar results could be obtained for Brouwerian (dual-intuitionistic) al-
gebras, where too the symmetric difference operation is defined and is known to
possess pseudometric properties (Nordhaus & Lapidus 1954, Miller 1978, § 3). This
was a conspicuous omission. The algebra of deductive theories of a language in-
corporating classical logic, whose Lindenbaum–Tarski algebra is Boolean, is known
to be Brouwerian in general (provided that, as in the Lindenbaum–Tarski algebra,
X  Z means that X is logically stronger than Z and not, as in the original presenta-
tion in Tarski 1935/1936, that X ⊆ Z). Since the problem that prompted the original
investigation was to say something enlightening about the idea of the distance of a
theory from the truth, or its distance from another theory (Miller 1977, 1978), it is
plainly of interest to develop further the Brouwerian case.
Another matter that needs to be addressed is the question of the existence of a
metric d and of a strictly positive measure µ. Pseudometrics satisfying the systems
A and B abound, as do measures satisfying the system C. We certainly need to be
more specific. But as recorded in Miller (1986), p. 174, these two goals are not
easily attainable together. Unless the Brouwerian algebra L is Boolean, there exists
no metric d obeying the axiom system A, even if L is finite.
In this paper the axiom systems A and C will be replaced by systems A? and C?
that overcome the main difficulties sketched. A? and C? will be shown to be logically
equivalent in Boolean algebras to A and C respectively (and therefore to B), and to
be logically equivalent in Brouwerian algebras to each other. The system B will be
shown to be equivalent in Boolean algebras to a somewhat more transparent system
B\ , which in its turn can be modified to a system B\? that is appropriate to Brouw-
erian algebras. We shall note that there exists another way of providing axioms
for pseudometric operations in Brouwerian algebras, which abandons the system A
and retains the system B without modification (and makes a trifling change to the
system C here presented). Results obtained by Mormann (2006) under this alter-
native approach will be adapted to show that every Brouwerian algebra of theories
(of a denumerable language) can be metrized by a metric that complies with the
characteristic axiom (A? 1) of the system A? .

Principia 13(3): 339–56 (2009).

A Refined Geometry of Logic 341

1. Three Axiom Systems

The letters a, b, c are used as variables for the elements of an arbitrary lattice L
with ordering . The join and meet operations of L are represented by + and ·
respectively, and the unit and zero elements (if these elements exist) by > and ⊥
respectively. The symbols + and = are used also for operations in some partially
ordered semigroup with unit, the non-negative real numbers unless otherwise stated,
which is the space of values for two functions d and µ defined on L. There should
not be any danger of confusion between the different uses of these symbols.
We first present the three axiom systems A, B, C already mentioned. The axioms
of A are:

(A0) µ(b) =Df d(b, ⊥)

(A1) d(a, c) + d(ϕa, b) ≥ d(ϕc, b)
(A2) abc ⇒ d(a, c) = d(a, b) + d(b, c),

where in (A1), which is a scheme of axioms, the term ϕc is identical with the term
ϕa except perhaps for containing c at one or more places at which ϕa contains a.
The axioms of B are:

(B0) µ(b) =Df d(b, ⊥)

(B1) d(b, a) + d(b, c) ≥ d(a, c)
(B2) d(a, c) ≥ d(b + a, b + c) + d(b · a, b · c).

This system of axioms (B1 is just the triangle inequality) derives from Theorem 1
of Chapter X of Birkhoff (1967). Note that the definition (B0) of µ in terms of d
is identical with (A0). C is a familiar system of axioms for a measure µ. The term
a 4 c in the definition (C0) of d in terms of µ is the Boolean symmetric difference
(exclusive disjunction) a · c 0 + c · a0 of a and c.

(C0) d(a, c) =Df µ(a 4 c)

(C1) µ(⊥) = 0
(C2) µ(b) ≥ 0
(C3) µ(a) + µ(c) = µ(a + c) + µ(a · c).

If the lattice L is a Boolean algebra, the systems A, B, and C are logically equiv-
alent (Miller 1984). In § 3 we shall add a fourth equivalent system B\ , like B but
somewhat more transparent.

Principia 13(3): 339–56 (2009).

342 David Miller

2. Degrees of Diversity

The axiom systems B and C need no immediate discussion, but something must be
said about A. In elementary logic the relation of identity or equality is standardly
introduced by the axioms

(E0) b = b
(E1) a = c ∧ Φa ⇒ Φc,

where Φa is a formula perhaps involving a, and Φc is like Φa except perhaps for

containing c free at one or more places where Φa contains a free. If we write O for
the falsum or absurdity (of the metalanguage), and ⇐ for the converse of ⇒ (⇐ is
to be read if and ⇒ is to be read if . . . then), then we may accordingly characterize
also the relation 6= of diversity by the axioms

(E2) b 6= b ⇔ O
(E3) a 6= c ∨ ¬Φa ⇐ ¬Φc.

In the case of algebraic languages, in which all atomic formulas are identities, the
second principle (E3), the principle of the diversity of discernibles, may be rendered
in one of the forms

(E4) a 6= c ∨ ϕa 6= ψa ⇐ ϕc 6= ψc,
(E5) a 6= c ∨ ϕa 6= b ⇐ ϕc 6= b,

where ϕa and ψa are terms perhaps involving a and the conventions on substitu-
tion are as before. In general these statements are not logically equivalent; (E5) is
weaker than (E4). A theory of the partial diversity or degree of dissimilarity of two
elements of an algebra may be obtained by generalizing these axioms (E0) and (E4)
in terms of a (typically real-valued) function d satisfying

(E6) d(b, b) = 0
(E7) d(a, c) + d(ϕa, ψa) ≥ d(ϕc, ψc).

In Boolean algebras, the scheme (E7) is equivalent to the restricted version called
(A1) above:

(E8) d(a, c) + d(ϕa, b) ≥ d(ϕc, b).

When the axiom (E6) and the scheme (E7) are augmented by the obvious definition
of identity,

(E9) a = c =Df d(a, c) = 0,

Principia 13(3): 339–56 (2009).

A Refined Geometry of Logic 343

they yield a set of axioms for degrees of dissimilarity or diversity. In the bulk of this
paper the westward half of (E9) is not assumed, and it is only in § 6 that metrics
come to the fore.
Consider first a degenerate algebra, with no operations. As a is the only term
that involves a,

(E10) d(a, c) + d(a, a) ≥ d(c, a), E7

(E11) d(a, c) = d(c, a), E6, E10
(E12) d(a, c) + d(a, b) ≥ d(c, b), E7

are all we can extract from (E6) and (E7). In other words, d is a pseudometric
operation. If (E9) holds too, then d is a metric operation. If d is defined on an
algebra L that contains operations, a strong condition of uniform continuity (E14),
the Lipschitz condition, is satisfied:

(E13) d(a, c) + d(ϕa, ϕa) ≥ d(ϕc, ϕa) E7

(E14) d(a, c) ≥ d(ϕa, ϕc) E6, E11, E13
(E15) d(a, c) ≥ d(b, b) = 0 E6, E14

Inequality (E14), although weaker than (E7), is easier to work with than either (E7)
or (E8). It follows from (E8) alone, and not until step (A5) below is the full strength
of (E7) needed. Note also that from (E14) and the triangle inequality (E12) we can
derive (E8), but not (E7).
The next results invoke the assumption that the algebra L is a lattice with join +
and meet ·.

(E16) d(a, c) ≥ d(b + a, b + c) E14

(E17) d(a, c) ≥ d(a · b, c · b) E14
(E18) d(c, a · c) ≥ d(a + c, a + a·c) E16
(E19) d(c, a · c) ≥ d(a + c, a) E18
(E20) d(a + c, a) ≥ d((a + c)·c, a · c) E17
(E21) d(a + c, a) ≥ d(c, a · c) E20
(E22) d(a + c, a) = d(c, a · c) E19, E21

The lattice quadrangles are parallelograms. The isotony principles (E23) and (E24)
follow easily.

(E23) a  b  c ⇒ d(a, c) ≥ d(a, b) E16

(E24) a  b  c ⇒ d(a, c) ≥ d(b, c). E17

Principia 13(3): 339–56 (2009).

344 David Miller

3. Radial Convexity

The metric part of the axiomatic system A presented in § 1 above is obtained by

adding to the scheme (A1), which was labelled (E8) in § 2, the following principle
of additivity along chains:

(A2) a  b  c ⇒ d(a, c) = d(a, b) + d(b, c).

Since all the metrical formulas (E6)–(E24) of § 2 hold for the trivial metric (whose
value is 0 when a = c and 1 when a 6= c), (A2) is an independent assumption.
Mormann (2006), definition 2.6, calls those (pseudo)metric operations d for which
(A2) holds radially convex (pseudo)metrics.
In Miller (1984) there is a short proof that, if L is a Boolean algebra, the scheme
(A1) and the axiom (A2) imply that the lattice quadrilaterals are rectangles (see
§ 1, lines 14–21). It is shown also that this result, which is proved in formula (A8)
below, is not forthcoming for arbitrary lattices, even Boolean lattices. Yet the re-
sult is demonstrable for all lattices once the scheme (A1), formerly called (E8), is
strengthened to (E7), which we here formally relabel:

(A? 1) d(a, c) + d(ϕa, ψa) ≥ d(ϕc, ψc).

Since these two schemes (A1) and (A? 1) are logically equivalent in Boolean algebras
(Miller op.cit., formula 23), and it is (A? 1) that we shall eventually want to adopt
for other algebras, we shall work in this section in the axiomatic system A+ whose
metrical postulates are (A? 1) and (A2) (the definitional axiom (A0) will not be
needed). Our purpose is to show how Birkhoff’s complicated axiom (B2) may be
simplified, whatever the lattice; in the presence of (B1), it obviously implies radial
convexity (A2), but it is itself a consequence of (A2) and rectangularity.
(E22) states that the opposite sides of a typical lattice quadrilateral are of equal
length. We show now, within the system A+ , that the diagonals of the quadrilateral
also are of equal length.

(A3) d(b, b) + d(b, b) = d(b, b) A2

(A4) d(b, b) = 0 A3
(A5) d(a, c) + d(a + a, a · a) ≥ d(a + c, a · c) A? 1
(A6) d(a + c, a · c) + d(a + c + a, a + c + c) ≥ d(a · c + a, a · c + c) A? 1
(A7) d(a + c, a · c) ≥ d(a, c) A6
(A8) d(a, c) = d(a + c, a · c). A4, A5, A7

Principia 13(3): 339–56 (2009).

A Refined Geometry of Logic 345

t !@
a+c ! L
L L@
L L @
L L @
L L @
L L @
L L @
L L @
L L @t (b + a)·(b + c)
L L t b+a·c
L L !!L
L !
t ! Lb
b ·(a + c) L! L
b·a+b·c t
@ L L
@ L L
@ L L
@ L !Lt a · c
@L !!

Figure 0: A diagrammatic demonstration that (B2) follows from radial convexity

plus the rectangularity of the lattice quadrilaterals.

We are now able to give a proof of Birkhoff’s axiom (B2) within the system A+ ,
and further to show that, if the lattice L is distributive, then the inequality in (B2)
is turned into an identity:

(A9) d(a, c) = d(b + a, b + c) + d(b · a, b · c).

By (A8), d(a, c) = d(a + c, a · c). Figure 0 shows that if to the two ends of the interval
[a + c, a · c] the intervals [b + a + c, a + c] and [a · c, b · a · c] are added, we obtain a
path through the lattice with length, by radial convexity (A2), equal to

d(b + a + c, (b + a) · (b + c)) + d((b + a) · (b + c), b + a · c)+

+d(b + a · c, b) + d(b, b · (a + c))+
+d(b · (a + c), b · a + b · c) + d(b · a + b · c, b · a · c).

The two middle terms here are, by (E22), equal in length to the two added inter-
vals, while the two outer terms are, by (A8), equal in length to d(b + a, b + c) and
d(b · a, b · c). The other two terms are non-negative, and if L is distributive, zero.
Birkhoff’s axiom (B2) is therefore derivable from the formulas (A2), (E22), and
(A8). It is clear that (B2) is satisfied by any function d that has a constant negative
value (say, −1), and hence it cannot ensure the triangular inequality, which above

Principia 13(3): 339–56 (2009).

346 David Miller

was called (B1), or even d(b, b) = 0, which was called (E6). Since (E6) follows from
(A2), we have to assume more than (B2) if we are to prove (A2). We shall show that
the three formulas (A2), (E22), and (A8) are all derivable from (B2) together with
(B1); that is, that they are theorems of the system B. The following lines correct an
error in the derivation on lines 0–15 and 26–31 of § 2 of Miller (1984).

(B1) d(b, a) + d(b, c) ≥ d(a, c)

(B2) d(a, c) ≥ d(b + a, b + c) + d(b · a, b · c)
(B3) a  b  c ⇒ d(a, c) ≥ d(a, b) + d(b, c) B2
(B4) d(b, b) + d(b, b) = d(b, b) B1, B2
(B5) d(b, b) = 0 B4
(B6) d(b, a) = d(a, b) B1, B5
(B7) a  b  c ⇒ d(a, c) = d(a, b) + d(b, c) B1, B3, B6
(B8) d(b, a) ≥ 0 B1, B5
(B9) d(a, c) ≥ d(b · a, b · c) B2, B8
(B10) d(a + c, a) ≥ d(c · (a + c), c · a) B9
(B11) d(a, c) ≥ d(b + a, b + c) B2, B8
(B12) d(c, a · c) ≥ d(a + c, a + a · c) B11
(B13) d(a + c, a) = d(c, a · c) B10, B12
(B14) d(a, c) ≥ d(a + a, a + c) + d(a · a, a · c) B2
(B15) d(a, c) ≥ d(a + c, a · c) B1, B14
(B16) d(a + c, a · c) ≥ d((a · c) + a, (a · c) + c) +
+ d((a · c) · a), (a · c) · c) B2
(B17) d(a, c) = d(a + c, a · c) B5, B15, B16

The formulas to be proved, (A2), (E22), and (A8), are at lines (B7), (B13), and
(B17), respectively.
These results encourage the recognition of a fourth axiom system B\ , a notable
simplification of the system B, and equivalent to each of A, B, and C in Boolean
algebras. B\ has five axioms, the definition (B\ 0), and the principles of triangularity
(B\ 1), translation invariance (B\ 2), torsion invariance (B\ 3), and radial convexity
(B\ 4):

(B\ 0) µ(b) =Df d(b, ⊥)

(B\ 1) d(b, a) + d(b, c) ≥ d(a, c)
(B 2) d(a + c, a) = d(c, a · c)

Principia 13(3): 339–56 (2009).

A Refined Geometry of Logic 347

(B\ 3) d(a, c) = d(a + c, a · c)

(B 4) abc ⇒ d(a, c) = d(a, b) + d(b, c).

4. Brouwerian Algebras

Although, as we have seen, a good number of the derivations succeed in all lat-
tices, the investigation in Miller (1984) was carried out, for the most part, under
the hypothesis that the lattice L is a Boolean algebra. For the rest of the paper we
shall loosen this hypothesis, and assume only that the lattice on which d is defined
is a Brouwerian algebra; that is to say, L is a lattice with unit > in which for ev-
ery two elements a, c there exists a smallest element b such that b + c  a. This
element, written a − c, is called the remainder or difference when c is subtracted
from a. Brouwerian algebras are dual to Heyting algebras, the algebras correspond-
ing to intuitionistic logic (in which for every two elements a, c there is a largest
element b, the conditional a → c, such that b · a  c). Brouwerian algebras, like
Heyting algebras, are invariably distributive, and all finite distributive lattices, in
particular finite chains, are Brouwerian algebras. Just as in a Heyting algebra each
element b has a pseudocomplement b → ⊥ that obeys the law of non-contradiction
b · (b → ⊥) = ⊥, so in a Brouwerian algebra each element b has an authentic com-
plement >− b that obeys the law of excluded middle b+(>− b) = >; it will be called
the authocomplement of b, and written b 0 . The symmetric difference of a and c, which
is defined by

(D0) a 4 c =Df (a − c) + (c − a),

will continue to play a crucial role in what we do. It must not be forgotten that
in Brouwerian algebras neither of the Boolean identities a − c = a · c 0 and a 4 c =
a · c 0 + c · a0 is generally valid.
Because of the relative unfamiliarity of the laws of the non-Boolean remainder
and symmetric difference (D0), we list below without proof the main ones appealed
to in the rest of the paper. They are duals of perhaps more homely laws of the intu-
itionistic conditional and biconditional. Because of its importance, attention should
be drawn to the law (D10), in which χ c is a term like χ a except for containing c
at one or more places where χ a contains a. This is the dual of the intuitionistic
law of replacement: p ↔ r ` X p ↔ X r, where X p is a formula that contains free
the variable p and X r is the result of replacing one or more instances of p by the
variable r.

(D1) a−c = (a + c) − c
(D2) a−c = a−a·c

Principia 13(3): 339–56 (2009).

348 David Miller

(D3) (a − c) · c = (a − c) · (a · c)
(D4) a4c = (a + c) 4 (a · c)
(D5) a+c = (a 4 c) + a · c
(D6) a4c ≥ (a 4 b) + (b 4 c)
(D7) a  b  c ⇒ a 4 c = (a 4 b) + (b 4 c)
(D8) a  c ⇔ (a 4 c) + c = a
(D9) ac⇒a4c = a−c
(D10) a4c  χa 4 χc
(D11) > = ⊥
0 0
(D12) (b · b ) = >
(D13) b−b = ⊥
(D14) b−⊥ = b
(D15) b4⊥ = b
(D16) b4b = ⊥

In the deductively more exiguous setting of Brouwerian

algebras the equivalence of A, B, and C breaks down at 1 u µ(1) = 2
a number of points. The most significant difficulty is that
what distinguishes Boolean from non-Boolean Brouwerian
algebras is the presence in the latter of at least one ele-
ment b for which the law of non-contradiction b · b 0 = ⊥
fails. But (D12) (b · b 0 ) 0 = > is universally valid. As
(E11) and (E14) above show, it follows from (A1) that
d(a, c) ≥ d(ϕc, ϕa), and therefore by (D11) that d(>, b · b 0 ) b u µ(b) = 1
0 0 0
≥ d((b · b ) , > ) = d(>, ⊥). By (A2) we may conclude that
d(b · b 0 , ⊥) = 0. That is to say, on a proper Brouwerian alge-
bra L, as we shall henceforth call a Brouwerian algebra that
is not Boolean, there exists no genuine metric (as opposed to
pseudometric) satisfying the axiom system A. If conformity
with A is required, then the Brouwerian algebra is effectively
booleanized. Yet system B certainly admits a metric d, and 0 u µ(0) = 0
system C admits a measure µ that is positive isotone (and
Figure 1: The three-
therefore strictly positive). For the simplest of all proper
element chain
Brouwerian algebras, the three-element chain, possible val-
ues of µ are recorded in Figure 1. For each a ≥ c, we may set d(a, c) = µ(a) − µ(c).
In this algebra, with this metric, the authocomplement of b is the unit element 1,
and so b·b0 = b. The axioms of A are therefore transgressed.

Principia 13(3): 339–56 (2009).

A Refined Geometry of Logic 349

In a Brouwerian algebra, that is, neither B nor C implies A. Nor does A imply
either B or C, though this can be taken care of by strengthening (A1) to the already
familiar generalization,

(A? 1) d(a, c) + d(ϕa, ψa) ≥ d(ϕc, ψc),

It may be noted again that this alternative formulation makes no difference in Bool-
ean algebras, where (A1? ) follows from (A1) (without the use of A2). Another
shortcoming, which is relatively easily taken care of, is the failure of C to imply B.
To recover for Brouwerian algebra this implication we have to make two reforms;
one is to replace the definition (C0) with the variant

(C+ 0) d(a, c) =Df µ(a + c) − µ(a · c),

(Birkhoff op.cit., Theorem X.1), and the other is to add an axiom (to be called (C? 4)
below) that states that the measure µ respects the ordering  (that is, µ is iso-
tone). Neither of these changes can be thought of as momentous amendments to the
original system C, since in Boolean algebras (C0) and (C+ 0) are logically equivalent
given the other axioms (since a · c and a 4 c are disjoint), and (C? 4) is a consequence
of these other axioms (since a · c and a · c 0 are disjoint).
It is not difficult to see in what way the definition (C+ 0) differs from (C0) in
Brouwerian algebras. (D5) (a 4 c) + a · c = a + c is a theorem of Brouwerian algebra.
By additivity (C3),

(C+ 1) µ(a 4 c) + µ(a · c) = µ(a + c) + µ((a 4 c) · (a · c))

(C 2) µ(a + c) − µ(a · c) = µ(a 4 c) − µ((a 4 c) · (a · c)).

Since (a 4 c) · (a · c) = ⊥ is not a valid identity in Brouwerian algebra, in general

(C+ 0) and (C0) are distinct ways to define the function d. An explicit axiom of
isotony, such as (C? 4), is needed in Brouwerian algebras because axiom (C3) is
vacuous in a chain. In the chain depicted in Figure 1, for example, µ(b) and µ(1)
may be independently assigned any non-negative values.
It is apparent that the kinship in Boolean algebras between distance functions
(and measures) and the more abstract degrees of dissimilarity characterized by the
axiom system A cannot exist in Brouwerian algebras unless these axioms are some-
how weakened. The natural way to proceed is to weaken axiom (A1), even though,
as we have just noted, it has to be strengthened as well. After all, the central prob-
lem arises from the discontinuity within Brouwerian algebras of the operations of
authocomplementation and remainder (Miller 1986); from the fact, already visible
in Figure 1, that adjacent elements (there 1 and b) can have authocomplements
(there 0 and 1) that are not adjacent. This phenomenon seems to be blatantly in

Principia 13(3): 339–56 (2009).

350 David Miller

conflict with the intuitive idea behind (A1), that if a and c are similar then there is
an upper bound to how dissimilar ϕa and ϕc can be. None of this can be denied.
Note however that the three-element chain is quite representative of all finite chains:
if y is the immediate inferior of the unit element 1, then y 0 = 1 and 10 = 0 are as
widely separated by the lattice ordering  as they can be. In other words, there is
no upper bound in Brouwerian algebras to the number of elements (or equivalently
the number of non-zero intervals), between the authocomplements of adjacent el-
ements; the intuitive distance d(ϕa, ϕc), that is, is generally unconstrained by the
value of d(a, c). In these circumstances the idea of adding a correction term to axiom
(A1) seems distinctly unpromising.
The simplest policy, which is also the most conservative one, therefore seems
to be to abandon the axiom system A altogether, to replace (C0) by (C0+ ), and to
add the isotony axiom (C? 4). It can then be shown, in a standard manner, that the
new system, which we may call C+, is logically equivalent to the system B: that
C+ implies B is the content of Birkhoff op.cit., Theorem X.1, while the proof that B
implies C+ is contained within lines 0–15 and 35–40 of Miller (1984), § 2. (In these
lines nothing more is assumed about L than that it is a lattice. There is an error at
lines 7f., which is implicitly corrected on p. 346 above.) But as announced in § 0,
our purpose here is to preserve the spirit of the system A by introducing a variant
of axiom (A2), and at the same time strengthening (A1) to (A? 1). This variation
will enable us to prove the equivalence of the new system A? with the system C? ,
which consists of the same axioms as the old system C, together with the isotony
axiom (C? 4). What we have to surrender, in order to achieve this, is equivalence
with the system B, since its characteristic axiom (B2) has to be so harshly mutilated
that it no longer qualifies for participation in a respectable axiomatic system. But
fortunately the variant B\ is at hand. If we modify its axiom (B\ 4), which is identical
with axiom (A2), in an identical way, we obtain a system B\? completely equivalent
with A? and C? .

5. Radial Convexity Refined

The axioms of the new system C? are those of C, augmented with an axiom of
isotony (C? 4).

(C? 0) d(a, c) =Df µ(a 4 c)

(C 1) µ(⊥) = 0
(C 2) µ(b) ≥ 0
(C 3) µ(a) + µ(c) = µ(a + c) + µ(a · c)
(C 4) ac ⇒ µ(a) ≥ µ(c).

Principia 13(3): 339–56 (2009).

A Refined Geometry of Logic 351

The following derivations are conducted within this system C? , together with
some of the laws of Brouwerian algebra listed above. Applying (C? 3) and the defi-
nition (C? 0) to (D7), we obtain

(C? 5) a  b  c ⇒ µ(a 4 c) = µ(a 4 b) + µ(b 4 c) − µ((a 4 b) · (b 4 c)),

(C? 6) a  b  c ⇒ d(a, c) = d(a, b) + d(b, c) − µ((a 4 b) · (b 4 c)).

Thanks to the laws (D15) b 4 ⊥ = b and (D9) a  c ⇒ a 4 c = a − c, (C? 6) may be


(C? 7) a  b  c ⇒ d(a, c) = d(a, b) + d(b, c) − d((a 4 b) · (b 4 c), ⊥);

(C 8) a  b  c ⇒ d(a, c) = d(a, b) + d(b, c) − d((a − b) · (b − c), ⊥).

Since (a − b) · (b − c) = ⊥ is not a Brouwerian theorem, even when a  b  c

(its dual is: if p ` q ` r then ` (r → q) ∨ (q → p), which is intuitionistically invalid,
though classically valid), the final term in (C? 8) will be greater than 0 for any metric
d; that is, for any pseudometric that satisfies (E9). In Brouwerian algebras no metric
constrained by C? can be radially convex.
Our proposal — it may seem quite ad hoc — is to modify the system A by re-
placing (A2) with (C? 8), which is a Boolean equivalent. This new axiom, unlike
(A2), makes sense only in lattices in which the remainder operation is defined. The
axioms of the revised system A? are:

(A? 0) µ(b) =Df d(b, ⊥)

(A 1) d(a, c) + d(ϕa, ψa) ≥ d(ϕc, ψc)
(A 2) abc ⇒ d(a, c) = d(a, b) + d(b, c) −
− d((a − b) · (b − c), ⊥)
(A 3) d(b, b) = 0.

Note that axiom (A? 3) must be stated independently. Without it, d could be any
constant non-zero function. (Thanks to (A? 2), and (D13), it suffices to postulate
d(⊥, ⊥) = 0.) The abbreviation

(A? 4) κ(a, b, c) =Df d((a − b) · (b − c), ⊥)

allows (A? 2) to be expressed more concisely. The correction term κ(a, b, c) may be
thought of as a measure of the curvature (the deviation from linearity) of chains
from a to c that pass through the point b. Since (> − b) · (b − ⊥) = b 0 · b 
(a− b)·(b−c), the curvature κ(a, b, c) is bounded above by d(b· b 0 , ⊥), which implies
that κ(a, b, c) = 0 if b has a Boolean complement (that is, if b· b 0 = ⊥). As explained

Principia 13(3): 339–56 (2009).

352 David Miller

in § 0 above, the deductive theories of a language form a Brouwerian algebra, while

according to Tarski (1935/1936), Theorem 17, the axiomatizable deductive theories
are precisely those that conform to the law of non-contradiction. We may therefore
combine the languages of general metamathematics and general relativity to say
(not entirely seriously) that Boolean elements (propositions) possess zero mass, and
only unaxiomatizable theories can engender deflections from straightness. Note,
however, that the converse fails: (a − b) · (b − c) = ⊥ does not imply b · b 0 = ⊥. For
example, if a = b or b = c, then by (D13) and (A? 0), κ(a, b, c) = 0.
In Boolean algebras the systems A? and C? reduce to A and C respectively. We
turn now to the task of proving that in Brouwerian algebras A? and C? are equivalent
to each other, starting with the proof that C? implies A? . It follows at once from
(D15) and (C? 0) that (A? 0) holds, and from (D16), (C? 0), and (C? 1) that (A? 3)
holds. (A? 2) is (C? 8). It remains to prove (A? 1).
We have seen that (A? 0) follows from (C? 0) and (D15), and from the three of
them together,

(C? 9) d(a, c) = d(a 4 c, ⊥).

By appealing to (D7), (C? 4), and (A? 0) we establish easily the second isotony prin-
ciple (E24),

(C? 10) a  b  c ⇒ d(a, c) ≥ d(b, c),

and by appealing to (C? 0), (C? 3), (C? 2), (D6), and (C? 4), we establish triangularity
as easily,

(C? 11) d(b, a) + d(b, c) = µ(a 4 b) + µ(b 4 c)

≥ µ((a 4 b) + (b 4 c) ≥ µ(a 4 c) = d(a, c).

Writing for χ in (D10) the term ϕ 4 ψ, and then applying (C? 10), (C? 9), and
(C? 11), we obtain

(C? 12) d(a 4 c, ⊥) ≥ d((ϕa 4 ψa) 4 (ϕc 4 ψc), ⊥),

(C 13) d(a 4 c, ⊥) ≥ d(ϕa 4 ψa, ϕc 4 ψc),
(C 14) d(a 4 c, ⊥) + d(ϕa 4 ψa, ⊥) ≥ d(ϕc 4 ψc, ⊥).

A further three-fold application of (C? 9) establishes our old principle of partial di-
versity (A? 1):

(C? 15) d(a, c) + d(ϕa, ψa) ≥ d(ϕc, ψc).

We turn now to the converse derivation, the derivation of the axioms of C? within
the system A? . Since (A? 3) is the same as (E6), and (A? 1) is the same as (E7), the

Principia 13(3): 339–56 (2009).

A Refined Geometry of Logic 353

metrical formulas (E6)–(E24) of § 2 can all be called on for this purpose. Axiom
(C? 1) is an immediate consequence of (A? 0) and (A? 3). Formula (E15) establishes
that d is non-negative, and by (A? 0) the same holds for µ, establishing (C? 2). It
follows from (E24) that if a  c then d(a, ⊥) ≥ d(c, ⊥), and hence by (A? 0) that
µ(a) ≥ µ(c); that is to say, µ is isotone, as stated in (C? 4). For (C? 0) and (C? 3)
rather more work is needed.
To establish (C? 0), we first prove within the system A? that d(a + c, a · c) =
d(a 4 c, ⊥). By the Lipschitz condition (E14), d(a 4 c, ⊥) ≥ d((a 4 c) + c, ⊥ + c),
and so by (D8), if a  c then d(a 4 c, ⊥) ≥ d(a, c). By (E14) again, and (D16),
d(a, c) ≥ d(a 4 c, c 4 c) = d(a 4 c, ⊥). Since a + c  a · c, these last two results
establish that d((a + c) 4 a · c, ⊥) = d(a + c, a · c). By (D4), we conclude that
d(a 4 c, ⊥) = d(a + c, a · c). It was shown in formula (A8) on p. 344 that the equality
of the diagonals of the lattice quadrangles is derivable from (A? 1) alone: d(a, c) =
d(a + c, a · c). It follows that d(a, c) = d(a 4 c, ⊥) and, by application of (A? 0), the
proof of (C? 0) is complete.
In any Brouwerian algebra, by (D1) and (D14), ((a + c) − c) · (c − ⊥) = (a − c) · c.
By (D2), (D14), and (D3), (a − a · c) · (a · c − ⊥) = (a − c) · (a · c) = (a − c) · c. In
other words, by (A? 4), κ(a + c, c, ⊥) = κ(a, a · c, ⊥). The chain a + c   c   ⊥ has
the same curvature at c as the chain c   a · c   ⊥ has at a · c, and if we apply (A? 2)
to each chain, and subtract, the curvature terms cancel out:

(A? 5) d(a + c, ⊥) − d(a, ⊥) = d(a + c, c) + d(c, ⊥) − d(a, a · c) − d(a · c, ⊥).

By (E22), we may infer that d(a+c, ⊥)− d(a, ⊥) = d(c, ⊥)− d(a·c, ⊥). The definition
(A? 0) may now be applied to each term, with the conclusion (C? 3): µ(a) + µ(c) =
µ(a + c) + µ(a · c).
The modified axiomatic systems A? and C? are thus interderivable. They are
interderivable also with the system B\? , which is obtained from B\ by modifying
(B\ 4) in the same way as (A2) was modified ((B\? 2) is indeed the same formula as
(A? 2)). The proofs of B\? from A? and of C? from B\? are minor variations on proofs
that have been given earlier. The details are omitted.

6. Metrization

A measure µ for which the system (C? ) holds is called strictly positive whenever
µ(b) = 0 implies that b = ⊥. That is, by (C? 0), the pseudometric d is a metric that
obeys (E9). Conversely, if d is a metric, the measure µ defined by (A? 0) is positive
isotone. The Brouwerian algebra L can be metrized in accordance with system A? if
and only if it admits a metric that satisfies the axioms of A? ; that is, if and only if it
admits a strictly positive measure µ.

Principia 13(3): 339–56 (2009).

354 David Miller

The algebras that concern us, the algebras of theories of a classical logical cal-
culus based on a denumerably infinite language, are all atomic, the atoms being
the maximal theories. These algebras either contain a theory that cannot be finitely
maximized (that is, extended to a maximal theory by adjoining a single proposition),
or contain no such theory. According to a theorem of Mostowski (1937), the alge-
bras in the first class, which are epitomized by any calculus vulnerable to Gödel’s
theorem, have 2ℵ0 atoms, while those in the second class have ℵ0 atoms. For details
of Mostowski’s results, and a relatively elementary proof, see Miller (1992).
Theorem 2.5 of Horn & Tarski (1948) states that an algebra T of theories that has
denumerably many atoms admits a strictly positive measure, and is thus metrizable
in accordance with the system A? . But a theory algebra with 2ℵ0 atoms admits no
such measure, and is not metrizable in accordance with the system A? .
It does not follow that there are algebras of classical theories that are not metriz-
able. In general, metrization demands less than measurability, as the following in-
tuitive remarks may help to make clear. Because distinct theories imply distinct
propositions, of which there are only denumerably all told, every path from > to ⊥
in an algebra of theories has at most ℵ0 links. A radially convex metric therefore
must divide the interval [>, ⊥] into at most ℵ0 disjoint pieces. But an algebra with
continuum many atoms is pear-shaped, very much wider at the bottom than it is tall,
and the total available measure on it must be divided into non-denumerably many
non-zero pieces.
Making use of results of Urysohn, Carruth, and Vietoris, Mormann (2006), Theo-
rem 5.4, has demonstrated that each algebra of theories T is metrizable by a radially
convex metric ∂ (a metric satisfying A2). We shall use Mormann’s theorem to show
that each algebra T of theories is metrizable in another way by a metric d that satis-
fies (A? 1) and for which, for every a, c ∈ T,
(A? 6) d(a, c) = ∂ (a 4 c, ⊥).
Since T may not be measurable, we cannot use the definition (A? 0) to define a
measure µ in terms of d (or in terms of ∂ either). It follows that (A? 2) must fail
for d. The system A? was designed specifically to be interderivable with C? , but the
design must now be partly unstitched.
The correction term κ in (A? 2) was calculated (p. 351) by applying (C? 3) and
(C 0) to (D7). If we are given a bounded function ν, not a measure, nothing need
stop us from recalculating the correction term to suit a different purpose. Let λ(a, c)
record the extent to which the function ν deviates from additivity, so that
(Cλ 3) ν(a) + ν(c) = ν(a + c) + ν(ac) + λ(a, c).
We can preserve the axiom scheme (A? 1) if we replace (A? 2) by
(Aλ 2) a  b  c ⇒ d(a, c) = d(a, b) + d(b, c) − κ(a, b, c) − λ(a − b, b − c).

Principia 13(3): 339–56 (2009).

A Refined Geometry of Logic 355

If the function λ can be expressed in terms of d, then of course it should be so

expressed. It follows that the theory Cλ , whose axioms are (C? 0), (C? 1), (C? 2),
(Cλ 3), implies the theory Aλ , whose axioms are (A? 0) (A? 1), (Aλ 2), and (A? 3).
The converse implication does not hold.
Mormann’s proof is non-constructive, and in view of the extraordinary variety
of theory algebras with 2ℵ0 atoms, it is unlikely that much can be said in general
about the metric ∂ . My interest here is only in the values ν(b) = ∂ (b, ⊥), which are
all positive and bounded by ∂ (>, ⊥). Thanks to (C? 0) and (A? 6), we can define in
terms of this function ν a metric d that, in almost the same way as before, may be
proved to satisfy the theory Aλ , and in particular axiom (A? 1).

THEOREM: Let T be a Brouwerian algebra of theories. If ∂ is a metric on T, then there

exists a metric d on T that satisfies axiom (A? 1) and for which d(a, c) = ∂ (a 4 c, ⊥)
for each a, c ∈ T.1

Birkhoff, G. 1967. Lattice Theory. 3rd edition. Providence RI: American Mathematical Society.
Horn, A. & Tarski, A. 1948. Measures in Boolean Algebras. Transactions of the American
Mathematical Society 64: 467–97.
Miller, D. W. 1977. On Distance from the Truth as a True Distance. (short version) Bulletin
of the Section of Logic, Institute of Philosophy & Sociology, Polish Academy of Sciences,
Wrocław 6(1): 15–26.
———. 1978. On Distance from the Truth as a True Distance. In K. J. J. Hintikka, I. M. O.
Niiniluoto, & E. Saarinen (eds.) Essays on Mathematical & Philosophical Logic, pp. 415–
435. Dordrecht: D. Reidel Publishing Company.
———. 1979. Metric Postulates for Modular, Distributive, and Boolean Lattices. Bulletin of
the Section of Logic, Institute of Philosophy & Sociology, Polish Academy of Sciences,
Wrocław 8(4): 191–6.
———. 1984. A Geometry of Logic. In H. J. Skala, S. Termini, & E. Trillas (eds.) Aspects of
Vagueness, pp. 91–104. Dordrecht: D. Reidel Publishing Company.
———. 1986. Continuous Connectives. Fisal–84 [Fall International Seminar on Applied Logic]
Proceedings, pp. 173–179. Palma de Mallorca: Govern Balear Conselleria d’Educació i Cul-
tura, Universitat de les Illes Balears.
———. 1992. The Disposition of Complete Theories. Coleção Documentos, Série Lógica e
Teoria da Ciência, 10, agosto de 1992, Instituto de Estudos Avançados, Universidade de
São Paulo. Available at
———. 1994. Critical Rationalism. A Restatement and Defence. Chicago & La Salle: Open
Court Publishing Company.
Mormann T. 2006. Truthlikeness for Theories on Countable Languages. In I. C. Jarvie,
K. M. Milford, & D. W. Miller (eds.) Karl Popper: A Centenary Assessment. Volume III:
Science, pp. 3–15. Aldershot & Burlington VT: Ashgate.

Principia 13(3): 339–56 (2009).

356 David Miller

Mostowski, A. 1937. Abzählbare Boolesche Körper und ihre Anwendung auf die allgemeine
Metamathematik. Fundamenta Mathematicae 29(1): 34–53. English translation 1979.
Foundational Studies. Selected Works, Volume II, pp. 1–17. Amsterdam: North-Holland
Publishing Company, and Warsaw: Polish Scientific Publishers.
Nordhaus, E. A. & Lapidus, L. 1954. Brouwerian Geometry. Canadian Journal of Mathematics
6: 217–229.
Tarski, A. 1935/1936. Grundzüge des Systemenkalkül. Fundamenta Mathematicae 25(4):
503–26, and 26(2): 283–301. English translation 1956. Logic, Semantics, Metamathemat-
ics, pp. 342–383. Oxford: Clarendon Press. Second edition 1983. Indianapolis: Hackett
Publishing Company.

Department of Philosophy
University of Warwick
Coventry CV4 7AL UK

Resumo. A fim de medir o grau de dessemelhança entre elementos de uma álgebra boole-
ana, o autor propôs em (1984) usar pseudométricas satisfazendo generalizações dos axiomas
usuais para a identidade. A proposta é estendida, na medida em que é exequível, de álgebras
booleanas (álgebras de proposições) para álgebras de Brouwer (álgebras de teorias deduti-
vas). A relação entre geometrias booleanas e de Brouwer da lógica resulta semelhante, de
maneira curiosa, à relação entre geometrias euclidianas e não-euclidianas do espaço físico.
O artigo conclui com uma breve consideração do problema da metrização da álgebra de

Palavras-chave: Álgebra booleana, álgebra de Brouwer, métrica, mensuração, identidade


Some of this material has been presented at the INTERNATIONAL CONFERENCE ON ALGEBRAIC
& TOPOLOGICAL METHODS IN NON-CLASSICAL LOGICS at Tbilisi State University, in 2003, and
at the XIV ENCONTRO BRASILEIRO DE LÓGICA at Itatiaia in 2006; and also at seminars at
the Department of Mathematics, University of Warwick, in 2005, at the Departamento de
Filosofía, Universidad Nacional de Córdoba, Argentina, and the Departamento de Matemáti-
cas, Universidad Nacional de Colombia, Bogotá, in 2006. It is a real pleasure to be able to
dedicate this written version to Newton da Costa on his 80th birthday. In preparing it, I have
been much encouraged by the interest shown by Thomas Mormann in the geometry of logic.
As always, responsibility for errors is reserved.

Principia 13(3): 339–56 (2009).

Universidad de Buenos Aires/Universidad Nacional de La Plata
Universidad Nacional de La Plata

Abstract. This work agrees and supports the I. Hacking’s thesis regarding the meaning
of the logical constants accordingly with Gentzen’s Introduction and Elimination Rules of
Sequent Calculus, corresponding with the abstract conception of the notion of logical conse-
quence. We would like to ask for the minimum rules that must satisfy a connective in order to
be considered as a genuine negation. Mainly, we will refer to both da Costa’s C-Systems and
Priest’s LP system. Finally, we will analyze the presentations of these systems within the Se-
quent Logic to show that paraconsistent negation lacks of pure rules of negation-elimination
and negation-introduction rules or that they involve other connectives, thus making difficult
to assign an univocal meaning to paraconsistent negation.

Keywords: Paraconsistent logic, negation, sequent calculus.

1. Introducción

En un artículo fundacional What is Logic? (1979), Ian Hacking, arroja luces decisi-
vas sobre el problema clásico en filosofía de la lógica tal como lo es el significado de
las constantes lógicas. Hacking comienza definiendo la lógica como transiciones de
oraciones a otras oraciones. Estas transiciones deben expresarse en el metalenguaje y
deben considerarse solamente descripciones o codificaciones de cómo nosotros cree-
mos que deben hacerse las mismas cuando deseamos decir que ellas son de carácter
lógico. Además, pueden entenderse como casos especiales de transformaciones y
pueden ser pensadas tanto semántica como sintácticamente y, de acuerdo con Witt-
genstein, nunca pueden calificarse de justificaciones. Para ello adopta el cálculo de
secuentes de Gentzen, en el cual las reglas estructurales caracterizan el concepto
abstracto de consecuencia lógica mientras que las reglas operatorias caracterizan el
uso o significado de las constantes lógicas en concordancia con las reglas estructu-
rales. A su vez, las reglas operatorias que caracterizan las constantes lógicas sólo
tienen sentido leídas en el marco de deducibilidad generado por las reglas estructu-
rales y sólo dentro de esta estructura de conjunto puede decirse que constituyen el
significado de las constantes lógicas. Esta tesis ha sido rigurosamente fundamenta-
da por R. Wójcicki (1988) quien sostiene que toda operación de consecuencia está
Principia 13(3): 357–70 (2009).
Published by NEL — Epistemology and Logic Research Group, Federal University of Santa Catarina (UFSC), Brazil.
358 Gladys Palau & Cecilia Duran

unívocamente determinada por el conjunto de reglas de inferencia consistentes con

ella y muestra que hay lógicas que coinciden en el conjunto de sus teoremas pero
no coinciden en el conjunto de sus reglas de inferencia, tal como es el caso de la
lógica intuicionista y de la lógica de la relevancia respecto de la lógica clásica.1 Es
precisamente dentro de este enfoque que analizaremos el significado de la negación
en la lógica paraconsistente ya que actualmente se acuerda que ella es una lógica
subclásica. Hoy en día se acepta en considerar que una lógica es paraconsistente si
y sólo si no satisface la regla conocida con el nombre de Ex Contradictione Sequitur
Γ, A, ¬A ⇒ B, ∆ (ECQ).

Es decir, los sistemas de lógica paraconsistente son aquellos que admiten con-
tradicciones pero no por ello originan sistemas triviales, lo cual implica distinguir
nítidamente inconsistencia de trivialidad. En palabras de W. A. Carnielli y J. Marcos
(2007: 11), una lógica es paraconsistente si es inconsistente pero no trivial. Coinci-
dimos con estos autores en que uno de los argumentos más fuertes a favor de la
importancia de estos sistemas sea la existencia efectiva de teorías inconsistentes pe-
ro no triviales, como por ejemplo, la teoría intuitiva de conjuntos.
En el enfoque de Hacking el significado de las conectivas lógicas, incluso el de
la negación, depende de las reglas admitidas o rechazadas que rigen el uso de cada
conectiva. Precisamente el Cálculo de Secuentes de Gentzen nombra las reglas que
rigen la introducción y la eliminación de las conectivas como Introducción en el pos-
tsecuente y en el prosecuente respectivamente. Por ello, dada una regla que no debe
quedar validada por la teoría, hay que explorar qué es lo que se debe alterar en el
cálculo de secuentes a fin de filtrarla. En el caso que nos ocupa, dado que el rechazo
de ECQ es común a todo sistema de lógica paraconsistente, todo sistema paraconsis-
tente deberá rechazar también las reglas que permiten derivar ECQ. En particular,
desde el marco de la lógica de secuentes, se deberían rechazar o modificar las reglas
estructurales o las operatorias que posibilitan la derivación de tal regla.
Si se explora en esta lógica qué reglas permiten la demostración de ECQ, se
encuentra que la demostración en lógica de secuentes de ECQ involucra Introduc-
ción de la Negación en el prosecuente (¬ ⇒), Introducción de la Conjunción en el
prosecuente (∧ ⇒) y Atenuación en el postsecuente (⇒ At).3 Por ello, se abren dos
caminos, o bien se pone en duda (⇒ At) o bien falla (¬ ⇒). Exploraremos ambas
La primera opción no resulta conveniente ya que si se descarta (⇒ At) se res-
tringe la potencia lógica de los sistemas paraconsistentes, puesto que esta regla es
necesaria para la demostración en lógica de secuentes del axioma del sistema Cω
de N. C. A. da Costa, D. Krause y O. Bueno (2007: 798), (A → C) → ((B → C) →
((A ∨ B) → C)) y para la derivación de la Ley de Dummett, (A ∨ (A → B)).4 Además,

Principia 13(3): 357–70 (2009).

El Significado de la Negación Paraconsistente 359

quitar (⇒ At) no es condición suficiente para evitar explosión, ya que ello no impide
la derivación de otra forma de explosión, como es el caso de la regla Γ, A, ¬A ⇒ ¬B, ∆
cuya derivación no emplea (⇒ At).
La segunda opción nos conduce a explorar qué sucede en una lógica paraconsis-
tente con la regla operatoria de Introducción de la negación en el prosecuente (¬ ⇒).
Si esta regla fallara en todas sus aplicaciones, la desviación de la lógica paraconsis-
tente respecto de la lógica clásica no estaría dada por las reglas estructurales sino por
alguna de las reglas operatorias que involucran la negación, lo cual condicionaría el
significado de esta conectiva en la mencionada lógica. Elucidar estas cuestiones y sus
consecuencias para la filosofía de la lógica es uno de nuestros objetivos principales.
Para nuestro trabajo hemos elegido como referencia los sistemas que considera-
mos más representativos de este tipo de lógica en la literatura actual, a saber, los
C-sistemas de N. C. A. da Costa et al. (2007), y el sistema LP de Graham Priest
(2001), ya que ellos emplean dos estrategias totalmente diferentes, la vía de una
lógica no veritativo funcional por parte de la escuela de da Costa y la vía de una
lógica multivaluada por parte de Priest.

2. La Negación en los C-sistemas de da Costa et al.

Es sabido que en todo sistema lógico paraconsistente, a los efectos de obtener sis-
temas no triviales pero que admitan contradicciones se debe satisfacer las reglas de
las conectivas positivas pero no debe satisfacerse la ya citada regla ECQ. Respecto
del Principio de No Contradicción la situación es distinta ya que si se desea que el
sistema sea una lógica dialéctica, como el sistema DL de da Costa-Wolf (1980) este
principio no puede ser válido, de donde se sigue que su rechazo no constituye una
condición necesaria para la paraconsistencia. Los sistemas paraconsistentes de da
Costa conforman una colección jerárquica Cn , 0 ≤ n ≤ ω construida a partir de los
axiomas positivos de la axiomática de Hilbert. Cω contiene la negación paraconsis-
tente más débil y por ello está caracterizado solamente por 11 axiomas, entre los
que se incluye el Modus ponens (MP) en tanto regla y dos axiomas específicos para
la negación: la doble negación no intuicionista ¬¬A → A (DN) y el principio clási-
co Tertium non datur A ∨ ¬A (TND). Como en todo C-sistema, además de ECQ, en
Cω tampoco puede ser válida la regla clásica Reductio ad absurdum (RAb), A → B,
A → ¬B ⇒ ¬A debida a Johanssen y Kolmorogov y tan exitosamente usada en la
matemática, porque si valiera, sería posible demostrar ¬(A ∧ ¬A) como teorema por
Sustitución Uniforme a partir de los dos primeros axiomas positivos de Cn con dos
aplicaciones de MP (Maranhão 2001: 13).5
Tampoco puede valer el principio de doble negación intuicionista A ⇒ ¬¬A, por-
que si valiera conjuntamente con el axioma ¬¬A ⇒ A, la negación paraconsistente

Principia 13(3): 357–70 (2009).

360 Gladys Palau & Cecilia Duran

colapsaría nuevamente en la negación clásica. La falla de la regla (RAb) hace caer to-
das las formas de la Transposición e incluso al Silogismo Disyuntivo (SD)6 y al Modus
tollens. Tampoco resultan teoremas ¬A → (A → B), ¬A → (A → ¬B), A → (¬A → B),
A → (¬A → ¬B), (A∧ ¬A) → B) y (A∧ ¬A) → ¬B) cuyas demostraciones requieren de
ECQ. Dado que en lógica de secuentes todas estas reglas utilizan en su demostración
la regla Introducción de la negación en el prosecuente (¬⇒), se podría entonces supo-
ner que en todo sistema paraconsistente se debería descartar esta regla. Sin embargo
esta alternativa plantea los siguientes dos problemas (i) si se aplica Sustitución Uni-
forme en el primer axioma positivo de Cω , i.e., ⇒ A → (B → A), entonces resultan
teoremas los casos de Causa mirabilis ¬A → (A → ¬A) y A → (¬A → A), cuyas demos-
traciones en lógica de secuentes involucran la regla (¬ ⇒) y (ii) la demostración en
lógica de secuentes del axioma de Cω , ¬¬A → A requiere tanto de Introducción de la
negación en el postsecuente (⇒ ¬) como de la regla de Introducción en el prosecuente
(¬⇒). Sin embargo se debe tener en cuenta que si se descarta la regla (¬⇒) enton-
ces todas las formas de explosión quedan obstruidas, lo cual muestra la utilidad de
su rechazo. Por otra parte, es posible resolver los inconvenientes planteados en (i)
y (ii) prescindiendo de (¬ ⇒). En efecto, los casos de Causa mirabilis planteados en
(i) quedan validados en los C-sistemas si se admite que las fórmulas fundamentales
o axiomas de los que parte toda derivación, puedan ser literales (i.e., fórmulas ató-
micas o fórmulas atómicas negadas) del tipo ¬A ⇒ ¬A7 y además se acepte atenuar
con literales. En relación con el inconveniente (ii) i.e., el referido a la Eliminación
de Doble Negación, Andrés Raggio (1968) realiza un análisis de la negación en los
sistemas Cω desde la lógica de Secuentes y propone caracterizar la negación para-
consistente de Cω por la siguiente regla de Introducción de la doble negación en el
prosecuente (¬¬⇒) la cual se conserva en todos los C-sistemas, a saber:
A, Γ ⇒ Θ
¬¬A, Γ ⇒ Θ
En el cálculo de Raggio para Cω , además de la regla de la doble negación se efec-
túan dos modificaciones más: (i) se elimina la regla (¬ ⇒) de la lógica de secuentes
de Gentzen, dejando solamente la mencionada (¬¬ ⇒) y (ii) se modifica la regla
(⇒→) de Gentzen mediante una restricción según la cual en el postsecuente de la
premisa sólo puede aparecer la fórmula que es el consecuente del condicional que a
su vez es conclusión de la regla. O sea, la regla (⇒→)R :
A, Γ ⇒ B
Γ⇒A→ B
Esta modificación filtra perfectamente tanto la ley de Dummet como la ley de
Peirce que no son teoremas de Cω dado que éste fue construido sobre el fragmen-
to positivo del sistema de Hilbert, según lo afirman sus propios autores. Además,

Principia 13(3): 357–70 (2009).

El Significado de la Negación Paraconsistente 361

(⇒→)R no cumple con toda la restricción intuicionista (i.e. que el postsecuente no

pueda tener más de una fórmula) ya que si lo hiciera quedaría comprometida la
validez de los axiomas para la negación de Cω .
Corresponde preguntarnos ahora qué resultados se siguen de nuestro análisis a
los efectos de caracterizar la negación en Cω . Ya hemos afirmado que en Cω no se
valida ECQ pero sí DN y TND. Sin embargo, de estos axiomas no se sigue ninguna
otra propiedad interesante respecto de la negación en Cω . Por ello, ahí que lo único
que parece caracterizar positivamente un sistema tan débil respecto de la negación
como Cω es que Cω no es finitamente trivializable (da Costa et al. 2007: 809). Sin
embargo, a partir de la enumeración ya dada de las reglas que no se satisfacen en Cω ,
es posible aventurar una caracterización negativa de la negación. Otra posibilidad
consiste en encontrar reglas mínimas o puras con las que se pueda determinar el
significado de la negación en Cω . La primera alternativa nos enfrenta al problema
lógico filosófico de si es posible inferir lo que cierta cosa es partiendo solamente
de lo que ella no es, dado que en una metateoría clásica sigue valiendo el dictum
aristotélico de que no hay inferencias válidas con todas sus premisas negativas.
Una caracterización precisa de la negación en Cω debería también tener en cuen-
ta ciertas peculiaridades. Cω carece de una negación fuerte, es decir no es definible
como ¬(n) A =df ¬A ∧ A(n) . Además, si bien Cω no es finitamente trivializable, es
posible definir una negación a partir del fragmento implicativo de Cω siempre que
el conjunto de fórmulas de Cω se construya con elementos de un conjunto finito
de variables proposicionales8 (Maranhão 2001: 29). Resulta sencillo advertir que
al carecer de la regla (¬ ⇒) no se valida Introducción de de la Doble Negación en
el postsecuente (⇒ ¬¬) ni tampoco es posible validar teoremas negados o probar
equivalencia de fórmulas negadas.
La escuela de da Costa y él mismo entendieron la debilidad del sistema ω desde
un principio y por ello, para construir la jerarquía de los C-sistemas, en particular C1 ,
adicionan un nuevo operador denominado estabilizador para “estabilizar” fórmulas
y simbolizado por ◦ de forma tal que A◦ =df ¬(A ∧ ¬A) y dos nuevos axiomas, uno
que garantiza que las fórmulas compuestas por fórmulas estabilizadas también “se
portan bien”, llamado Propagación del Buen Comportamiento A◦ ∧ B ◦ → (A◦ ∧ B ◦ ) ∧
(A◦ ∨B ◦ )∧(A◦ → B ◦ ) y otro que dota a la negación de nuevas propiedades B ◦ → (A →
B) → ((A → ¬B) → ¬A)). Este último axioma mediante el converso del Metateorema
de a Deducción, se traduce en la regla conocida como Absurdo Paraconsistente (PAb).
Dado que la invalidez de (RAb) en Cω precisamente impedía todas las formas
de explosión, (PAb) debería conservar esa propiedad. Efectivamente, (PAb) expresa
la explosión frente a fórmulas que respetan el Principio de No Contradicción pero
no para fórmulas cualesquiera. Así, la negación de una fórmula estabilizada es una
negación fuerte en el sentido de que es explosiva. No es necesario discriminar la
graduación de propiedades de la negación en los diversos sistemas Cn , pero sí se

Principia 13(3): 357–70 (2009).

362 Gladys Palau & Cecilia Duran

requiere aclarar que An abrevia fórmulas con un grado de complejidad creciente,

siendo que ello hace variar el tipo de fórmula que trivializa esos sistemas. O sea, a
medida que incrementa n en Cn tal que 0 ≤ n ≤ ω la lógica se debilita, C0 es la lógica
clásica y C1 el sistema más potente dentro de la jerarquía de cálculos paraconsisten-
tes. Analizaremos pues la negación en C1 a fin de completar lo expresado respecto
de ella. Al adicionar (PAb) a C1 , se obtienen las reglas clásicas para la negación pero
a condición de que queden estabilizadas las fórmulas que producen explosión. Sin
embargo en C1 sigue sin ser válido el Silogismo disyuntivo y, a menos que A esté
estabilizada tampoco es teoremas la fórmula (A ∨ B → (¬A → B). Más aún, también
se satisface en C1 la regla según la cual, A◦ , A, ¬A ⇒ B. Debemos preguntarnos en-
tonces qué reglas en secuentes quedan involucradas en la demostración de (PAb).
La solución nos lleva a reintroducir en el cálculo de secuentes la negación en el pro-
secuente que fuera cancelada para Cω . Pero, si se introduce la regla de Gentzen,
la negación paraconsistente colapsaría con la clásica. Para evitar esto es suficiente
emplear la regla (◦ ¬⇒) (Benassi et al. 2007).

A◦ , Γ ⇒ Θ, A (◦ ¬⇒)
A◦ , ¬A, Γ ⇒ Θ

Aplicando (◦ ¬ ⇒) es derivable (PAb) en el Cálculo de Secuentes.9 Nótese que

esta modificación no reintroduce Pseudo Scotus ni Explosión puesto que A está esta-
bilizada. Además, es posible redefinir la regla para cada uno de los Cn en función
de la complejidad de las fórmulas para las que vale ◦ . Para completar el cálculo de
secuentes se deberían construir reglas para la introducción del estabilizador, bajo la
guía de la definición de A◦ como ¬(A∧¬A) de modo que se cumpla ⇒ A◦ → ¬(A∧¬A)
y ⇒ ¬(A ∧ ¬A) → A◦ .
Agregando la definición de la negación fuerte, i.e., ¬∗ A =df ¬A ∧ A◦ se completa
el sistema C1 . Esta negación permite demostrar en C1 la fórmula ((A → B) → ((A →
¬∗ B) → ¬∗ A)). Sin embargo, en C1 el teorema 20 afirma que el esquema A ↔ (A →
B), donde B es una fórmula cualquiera, trivializa a la teoría, por lo cual la regla
A ↔ B ⇒ ¬A ↔ ¬B no puede ser válida en C1 . Por ello, la negación en C1 no
colapsa en la clásica, sino que evidencia solamente fallas en la simetría. De ahí que,
la negación paraconsistente de C1 impide que se satisfaga el principio de sustitución
de modo que no siempre será válido el Principio de Reemplazo de Equivalentes, puesto
que la negación de fórmulas equivalentes no siempre es equivalente, al igual que en
Cω . Esto equivale en lógica de secuentes a modificar (¬⇒).
La mayor diferencia entre Cω y C1 reside en que Cω no es finitamente trivializable
(da Costa et al. 2007 p.809) mientras que C1 es finitamente trivializable. De hecho,
la fórmula A ∧ ¬∗ A trivializa C1 ya que siendo ¬∗ clásica, entonces se cumple ⇒
(A ∧ ¬∗ A) → B (da Costa et al. 2007: 806). Respecto de la negación fuerte se validan

Principia 13(3): 357–70 (2009).

El Significado de la Negación Paraconsistente 363

todos los teoremas clásicos, es decir siempre que las fórmulas que pudieran generar
explosión estén estabilizadas. 10

3. La negación en el sistema LP de G. Priest

Pasemos ahora a analizar la posición realista de G. Priest, llamada por él mismo

dialeteismo y expresada esencialmente en su sistema LP y cuyo supuesto filosófico
principal consiste en sostener que el mundo real es inconsistente ya que en él existen
dialetheias, i.e., proposiciones que son verdaderas y falsas al mismo tiempo las cuales
constituyen contradicciones genuinas.
El sistema LP de G. Priest (2001) es un sistema trivalente, ya que contiene un
tercer valor de verdad “i”. Respecto de las interpretaciones que ha recibido este ter-
cer valor en los sistemas trivalentes, el mismo Priest las agrupa en dos clases. Por un
lado, están aquellos sistemas que interpretan este valor como “ni verdadero ni falso”,
como por ejemplo los sistemas que intentan dar cuenta de los futuros contingentes o
de los enunciados que contienen propiedades vagas o conceptos que no denotan. Por
el otro lado, están los sistemas que sostienen que hay proposiciones que pueden ser
verdaderas y falsas al mismo tiempo. Priest da como ejemplos de estas últimas a las
proposiciones “contradictorias” de la tradición hegeliano- marxista, las paradojas de
la autorreferencia y ciertas afirmaciones sobre micro objetos de la mecánica cuánti-
ca. Precisamente para las conectivas de estas proposiciones Priest propone matrices
con un tercer valor, el cual ya no puede ser representado por un lugar vacío, sino
por un “atasco” (“glut”), y que se encuentra mejor representado por el valor “01”,
que significa que las proposiciones atómicas pueden tener el valor verdadero y falso
simultáneamente. Los restantes valores tienen el sentido habitual, es decir que el va-
lor 1 debe leerse como verdadero y el valor 0 como falso. Lo más significativo de su
sistema está dado por la matriz para la negación ya que ella apareja consecuencias
radicales. Su sistema de lógica de las paradojas LP, es el mismo que el sistema K3
de Kleene pero tiene la particularidad de que el conjunto de valores distinguidos es
{1, 01}. De esta forma el significado de la negación de LP queda determinado por la
siguiente matriz:

A ¬A
0 1
01 10
1 0

La matriz característica de LP para el conjunto de las conectivas proposicionales

positivas,11 conjuntamente con el considerar el valor 01 como valor distinguido hace
que la regla ECQ no resulte válida y esto se comprueba cuando en la valuación de la

Principia 13(3): 357–70 (2009).

364 Gladys Palau & Cecilia Duran

premisa A ∧ ¬A se toma el valor 1 de la valuación 01 y el valor 0 de la conclusión B.

Si bien este resultado es esperado, el sistema produce otros resultados, algunos de
ellos bastante extraños.
(i) El Modus Ponens y la Transitividad (TR) no son inferencias válidas. En efecto
MP resulta inválido cuando V (A) = 01 y V (B) = 0 y TR = 0 cuando V (A) = 1,
V (C) = 0 y V (B) = 01.
(ii) La contraposición débil (A → B) → (¬B → ¬A) resultan válida porque su tabla
ostenta solamente apariciones de los valores 1 y 01, y en LP, 01 es un valor
(iii) El principio Tercero Excluido y el Principio de No Contradicción resultan váli-
dos, porque su matriz final contiene solo un caso 01 y este es un valor distin-
(iv) La doble negación A ↔ ¬¬A. es válida porque su matriz final contiene un caso
01 y éste es valor distinguido
(v) Lo “paradójico” radica en que su sistema permite contradicciones verdaderas.
Es decir A ∧ ¬A, es verdadera ya que si V (A) = 01 y V (¬A) = 01, entonces,
tomando en cuenta el 1 del valor 01, A ∧ ¬A resulta con valor 1 y, por lo tanto
¡hay contradicciones verdaderas!
Así, la base deductiva de LP queda establecida por las siguientes reglas:12
1) B ⇒ (A → B)
2) (A → B) ∧ (C → D) ⇒ (A → D) ∨ (C → B)
3) A → C ⇒ (A ∧ B) → C
4) (A ∧ B) → C ⇒ (A → C) ∨ (B → C)
5) ¬(A → B) ⇒ A
6) ¬A ⇒ (A → B)
7) A → B ⇒ ¬B → ¬A
8) ¬¬A ⇒ A
9) A ⇒ ¬¬A
10) (A → B) ∧ (A → ¬B) ⇒ ¬A (RAb)
Sin embargo se produce el hecho curioso de que son verdades lógicas (teoremas)
pero no inferencias válidas MP, ECQ y el SD.
⇒ ((A → B) ∧ A) → B)
⇒ (A ∧ ¬A) → B
⇒ ((A ∨ B) ∧ A) → B
Al igual que otras lógicas subclásicas, en LP el conjunto de las reglas de inferen-
cia o inferencias válidas es menor que el conjunto de verdades lógicas (o teoremas).

Principia 13(3): 357–70 (2009).

El Significado de la Negación Paraconsistente 365

Ahora bien, varias de las reglas de LP requieren de (⇒At), (¬ ⇒) y (⇒ ¬), simi-

larmente a lo que sucedía en varios C-sistemas, en particular con C1 . Además, la
validez de la transposición subminimal necesariamente conduce a aceptar en deduc-
ción natural la regla RAb y en lógica de secuentes la regla (¬ ⇒) rechazada en los
C-sistemas. Sin embargo, la diferencia fundamental de LP con estos sistemas radica
en la falla en LP del MP como regla. Si ahora nos preguntamos qué implica en una
deducción no contar con MP como regla de inferencia, estamos en condiciones de
responder que se pierde la regla de Eliminación del Condicional en Deducción Na-
tural y consecuentemente con la falla de regla de Introducción del Condicional en el
prosecuente (→⇒) en el cálculo de secuentes, pero preservándose ambas reglas de
atenuación. Así, es posible afirmar que la caracterización del condicional en tanto
regla queda “manca”, lo cual a su vez conlleva a afirmar que el condicional de LP no
es el condicional material clásico, sino otro tipo de condicional.
En un trabajo anterior C. Oller y G. Palau (2008) propusieron un sistema de
secuentes LP (SLP) sin reglas estructurales En él la negación de SLP queda caracte-
rizada por dos de los siguientes tres axiomas:
Ax.1 Γ, A ⇒ ∆, A,
Ax.2 Γ, ¬A ⇒ ∆, ¬A,
Ax.3 ⇒ ∆, A, ¬A.
Los dos primeros axiomas garantizan la propiedad de Identidad mientras que el
tercer axioma compensa la ausencia de las reglas estándares de la negación permi-
tiendo la prueba de TND mediante la regla operatoria de ⇒ ∨. Dado que LP carece
de MP y que A → B es entendida como ¬A ∨ B, la regla operatoria (⇒→) de SLP es
sustituida por:

Γ ⇒ ∆, ¬A, B
Γ ⇒ A → B, ∆

Las reglas en SLP que contienen solamente la negación son:

(¬¬ ⇒) A, Γ ⇒ ∆ (⇒ ¬¬) Γ ⇒ ∆, A
¬¬A, Γ ⇒ ∆ Γ ⇒ ∆, ¬¬A

Similarmente a las reglas que A. Avron (2005) enumera como reglas comunes a
la negación de varias lógicas subclásicas, en este sistema se emplean reglas para in-
troducir la negación en combinación con otras conectivas, a los efectos de bloquear
la derivación en SLP de la regla ECQ y de MP ya que ninguna de ellas permite el
pasaje de una fórmula del prosecuente al postsecuente ni del postsecuente al prose-
cuente. Dado que LP carece de MP, en particular nos interesa destacar las referidas
al condicional, ya que ellas se presentan usando la negación:13

Principia 13(3): 357–70 (2009).

366 Gladys Palau & Cecilia Duran

(¬ →⇒) Γ, A, ¬B ⇒ ∆ (⇒ ¬ →) Γ ⇒ ∆, A Γ ⇒ ∆, ¬B
Γ, ¬(A ⊃ B) ⇒ ∆ Γ ⇒ ∆, ¬(A ⊃ B)

Al igual que las reglas presentadas por Avron (2005) la formulación de estas
reglas rompe drásticamente la tesis filosófica que hemos propuesto al comienzo de
este trabajo, a saber, que el significado de cada conectiva lógica está dado por las
reglas de eliminación e introducción correspondientes, las cuales a su vez definen la
noción de consecuencia lógica del sistema lógico en cuestión. En síntesis, toda lógica
queda determinada por la base deductiva formada por el conjunto de reglas que se
satisfacen. En la literatura actual esta tesis ha dado lugar a la propiedad conocida
como separabilidad de las conectivas lógicas tanto respecto de sus condiciones de
verdad como de las reglas que determinan su uso. Se dice que una base deductiva
Q es separable si y sólo si ella permite caracterizar cada conectiva del lenguaje en
forma independiente de las restantes. Más precisamente una base deductiva Q es
separable si (i) ninguna regla de Q involucra más de una conectiva y (ii) para cual-
quier conjunto de conectivas Ci de Q, las consecuencias lógicas de Ci son iguales al
conjunto de las consecuencias lógicas de Q restringidas a las conectivas de Ci (Wój-
cicki 1998: 183). La base deductiva de la lógica clásica es en efecto separable, tal
como queda mostrado tanto en la presentación axiomática de Hilbert, como en los
sistemas de deducción natural y de secuentes de Gentzen. Por el contrario, la base
deductiva propuesta para LP adolece de ciertas dificultades: 1) No tiene reglas para
la negación sino que en su lugar figuran las de Introducción y Eliminación de la Do-
ble Negación y 2) las reglas del condicional (y de las restantes conectivas) involucran
para su eliminación o introducción de la negación. Ergo, la base deductiva de LP, no
es separable y por lo tanto el significado de sus conectivas no queda unívocamente
determinado. Sin embargo, dado que este sistema y en general los sistemas para-
consistentes han sido construidos para dar cuenta de la lógica de dominios vagos o
difusos o de los que admiten cierto tipo de contradicciones, que las conectivas lógicas
de tales sistemas no tengan la propiedad de separabilidad, no resulta un resultado

4. Comentarios finales

Ahora bien, hasta aquí, hemos analizado las propiedades que necesariamente no
debe cumplir un operador de negación en un sistema lógico que pretenda tener en
cuenta el carácter dialéctico de la realidad. Sin embargo, las leyes sobre la negación
rechazadas tanto en los cálculos C1 , . . . , Cn , Cω como en el sistema LP de Priest han
generado una negación más débil que la negación clásica, lo cual ha derivado en
que muchos lógicos se pregunten si constituyen realmente una negación y nos ha
conducido a preguntarnos qué propiedades debe cumplir un operador de negación

Principia 13(3): 357–70 (2009).

El Significado de la Negación Paraconsistente 367

para ser considerado una genuina negación. Cuestión esta que pasaremos ahora a
En el libro tradicional sobre lógicas multivaluadas de N. Rescher, éste propone
una condición semántica para que un operador sea realmente una negación. Tal
requisito consiste en prohibir que A y ¬A tengan el mismo valor distinguido. Obvia-
mente, este requisito no es satisfecho por la negación del sistema LP de Priest ya
que, como se mostró anteriormente, A y ¬A ambas pueden tener el valor 01. Por
su parte, con la negación del sistema paraconsistente C1 de da Costa et al., aunque
ésta no sean veritativo funcional, si la V (A) = 0 entonces V (¬A) = 1 (da Costa et al.
2007: 821) y por ello este requisito se cumple parcialmente. En Cω sólo se satisface
que si V (¬¬A) = 1 entonces V (A) = 1 lo cual determina que el requisito se cumpla
pero respecto de la doble negación. En otras palabras por una u otra razón en los
sistemas estudiados A y ¬A no aparecen distinguibles respecto de la verdad en un
sentido clásico.
Creemos oportuno preguntarnos entonces a manera de conclusión al igual que
Wolfang Lenzen en el trabajo titulado Necessary conditions for negation operators
(with particular applications to paraconsistent negation) cuáles son las propiedades
esenciales o indispensables para cualquier operador de negación Según este autor,
debe cumplir con el siguiente principio:
CP: Si A es al menos tan verdadero como B, i.e. si A implica lógicamente a B,
entonces conversamente B es a lo sumo tan falso como A, i.e. la falsedad de B
implica lógicamente la falsedad de A. En símbolos: Si A → B entonces ¬B →
Pedir como condición indispensable (o suficiente) la contraposición parece bas-
tante sensato ya que éste tiene su origen en el esquema de argumento refutatorio
empleado por Platón en sus diálogos. Tiene varias formulaciones y la mencionada
corresponde a la versión característica de la negación minimal. Además, esta versión
es considerada la más débil, ya que su demostración sólo requiere el uso de la regla
de Reducción al Absurdo que no implica la doble negación clásica. Ya hemos dicho
que la contraposición minimal resulta inválida en el sistema Cω , y sí es válida en C1 ,
pero recordando que este resultado se basa en el supuesto de de que la fórmula B
esté estabilizada. El sistema dado LP de Priest cumple con este principio, dado que,
(p → q) → (¬q → ¬p) nunca toma el valor 0 y toma el valor 01 en los casos en los
que V (p) = 1 y V (q) = 01, V (p) = 01 y V (q) = 0 y cuando V (p) = V (q) = 01. Pero,
como el valor 01 es distinguido, entonces el requisito de contraposición es válido y,
aún cuando no cumpla con la natural exigencia de Rescher, la negación del siste-
ma LP de Priest aparece como constituyendo una negación genuina, pese a que no
posibilite la derivación de ECQ pero sí permita la validez del principio de No Contra-
dicción. Finalmente, dado que tanto C1 como LP cuentan con la validez de la forma

Principia 13(3): 357–70 (2009).

368 Gladys Palau & Cecilia Duran

de Reducción ad absurdum mencionada, es posible concluir que la negación de los

sistemas C1 y LP, aunque no sea la clásica y su formulación en lógica de secuentes
tenga los inconvenientes y restricciones señaladas, constituye una genuina negación.

Avron, A. 1999. Negation: Two Points of View. In D .Gabbay & H. Wansing (eds.) What is
Negation? Dordrecht: Kluwer Academic Publishers, p. 3–21.
———. 2003. Classical Gentzen-type Methods in Propositional Many-Valued Logics. En M.
Fitting & E. Orlowska (eds.) Beyond Two: Theory and Application of Multiple-valued Logic.
Heidelberg, New York: Physica-Verlag, p. 117–55.
———. 2005. A Non deterministic View on Non-classical Negation. Studia Logica 80: 159–
Benassi, C. & Gentilini, P. 2007. Paraconsistent Provability Logic and Rational Epistemic
Agents. En J.-Y. Béziau, W. Carnielli, and D. Gabbay (eds.) Handbook of Paraconsistency.
Studies in Logic, vol. 9. London: King’s College Publications, p. 215–57.
Béziau, J.-Y. 2000. What is paraconsistent logic? Frontiers of Paraconsistent Logic Research
Studies Press, p. 95-112.
———. 2005. The future of paraconsistent logic. Publicación electrónica:
———. 2007. Adventures in the Paraconsistent Jungle. En J.-Y. Béziau, W. Carnielli, and
D. Gabbay (eds.) Handbook of Paraconsistency. Studies in Logic, vol. 9. London: King’s
College Publications, p. 63–80.
Carnielli, W. A. & Marcos, J. 2001. A Taxonomy of C-systems.
Carnielli, W. A.; Coniglio, M.; Marcos, J. 2007. Logics of Formal Inconsistency. In D. Gabbay
and F. Guenthner (eds.) Handbook of Philosophical Logic, 2nd. ed., vol. 4, p.1–93.
da Costa, N. C. A. & Wolf, R. 1980. Studies in Paraconsistent Logic I: Dialectical Principle of
the Unity of Opposites. Philosophia (Philosophical Quarterly of Israel) IX(2): 189–217.
da Costa, N. C. A.; Krause, D.; Bueno, O. 2007. Paraconsistent Logics and Paraconsistency.
In D. M. Gabbay, P. Thagard, and Woods J. (eds.) Handbook of the Philosophy of Science.
Philosophy of Logic Volume, Dale Jaquette (ed.) Amsterdam: Elsevier, p. 791–911.
Duran, C. & Palau, G. 2008. Ley de Peirce, negación y lógica paraconsistente. In C. Duran
and A. Hebrard (eds.) Actas de las VI Jornadas de Investigación en Filosofía, vol. II. La
Plata, Argentina: Ediciones al Margen, p.207–14.
Gabbay, D. M.; Thagard P.; Woods, J. (eds.) 2007. Handbook of the Philosophy of Science.
Philosophy of Logic. Amsterdam: North-Holland, Elsevier.
Gabbay, D. & Gunthner, F. (eds.) 2007. Handbook of the Philosophical Logic, vol. 14. Nether-
lands: Kluwer Academic Publishers.
Hacking, I. 1994. What is Logic? In D. Gabbay (ed.) What is a logical System? Oxford: Claren-
don Press, p. 1–33. Publicado por primera vez en The Journal of Philosophy 6: 285–318.
Hunter, A. 1996. Paraconsistent Logics. In J.-Y. Beziau and W. A. Carnielli (eds.) Paraconsis-
tency with No Frontiers. Studies in Logic and Practical Reasoning vol.4. Amsterdam: North
Holland-Elsevier, p. 301–11.

Principia 13(3): 357–70 (2009).

El Significado de la Negación Paraconsistente 369

Lenzen, W. 1998. Necesary conditions for negation operators (with particular applications
to paraconsistent negation). In D. Gabbay et al. (eds.) Handbook of Defeasible Reasoning
and Uncertainty Management, vol. 2, Reasoning with Potential Contradictions. Dordrecht:
Kluwer Academic Publishers, p. 211–35.
Maranhão, J. S. 2001. Some propositional paraconsistent logics. Electronic publication:˜logik/logikalt/max/nkl2/ks16-10.pdf
Marcos, J. 2003. On negation: Pure local rules. Electronic publication:
Oller, C. & Palau, G. 2008. A Sequent System for LP. CLE-e-Prints 8(6).
Priest, G. 1999. What not? A defence of Dialetheic Theory of Negation. In D. Gabbay and H.
Wansing (eds.) What is Negation? Dordrecht: Kluwer Academic Publishers, p. 101–20.
Priest, G. 2001. An Introduction to Non- Classical Logic. New York: Cambridge University
Raggio, A. 1968. Propositional sequence-calculi for inconsistent sytems. In A. Raggio (2002),
Andrés Raggio. Escritos Completos (1927-1991), Moreno, A. & Doffi, M. (eds.) Buenos
Aires: EUDEBA, p. 179–91.
Rescher, N. 1969. Many-valued logics. New York: McGraw-Hill Company.
Wójcicki, R. 1988. Theory of Logical Calculi, Basic Theory of Consequence Operations. Dordre-
cht: Kluwer Academic Publishers.

Departamento de Filosofía
Facultad de Humanidades y Ciencias de la Educación (FaHCE)
Universidad Nacional de La Plata (UNLP)
Universidad de Buenos Aires

Departamento de Filosofía
Facultad de Humanidades y Ciencias de la Educación (FaHCE)
Universidad Nacional de La Plata (UNLP)

Resumo. Neste trabalho concorda-se com a tese de I. Hacking segundo a qual o significado
das constantes lógicas é dado pelas Regras de Introdução e Eliminação do cálculo de sequen-
tes de Gentzen que caracterizam a concepção da noção de consequência lógica abstrata.
Perguntamos quais são as regras mínimas que um conectivo deve satisfazer para que seja
considerado uma negação genuína. Tomaremos como referência para tratar dessa questão
os C-sistemas de Newton da Costa e o sistema LP de Graham Priest. Finalmente, analisare-
mos esses sistemas na lógica de sequentes a fim de mostrar que a negação paraconsistente
ou bem carece das regras puras de eliminação e negação da negação ou ela envolve outros
conectivos, o que torna difícil atribuir um significado unívoco à negação paraconsistente.
Palavras-chave: Lógica paraconsistente, negação, cálculo de sequentes.

Principia 13(3): 357–70 (2009).

370 Gladys Palau & Cecilia Duran

La demostración primera se debe en realidad a Gödel, quien demuestra que todas las le-
yes clásicas expresadas en términos de la conjunción y negación son leyes intuicionistas y
la segunda a Anderson y Belnap (1975) según quienes las leyes clásicas expresadas con
conjunción, disyunción y negación son todas leyes relevantistas.
En la literatura sobre el tema también se menciona que la lógica paraconsistente no cumple
Γ, A, ¬A ⇒ ∆, denominada Explosión (o Pseudo Scotus), pero en el texto nos referiremos
siempre a ECQ ya que ellas no se distinguen en una lógica de conclusión única.
Hay otra, de origen medieval, que en deducción Natural utiliza Adición y Silogismo Dis-
yuntivo y que su demostración en secuentes involucra también la regla (¬ ⇒).
En Carnielli y Marcos (2001) la Ley de Dummett, se agrega al conjunto de axiomas de Cω
para obtener Cmin (considerado por ellos el más débil de la jerarquía de los C- sistemas).
Otra razón para rechazar Reductio ad absurdum reside en que esta regla permite derivar
reglas de la negación clásica que harían colapsar la negación paraconsistente con la clásica.
El rechazo de SD conduce a un resultado paradójico ya que en los C-sistemas no es posible
derivar la fórmula B por SD, pero sí puede derivarse a partir de ((A → B) ∧ A) por MP, lo cual
implica el rechazo de la definición del condicional en términos de disyunción y negación.
Como se señala en Benassi y Gentilini (2007: 222), si esto no fuera posible, tampoco podría
demostrarse ¬A ⇒ ¬A a partir de A ⇒ A. También se requiere de este recurso para derivar
(A → ¬A) → ¬A.
Dicha negación permite definir una trivialización parcial (o positiva).
También se puede optar por agregarla a las ya dadas como regla derivada:
B ◦ , Γ1 , A → B ⇒ ∆1 B ◦ , Γ2 , A → ¬B ⇒ ∆2
Γ1 , Γ2 ⇒ ¬A, ∆1 , ∆2
Otros análisis (Carnielli et al. 2007) se han dedicado al detallado análisis de los sistemas
de la Lógica de la Inconsistencia Formal determinando que ya no es posible identificar a los
C-sistemas con los sistemas Cn de da Costa sino que abarcan un vastísimo campo de sistemas
paraconsistentes dentro de los cuales es posible expresar consistencia.
La matriz de Priest para las restantes conectivas es la siguiente:
A∧ B A∨ B A→ B
LP A@ @ 1 01 0 1 01 0 1 01 0
1 1 01 0 1 1 1 1 01 0
01 01 01 0 1 01 01 1 01 01
0 0 0 0 1 01 0 1 1 1
Si se quiere la validez MP como regla de inferencia, hay que cambiar la matriz del condi-
cional y se obtendría el sistema RM3.
Las reglas propuestas para las restantes conectivas son:
(¬∧ ⇒) Γ, ¬A ⇒ ∆ Γ, ¬B ⇒ ∆ (⇒ ¬∧) Γ ⇒ ∆, ¬A, ¬B
Γ, ¬(A ∧ B) ⇒ ∆ ⇒ ∆, ¬(A ∧ B)
(¬∨ ⇒) Γ, ¬A, ¬B ⇒ ∆ (⇒ ¬∨) Γ ⇒ ∆, A Γ ⇒ ∆, B
Γ, ¬(A ∨ B) ⇒ ∆ Γ ⇒ ∆, ¬(A ∨ B)

Principia 13(3): 357–70 (2009).

Universidade do Estado do Rio de Janeiro/CNPq

Abstract. This article defends the hypothesis that Ernst Mach (1838-1916) employed nat-
uralism (a word he never used) — an attitude which has its origin in Mach’s acceptance
of Darwinism — as a weapon against metaphysics, since the former would try to establish
evalutions criteria, including those necessary to choose scientific theories, capable of being
used by human beings.

Keywords: Mach, naturalism, metaphysics, history of science.

Quem conhece a evolução completa do desenvolvimento

científico, julgará o significado de um [certo] movimento
científico atual de uma maneira mais independente e
exata do que aquele que, por ter o seu juízo limitado ao
período de tempo em que vive, somente poderá basear-se
na direção momentânea que tomou tal movimento.
E. Mach, História de la Mecánica, p. 18

Uma das mais importantes “verdades” da história da ciência moderna afirma que os
cientistas, em especial os físicos, sempre alimentaram uma profunda desconfiança
pela metafísica, quando não uma autêntica repulsa. Tomada em sentido estrito e
como se desfrutasse de valor univeral, penso que essa tese é falsa. Ainda que seja
impossível formular uma posição comum para os cientistas-filósofos atuantes entre
fins dos oitocentos e início dos novecentos no que diz respeito à metafísica — se ela
é efetivamente um obstáculo para o desenvolvimento da ciência, por exemplo —,
creio ser inquestionável que um bom número deles não apenas aceitava a existência
da metafísica, mas pensava que esta poderia dar algum tipo de contribuição positiva
para a ciência. No entanto, creio que este é um trabalho que está por ser feito. Em
outras palavras, penso que seria muito útil e instrutivo conhecer o “lugar” atribuído
à metafísica pelos cientistas deste período. Apesar de não ser difícil mostrar opiniões
positivas e favoráveis, a historiografia mais conhecida e divulgada ainda permanece
aferrada à crença de que a metafísica sempre foi percebida como um problema ou
um obstáculo. Por exemplo, entre aqueles que publicamente defendiam o completo
afastamento entre ciência e metafísica, encontra-se a célebre figura de Ernst Mach
Principia 13(3): 371–84 (2009).
Published by NEL — Epistemology and Logic Research Group, Federal University of Santa Catarina (UFSC), Brazil.
372 Antonio Augusto Passos Videira

(1838-1916). Constitui um dos lugares comuns da história e da filosofia da ciência a

tese de que Mach, desde a sua juventude, quando tomou aversão pela “coisa em si”
kantiana, dedicou-se à elaboração de um pensamento científico-filosófico dentro do
qual a metafísica não teria lugar. E mais: toda a sua “filosofia” deve ser compreendida
a partir do seguinte “princípio”: “É imperioso lutar pela exclusão da metafísica!”; este
seria o seu principal leitmotiv. Em parte, tal leitmotiv foi disseminado pelos membros
do Círculo de Viena a partir dos anos 1910 do século passado.
A repulsa que Mach nutria pela metafísica, todavia, não se explica apenas por sua
rejeição do sistema kantiano, no que ele foi acompanhado por muitos cientistas seus
contemporâneos. Além desta razão, deve-se mencionar uma outra, tão importante
quanto aquela: sua preocupação em evitar que a separação entre ciência e senso
comum se tornasse irreversível. O próprio principio de economia, talvez o mais cé-
lebre formulado por Mach, pode ser entendido como integrando a sua estratégia
para evitar aquela separação. A importância que Mach concedia à teoria da evolu-
ção de Charles Darwin explica-se pelo mesmo motivo. Diferentemente do primeiro
motivo, a rejeição a Kant, o segundo não é estritamente negativo; ele é construtivo,
fixando-se a sua contribuição mais relevante na tese relativa à fundamentação, ou
justificação, daquilo que se considera ser o conhecimento em parâmetros acessíveis
à espécie humana.
O ponto de vista adotado no presente trabalho não procura, como é habitual-
mente o caso, discutir as críticas de Mach à versão newtoniana da mecânica ou ainda
se a sua filosofia da ciência, marcadamente empirista e sensualista, impediria o pro-
gresso científico. Acredito que a maioria das análises históricas e filosóficas sobre
Mach são formuladas a partir do desenvolvimento da física posterior à sua morte;
desenvolvimento, que, a se crer nas opiniões correntes, desmentiu as suas princi-
pais teses, gerando a conclusão de que ele teria falhado no seu objetivo de formular
uma concepção de ciência capaz de contribuir para o progresso desta última. Sua
resistência ao atomismo é muitas vezes citada como prova de sua incapacidade de
adaptar ao novo. Filósofos de inspiração marxista, os quais normalmente associam
atomismo a materialismo, estiveram entre os principais críticos de Mach. Contudo, o
que ocorre é que não se dá a devida atenção ao interesse de Mach pela promoção de
mudanças, isto é, essas análises não levam em consideração o “motor” responsável
pela possibilidade de obtenção de progresso: “. . . o fator que mais promove o pen-
samento científico é a ampliação gradual do campo de experiências.” (Mach 1943:
222) A vontade e a necessidade de conhecer o novo, o incomum, ou ainda: aquilo
que ainda não foi compreendido.
Não pretendo no escopo deste trabalho discutir se tais análises e conclusões são
corretas. Uma análise historiográfico-filosófica do “destino” do pensamento de Mach
ficará para uma ocasião futura. Ainda que não tenha espaço para justificar esta posi-
ção, parece-me que as análises sobre esse importante pensador austríaco de fins do

Principia 13(3): 371–84 (2009).

O Naturalismo como Atitude: Mach em Disputa com a Metafísica 373

século XIX e dos primeiros anos do século seguinte, não são completamente isentas,
já que o compreendem a partir dos resultados da ciência. Mach parece-me sempre
ser “acusado” de algum crime que supostamente teria cometido. Como no caso da
sociedade brasileira, onde os acusados sempre têm que se defender, Mach, ou me-
lhor, aqueles que acreditam em suas teses, devem, sob pena de receber um veredicto
negativo, apresentar provas em favor das idéias do físico austríaco de origem checa.
É possível que o próprio Mach não se importasse com isso, uma vez que sempre se
considerou um físico e não um filósofo. As atitudes dos filósofos lhes pareciam equi-
vocadas; elas o seriam, posto que à filosofia, conforme Mach compreendia aquela que
se fazia majoritariamente em seu tempo, caberia a determinação de um fundamento
último e definitivo para o conhecimento. Em outras palavras, a filosofia parecia se
ocupar principalmente com metafísica. Mach sempre se pautou pela busca de uma
concepção de conhecimento que fosse livre de todo e qualquer traço de metafísica,
já que não acreditava na capacidade da filosofia em dar respostas definitivas para
questões referentes a temas, como, por exemplo, a origem do universo ou se existi-
ria a alma. Ao escolher os cientistas como alvo principal para as suas publicações,
Mach pretendia mostrar-lhes a inutilidade, quando não o perigo, das interrogações
sem fim sobre a natureza última das coisas, como, por exemplo, da matéria ou do
calor. O único meio para por fim a essas reflexões seria através do reconhecimento
de que a matéria não é uma substância, mas, sim, um conceito, isto é, um símbolo de
pensamento, que substitui um conjunto de relações que mantemos com os objetos
e que estes mantêm entre si. A formação de tais conceitos é uma tendência natural
nos seres humanos, tal tendência, todavia, pode se revelar desastrosa, donde peri-
gosa, caso se considere que o termo designa uma substância verdadeira que existe de
maneira incondicionada. É nesta crítica que se pode encontrar a crítica mais radical
que Mach dirigia à metafísica. Uma verdade incondicionada estaria para além das
potencialidades intelectuais dos seres humanos. Em outros termos, a metafísica de-
veria ser criticada, uma vez que ela dava origem a uma série de falsos problemas,
irresolúveis porque mal formulados. Assim, pode-se pensar que Mach nutria como
um de seus principais objetivos em mostrar o sem sentido de alguns dos proble-
mas mais antigos sobre os quais a humanidade se debruçava tentando fornecer-lhes
Um ponto que deve ser enfatizado quando se analisa o pensamento de Mach é
a sua preocupação em gerar uma imagem de conhecimento que esteja na medida
do ser humano. Mach sempre lutou para que se reconhecesse que o conhecimento
é humano, o que significa afirmar que ele foi gerado por homens para os seus se-
melhantes: “[O] conhecimento . . . é um produto da natureza orgânica (Mach 1943:
217). De certo modo, pode-se considerar Mach um adepto da antiga fórmula de Pro-
tágoras: “O homem é a medida de todas as coisas.” É aqui, na obediência a essa
fórmula, que ele mais se aproxima do positivismo. Se o conhecimento deve ser ade-

Principia 13(3): 371–84 (2009).

374 Antonio Augusto Passos Videira

quado ao homem nada melhor do que recorrer às sensações, produtos da interação

de todo e qualquer ser humano com o meio ambiente, uma vez que seria sobre elas
que se poderia construir uma imagem de mundo comum aos homens, ao menos a
uma maioria dentre eles.
Se filosofia fosse a produção de reflexões sobre a natureza última e definitiva
do conhecimento, da verdade, do bem, entre outros temas, ela deveria ser recusada
em favor de algo diferente, uma vez que a análise do conhecimento, bem como as
produções intelectuais dos seres humanos, devem obrigatoriamente acontecer numa
esfera acessível à fisiologia da nossa espécie. A suspeita de Mach com relação à filo-
sofia era tão forte que ele repetidas vezes afirmou não ter criado nenhum conjunto
de idéias e teses que merecesse receber tal denominação. Ao defender a crítica como
instrumento privilegiado para a compreensão do conhecimento, Mach, mais do que
se afastar da filosofia, pretendia reformulá-la. Tal reformulação deveria acontecer
por meio da incorporação de uma das principais marcas distintivas da ciência, a sa-
ber: a sua enorme capacidade de comunicação. Ou seja: a filosofia deveria se dar por
meio do diálogo. E o diálogo somente poderia ser uma realidade quando os filósofos
reconhecessem que deveriam se reunir para trocar idéias.
O máximo que Mach dizia ter desenvolvido era uma metodologia de análise
e uma psicologia do conhecimento. Tanto a primeira, como a segunda denomina-
ções para o pensamento de Mach, procuravam fixar as origens de certas idéias, o
que seria feito no interior do desenvolvimento da história da espécie humana. Fora
desta última, Mach pensava ser impossível compreender o surgimento e as trans-
formações sofridas pelas teorias concebidas pelos homens. Os filósofos poderiam,
segundo o próprio Mach, ver sua empreitada como sendo uma “clarificação filosó-
fica da metodologia científica”, a qual, entre outros resultados importantes, deveria
fornecer os seguintes: (1) fixar que a coisa era uma abstração, enquanto que eram
as sensações aquilo que existia realmente; e (2) Ao invés de causa, é necessário em-
pregar o conceito de função; na natureza não existem causas ou efeitos, uma vez
que ela se apresenta uma única vez. Com relação à causalidade, também o novo,
ou o desconhecido, poderiam mostrar os seus limites: “. . . a crença nesse agente
misterioso chamado causalidade, o qual mantém unidos pensamento e evento, é vio-
lentamente sacudido quando alguém entra pela primeira vez numa província da qual
ele não possui nenhuma experiência prévia.” (Mach 1943: 221, itálico no original.)
Em suma, Mach é daqueles que, ao final dos oitocentos, defendeu a naturalização
da filosofia, o que deveria promover a aproximação entre as ciências. Mas, não eram
apenas as diversas ciências que se afastavam umas das outras, num movimento na-
tural e desejável, posto que ele contribuísse para o progresso. A especialização não
merecia ser condenada. Também a filosofia se afastava das ciências, evento que de-
veria ser combatido, pois ele geraria uma especialização excessiva fundada numa
análise unilateral da prática e dos objetivos da ciência. Mach sempre desejou reapro-

Principia 13(3): 371–84 (2009).

O Naturalismo como Atitude: Mach em Disputa com a Metafísica 375

ximar ciência e filosofia, realizando isso através da determinação de que à segunda

caberia compreender como as idéias da ciência surgiram e se desenvolveram.
A “clarificação filosófica da metodologia científica” está fundada na tese de que
a comparação é um princípio básico para a formulação de conclusões a respeito do
comportamento da natureza. Além de tornar possível o conhecimento, a comparação
permitia a comunicação: “A comparação, como a condição fundamental da comuni-
cação, é o elemento vital interno da ciência mais poderoso.” (Mach 1943: 238). O
lugar fundamental arrisco-me a dizer que Mach dava à comparação, levou-o a de-
fender a necessidade de se empregar analogias e metáforas, outro ponto em que
se distanciava de muitos dos seus colegas, com os quais mesmo comungando cer-
tos princípios gerais, afastava-se dos mesmos em função das excessivas inclinações
“matematizantes” destes:
Essa relação entre sistemas de idéias, nos quais a dissimilaridade entre dois
conceitos homólogos, bem como a concordância nas relações lógicas entre
dois pares de conceitos homólogos, é claramente exposta, é denominada
de analogia. [Ela] consiste num meio efetivo para o domínio, numa com-
preensão unitária [unificada], de diferentes áreas. A trajetória [para esse
domínio] é simplesmente mostrada através da elaboração de uma fenome-
nologia física universal, a qual envolverá todas as áreas. (Mach 1943: 250,
itálico no original)

Como sugerido acima, uma outra característica importante da atitude de Mach

— penso ser melhor referir-se ao seu pensamento como sendo uma defesa de uma
atitude muito mais do que de um conjunto de idéias – é a sua tentativa de unificar
a ciência, o que somente seria possível se se recorresse à crítica, construída sobre
uma genealogia histórica. Mach sempre se manteve fiel à idéia de que a análise
histórico-crítica deveria ser considerada como completa com o reconhecimento de
que sensações encontram-se na sua origem. Em outros termos, a análise não deveria
ir além da determinação das sensações que estão na base das idéias com conteúdo
cognitivo. São os pensamentos que devem se adaptar aos fatos, adaptar-se entre
eles e não os fatos que devem se adaptar aos pensamentos. Esta tese restringiria
em muito o campo de ação de afirmações sem base empírica. Ainda assim, não se
deve concluir que Mach fosse contrário ao uso de hipóteses, sendo, ao contrário,
recomendável, desde que se tomasse cuidado com o seu emprego:
A construção de hipóteses, portanto, não é um produto de métodos científi-
cos artificiais. Este processo é, inconscientemente, transmitido desde a mais
antiga infância da ciência. Mesmo mais tarde, as hipóteses só se tornam pe-
rigosas e danosas para o progresso quando mais confiança lhes é atribuída
do que aos próprios fatos. (1943: 229)

Todas as idéias científicas encontram sua origem na interação do homem com o

seu ambiente natural. A história da ciência funcionaria como um sustentáculo da teo-

Principia 13(3): 371–84 (2009).

376 Antonio Augusto Passos Videira

ria machiana do conhecimento. Ao mostrar de que modo as idéias cientificas apare-

ceram e se transformaram, Mach desejava contribuir para a libertação dos bloqueios
epistemológicos que obstaculizavam a prática da ciência. Tal preocupação relembra-
nos de que Mach se dirigia principalmente aos seus colegas cientistas e menos aos
filósofos. Ou seja, o seu público leitor, tal como ele mesmo o imaginava, deveria ser
composto primeira e principalmente por cientistas; era sobre estes últimos que Mach
procurava influenciar.
No escopo deste texto, procuro adotar uma postura “simpática” à atitude de
Mach. Não que eu queira criticar a filosofia, como ele o fez, mesmo reconhecendo
que algumas de suas opiniões fazem sentido. O meu objetivo consiste em apresentar
e discutir aquilo que me parece ser o cerne da atitude filosófica de Mach com relação
ao “desvelamento” dos motivos que o levaram a se opor à metafísica. Como não era
incomum em sua época, a postura anti-metafísica de Mach ligava-se à sua rejeição à
tese de que existira um ser onipotente, onipresente e oniconsciente. É bem conhecido
que Mach era ateu. A dupla rejeição da metafísica e da religião o influenciou sobre
como deveria ser a visão correta do que é o conhecimento. É interessante observar
a relação que Mach estabelece entre a rejeição à existência de entidades as quais
os seres humanos não podem ter acesso e liberdade daí resultante. Essa relação
certamente não era inovadora, podendo ser considerada como herdeira direta do
Iluminismo. Quanto a mim, o interesse contido nessa relação negativa encontra-
se na coerência com que Mach a defendeu, bem como na defesa que fez das suas
Nós devemos ao Iluminismo a nossa crença na liberdade de pensamento, o que
já era defendido por muitos círculos intelectuais como um valor universal ao final
do século XIX. Nessa época, disseminava-se a tese que de que seria necessário liber-
tar a mente humana do ponto de vista que afirmava que a visão de mundo físico-
mecanicista teria primazia na tarefa de compreender o mundo. Esse “movimento de
libertação” alcançaria a física, sendo uma consequência imposta à humanidade pelas
novas tarefas do trabalho e da organização sócio-econômica. Se a física conseguisse
deixar de conceder ao mecanicismo o primeiro lugar como fonte de inspiração para
a explicação do mundo, ela realizaria aquilo que lhe cabe. Mach reconhece que a
análise das tarefas e dos métodos da física não é exatamente uma ocupação do fí-
sico, mas, sim, do filósofo. Reconhecendo a existência de fronteiras disciplinares, ele
pede desculpas por ultrapassar a fronteira da física com a filosofia, justificando-se
com a afirmação de que o conhecimento é um problema que atinge e interessa a
todas as áreas e que, por isso mesmo, fronteiras fixas e rígidas não podem ser traça-
das: “Toda razão [motivo] que [nos] impele e estimula a modificar e a transformar
nossos pensamentos procede daquilo que é novo, incomum e não compreendido.”
(Mach 1943: 222)
Além de nos levar a conhecer coisas até então desconhecidas, a natureza imporia

Principia 13(3): 371–84 (2009).

O Naturalismo como Atitude: Mach em Disputa com a Metafísica 377

à ciência uma atitude pluralista, como se pode perceber na citação abaixo:

Muitos lados, no entanto, possui a natureza. (. . . ) Mas nós não precisamos
imaginar que o progresso, por meio de trajetórias uma vez percorridas, é o
único meio de atingirmos a verdade. (1943: 217, itálicos no original)

Descrevendo de modo direto, talvez excessivamente direto, o verdadeiro obje-

tivo das teses científico-filosóficas de Mach, creio que o seu fim mais importante
consistia no estabelecimento do conhecimento em bases puramente humanas. Se o
conhecimento é produzido pelo homem e para o seu próprio uso, o mesmo deveria
acontecer com a sua justificação. Neste ponto, atua, uma vez mais, a sua aceitação
radical da teoria da evolução de Darwin.
A questão que mais o preocupava pode ser formulada do seguinte modo: se a
mecânica não deve ser considerada como o fundamento da ciência, e nem mesmo o
da física, como então determinar os critérios que serão usados para a escolha entre
teorias científicas rivais, ou ainda: como reconhecer que uma teoria conseguiu reali-
zar os seus objetivos de descrever os fenômenos naturais? Esta questão foi formulada
por ele do seguinte modo:
Suponha-se, agora, que [se realizou] o ideal de alcançar uma dada provín-
cia de fatos. Realiza a descrição tudo aquilo que o investigador pode per-
guntar? Em minha opinião, sim. A descrição é a construção no pensamento
de fatos e essa construção, nas ciências experimentais, é frequentemente a
condição da execução real. Nossas imagens mentais quase são um substituto
completo para os fatos e o meio pelo qual nós podemos assegurar todas as
suas propriedades. (. . . ) [A] ciência completa no pensamento fatos, os quais
são dados apenas parcialmente. Isso é possível graças à descrição, já que a
descrição pressupõe a interdependência dos elementos descritivos: de outro
modo, nada poderia ser descrito. (Mach 1943: 252–3, itálicos no original)

Em suma, para Mach, era fundamental determinar o que poderia ser obtido por
uma descrição.
O reconhecimento de que não seria possível evitar o uso de critérios para a avali-
ação do estágio alcançado pelo conhecimento não implica que a explicação para eles
encontra-se na estrutura da razão. Uma vez que aquilo que os filósofos chamam de
razão tem uma história, isto é, fruto de um processo evolutivo, não é mais factível
pensar que ela é capaz de gerar uma crença insuspeita em seu funcionamento. A
razão humana seria falível.
Mach não acreditava ser possível responder de modo definitivo à questão acerca
dos critérios que deveriam justificar aqueles outros efetivamente empregues pelos
físicos e demais cientistas naturais. De modo breve, estabelecer critérios (problema
normalmente atribuído à epistemologia) é sinônimo de fazer metafísica, pois, para
que os critérios pudessem realizar aquilo que se espera deles, sendo que uma das

Principia 13(3): 371–84 (2009).

378 Antonio Augusto Passos Videira

mais importantes de suas atribuições é a determinação do valor de verdade, seria

preciso que eles estivessem ou fundados numa entidade transcendente, como Deus,
ou fundados na natureza íntima das coisas. Reconhecendo, todavia, a impossibili-
dade de abdicar do uso de critérios, Mach procurou fixá-los de modo a que fossem
o mais possível naturais aos seres humanos. Em outras palavras, Mach defendia que
os critérios se originariam com a espécie humana; eles seriam fruto dos mesmos
processos evolutivos pelos quais passou a espécie humana.
Para mostrar a naturalidade dos critérios, ou seja, determinar a sua origem no
mesmo processo evolutivo que fixou (temporariamente) aquilo que somos atual-
mente, Mach recorreu à história. Ele é conhecido como um dos mais importantes
historiadores da ciência do final do século XIX. Possivelmente apenas Duhem possa
rivalizar com ele no quesito referente à relevância da história. No entanto, Mach,
diferentemente do seu colega e companheiro de disputas filosóficas e científicas,
nunca explicou com clareza o porquê de seu recurso à história. Mach cala-se sobre
esse tema, talvez, assim, evitando construir uma teoria da história, o que lhe levaria
a formular e a defender a necessidade de uma base apriorística para as suas descri-
ções históricas. Se tivesse que construir uma teoria da história, as suas leis teriam
que estar fora da história e, fora desta, nada poderia existir. Mach gostava de repetir
a seguinte frase: “A história tudo fez, tudo pode ela transformar.” (A citação não é
literal.) A história parece não possuir leis, o que nos permite conjecturar que ela es-
capa à explicação, restringindo-se somente à descrição. Como justificativa implícita
para o seu silêncio sobre a natureza da história, Mach, ao final do seu livro sobre
o desenvolvimento histórico da mecânica, afirma que a sua concepção de economia
do pensamento foi desenvolvida na sua experiência didática, através da sua prática
de professor.
Ao acreditar que tudo foi feito pela história e que ela pode modificar tudo, Mach
parece esposar uma noção restrita de previsão. Se a previsão tem limites, melhor se-
ria não confundi-la com o conhecimento. Este deveria ser “maior” do que a previsão.
Mach concedia um lugar especial à contingência como no caso da história, que seria
o domínio da contingência e não da necessidade, tão cara aos filósofos. Ao ligar ciên-
cia e história, Mach procurava romper com o vínculo entre leis e teorias científicas e
necessidade filosófica. Mas, postar-se ao lado daqueles que defendiam a contingên-
cia gerava um problema importante. Ainda que a história pudesse ser o domínio da
contingência, ela não era arbitrária, o que poderia retirar-lhe toda inteligibilidade
dos esquemas descritivos que dela decorriam.
A história também é importante, pois permite que nos familiarizemos com as
descrições que construímos dos fenômenos analisados. Para Mach, explicar significa
reconduzir fenômenos complexos a relações entre fatos e que são consideradas como
mais simples. Uma das tarefas mais importantes da história da ciência seria mostrar,
contra toda a aparência de arbitrariedade que se poderia imaginar em função da con-

Principia 13(3): 371–84 (2009).

O Naturalismo como Atitude: Mach em Disputa com a Metafísica 379

tingência que rege os eventos históricos, as motivações que presidiram à formulação

desse conceito ou daquele enunciado.
Segundo Mach, a história da ciência era inseparável de uma epistemologia natu-
ralizada, a qual ligaria a seleção de fatos, classificados como fundamentais, ao hábito
e à história da nossa espécie. A análise histórica da ciência ensina-nos que é difícil
estabelecer onde, quando e de que maneira começou realmente o desenvolvimento
da ciência, parecendo natural admitir que o agrupamento instintivo das experiências
fosse anterior à ordenação científica.
Mach, em suas análises sobre a ciência, defendia que os dados sensoriais, que
ele denominava de elementos, formavam o bloco constituinte da experiência e da
realidade. Assim, esperava impedir todo e qualquer discurso sobre entidades que
existiriam para além da experiência humana. A ciência nunca deveria fazer uso de
entidades que escapam à experiência humana, donde a sua aversão às teorias que
empregavam átomos. Os conceitos fundamentais da física deveriam ser entendidas
como cópias de experiências reais. O processo de construção de conceitos deve-se às
nossas interações com o meio ambiente, estando em estreita conexão com as ações
que realizamos cotidianamente:
Assim, pois, a ciência natural se propõe a investigar o que há de cons-
tante nos fenômenos, quais são os seus elementos e qual é a sua vinculação
e dependência mútua entre eles. Ela [a ciência] se esforça em economi-
zar, mediante uma descrição clara e completa, novos experimentos, [para]
despreocuparmo-nos deles . . . (Mach 1949: 17)
As observações anteriores não devem ser compreendidas como se para Mach a
ciência nada mais seria do que mero refinamento do senso comum. A ciência era
diferente do senso comum, na medida em que era abstrata e formalizada, sendo que
essa segunda característica pressupunha o uso da matemática. No entanto, esse uso
deveria ser parcimonioso para que se evitasse a elaboração de idéias sem referência a
sensações, tomando como base apenas as idéias e equações matemáticas. A força da
matemática era devida à sua capacidade em ser um instrumento útil na formulação
concisa das observações, o que era fundamental para o esforço de poupar energia.
Nas suas próprias palavras:
O conhecimento instintivo, involuntário, precede sempre ao seu conheci-
mento científico, consciente, vale dizer à investigação dos fenômenos. O pri-
meiro nasce das relações entre os fenômenos naturais e a satisfação das
nossas necessidades. A aquisição dos conhecimentos mais elementares não
compete, sem dúvida [alguma], somente ao indivíduo, mas também ela é
esperada para o desenvolvimento da espécie. (Mach 1949: 13, itálico no
Mach, como alguns de seus contemporâneos, reconhecia explicitamente que ha-
via se instalado um fosso preocupante entre as diversas ciências. O final do século

Principia 13(3): 371–84 (2009).

380 Antonio Augusto Passos Videira

XIX já reconhecia publicamente que a prática cientifica se transformara em atividade

de especialistas. Assim urgia, para Mach, bem como para outros de seus companhei-
ros empiristas, construir uma base capaz de reunificar a ciência, o que não obrigava
a elaboração de uma base comum para ela: o fenômeno da especialização, isto é,
de subdivisão em disciplinas particulares, era um “resultado” normal do processo de
transformação sofrido pela ciência moderna. A especialização ganharia um aspecto
francamente negativo caso impedisse a comunicação entre as diferentes especiali-
dades. Mas esse fosso não apenas atrapalhava o desenvolvimento da ciência, mas
também o da própria humanidade, uma vez que esta ultima dependia, e cada vez
mais, da ciência. Sem essa base unificadora, seria impossível à ciência realizar a sua
obra maior, a saber: promover a melhoria da condição humana.
Como visto anteriormente, não constitui um exagero afirmar que Mach defen-
dia que cientistas e filósofos reconhecessem que os problemas somente poderiam
ser considerados como resolvidos, caso suas respostas fossem dadas em dimensões
humanas, o que excluiria a possibilidade de serem definitivas. A dimensão humana
das respostas apresenta-se, inclusive, na formulação dos seus problemas originais.
Também os problemas devem ser formulados em termos humanos e não em termos
absolutos como se o homem pudesse desfrutar da mesma posição dos deuses. O ab-
soluto, ou em si, corresponderia a uma situação inatingível para todo e qualquer
ser humano. Como o absoluto é inatingível, ele não serve como fundamento para
o conhecimento. Ou ainda, a unificação do conhecimento não pode ser tentada em
termos que não são compreensíveis aos seres humanos, ou seja, passíveis de serem
vividos por eles. Neste ponto se apresenta uma característica fundamental do pensa-
mento de Mach: todo e qualquer conhecimento nada mais é do que um instrumento
de adaptação da espécie humana ao ambiente. Uma vez mais, percebe-se a influên-
cia do darwinismo no pensamento de Mach. É aqui que se encontra a origem do seu
Para Mach, a ciência não deveria alimentar o objetivo de explicar as engrena-
gens de uma hipotética realidade em si. Repetindo: o seu objetivo seria a economia
de pensamento, isto é, a representação econômica daquilo que é factual. Para isso,
dever-se-ia adaptar os pensamentos aos fatos, além de estabelecer uma adaptação
entre os próprios pensamentos. As representações, por exemplo, os conceitos, não
são definitivas, ou seja, não podem ser consideradas como prontas quando formula-
das. Sua característica principal encontra-se na sua capacidade de ser continuamente
aperfeiçoada. Sobre as representações, Mach afirmava o seguinte:

A representação mental que nos formamos dos fatos nunca são o fato total,
mas, sim, aquele aspecto do mesmo que é, para nós, importante (. . . ). Nos-
sas representações são sempre abstrações. Também aqui podemos ver uma
faceta do caráter econômico. (Mach 1949: 400)

Principia 13(3): 371–84 (2009).

O Naturalismo como Atitude: Mach em Disputa com a Metafísica 381

As representações são reproduções das interações entre os seres humanos e o

seu ambiente. Elas tornam possível a reconstrução da experiência com o objetivo de
economizar esforços aos homens. Seria a economia de pensamento, para Mach, o
reconhecimento da finitude do intelecto humano? Penso que sim. Segundo o filó-
sofo austríaco, a mente humana é limitada e, desse modo, é como que obrigada a
proceder economicamente para que possa espelhar no seu interior a imensa riqueza
existente no mundo externo. Sendo impossível ao ser humano apreender a realidade
em toda a sua totalidade, só lhe restaria formular uma descrição do real que esti-
vesse ao seu alcance, donde a exigência de que tal descrição fosse econômica. Em
outras palavras, a economia aqui significa a possibilidade de o homem trazer con-
sigo, ou seja, memorizar, a natureza “concretizada” nas representações, como as leis
A ciência seria, portanto, o conjunto de representações capazes de economizar
os nossos esforços, uma vez que ela incorpora ao seu conjunto de leis aquelas repre-
sentações que foram bem sucedidas, tornado possível que os homens se adaptassem
ao seu ambiente:

Toda ciência tem que substituir, ou economizar, a experiência mediante ima-

gens e representações mentais dos fatos; imagens que são mais fáceis de
manejar que a própria experiência . . . Com o reconhecimento do caráter
econômico desaparece também toda a mística da ciência. (Mach 1949: 399)

Ou ainda: “Toda ciência tem por missão substituir a experiência.” (Mach 1949: 406)
Mach sempre foi um defensor apaixonado do caráter econômico da ciência. Para
ele, a ciência deveria principalmente nos ajudar na economia de esforços por meio da
codificação dos processos adaptativos, os quais foram bem sucedidos anteriormente.
De certo modo, a ciência é uma forma simples, fácil e coletiva de memorização
daqueles resultados adaptativos que os seres humanos obtiveram ao longo da sua
história. Em suas palavras, “[A] física é experiência arrumada de modo econômico
(Mach 1943: 197)
A ciência para Mach é criada pelas e nas necessidades mais cotidianas dos seres
humanos. Ou ainda: a origem da ciência encontra-se na prática e não na especula-
ção. Em suma, a ciência origina-se na sociedade. Isso gera a seguinte implicação:
não existe ruptura radical, ou abismo intransponível, entre ciência e senso comum.
A prática científica mantém uma vinculação contínua com a prática cotidiana. A me-
lhor maneira de compreendermos a natureza da ciência é através dos seus inícios
“caseiros”, ou seja, do papel que ela desempenhou, como ainda o faz, na luta do
homem pela sua sobrevivência. É curioso observar que justamente na época em que
se dissemina a tese de que o pensamento está vinculado à sobrevivência surgia, de
modo avassalador, o sentimento de segurança, uma das marcas características do
século XX. O conhecimento tem uma origem primitiva, o responsável pela solidez do

Principia 13(3): 371–84 (2009).

382 Antonio Augusto Passos Videira

pensamento científico: “Os maiores avanços da ciência têm sempre consistido [na
obtenção] de uma formulação bem sucedida em termos claros, abstratos e comu-
nicáveis daquilo que era há muito tempo conhecido [de modo] instintivo e, assim,
tornando-o uma propriedade permanente da humanidade.” (Mach 1943: 191) Ou
ainda: “. . . as leis mais felizes [i.e. bem sucedidas] não caem do céu, mas emergem
(saltam) de noções já existentes.” (Mach 1943: 226)
A abstração e a formalização, marcas próprias da ciência moderna, em particular,
daquelas que sofreram o impacto da matematização das suas fórmulas e representa-
ções como é o caso da física, não devem ser vistas como desconectando o mundo da
vida daquele outro construído graças ao recurso do conhecimento científico. Para o
nosso cientista-filósofo, a ciência é, acima de tudo, um instrumento de adaptação do
ser humano ao seu ambiente. A ciência nunca poderá ser considerada pronta e com-
pleta; ela se encontra continuamente submetida a um processo de melhoramento.
Jamais seria possível construir uma teoria científica definitivamente verdadeira do
mundo. Todos os enunciados empíricos das ciências naturais têm um caráter hipo-
tético. A ciência se move no domínio das conjecturas. As próprias teorias científicas
nada mais são do que instrumentos provisórios.

À Guisa de Conclusão

Sou da opinião que Mach sempre defendeu uma concepção falibilista e revisionista
de ciência. No entanto, isso não significa, ou melhor, não implica que a ciência seja
instável. Ao contrário, uma de suas mais marcantes características é a sua estabili-
dade. Mas, o que gera tal estabilidade? Se não nos é possível conhecer a natureza
última das coisas, não devemos procurar a base, ou o mecanismo, dessa estabilidade
nas coisas ou na substância como uma filosofia essencialista procuraria realizar. Para
Mach, a estabilidade deveria ser explicada por algum mecanismo que envolvesse di-
retamente os seres humanos. A ciência não é obra de deuses, mas, sim, de seres que
são finitos, falíveis e contingentes. São os interesses humanos, e que se desenvol-
veram junto com a própria espécie humana, que vão, em última instância, explicar
de que modo a ciência torna-se estável. O mecanismo humano que torna possível
a estabilidade da ciência são os seus interesses, entre os quais se sobressai aquele
ligado à sobrevivência da espécie. A estabilidade, tal como a existência da ciência,
deve receber uma explicação naturalista: “A transformação das idéias aparece assim
como uma parte da evolução geral da vida, como uma parte da sua adaptação a uma
esfera de ação que constantemente se amplia.” (Mach 1943: 233)
Alguns cientistas-filósofos da estirpe de Mach, Poincaré, Bolztmann, Duhem,
Einstein e Heisenberg entre outros, não apenas colocaram em discussão a existência
da metafísica, mas, ainda que de modo conflituoso, defenderam, em certos casos,
uma relação positiva entre esta e a própria ciência natural. Aqueles que defende-

Principia 13(3): 371–84 (2009).

O Naturalismo como Atitude: Mach em Disputa com a Metafísica 383

ram essa segunda posição, desconfiavam que a metafísica não se reduzisse a uma
mera teoria sobre a constituição de objetos, principalmente daqueles invocados para
explicar seja a composição, seja a estrutura dos fenômenos visíveis. Para além des-
ses temas, a metafísica teria relação com o antigo e bem conhecido problema da
orientação, a saber: qual o sentido da existência humana?
De que modo é possível perceber e mostrar a preocupação por parte de “nossos”
cientistas-filósofos com a questão do sentido? Penso que uma possibilidade encontra-
se se direcionarmos o foco para as suas preocupações com o ensino, mais especifica-
mente com os rumos que deveriam ser dados ao ensino de línguas clássicas e ciências
naturais nos liceus e ginásios de seus países e épocas. Recorrendo a uma termino-
logia que se tornou famosa na segunda metade do século XX, os cientistas de finais
dos oitocentos percebiam que o fosso crescente entre as duas culturas — científica e
humanista - era preocupante.
Atendo-me especificamente ao caso de Mach, se, por um lado, é correto afirmar
que ele favoreceu a posição que pregava a diminuição do número de horas de au-
las de latim e grego, ele nunca o fez, defendendo as suas completas eliminações. O
grego e o latim fazem parte da história de todo aquele que se considera um habitan-
tes do espaço cultural e ideológico definido pelo Ocidente. Não ensinar as línguas
clássicas e as idéias a elas associadas geraria uma situação inaceitável, uma vez que
os resultados obtidos através do método histórico-crítico não poderiam nem mesmo
ser reconhecidos, o que retiraria todo o seu valor. O que Mach desejava para o ensino
era que este fosse capaz de treinar os jovens no exercício consciente e quotidiano da
interrogação, da investigação e da formulação ousada de soluções para os problemas
que atraíam a sua atenção. Em suma, o que me parece ser válido para aquilo que
Mach entende ser ciência ou conhecimento, seja no que diz respeito ao seu ensino
e à sua transmissão, seja no que diz respeito à sua criação, sintetiza-se na seguinte
conclusão: o ser humano não pode esquecer-se de que o conhecimento e a sua trans-
missão existem para a promoção da sua liberdade e não o inverso. Finalmente, a
liberdade humana somente poderia ser plena e autenticamente exercida caso pro-
curasse conhecer e acolher o novo, consertando nas tradições recebidas aquilo que
fosse necessário. Inversamente, o melhor critério para sabermos se um conhecimento
merecesse ser visto como tal é se ele promoveria o sentimento de liberdade entre os
seres humanos. Em suma, liberdade e conhecimento seriam complementares e in-


Baç, M. 2000. Strucuture versus Process: Mach, Hertz, and the Normative Aspect of Science.
Journal for General Philosophy of Science 31: 39–56.

Principia 13(3): 371–84 (2009).

384 Antonio Augusto Passos Videira

Blackmore, J. 1972. Ernst Mach — His Work, Life, and Influence. Berkeley, Los Angeles, Lon-
dres: University of California Press.
Boltzmann, L. 2004. Escritos Populares. (Seleção, Tradução e Introdução de Antonio Augusto
Passos Videira) São Leopoldo: Editora da Unisinos.
Garreta, G. 2002. Ernst Mach: l’épistémologie comme historie naturelle de la science. In P.
Wagner (org.) Les philosophes et la science. Paris: Gallimard (Folio/Essais): 624–58.
Mach, E. 1911. History and Root of the Principle of the Conservation of Energy. (Tradução de
Philip. E. B. Jourdain) Chicago/Londres: The Open Court Publishing Co.
———. 1943. Popular Scientific Lectures. (Tradução de Thomas J. McCormack) La Salle (Ill.):
The Open Court Publishing Co.
———. 1949. Desarrolo Historico-Critico de la Mecanica. (Tradução de José Babini) Buenos
Aires/México: Espasa/Calpe Argentina.
Miguel, L. & Videira, A. A. P. 2008. A Idéia de Evolução como Ponte entre Ciência, História
e Evolução. Ciência & Ambiente (Pensando a Evolução) 36: 71–85.
Wolters, G. 2001. Mach. In W. H. Newton-Smith (ed.) A Companion to the Philosophy of
Science (Blackwell Companions to Philosophy). Londres: Blackwell Publishers: 252–6.


Departamento de Filosofia
Universidade do Estado do Rio de Janeiro
Rua São Francisco Xavier 524, sala 9027B
20550-013 Rio de Janeiro, RJ

Resumo. Este artigo defende a hipótese de que Ernst Mach (1838-1916) recorreu ao natu-
ralismo (termo que ele não usou) — atitude derivada da sua aceitação da teoria da evolução
de Darwin — como arma contra a metafísica (aqui compreendida como uma atitude em
favor da busca por fundamentação última, fixa e definitiva), já que procuraria estabelecer
critérios de avaliação, incluindo os relativos à escolha de teorias científicas, efetivamente
capazes de serem usados pelos seres humanos.

Palavras-chave: Mach, naturalismo, metafísica, história da ciência.

Principia 13(3): 371–84 (2009).