Possibility theory

From Wikipedia, the free encyclopedia

Jump to: navigation, search

Possibility theory is a mathematical theory for dealing with certain types of

uncertainty and is an alternative to probability theory. Professor Lotfi Zadeh
first introduced possibility theory in 1978 as an extension of his theory of
fuzzy sets and fuzzy logic. D. Dubois and H. Prade further contributed to its
development. Earlier in the 50s, economist G.L.S. Shackle proposed the
min/max algebra to describe degrees of potential surprise.

Formalization of possibility

For simplicity, assume that the universe of discourse Ω is a finite set, and
assume that all subsets are measurable. A distribution of possibility is a
function from Ω to [0, 1] such that:

Axiom 1:
Axiom 2:
Axiom 3: for any disjoint
subsets U and V.

It follows that, like probability, the possibility measure on finite set is

determined by its behavior on singletons:

provided U is finite or countably infinite.

Axiom 1 can be interpreted as the assumption that Ω is an exhaustive

description of future states of the world, because it means that no belief
weight is given to elements outside Ω.

Axiom 2 could be interpreted as the assumption that the evidence from

which was constructed is free of any contradiction. Technically, it
implies that there is at least one element in Ω with possibility 1.
Axiom 3 corresponds to the additivity axiom in probabilities. However there
is an important practical difference. Possibility theory is computationally
more convenient because Axioms 1-3 imply that:

for any subsets U and V.

Because one can know the possibility of the union from the possibility of
each component, it can be said that possibility is compositional with respect
to the union operator. Note however that it is not compositional with respect
to the intersection operator. Generally:

Remark for the mathematicians:

When Ω is not finite Axiom 3 can be replaced by:

For all index sets I, if the subsets are pairwise disjoint,

Necessity

Whereas probability theory uses a single number, the probability, to describe

how likely an event is to occur, possibility theory uses two concepts, the
possibility and the necessity of the event. For any set U, the necessity
measure is defined by

In the above formula, denotes the complement of U, that is the elements

of Ω that do not belong to U. It is straightforward to show that:

for any U

and that:

Note that contrary to probability theory, possibility is not self-dual. That is,
for any event U, we only have the inequality:
However, the following duality rule holds:

For any event U, either , or

Accordingly, beliefs about an event can be represented by a number and a

bit.

Interpretation

There are four cases that can be interpreted as follows:

means that U is necessary. U is certainly true. It implies that

.

means that U is impossible. U is certainly false. It implies that

.

means that U is possible. I would not be surprised at all if U

occurs. It leaves unconstrained.

means that U is unnecessary. I would not be surprised at all if U

does not occur. It leaves unconstrained.

The intersection of the last two cases is and

meaning that I believe nothing at all about U. Because it allows for
indeterminacy like this, possibility theory relates to the graduation of a
many-valued logic, such as intuitionistic logic, rather than the classical two-
valued logic.

Note that unlike possibility, fuzzy logic is compositional with respect to both
the union and the intersection operator. The relationship with fuzzy theory
can be explained with the following classical example.

• Fuzzy logic: When a bottle is half full, it can be said that the level of
truth of the proposition "The bottle is full" is 0.5. The word "full" is
seen as a fuzzy predicate describing the amount of liquid in the bottle.
 • Possibility theory: There is one bottle, either completely full or totally
empty. The proposition "the possibility level that the bottle is full is
0.5" describes a degree of belief. One way to interpret 0.5 in that
proposition is to define its meaning as: I am ready to bet that it's
empty as long as the odds are even (1:1) or better, and I would not bet
at any rate that it's full.

Possibility theory as an imprecise probability theory

There is an extensive formal correspondence between probability and

possibility theories, where the addition operator corresponds to the
maximum operator.

A possibility measure can be seen as a consonant plausibility measure in

Dempster–Shafer theory of evidence. The operators of possibility theory can
be seen as a hyper-cautious version of the operators of the transferable belief
model, a modern development of the theory of evidence.

Possibility can be seen as an upper probability: any possibility distribution

defines a unique set of admissible probability distributions by

This allows one to study possibility theory using the tools of imprecise
probabilities.

Necessity logic

We call generalized possibility every function satisfying Axiom 1 and

Axiom 3. We call generalized necessity the dual of a generalized possibility.
The generalized necessities are related with a very simple and interesting
fuzzy logic we call necessity logic. In the deduction apparatus of necessity
logic the logical axioms are the usual classical tautologies. Also, there is
only a fuzzy inference rule extending the usual Modus Ponens. Such a rule
says that if α and α → β are proved at degree λ and μ, respectively, then we
can assert β at degree min{λ,μ}. It is easy to see that the theories of such a
logic are the generalized necessities and that the completely consistent
theories coincide with the necessities (see for example Gerla 2001).
Possibility theory

Possibility theory is an uncertainty theory devoted to the handling of

incomplete information. As such, it complements probability theory. It
differs from the latter by the use of a pair of dual set-functions (possibility
and necessity measures) instead of only one. This feature makes it easier to
capture partial ignorance. Besides, it is not additive and makes sense on
ordinal structures. The name Theory of Possibility was coined in (Zadeh
1978), inspired by (Gaines and Kohout 1975). In Zadeh's view, possibility
distributions were meant to provide a graded semantics to natural language
statements. However, possibility and necessity measures can also be the
basis of a full-fledged representation of partial belief that parallels
probability (Dubois and Prade 1988). Then, it can be seen either as a coarse,
non-numerical version of probability theory, or as a framework for reasoning
with extreme probabilities, or yet as a simple approach to reasoning with
imprecise probabilities (Dubois, Nguyen and Prade, 2000).

Basic Notions

A possibility distribution is a mapping from a set of states of affairs S to

a totally ordered scale such as the unit interval . The function
represents the knowledge of an agent (about the actual state of affairs)
distinguishing what is plausible from what is less plausible, what is the
normal course of things from what is not, what is surprising from what is
expected. It represents a flexible restriction on what the actual state of affairs
is, with the following conventions:

• means that state s is rejected as impossible;

• means that state s is totally possible (= plausible or
unsurprizing).

If the state space is exhaustive, at least one of its elements should be the
actual world, so that at least one state is totally possible (normalisation).
Distinct values may simultaneously have a degree of possibility equal to 1.

Possibility theory is driven by the principle of minimal specificity. It states

that any hypothesis not known to be impossible cannot be ruled out. A
possibility distribution is said to be at least as specific as another one if and
only if each state is at least as possible according to the latter as to the
former (Yager 1983). Then, the most specific one is the most restrictive and
informative.

In the possibilistic framework, extreme forms of partial knowledge can be

captured, namely:

• Complete knowledge: for some state and for

other states s (only is possible)
• Complete ignorance: , (all states are totally possible).

Given a simple query of the form does an event A occur?, where A is a

subset of states, or equivalently does the actual state lie in A, a response to
the query can be obtained by computing degrees of possibility and
necessity, respectively (if the possibility scale is ):

The possibility degree evaluates to what extent event A is consistent

with the knowledge , while evaluates to what extent is certainly
implied by the knowledge. The possibility-necessity duality is expressed by
where is the complement of Generally,
and . Possibility measures satisfy the basic
maxitivity property:

Necessity measures satisfy an axiom dual to that of possibility measures,

namely On infinite spaces, these axioms must
hold for infinite families of sets.

Human knowledge is often expressed in a declarative way using statements

to which some belief qualification is attached. Certainty-qualified pieces of
uncertain information of the form is certain to degree can then be
modelled by the constraint The least specific possibility
distribution reflecting this information assign possibility 1 to states where
is true and to states where A is false.
Apart from , which represents the idea of potential possibility, another
measure of guaranteed possibility can be defined (Dubois, Hajek and
Prade, 2000) : It estimates to what extent all states in are
actually possible according to evidence.

Notions of conditioning and independence were studied for possibility

measures. Conditional possibility is defined similarly to probability theory
using a Bayesian like equation of the form (Dubois and Prade, 1988):

However, in the ordinal setting the operation cannot be a product and is

changed into the minimum. In the numerical setting, there are several ways
to define conditioning, not all of which have this form (Walley, 1996). There
are several variants of possibilistic independence (De Cooman, 1997;
Dubois et al. 1997; De Campos and Huete, 1999). Generally, independence
in ordinal possibility theory is neither symmetric, nor insensitive to negation.
For non Boolean variables, independence between events is not equivalent
to independence between variables. Joint possibility distributions on
Cartesian products of domains can be represented by means of graphical
structures similar to Bayesian networks for joint probabilities (see Borgelt et
al. 2000; Benferhat Dubois Garcia and Prade 2002). Such graphical
structures can be taken advantage of for evidence propagation (Ben Amor et
al, 2003) or learning (Borgelt and Kruse, 2003).

Historical Background

Zadeh was not the first scientist to speak about formalising notions of
possibility. The modalities possible and necessary have been used in
philosophy at least since the Middle-Ages in Europe, based on Aristotle's
works. More recently these concepts became the building blocks of modal
logic that emerged at the beginning of the XXth century. In this approach,
possibility and necessity are all-or-nothing notions, and handled at the
syntactic level. Independently from Zadeh's view, the notion of possibility as
opposed to probability was central in the works of one economist, Shackle,
and is present in those of two philosophers, Cohen and Lewis:

• The English economist G. L. S. Shackle (1962) introduced a full-

fledged approach to uncertainty and decision in the 1940-1970's based
on degrees of potential surprise of events. They are degrees of
 impossibility, that is, degrees of necessity of the opposite events. The
degree of surprise of an event made of a set of elementary hypotheses
is the degree of surprise of its least surprising realisation. Potential
surprise is understood as disbelief. The disbelief notion introduced
later in (Spohn 1990) employs the same type of convention as
potential surprise, using the set of ordinals as a disbelief scale.
Shackle also introduces a notion of conditional possibility. A
framework very similar to the one of Shackle was proposed by the
philosopher L. J. Cohen who considered the problem of legal
reasoning (Cohen, 1977).
• The philosopher David Lewis introduced a graded notion of
possibility in the form of a weak order between possible worlds he
calls comparative possibility (Lewis, 1973). He relates this concept of
possibility to a notion of similarity between possible worlds and a
most plausible one. Comparative possibility relations are instrumental
in defining the truth conditions of counterfactual statements.
Comparative possibility relations obey the key axiom, for all
events A, B, C: This axiom was later
independently proposed in (Dubois 1986) in an attempt to derive a
possibilistic counterpart to De Finetti and Savage comparative
probabilities.

• Zadeh (1978) proposed an interpretation of membership functions of

fuzzy sets as possibility distributions encoding flexible constraints
induced by natural language statements. Zadeh articulated the
relationship between possibility and probability, noticing that what is
probable must preliminarily be possible. However, the view of
possibility degrees developed in his paper refers to the idea of graded
feasibility (degrees of ease, as in the example of how many eggs can
Hans eat for his breakfast) rather than to the epistemic notion of
plausibility laid bare by Shackle. Nevertheless, the key axiom of
maxitivity for possibility measures is highlighted. Later on, Zadeh
(1979) acknowledged the connection between possibility theory,
belief functions and upper/lower probabilities, and proposed their
extensions to fuzzy events and fuzzy information granules.

This state of fact lays bare two branches of possibility theory: the qualitative
and the quantitative one (Dubois and Prade, 1998).

Qualitative possibility theory

Qualitative possibility relations can be partially specified by a set of
constraints of the form , where A and B are events that may take the
form of logical formulae. If these constraints are consistent, this relation can
be completed through the principle of minimal specificity. A plausibility
ordering on S can be obtained by assigning to each state of affairs its highest
possibility level in agreement with the constraints (see Benferhat et al.1997,
1998). A general discussion on the relation between possibility relations and
partial orders on state of affairs is in (Halpern, 1997).

Qualitative possibility relations can be represented by (and only by)

possibility measures ranging on any totally ordered set (especially a finite
one, Dubois 1986). This absolute representation on an ordinal scale is
slightly more expressive than the purely relational one. When the finite set S
is large and generated by a propositional language, qualitative possibility
distributions can be efficiently encoded in possibilistic logic (Dubois, Lang,
Prade, 1994). A possibilistic logic base K is a set of pairs , where is
a Boolean expression and takes on values on an ordinal scale. This pair
encodes the constraint where is the degree of necessity of the
set of models of . Each prioritized formula expresses a necessity-
qualified statement. It is interpreted as the least specific possibility
distribution on interpretations where this statement holds. Thus a prioritized
formula has a fuzzy set of models. The fuzzy intersection of the fuzzy sets of
models of all prioritized formulas in K yields an associated plausibility
ordering on S.

Syntactic deduction from a set of prioritized clauses is achieved by

refutation using an extension of the standard resolution rule, whereby
can be derived from and This rule,
which evaluates the validity of an inferred proposition by the validity of the
weakest premiss, goes back to Theophrastus, a disciple of Aristotle.
Possibilistic logic is an inconsistency-tolerant extension of propositional
logic that provides a natural semantic setting for mechanizing non-
monotonic reasoning (Benferhat, Dubois, Prade, 1998), with a
computational complexity close to that of propositional logic.

The main idea behind qualitative possibility theory is that the state of the
world is by default assumed to be as normal as possible, neglecting less
normal states. In particular, the events accepted as true are those true in all
the most plausible states, namely the ones with positive degrees of necessity
These assumptions lead us to interpret the plausible inference of a
proposition B from another A, under knowledge as follows: B should be
true in all the most plausible states were A is true. It also means that the
agent considers B as an accepted belief in the context A. This kind of
inference is nonmonotonic in the sense that in the presence of additional
information C, B may fail to remain an accepted belief. This is similar to the
fact that a conditional probability may be low even if
is high. The properties of this consequence relation are now well-understood
(Benferhat, Dubois and Prade, 1997).

Decision-theoretic foundations of qualitative possibility were established by

Dubois et al. (2001) in the act-based setting of Savage. Two qualitative
decision rules under uncertainty can be justified: a pessimistic one and an
optimistic one, respectively generalizing Wald's maximin and maximax
criteria under ignorance, refined by accounting for a plausibility ordering on
the state space.

Quantitative possibility theory

The phrase "quantitative possibility" refers to the case when possibility

degrees range in the unit interval. In that case, a precise articulation between
possibility and probability theories is useful to provide an interpretation to
possibility and necessity degrees. Several such interpretations can be
consistently devised (see (Dubois, 2006) for a detailed survey):

• a degree of possibility can be viewed as an upper probability bound

(Walley, 1996; De Cooman and Aeyels, 1999), Especially,
probabilistic inequalities such as Chebychev one, can be interpreted as
defining possibility distributions (Dubois et al., 2004).
• a possibility measure is also a special case of a plausibility function in
the theory of evidence (Shafer, 1987). It is then equivalent to a
consonant random set.
• a possibility distribution can be viewed as a likelihood function
(Dubois, Moral and Prade, 1997).
• Confidence or dispersion intervals are often extracted from statistical
information and are attached a confidence level like 0.95 per cent.
Varying this confidence level yields a family of nested intervals that
can be represented as a possibility distribution (Dubois et al. 2004).
• Following a very different approach, possibility theory can represent
probability distributions with extreme, infinitesimal values (Spohn,
1990).
 • The theory of large deviations in probability theory also handles set-
functions that look like possibility measures (Nguyen and Bouchon-
Meunier, 2003).

There are finally close connections between possibility theory and

idempotent analysis (Kolokoltsov and Maslov, 1997).

Applications

Possibility theory has not been the main framework for engineering
applications of fuzzy sets in the past. However, two directions where it
proved useful can be highlighted:

• Interval analysis has been generalized in the setting of possibility

theory. This is the calculus of fuzzy numbers; see (Dubois, Kerre,
Mesiar and Prade, 2000) for a survey. It is then analogous to random
variable calculations. Finding the potential of possibilistic
representations in computing conservative bounds for probabilistic
calculations is certainly a major challenge.

• Possibility theory also offers a framework for preference modeling in

constraint-directed reasoning. Both prioritized and soft constraints can
be captured by possibility distributions expressing degrees of
feasibility rather than plausibility (Dubois, Fargier and Prade, 1996).

Other applications of possibility theory can be found in fields such as data

analysis (Wolkenhauer, 1998; Tanaka and Guo, 1999), structural learning
(Borgelt et al., 2000), database querying (Bosc and Prade, 1997), diagnosis
(Cayrac et al., 1996), belief revision (Benferhat, Dubois, Prade and
Williams, 2002), argumentation (Amgoud and Prade, 2004), case-based
reasoning (Huellermeier, 2007).

Lastly, possibility theory is also being studied from the point of view of its
relevance in cognitive psychology. Experimental results in cognitive
psychology (Raufaste et al., 2003) suggest that there are situations where
people reason about uncertainty using the rules or possibility theory, rather
than with those of probability theory.
Exploring Extensions of Possibilistic Logic over
G¨odel Logic
Pilar Dellunde1,2, Llu´ıs Godo2, and Enrico
Marchioni2
1 Universitat Aut`onoma de Barcelona,
08193 Bellaterra, Spain
pilar.dellunde@uab.cat
2 IIIA – CSIC,
08193 Bellaterra, Spain
{enrico,pilar,godo}@iiia.csic.es
Abstract. In this paper we present completeness
results of several fuzzy
logics trying to capture different notions of necessity
(in the sense of
Possibility theory) for G¨odel logic formulas. In a first
attempt, based on
different characterizations of necessity measures on
fuzzy sets, a group of
logics, with Kripke style semantics, are built over a
restricted language,
indeed a two level language composed of non-modal
and modal formulas,
the latter moreover not allowing for nested
applications of the modal
operator N. Besides, a full fuzzy modal logic for
graded necessity over
G¨odel logic is also introduced together with an
algebraic semantics, the
class of NG-algebras.
1 Introduction
The most general notion of uncertainty is captured
by monotone set functions
with two natural boundary conditions. In the
literature, these functions have
received several names, like Sugeno measures [24]
or plausibility measures [20].
Many popular uncertainty measures, like
probabilities, upper and lower probabilities,
Dempster-Shafer plausibility and belief functions, or
possibility and
necessity measures, can be therefore seen as
particular classes of Sugeno measures.
In this paper, we specially focus on possibilistic
models of uncertainty. A possibility
measure on a complete Boolean algebra of events U
= (U, !, ", ¬, 0U, 1U)
is a Sugeno measure μ! satisfiying the following "-
decomposition property for
any countable set of indices I
μ!("i"I ui) = sup
i"I
μ!(ui),
while a necessity measure is a Sugeno measure μ!
satisfying the !-decomposition
property
μ!(!i"I ui) = inf
i"I
μ!(ui).
Possibility and necessity are dual classes of
measures, in the sense that if μ!
is a possibility measure, then the mapping μ!(u) = 1
− μ!(¬u) is a necessity
C. Sossai and G. Chemello (Eds.): ECSQARU 2009,
LNAI 5590, pp. 923–934, 2009.
#c Springer-Verlag Berlin Heidelberg 2009
924 P. Dellunde, L. Godo, and E. Marchioni
measure, and vice versa. If U is the power set of a
set X, then any dual pair of
measures (μ!, μ!) on U is induced by a normalized
possibility distribution, i.e. a
mapping ! : X $ [0, 1] such that, supx"X !(x) = 1, and,
for any A % X,
μ!(A) = sup{!(x) | x & A} and μ!(A) = inf{1 − !(x) | x
'& A}.
Appropriate extensions of uncertainty measures on
algebras of events more
general than Boolean algebras need to be considered
in order to represent and
reason about the uncertainty of non-classical events.
For instance, the notion of
(finitely additive) probability has been generalized in
the setting of MV-algebras
by means of the notion of state [22]. In particular,
the well-known Zadeh’s notion
of probability for fuzzy sets (as the expected value of
the membership function)
defines a state over an MV-algebra of fuzzy sets.
States on MV-algebras have
been used in [12] to provide a logical framework for
reasoning about the probability
of (finitely-valued) fuzzy events. Another
generalization of the notion of
probability measure has been recently studied in
depth by defining probabilistic
states over G¨odel algebras [1].
On the other hand, extensions of the notions of
possibility and necessity measures
for fuzzy sets have been proposed under different
forms and used in different
logical systems extending the well-known Dubois-
Lang-Prade’s possibilistic logic
to fuzzy events, see e.g. [7,9,16,3,2,4]. All the
notions of necessity for fuzzy sets
considered in the literature turn out to be of the form
N(A) = infx"U !(x) ( A(x) (*)
where A is a fuzzy set in some domain U, ! : U $ [0,
1] is a possibility distribution
on U and ( is some suitable many-valued implication
function. In
particular, the following notions of necessity have
been discussed:
(1) x (KD y = max(1 − x, y) (Kleene-Dienes);
(2) x (RG y = 1 if x ) y, and x (RG y = 1−x otherwise
(reciprocal of G¨odel);
(3) x (!L y = min(1, 1 − x + y) ("Lukasiewicz).
All these definitions actually extend the above
definition over classical sets or
events.
In the literature different logical formalizations to
reason about such extensions
of the necessity of fuzzy events can be found. In [19],
and later in [17], a full
many-valued modal approach is developed over the
finitely-valued "Lukasiewicz
logic in order to capture the notion of necessity
defined using (KD. A logic
programming approach over G¨odel logic is
investigated in [3] and in [2] by relying
on (KD and (RG, respectively. More recently,
following the approach of
[12], modal-like logics to reason about the necessity
of fuzzy events in the framework
of MV-algebras have been defined in [13], in order to
capture the notion of
necessity defined by (KD and (!L.
The purpose of this paper is to explore different
logical approaches to reason
about the necessity of fuzzy events over G¨odel
algebras. In more concrete terms,
our ultimate aim is to study a full modal expansion of
the [0, 1]-valued G¨odel logic
with a modality N such that the truth-value of a
formula N" (in [0, 1]) can be
Exploring Extensions of Possibilistic Logic over
G¨odel Logic 925
interpreted as the degree of necessity of ", according
to some suitable semantics.
In this context, although this does not extend the
classical possibilistic logic,
it seems also interesting to investigate the notion of
necessity definable from
G¨odel implication, which is the standard fuzzy
interpretation of the implication
connective in G¨odel logic:
(4) x (G y = 1 if x ) y, and x (G y = y otherwise
(G¨odel);
This work is structured as follows. After this
introduction, in Section 2 we
recall a characterization of necessity measures on
fuzzy sets defined by implications
(KD and (RG and provide a (new) characterization of
those defined by
(G. These characterizations are the basis for the
completeness results of several
logics introduced in Section 3 capturing the
corresponding notions of necessity
for G¨odel logic formulas. These logics, with Kripke
style semantics, are built over
a two-level language composed of modal and non-
modal formulas, moreover the
latter not allowing nested applications of the modal
operator. In Section 4 a
full fuzzy modal logic for graded necessity over
G¨odel logic is introduced together
with an algebraic semantics. Finally, in Section 5 we
mention some open
problems and new research goals we plan to address
in the near future.
Due to lack of space, we cannot include preliminaries
on basic notions regarding
G¨odel logic and its expansions with truth-constants,
with Monteiro-Baaz’s
operator # and with an involutive negation, that will
be used throughout the
paper. Instead, the reader is referred to [17,10,11]
for the necessary background.
2 Some Necessity Measures over G¨odel Algebras of
Fuzzy Sets and Their Characterizations
Let X be a (finite) set and let F(X) = [0, 1]X be the set
of fuzzy sets over
X, i.e. the set of functions f : X $ [0, 1]. F(X) can be
regarded as a G¨odel
algebra equipped with the pointwise extensions of
the operations of the standard
G¨odel algebra [0, 1]G. In the following, for each r &
[0, 1], we will denote by r
the constant function r(x) = r for all x & X.
Definition 1. A mapping N : F(X) $ [0, 1] satisfying
(N1) N(!i"Ifi) = infi"I N(fi)
(N2) N(r) = r, for all r & [0, 1]
is called a basic necessity.
If N : F(X) $ [0, 1] is a basic necessity then it is easy
to check that it also
satisfies the following properties:
(i) min(N(f),N(¬Gf)) = 0
(ii) N(f (G g) ) N(f) (G N(g)
The classes of necessity measures based on the
Kleene-Dienes implication
and the reciprocal of G¨odel implication have been
already characterized in the
literature. We do not consider here the one based on
"Lukasiewicz implication.
926 P. Dellunde, L. Godo, and E. Marchioni
Lemma 2 ([3,2]). Let N : F(X) $ [0, 1] be a basic
necessity. Consider the
following properties:
(NKD) N(r (KD f) = r (KD N(f), for all r & [0, 1]
(NRG) N(r (RG f) = r (RG N(f), for all r & [0, 1]
Then, we have:
(1) N satisfies (NKD) iff N(f) = infx"X !(x) (KD f(x)
(2) N satisfies (NRG) iff N(f) = infx"X !(x) (RG f(x)
for some possibility distribution ! : X $ [0, 1] such
that supx"X !(x) = 1.
The characterization of the necessity measures
based on G¨odel implication is
somewhat more complex since it needs to consider
also an associated class of
possibility measures which are not dual in the usual
strong sense.
Definition 3. A mapping $ : F(X) $ [0, 1] satisfying
($1) $("i"Ifi) = supi"I $(fi)
($2) $(r) = r, for all r & [0, 1]
is called a basic possibility.
Note that if $ : F(X) $ [0, 1] is a basic possibility then
it also satisfies
max($(¬f),$(¬¬f)) = 1.
For each x & X, let us denote by x its characteristic
function, i.e. the function
from F(X) such that x(y) = 1 if y = x and x(y) = 0
otherwise. Observe that
each f & F(X) can be written as
f=
!
x"X
x (G f(x) =
"
x"X
x ! f(x).
Therefore, if N and $ are a pair of basic necessity and
possibility on F(X)
respectively, by (N1) and ($1) we have
N(f) = inf
x"X
N(x (G f(x)) and $(f) = sup
x"X
$(x ! f(x)).
Then we obtain the following characterizations.
Proposition 4. Let $ : F(X) $ [0, 1] be a basic
possibility. $ further satisfies
($3) $(f ! r) = min($(f), r), for all r & [0, 1]
iff there exists ! : X $ [0, 1] such that supx"X !(x) = 1
and, for all f & F(X),
$(f) = supx"X min(!(x), f(x)).
Proof. One direction is easy. Conversely, assume that
$ : F(X) $ [0, 1] satisfies
($1) and ($3). Then, taking into account the above
observations, we have
$(f) = sup
x"X
$(x ! f(x)) = sup
x"X
min($(x), f(x)).
Hence, the claim easily follows by defining !(x) = $(x)
.!
Exploring Extensions of Possibilistic Logic over
G¨odel Logic 927
Proposition 5. Let N : F(X) $ [0, 1] be a basic
necessity and $ : F(X) $ [0, 1] be a basic possibility
satisfying ($3). N and $ further satisfy
($N) N(f (G r) = $(f) (G r, for all r & [0, 1]
iff there exists ! : X $ [0, 1] such that supx"X !(x) = 1
and
N(f) = inf
x"X
!(x) (G f(x) and $(f) = sup
x"X
min(!(x), f(x)).
Proof. As for the possibility $, this is already shown
above in Proposition 4. Let
N be defined as N(f) = infx"X !(x) (G f(x) for the
possibility distribution
! : F(X) $ [0, 1] determined by $. We have N(f (G r) =
infx"X(!(x) (G
(f(x) (G r)) = infx"X((!(x) ! f(x)) (G r) = (supx"X !(x) !
f(x)) (G r =
$(f) (G r. Hence, $ and N satisfy ($N).
Conversely, suppose that N and $ satisfy ($N). Then,
using the fact that
$(x) = !(x) for each x & X, we have N(f) = infx"X N(x
(G f(x)) =
infx"X $(x) (G f(x) = infx"X !(x) (G f(x). !
3 Four Complete Logics: The Two-Level Language
Approach
The language of the logics we are going to consider
in this section consists of
two classes of formulas:
(i) The set Fm(V ) of non-modal formulas ", % . . .,
which are formulas of G!(Q)
(G¨odel logic G expanded with Baaz’s projection
connective # and truth
constants r for each rational r & [0, 1]) built from the
set of propositional
variables V = {p1, p2, . . .};
(ii) And the set MFm(V ) of modal formulas &, ' . . .,
built from atomic modal
formulas N", with " & Fm(V ), where N denotes the
modality necessity,
using the connectives from G! and truth constants r
for each rational r & [0, 1]. Notice that nested
modalities are not allowed.
The axioms of the logic NG0 of basic necessity are
the axioms of G!(Q) for
non-modal and modal formulas plus the following
necessity related modal
axioms:
(N1) N(" $ %) $ (N" $ N%)
(N2) N(r) * r, for each r & [0, 1] + Q.
The rules of inference of NG0 are modus ponens (for
modal and non-modal
formulas) and necessitation: from , " infer , N".
It is worth noting that NG0 proves the formula N(" !
%) * (N" ! N%),
which encodes a characteristic property of necessity
measures.
928 P. Dellunde, L. Godo, and E. Marchioni
As for the semantics we consider several classes of
possibilistic Kripke models.
A basic necessity Kripke model is a system M= -W, e,
I. where:
– W is a non-empty set whose elements are called
nodes or worlds,
– e : W × V $ [0, 1] is such that, for each w & W, e(w,
·) : V $ [0, 1]
is an evaluation of propositional variables which is
extended to a G!(Q)-
evaluation of (non-modal) formulas of Fm(V ) in the
usual way.
– For each " & Fm(V ) we define its associated
function ˆ "W : W $ [0, 1],
where ˆ"W(w) = e(w, "). Let #Fm = { ˆ" | " & Fm(V )}
– I : #Fm $ [0, 1] is a basic necessity over #Fm (as a
G-algebra), i.e. it satisfies
(i) I(ˆrW) = r, for all r & [0, 1] + Q
(ii) I(!i"I"ˆiW) = infi"I I( ˆ "iW).
Now, given a modal formula &, the truth value of & in
M = -W, e, I., denoted
0&0M, is inductively defined as follows:
– If & is an atomic modal formula N", then 0N"0M = I(
ˆ"W)
– If & is a non-atomic modal formula, then its truth
value is computed by evaluating
its atomic modal subformulas, and then by using the
truth functions
associated to the G!(Q) connectives occurring in &.
We will denote by N the class of basic necessity
Kripke models. Then, taking
into account that G!(Q)-algebras are locally finite,
following the same approach
of [13] with the necessary modifications, one can
prove the following result.
Theorem 6. NG0 is sound and complete for modal
theories w.r.t. the class N of basic necessity
structures.
Now our aim is to consider extensions of NG0 which
faithfully capture the
three different notions of necessity measure
considered in the previous section.
We start by considering the following additional
axiom:
(NKD) N(r " ") * (r "N"), for each r & [0, 1] + Q.
Let NGKD be the axiomatic extension of NG0 with
(NKD). Then, using Lemma
2, it is easy to prove that indeed NGKD captures the
reasoning about KDnecessity
measures.
Theorem 7. NKD is sound and complete for modal
theories w.r.t. the subclass
NKD of necessity structures M = -W, e, I. such that
the necessity measure I is
defined as, for every " & Fm(V ), I( ˆ"W) = infw"W !
(w) (KD ˆ"W(w) for some
possibility distribution ! : W $ [0, 1] on the set of
possible worlds W.
To capture RG-necessities, we need to expand the
base logic G!(Q) with an
involutive negation 1. This corresponds to the logic
G$(Q), as defined in [10].
So we define NGRG as the axiomatic extension of
NG0 over G$(Q) (instead of
over G!(Q)) with the following axiom:
(NRG) N(1" $ 1 − r) * (1N" $ 1 − r), for each r & [0, 1]
+ Q.
Exploring Extensions of Possibilistic Logic over
G¨odel Logic 929
Then, using again Lemma 2 and the fact that also G$
(Q)-algebras are locally
finite, one can also prove the following result.
Theorem 8. NGRG is sound and complete for modal
theories w.r.t. the subclass
NRG of necessity structures1 M= -W, e, I. such that
the necessity measure I is
defined as, for every " & Fm(V ), I( ˆ"W) = infw"W !
(w) (RG ˆ"W(w) for some
possibility distribution ! : W $ [0, 1] on the set of
possible worlds W.
It is worth pointing out that if we added the Boolean
axiom " " ¬" to the
logics NKD and NRG, both extensions would basically
collapse into the classical
possibilistic logic.
Finally, to define a logic capturing NG-necessities, we
need to expand the
language of NG0 with an additional operator $ to
capture the associated
possibility measures according to Proposition 5.
Therefore we consider the
extended set MFm(V )+ of modal formulas &, ' . . . as
those built from atomic
modal formulas N" and $", with " & Fm(V ), truth-
constants r for each
r & [0, 1] + Q and G! connectives. Then the axioms
of the logic N$G are
those of G!(Q) for non-modal and modal formulas,
plus the following necessity
related modal axioms:
(N1) N(" $ %) $ (N" $ N%)
(N2) N(r) * r,
($1) $(" " %) * ($" " $%)
($2) $(r) * r,
($3) $(" ! r) * ($" ! r)
(N$) N(" $ r) * ($" $ r)
where (N2), ($2), ($3) and (N$) hold for each r & [0,
1] + Q. Inference rules
of N$G are those of G!(Q) and necessitation for N and
$.
Now, we also need to consider expanded Kripke
structures of the form M=
-W, e, I, P., where W and e are as above and the
mappings I, P :$ [0, 1] are such
that, for every " & Fm(V ), I( ˆ"W) = infw"W !(w) (G
ˆ"W(w) and P( ˆ"W) =
supw"W min(!(w), ˆ "W(w)), for some possibility
distribution ! : W $ [0, 1]. Call
NPG the class for such structures. Then, using
Proposition 5 we get the following
result.
Theorem 9. N$G is sound and complete for modal
theories w.r.t. the class
NPG of structures.
4 Possibilistic Necessity G¨odel Logic and Its
Algebraic
Semantics: The Full Modal Approach
The logics defined in the previous section are not
proper modal logics since
the notion of well-formed formula excludes those
formulas with occurrences of
1 With the proviso that the evaluations e of
propositional variables extend to G!(Q)-
evaluations for non-modal formulas and not simply to
G!(Q)-evaluations.
930 P. Dellunde, L. Godo, and E. Marchioni
nested modalities. Our aim in this section is then to
explore a full (fuzzy) modal
approach.
We start as simple as possible by defining a fuzzy
modal logic over G¨odel
propositional logic G to reason about the necessity
degree of G-propositions.
The language of Possibilistic Necessity G¨odel logic,
PNG, is defined as follows:
formulas of PNG are built from the set of G-formulas
using G-connectives and
the operator N. Axioms of PNG are those of G¨odel
logic plus the following
modal axioms:
1. N(" $ %) $ (N" $ N%).
2. N% * NN%.
3. ¬N0.
Deduction rules for PNG are Modus Ponens and
Necessitation for N (from %
derive N%). These axioms and rules define a notion
of proof ,PNG in the usual
way.
Notice that in PNG the Congruence Rule “from " * %
derive N" * N%”
as well as the theorems N1 and N(" ! %) * (N" ! N%)
are derivable. Also
observe that, if we had restricted the Necessitation
Rule only to theorems, we
would have obtained a local consequence relation
(instead of the global one
we have introduced here). For this weaker version of
the logic, the Deduction
Theorem in its usual form would holds, nevertheless
this logic turns out not to
be algebraizable.
Theorem 10. [Deduction Theorem] If T 2{", (} is any
set of PNG-formulas,
then T 2 {"} ,PNG ( iff T ,PNG (" !N") $ (.
Kripke style semantics based on possibilistic
structures (W, e, I) could be also
defined as in Section 3, but now the situation is more
complex due to the fact
that we are dealing with a full modal language.
Moreover, it seems even more
complex to try to get some completeness results
with respect to this semantics
so this is left for future research. This is the reason
why in the rest of the paper
we will turn our attention to the study of an algebraic
semantics, following the
ideas developed in [15,14] for the case of a
probabilistic logic over "Lukasiewicz
logic, and see how far we can go.
We start by defining a suitable class of algebras
which are expansions of G¨odel
algebras with a new unary operator trying to capture
the notion of necessity.
Definition 11. An NG-algebra is a structure (A,N)
where A is a G-algebra
and N : A $ A is a monadic operator such that:
1. N(x ( y) ( (Nx ( Ny) = 1
2. Nx = NNx
3. N1 = 1
The function N is called an internal possibilistic state
on the G-algebra A.
Observe that, so defined, the class of NG-algebras is
a variety. Examples of
internal possibilistic states are the identity function
Id, the # operator and the
Exploring Extensions of Possibilistic Logic over
G¨odel Logic 931
¬¬ operator. The variety of G-algebras can be
considered as a subvariety of NGalgebras,
namely the subvariety obtained by adding the
equation N(x) = x. It
is easy to check, using the definition of NG-algebra
that, for every NG-algebra
(A,N) such that N(A) = A we have N = Id, and that,
given a, b & A, a ) b
implies Na ) Nb.
Definition 12. An NG-filter F on an NG-algebra (A,N) is
a filter on the Galgebra
A with the following property: if a & F, then Na & F.
By an argument analogous to the one in Lemma
2.3.14 of [17], if 1F is the
relation defined by: for every a, b & A, a 1F b iff (a (
b) & F and (b ( a) & F,
then 1F is a congruence on (A,N) and the quotient
algebra (A,N)/ 1F is an
NG-algebra.
Lemma 13. Let F be an NG-filter on an NG-algebra
(A,N). Then, the least
NG-filter containing F as a subset and a given a & A
is
F% = {u & A : 3v & F such that u 4 v 5 a 5 Na}
By Corollary 4.8 of [5], it is easy to check that PNG is
finitely algebraizable and
that the equivalent algebraic semantics of PNG is the
variety of NG-algebras.
As a corollary we obtain the following general
completeness result.
Theorem 14. The logic PNG is strongly complete with
respect to the variety of
NG-algebras. This means that for any set of formulas
) 2 {&}, ) ,PNG & iff,
for all NG-algebra A and for all evaluation e on A, if
e(') = 1A for all ' & ),
then e(&) = 1A.
Observe that it is not possible to prove completeness
with respect to linearly
ordered NG-algebras. Otherwise N(& " ') * (N& "N')
would be a theorem.
Now we prove some satisfiability results of formulas
of the logic PNG.
Formulas of the language of PNG can be seen also as
terms of the language of
NG-algebras. Therefore for the sake of clarity, in the
following proofs we work
with first-order formulas of the language of NG-
algebras proving that they are
satisfiable, if the corresponding formulas of the
language of PNG are satisfiable.
Proposition 15. Let ((x1, . . . , xn) be a PNG-formula. If
( is satisfiable, then
( = 1 is satisfiable in an NG-algebra (B,*), by a
sequence (b1, . . . , bn) of elements
of B such that, for every 0 < i ) n, we have either bi =
1 or *(bi) = 0.
Proof. Let (A,N) be an NG-algebra such that ( = 1 is
satisfiable in (A,N) by
(a1, . . . , an). Without loss of generality we assume
that there is k ) n such that
for every 0 < i ) k, N(ai) '= 0 and for i > k, N(ai) = 0.
Now we build a finite sequence of NG-algebras
(B1, . . . ,Bk) and homomorphisms
(h1, . . . , hk) such that for every 0 < i ) k, ( is
satisfied in Bi by
(c1, . . . , ci−1, hi 6 hi−1 6 · · · 6 h1(ai), . . . , hi 6
hi−1 6 · · · 6 h1(an))
where each ci & {0, 1}. We define only the first
homomorphism, the others can
be introduced analogously. Let F = {x & A : Nx 4
Na1}. So defined, it is easy
932 P. Dellunde, L. Godo, and E. Marchioni
to check that F is an NG-filter. And since, by a
previous assumption, Na1 '= 0,
the filter F is proper. Thus, (A,N)/ 1F is an NG-
algebra. Now let h1 be the
canonical homomorphism from (A,N) to (A,N)/1F ,
and let B1 = (A,N)/1F .
It is easy to check that ( = 1 is satisfied in B1 by
(h1(a1), . . . , h1(an)), that
h1(a1) = 1 and that for i > k, N(h1(ai)) = 0. Finally,
take (B,*) = (Bk, hk 6
· · · 6 h1 6 N). !
Definition 16. An unnested atomic formula of the
language of NG-algebras,
is an atomic formula of one of the following four
forms: x = y, c = y (where c
is a constant c & {0, 1}), Nx = y or F(x) = y (for
some function symbol F of
the language of the G¨odel algebras).
Lemma 17. Let ( be a term of the language of NG-
algebras. Then there is a
set )" of unnested atomic formulas such that, for
every NG-algebra (A,N):
( = 1 is satisfiable in (A,N) iff )" is satisfiable in (A,N).
Proof. It is a direct consequence of Theorem 2.6.1 of
[21]. !
Example: Let ( be the term x1 " N(x2 ( N(x3 ( 0)),
take )" to be the
following set of unnested atomic formulas:
{x1 " y = z,1 = z,Nw = y, (x2 ( v) = w,Nq = v, (x3 (
p) = q,0 = p}
Theorem 18. Let ((x1, . . . , xn) be a PNG-formula. If (
is satisfiable, then ( =
1 is satisfiable in the NG-algebra ([0, 1]G,#) by a
sequence of rational numbers.
Proof. Let (A,*) be an NG-algebra in which ((x1, . . . ,
xn) = 1 is satisfiable by
an n-tuple (a1, . . . , an). Without loss of generality
we may assume that:
– ( is a conjunction of unnested atomic formulas (by
using Lemma 17);
– for every 0 < i ) n, ai '= 0 and ai '= 1 (otherwise we
can work with the
formulas ((xi/1) or ((xi/0), by substituting the
corresponding variables by
the constants 0 or 1);
– for every i, we have *(ai) = 0 (by Proposition 15).
Now we consider the unnested conjuncts of (. For the
sake of simplicity, assume
that there is k ) n such that only in case that 0 < i ) k,
the variable xi has an
occurrence in an unnested atomic formula of the
form Nxi = y. We work now
with the formula + = ((Nxi/0), obtained by
substituting in ( all the occurrences
of Nxi by the constant 0, for every 0 < i ) k.
Observe that, so defined, + is a conjunction of
unnested atomic formulas
in the language of the G-algebras which is satisfied in
(A,*) by (a1, . . . , an).
Therefore, the conjunction +0 = + !
$
0<i'k(xi '= 1) is also satisfied in (A,*) by
(a1, . . . , an) (by our assumptions at the beginning of
this proof). Finally, since
+0 is a formula in the language of the G-algebras, it
is satisfied in [0, 1]G by a
sequence of rational numbers, and thus, by definition
of +0, it is easy to check
that ( is also satisfied in ([0, 1]G,#). !
Exploring Extensions of Possibilistic Logic over
G¨odel Logic 933
5 Future Work
Several issues related to the logic PNG deserve
further investigation. A topic that
is worth studying in depth is the relation between the
algebraic semantics for
the logic PNG (and of some meaningful axiomatic
extensions) and the Kripke
style semantics of the kind used in Section 3. This is
crucial if one wants to
keep as the intended graded semantics of the N
operator one of the possibilistic
necessities of the families described in Section 2.
Actually, the PNG logic might
only capture the logic of basic necessities, and so,
different axioms (and possibly
operators as well) must be added in order to capture
other more specific families
of necessities, somehow related to axioms (NKD),
(NRG), ($3) or the axiom
(N$).
Also as a future task, we aim at studying the
complexity of the sets of satisfiable
formulas for both NGKD, NGRG and N$G. Given the
results in [18],
we conjecture that the problem of checking
satisfiability for those logics is in
PSPACE. As for PNG, notice that from the results in
the above section and the
fact that satisfiability for G! is an NP-complete
problem (easily derivable from
[17]), we immediately obtain that the set of
satisfiable PNG-formulas is in NP.
Acknowledgments. The authors are grateful to the
anonymous referees for their
valuable comments for improving the final version of
this paper. They also acknowledge
partial support from the Spanish projects MULOG2
(TIN2007-68005-
C04) and Agreement Technologies (CONSOLIDER
CSD2007-0022, INGENIO
2010), as well as the ESF Eurocores-
LogICCC/MICINNproject (FFI2008-03126-
E/FILO). Marchioni also acknowledges partial support
of the Juan de la Cierva
Program of the Spanish MICINN.
References
1. Aguzzoli, S., Gerla, B., Marra, V.: De Finetti’s no-
Dutch-book criterion for G¨odel
logic. Studia Logica 90, 25–41 (2008)
2. Alsinet, T.: Logic Programming with Fuzzy
Unification and Imprecise Constants:
Possibilistic Semantics and Automated Deduction.
Monografies de l’Institut
d’Investigaci´o en Intel·lig`encia Artificial, CSIC,
Barcelona (2003)
3. Alsinet, T., Godo, L., Sandri, S.: On the Semantics
and Automated Deduction for
PLFC: a logic of possibilistic uncertainty and
fuzziness. In: Proc. of 15th Conference
on Uncertainty in Artificial Intelligence Conference
UAI 1999, Stokholm, Sweden,
pp. 3–12 (1999)
4. Alsinet, T., Godo, L., Sandri, S.: Two formalisms of
extended possibilistic logic
programming with context-dependent fuzzy
unification: a comparative description.
Electr. Notes Theor. Comput. Sci. 66(5) (2002)
5. Blok, W.J., Pigozzi, D.: Algebraizable logics.
Memoirs of the American Mathematical
Society A.M.S. 396 (1989)
6. Burris, S., Sankappanavar, H.P.: A Course in
Universal Algebra. Graduate texts
in mathematics, vol. 78. Springer, Heidelberg (1981)
7. Dubois, D., Prade, H.: Possibility theory. Plenum
Press, New York (1988)
8. Dubois, D., Prade, H.: Resolution principles in
possibilistic logic. International
Journal of Approximate Reasoning 4(1), 1–21 (1990)
934 P. Dellunde, L. Godo, and E. Marchioni
9. Dubois, D., Lang, J., Prade, H.: Possibilistic logic. In:
Gabbay, et al. (eds.) Handbook
of Logic in Artificial Intelligence and Logic
Programming. Nonmonotonic
Reasoning and Uncertain Reasoning, vol. 3, pp. 439–
513. Oxford University Press,
Oxford (1994)
10. Esteva, F., Godo, L., H´ajek, P., Navara, M.:
Residuated fuzzy logics with an involutive
negation. Archive for Mathematical Logic 39(2), 103–
124 (2000)
11. Esteva, F., Gispert, J., Godo, L., Noguera, C.:
Adding truth-constants to logics
of continuous t-norms: Axiomatization and
completeness results. Fuzzy Sets and
Systems 158, 597–618 (2007)
12. Flaminio, T., Godo, L.: A logic for reasoning about
the probability of fuzzy events.
Fuzzy Sets and Systems 158, 625–638 (2007)
13. Flaminio, T., Godo, L., Marchioni, E.: On the
Logical Formalization of Possibilistic
Counterparts of States over n-Valued "Lukasiewicz
Events. Journal of Logic and
Computation (in press), doi:10.1093/logcom/exp012
14. Flaminio, T., Montagna, F.: MV-algebras with
internal states and probabilistic
fuzzy logics. International Journal of Approximate
Reasoning 50, 138–152 (2009)
15. Flaminio, T., Montagna, F.: An algebraic approach
to states on MV-algebras. In:
Proc. of EUSFLAT 2007, Ostrava, Czech Republic, vol.
II, pp. 201–206 (2007)
16. Godo, L., Vila, L.: Possibilistic temporal reasoning
based on fuzzy temporal constraints.
In: Proceedings of IJCAI 1995, pp. 1916–1922. Morgan
Kaufmann, San
Francisco (1995)
17. H´ajek, P.: Metamathematics of Fuzzy Logic.
Trends in Logic, vol. 4. Kluwer Academic
Publishers, Dordrecht (1998)
18. H´ajek, P.: Complexity of fuzzy probability logics
II. Fuzzy Sets and Systems 158,
2605–2611 (2007)
19. H´ajek, P., Harmancov´a, D., Esteva, F., Garcia,
P., Godo, L.: On Modal Logics
for Qualitative Possibility in a Fuzzy Setting. In: Proc.
of the 94 Uncertainty in
Artificial Intelligence Conference (UAI 1994), pp. 278–
285. Morgan Kaufmann, San
Francisco (1994)
20. Halpern, J.Y.: Reasoning about uncertainty. MIT
Press, Cambridge (2003)
21. Hodges, W.: Model Theory. Encyclopedia of
Mathematics and its Applications,
vol. 42. Cambridge University Press, Cambridge
(1993)
22. Mundici, D.: Averaging the truth-value in
"Lukasiewicz logic. Studia Logica 55(1),
113–127 (1995)
23. Shafer, G.: A mathematical theory of evidence.
Princeton University Press, Princeton
(1976)
24. Sugeno, M.: Theory of Fuzzy Integrals and its
Applications. PhD thesis, Tokyo

Institute of Technology, Tokio, Japan (1974)