# Fuzzy logic and topos theory

Why doesn't one develop fuzzy logic by extending topos theory, by simply extending the subobject classifier $\Omega$ to the unit interval [0,1]? Have people done that?

• The topos of sheaves on $[0, 1]$ has as its subobject classifier the lattice of all open subsets of $[0, 1]$. But as far as I know this is not what is meant by fuzzy logic. Can you please elaborate on your idea? Aug 6, 2011 at 13:41
• Also, see this MathOverflow thread. Aug 6, 2011 at 13:45
• Zhen: Thanks for your reply. I'll need some time to learn all this. I don't really have a detailed idea, except that I use a new form of fuzzy logic different from Zadeh's in my A.I. research. Basically my category of fuzziness is what people would call "degrees" -- ie, a degree is a number in [0,1]. And probabilities can be distributed over degrees, resulting in fuzzy-probabilistic logic. I'm just wondering, could the concept of "degree" coincide with the topos with \Omega = [0,1]? Aug 8, 2011 at 5:08
• PS: I think a topos with $\Omega$ = [0,1] can be regarded as a fuzzy set (or predicate). For example, "tallness" is a fuzzy predicate, mapping objects like "john" to, say, 0.7. This is exactly Zadeh's definition of fuzzy sets. But my interpretation of it is different -- it is regarded as a fuzzy predicate returning a degree of something, NOT set membership. Thus, Barr's worry about fuzzy set equality need not bother us -- we're just talking about the equality of 2 degrees, which can be crisp. You get my idea? Aug 8, 2011 at 5:37
• Oh, I realize there is a conceptual problem... topos is a generalization of set, but if I take the interpretation of fuzziness as "degree", then fuzziness is no longer a set or similar to sets. So my idea of making topos fuzzy may be unsound. I just think that fuzzy sets are not the right way to formulate fuzzy logic(s). Aug 8, 2011 at 5:55

Here are some papers, which I have not read in detail. I'm somewhat troubled that there aren't any Math Reviews for most of them, and I certainly can't vouch for their correctness. (I know nothing of fuzzy logic, and only a miniscule bit of topos theory; I'm merely an enthusiast.)

• Pitts, Andrew M. (1982). Fuzzy sets do not form a topos. In: Fuzzy Sets and Systems 8.1, pp. 101–104. DOI: 10.1016/0165-0114(82)90034-3

Let $H$ be a complete Heyting algebra. The main result is that the category of $H$-valued fuzzy sets $\textrm{Fuz}(H)$ defined by Eytan is a topos if and only if $H$ is a Boolean algebra. Apparently, the problem is that $H$-valued fuzzy sets are insufficiently ‘fuzzified’ (Pitt's word!) to be well-behaved enough to form a topos. To be precise, $H$-valued fuzzy sets ‘fuzzify’ only the $\in$ predicate, while $H$-valued sets ‘fuzzify’ both $=$ and $\in$. It can be shown that the category of $H$-valued sets is equivalent to the category of sheaves over $H$, and hence do form a topos $H\textrm{-Set}$. Pitt shows that $H$-valued fuzzy sets are equivalent to a certain full subcategory of $H\textrm{-Set}$

• Stout, Lawrence Neff (1984). Topoi and categories of fuzzy sets. In: Fuzzy Sets and Systems 12.2 pp. 169–184. DOI: 10.1016/0165-0114(84)90036-8

Let $L$ be a completely distributive lattice. (Then it is also a complete Heyting algebra $H$.) The paper starts by showing that Goguen's characterisation of categories of sets having ‘fuzzy boundary with fuzziness measured in $L$’ $\textrm{Set}(L)$ are essentially incompatible with the axioms of elementary toposes: $\textrm{Set}(L)$ is a topos if and only if $L$ is the trivial lattice, in which case the category is the usual category of sets. As above, Goguen suggested that a ‘nicer category may result by taking maps which are fuzzy as well’; apparently this is what Eytan did with his $\textrm{Fuz}(H)$.

Stout continues with a brief examination of the internal logic of $\textrm{Set}(H)$, $\textrm{Fuz}(H)$, $\textrm{Sh}(H)$, noting that the internal logic of these categories and the usual fuzzy logic operations do not necessarily coincide, and $\textrm{Set}^{H^{\textrm{op}}}$ and concludes with some results about change-of-base for these categories.

• Barr, Michael (1986) Fuzzy set theory and topos theory. In: Canad. Math. Bull. 29.4, pp.501–508. DOI: 10.4153/CMB-1986-079-9.

There is a Math Review for this paper: MR860861.

Let $L$ be a locale. Like the above two papers, the main contention is that $L$-valued fuzzy sets are not fuzzy enough for their categories to be toposes. Barr describes one generalisation and shows that it is equivalent to a sheaf on a locale $L^+$, obtained by extending $L$ with a new bottom element, and also shows that Eytan's $\textrm{Fuz}(L)$ is equivalent to a certain subcategory of $\textrm{Sh}(L^+)$.

• Stout, Lawrence Neff (1991). A survey of Fuzzy Set and topos theory. In: Fuzzy Sets and Systems 42.1, pp. 3–14. DOI: 10.1016/0165-0114(91)90085-5.

I'll just quote the paper directly: ‘Our survey of the various categorical formulations of Fuzzy Set theory and a variety of kinds of topoi will suggest that a synthesis of the two fields, involving some refinement of the notion of a category of fuzzy sets to give something weaker than a topos but with a richer logic, is desirable.

[...]

‘Throughout this process we should remember that the object is to find a proper foundation for Fuzzy Set theory, so that the original interval-valued fuzzy sets should at least be embeddable into our structure. The choice of definition should be based on what is natural, what is powerful, and what gives an elegant and complete foundation for fuzzy mathematics.’

Fuzzy set theory is a mathematical model of vagueness, which is assumed to be a basic property of our world. In different words, it is assumed that we live in a world with vague objects where nothing is actually crisp. Ordinary mathematics assumes that everything is crisp. Thus I find it at least strange to try to "embed" mathematics of vagueness into universes of crisp mathematics. So I do not really understand what a topos of fuzzy sets would be? Arrows should connect objects up to some degree but this does not happen when one deals with say categories of fuzzy sets. Since we have a new set theory, which was also formalized, we need a new catetory theory and why not a new type theory.