Let $A,B$ be subsets of a set $X$. Then there is a largest subset $C \subseteq X$ such that $C \cap A \subseteq B$. Explicitly, we have $C = \{x \in X : x \in A \Rightarrow x \in B\} = (X \setminus A) \cup B$. Does $C$ have a name? I would call $C$ the set quotient and denote it by $(B:A)$. Namely, there is an analogy$^1$ to ideal quotients: If $A,B$ are ideals of a commutative ring $R$, then there is a largest ideal $C$ such that $C \cdot A \subseteq B$, namely $C=(B:A) = \{x \in R : x \cdot A \subseteq B\}$.

$^1$ Actually it is more than just analogy. Both quotients are internal homs in monoidal preorders, the one being $(\wp(X),\subseteq,\cap)$ and the other one $(\mathrm{Id}(R),\subseteq,*)$.

  • 3
    $\begingroup$ It has many names: Heyting implication, relative pseudocomplement... $\endgroup$ – Zhen Lin Oct 31 '13 at 9:25
  • $\begingroup$ $(B:A)$ isn't good notation, because it omits the dependence on $X$. The notation $X\!\setminus\!\! A\cup B$ has the advantage that everyone can understand it immediately (add parentheses if you will). $\endgroup$ – John Bentin Oct 31 '13 at 12:17
  • $\begingroup$ @John: Since I regard $A,B$ as objects of the partial order of subsets of $X$, this $X$ implicitly belongs to the datum. If $X \subseteq X'$, then (for a category theorist) $A$ isn't really a subset of $X'$, but rather there is a functor $\wp(X) \to \wp(X')$ and $A$ shouldn't be confused with its image (of course set theory is full of such confusions). $\endgroup$ – Martin Brandenburg Nov 1 '13 at 10:32
  • $\begingroup$ @John: Besides, if $A,B$ are ideals of a ring $R$, then $(B:A)$ is a univerally accepted notation for the ideal quotient. Note that also here $R$ belongs implicitly to the datum. If $R \subseteq R'$ is a ring extension with $R'A \subseteq A$ (likewise $B$), then of course $A,B$ can be "seen" as ideals of $R'$ and in the partial order of ideals of $R'$ we can form $(B:A)$. $\endgroup$ – Martin Brandenburg Nov 1 '13 at 10:38
  • $\begingroup$ @John: This confusing abuse of notation can be avoided if we really stick to the general rule (which I try to promote here on math.SE (see for example math.stackexchange.com/questions/389328) and other places) that different categories should be seen as disjoint, and that forgetful functors shouldn't be seen as identities. This clarifies and organizes a lot of mathematics! Objects of mathematics shouldn't "swim" between different categories, but rather each object lies in a distinguished category (and will never leave it). $\endgroup$ – Martin Brandenburg Nov 1 '13 at 10:38

Let $L$ be a lattice that is cartesian closed as a category. Then $L$ is a Heyting algebra, and the operation you describe is sometimes called the Heyting implication or relative pseudocomplement. If $L$ is moreover a boolean algebra then the Heyting implication $a \to b$ is given by the classical formula $(\lnot a) \vee b$, as you described.

  • 2
    $\begingroup$ An excellent book covering these issues completely in the first two chapters is "Stone Spaces" by Johnstone. $\endgroup$ – Ittay Weiss Oct 31 '13 at 10:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.