Metamath Proof Explorer


Table of Contents - 21.19.4. Stronger axioms of regularity

This section contains some experiments related to the Axiom of Regularity ax-reg. As written, ax-reg cannot guarantee that all sets are well-founded unless we further assume ax-inf / ax-inf2; in particular, ax-reg alone is insufficient to assert that every set has a transitive closure (tz9.1), even though this is true among the hereditarily finite sets.

The underlying cause of this issue is that ax-reg requires a witness set to detect non-well-foundedness, but if all sets are hereditarily finite, then there may be no such witness set for an infinite descending -chain. The question is, how can we strengthen ax-reg so that we get a true "Axiom of Foundation" even in the absence of ax-inf / ax-inf2 (e.g., so that we can prove unir1)?

There are a few possible solutions. First, we can directly strengthen ax-reg into ax-regs, which asserts that every class has an -minimal element. Second, we can keep ax-reg and add ax-tco, which asserts that every set is a member of a transitive set. Third, we can replace ax-reg with a set-induction axiom mh-setind. Fourth, we can take unir1 as an axiom and derive everything from that. This list is far from exhaustive.

In this section, we prove that these four listed principles are equivalent. We see that ax-regs implies the other three principles: ax-reg + ax-tco via axreg + tz9.1regs, mh-setind via setindregs, and unir1 via unir1regs. So we just have to show that ax-regs is implied by each of the other three.

Some questions: When expanded to primitives, what is the shortest single axiom equivalent to these, over ZF minus ax-reg and ax-inf / ax-inf2? One candidate is mh-setind, with 19 primitives. What is the shortest single axiom not using any wff variables? The conjunction of ax-reg + ax-tco, expanded using mh-regprimbi and slightly simplified, comes out to 42 primitives. Can we do better?

  1. mh-setind
  2. mh-setindnd
  3. regsfromregtco
  4. regsfromsetind
  5. regsfromunir1