Hacker News Reader

Comment by appplication | Hacker News Reader

I’m not a mathematician, so I’d be interested in any perspectives if the two are ultimately so similar in this regard, or is there is something foundationally different about the traditional mathematical approach.

I suppose any proof is ultimately a matter of review and consensus, and potentially would rely on some fundamentally unprovable assumptions at a basic level (e.g. the underlying laws of the universe).

View on HN

What I was referring to is that we don't know if ZFC, for example, is consistent. And it is only consistent if there are unprovable statements. And of course, you can't know which statements are unprovable (because then you would prove its consistency). So, in some sense you're always chasing your own tail. Theorems might be proven true only because ZFC is inconsistent and therefore everything can be proven.

The Lean language is known to be equivalent to ZFC + {existence of n < omega inaccessible cardinals}, which is roughly the common language of mathematics. The actual implementation of the kernel _may_ have bugs, but that's sort of separate from the language definition.

The sentiment I see for the purpose of Lean now and in the near future is for more collaborative mathematics. Theorems and structures are more transparent and concrete (the code is right there). Results are more consumable. The community does not seem to see its main purpose being a way to quell doubts of a paper proof.

It is a fundamentally different manner. Proofs in Lean are formal all the way down. The kernel checks the validity of a proof. This is a finite program checking the validity of the kernel to high degree once is different to checking the validity of every new paper that comes out to high degree.

Yes, the human effort of checking proofs written in Lean is O(1) as you only need to verify the kernel. The human effort of checking proofs written traditionally is O(N). Of course writing proofs in Lean is much more difficult (for now) and, checking the proof is not the only interesting part. Other mathematicians also want to understand the proof, and this will always be O(n).

Also, while it's true that bugs in mathlib cannot cause lean to accept invalid proofs as long as the kernel is correct, bugs in mathlib can still be annoying (I have only used lean briefly and I got stuck for a long time on a buggy tactic that was silently messing up my proof state).

My personal belief is that translating an informal proof from a paper to a proof in Lean, given that the prerequisites are already formalized, is an O(n) operation where n is the length of the proof. It's just translation. Right now the constant is something like 10-100, but I think with more work on matlib and AI assistance it can be brought down to maybe 1-3, meaning formalising proofs right as you write them becomes feasible. In fact the constant may go below 1 as formalization in Lean can prevent unseen errors from slowing your progress.

Reply on news.ycombinator.com