• Bustedknuckles@lemmy.world
    link
    fedilink
    English
    arrow-up
    110
    arrow-down
    3
    ·
    1 day ago

    Isn’t this “Independent mathematicians who also used Lean software to check the proof that the author wrote/checked with Lean”?

    It could still be valid of course, but I thought the proof was like 250000 pages and has only been looked at by the same automated framework. This is out of my wheelhouse and I don’t have a dog in the fight, but I’d still like some expert human eyes on it

    • bss03@infosec.pub
      link
      fedilink
      English
      arrow-up
      27
      ·
      21 hours ago

      They used 2 checkers that are maintained and developed independently of the Lean kernel. But they did use Lean formats/syntax. They’ve also done some manual verification, though there might be more to do. The proof did use the statement if RH directly from the well established Mathlib.

      I’ve also seen claims that the proof has some weird things in it that might be hints that it is finding holes in the theory Lean models.

      Honestly, even if it ends up not being a proof of RH, it will tell us something about computable / constructive maths.

      • redjard@reddthat.com
        link
        fedilink
        English
        arrow-up
        8
        ·
        edit-2
        15 hours ago

        It is 2028. Another lean CVE was registered; GRH was proven again. I am excited that yet another logic bug will be fixed.


        Scott Optimized (Aaronson) declared the proof likely correct on friday, before this news, so I’d take their word for it.

        Edit: Update from Scott Optimized:

        [Update (Nov 21): The proof has now been verified independently by the Lean community, using con-leche, a piece of Lean that itself is formally verified all the way down to the axioms of set theory. This is a much higher degree of verification than a typical human paper accepted to a math journal. The proof is correct.]

    • gandalf_der_13te@feddit.org
      link
      fedilink
      English
      arrow-up
      14
      arrow-down
      2
      ·
      19 hours ago

      i get what you’re saying but these automated proof-checking engines are solid. they don’t make mistakes. it’s like saying “ok so this compiler compiled 10 MB of C++ code into machine binary code, but i’d still like some human to confirm that it compiled correctly”. this is just not reasonable from a cost/benefit perspective.

      what is interesing though is to look at the ideas nevertheless, but not to check whether every step was done correctly, but to figure out the important ideas and concepts used during the proof. maybe they can be re-used somewhere else?

      • redjard@reddthat.com
        link
        fedilink
        English
        arrow-up
        7
        ·
        15 hours ago

        There have been bugs in lean before. I think there still are known issues now. I recall specifically an “unsound type system”.

        This reverification now though seems proven correct fully, according to Scott Aaronson

    • TapedeckNinja@lemmy.worldOP
      link
      fedilink
      English
      arrow-up
      80
      arrow-down
      5
      ·
      edit-2
      1 day ago

      There are a bunch of ways that a proof written in Lean can be invalid.

      The most obvious is that the problem statement is invalid or incomplete. This was not really a concern here as the proof used Mathlib’s built-in definitions rather than restating the problem, but nonetheless it has been verified.

      Beyond that Lean proofs can contain custom axioms/assumptions, shortcuts, tricks, exploits. And there’s the possibility that the Lean compiler used in the proof had a bug.

      Anyway all that has been tested and verified. The problem statement is valid. The proof doesn’t contain any tricks or assumptions. It compiles on every major Lean kernel.

      Basically: it is almost certainly mathematically true. The next step is humans actually understanding it.

      • Telorand@reddthat.com
        link
        fedilink
        English
        arrow-up
        24
        arrow-down
        2
        ·
        edit-2
        20 hours ago

        The next step is humans actually understanding it.

        So what we got here is “42.”

        • Lettuce eat lettuce@lemmy.ml
          link
          fedilink
          English
          arrow-up
          10
          arrow-down
          1
          ·
          17 hours ago

          “Claude, explain this proof to me in a way that I can perfectly comprehend in totality without requiring me to understand any mathematics beyond high school. Make no mistakes.”

          • brianary@piefed.zip
            link
            fedilink
            English
            arrow-up
            3
            ·
            3 hours ago

            I’d settle for a human mathematician understanding it. What good is a proof that only other software understands? Am I missing something? Does it give us some sort of useful or more widely applicable formula or constant?

    • wewbull@feddit.uk
      link
      fedilink
      English
      arrow-up
      39
      arrow-down
      11
      ·
      1 day ago

      It’s 250,000 pages of explanation for 17 million (I think) lines of lean code.

      It’s incomprehensible, and therefore probably of little value.

      • TapedeckNinja@lemmy.worldOP
        link
        fedilink
        English
        arrow-up
        11
        arrow-down
        6
        ·
        20 hours ago

        It’s incomprehensible as a whole but not in pieces.

        It will be simplified, refactored, maybe rebuilt from the ground up. It won’t be incomprehensible for long.

        The last I looked the size had already been reduced by 25% based on community efforts.

        Kind of fun watching the Lean Prover Zulip chat play with it.

    • HexesofVexes@lemmy.world
      link
      fedilink
      English
      arrow-up
      3
      arrow-down
      5
      ·
      10 hours ago

      Truth is proof - different people have different burdens of proof. What use a proof that relies on faith in a system rather than mental construction?