[HN Gopher] Almost all Collatz orbits attain almost bounded values
       ___________________________________________________________________
        
       Almost all Collatz orbits attain almost bounded values
        
       Author : measurablefunc
       Score  : 91 points
       Date   : 2025-11-17 20:28 UTC (6 days ago)
        
 (HTM) web link (mathvideos.org)
 (TXT) w3m dump (mathvideos.org)
        
       | throwaway81523 wrote:
       | youtube link https://youtu.be/k-dtx8s2ehM
        
       | tux3 wrote:
       | The paper (2019): https://arxiv.org/abs/1909.03562
        
       | noduerme wrote:
       | Does this have some significance for back propagation or
       | something, or is it just an interesting trick of arithmetic?
       | //not that it needs to have a technical use, it's still neat.
        
         | huhtenberg wrote:
         | Hailstone numbers has been a popular subject in computing
         | circles since forever. Not much practical application, just a
         | very simple, but curious construct.
        
         | robot-wrangler wrote:
         | Collatz, busy-beavers, and algorithmic information theory are
         | all related. To the extent they offer insight into the
         | sparseness or density of irreducible complexity in the space of
         | all computation.. this has many implications for what can be
         | computed efficiently, what can be learned efficiently, program-
         | synthesis, what can be analyzed "at a distance" without just
         | trying it and potentially needing to wait forever, etc.
         | 
         | Whether it will say anything very significant _practically_ or
         | only philosophically is a different question. Maybe it is
         | something like the discovery of transcendentals.. finding out
         | that most of the number line won 't have a tidy algebraic
         | closed-form isn't exactly a make-or-break deal for the program
         | of mathematics itself, and it also doesn't matter much to
         | people who are doing engineering
        
         | ur-whale wrote:
         | > Does this have some significance for back propagation or
         | something
         | 
         | You're right!
         | 
         | What could darn possibly matter these days, in the whole
         | entirety of the realm Mathematics, if it does now somehow have
         | a measurable impact on backprop ?
        
           | WhyOhWhyQ wrote:
           | This guy's kid hits a homerun in the little league game and
           | he bemoans that the kid is wasting his talents not working on
           | backprop.
        
         | phyzome wrote:
         | Crazy how AI has infected every conversation these days.
        
       | tromp wrote:
       | The closely related function Col' which also divides 3n+1 by 2 in
       | the odd case, is concisely represented by the 65-bit lambda
       | calculus term l1(lll31(ll2(421)))(ll1)1(ll1) operating on Church
       | numerals [1]. It starts from the pair of numbers n and 0 and then
       | performs n iterations of swapping the numbers after incrementing
       | the first. Its lambda diagram is
       | +---------------+--         | +---------- - | -         |
       | +-----+---- + | +         | +-+---+---- | | |         | +-+
       | +-+-+-- | | |         |   | +-+-+-+ | | |         |   | | +-+ | |
       | | |         |   | |   +-+ | | |         |   | +---+   | | |
       | |   +-+       | | |         +---+         | | |
       | +---------+ | |                       +-+ |
       | +-+
       | 
       | [1]
       | https://github.com/tromp/AIT/blob/master/fast_growing_and_co...
        
         | measurablefunc wrote:
         | Does it terminate for all n?
        
           | tromp wrote:
           | Yes; the Col' function trivially terminates for all n, since
           | Col n' is just                   (if odd n then 3*n+1 else n)
           | `div` 2
        
             | measurablefunc wrote:
             | That's good then. It shouldn't difficult to bootstrap that
             | to the full Collatz conjecture.
        
               | tromp wrote:
               | Since the Collatz conjecture is not (known to be)
               | finitely refutable, we cannot encode it as a program
               | whose termination decides the conjecture. If the
               | existence of diverging orbits were disproven though, then
               | we could.
        
               | measurablefunc wrote:
               | That's a good point & also surprising that such a simple
               | dynamical process can not be proven one way or the other
               | to be an instance of a terminating or non-terminating
               | computation.
        
         | im3w1l wrote:
         | I thought that diagram was giving me crazy strong synaesthesia,
         | but turns out it was subpixel rendering and it really did have
         | color.
        
       | anonymous2024 wrote:
       | Someone has also noticed another curiosity: The number of bits of
       | the biggest number (in binary notation) in a path is less than
       | the number of bits of the initial number (in binary notation) * 3
       | + 1
        
         | tromp wrote:
         | If that were true, then every number would lead to a cycle
         | (possibly a different one from 1-4-2). It would make the
         | decision problem (whether a given initial number leads to 1)
         | decidable, and the Collatz conjecture finitely refutable.
        
         | wizzwizz4 wrote:
         | This isn't true. Take 9_A = 1001_2. 28_A = 11100_2, which is 5
         | bits long (3 set). The biggest number in this path is 52_A =
         | 110100_2, which is 6 bits long (3 set). 5 [?] 6, and 3 [?] 3:
         | neither of my interpretations of your statement holds.
        
           | taberiand wrote:
           | Not to say their statement is true but I don't see any reason
           | to count the initial zeroes.
           | 
           | 11100 == 111 == 11100000000, in terms of the next odd
           | iteration
           | 
           | Even numbers don't really count in the process surely? All
           | collatz does is essentially ignore those zeroes
        
             | wizzwizz4 wrote:
             | Valid point: it depends what invariants you're trying to
             | construct. Considering only the odd elements of the
             | sequence does yield a slightly different set of insights
             | compared to other approaches. (9 / 28 / 52 still describe a
             | counterexample to the proposed invariant, even in this
             | scheme.)
        
           | fhars wrote:
           | There is another interpretation, reading "bits" as "set bits"
           | and assuming that textual description (especially the
           | operator "of the") has a higher precedence than
           | multiplication, then your initial number is 9 with 2 bits
           | set, and the largest number is 52 with 3 bits set, and 3 < 2
           | * 3 + 1 = 7.
        
           | anonymous2024 wrote:
           | What I understood was: 9_A = 1001_2 needs 4 bits, set or not
           | set as the minimum length of the binary representation. 52_A
           | = 110100_2 needs 6 bits 6 bits is less than 4*3+1=13 bits
        
             | wizzwizz4 wrote:
             | At that point, the conjecture's just numerology: 27 takes 5
             | bits, and 9232 takes 14 bits (two shy of 3x5+1 = 16). 27 is
             | the peak of the _average_ ratio between start and maximum,
             | because the +1s are so significant when the numbers are
             | small: past that point, we 're relying on _extreme_ outlier
             | behaviour to get each new high-score. Those only start
             | showing up often enough to matter once we get into the
             | thousands.
             | 
             | Plugging in values from OEIS A006884, it looks like the
             | maximum ratio between the maximum and starting values goes
             | down until around 4255, then picks up again, gradually
             | increasing from there. Eyeballing the growth rate, I
             | suspect there's a counterexample to _this_ interpretation
             | somewhere before 10^1000. (Does anyone have an element of
             | A006884 greater than 2358909599867980429759? That 's 140
             | bits maximum to 71 bits starting.)
        
         | littlestymaar wrote:
         | Do you have a source for that?
        
       | exomonk wrote:
       | In case people don't know the Collatz Conjecture, It's based on a
       | simple rule: take any natural number n. If it is even, divide it
       | by 2. If it is odd, multiply by 3 and add 1. The conjecture
       | states that no matter what number you start with, you will
       | eventually reach the number 1.
       | 
       | It would seem simple, but many simple iterative calculations get
       | us to Turing machine territory regarding computability.
        
       | keepamovin wrote:
       | It's like gravity, collisions, and interstellar objects. Some
       | starting points escape and go on ... forever.
       | 
       | Some kind of structure there that Collatz probing is sketching
        
         | apetresc wrote:
         | Except the Collatz Conjecture is almost certainly true, and
         | there are no starting points that go on forever. Or did I miss
         | the point of the analogy?
        
           | keepamovin wrote:
           | No you didn't misunderstand, I think. I thought there were
           | some values that went on forever!
           | 
           |  _edit:_ however we could consider the weaker definition of
           | "forever", and consider there are some outliers that go on
           | "for a long time" per post title, probing structure with
           | these loops and spokes. :D
        
             | sfblah wrote:
             | The conjecture is that there are no values that go on
             | forever, but it's as yet unproven.
        
       | madars wrote:
       | Good background reading/watching - Terence Tao's "The Notorious
       | Collatz conjecture" talk.
       | https://www.youtube.com/watch?v=X2p5eMWyaFs Slides:
       | https://terrytao.wordpress.com/wp-content/uploads/2020/02/co...
       | 
       | I especially like how he highlights that Collatz conjecture shows
       | that a simple dynamical system can have amazingly complex
       | behavior; also 3n-1 variant has two known cycles - so "any proof
       | of the Collatz conjecture must at some point use a property of
       | the 3n+1 map that is not shared by the 3n-1 map." And this
       | property can't be too general either - questions about FRACTRAN
       | programs (of which Collatz conjecture is a special case) can
       | encode the halting problem.
       | 
       | If you haven't seen it, FRACTRAN itself is amazing -
       | https://www.cs.unc.edu/~stotts/COMP210-s23/madMath/Conway87....
       | and the paper is pure joy to read.
        
       | Xcelerate wrote:
       | As a non-mathematician, I'm confused why so many people think the
       | conjecture (whether true or false) is provable within PA. To me,
       | it seems like something that would be very nicely just right
       | outside the boundary of PA's capability, sort of like how proving
       | all Goodstein sequences terminate requires transfinite induction
       | up to e_0. Add that to the fact that the Collatz Conjecture seems
       | to fall in the same "category" of problem as the Turing machines
       | that the Busy Beaver project is having a hard time proving non-
       | halting behavior of, and the heuristic arguments all seem to
       | point to: Collatz is independent of PA.
       | 
       | But I'm interested in hearing the counterarguments that Collatz
       | likely _is_ provable within PA and why this would be the case.
        
       ___________________________________________________________________
       (page generated 2025-11-23 23:01 UTC)