[HN Gopher] A new, faster algorithm for the group isomorphism pr...
       ___________________________________________________________________
        
       A new, faster algorithm for the group isomorphism problem
        
       Author : theafh
       Score  : 88 points
       Date   : 2023-06-23 13:46 UTC (9 hours ago)
        
 (HTM) web link (www.quantamagazine.org)
 (TXT) w3m dump (www.quantamagazine.org)
        
       | kjhughes wrote:
       | The "major algorithmic goal" is better time complexity for group
       | isomorphism.
       | 
       | Xiaorui Sun improved on Robert Tarjan's (50-year-old) result,
       | n^(log n)
       | 
       | achieving                 n^((log n)^(5/6))
       | 
       | for certain types of groups that appear to be easier to compare
       | but had been defying improvement attempts for decades.
       | 
       | ---
       | 
       | Anyone see intuitively how the 5 and 6 come into the improved
       | complexity -- why those two particular constants?
        
         | scythe wrote:
         | I can't say for "intuitive", but the introduction to the paper
         | involves breaking p-groups -- groups having p^k members where p
         | is a prime -- into the cases k > lg(p)^5 and k <= lg(p)^5,
         | which likely explains why the number 5 has crashed the party.
         | (Here lg() denotes the log base 2)
        
           | schoen wrote:
           | I've wished for a long time that mathematicians would bring
           | back the notation "ld" (logarithmus dualis), which I find
           | very elegant, for log2. The "ln" is already from Latin
           | (that's why it's ln instead of nl), I think established by
           | Gauss or something.
        
             | _a_a_a_ wrote:
             | Name's new to me.
             | 
             | https://en.wikipedia.org/wiki/Binary_logarithm
        
         | klyrs wrote:
         | Strassen's algorithm is, I think, one of the easier ones to
         | understand along these lines. It uses a recursive divide and
         | conquer strategy. The naive algorithm does 1 (N x N)
         | multiplication by adding the results of 8 multiplications of
         | size (N/2 x N/2), and (sweeping details under the rug here)
         | that takes time N^log2(8) = N^3. The improved algorithm uses 7
         | multiplications of size (N/2 x N/2) and takes N^log2(7). For
         | those details, the wiki page does a nice job.
         | 
         | https://en.m.wikipedia.org/wiki/Strassen_algorithm
         | 
         | So intuitively speaking: when I see a logarithm like that with
         | an integer ratio, I think "oh, somebody saved 1/6th of the work
         | in a recursive algorithm."
        
           | dataflow wrote:
           | This is a power of a logarithm though? Whereas you're talking
           | about the base and its argument.
        
             | klyrs wrote:
             | No, Strassen's isn't an exact match in this case. Just an
             | example how strange exponents can crop up in complexity
             | analysis. I was aiming to satisfy the original comment's
             | ask for "intuition" rather than "read the entire paper and
             | digest it in time to make an unassailable ELI5 while
             | waiting for the bus."
        
       | gilleain wrote:
       | "Sun's method takes an approach called individualization and
       | refinement..."
       | 
       | Which is what many standard methods for graph isomorphism use, if
       | we are talking about the same thing.
        
       | boxfire wrote:
       | Scroll scroll scroll and...
       | 
       | https://arxiv.org/abs/2303.15412
        
         | matthewdgreen wrote:
         | > Scroll scroll scroll and...
         | 
         | The new result (with explicit link to the arXiv as well as the
         | author's home page) is linked in _the fourth paragraph_ , and
         | it only appears that far down because the first three
         | paragraphs very efficiently provide background on the problem
         | and recent results. The whole thing is an excellent general-
         | audiences article explaining a complex theoretical result with
         | illustrations and accessible links to all the relevant source
         | material. I'm really glad we have Quanta and I'm not sure how
         | this reporting could have been handled better.
        
           | bluepod4 wrote:
           | Exactly.
           | 
           | EDIT: I was going to write a snark-ish comment that someone
           | would eventually complain about the post title only to
           | refresh a second later and see that someone changed the title
           | already.
           | 
           | "Major algorithmic goal" was completely fine and says quite a
           | few different things than "new, faster".
           | 
           | Also, according to HN's guidelines, the post title shouldn't
           | have changed in this situation:
           | 
           | "Otherwise please use the original title, unless it is
           | misleading or linkbait; don't editorialize."
           | 
           | The original title was not misleading or linkbait.
        
           | ouid wrote:
           | [dead]
        
       | samsquire wrote:
       | Could this be used for computer program equivalence?
        
         | ouid wrote:
         | [dead]
        
         | henrydark wrote:
         | Don't think so, doesn't computer program equivalence require
         | solving the halting problem and undecidable problems?
         | 
         | For example, consider the empty program, and the program that
         | print "hello, world!" if an undecidable condition is met. I
         | think checking if these two programs are equivalent is
         | undecidable
        
           | maweki wrote:
           | For these two programs it's easy to check that they are not
           | equivalent.
           | 
           | The magic words are "in general" and in general, program
           | equivalence, as well as all other interesting program
           | properties, are in general undecidable (Rice's Theorem).
        
       ___________________________________________________________________
       (page generated 2023-06-23 23:01 UTC)