LLL improvements

15 views
Skip to first unread message

Fredrik Johansson

unread,
Aug 21, 2025, 11:19:15 AMAug 21
to flint-devel
Hi all,

I have recently improved the multiprecision code in fmpz_lll, mainly by using nfloat instead of mpf and mpfr and in some cases using arb instead of fmpq.

As a result, factoring high-degree polynomials in Z[x] with many local factors is now often ~5x faster, and finding integer relations in high dimension is often ~10x faster (one example is 60x faster).


Our LLL code can certainly be optimized a lot more (there are some ideas in the blog post), but it's a big project.

Fredrik

Bill Hart

unread,
Aug 21, 2025, 12:02:31 PMAug 21
to flint...@googlegroups.com
This is really nice work! It's also written up very well too.

The original code probably represents nearly a year of work between
Andy and I (to say nothing of the original work of people whose work
we were building on). It's really fantastic to see it taken seriously
and live on with such impressive speedups!

Bill.
> --
>
> ---
> You received this message because you are subscribed to the Google Groups "flint-devel" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to flint-devel...@googlegroups.com.
> To view this discussion, visit https://groups.google.com/d/msgid/flint-devel/CAJdUXTK53JA_89kreKzfwXub9Sa%3Drq7nPymHfpcpQ2X%3DEFSWmA%40mail.gmail.com.

Jeffrey Sarnoff

unread,
Aug 21, 2025, 9:29:58 PMAug 21
to flint...@googlegroups.com
Reply all
Reply to author
Forward
0 new messages