Repository navigation
Radix p-adic numbers - #2719
Radix p-adic numbers#2719
Conversation
|
Added square root functions. Lots of things to tighten up, but the asymptotic speed is good. |
|
Now moved to its own module. I'm still undecided about whether this should actually be named The latter would match the Or maybe we should use the shorter |
|
There is now decent dot product code with delayed limb reduction. Some timings for FLINT ( Matrix multiplication: M * M2 where M = nxn Hilbert matrix, M2 = M+1. Determinant: det(M) where M = nxn Hilbert matrix This should be even better once there is actual fast matrix multiplication code. |
|
Renamed to |
This PR (co-written with Claude Opus) implements$p$ -adic numbers and fields for word-size $p$ in the $p$ -adic ball arithmetic, and this works out of the box with GR generics:
radixrepresentation. Unlike the existingpadicmodule,radix_padicelements track error bounds, i.e. we implementOther advantages of
radix_padic:padic(see benchmarks below), and there is no need to precompute huge powers ofOther differences:
padicthe output precision is controlled by the variable (MPFR-style); inradix_padicthe default precision is set in the context object. TODO: add arithmetic functions that take precision parameters as input (Arb-style).Known issues:
radix_padica bit for this use case too.padic_polyandpadic_matfor most things as there is no fast polynomial multiplication and matrix multiplication yet.The PR mostly consists of wrapper code around existing$p$ -adic division.
radixalgorithms, but also implements block-basecase and Karp-MarksteinOutput of
radix/profile/p-padicbelow comparingpadicandradix_padicoperations; x and y are full-precision 7-adic numbers (with big-O error bounds attached in theradix_padiccase).The only operations that are slower are:
PR marked as draft as this needs review, code cleanup, and work on the interfaces.
@thofma @fieker @r-mb