arb

[orkolorko]

I'm currently working on a wrapper for arb (pretty similar to the wrapper for mpfi).
Is there any interest for the library?

[NAThompson]

I love Arb! But there seems to be some differences in goals. Arb uses arbitrary precision interval arithmetic for evaluation of special functions, whereas Multiprecision is purely for high-precision arithmetic. Presumably you could instantiate a boost::numeric::interval on a boost::multiprecision type and then use boost::math to evaluate special functions using the interval, but I suspect there is much pain associated with that route.

Also, doesn't arb just use MPFR? Boost.Multiprecision already has an MPFR wrapper.

[jzmaddock]

Actually I think this could work rather well: we already have a wrapper for mpfi and support for interval types within the library - so I would say to go for it :)

[orkolorko]

Answering NAThompson:

Arb is a multiprecision library for validated numerics, as mpfi.
The difference is that, while MPFI stores an interval as two endpoints, ARB stores an interval as a center point and a radius.
So, ARB is a multiprecision library.

Both libraries have some flaws and some advantages... since there is already a boost multiprecision support for MPFI, I think ARB would also be cool :-)

[NAThompson]

@orkolorko : Got it; I guess that means that there's some significant focus on the complex plane?

[orkolorko]

@NAThompson
ARB has two certified types, the arb type for real balls and the acb type for complex balls.
I'm initially thinking of wrapping the real balls and see what is possible to do for complex balls.

[NAThompson]

@orkolorko : I recant my previous statement! This is a cool idea.

[jzmaddock]

Ping? Any progress?

If it helps any, there is at least now a short guide to adding new backends: 6f73233

@ckormanyos : Is this yet another possible for GSOC (and yes I know we have more than enough already, just flagging it up).

[orkolorko]

I paused the work for some time, but I'm planning to get back to it this week,

…

On Wed, Mar 11, 2020 at 3:45 PM jzmaddock ***@***.***> wrote: Ping? Any progress? If it helps any, there is at least now a short guide to adding new backends: 6f73233 <6f73233> @ckormanyos <https://github.com/ckormanyos> : Is this yet another possible for GSOC (and yes I know we have more than enough already, just flagging it up). — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#135 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABNNBKTAB6Q74PB3XTIISFLRG7L4HANCNFSM4H35VZIQ> .

-- Isaia Nisoli

[ckormanyos]

Is this yet another possible for GSOC...

Yes. But indeed, as you mention, we have more fascinating ideas than we have time and capacity at the moment to fully handle them all with the depth these issues deserve.

One thing to consider may be to combine 6f73233 with either the issue here, as mentioned, or with quad_double and friends in #190

[orkolorko]

Another interesting idea would be to introduce some error-free multiprocessing type (similar to qd, but can be made much more general, i.e., Campary)

[kaba2]

Just wanted to say I'd love Arb support in Boost.Multiprecision, and was just about to suggest it. Has there been any progress?