I've seen your comments that you feel that the various non-number values in arb are not well enough defined, but I think this should be a case where there is a reasonable mathematically correct interpretation.
In arb_mul (and arb_mul_arf), if you multiply an arb_t (or an arf_t) representing a signed infinity by a non-exact finite arb_t, then the result's radius is infinite. It seems like it would be mathematically sound to test whether the finite arb_t is either all positive or all negative, and if so, set the result's radius to be finite instead -- correct? That would give answers that users might expect more, even though of course the current result is also correct. I would be happy to prepare a patch.
I've seen your comments that you feel that the various non-number values in arb are not well enough defined, but I think this should be a case where there is a reasonable mathematically correct interpretation.
In
arb_mul
(andarb_mul_arf
), if you multiply anarb_t
(or anarf_t
) representing a signed infinity by a non-exact finitearb_t
, then the result's radius is infinite. It seems like it would be mathematically sound to test whether the finitearb_t
is either all positive or all negative, and if so, set the result's radius to be finite instead -- correct? That would give answers that users might expect more, even though of course the current result is also correct. I would be happy to prepare a patch.