After installing 13.3 and testing my code I ran into an issue with Arg which i guess has to do with something in the handling of the numerical precision of complex numbers. To prevent stuff happing in the background of my data i always mask that out by multiplication with exact 0 values. However in …

From the documentation on Numbers: Whenever you give a number with an explicit decimal point, the Wolfram Language produces an approximate numerical result Here is what happens by multiplying an approximate number by the exact zero: In[2]:= 0*0. Out[2]= 0.

The idea behind having 0*0.0 evaluate to 0.0 is to maintain the type rule int times real equal real. One can question this of course. But it has advantages particularly for type consistency and inferencing, as is needed in Compile and the newer FunctionCompile world.

I agree that 0*0.0+0.0 I gives 0.0+0.0 I I also expect that 0 * 0.0 + 10.0^-200 should give 0.0+0.0 I I also expect that Arg[-10.^-200 + (0.)] = Pi and Arg[0.0+0.0 I] = 0.0 But that was not my point, Arg[0 *(-10.0^-200+0.0I)] should and has always evaluated to 0.0 but with Mathematica 13.3 it no l…

I am working with version 13.3 on a Mac. The command c = Unitize[arg] gives a list of zeros every time I try it.

Martijn, I doubt there is anything I can change about this, but I’ll take a look because it does seem suspicious. Your explanation in this recent response, and isolation of clear examples, is a big help.

Thanks Daniel! For now i know the reason why my code changed outputs and i have a fix for it which works in all versions and i learned some things on the way so I’m happy for now.

I did check. It appears to be simply a bug. I filed a report for it.

Can you check mine too? I recently downloaded 13.3 on my Debian linux and checked there is the same issue as in cloud.

Unexpected behavior from Arg in v13.3

Questions

valzapodvz (ZAQU zaqu) July 4, 2023, 11:41pm 4

The C99 complex.h requires a separate implementation, that is in glibc and Intel libm, even less accurate than real numbers math.h glibc and Microsoft math library. Mathematica did not bother, it just uses naive implementation for complex multiplication. In practice the code using glibc is less accurate than from Windows:

Topic		Replies	Views
A complex number, Replace, Simplifiy, Refine .... Questions computer-science , finance , physics , algebra , calculus , equation-solving , numerical-computation , computational-linguistics	5	13235	September 13, 2015
Why Arg[z] is not always the same as Im[Log[z]] Questions numerical-computation	2	7474	February 2, 2017
PlanarAngle issues? Questions geometry , wolfram-cloud	13	8523	July 20, 2021
ArgMax produces wrong result Questions calculus , optimization	2	6082	March 25, 2021
Why Simplify[Arg[z]+Arg[Conjugate[z]], Element[z,Complexes]&& Im[z] !=0] does not evaluate to 0? Questions	3	3915	July 21, 2024

Unexpected behavior from Arg in v13.3

Related topics