arpra-project/arpra

Arpra is a C library for analyzing the propagation of numerical error in arbitrary precision IEEE-754 floating-point computations.

arpra's Issues

New name and logo for the MPFA project

The current name, MPFA, is a twist on MPFR (GNU Multiple Precision Floating-Point Reliably), on which this software is based; the A stands for Affine Arithmetic. I picked the name without much thought, following the example of the MPFI Interval Arithmetic library, however I am concerned that these similarly named libraries will be easily mistaken for each other.

As such, I think MPFA should be changed to something more unique and memorable, whilst this is still possible without disruption, and propose a vote for the new name. I will update this post with the candidates so far, and their vote counts, but please suggest alternatives if none appeal to you. I will set a provisional deadline at the end of 31st January 2018, UTC time.

A new logo for the website must also be eventually chosen, and will throw in some options as I think of them, but I welcome all other submissions. Again, I will update this post with candidates.

A couple of names, and my vote, to begin with:

MPFA (keep the existing name)
0 votes

ArPRA ([Ar]bitrary [P]recision [R]ange [A]nalysis)
1 vote

MP Range ([M]ultiple [P]recision [Range] Analysis)
0 votes

AP Range ([A]rbitrary [P]recision [Range] Analysis)
0 votes

RAMP ([R]ange [A]nalysis [M]ultiple [P]recision)
0 votes

division by zero in transcendental functions

Division by zero is possible in transcendental functions when x.radius > 0, but width(true_range) == 0.

Gaussian noise symbols

Noise symbols ε are currently all uniformly distributed in [-1, 1]. The possibility of having ε normally distributed about 0, with standard distribution σ, might also be useful, for statistical applications.

Normally distributed ε could be stored and computed in a separate array, in a similar manner to uniformly distributed ε. All σ for each ε would need to be stored globally.

Modernising the Arpra buildsystem

This issue tracks the Arpra project buildsystem change.

In this change, the current GNU Make buildsystem will be replaced with a modern alternative. Candidates are:

• CMake
• Meson

Further research required.

Add split and join functions for ranges

The 'split' function should split a range into two or more new ranges.
The 'join' function should merge two or more ranges into one new range.

The resulting ranges of both functions should be as tight as possible, while preserving as much correlation information as possible. The joint range of all output ranges should fully contain the joint range of all input ranges, preserving the fundamental invariant of range analysis (see http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.36.8089, section 1.2.3).

Taylor series deviation terms

Add support for Taylor series deviation terms.

Currently, Arpra only supports first-order deviation terms (affine arithmetic).

configure is missing

Following the instructions in README:

kbriggs:~/Downloads/mpfa-0.0.0> ./configure
bash: ./configure: No such file or directory