search

SKopecz / PositiveIntegrators.jl

A Julia library of positivity-preserving time integration methods
https://skopecz.github.io/PositiveIntegrators.jl/
MIT License
13 stars 4 forks source link

Sk/update basic examples and docstrings #96

Closed SKopecz closed 2 months ago

SKopecz commented 2 months ago

This is to take care of #7.

codecov-commenter commented 2 months ago

:warning: Please install the 'codecov app svg image' to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 98.32%. Comparing base (2389d44) to head (c37dbba).

:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files ```diff @@ Coverage Diff @@ ## main #96 +/- ## ======================================= Coverage 98.32% 98.32% ======================================= Files 6 6 Lines 1250 1250 ======================================= Hits 1229 1229 Misses 21 21 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

SKopecz commented 2 months ago

I have some issues regarding PDSProblem. See https://github.com/SKopecz/PositiveIntegrators.jl/blob/2389d44d8037e2d3a2c1ba0d3be67a179dfd3569/src/proddest.jl#L7C24-L7C45.

  1. I don't see why we need d_prototype in PDSProblems. D will always be a vector with the same floating point type as P. Or are there other reasonable choices?
  2. The docu says "If P is given in in-place form, p_prototype is used to store evaluations of P". This is true for PDSFunctions, but these are only used internally and not documented. Users might think that p_prototype is used to store evaluations of P inside an algorithm like MPE() while solving a PDS. But this is not the case. All algorithms allocate new memory using something like P=zero(f.p_prototype).
  3. Would it be a good idea to actually use f.p_prototype to store evaluations of P inside the algorithms?
ranocha commented 2 months ago
  1. I don't see why we need d_prototype in PDSProblems. D will always be a vector with the same floating point type as P. Or are there other reasonable choices?

Maybe someone wants to use a SparseVector (from SparseArrays) or some custom struct encoding sparsity? Or we may want to have the flexibility to run on GPUs in the future?

  1. The docu says "If P is given in in-place form, p_prototype is used to store evaluations of P". This is true for PDSFunctions, but these are only used internally and not documented. Users might think that p_prototype is used to store evaluations of P inside an algorithm like MPE() while solving a PDS. But this is not the case. All algorithms allocate new memory using something like P=zero(f.p_prototype).

Maybe we should write "p_prototype or copies thereof are used to store evaluations of P" to clarify this aspect?

  1. Would it be a good idea to actually use f.p_prototype to store evaluations of P inside the algorithms?

I don't think so. Right now, it is possible to solve the same problem concurrently using different time integration methods. If we overwrite some memory of the problem, this is not possible anymore.

SKopecz commented 2 months ago

I created #97 for the implementation of d_prototype in PDSProblems.

SKopecz commented 2 months ago

I created #98 to improve the efficiency of sparse matrices.