Math::NIntegrate ā 04-Fully-symmetric-rules-implementation
Reading the section "MultidimensionalRule" from the advanced documentation.
Finding the articles by Espelid, Genz.
Finding the corresponding FORTRAN 77 code.
Programming D07HRE and D09HRE to Raku by AI-agent Codex using "GPT-5.6-Terra".
Programming
method integrateand disliking the implementation.Using Codex tests and making 2D test on my own.
That implementation did not figure out the computation of the largest-error axis.
Reading the code and coming with related code outline / pseudocode.
Raku implementation of the pseudocode by Codex with "GPT-6.0-Sol".
Testing with region integration worked.
But with "GlobalAdaptive" the results were not good.
One reason is that the 7-generator rule (D07HRE) is not that robust.
So switched to 9-generator rule by default.
Another reason is that the largest-error axis was not passed/assigned after the rule application computation.
Currently, works very well.
Tested both numerically and visually.
Used a lot the singular integral
nintegrate({1 / ($^x + $y).sqrt}, <x 0 1>, <y 0 1>)