matrix log

[jacksonloper]

Working on a project about SO(n) and needed matrix log. Matrix log is tricky, and so I think it will add about 6kb to the bundle, because (at least to follow the scipy way of doing it) you have to have a function that converts quasi-triangular real Schur to complex triangular Schur and then another step for the actual logging of the triangular matrix.

A draft of what this might look like, currently passing tests, and including typescript and so on, is at

develop...jacksonloper:mathjs:feat/logm

Do you think it might be worth submitting an issue/PR?

Thanks!

Jackson

[josdejong]

Thanks Jackson, your code for logm looks well taken care off! I definitely think it would be a valuable addition to mathjs. It makes sense that such a function requires some KB's, though in the end it may be not that bad when gzipping the bundle. Do you maybe see any opportunities for code-reuse with existing functions?

[gwhitney]

The potential PR looks reasonable to me, and since we have expm, it would be very welcome to have logm as well. Please do submit a PR. (You can certainly look for re-use as Jos suggests, but I see for example you have taken advantage of the existing schur() function.)

My only comments/suggested changes are:

1. I do not think logm_triu() should be an independent library function for client calling, but an internal helper to logm(). As far as I can see, the user should not need to know that there is a separate logm call for upper triangular matrices, or have to worry about which logm to call on their matrix to get its log. Rather, logm should check for upper triangularity as appropriate, and reduce to logm_triu() when possible.

2. Similarly, please consider whether there is enough independent utility for rsf2csf to warrant it being a separate user-level function of the package rather than an internal helper to logm(). That's hard for me to judge, as this is not my area of math.

In any case, please do submit a PR! Thanks.

[jacksonloper]

Great! I realized I need to get the Schur decomposition more reliable first, so I'm suggesting #3634 first. Once that is all set, the logm should be straightforward. I will take logm_triu and rsf2csf out of user-land.