Why square Jacobians?

[gaperez64]

Is there a particular reason why you insist on Jacobian matrices being square?

TaylorSeries.jl/src/calculus.jl

Line 255 in d9ca589

@assert length(vf) == numVars

The usual definition allows for nonsquare Jacobians. Also, in the context of ODEs, if you have a time variable t then you'll end up with more variables than equations, right?

[lbenet]

Thanks for opening this issue, and good catch! There is no real reason... If you have some time, please, open a PR solving this.

[gaperez64]

Sure, I'll do that soon. But while on the topic, I am wondering if jacobian is the best name for this function since you are evaluating the matrix on a point. For me the Jacobian is a matrix of functions, and to obtain this using your API I instead need to feed it as evaluation point an array of variables, e.g.:

TaylorSeries.jacobian(g, [x, y])

[lbenet]

I think jacobian is used in both contexts: as a matrix of the derivatives of a vector of functions, or as that matrix evaluated at a given point, which is what we consider. If you have a better name, I'm open for suggestions, which the could allow us to have it exported (the reason it is not exported is to avoid naming collisions with other packages).

[lbenet]

Closed by #372