Currently what I am doing is: S(v) is a d-dimensional positive definite matrix determined by the d-dimensional vector v.

I want to optimize the maximum diagonal entry of the inverse of S(v) subject to

entrywise-sum of v equal to 1.

(See https://mathoverflow.net/questions/416095/cvxpy-maximum-diagonal-entry-of-the-inverse-matrix for details)

This is a convex problem, but CVXPY does not support a matrix inverse including a variable, as far as I know.

Using matrix_frac for a canonical vector e (will return e @ S(v) @ e) and

taking maximum for d different canonical vectors will work,

but it will create d-inverse of d by d matrix for each computation, which is really heavy.

Any other good solution?