We form a quadratic-separable problems, which is great for CLARABEL, QSS, and OSQP, but not what MOSEK expects. Currently, we're forcing the data through cvxpy to get the MOSEK canonicalization with a linear objective and expanded cone constraints. This can be improved by linearizing the quadratic term directly, e.g., with a sparse Cholesky factorization, as shown below.
We form a quadratic-separable problems, which is great for CLARABEL, QSS, and OSQP, but not what MOSEK expects. Currently, we're forcing the data through cvxpy to get the MOSEK canonicalization with a linear objective and expanded cone constraints. This can be improved by linearizing the quadratic term directly, e.g., with a sparse Cholesky factorization, as shown below.