Your code has log of cvx_expression, so CVX uses its unreliable Successive Approximation method. if you use CVX 2.2 with Mosek 9.x as solver, CVX will make use of Mosek’s native exponential cone capability to handle the log; that is more reliable and robust, and likely faster. if Mosek is not available to you, install CVXQUAD and follow the instructions at CVXQUAD: How to use CVXQUAD's Pade Approximant instead of CVX's unreliable Successive Approximation for GP mode, log, exp, entr, rel_entr, kl_div, log_det, det_rootn, exponential cone. CVXQUAD's Quantum (Matrix) Entropy & Matrix Log related functions