“…With this choice, the block-diagonal entries of the Jacobian reduce to diagonal matrices, enabling the construction of implicit schemes particularly robust and free of matrix inversions. However, it was shown that this choice does degrade the achievable rate of convergence [46,7,15], and using characteristic-splitting of the flux Jacobians is preferable. Acknowledging this fact, we favor the matrix dissipation scheme in the definition of the residual, whose first-order equivalent is the upwind flux of Roe [33].…”