“…However, in this case, the residual Jacobian consists of 12×12 blocks for 4 conservative variables and 8 primitive-variable gradient components (20×20 blocks in three dimensions). The size of the discrete problem is equivalent to the P 1 discontinuous Galerkin method, where the solution gradients are introduced as additional discrete unknowns, and to the hyperbolic Navier-Stokes method [26,27,28,29,30,31], where the solution gradients are introduced as additional unknowns in the differential-equation-level. Clearly, it would require much more memory to store the Jacobian matrix although it is quite feasible and can be very effective for iterative convergence.…”