2.4. Grid dependence and computer storage

The requirements set out in two recent guidelines referring to good CFD practice [35, 36] were followed in the present work. Non-uniform structured Cartesian grids were used throughout this work with extensive tests for independence of grid size checked by increasing the grid numbers until further refinement was shown not to be of significance. Close to solid surfaces, the grid was refined using geometric progression with an expansion ratio from the solid surfaces of less than 1.1 and with the Y<sup>þ</sup> values adjacent to a solid boundary held around a


Table 2. Various settings used during the calculations.


Table 3. Boundary conditions used during the calculations.

value of 20, so complying with recommendations given in the literature that Y<sup>þ</sup> should be between 11.5 and 300 to ensure accuracy when using a high-Reynolds turbulence model [37]. Tests for grid independent solutions were carried out using 2.0 �105 , 2.0�10<sup>6</sup> and 1.0� 107 . The overall change in the residual for each variable between the last two numbers of elements was less than 0.1%, indicating grid independence had been achieved. The calculations were performed on a Dell T5500 workstation with 32 nm six-core Intel Xeon 5600 series processor and main memory of 24 GB. A typical CPU time for a transient run with a grid having 10<sup>6</sup> nodes was just over 24 h.
