The advantages of solution-adaptive refinement, when used properly as in the turbine cascade example in Section  26.1.1, are significant. However, this capability must be used carefully to avoid certain pitfalls. Some guidelines for proper usage of solution-adaptive refinement are as follows:

• The surface mesh must be fine enough to adequately represent the important features of the geometry.

For example, it would be bad practice to place too few nodes on the surface of a highly-curved airfoil, and then use solution refinement to add nodes on the surface. The surface will always contain the facets contained in the initial mesh, regardless of the additional nodes introduced by refinement.

• The initial mesh should contain sufficient cells to capture the essential features of the flow field.

Consider the following example, in which you want to predict the shock forming around a bluff body in supersonic flow. To obtain a reasonable first solution, the initial mesh should contain enough cells and also have sufficient resolution to represent the shape of the body. Subsequent gradient adaption can be used to sharpen the shock and to establish a grid-independent solution.

• Polyhedral cells are not eligible for adaption. The presence of polyhedral cells in a mesh may or may not limit the eligibility of other cells for adaption, depending on the manner in which the polyhedral cells were created:
• If the domain was converted to polyhedra (see Section  6.7.1), then no part of the mesh can be adapted (even if hexahedral cells are present in the mesh after conversion).

• If the polyhedra are a result of converting skewed tetrahedral cells (see Section  6.7.2) or converting the transitional cells of a hexcore mesh (see Section  31.5.2), then the nonpolyhedral cells may be adapted. The polyhedral cells, however, will be automatically unmarked from the register when adaption is initiated and will remain unchanged.

• Obtain a reasonably well-converged solution before performing an adaption. If you adapt to an incorrect solution, cells will be added in the wrong region of the flow.

Use careful judgment in deciding how well to converge the solution before adapting, because there is a trade-off between adapting too early to an unconverged solution and wasting time by continuing to iterate when the solution is not changing significantly. This does not directly apply to dynamic adaption, because here the solution is adapted either at every iteration or at every time-step, depending on which solver is being used.

• Write a case and data file before starting the adaption process. If you generate an undesirable mesh, you can restart the process with the saved files. This does not directly apply to dynamic adaption, because here the solution is adapted either at every iteration or at every time-step, depending on which solver is being used.

• Select suitable variables when performing gradient adaption. For some flows, the choice is clear. For instance, adapting on gradients of pressure is a good criterion for refining in the region of shock waves. In most incompressible flows, however, it makes little sense to refine on pressure gradients. A more suitable parameter in an incompressible flow might be mean velocity gradients. If the flow feature of interest is a turbulent shear flow, it will be important to resolve the gradients of turbulent kinetic energy and turbulent energy dissipation, so these might be appropriate refinement variables. In reacting flows, temperature or concentration (or mole or mass fraction) of reacting species might be appropriate.

• Do not over-refine a particular region of the solution domain. It causes very large gradients in cell volume. Such poor adaption practice can adversely affect the accuracy of the solution.