Improving solution for heat transfer with tetrahedral mesh
The solution accuracy for heat transfer prediction for laminar flow in channels with full tetrahedral mesh can be lower compared to a full hexahedral mesh. This is true even when the second order discretization scheme is used. This can be a problem in modeling complex geometries for heat transfer calculations where a full hexahedral mesh or even a boundary layer approach is not practical.
Fluent uses a cellcentered Finite Volume scheme and, in order to go second order, one performs a solution reconstruction from the cellcenter towards the integration points (the cell's facecenter). This reconstruction uses the cellcenter value and the reconstruction gradient of that variable to compute the face center values (on either sides of any cellface). Well, the flow field can have discontinuities or even large local spatial variations, so the local gradients can be big, and the reconstruction can fail to produce the expected results. To 'fix' this problem one uses a limiter. The limiter is built such that it satisfies the so called "maximum principle". Basically, the limiter will switch to a first order reconstruction any time it detects a 'discontinuity'. Effectively, although the user specifies to use a second order scheme, at least for temperature the limiter can switch the scheme almost everywhere to first order, the scheme with extra dissipation. The Fluent default limiter works fine for regular hex grids but obviously not that well for tetrahedral meshes. To alleviate this problem, a new limiter is implemented in Fluent 6.1, called "Multidimensional Limiter", which does the same job of enforcing the 'maximum principle', but it will not introduce extra dissipation for smooth flows. Solution. Use TUI and type: > sol <enter> > set <enter> > slo <enter> and choose the MD limiter [1]. The results for heat transfer with full tet mesh with this limiter will be closer to the solution obtained for full hex mesh. 

