Hi

looking at current (limited) possibilities for solving the Navier-Stokes equations in the case of highly turbulent air flow ("traditional" CFD with turbulence model, Lattice-Boltzmann, Lattice-Gas, Large-Eddy, DES, DNS) I feel that this is all not really appropriate to the problem. Todays computers (and the "second one" on the graphics card) are very powerful and I suppose that there should exist methods that allow a more precise and faster turbulent flow simulation. CFD seems to stick in either discrete field or particle methods but both seem to be not the end of the story.

It is just a feeling, but it tells me that my up-to-date computer _should_ be able to simulate a turbulent flow, even at interactive frame rates.

Is there anybody interested in discussing this issue? I know there are completely different methods that still have to be discovered. The often cited much-too-large resources for a real DNS at high Reynolds numbers should not be the end of the story. Why can the resources not be limited only to the states that are valid solutions of the NS equations? Storing a huge grid with floating point field variables is one thing but almost all of the states that such an ensemble could represent are not valid solutions of the NS equations. Therefore there must actually not be storage space for those. This is just one thing I was thinking about, maybe it is nonsense however.

regards

Rob

I think the point is that the instantaneous flow field prediction will be invalid (compared to reality) due to sensitivity to initial conditions of this non-linear open system. i.e. even if the modelling equation were 'perfect', imperfections in the initial and boundary conditions would produce errors that would lead to a divergence between the real evolution of the turbulent flow and the modelled one (the largest positive Lyaponov exponent of the system is a good measure of this). This harks back to Ed Lorenz's original 'butterfly effect' paper.

However, despite the fact that the system is chaotic and thus unpredictable, the time averaged properties of the flow (e.g. u_bar) should themsleves be predictable. So, the validity in such simulations are in the prediction of engineering useful metrics such as time averaged velocity etc.

Now, if you're just after a qualitative prediction of a turbulent flow, such as for CGI effects then yes, maybe there are assumptions you could make, or new approaches that could be taken, to simplify the governing equations into a form that would result in the 'real time' visualisation (I don't want to use the term 'prediction'!) of a highly turbulent flow. What these might be I've no idea!

Umm... I read a paper a long time ago about using a model of many rotating cylinders, of a range of different diameters, all abutting and counter rotating. This model self-organised itself into a form that had similar characteristics to the turbulent energy spectrum of a turbulent flow. It looked pretty as well!