A car with a single suspension setting for all wheels — smooth on asphalt, but bumpy on dirt roads. That's roughly how modern computer models of fluids and gases work: all disturbances, from large whirlpools to fine ripples, are calmed with one common "brake." Stochastic modeling of flows has long been seeking a more refined tool.
A team of developers took the popular lattice scheme D3Q125 (125 fixed velocities in three-dimensional space) as a basis and divided turbulence into floors. Instead of a single nonequilibrium sensor, they introduced three separate ones: K2 for large vortices, K3 for medium ones, and K4 for the smallest. Each sensor controls its own damping spectrum, like a separate shock absorber for each type of bump. In tests, this separation worked selectively: when, say, only K3 was activated, the other sectors remained undisturbed to within rounding errors.
The method showed sensitivity to the frame of reference: the effect depends on whether we view the flow from the laboratory or move along with it. But the sign of the correction (that is, the improvement itself) remained stable across all tested configurations. The main physical invariants (mass, momentum) were preserved to machine precision, and the numerical "fluid" remained positive — no negative densities appeared.
For now, this is only a demonstration of the mechanism on a fixed set of velocities. To be confident in universal accuracy, an independent verification against reference kinetic data is needed. But it's already clear: the ion channel of turbulence can be tuned floor by floor, and this reduces noise.
The number 125 in the scheme name D3Q125 is the number of allowed velocities for particles: three dimensions, five levels in each direction, plus many diagonal combinations. This set allows for more accurate description of complex flows but requires more computation.