Kolmogorov flow — a sinusoidal forcing on a two-dimensional Navier-Stokes fluid with periodic boundary conditions — has been a workhorse model for studying turbulence since the 1960s. Its simplicity makes it computationally tractable; its richness makes it physically relevant.
Note (2026-11-14; read against the paper's full text): the paper is a proof, and the numerical comparisons the essay reports come from the author's earlier work that it cites. What it proves is that the solution of the two-dimensional Kolmogorov flow with periodic boundary conditions keeps the spatial symmetry (rotation and/or translation) of its smooth initial condition for all later times, and the remark on numerical simulations assumes an even forcing wavenumber; a proof for the three-dimensional equations is in an appendix. It says clean numerical simulation keeps the symmetry in a finite but long enough time interval, not for all time, and that direct numerical simulation results keep it at first and then lose it quickly, so the symmetry check is a test for a finite window; the essay's 'time-unlimited diagnostic' overstates this. The paper presents the theorem as a way to check the correctness of simulations of turbulence, and the 'thermometer' image is the essay's.
Liao proves a theorem: solutions of 2D Kolmogorov flow preserve the spatial symmetry of their initial conditions for all time. If the initial velocity field has a specific reflection or rotation symmetry, that symmetry persists through the entire evolution, however turbulent the flow becomes. This is an exact, rigorous result about the Navier-Stokes equations.
The theorem becomes a diagnostic tool. Standard direct numerical simulations (DNS) lose this symmetry rapidly — within short integration times, the computed solution breaks symmetries that the mathematics guarantees must be preserved. The symmetry violation is not physical; it is numerical noise contaminating the trajectory. The flow has not actually broken symmetry; the simulation has.
Clean numerical simulations (CNS), which explicitly control numerical error to within specified bounds, preserve the symmetry throughout, consistent with the theorem.
The structural point: a mathematical property of the equations becomes a thermometer for the quality of their numerical solutions. The theorem tells you what the simulation must preserve, and any deviation is a direct measurement of computational infidelity. Physics provides the test that no benchmark suite can — an exact, problem-specific, time-unlimited diagnostic of whether the numbers you are computing bear any relation to the equations you wrote down.