Source-linked AI summary
Concentration of Stationary Measures onto Large Scales in the Fast-Advection Limit of the Stochastically Forced Two-Dimensional Navier-Stokes Equations
Tobias Rohner
TL;DR
The paper asks whether stationary measures of stochastically forced two-dimensional Navier–Stokes concentrate on the lowest Fourier modes as advection becomes fast. Numerical evidence shows increasing concentration as ε decreases, while extrapolation suggests the limiting values may be independent of forcing shell.
Problem
The paper asks whether stationary distributions concentrate on the lowest Fourier modes as ε → 0 and whether this behavior depends on forcing shell or torus geometry.
Source: §1 · p. 3Method
The authors examine numerical evidence for the fast-advection limit of statistically stationary stochastically forced two-dimensional Navier–Stokes equations.
Source: §4 · pp. 15–16Results
Both E[E/Ω] and E[R] increase monotonically as ε → 0, and the stationary distribution concentrates on the lowest Fourier modes, with approximately 95% of energy there at the smallest simulated ε.
Source: §3.3 · p. 15Takeaways & Limitations
At finite ε concentration depends on forcing shell through B1/B0, while extrapolation suggests limiting values may be forcing-independent and the same behavior persists on the thin torus.
Source: §3.3 · p. 15, §4 · p. 16All simulations use N = 128, and extrapolations for the two highest forcing shells rely on four nonzero points and slowly drifting data.
Source: §3.1.2 · p. 7, §3.3 · p. 15
Abstract
from arXiv · showhide
We investigate the fast-advection limit ($\varepsilon \to 0$) of the stochastically forced two-dimensional incompressible Navier-Stokes equations on the torus. Numerical simulations are performed at resolution $N = 128$ for four different forcing shells on both the square torus and a thin torus, with the advection parameter $\varepsilon$ varied over approximately one decade. We find that the stationary distributions become increasingly concentrated on the lowest Fourier modes as $\varepsilon \to 0$, with the energy-to-enstrophy ratio $E/\Omega$ and the fraction $R$ of enstrophy in the lowest modes both increasing monotonically. At finite $\varepsilon$ the degree of concentration depends on the forcing shell through the effective spectral value $B_1/B_0$, in agreement with the condensation bound of Sznitman and Widmayer. Linear extrapolation to $\varepsilon = 0$ suggests that the limiting values may be forcing-independent, with $B_1/B_0$ governing only the rate of convergence. The same qualitative behaviour persists on the thin torus.
1 Introduction
In two-dimensional turbulence, energy and enstrophy conservation support an inverse energy cascade and large-scale condensates on finite domains. This paper studies whether stationary condensates concentrate on the lowest Fourier modes in the fast-advection limit, and how forcing and torus geometry affect that behavior.
1 Introduction
- On finite domains, the inverse cascade is arrested at the largest scales, accumulating energy in the lowest Fourier modes and forming condensates.The dual-cascade scenario transfers enstrophy toward small scales while transporting energy upscale.Source: §1 · p. 1
- Prior work found that condensate onset depends on the forcing mechanism, with prescribed-energy stochastic forcing producing a continuous transition and linear-instability forcing a discontinuous one with hysteresis.This motivates distinguishing onset behavior from the stationary-state limit studied here.Source: §1 · p. 1
- Sznitman and Widmayer [11] [12] bound the expected energy-to-enstrophy ratio using the forcing's effective spectral value B1/B0, predicting stronger concentration when it is small relative to the highest resolved wavenumber.Their framework leaves open questions about limiting motion and the role of small stirring.Source: §1 · p. 2
- The study examines statistically stationary condensates after formation, rather than the onset of condensation, using simulations evolved to approximate statistical stationarity.Its numerical findings are compared with the Sznitman–Widmayer condensation bound.Source: §1 · p. 2
- The paper asks whether stationary distributions concentrate on the lowest Fourier modes as ε → 0, whether key quantities have positive limits, and how forcing shell and torus geometry affect the behavior.It also asks how variances of global quantities and individual low modes behave in this limit.Source: §1 · p. 3
2 Numerical Method
The numerical method uses a Fourier-space pseudospectral solver with additive stochastic forcing, spectral viscosity, and Euler–Maruyama timestepping. Thin-torus simulations restrict the dynamics to a Fourier sublattice corresponding to the thinner periodic geometry.
2 Numerical Method
- The GPU-accelerated pseudospectral solver azeban [9] operates on velocity fields in Fourier space, enforces incompressibility with the Leray projection, and computes advection using 2/3-dealiasing.The solver was extended for additive stochastic forcing; data are publicly available.Source: §2 · p. 3, §2.1 · p. 4
2.1 Stochastic forcing
- Forcing is translated from vorticity to velocity variables using the Biot–Savart operator, with the implementation multiplying each forced mode by k⊥/|k|.The forcing normalization keeps the prescribed energy injection rate B0 constant across forcing shells; ε controls advection strength, viscosity, and forcing amplitude.Source: §2.1 · p. 4
2.2 Viscosity
- Spectral viscosity uses s = 1 and a step damping profile that equals one for every nonzero mode, reducing to standard Laplacian dissipation.The viscosity coefficient is εN, with 1/N scaling consistent with spectral-viscosity convergence theory [6].Source: §2.2 · p. 4
2.3 Timestepping
- Because the additive forcing has state-independent diffusion, Euler–Maruyama is equivalent to Milstein and yields strong order-one convergence.Each Fourier mode is updated with a single Euler–Maruyama step per timestep, subject to advective and viscous CFL constraints.Source: §2.3 · p. 5
2.4 Thin torus
- Thin-torus simulations restrict forcing to modes with k1 divisible by δ, and the dynamics remain on that sublattice, representing functions periodic with period 2π/δ in x.The per-mode forcing amplitude is rescaled to compensate for the reduced number of active modes; this work uses δ = 2.Source: §2.4 · p. 5
3 Results
Across the square and thin tori, concentration measures rise as ε decreases, and extrapolations suggest positive finite limits with reduced variation across forcing shells. Energy concentrates more strongly than enstrophy, while fluctuations in global and low-mode quantities decrease.
3.1 Square Torus
- The square-torus simulations use N = 128, four forcing shells with B1/B0 ranging from approximately 134 to 687, and ε values spanning roughly one decade.Stationary averages are computed over the final 1/16 of each simulation.Source: §3.1 · pp. 5–6
3.3 Discussion
- The extrapolations for the (20, 21) and (20, 26) shells are less certain because they use only four nonzero points and the observables still drift slowly at the smallest ε.The last-quarter relative standard deviation is 2–3%; the discussion also notes the single resolution, single realization, and assumed regular ε →0 asymptotics as limitations.Source: §3.1.2 · p. 7, §3.3 · p. 15
3.1.3 Spectral distribution and variances
- At the smallest ε, approximately 95% of total energy but only 59% of enstrophy lies in |k|∞ = 1, while the vorticity field develops a coherent dipole.The spectra show energy accumulating in the lowest modes while enstrophy remains more broadly distributed.Source: §3.1.3 · p. 8, Figure 2 · p. 9, Figure 3 · p. 10
3.2 Thin Torus
- On the thin torus, extrapolated E[E/Ω] lies in 0.57–0.67, with a cross-shell ratio of ≈1.18, comparable to the square-torus range of 0.56–0.66.The thin-torus setup uses B0 = 4000 to compensate for fewer active modes under the sublattice restriction.Source: §3.2 · p. 12, Table 2 · p. 12
3.3 Discussion
- Both E[E/Ω] and E[R] increase monotonically as ε →0 across forcing shells and both torus geometries, with extrapolation supporting positive finite limits.The thin-torus results show the same qualitative behaviour as the square-torus results.Source: Figure 1 · p. 6, Figure 6 · p. 12, §3.3 · p. 15
- At the smallest simulated ε, approximately 95% of energy resides in |k|∞ = 1, while R reaches 0.59 on the square torus and 0.61 on the thin torus.Energy is more sharply peaked at the lowest mode than enstrophy.Source: Figure 2 · p. 9, §3.3 · p. 15
3.1.2 Extrapolation to the inviscid limit
- At finite ε, smaller B1/B0 yields stronger condensation; the spread in E[E/Ω] narrows from ≈1.7 to ≈1.17 after extrapolation, suggesting forcing-independent limits.This agrees qualitatively with the Sznitman–Widmayer condensation bound [11, 12], which controls finite-ε distance from the condensed regime but does not predict distinct limiting values.Source: §3.1.2 · p. 7, §3.3 · p. 15
3.1.3 Spectral distribution and variances
- The normalized variance Var(Ω)/B0^2 and the relative variance of the lowest modes both decrease as ε →0, indicating increasingly deterministic large-scale observables.For the (10, 11) shell, the lowest-mode relative variance falls from approximately 1.5 to below 0.1.Source: §3.1.3 · pp. 8, 11, §3.3 · p. 15
4 Conclusion
Numerical evidence indicates that stationary measures concentrate on the lowest Fourier modes as ε → 0, with reduced fluctuations and a coherent dipole condensate. At finite ε, concentration depends on the forcing shell, while extrapolation suggests the limiting values may be forcing-independent; this behaviour also persists on the thin torus.
4 Conclusion
- As ε → 0, stationary measures concentrate monotonically on the lowest Fourier modes, with concentration stronger for energy than enstrophy and fluctuations reduced in global and low-mode quantities.At finite ε, concentration depends on B1/B0; extrapolation suggests forcing-independent limiting values, with B1/B0 governing the rate of convergence.Source: §4 · pp. 15–16
- The same concentration behaviour persists on the thin torus, while extrapolation suggests the limiting values may be forcing-independent.The forcing-shell dependence at finite ε agrees qualitatively with the condensation bound of Sznitman and Widmayer [11] [12].Source: §4 · p. 16
- The condensate appears as a coherent dipole, consistent with dominance of the |k| = 1 Fourier modes and with geometry shaped by the domain and its lowest modes.Earlier stochastic-forcing studies also observed large-scale coherent structures.Source: §4 · p. 16
- These simulations complement the theoretical framework of [11] [12] with direct numerical evidence for condensation in a high-dimensional setting.They raise whether the apparent forcing-independence of the limit can be established rigorously.Source: §4 · p. 16