greatitself.
Continue
advanced32 min studyLesson 929 in Physics

Spectral fluid methods and aliasing

Recognize nonlinear mode folding and distinguish a dealiased finite computation from continuum fluid claims.

Before you begin
On this page

The question for today

What you’ll understand

  • Identify Fourier modes that become indistinguishable on a sampled grid.
  • Remove a quadratic alias in an explicit tiny example.
  • Separate numerical resolution, turbulence closure and exact continuum existence statements.

Fourier methods represent a periodic fluid field through spatial modes. Differentiation becomes simple multiplication by wave number, but nonlinear products create new modes beyond the retained range. Evaluating those products on too small a grid folds unresolved modes back into resolved ones. This aliasing is a deterministic numerical error rather than a turbulent fluctuation. A spectral calculation must specify its periodic domain, retained modes, product evaluation and time integration before its fine spatial appearance can be interpreted as accurate fluid dynamics.

Sampling identifies modes modulo the grid size

Take a dimensionless periodic coordinate x on an interval of length two pi, sampled at N uniformly spaced points without a repeated endpoint. A Fourier mode whose integer wave number differs by N has identical samples. Linear differentiation of retained coefficients does not create new wave numbers, but multiplying two retained fields convolves their spectra. Their sum wave number can exceed the grid's representable range and then appear as an unrelated lower wave number.

This mechanism is particularly important for quadratic advective terms in fluid equations. A spurious resolved contribution can feed back into subsequent time steps even if the original fields were resolved. More accurate arithmetic does not remove the sampling identity. Increasing the time integrator's order does not repair it either. One must either restrict the participating modes appropriately or evaluate the product on a sufficiently padded grid and project it back to the intended coefficient space.

xj=2πj/N:ei(k+mN)xj=eikxj,m∈Z,uv^k=∑p+q=ku^pv^qx_j=2\pi j/N:\qquad e^{i(k+mN)x_j}=e^{ikx_j},\quad m\in\mathbb Z,\qquad \widehat{uv}_k=\sum_{p+q=k}\widehat u_p\widehat v_q
x and the integer mode indices are dimensionless here; a physical interval length rescales wave numbers. The convolution is the continuum Fourier-series product; an unpadded discrete product instead includes aliased congruence classes.↔ Scroll wide equations horizontally.

Dealias the product and enforce the chosen constraints

For quadratic products, a standard approach pads coefficient arrays to three halves the original transform size, transforms to the larger grid, multiplies there and projects back. Equivalent two-thirds truncation rules require precise endpoint conventions; retaining a marginal cutoff mode without checking its aliases can reintroduce contamination. Cubic nonlinearities require different padding. Dealiasing makes a specified product projection accurate; it does not recover every physically unresolved mode or prove a turbulent spectrum converged.

In periodic incompressible flow, pressure can be eliminated for nonzero wave vectors by projecting velocity forcing perpendicular to the wave vector. The zero mode needs its own mean-flow and pressure convention. This Fourier projection does not implement no-slip walls on a nonperiodic container. Time integration still faces advective restrictions and, for explicit viscous stepping, a stiffness limit proportional to inverse wave number squared. A spatially excellent spectral representation can therefore fail through unstable time stepping or mismatched boundaries.

Pk=I−kkT∣k∣2(k≠0),∂tu^k=−ν∣k∣2u^k ⟹ ν∣k∣2Δt≤2 for explicit EulerP_k=I-\frac{k k^{\mathsf T}}{|k|^2}\quad(k\ne0),\qquad \partial_t\widehat u_k=-\nu|k|^2\widehat u_k\ \Longrightarrow\ \nu |k|^2\Delta t\le2\ \text{for explicit Euler}
In the projection k is a physical wave vector, ν is kinematic viscosity with length²/time units, and the final restriction concerns only the scalar viscous decay substep. It is not the full nonlinear fluid stability criterion.↔ Scroll wide equations horizontally.

Separate computation, closure and continuum research status

A finite resolved calculation is different from a turbulence closure, which models effects of unresolved scales in averaged or filtered equations. Both differ from an exact continuum existence statement. Fefferman's official formulation separates unforced existence on three-dimensional whole space with smooth, rapidly decaying, divergence-free initial data from the smooth divergence-free periodic-torus version. The breakdown alternatives allow smooth forcing with stipulated decay and periodicity conditions. Regularity and whole-space finite-energy requirements remain part of the target. A finite periodic run does not establish a theorem for every permitted initial field or for the whole-space alternative.

As checked on 18 October 2026, CMI's 11 September announcement reports an apparent settlement and an evaluation process still continuing. It does not report an accepted evaluation or awarded prize. This status changes neither the numerical aliasing identity nor the separate need to assess turbulence closures. The spectral lab displays a prescribed finite sine sum with a common temporal frequency. It has no nonlinear product, pressure projection, viscous solver or dealiased fluid evolution, so its additional harmonics illustrate resolution choices only.

∂tu+(u ⁣⋅ ⁣∇)u=−∇p/ρ+ν∇2u+f,∇ ⁣⋅u=0\partial_t u+(u\!\cdot\!\nabla)u=-\nabla p/\rho+\nu\nabla^2u+f,\qquad \nabla\!\cdot u=0
This is the constant-density incompressible model in physical units, with force per unit mass f and specified initial and boundary conditions. Writing it does not assert that a finite discretization settles its continuum existence or closure questions.↔ Scroll wide equations horizontally.

Work it through

A quadratic mode folds into the wrong location

Use N = 6 samples at xj=2πj/6x_j=2\pi j/6. A retained complex mode is u(x) = exp(2ix). Identify the samples of u², their aliased wave number, and what happens if the product is evaluated on nine points and projected back to retained modes |k| ≤ 2.

  1. The exact product has wave number four: u² = exp(4ix). On six points, four and minus two differ by six, so the samples also equal exp(−2ix).

  2. At j = 1, both values are exp(4πi/3) = exp(−2πi/3), directly displaying the alias. An unpadded product would contaminate the retained minus-two coefficient.

  3. A nine-point product grid resolves wave number four because its modes extend through ±4. Projecting the product back to |k| ≤ 2 discards that genuine high mode and leaves zero, rather than an incorrect minus-two component. This is a projection of a quadratic product, not an assertion that the physical product vanishes.

The result

The six-point product aliases k = 4 to k = −2. Nine-point evaluation followed by the declared low-mode projection removes that false retained contribution.

The living notebook

Explore the model

Interactive lab

Interactive lab · spectral

Retained harmonics without nonlinear evolution

Reset and set BOTH Mode count N = 3 and Coefficient decay p = 1; inspect the three prescribed harmonics. Reset and set N = 10 and p = 1; predict added shorter spatial harmonics weighted by 1/n. Both displays use the analytic sum sin(nx − t)/n^p, not a nonlinear product, fluid solution, dealiased transform or physical dispersion relation.

Retained harmonics without nonlinear evolutionA finite sum of sin(nx − t)/nᵖ. Playback changes t; increasing the cutoff includes finer spatial scales. This is not a proof about a continuum limit or renormalization. Current readings: Sum at x = π/2: 0.867; Smallest wavelength: 2π / 5; Highest-mode coefficient: 0.2. All parameters are adjustable with the labeled controls below.01.573.144.716.28−2.56−1.2801.282.56x · radiansFinite sine sum
Finite sumHighest included mode
Sum at x = π/2
0.867
Smallest wavelength
2π / 5
Highest-mode coefficient
0.2
t = 0 s

A finite sum of sin(nx − t)/nᵖ. Playback changes t; increasing the cutoff includes finer spatial scales. This is not a proof about a continuum limit or renormalization. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.

Look forAdding represented harmonics differs from correctly evaluating nonlinear interactions among them.

Pause & reason

Check your understanding

Which operation directly addresses quadratic aliasing in a periodic pseudospectral product?

Your turn

Try it yourself

For u(x) = cos(2x) on the same six-point grid, use the double-angle identity to find the false resolved mode in an unpadded evaluation of u². What remains after evaluating on nine points and projecting to |k| ≤ 2?

Take a moment to make a prediction or work it out on paper. Open a hint when you need one.

Give me a hint

The exact product is one half plus one half cos(4x); sampled cos(4x) equals cos(2x) on this six-point grid.

Compare with the solution

The exact square is 0.5 + 0.5 cos(4x), but six-point samples cannot distinguish the second term from 0.5 cos(2x). An unpadded product therefore inserts a false cosine-two amplitude of 0.5 into the retained band. Nine points distinguish the genuine ±4 modes from ±2. Projecting that product to |k| ≤ 2 removes both fourth harmonics and retains only the constant 0.5. Dealiasing does not remove the true mean; deleting the entire product would incorrectly erase it.

Keep the idea

Take it with you

  • Nonlinear products create wave numbers outside a retained band.
  • Padding and projection address aliasing, while time stability and boundaries remain separate.
  • A finite numerical result, a closure model and a continuum theorem make different claims.

Read further

Original sources and deeper explanations to follow at your own pace.

Optional progress, saved in this browser when available. Revisit any lesson at any time.

Lesson not marked complete.