pith. sign in

arxiv: 1907.05554 · v1 · pith:ASFYY2KGnew · submitted 2019-07-12 · ⚛️ physics.ao-ph · physics.flu-dyn

Residual cross-grid flow numerical error in large-eddy simulations of cumulus-topped boundary layers

Pith reviewed 2026-05-24 22:35 UTC · model grok-4.3

classification ⚛️ physics.ao-ph physics.flu-dyn
keywords large-eddy simulationshallow convectioncumulus-topped boundary layernumerical dispersion errorGalilean invariancefinite difference schemedomain translation velocity
0
0 comments X

The pith

LES results of shallow convection depend on domain translation velocity even in Galilean-invariant formulations because of residual cross-grid flow error from finite-difference dispersion biases.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that large-eddy simulations of cumulus-topped boundary layers produce different outcomes depending on the chosen domain translation velocity, despite the equations being Galilean invariant. This dependence stems from biases in the dispersion properties of finite-difference schemes, which grow when the flow contains strong asymmetries such as cumulus updrafts. The size of the error shrinks steadily as the numerical scheme order rises from second to sixth. Readers interested in atmospheric modeling would care because translation is commonly used to enlarge time steps, so any hidden sensitivity undermines the trustworthiness of the computed statistics.

Core claim

LES results of shallow convection depend on the domain translation velocity even when a Galilean invariant formulation is used. This residual cross-grid flow error is caused by biases in finite difference dispersion errors that are amplified by large-scale flow asymmetries such as strong updrafts in cumulus-cloud layers. The cross-grid flow error strongly depends on the order of accuracy of the numerical scheme, progressively becoming negligible as the order of accuracy is increased from second to sixth.

What carries the argument

Residual cross-grid flow error arising from biased finite-difference dispersion errors that are amplified by large-scale flow asymmetries such as strong updrafts.

If this is right

  • Low-order schemes produce larger cross-grid flow errors than higher-order schemes.
  • The error becomes negligible once the scheme order reaches sixth in the tested configurations.
  • Flow features with strong vertical asymmetry, such as cumulus updrafts, magnify the dispersion biases.
  • Domain translation can be used safely only when the numerical scheme is sufficiently accurate.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Simulations that rely on mean-flow translation to increase time-step size may need to verify results at multiple translation speeds when using low-order schemes.
  • The same dispersion-bias mechanism could appear in other flow configurations that combine mean advection with localized strong gradients.
  • An alternative fix would be to replace the finite-difference advection operator with a dispersion-free method and re-test the translation-velocity sensitivity.

Load-bearing premise

The dependence on translation velocity is produced by amplification of numerical dispersion biases by strong updrafts in the cloud layer.

What would settle it

Running the same shallow-convection case with a sixth-order or higher scheme and finding that statistics remain unchanged across a range of translation velocities would support the claim; persistent dependence at sixth order would falsify it.

read the original abstract

A computational domain translation velocity is often used in LES simulations to improve computational performance by allowing longer time-step intervals. Even though the equations of motion are Galilean invariant, LES results have been observed to depend on the translation velocity. It is found that LES results of shallow convection depend on the domain translation velocity even when a Galilean invariant formulation is used. This type of model error is named residual cross-grid flow error, to emphasize the expectation that it should be negligible or zero. The residual gross-grid flow error is caused by biases in finite difference dispersion errors. Schemes with low resolving power (typically low order of accuracy) produce larger dispersion errors that can be amplified by large-scale flow asymmetries, such as strong updrafts in cumulus-cloud layers. Accordingly, the cross-grid flow error strongly depends on the order of accuracy of the numerical scheme progressively becoming negligible as the order of accuracy is increased from second to sixth in the present simulations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims that LES results for shallow convection depend on domain translation velocity despite a Galilean-invariant formulation. This dependence is termed 'residual cross-grid flow error' and is attributed to biases in finite-difference dispersion errors. These biases are larger for low-order schemes and are amplified by flow asymmetries such as strong updrafts; the error is reported to decrease progressively and become negligible as scheme order increases from second to sixth.

Significance. If the central claim is supported by the simulations, the work identifies a subtle but potentially systematic numerical artifact in atmospheric boundary-layer LES that affects the fidelity of cumulus simulations. It supplies a concrete recommendation on advection-scheme order that could be adopted in operational models, and the emphasis on Galilean invariance plus quantitative order-dependence tests would constitute a useful contribution to the numerical-methods literature in the field.

major comments (2)
  1. [Abstract] Abstract and § on numerical experiments: the claim that the error 'strongly depends on the order of accuracy' and 'becomes negligible' at sixth order is presented without reported error magnitudes, convergence rates, or quantitative comparison across the second-, fourth-, and sixth-order runs. Without these data the load-bearing assertion that the dependence vanishes at sixth order cannot be evaluated.
  2. [Mechanism discussion] Mechanism discussion: the attribution of the observed translation-velocity dependence specifically to finite-difference dispersion biases is not isolated from other possible sources. No dispersion-relation analysis, no symmetric-flow control runs, and no tests that hold the pressure solver, time-stepping, and subgrid model fixed while varying only advection order are described; therefore the causal link remains unverified.
minor comments (2)
  1. Define 'residual cross-grid flow error' quantitatively (e.g., via a specific norm or diagnostic) at first use so that later statements about its magnitude are unambiguous.
  2. Provide the grid resolution, domain size, and time-step criteria used in the reported simulations so that the practical impact of the error can be assessed by readers.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address the two major comments below and will revise the manuscript to improve quantitative support and clarify the mechanistic discussion.

read point-by-point responses
  1. Referee: [Abstract] Abstract and § on numerical experiments: the claim that the error 'strongly depends on the order of accuracy' and 'becomes negligible' at sixth order is presented without reported error magnitudes, convergence rates, or quantitative comparison across the second-, fourth-, and sixth-order runs. Without these data the load-bearing assertion that the dependence vanishes at sixth order cannot be evaluated.

    Authors: We agree that explicit numerical values would strengthen the claim. The manuscript presents the order dependence primarily through figures showing reduced sensitivity at higher orders, but does not tabulate error magnitudes or convergence rates. In the revised version we will add a table (or expanded text in the numerical experiments section) that reports the magnitude of the translation-velocity dependence (e.g., maximum difference in domain-averaged cloud fraction and liquid-water path across the tested velocities) for the second-, fourth-, and sixth-order schemes, together with the observed reduction ratios between successive orders. revision: yes

  2. Referee: [Mechanism discussion] Mechanism discussion: the attribution of the observed translation-velocity dependence specifically to finite-difference dispersion biases is not isolated from other possible sources. No dispersion-relation analysis, no symmetric-flow control runs, and no tests that hold the pressure solver, time-stepping, and subgrid model fixed while varying only advection order are described; therefore the causal link remains unverified.

    Authors: The attribution rests on the systematic decrease of the observed error with increasing formal order, which aligns with the known improvement in dispersion relations for higher-order finite-difference schemes. We acknowledge, however, that the manuscript does not include dedicated isolation experiments (dispersion analysis, symmetric-flow controls, or fixed-solver advection-order sweeps). In revision we will expand the mechanism discussion to state this inference explicitly and note that additional targeted tests would be required to rule out contributions from other numerical components. No new simulations will be added. revision: partial

Circularity Check

0 steps flagged

No significant circularity: empirical observation of scheme-order dependence

full rationale

The paper reports an observed dependence of LES shallow-convection results on domain translation velocity even under a Galilean-invariant formulation, naming the phenomenon 'residual cross-grid flow error' and linking it to dispersion biases that diminish with increasing finite-difference order (2nd to 6th). This is presented as a direct outcome of controlled numerical experiments rather than any self-definitional loop, fitted-parameter prediction, or self-citation chain. No load-bearing step reduces by construction to its own inputs; the attribution rests on the simulation results themselves, which remain externally falsifiable against other codes or analytic dispersion relations.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The claim rests on standard numerical analysis of finite-difference dispersion and the domain assumption of Galilean invariance; the named error is introduced as an explanatory label for observed simulation behavior without independent falsifiable evidence outside the runs.

axioms (2)
  • domain assumption The equations of motion are Galilean invariant.
    Explicitly invoked in the abstract as background for the unexpected dependence.
  • standard math Finite-difference schemes exhibit dispersion errors whose magnitude decreases with increasing order of accuracy.
    Standard property of numerical methods used to explain the error reduction from second to sixth order.
invented entities (1)
  • residual cross-grid flow error no independent evidence
    purpose: Label for the observed dependence of LES results on domain translation velocity despite Galilean invariance.
    Newly coined term to emphasize that the error should ideally be zero but is not.

pith-pipeline@v0.9.0 · 5697 in / 1502 out tokens · 62835 ms · 2026-05-24T22:35:23.801602+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.