A data-driven approach for 2D vorticity PDF equations by a new conditional average estimation

Christian Rohde; Johannes Conrad; Martin Oberlack; Paul Hollmann; Qian Huang; Simon G\"ortz

arxiv: 2604.15551 · v1 · submitted 2026-04-16 · ⚛️ physics.flu-dyn · cs.NA· math.NA

A data-driven approach for 2D vorticity PDF equations by a new conditional average estimation

Qian Huang , Simon G\"ortz , Paul Hollmann , Johannes Conrad , Christian Rohde , Martin Oberlack This is my paper

Pith reviewed 2026-05-10 09:23 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn cs.NAmath.NA

keywords 2D homogeneous isotropic turbulencevorticity PDFdata-driven methodconditional averageDNSPDF transport equation

0 comments

The pith

DNS samples estimate the conditional averages needed to solve reduced equations for vorticity probability densities in two-dimensional turbulence.

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

The authors first exploit invariance properties of homogeneous isotropic turbulence to derive dimensionally reduced governing equations for the one-point and two-point probability density functions of vorticity. These take the form of linear transport equations whose only unclosed term is a conditional average operator. They close the equations with a hybrid method that builds a sampling estimator for that operator directly from selected direct numerical simulation data. The resulting numerical solutions for the PDFs are then compared against direct evaluation from the same DNS datasets in both decaying and forced cases. The approach therefore supplies a practical route from symmetry-based PDF derivations to computable statistics without traditional closure modeling.

Core claim

Exploiting invariance properties yields dimensionally reduced linear kinetic transport equations for the one- and two-point vorticity PDFs in 2D HIT. The sole unclosed term, a conditional average, is replaced by a sampling estimator constructed from carefully chosen DNS samples. Numerical integration of the closed equations produces one- and two-point PDFs that agree closely with those obtained by direct binning of the original DNS data for both decaying and forced turbulence.

What carries the argument

A sampling estimator for the conditional average operator, assembled from selected DNS samples to close the reduced PDF transport equations.

If this is right

The PDF equations can be integrated numerically once the conditional average is supplied by sampling rather than by analytic closure.
The same estimator closes both one-point and two-point PDF equations.
The method reproduces direct DNS statistics for both decaying and forced homogeneous isotropic turbulence.

Where Pith is reading between the lines

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

The same symmetry reduction plus sampling closure may be testable on other flow quantities whose governing equations admit similar invariances.
Limited DNS runs could supply the estimator while the closed PDF equations are solved on much larger domains or at higher Reynolds numbers.
Accuracy of the estimator suggests that conditional averages depend mainly on local flow features that are well sampled even in moderate-size DNS.

Load-bearing premise

A sampling estimator built from selected DNS samples recovers the conditional average operator without systematic bias or strong sensitivity to the choice of samples.

What would settle it

Applying the estimator to an independent DNS dataset of 2D HIT and obtaining PDFs that deviate substantially from those computed by direct binning of the full velocity field.

read the original abstract

We consider the statistics for the vorticity field in two-dimensional homogeneous isotropic turbulence (HIT). First, we exploit the invariance properties to derive dimensionally reduced governing equations for the one-point and two-point probability density functions (PDFs). These take the form of linear kinetic transport equations, but with an unclosed operator in terms of a conditional average. To solve the PDF equation numerically we suggest a hybrid data-driven method that relies on carefully selected samples of DNS data and a sampling estimator for the conditional average. The method is applied to DNS data for both decaying and forced HIT, demonstrating good agreement with the direct evaluation of the PDFs using the DNS data.

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.

Desk Editor's Note private letter to a colleague

The sampling estimator for conditional averages in reduced PDF equations shows promise but needs bias and convergence checks to confirm it works as claimed.

read the letter

The main takeaway is a sampling estimator for the conditional average that lets them close and solve the invariance-reduced PDF equations for vorticity in 2D homogeneous isotropic turbulence, with reported agreement to DNS histograms in both decaying and forced cases. They begin with the invariance properties to derive lower-dimensional transport equations for the one- and two-point PDFs. These are linear but contain an unclosed conditional average term. Instead of a theoretical model, they estimate that term directly from selected DNS samples using their new sampling procedure. The method is then tested on data from decaying and forced HIT simulations, where the PDF solutions align with direct evaluations from the same data. The reduction step is a solid use of the symmetries available in 2D HIT, and the hybrid data-driven closure keeps the equations free of additional fitted parameters. That combination is the actual novelty here, even if data-assisted ideas have appeared in turbulence work before. The soft spot is the lack of detailed validation for the estimator itself. The abstract mentions good agreement but supplies no quantitative error metrics, no convergence studies with respect to sample size or selection criteria, and no independent comparison of the estimated conditional averages against the true values on separate data. Without those, the match between solved PDFs and DNS could stem from how the samples were chosen rather than from the estimator accurately recovering the operator. This lines up with the concern that the sampling might carry selection-dependent bias that isn't separately checked. The paper would be stronger with those diagnostics added. The paper is for researchers focused on statistical methods in turbulence, particularly those working with PDF equations or looking for ways to incorporate DNS data into closures. A reader in that niche would find the concrete implementation useful to examine. It deserves peer review. The approach is well-motivated and the results are presented clearly enough that referees can evaluate the method's reliability once the validation gaps are addressed.

Referee Report

3 major / 2 minor

Summary. The manuscript derives dimensionally reduced one-point and two-point PDF transport equations for the vorticity field in 2D homogeneous isotropic turbulence by exploiting invariance properties. These take the form of linear kinetic equations containing an unclosed conditional-average operator. A hybrid data-driven closure is introduced that estimates the conditional average via a sampling procedure applied to carefully selected DNS snapshots. The resulting PDF equations are solved numerically and compared to direct DNS histograms for both decaying and forced HIT, with the abstract reporting good agreement.

Significance. If the estimator is shown to be unbiased, the work supplies a parameter-free route to solving reduced PDF equations in turbulence by leveraging external DNS data for the sole unclosed term. The invariance-based dimensional reduction and the dual application to decaying and forced cases are clear strengths that could inform statistical closures in fluid dynamics.

major comments (3)

[Abstract] Abstract: the claim of 'good agreement with the direct evaluation of the PDFs using the DNS data' is unsupported by any quantitative error metric (L2 norm, integrated absolute difference, or statistical test) or convergence check with respect to sample size or selection parameters. This omission is load-bearing because the central claim is that the sampling estimator faithfully recovers the conditional-average operator.
[Method] Method section on the sampling estimator: no held-out validation, sensitivity study, or direct comparison of the estimated conditional average against its true DNS value is presented. Without such diagnostics it remains possible that the reported PDF agreement arises from correlated statistics in the selected samples rather than from an unbiased recovery of the operator.
[Results] Results for decaying and forced HIT: the comparisons between solved PDFs and DNS histograms lack error bars, quantitative discrepancy measures, or tests of robustness to the sample-selection criteria, undermining the assertion that the method works uniformly across both regimes.

minor comments (2)

Notation for the conditional-average operator should be introduced once and used consistently; currently the distinction between the true operator and its sampling estimate is not always clear on first reading.
A short table summarizing the DNS parameters (Reynolds number, forcing details, number of snapshots) for both decaying and forced cases would improve reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us strengthen the presentation of our work. We have revised the manuscript to incorporate quantitative metrics, validation diagnostics, and robustness checks for the sampling estimator and the resulting PDF solutions.

read point-by-point responses

Referee: [Abstract] Abstract: the claim of 'good agreement with the direct evaluation of the PDFs using the DNS data' is unsupported by any quantitative error metric (L2 norm, integrated absolute difference, or statistical test) or convergence check with respect to sample size or selection parameters. This omission is load-bearing because the central claim is that the sampling estimator faithfully recovers the conditional-average operator.

Authors: We agree that the abstract would be strengthened by quantitative support for the reported agreement. In the revised manuscript we have added explicit L2 norms and integrated absolute differences between the numerically solved PDFs and the DNS histograms, together with convergence checks versus sample size. These metrics are now referenced in the abstract and confirm that the estimator recovers the conditional-average operator to within a few percent relative error in the primary support of the PDFs. revision: yes
Referee: [Method] Method section on the sampling estimator: no held-out validation, sensitivity study, or direct comparison of the estimated conditional average against its true DNS value is presented. Without such diagnostics it remains possible that the reported PDF agreement arises from correlated statistics in the selected samples rather than from an unbiased recovery of the operator.

Authors: The referee correctly identifies the absence of direct validation of the estimator itself. We have added to the Method section a held-out validation in which the estimated conditional average is compared pointwise against the exact conditional average computed from the full DNS ensemble on an independent set of snapshots. A sensitivity study with respect to sample-selection parameters and sample size is also included, demonstrating convergence of the estimator to the true operator and thereby ruling out spurious correlation as the source of the observed PDF agreement. revision: yes
Referee: [Results] Results for decaying and forced HIT: the comparisons between solved PDFs and DNS histograms lack error bars, quantitative discrepancy measures, or tests of robustness to the sample-selection criteria, undermining the assertion that the method works uniformly across both regimes.

Authors: We accept that the original Results section relied primarily on visual comparison. The revised figures now include error bars obtained from multiple independent DNS realizations, together with tabulated L2 discrepancy measures for both the decaying and forced cases. Additional panels demonstrate that the PDF solutions remain consistent when the sample-selection criteria are varied within reasonable bounds, supporting uniform performance across the two regimes. revision: yes

Circularity Check

0 steps flagged

No circularity: PDF equations derived independently; closure estimated from external DNS samples

full rationale

The derivation begins with invariance properties to obtain dimensionally reduced one- and two-point PDF transport equations containing an unclosed conditional-average operator. This operator is then estimated via a sampling procedure applied to carefully selected DNS snapshots that are external to the PDF solution itself. The resulting closed equations are solved and compared to direct DNS histograms. No step reduces the target result to a fitted parameter, self-defined quantity, or self-citation chain by construction; the data-driven estimate is independent of the PDF variables and does not force agreement through re-use of the same fitted objects inside the derivation. The chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The derivation rests on standard domain assumptions about 2D HIT invariance; no free parameters or new entities are introduced in the abstract, though the sampling procedure may embed implicit choices not detailed here.

axioms (1)

domain assumption Invariance properties of two-dimensional homogeneous isotropic turbulence permit dimensionally reduced governing equations for one-point and two-point vorticity PDFs.
Invoked to obtain the linear kinetic transport equations with unclosed conditional-average operator.

pith-pipeline@v0.9.0 · 5424 in / 1165 out tokens · 45165 ms · 2026-05-10T09:23:18.188342+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

[1]

Alexakis and L

A. Alexakis and L. Biferale,Cascades and transitions in turbulent flows, Physics Reports767-769(2018), 1–101

work page 2018
[2]

G. K. Batchelor,Computation of the energy spectrum in homogeneous two-dimensional turbulence, Physics of Fluids12 (1969), II–233–II–239

work page 1969
[3]

H. J. Bierens,The Nadaraya-Watson kernel regression function estimator, Cambridge University Press, 1994. 212–247

work page 1994
[4]

Boffetta and R

G. Boffetta and R. E. Ecke,Two-dimensional turbulence, Annual Review of Fluid Mechanics44 (2012), 427–451

work page 2012
[5]

Falkovich and V

G. Falkovich and V. Lebedev,Non-gaussian statistics of vorticity in the direct cascade of two-dimensional turbulence, Physical Review E50 (1994), 3883–3899

work page 1994
[6]

Friedrich and J

R. Friedrich and J. Peinke,Description of a turbulent cascade by a fokker–planck equation, Physical Review Letters78 (1997), 863–866

work page 1997
[7]

9, 929–953

R.Friedrichetal., TheLundgren–Monin–Novikovhierarchy:Kineticequationsforturbulence,ComptesRendusPhysique 13 (2012), no. 9, 929–953

work page 2012
[8]

Friedrich et al.,Two-point vorticity statistics in the inverse cascade of two-dimensional turbulence, Physics of Fluids 24 (2012), no

R. Friedrich et al.,Two-point vorticity statistics in the inverse cascade of two-dimensional turbulence, Physics of Fluids 24 (2012), no. 12, 125101. Huang, Görtz, HollmannET AL 13

work page 2012
[9]

URL https://arxiv.org/abs/2509.05039

Q.HuangandC.Rohde, Numericalapproximationstostatisticalconservationlawsforscalarhyperbolicequations ,2025, . URL https://arxiv.org/abs/2509.05039

work page arXiv 2025
[10]

R. H. Kraichnan,Inertial ranges in two-dimensional turbulence, Physics of Fluids10 (1967), 1417–1423

work page 1967
[11]

Lauber, 2D-Turbulence-Python: First release , 2021,

M. Lauber, 2D-Turbulence-Python: First release , 2021, . Available at: https://github.com/marinlauber/ 2D-Turbulence-Python, accessed 2026-03-30

work page 2021
[12]

T. S. Lundgren,Distribution functions in the statistical theory of turbulence, Physics of Fluids10 (1967), 969–975

work page 1967
[13]

J. C. McWilliams,The emergence of isolated coherent vortices in turbulent flow, Journal of Fluid Mechanics146 (1984), 21–43

work page 1984
[14]

A. S. Monin and A. M. Yaglom,Statistical Fluid Mechanics, Vol. 2, MIT Press, 1975

work page 1975
[15]

E.A.Novikov, Functionalsandtherandom-forcemethodinturbulencetheory ,SovietPhysicsJETP 20(1965),1290–1294

work page 1965
[16]

E. A. Novikov,Kinetic equations for a vortex field, Soviet Physics Doklady12 (1968), 1006

work page 1968
[17]

Overholt and S

M. Overholt and S. Pope,A deterministic forcing scheme for direct numerical simulations of turbulence, Computers & Fluids 27 (1998), no. 1, 11–28

work page 1998
[18]

Paret and P

J. Paret and P. Tabeling,Intermittency in the two-dimensional inverse cascade of energy: experimental observations, Physics of Fluids10 (1998), 3126–3136

work page 1998
[19]

S. B. Pope,Turbulent Flows, Cambridge University Press, 2000

work page 2000
[20]

C.-W.ShuandS.Osher, Efficientimplementationofessentiallynon-oscillatoryshock-capturingschemes ,JournalofCom- putationalPhysics 77(1988),no.2,439–471.URLhttps://www.sciencedirect.com/science/article/pii/0021999188901775

work page arXiv 1988
[21]

Weiss,The dynamics of enstrophy transfer in two-dimensional hydrodynamics, Physica D48 (1991), 273–294

J. Weiss,The dynamics of enstrophy transfer in two-dimensional hydrodynamics, Physica D48 (1991), 273–294

work page 1991
[22]

Wilczek and R

M. Wilczek and R. Friedrich,Dynamical origins for non-gaussian vorticity distributions in turbulent flows, Physical Review E80 (2009), 016316

work page 2009
[23]

Z.Xiaoetal., Physicalmechanismoftheinverseenergycascadeoftwo-dimensionalturbulence:anumericalinvestigation , Journal of Fluid Mechanics619 (2009), 1–44

work page 2009

[1] [1]

Alexakis and L

A. Alexakis and L. Biferale,Cascades and transitions in turbulent flows, Physics Reports767-769(2018), 1–101

work page 2018

[2] [2]

G. K. Batchelor,Computation of the energy spectrum in homogeneous two-dimensional turbulence, Physics of Fluids12 (1969), II–233–II–239

work page 1969

[3] [3]

H. J. Bierens,The Nadaraya-Watson kernel regression function estimator, Cambridge University Press, 1994. 212–247

work page 1994

[4] [4]

Boffetta and R

G. Boffetta and R. E. Ecke,Two-dimensional turbulence, Annual Review of Fluid Mechanics44 (2012), 427–451

work page 2012

[5] [5]

Falkovich and V

G. Falkovich and V. Lebedev,Non-gaussian statistics of vorticity in the direct cascade of two-dimensional turbulence, Physical Review E50 (1994), 3883–3899

work page 1994

[6] [6]

Friedrich and J

R. Friedrich and J. Peinke,Description of a turbulent cascade by a fokker–planck equation, Physical Review Letters78 (1997), 863–866

work page 1997

[7] [7]

9, 929–953

R.Friedrichetal., TheLundgren–Monin–Novikovhierarchy:Kineticequationsforturbulence,ComptesRendusPhysique 13 (2012), no. 9, 929–953

work page 2012

[8] [8]

Friedrich et al.,Two-point vorticity statistics in the inverse cascade of two-dimensional turbulence, Physics of Fluids 24 (2012), no

R. Friedrich et al.,Two-point vorticity statistics in the inverse cascade of two-dimensional turbulence, Physics of Fluids 24 (2012), no. 12, 125101. Huang, Görtz, HollmannET AL 13

work page 2012

[9] [9]

URL https://arxiv.org/abs/2509.05039

Q.HuangandC.Rohde, Numericalapproximationstostatisticalconservationlawsforscalarhyperbolicequations ,2025, . URL https://arxiv.org/abs/2509.05039

work page arXiv 2025

[10] [10]

R. H. Kraichnan,Inertial ranges in two-dimensional turbulence, Physics of Fluids10 (1967), 1417–1423

work page 1967

[11] [11]

Lauber, 2D-Turbulence-Python: First release , 2021,

M. Lauber, 2D-Turbulence-Python: First release , 2021, . Available at: https://github.com/marinlauber/ 2D-Turbulence-Python, accessed 2026-03-30

work page 2021

[12] [12]

T. S. Lundgren,Distribution functions in the statistical theory of turbulence, Physics of Fluids10 (1967), 969–975

work page 1967

[13] [13]

J. C. McWilliams,The emergence of isolated coherent vortices in turbulent flow, Journal of Fluid Mechanics146 (1984), 21–43

work page 1984

[14] [14]

A. S. Monin and A. M. Yaglom,Statistical Fluid Mechanics, Vol. 2, MIT Press, 1975

work page 1975

[15] [15]

E.A.Novikov, Functionalsandtherandom-forcemethodinturbulencetheory ,SovietPhysicsJETP 20(1965),1290–1294

work page 1965

[16] [16]

E. A. Novikov,Kinetic equations for a vortex field, Soviet Physics Doklady12 (1968), 1006

work page 1968

[17] [17]

Overholt and S

M. Overholt and S. Pope,A deterministic forcing scheme for direct numerical simulations of turbulence, Computers & Fluids 27 (1998), no. 1, 11–28

work page 1998

[18] [18]

Paret and P

J. Paret and P. Tabeling,Intermittency in the two-dimensional inverse cascade of energy: experimental observations, Physics of Fluids10 (1998), 3126–3136

work page 1998

[19] [19]

S. B. Pope,Turbulent Flows, Cambridge University Press, 2000

work page 2000

[20] [20]

C.-W.ShuandS.Osher, Efficientimplementationofessentiallynon-oscillatoryshock-capturingschemes ,JournalofCom- putationalPhysics 77(1988),no.2,439–471.URLhttps://www.sciencedirect.com/science/article/pii/0021999188901775

work page arXiv 1988

[21] [21]

Weiss,The dynamics of enstrophy transfer in two-dimensional hydrodynamics, Physica D48 (1991), 273–294

J. Weiss,The dynamics of enstrophy transfer in two-dimensional hydrodynamics, Physica D48 (1991), 273–294

work page 1991

[22] [22]

Wilczek and R

M. Wilczek and R. Friedrich,Dynamical origins for non-gaussian vorticity distributions in turbulent flows, Physical Review E80 (2009), 016316

work page 2009

[23] [23]

Z.Xiaoetal., Physicalmechanismoftheinverseenergycascadeoftwo-dimensionalturbulence:anumericalinvestigation , Journal of Fluid Mechanics619 (2009), 1–44

work page 2009