It\^o tracers: continuous-trajectory Lagrangian particles for Eulerian hydrodynamics

Eric R. Moseley; R. Teyssier; Tom Abel

arxiv: 2604.23041 · v2 · submitted 2026-04-24 · 🌌 astro-ph.GA

It\^o tracers: continuous-trajectory Lagrangian particles for Eulerian hydrodynamics

Eric R. Moseley , R. Teyssier , Tom Abel This is my paper

Pith reviewed 2026-05-08 10:33 UTC · model grok-4.3

classification 🌌 astro-ph.GA

keywords Itô tracersLagrangian particlesEulerian hydrodynamicsnumerical diffusionMonte Carlo tracersstochastic differential equationsastrophysical fluid dynamicsturbulence simulations

0 comments

The pith

Itô tracers evolve via stochastic differential equations to match the advection and numerical diffusion of Eulerian gas on a grid.

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

The paper develops Itô tracers as continuous-time Lagrangian particles whose positions follow a stochastic differential equation whose drift and diffusion terms are calibrated to reproduce the first and second moments of the gas distribution, including numerical diffusion from the hydrodynamics scheme. This removes the bias of classical advected particles that ignore grid diffusion while avoiding the discontinuous jumps required by Monte Carlo tracers that sample fluxes probabilistically. Validation on a one-dimensional square-pulse advection problem and three-dimensional decaying turbulence at Mach 15 shows that the new particles recover column-density maps, joint density histograms, log-density-ratio PDFs, and density power spectra at least as accurately as Monte Carlo tracers. The continuous formulation also admits variance-reduction methods, higher-order integrators, and direct coupling to other continuous Lagrangian processes such as dust or cosmic rays.

Core claim

Itô tracers are continuous-trajectory Lagrangian particles governed by an Itô stochastic differential equation whose coefficients are set to match the advection, numerical diffusion, and dispersion operators acting on the underlying Eulerian gas. In the one-dimensional advection test the particles follow the square pulse without artificial spreading beyond the grid diffusion; in three-dimensional decaying turbulence they improve the gas-tracer density correlation by more than 3 percent and halve the width of the log-density-ratio PDF relative to Monte Carlo tracers, with gains of at least 30 percent and 230 percent over classical tracers.

What carries the argument

The Itô stochastic differential equation for particle position, with drift set by the cell-centered velocity and diffusion coefficient set by the magnitude of numerical or Smagorinsky-Lilly turbulent diffusivity, ensuring moment matching to the Eulerian scheme.

If this is right

Itô tracers improve the gas-tracer density correlation by more than 3 percent and reduce the log-density-ratio PDF width by nearly 50 percent relative to Monte Carlo tracers in decaying turbulence.
Relative to classical advected particles the same metrics improve by at least 30 percent and 230 percent, respectively.
A subgrid-scale variant using Smagorinsky-Lilly diffusivity shows that the magnitude of diffusion dominates its precise functional form.
The continuous trajectory formulation directly supports variance-reduction techniques and higher-order integrators unavailable to discrete-jump schemes.
The same stochastic differential equation maps onto other continuous Lagrangian processes such as dust grains or cosmic rays.

Where Pith is reading between the lines

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

The moment-matching construction suggests that simple isotropic diffusion models may be adequate for tracer accuracy even when the underlying turbulence is anisotropic.
Continuous trajectories open the possibility of coupling Itô tracers to additional forces or source terms (gravity, chemistry, radiation) within a single integrator.
Variance-reduction methods applied to the stochastic equation could reduce the number of particles needed to achieve a target statistical precision in large-volume simulations.
The approach could be tested in driven turbulence or adaptive-mesh-refinement runs to check whether the reported improvements persist when numerical diffusion changes with resolution.

Load-bearing premise

Matching only the first two moments of the tracer distribution via the stochastic differential equation is sufficient to reproduce all statistical effects of the discrete grid operations, including any higher-order biases.

What would settle it

A 3D turbulence run in which the joint density histogram or small-scale power spectrum of tracer density deviates from the gas distribution by more than sampling noise after the reported correlation and PDF-width improvements are achieved.

Figures

Figures reproduced from arXiv: 2604.23041 by Eric R. Moseley, R. Teyssier, Tom Abel.

**Figure 1.** Figure 1: Thus, the Itˆo-3 tracer matches the behavior of the gas up to the third order dispersive term, and evolves according to the following Itˆo process: dXt = u dt + √ 2κ dWt , (40) where u and κ are defined according to Eqs. 28 and 29, and the stochastic increment Wt is a L´evy process with its increments PSU distributed with mean 0, variance δt, and skewness γ(δt). 2.1.7 The Itˆo-n tracer hierarchy As brief… view at source ↗

**Figure 2.** Figure 2: The square pulse advection test using the first-order Godunov method, presented against classical tracers and a “trivial” Itˆo tracer with constant advection and diffusion (no interpolation of gas quantities). Initial conditions are as described in Sec. 3.2. The square pulse has been advected to the right for one period. The effective particle count here is between 262, 144 and 524, 288 particles per cel… view at source ↗

**Figure 3.** Figure 3: Top: The same square pulse test as view at source ↗

**Figure 5.** Figure 5: Particle trajectories for the MC tracer, classical tracer, and Itˆo-2 and Itˆo-3 tracers for the square pulse advection test described in Sec. 3.2. As the classical tracer (blue) simply interpolates cell-centered velocities, it uniformly proceeds from left to right. The MC tracer (green) can only stay put or jump to the right at each timestep in this test, and so its trajectory looks like a staircase with… view at source ↗

**Figure 6.** Figure 6: The distribution of particles initially located in a sheet located from z = 63h → 64h after one advection period. These particles are part of the square pulse test in Sec. 3.2. The distribution of all stochastic particle types is approximately equal to the analytic Gaussian, differing mainly at the tails where the advection and diffusion coefficients vary within the transition regions of the pulse from h… view at source ↗

**Figure 7.** Figure 7: Top: Column densities (normalized by the mean) of gas, MC, Itˆo-3, Itˆo-2, SGS-Itˆo, and classical tracers for the decaying turbulence test described in Sec. 3.3, taken at t = 0.2 L0/cs . Bottom: ratio of tracer column density to gas column density for the same particles. The MC tracers are those implemented and run in RAMSES (Cadiou et al. 2019; Teyssier 2002), while the other tracers here are run with id… view at source ↗

**Figure 8.** Figure 8: Joint density histograms of MC, Itˆo-3, Itˆo-2, SGS-Itˆo, and classical tracers vs. gas. The solid gray line shows the 1:1 tracer:gas density. Dashed lines show the 90% confidence region as computed from the Poisson distribution, while dotted lines show the 99.9% confidence region. Each plot is also labeled with a best fit slope, an R 2 value for that slope, and an R 2 1:1 value computed assuming that the … view at source ↗

**Figure 11.** Figure 11: Similar to view at source ↗

**Figure 10.** Figure 10: Density power spectra vs. wavenumber k for the gas, as well as the various tracer methods. Classical tracers show elevated power across a large range of scales, while Itˆo tracers tend to show suppressed power. MC tracers tend to have exact agreement in power until one reaches small scales, where Poisson noise creates increased power. On sufficiently large scales, all methods tend to agree with the gas. t… view at source ↗

read the original abstract

Lagrangian tracer particles have long been used to track the history of individual gas parcels in hydrodynamical codes. Particles advected by the cell-centered velocity carry no representation of underlying numerical diffusion, and thus exhibit systematic bias. The Monte-Carlo (MC) tracer resolves this with discrete probabilistic cell-to-cell, flux-based jumps, at the cost of trajectories that are discontinuous in time. We introduce the It\^o tracer, a continuous-time Lagrangian particle with moments matched to the advection, diffusion, and dispersion of the gas. A subgrid-scale variant (SGS-It\^o) replaces the numerical diffusion with a Smagorinsky--Lilly turbulent diffusivity, illustrating that the form of the diffusion matters less than its magnitude. We validate these methods with a 1D square-pulse advection test and 3D decaying turbulence at $\sigma_{\rm rms} = 15\,c_{\rm s}$. We compare the different tracer particle methods using several statistical tests. It\^o tracers largely reproduce or improve upon MC tracers statistics across column-density maps, joint density histograms, log-density-ratio PDFs, and density power spectra. In the turbulence test, It\^o tracers improve the correlation between tracers and gas over the MC tracers by >3\%, and reduce the width of the log-density ratio PDF by nearly 50\%. Relative to classical tracers, these improvements are $\gtrsim$30\% and 230\%, respectively. Because It\^o tracers follow a stochastic differential equation, the method maps onto other continuous-trajectory Lagrangian processes (e.g. dust grains, charged particles, cosmic rays), admits variance-reduction techniques, higher-order integrators, and GPU-friendly implementations -- all of which are unavailable to discrete-jump schemes.

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

Itô tracers give continuous paths that beat Monte Carlo tracers on the reported stats in advection and turbulence tests, though the moment-matching approach leaves open questions about higher-order grid effects.

read the letter

The main takeaway is that the authors have developed Itô tracers, which are continuous-trajectory Lagrangian particles whose motion follows a stochastic differential equation tuned to reproduce the advection and numerical diffusion of the underlying Eulerian hydrodynamics. This is new because previous methods either ignored diffusion entirely or used discrete probabilistic jumps. The continuous form allows for things like higher-order integrators and variance reduction that discrete schemes cannot use. They also show a subgrid-scale variant that swaps in a turbulent diffusivity model. The paper does well on the validation side. In both the 1D advection test and the 3D turbulence simulation, the Itô tracers produce statistics closer to the gas than Monte Carlo tracers do. The gains show up in correlation measures, density ratio distributions, and power spectra, with the numbers indicating clear improvements over classical tracers and modest ones over MC. A soft spot is that the method rests on matching only the first and second moments. Discrete grid effects from the hydro solver, such as limiters or dispersion, might affect higher-order behavior in ways not captured by the SDE. The tests are limited to specific cases, so the generality is not fully established yet. Still, the reported results suggest the approach works reasonably for the quantities they checked. This is relevant for anyone doing astrophysical fluid simulations where tracking the history of gas parcels matters, like in studies of chemical enrichment or star formation. Readers focused on numerical methods will appreciate the direct comparisons. It deserves a serious referee. The work is grounded in standard stochastic calculus and provides usable improvements. I would recommend sending it for peer review.

Referee Report

3 major / 2 minor

Summary. The manuscript introduces Itô tracers as continuous-trajectory Lagrangian particles whose stochastic differential equation is constructed to match the first and second moments (advection, diffusion, and dispersion) of the underlying Eulerian hydrodynamics, including numerical effects. This is positioned as an improvement over classical velocity-advected tracers (which ignore numerical diffusion) and Monte-Carlo tracers (which use discontinuous flux-based jumps). Validation consists of a 1D square-pulse advection test and 3D decaying turbulence at σ_rms = 15 c_s, with comparisons via column-density maps, joint density histograms, log-density-ratio PDFs, and density power spectra. The central empirical claim is that Itô tracers largely reproduce or improve MC-tracer statistics, with quantified gains (>3% correlation improvement and ~50% narrower PDF width) relative to MC tracers and larger gains relative to classical tracers. A subgrid-scale (SGS-Itô) variant using Smagorinsky-Lilly diffusivity is also presented.

Significance. If the moment-matching approach generalizes, the method supplies a continuous Lagrangian tracer that better captures numerical diffusion while enabling direct mapping to other continuous processes (dust, cosmic rays, charged particles), variance-reduction techniques, higher-order integrators, and GPU implementations unavailable to discrete-jump schemes. The reported improvements in turbulence statistics would be valuable for applications requiring accurate gas-parcel histories, such as chemical evolution or mixing studies.

major comments (3)

[Itô tracer formulation and method description] The core modeling assumption—that matching the first two moments of the tracer probability distribution via an Itô SDE is sufficient to reproduce the full set of reported statistics, including higher-order effects arising from discrete finite-volume operations (slope limiters, Riemann-solver dispersion, grid interpolation)—is stated without formal justification or analysis of possible discrepancies. This assumption is load-bearing for the claim that the observed improvements are a general consequence of the method rather than test-specific.
[3D turbulence test results] In the turbulence validation results, the reported improvements (>3% correlation gain and ~50% reduction in log-density-ratio PDF width relative to MC tracers) are given as point values without error bars, bootstrap uncertainties, or details on the number of independent realizations or tracer counts, preventing assessment of whether the gains exceed statistical fluctuations.
[Itô tracer formulation] The standard Itô implementation requires the diffusion coefficient to be taken from the numerical fluxes of the hydro solver, yet the manuscript provides no explicit algorithm, pseudocode, or equation showing how this coefficient is extracted or interpolated onto particle positions. This choice is load-bearing for reproducibility and for understanding why the SGS-Itô variant (which replaces it with a Smagorinsky-Lilly model) yields comparable results.

minor comments (2)

[Abstract] The abstract refers to 'dispersion of the gas' without a brief parenthetical definition or reference to the relevant term in the SDE; adding this would improve accessibility.
[Throughout] Notation for the stochastic terms (drift vector, diffusion tensor) should be made fully consistent between the textual description and any displayed equations.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We address each major comment below, agreeing where revisions are needed to improve clarity, reproducibility, and statistical rigor. We propose targeted changes to the manuscript while defending the core approach on substantive grounds.

read point-by-point responses

Referee: [Itô tracer formulation and method description] The core modeling assumption—that matching the first two moments of the tracer probability distribution via an Itô SDE is sufficient to reproduce the full set of reported statistics, including higher-order effects arising from discrete finite-volume operations (slope limiters, Riemann-solver dispersion, grid interpolation)—is stated without formal justification or analysis of possible discrepancies. This assumption is load-bearing for the claim that the observed improvements are a general consequence of the method rather than test-specific.

Authors: We acknowledge that the manuscript presents the moment-matching construction without a formal theorem proving equivalence for all higher-order statistics induced by discrete hydro operations. The Itô SDE is derived to reproduce the exact advection and diffusion operators of the underlying finite-volume scheme in the continuous limit (via the Fokker-Planck equation), which is the standard justification in stochastic calculus for advection-diffusion processes. In practice, the 1D and 3D tests show that this captures the dominant numerical effects, including dispersion from the Riemann solver and interpolation. We agree that a more explicit discussion of the assumptions and potential discrepancies would strengthen the paper. We will add a dedicated paragraph in the methods section referencing the central-limit behavior of the underlying stochastic process and noting that full equivalence for arbitrary higher moments is not guaranteed, while emphasizing that the reported statistics (density PDFs, power spectra, correlations) are primarily sensitive to the first two moments. revision: partial
Referee: [3D turbulence test results] In the turbulence validation results, the reported improvements (>3% correlation gain and ~50% reduction in log-density-ratio PDF width relative to MC tracers) are given as point values without error bars, bootstrap uncertainties, or details on the number of independent realizations or tracer counts, preventing assessment of whether the gains exceed statistical fluctuations.

Authors: The referee is correct that the turbulence results are presented as single-point estimates without uncertainty quantification. The simulations used a single realization of decaying turbulence with 10^6 tracers per method; the quoted improvements are measured directly from that run. We will revise the results section to include bootstrap uncertainties on the correlation coefficients and PDF widths (computed by resampling the tracer ensemble), report the exact tracer count and number of independent hydro realizations (one for the main run, with a second lower-resolution check), and add error bars to the relevant figures and text. This will allow readers to assess whether the gains are statistically significant. revision: yes
Referee: [Itô tracer formulation] The standard Itô implementation requires the diffusion coefficient to be taken from the numerical fluxes of the hydro solver, yet the manuscript provides no explicit algorithm, pseudocode, or equation showing how this coefficient is extracted or interpolated onto particle positions. This choice is load-bearing for reproducibility and for understanding why the SGS-Itô variant (which replaces it with a Smagorinsky-Lilly model) yields comparable results.

Authors: We agree that an explicit description of how the local diffusion coefficient is obtained from the hydro solver's numerical fluxes and interpolated to particle locations is essential for reproducibility. The current manuscript describes the principle but omits the concrete extraction step. We will add a new subsection (or appendix) containing the precise equation for the diffusion coefficient D_i = (1/2) * (flux-based variance per time step) evaluated at cell centers, together with the interpolation scheme (linear or higher-order) used to evaluate D at particle positions. We will also include pseudocode for the full Itô update step. This addition will clarify why the SGS-Itô variant, which substitutes a Smagorinsky-Lilly estimate for the same magnitude, produces statistically similar results. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper constructs the Itô tracer SDE by directly matching the first and second moments (advection, diffusion, dispersion) of a continuous stochastic process to the Eulerian hydro equations, using standard Itô calculus applied to the advection-diffusion PDE. This is an independent derivation, not a self-definition or relabeling of fitted inputs as predictions. Validation proceeds via separate numerical experiments (1D square-pulse advection and 3D decaying turbulence) that compare independent statistical diagnostics (column-density maps, joint histograms, log-density-ratio PDFs, power spectra) against MC and classical tracers, with no reduction of the reported improvements (>3% correlation, ~50% narrower PDF) to the input assumptions by construction. No load-bearing self-citations, imported uniqueness theorems, or ansatzes smuggled via prior author work appear in the derivation; the SGS-Itô variant simply substitutes a standard Smagorinsky-Lilly diffusivity. The method remains externally falsifiable and self-contained against the benchmark tests.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach relies on standard stochastic differential equation theory and the assumption that moment matching captures tracer statistics; no new physical entities are introduced.

axioms (2)

domain assumption The gas evolution in the Eulerian hydro code is governed by an advection-diffusion equation including numerical diffusion.
This is the standard foundation for grid-based hydrodynamics codes.
ad hoc to paper Matching the first two moments of the tracer probability distribution is sufficient to reproduce the statistical properties of gas parcels.
This is the core design choice for the Itô tracer and SGS variant.

pith-pipeline@v0.9.0 · 5630 in / 1330 out tokens · 64237 ms · 2026-05-08T10:33:22.636430+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

Kloeden, E

Agertz O., et al., 2007, Monthly Notices of the Royal Astronom- ical Society, 380, 963 Birdsall C. K., Fuss D., 1969, Journal of Computational Physics, 3, 494 Boozer A. H., Kuo-Petravic G., 1981, The Physics of Fluids, 24, 851 Cadiou C., Dubois Y., Pichon C., 2019, Astronomy & Astro- physics, 621, A96 Colbrook M. J., Ma X., Hopkins P. F., Squire J., 2017,...

work page doi:10.1007/978-3-662-12616-5 2007
[2]

Princeton University Press Tillson H., Devriendt J., Slyz A., Miller L., Pichon C., 2015, Monthly Notices of the Royal Astronomical Society, 449, 4363 Tricco T

Optics, Fluids, Plasmas, Elasticity, Relativity, and Statistical Physics. Princeton University Press Tillson H., Devriendt J., Slyz A., Miller L., Pichon C., 2015, Monthly Notices of the Royal Astronomical Society, 449, 4363 Tricco T. S., Price D. J., 2012, Journal of Computational Physics, 231, 7214 Van Leer B., 1977, Journal of Computational Physics, 23...

work page arXiv 2015

[1] [1]

Kloeden, E

Agertz O., et al., 2007, Monthly Notices of the Royal Astronom- ical Society, 380, 963 Birdsall C. K., Fuss D., 1969, Journal of Computational Physics, 3, 494 Boozer A. H., Kuo-Petravic G., 1981, The Physics of Fluids, 24, 851 Cadiou C., Dubois Y., Pichon C., 2019, Astronomy & Astro- physics, 621, A96 Colbrook M. J., Ma X., Hopkins P. F., Squire J., 2017,...

work page doi:10.1007/978-3-662-12616-5 2007

[2] [2]

Princeton University Press Tillson H., Devriendt J., Slyz A., Miller L., Pichon C., 2015, Monthly Notices of the Royal Astronomical Society, 449, 4363 Tricco T

Optics, Fluids, Plasmas, Elasticity, Relativity, and Statistical Physics. Princeton University Press Tillson H., Devriendt J., Slyz A., Miller L., Pichon C., 2015, Monthly Notices of the Royal Astronomical Society, 449, 4363 Tricco T. S., Price D. J., 2012, Journal of Computational Physics, 231, 7214 Van Leer B., 1977, Journal of Computational Physics, 23...

work page arXiv 2015