Beyond the Diffusion Coefficient: Propagators and Memory in Cosmic Ray Transport

arXiv:2604.11898 · astro-ph.HE · Submitted 2026-04-13 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.

Vera: I'm Vera.

Jocelyn: And I'm Jocelyn, here with Subrahmanyan.

Vera: We're looking at the paper: "Beyond the Diffusion Coefficient".

Paper summary: Vera: So, to recap, this paper titled "Beyond the Diffusion Coefficient: Propagators and Memory in Cosmic Ray Transport" argues that using just one diffusion coefficient is insufficient for describing cosmic ray transport because it ignores distinct underlying mechanisms present in complex media.

Jocelyn: The authors move beyond that by developing a propagator based framework centered on P of x t, which is the probability distribution of particle positions over time, or its Fourier Laplace transform P of k s. They claim this compact object is statistically complete and inherently exposes memory effects in the cosmic ray flux because it allows dependence on earlier gradients that are otherwise lost when we use coarse-grained models.

Subrahmanyan: It matters because we see these deviations from standard diffusion—both superdiffusion and subdiffusion—in particle tracing studies, especially in environments like pulsar wind nebulae or supernova remnants where transport is known to be highly non-Gaussian.

Vera: They use the Montroll Weiss formalism to measure this P of k s directly from particle trajectories, and they show that in the large scale limit, this yields a frequency dependent diffusivity K of s and a time domain memory kernel K of t that captures history dependence.

Jocelyn: The paper demonstrates that slow regions in static multiphase media act as bottlenecks regulating escape without necessarily dominating the total residence-time budget for the particles, which is an important distinction from simpler models.

Subrahmanyan: They also tackle the complication where cell sizes vary by introducing a matrix Montroll Weiss description, and they introduce an accelerated Monte Carlo method to reconstruct memory kernels via a Prony expansion representing broad kernels as a small number of local relaxation modes.

Vera: The core idea they push is that the full propagator P of k s provides a more informative description than relying on just a single diffusion coefficient because it retains the full space and time dependent spreading of the particle distribution.

Jocelyn: They also show how to extract reduced descriptions from this P of k s, specifically recovering the waiting time transform psi of s, and the joint step size waiting time kernel Phi of k s. This lets us look at different aspects of the transport process separately.

Subrahmanyan: This framework connects directly to our cosmological structure formation models because understanding how these particles propagate through structured media helps us constrain how energy moves across different scales in the universe.

Conclusion: Vera: So, looking at "Beyond the Diffusion Coefficient: Propagators and Memory in Cosmic Ray Transport," it’s about moving past the simple diffusion coefficient to understand the full physics of cosmic ray transport by using these propagators and memory kernels.

Jocelyn: The authors are essentially saying that even when things look like standard diffusion on a large scale, there's a hidden history of events influencing where the particles end up, which is what this propagator captures.

Subrahmanyan: For the bigger picture, this means that when we look at gamma-ray emissivity across different scales in our universe, we can get a better handle on how energy moves through these structures because we are accounting for these memory effects.

Vera: In simple terms, it means that instead of just measuring how fast particles spread out and ignoring the process, you are getting a much richer picture of what’s actually happening at the level of individual transport events.

Jocelyn: The authors provide tools to extract specific kernel information—like step size or waiting time distributions—from that full propagator P of k s, which lets us probe the detailed physics of scattering and trapping.

Subrahmanyan: This work gives us a way to connect the microscopic details of how particles scatter in a medium to the macroscopic observations we make about cosmic rays across vast distances, which is what we need for cosmology.

University of California, Santa Barbara

astro-ph.HE

Submitted: 2026-04-13

Updated: 2026-11-17

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: The paper moves beyond modeling cosmic ray (CR) transport using a single diffusion coefficient, recognizing that this approach only captures variance growth and fails to account for distinct…

Key concepts

Propagator-based Framework
This is a mathematical tool centered on $P(x, t), the probability distribution of particle positions over time. It is more powerful than simple diffusion because it accounts for how particles move through the medium, capturing complex spreading and history dependence that single diffusion coefficients miss.
Memory Effects in CR Flux
This refers to the fact that cosmic ray flux depends on earlier gradients in space and time. The propagator framework exposes this by allowing dependence on past conditions, showing that transport is not instantaneous but has a temporal 'memory'.
Montroll–Weiss Formalism
This is a technique used to directly measure the propagator $P(k, s)$ from observed particle trajectories. It allows researchers to calculate frequency-dependent diffusivities and time-domain memory kernels that explicitly show the history dependence of the CR flux.

Terminology

Summary

The paper moves beyond modeling cosmic ray (CR) transport using a single diffusion coefficient, recognizing that this approach only captures variance growth and fails to account for distinct underlying transport mechanisms present in realistic multiphase, structured, and time-dependent media. p. 0

Where the paper says it

Peng Oh Department of Physics, University of California, Santa Barbara, CA 93106, USA 15 April 2026 ABSTRACT Cosmic ray (CR) transport is usually modeled with a single diffusion coefficient, but this description captures only the growth of the variance and not the full transport process. Distinct transport mechanisms can share the same effective diffusion coefficient while producing different particle distributions and approaches to the diffusive limit.

The core development is a propagator-based framework centered on P(x, t), the probability distribution of particle positions, or its Fourier–Laplace transform P(k, s). p. 0
Where the paper says it

Particle-tracing studies also show clear departures from standard diffusion, including both superdiffusion and subdiffusion. We therefore develop a propagator-based framework centered on 𝑃(𝑥, 𝑡), the probability distribution of particle positions, or equivalently its Fourier–Laplace transform 𝑃(𝑘, 𝑠).

This compact object is statistically complete and inherently exposes memory effects in the CR flux by allowing dependence on earlier gradients that are otherwise coarse-grained away. p. 0
Where the paper says it

We therefore develop a propagator-based framework centered on 𝑃(𝑥, 𝑡), the probability distribution of particle positions, or equivalently its Fourier–Laplace transform 𝑃(𝑘, 𝑠). This object is compact and statistically complete, and naturally exposes memory: the CR flux can depend on earlier gradients when unresolved trapping or phase changes are coarse-grained away.

The framework utilizes the Montroll–Weiss formalism to measure P(k, s) directly from particle trajectories. p. 0

Where the paper says it

Using the Montroll–Weiss formalism, we show how to measure 𝑃(𝑘, 𝑠) directly from trajectories, how to recover the associated memory kernel, and how to represent broad kernels efficiently with a Prony expansion. Applied to a multiphase medium, the framework shows that slow regions can regulate escape without dominating the total residence-time budget.

In the large-scale limit, this yields a frequency-dependent diffusivity K tilde(s) and a time-domain memory kernel K(t), which explicitly captures history dependence in the flux—the fact that CR flux depends on earlier gradients. p. 1
Where the paper says it

Rather, we develop a practical framework linking microscopic trajectory statistics, full propagators, memory kernels, and macroscopic transport laws in a form that can be applied directly to simulations and, ultimately, to observations. We show how to measure 𝑃(𝑘, 𝑠) directly from trajectories without first reconstructing jump and waiting-time distributions; how the large-scale limit yields a frequency-dependent diffusivity 𝐾˜ (𝑠) and a time-domain memory kernel 𝐾(𝑡); and how transport through a multiphase medium naturally generates bottlenecks, space–time coupling, and delayed convergence to ordinary diffusion.

The paper demonstrates that slow regions in static multiphase media act as bottlenecks regulating escape without necessarily dominating the total residence-time budget. p. 1
Where the paper says it

We show how to measure 𝑃(𝑘, 𝑠) directly from trajectories without first reconstructing jump and waiting-time distributions; how the large-scale limit yields a frequency-dependent diffusivity 𝐾˜ (𝑠) and a time-domain memory kernel 𝐾(𝑡); and how transport through a multiphase medium naturally generates bottlenecks, space–time coupling, and delayed convergence to ordinary diffusion. In particular, we will show that slow regions can control the overall resistance to escape without necessarily dominating the total residence-time budget, and that timedependent resetting can keep the effective transport above the usual static harmonic-mean limit.

Furthermore, when cell sizes vary, space and time become coupled, necessitating a matrix Montroll-Weiss description. p. 2
Where the paper says it

Integrating over all possible jump positions 𝑥 ′ and jump times 𝜏, we obtain: 𝑃(𝑥, 𝑡) = 𝑆(𝑡)𝛿(𝑥) + ∫ 𝑡 0 𝑑𝜏 𝜓(𝑡 − 𝜏) ∫ ∞ −∞ 𝑑𝑥′ 𝜆(𝑥 − 𝑥 ′ )𝑃(𝑥 ′ , 𝜏) (4) If we Fourier transform equation 4 in space and Laplace transform in time, then from the convolution theorem, the convolutions in (𝑥 ′ , 𝜏) become multiplications in (𝑘, 𝑠): 𝑃(𝑘, 𝑠) = 𝑆˜(𝑠) + 𝜓˜(𝑠)𝜆ˆ(𝑘)𝑃(𝑘, 𝑠). (5) Solving for 𝑃(𝑘, 𝑠) gives us the celebrated Montroll-Weiss formula (Montroll & Weiss 1965): 𝑃(𝑘, 𝑠) = 𝑆˜(𝑠) 1 − 𝜆ˆ(𝑘)𝜓˜(𝑠) = 1 − 𝜓˜(𝑠) 𝑠 1 1 − 𝜆ˆ(𝑘)𝜓˜(𝑠) , (6) where we use equation 3 in the second step.

An accelerated Monte Carlo method is introduced to coarse-grain unresolved intra-cell scattering into event-driven transitions, allowing for the reconstruction of memory kernels via a Prony expansion representing broad kernels as a small number of local relaxation modes. p. 1
Where the paper says it

In §4, we apply this formalism to transport in a multiphase medium, where slow regions act as bottlenecks that regulate escape and memory. We introduce an accelerated Monte Carlo method that coarse-grains unresolved intra-cell scattering into eventdriven transitions.

The central conclusion is that the full propagator P(k, s) provides a more informative description than a single diffusion coefficient because it retains the full space- and time-dependent spreading of the particle distribution. p. 1
Where the paper says it

Unlike 𝜅, the propagator retains the full statistical content of the transport: how broad the particle distribution is, how fast it spreads, whether the profile is Gaussian, and whether the response depends on what happened earlier. Thus, even when transport ultimately converges to ordinary diffusion at late times when jump-length variances and mean trapping times are finite (Liang & Oh 2025), the diffusion coefficient 𝜅 captures only that final asymptotic limit, whereas the propagator 𝑃(𝑘, 𝑠) retains the physically important crossover toward it.

The paper further details how to extract reduced descriptions from P(k, s), showing that one can recover the waiting-time transform psi(s), the joint step-size/waiting time kernel phi(k, s), and in the large-scale limit, the effective transport kernel K tilde(s). p. 1

Where the paper says it

We show how to measure 𝑃(𝑘, 𝑠) directly from trajectories without first reconstructing jump and waiting-time distributions; how the large-scale limit yields a frequency-dependent diffusivity 𝐾˜ (𝑠) and a time-domain memory kernel 𝐾(𝑡); and how transport through a multiphase medium naturally generates bottlenecks, space–time coupling, and delayed convergence to ordinary diffusion. In particular, we will show that slow regions can control the overall resistance to escape without necessarily dominating the total residence-time budget, and that timedependent resetting can keep the effective transport above the usual static harmonic-mean limit.

Analysis of physical parameters reveals how they influence memory characteristics. p. 8

Where the paper says it

Beyond the Diffusion Coefficient: Propagators and Memory in Cosmic Ray Transport 9 Figure 6. The memory kernel 𝐾˜ (𝑠) in the Laplace domain for different subdiffusion stability parameters 𝛼 = 0.

Increasing the dynamic range delta = L max/L min broadens the hierarchy of relaxation timescales and increases the memory width W K. p. 9
Where the paper says it

compact: a broad, non-local kernel is reduced to a small number of modes. Second, it is physically interpretable: rather than a single relaxation time, the transport is described by an effective hierarchy of timescales 𝜏𝑖 , each weighted by 𝑎𝑖 .

Conversely, increasing the spatial exponent alpha shifts the medium toward smaller patches, compressing trap lifetimes and narrowing K(t). p. 1
Where the paper says it

Thus, even when transport ultimately converges to ordinary diffusion at late times when jump-length variances and mean trapping times are finite (Liang & Oh 2025), the diffusion coefficient 𝜅 captures only that final asymptotic limit, whereas the propagator 𝑃(𝑘, 𝑠) retains the physically important crossover toward it. Time dependence of the medium can also matter: if the structures that trap CRs evolve while the particles are still sampling them, the large-scale transport need not approach the static asymptotic limit (§4.

Varying parameters like q R and q tau primarily shifts asymptotic endpoints rather than altering the breadth of the crossover. p. 1
Where the paper says it

Unlike 𝜅, the propagator retains the full statistical content of the transport: how broad the particle distribution is, how fast it spreads, whether the profile is Gaussian, and whether the response depends on what happened earlier. Thus, even when transport ultimately converges to ordinary diffusion at late times when jump-length variances and mean trapping times are finite (Liang & Oh 2025), the diffusion coefficient 𝜅 captures only that final asymptotic limit, whereas the propagator 𝑃(𝑘, 𝑠) retains the physically important crossover toward it.

Recoverability of K tilde(s) from P(k, s) is tied to the kernel's shape; large contrast between asymptotic diffusivities or a broad frequency span pushes scale-independent transport to smaller k and s, shrinking the valid extraction window delta 10 s valid. Larger q R and delta reduce recoverability by increasing kernel contrast and broadening trap lifetime hierarchies, whereas larger q tau and steeper alpha improve recoverability by reducing asymptotic diffusivity disparity and suppressing rare large traps. p. 1
Where the paper says it

Thus, even when transport ultimately converges to ordinary diffusion at late times when jump-length variances and mean trapping times are finite (Liang & Oh 2025), the diffusion coefficient 𝜅 captures only that final asymptotic limit, whereas the propagator 𝑃(𝑘, 𝑠) retains the physically important crossover toward it. Time dependence of the medium can also matter: if the structures that trap CRs evolve while the particles are still sampling them, the large-scale transport need not approach the static asymptotic limit (§4.

In the context of static one-dimensional chains of patches, the residence time fraction in a patch is proportional to its stored column density N i, which depends on both local resistance (R loc i) and downstream resistance (R d i). p. 1

Where the paper says it

We show how to measure 𝑃(𝑘, 𝑠) directly from trajectories without first reconstructing jump and waiting-time distributions; how the large-scale limit yields a frequency-dependent diffusivity 𝐾˜ (𝑠) and a time-domain memory kernel 𝐾(𝑡); and how transport through a multiphase medium naturally generates bottlenecks, space–time coupling, and delayed convergence to ordinary diffusion. In particular, we will show that slow regions can control the overall resistance to escape without necessarily dominating the total residence-time budget, and that timedependent resetting can keep the effective transport above the usual static harmonic-mean limit.

When R d i much greater than R loc i, the residence time in that slow phase depends mainly on its size and downstream resistance, showing why slow phases can contribute strongly to total series resistance while contributing only modestly to the total residence-time budget. p. 1
Where the paper says it

We show how to measure 𝑃(𝑘, 𝑠) directly from trajectories without first reconstructing jump and waiting-time distributions; how the large-scale limit yields a frequency-dependent diffusivity 𝐾˜ (𝑠) and a time-domain memory kernel 𝐾(𝑡); and how transport through a multiphase medium naturally generates bottlenecks, space–time coupling, and delayed convergence to ordinary diffusion. In particular, we will show that slow regions can control the overall resistance to escape without necessarily dominating the total residence-time budget, and that timedependent resetting can keep the effective transport above the usual static harmonic-mean limit.

This steady-state picture is noted as being broken by renewal resets. p. 8
Where the paper says it

resent this reduced dynamics when the coarse-grained evolution can be organized around discrete renewal events. In this picture, the complicated motion within a trapping region or phase is compressed into a waiting time, while the transition to the next region defines a jump.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Direct measurement of full propagator from trajectory data allows AI systems to diagnose memory, space-time coupling, and departures from standard diffusion without needing a priori knowledge of jump or waiting-time distributions.

  2. The framework enables AI to reconstruct the time domain memory kernel K(t) using a Prony expansion fitted to the measured Laplace-space kernel Ktilde(s), providing a compact representation suitable for embedding into two-fluid CR-MHD models.

  3. The system can perform coarse-grained transport using an accelerated Monte Carlo method, bypassing microscopic timestep constraints and achieving speedups by orders of magnitude in simulating complex multiphase environments.

  4. AI systems can distinguish between local escape problems and global transport effects by analyzing secondary-to-primary ratios like B/C, allowing them to infer different effective diffusion coefficients for near-source versus large-scale observables.

  5. The framework allows for the construction of predictive models where early-time transport history—captured by the kernel—is explicitly included, enabling better predictions for CR feedback that depend on when and where pressure is deposited before reaching late-time limits.

Abstract

Cosmic ray (CR) transport is usually modeled with a single diffusion coefficient, but this description captures only the growth of the variance. Distinct transport mechanisms can share the same effective diffusion coefficient while producing different particle distributions and approaches to the diffusive limit. The distinction matters in multiphase, structured and time-dependent media, and is reflected in the observed variation of CR transport near pulsar wind nebulae, supernova remnants and molecular clouds, and in particle-tracing studies that show both superdiffusion and subdiffusion. We therefore develop a propagator-based framework centered on the probability distribution of particle positions, or equivalently its Fourier-Laplace transform. This object is compact and complete for the observable spatial transport. It naturally reveals if large scale transport has memory: when the CR flux depends on earlier rather than just the instantaneous pressure gradient. Such memory arises when dynamically relevant transport states, such as trapping or phase changes, are not retained in the macroscopic description. Using the Montroll-Weiss formalism, we measure the propagator directly from trajectories, recover the associated memory kernel, and represent broad kernels with a Prony expansion. Applied to a multiphase medium, the framework shows that slow regions can regulate escape without dominating the total residence time budget. We also introduce an accelerated Monte Carlo method for coarse-grained transport, showing that if trapping structures evolve while particles are still sampling them, the static long-time limit need not be reached. This provides a foundation for observational applications, particle-tracing measurements and CR-MHD closures.

Sources

Related papers