Denoising Particle-In-Cell data via Smoothness-Increasing Accuracy-Conserving filters with application to Bohm speed computation

Matthew J. Picklo, Qi Tang, Yanzeng Zhang, Jennifer K. Ryan, Xian-Zhu Tang

TL;DR

This paper introduces the application of Smoothness-Increasing Accuracy-Conserving (SIAC) filters to denoise data from computationally expensive Particle-In-Cell (PIC) simulations. By effectively reducing statistical noise in plasma physics quantities, SIAC filters enable more accurate calculations of critical parameters like the Bohm speed with significantly less computational cost.

What Problem Does This Paper Address?

Simulating plasma physics is incredibly demanding due to the high dimensionality of the underlying equations. Particle-In-Cell (PIC) methods offer a scalable approach, but they suffer from inherent statistical noise arising from the finite number of particles used. This noise, which scales with 1/√N (where N is the number of particles per cell), can be particularly problematic when calculating higher-order velocity moments of the particle distribution function. These moments are crucial for understanding plasma behavior, such as determining thermodynamic variables, flow, and temperature. The noise can obscure subtle physical phenomena and necessitate extremely long simulation times or extensive data averaging to obtain reliable results, as demonstrated by the challenge of accurately computing the Bohm speed. This paper aims to address this by providing an effective denoising technique for PIC data that preserves physical accuracy and reduces computational burden.

Methods

The core of this work is the application of Smoothness-Increasing Accuracy-Conserving (SIAC) filters, originally developed for finite element method (FEM) data, to denoising PIC simulation outputs. The process involves several key steps:

  1. Data Initialization: Discrete PIC data is first converted into a continuous piecewise polynomial representation using Lagrange interpolants. This step creates a mesh over the pointwise grid, allowing for the application of convolution-based filtering.
  2. SIAC Kernel Construction: SIAC kernels, which are compactly supported piecewise polynomial functions, are designed. These kernels are characterized by parameters like spline order ($l$) and the number of moments ($r$) they can reproduce. For non-periodic data, position-dependent and generalized spline kernels are developed to handle boundary effects. An adaptive kernel scaling method is also introduced to better preserve sharp gradients near boundaries.
  3. Convolution: The initialized data is convolved with the constructed SIAC kernel. This operation selectively dampens high-frequency noise while preserving lower-frequency physical oscillations.
  4. Fourier Analysis: The Fourier transform of the SIAC kernel is analyzed to understand its frequency-damping characteristics and guide the selection of kernel parameters.

The novelty lies in adapting and extending SIAC filtering techniques, particularly the development of position-dependent and adaptive scaling kernels, to the specific challenges of denoising discrete, noisy PIC data.

Datasets

The paper utilizes data from existing PIC simulations. Specifically:

  • Periodic Boundary Condition Data: Data from VPIC code simulations of the whistler instability in magnetized plasma, driven by trapped electrons, is used. This data is described as having periodic boundary conditions.
  • Non-Periodic Boundary Condition Data: Data simulating a non-neutral sheath, which exhibits sharp gradients near boundaries, is used. This includes electron density ($n_e$), electron heat flux ($q_e$), and its spatial gradient ($dq_e/dx$).

The paper does not introduce new benchmark datasets but rather applies the SIAC filters to existing simulation outputs to demonstrate their effectiveness.

Key Insights and Ideas

  • SIAC Filters as a Powerful Denoising Tool for PIC Data: The paper convincingly demonstrates that SIAC filters can effectively remove the characteristic statistical noise in PIC simulations without distorting important physical features.
  • Preserving Physical Oscillations: A key success is the ability of SIAC filters to distinguish between high-frequency noise and genuine physical oscillations (like whistler waves), preserving the latter while attenuating the former.
  • Handling Boundary Gradients: The development of position-dependent kernels and adaptive scaling is crucial for accurately denoising data with sharp gradients near boundaries, which are common in plasma sheaths.
  • Reducing Computational Cost for Physics Discovery: The most significant implication is the potential to drastically reduce the computational cost of PIC simulations. By denoising data from shorter simulations or fewer time-averaged snapshots, critical physical laws (like the Bohm speed) can be reliably computed, saving significant resources.
  • Bohm Speed Sensitivity: The Bohm speed calculation is shown to be highly sensitive to noise, particularly in the electron heat flux gradient. SIAC filtering provides a robust way to handle this sensitivity.

Key Results

  • Periodic Data: SIAC filtering successfully damped high-frequency noise in periodic magnetic field data, matching the theoretical kernel response in Fourier space and preserving the physical whistler waves.
  • Non-Periodic Data (Boundary Gradients):
    • Position-dependent SIAC filters, especially when augmented with generalized splines and adaptive kernel scaling, were shown to preserve boundary layer behavior in electron density data.
  • Similar improvements were observed for electron heat flux and its gradient, although the gradient ($dq_e/dx$) remained highly oscillatory even after filtering, highlighting its inherent noisiness.
  • Bohm Speed Calculation: By applying SIAC filters with carefully chosen variable-specific scalings to short-time averaged data, the computed Bohm speed closely matched results obtained from extremely long-time averaged data (200,000 snapshots). This represents a significant reduction in computational cost, with the study suggesting a potential time saving of up to 98.7% (1.3% of the original time).

Limitations and Future Work

  • Limitations:
    • The paper notes that the limited smoothness of the shifting function in SIAC filters can restrict the smoothness of the filtered approximation at the interface of shifted and unshifted kernels.
  • The analytical Fourier representation of the SIAC kernel is only possible in the periodic case with constant scaling. For non-periodic and adaptively scaled filters, understanding their frequency effects requires further study.
  • The current approach does not automatically detect or resolve interior structures with sharp gradients like shocks.
  • Future Work:
    • Further investigation into the sensitivity of the Bohm speed calculation to noisy variables is planned.
  • Developing a method for SIAC filters to self-detect and resolve interior structures with sharp gradients (e.g., shocks) is a key future direction.
  • Quantifying the Fourier effects of adaptively scaled and position-dependent filters more thoroughly is also suggested.
  • Exploring the application of these filters to more complex dynamical systems.

Summary updated

Abstract

The simulation of plasma physics is computationally expensive because the underlying physical system is of high dimensions, requiring three spatial dimensions and three velocity dimensions. One popular numerical approach is Particle-In-Cell (PIC) methods owing to its ease of implementation and favorable scalability in high-dimensional problems. An unfortunate drawback of the method is the introduction of statistical noise resulting from the use of finitely many particles. In this paper we examine the application of the Smoothness-Increasing Accuracy-Conserving (SIAC) family of convolution kernel filters as denoisers for moment data arising from PIC simulations. We show that SIAC filtering is a promising tool to denoise PIC data in the physical space as well as capture the appropriate scales in the Fourier space. Furthermore, we demonstrate how the application of the SIAC technique reduces the amount of information necessary in the computation of quantities of interest in plasma physics such as the Bohm speed.
BibTeX
@article{picklo2024denoising,
  author = {Matthew J. Picklo and Qi Tang and Yanzeng Zhang and Jennifer K. Ryan and Xian-Zhu Tang},
  title = {{Denoising Particle-In-Cell data via Smoothness-Increasing Accuracy-Conserving filters with application to Bohm speed computation}},
  journal = {Journal of Computational Physics},
  year = {2024},
  doi = {10.1016/j.jcp.2024.112790},
}

Loading related content...

Figure