Comment on "Impact of particle number and cell-size in fully implicit charge- and energy-conserving particle-in-cell schemes" by N. Savard et al., Phys. Plasmas 32, 073903 (2025)

This paper refutes the conclusions of Savard et al. regarding the necessity of high particle counts in implicit particle-in-cell schemes by demonstrating that procedural diagnostic errors in their original study led to misleading results, which are corrected to show that their claims do not hold under independent scrutiny.

The Gist

The Big Picture: A Dispute Over a Recipe

Imagine two groups of chefs (scientists) trying to bake the perfect cake (a computer simulation of plasma physics).

• Chef Group A (Savard et al.) recently published a paper saying, "Our new baking method (ECC-IPIC) is great, but it fails miserably if you don't use a huge amount of flour (particles). If you try to save on flour, the cake turns into a mushy, inaccurate mess."
• Chef Group B (Chacón, Chen, and Ricketson) read that paper and said, "Wait a minute. We tried that same recipe, and we got a perfect cake even with less flour. We think Group A didn't mess up the recipe; they messed up the tasting."

This paper is Group B's formal rebuttal. They argue that Group A's conclusion—that the new method is inaccurate—is wrong because of how they measured the results, not because the method itself is flawed.

---

The Three Main Mistakes Group A Made

Group B identifies three specific "kitchen errors" that led Group A to the wrong conclusion.

1. The "Noisy Start" (Particle Initialization)

The Analogy: Imagine you are setting up a line of people to march in a parade.

• Group A's approach: They told the people to stand randomly. Some spots had 5 people, others had 20, just to get an "average" of 10. This created a bumpy, uneven line right from the start.
• Group B's fix: They carefully placed exactly the right number of people in every spot to match the density perfectly.
• Why it matters: If your starting line is messy, your parade will look messy later. Group B found that if you start with a clean, organized line (using a "mass-matrix" approach), the simulation works much better.

2. The "Blurred Photo" (Ensemble Averaging)

The Analogy: Imagine taking a photo of a fast-moving race car.

• Group A's approach: They took 10 photos of 10 different races, each with a slightly different car speed, and then stacked all 10 photos on top of each other to make one "average" image.
• The result: The sharp edges of the car (the shockwave) got blurred and smeared out. The car looked like a fuzzy ghost. They looked at this blurry ghost and said, "See? The car isn't sharp! Our method is bad!"
• Group B's fix: Instead of taking 10 photos with fewer details, they took one single photo with 10 times the detail (more particles).
• The result: The car was razor-sharp. Group B argues that Group A's "stacking" technique smoothed out the important details, making a good simulation look bad.

3. The "Wrong Ruler" (Diagnostic Errors)

The Analogy: Imagine measuring the distance a runner traveled.

• Group A's approach: They measured the runner's position exactly where they thought the runner should be. But because of tiny timing differences, the runner was actually 1 inch to the left. Group A measured the gap between the "expected" spot and the "actual" spot and called it a huge error.
• Group B's fix: They realized the runner was just slightly out of sync (a "phase shift"). They adjusted their ruler to slide left or right until it matched the runner's actual position, then measured the error.
• The result: When they accounted for that tiny shift, the error was tiny. Group A's method of measuring was too rigid and penalized the simulation for a tiny timing glitch rather than a real mistake.

---

The Verdict

After fixing the "starting line," taking a single high-quality "photo," and using a flexible "ruler," Group B ran the simulation again.

The Result: The new method (ECC-IPIC) worked just as well as the old, standard method. It produced sharp, accurate results even with fewer particles.

The Conclusion:
The problem wasn't the engine (the simulation algorithm); it was the dashboard (how they measured the speed). Group A's claim that the new method is "less accurate" is false. If you measure it correctly, the new method is just as good as the old one, but potentially faster and more efficient.

In short: Group A blamed the car for being slow because they were looking at a blurry, stacked photo and measuring the distance with a rigid ruler. Group B cleaned up the photo and used a better ruler, proving the car is actually a race winner.

Technical Summary

1. Problem Statement

The authors address a recent publication by Savard et al. (Phys. Plasmas 32, 073903, 2025), which concluded that Fully Implicit Charge- and Energy-Conserving Particle-in-Cell (ECC-IPIC) schemes are less accurate than their explicit counterparts when the cell size exceeds the Debye length and the number of particles per cell (PPC) is low. Savard et al. claimed that reproducing highly resolved, convergent solutions in implicit schemes requires a significantly higher particle count than explicit schemes.

Chacón et al. argue that this conclusion is flawed and stems from methodological errors in the initialization and diagnostic procedures used by Savard et al., rather than inherent deficiencies in the ECC-IPIC algorithm itself. They aim to demonstrate that, with proper implementation, ECC-IPIC achieves accuracy comparable to explicit PIC for equivalent PPC.

2. Methodology

To refute the previous findings, the authors re-simulated the Ion Acoustic Shock Wave (IASW) benchmark problem, a closed system with strict conservation of linear momentum and total energy. They employed their own implementation of the ECC-IPIC method (DPIC) and compared it against Savard's results.

Key methodological components include:

• Simulation Setup:

  • Used the same mesh mapping, time step ($dt = 2\omega_{pe}^{-1}$), and initial conditions as Savard et al.
  • Focused on the Implicit Cloud-in-Cell (ICIC) gather-scatter strategy.
  • Compared results on both uniform and adaptive meshes.

• Particle Initialization Improvements:

  • Density Matching: The authors identified that Savard's particle distribution likely introduced uncontrolled noise. They implemented two improved initialization strategies:
    1. Approximate Weighting: Setting particle weights proportional to the Jacobian and local cell density ($w_{p,i} \propto (Jn)_i / N_{ppc}$).
    2. Exact Mass-Matrix Inversion: Solving a linear system to determine particle weights ($W_j$) such that the initial density profile matches the analytical solution to machine precision.
  • Sampling Strategies: They compared standard Random Number Generation (RNG) with Quiet Start (QS) using Hammersley quasi-random low-discrepancy sequences to reduce initial statistical noise.

• Diagnostic Improvements:

  • Rejection of Ensemble Averaging: Savard et al. averaged 10 independent runs to reduce noise. Chacón et al. argue that because ECC-IPIC is nonlinear, ensemble averaging does not commute with the solution propagator. Instead, they performed single runs with $10\times$ more particles (equivalent computational cost) to achieve lower noise without the bias of averaging.
  • Phase-Shift Correction: The authors noted that implicit schemes may suffer from slight momentum conservation errors, causing the shock front to shift slightly in phase compared to the reference. Savard's error metric ($\sigma_{ion,RMS}$) penalized this shift heavily, obscuring true convergence. The authors introduced an optimized RMS error ($\sigma_{ion,RMS}^{opt}$) that minimizes the error over a range of phase shifts ($h$) to isolate particle noise from spatial/phase errors.

3. Key Contributions

• Identification of Diagnostic Artifacts: The paper identifies that the "poor accuracy" reported by Savard et al. was largely an artifact of ensemble averaging (which smoothed the shock profile and introduced systematic bias) and the use of an unshifted error metric that conflated phase errors with amplitude errors.
• Demonstration of Equivalent Accuracy: The authors prove that when initialized with exact density matching and diagnosed with phase-shift correction, ECC-IPIC achieves convergence rates and accuracy levels comparable to explicit PIC.
• Validation of Initialization Techniques: They demonstrate that exact density matching via mass-matrix inversion significantly improves convergence scaling and reduces RMS errors compared to standard random initialization.
• Scaling Analysis: The study clarifies the convergence scaling of ECC-IPIC, showing that with Quiet Start initialization, the error scales as $N_{ppc}^{-0.65}$ (better than standard Monte Carlo $N_{ppc}^{-0.5}$), whereas standard RNG recovers the $N_{ppc}^{-0.5}$ scaling.

4. Results

The re-analysis yielded the following specific findings (visualized in the paper's Figure 2 and 3):

1. Convergence Behavior: When using the optimized error metric (accounting for phase shifts), the ECC-IPIC RMS error converges with $N_{ppc}$ similarly to explicit PIC. Savard's original unshifted metric showed poor convergence because it was dominated by phase-shift errors.
2. Impact of Averaging: Figure 3 shows that Savard's ensemble-averaged solution exhibited a systematic smoothing bias, artificially broadening the shock front and underestimating downstream density. In contrast, the single-run ECC-IPIC solution (with higher PPC) maintained a sharp shock front and oscillated correctly around the reference solution.
3. Initialization Sensitivity: Exact density matching reduced RMS errors and produced cleaner scaling laws. However, even without exact matching, the qualitative conclusions of Savard et al. (that implicit is inferior) did not hold.
4. Error Sources: The study isolated that for high PPC, the dominant error in adaptive meshes was spatial discretization, not particle noise. This highlights the necessity of using reference solutions on the same mesh topology to isolate particle errors.

5. Significance

This paper is significant for the computational plasma physics community for several reasons:

• Restoring Confidence in Implicit PIC: It counters the narrative that implicit charge- and energy-conserving schemes are inherently less accurate than explicit ones, reaffirming their viability for large-scale simulations where time-step constraints make explicit methods prohibitive.
• Best Practices for PIC: It establishes critical best practices for future studies, specifically:
  • Avoiding ensemble averaging for nonlinear implicit solvers; prefer single runs with higher particle counts.
  • Implementing exact density matching during initialization.
  • Using phase-shift-corrected error metrics when analyzing shock-like or traveling wave solutions.
• Methodological Rigor: It serves as a cautionary tale regarding the interpretation of numerical convergence in PIC simulations, emphasizing that diagnostic choices (how error is measured) can drastically alter the perceived performance of an algorithm.

In conclusion, Chacón et al. demonstrate that the "negative results" reported by Savard et al. were preventable methodological artifacts. With careful initialization and diagnostics, ECC-IPIC schemes on adaptive meshes provide accuracy comparable to explicit PIC, validating their use for complex plasma simulations.