Hybrid Delta Tracking Schemes Using a Track-Length Estimator

This paper introduces a hybrid delta-tracking algorithm utilizing a track-length estimator for structured mesh flux tallying, demonstrating its ability to support hybrid-in-energy and hybrid-in-material schemes as well as continuously moving surfaces, while showing significant performance improvements over standard methods in problems with void regions and continuous energy reactor benchmarks.

The Gist

Imagine you are trying to map out how a crowd of people (neutrons) moves through a giant, complex building (a nuclear reactor) to see where they end up and how many bump into things. This is what scientists do with Monte Carlo simulations: they send thousands of virtual "people" on random walks to predict the behavior of real nuclear systems.

The paper you provided is about a new, smarter way to guide these virtual people through the building. Here is the breakdown using simple analogies.

The Old Ways: Two Different Maps

To get a particle from Point A to Point B, scientists usually use one of two "navigation strategies":

1. Surface Tracking (The "Door-to-Door" Method):

   • How it works: Imagine a person walking through a building. Every time they take a step, they check: "Am I about to hit a wall?" If yes, they stop, turn, and check the next room. If no, they check: "Did I bump into a person?"
   • The Problem: In a building with thousands of tiny rooms and walls (complex geometry), checking for walls takes a lot of time. It's like walking through a maze and checking every single inch for a wall before taking a step. It's accurate but slow.

2. Delta Tracking (The "Ghost Walk" Method):

   • How it works: Instead of checking for walls, this method assumes the whole building is made of the "thickest" material possible (the most crowded room). The walker takes a giant step based on that thick material.
   • The Catch: Since the real building has empty rooms (voids) and thin walls, the walker might take a step that would have hit a wall in the thick material, but in reality, they are in an empty room.
   • The Fix: The computer says, "Wait, you're in an empty room. That step was fake." It rejects the step and tries again. This is called rejection sampling.
   • The Problem: If the building has huge empty halls (voids), the walker keeps taking steps, getting rejected, and trying again. It gets stuck in a loop of "fake" steps, wasting a lot of time.

The New Innovation: The "Hybrid" GPS

The authors of this paper built a new system that combines the best of both worlds. They call it Hybrid Delta Tracking, and they added a special tool called a Track-Length Estimator.

1. The "Ghost Walk" with a Pedometer (Track-Length Estimator)

In the old "Ghost Walk" (Delta Tracking), if a step was rejected (because it was a "phantom" collision), the computer usually just ignored the distance the walker traveled. It was like a runner running a race but the timer only started when they actually hit a wall.

The authors realized: Why not count the distance even if the step was fake?
They added a pedometer (Track-Length Estimator) that counts every meter the walker travels, even the "fake" steps. This gives a much clearer, more accurate picture of where the crowd is, especially in empty rooms where the old method was terrible.

2. The "Smart Switch" (Hybrid Methods)

The authors realized that sometimes you need a Door-to-Door map, and sometimes a Ghost Walk is better. So, they built a Smart Switch that changes the navigation style based on the situation:

• Hybrid-by-Material (The "Room Type" Switch):

  • If the walker is in a dense, crowded room (like a solid block of fuel), use the Ghost Walk. It's fast because there are few walls to check.
  • If the walker is in a huge empty hall (a void), switch to Door-to-Door. It's faster to check the few walls than to keep getting rejected in the empty space.
  • Analogy: It's like driving a car. On a straight, empty highway, you cruise (Ghost Walk). When you hit a complex city intersection, you slow down and check every turn (Door-to-Door).

• Hybrid-by-Energy (The "Speed" Switch):

  • Fast particles (High energy) usually fly through things without stopping. For them, the Ghost Walk is perfect because they rarely hit anything.
  • Slow particles (Low energy) tend to get stuck in "resonances" (like hitting a specific frequency that stops them). For these, the Door-to-Door method is safer and more accurate.
  • Analogy: If you are running a sprint, you don't need to check every puddle; just run. If you are walking through a minefield, you need to check every single step carefully.

Why Does This Matter?

The authors tested these new methods on supercomputers (both giant CPU clusters and powerful GPUs) using four different "building" scenarios:

1. The "Kobayashi" Problem (A building with huge empty rooms):

   • The new method was 1.5 to 2.5 times faster and more accurate than the old ways. It solved the problem of the "Ghost Walk" getting stuck in empty space.

2. The "Dragon Burst" Problem (A moving slug of fuel):

   • This was a tricky test where a piece of fuel moves through a reactor. The new method proved it could handle moving walls perfectly, which is a big deal for safety simulations.

3. The "C5CE" Problem (A real-world reactor simulation):

   • This is where the Hybrid-by-Energy method shined. By switching strategies based on how fast the particles were moving, they achieved a 7 to 11 times speedup. This is massive. It means a simulation that used to take a week could now be done in a day.

The Bottom Line

Think of this paper as inventing a super-smart GPS for nuclear particles.

• Before, you had to choose between a slow, careful map (Surface Tracking) or a fast but glitchy map (Delta Tracking).
• Now, you have a GPS that switches modes automatically. It knows when to be careful and when to speed up. It also counts every step you take, even the ones that didn't count, to give you a perfect picture of the journey.

This allows scientists to simulate nuclear reactors, fusion experiments, and safety tests much faster and with greater accuracy, which is crucial for designing safer and more efficient energy systems.

Technical Summary

1. Problem Statement

In Monte Carlo (MC) radiation transport, two primary algorithms are used to sample particle paths: Surface Tracking and Woodcock Delta Tracking.

• Surface Tracking: Computes the distance to the nearest geometric surface and the distance to a collision. It is efficient for complex geometries but computationally expensive when many surface crossings occur.
• Delta Tracking: Uses a "majorant" cross-section (the maximum cross-section across all materials) to sample a distance to a potential collision. It avoids expensive surface distance calculations but introduces "phantom" (virtual) collisions that must be rejected via sampling.
  • Limitation 1: Standard delta tracking implementations typically rely on the collision estimator for flux tallies. In optically thin regions (voids or low-density materials), the collision estimator suffers from high variance because few physical collisions occur.
  • Limitation 2: While the track-length estimator (which scores the path length of particles through a mesh cell) offers lower variance, it has historically been difficult to combine with delta tracking due to algorithmic inefficiencies in determining which mesh cells a particle traverses without surface checks.
  • Limitation 3: Neither method is universally optimal; performance depends heavily on the problem's geometry, material composition, and particle energy.

The paper addresses the need for a flexible, high-performance tracking framework that can utilize the track-length estimator with delta tracking and combine both tracking methods (hybrid approaches) to optimize performance across diverse physical regimes.

2. Methodology

The authors implemented these concepts in MC/DC (Monte Carlo / Dynamic Code), an open-source, time-dependent MC neutron transport application designed for rapid numerical development on CPUs and GPUs.

A. Voxelized Track-Length Estimator with Delta Tracking

• Concept: The authors developed a "voxelized tally" algorithm that allows delta tracking to score the track-length estimator.
• Mechanism: Instead of stopping at material interfaces, the particle is "swept" through a structured rectilinear mesh (voxels). The algorithm calculates the distance traveled within each voxel and accumulates the track length.
• Innovation: This allows the use of the low-variance track-length estimator even when particles are undergoing delta tracking (where they do not stop at every material interface). This is particularly effective in void regions where collision estimators fail.

B. Hybrid Tracking Schemes

The authors implemented two hybrid strategies to leverage the strengths of both tracking algorithms:

1. Hybrid-in-Material: The tracking algorithm is selected based on the material region.
   • Logic: Delta tracking is used in regions with high cross-sections or complex geometries. Surface tracking is used in voids or regions with strong localized absorbers where delta tracking would suffer from excessive rejection sampling (virtual collisions).
2. Hybrid-in-Energy: The tracking algorithm is selected based on particle energy.
   • Logic: Delta tracking is used for high-energy particles (where mean free paths are long and cross-sections are relatively smooth). Surface tracking is used for low-energy particles (resonance regions) where cross-sections vary drastically between materials, making the majorant cross-section inefficient and rejection sampling costly.

C. Continuously Moving Surfaces

The implementation supports continuously moving surfaces (e.g., fuel pellets moving through a core). The authors verified that the delta tracking algorithm, combined with the voxelized tally, correctly handles dynamic geometry changes without requiring a restart or static mesh redefinition.

3. Key Contributions

1. First Published Implementation: This work presents the first published use of a track-length estimator in conjunction with full delta tracking on a structured mesh.
2. Hybrid Energy Tracking: It introduces a novel hybrid-in-energy method, dynamically switching between surface and delta tracking based on particle energy thresholds (specifically above/below resonance regions).
3. Dynamic Geometry Support: It demonstrates the first successful application of delta tracking with continuously moving surfaces.
4. Scalability: The methods were implemented and tested on both CPU (Intel Xeon) and GPU (Nvidia V100) architectures using the MC/DC framework.

4. Results

The methods were evaluated using four time-dependent fixed-source benchmark problems:

1. Kobayashi Problem: A problem with significant void regions.
   • Result: The voxelized delta tracking with a track-length estimator improved the Figure of Merit (FOM) by 1.5×–2.5× compared to standard delta tracking (collision estimator) and surface tracking.
2. Modified C5G7 (Multi-group): A pressurized water reactor benchmark.
   • Result: Standard delta tracking with a collision estimator performed best. The track-length estimator did not provide a significant advantage here, likely due to the dense, multi-group nature of the problem.
3. C5CE (Continuous Energy): A continuous-energy version of the C5G7 benchmark.
   • Result: The Hybrid-in-Energy method showed massive improvements, achieving a 7×–11× increase in FOM compared to standard methods. This confirms that switching to surface tracking in resonance regions and delta tracking at high energies is highly effective.
4. Modified Dragon Burst: A problem with a moving fuel slug through air (void) and fuel.
   • Result: Standard delta tracking failed to finish within the time limit due to excessive rejection sampling in the air region. The Hybrid-in-Material method (using surface tracking in air) succeeded, completing the simulation in ~4 hours, though standard surface tracking was still faster (1 hour) due to the problem's simple geometry.

Hardware Performance:

• GPU nodes (Lassen) generally outperformed CPU nodes (Dane) by factors of 1.2× to 2.7× depending on the problem and method.
• The hybrid-in-energy method showed the most consistent gains across both hardware architectures.

5. Significance

• Breaking the Dichotomy: The work demonstrates that surface and delta tracking are not mutually exclusive choices but can be mixed and matched to optimize performance based on local physical conditions (energy, material, geometry).
• Efficiency in Voids: By enabling the track-length estimator with delta tracking, the method significantly reduces variance in optically thin regions, a known weakness of standard delta tracking.
• Dynamic Systems: The successful integration with moving surfaces expands the applicability of delta tracking to time-dependent problems involving mechanical motion (e.g., control rod insertion, fuel movement), which are critical for safety analysis and transient simulations.
• Future Directions: The authors note that while reaction rate tallies are currently limited in this specific voxelized approach, future work will focus on efficient cross-section lookups on structured meshes to enable full reaction rate capabilities. The hybrid-in-energy approach is identified as a highly promising avenue for continuous energy reactor simulations.

In conclusion, this paper provides a robust framework for enhancing Monte Carlo transport efficiency by intelligently combining tracking algorithms and estimators, offering significant performance gains for complex, time-dependent nuclear simulations.