space-rha.comShort courseChapter 2 · Shielding and transport

Chapter 2 · Lesson 4 of 6 · about 10 minutes

Choosing a transport method

A simple dose-depth calculation is often enough for an early shielding trade. A path model adds placement and coverage. Detailed transport follows the interactions and the secondaries. More detail is worth the effort when it could change the design decision and when its assumptions can be supported.

A method for the engineering question

Early trade

A SHIELDOSE-2 or 2Q calculation, used within its supported range, can compare simple dose-depth cases. Keep its limits on geometry, source energy and scoring material in view.

Greater detail

Sector analysis resolves the material around a target direction by direction. Layered or three-dimensional transport becomes worth the effort when material interactions, openings or the local spectrum control the answer.

Practical choice

For each tool, know its inputs, its tradeoffs and the conditions under which you would move to a more detailed method. An established, qualified workflow is often the right choice when the question falls inside its supported scope.

Response libraries and particle histories

The methods differ in what they compute. Sectoring combines the geometry with a precomputed response library. Forward Monte Carlo samples individual particle histories. Deterministic solvers approximate the transport equation directly. Adjoint methods put the computing effort into the responses you have selected. The implementations also differ in which physics they cover, which sources they accept and which outputs they can give you.

Depth and sector tools. SHIELDOSE-2/2Q responses combined with SSAT or FASTRAD geometry can support fast trades, as long as the conventions match. A slant path, a normal depth and a sphere or slab response are different conventions, and they cannot be swapped freely.

Layered and three-dimensional tools. MULASSIS isolates the layered transport problem. A qualified GRAS/Geant4 or FASTRAD forward setup can represent the three-dimensional interactions that matter. In every case, check the module, the release, the particles covered, the scores available and the data actually used.

Efficiency trade. Forward sampling can spend most of its histories on particles that do not reach a small target. Adjoint or biasing methods can improve the statistics of the score you care about, but their weights, their source restrictions and the processes they support have to be verified.

Benchmark. Before running the full model, reproduce a simpler case whose answer is known. Jun et al. give historical electron/Al/Ta examples that serve this purpose. The differences they report are specific to those cases and do not rank today's software in general.

3D scene

Sector, forward and adjoint

The three method families drawn on the same enclosure: ray sectors, forward particle histories and adjoint sampling from the target outward.

  1. Rays from the target through the enclosure, each converted to an equivalent thickness and combined with a precomputed response library weighted by the solid angle of its sector. Fast enough for an early trade, as long as the conventions match: a slant path, a normal depth and a sphere or slab response are different conventions and cannot be swapped freely.

  2. Particles are sampled from the source outside and followed through the actual materials, interaction by interaction, with the secondaries they create. Most histories never reach a small target, which is where the computing time goes, so the score you will base the decision on is the one that has to converge, not the total number of histories.

  3. Adjoint methods start from the target and work outward, putting the computing effort into the responses you have selected. The weights, the source restrictions and the processes they support have to be verified, and before running the full model it is worth reproducing a simpler case whose answer is known.

The original space-rha.com environment and mission renderer (NASA imagery and spacecraft models, credits in the view), with the course scenes drawn by the same renderer. Trails, clocks and counts are illustrative; the scene computes no dose or rate.

Exercise

Explain, for each transport route available to you, which physical approximations it makes and what the computational tradeoffs are.

Match the geometry, the interactions, the source and the score to the question first. Only then compare software.

Going deeper

Transport operators and adjoint response

Transport keeps the particle balance while redistributing energy, direction and species under the chosen approximations. A response is a functional of the field that results. Forward, deterministic and adjoint methods agree only when they represent the same physical operator, the same boundary conditions and the same definition of the response.

Transport equation. Schematically, the linear problem is Lψ = q. The field ψ depends on position, direction, energy and species. The operator L contains streaming, removal, redistribution, production and any continuous-energy-loss approximation. Boundary sources, and time dependence where it is required, are part of the problem definition.

Physical approximation. A straight-ahead approximation drops the angular redistribution. Continuous slowing down replaces many small discrete energy losses with their mean. Whether either is adequate depends on the particle, the energy, the geometry and the score requested. Deterministic and Monte Carlo are two ways of solving the equation; neither is more accurate than the other by definition.

Adjoint identity. For a linear response R = ⟨h,ψ⟩, a compatible adjoint equation and boundary conditions give R = ⟨ψ†,q⟩, provided the source and surface terms are included consistently. The adjoint field describes how much each part of phase space matters to the response. It is a weighting, not a physically reversed shower.

Implementation. For the release you are using, check which reverse interactions are supported, what restrictions apply to the source angles and how the weights are normalized. Benchmark the same response against a forward calculation and look at the distribution of weights. The adjoint identity holds in general; whether a particular program implements every process you need is a separate question.

Takeaway

Choose the simplest method that is adequate, and write down what would make a deeper analysis necessary.

Check yourself

When is an adjoint calculation the better choice over a forward one?

When the quantity of interest is one response at one location and the sources are many or vary: the adjoint solves once from the detector outward and weights every source direction and energy by its importance. A forward calculation is the better choice when many responses are wanted from one source.

Two codes give different doses at the same point. What do you check first?

That they solved the same problem: the same source spectrum and normalization, the same geometry and materials, the same scoring volume and material, and the same physics switched on (secondaries, straggling, nuclear reactions). Differences in these usually explain more than differences in the solvers.

A sector tool reports slant path lengths through the box, and the dose-depth library it is paired with was built for normal depth into a slab. What is wrong with feeding one into the other?

A slant path, a normal depth and a sphere or slab response are different conventions, and they cannot be swapped freely; the response library was built for a particular convention, and the geometry has to be expressed in the same one (lesson 2.2). SHIELDOSE-2 responses with SSAT or FASTRAD geometry can support fast trades as long as the conventions match. A mismatch is an error in the problem definition, and it is not corrected by the solver.

When is a SHIELDOSE-2 dose-depth calculation enough for a shielding trade, and what signals that a layered or three-dimensional method is needed?

A SHIELDOSE-2 or 2Q calculation, used within its supported range of geometry, source energy and scoring material, can compare simple dose-depth cases for an early trade. Layered or three-dimensional transport becomes worth the effort when material interactions, openings or the local spectrum control the answer, when the extra detail could change the design decision, and when its assumptions can be supported. The course asks you to choose the simplest adequate method and to write down what would make a deeper analysis necessary, so that the move is made for a reason rather than by habit.

A colleague argues that a Monte Carlo result is more accurate than a deterministic one because it follows real particle histories. What is wrong with that?

Deterministic and Monte Carlo codes are two ways of solving the same transport problem, Lψ = q, and neither is more accurate than the other by definition. They agree only when they represent the same physical operator, the same boundary conditions and the same definition of the response, and the accuracy of either depends on the physics it carries (a straight-ahead approximation drops the angular redistribution, continuous slowing down replaces discrete losses with their mean), on the data it uses and on whether those approximations fit the particle, the energy, the geometry and the score. Reproducing a simpler case with a known answer before the full run is the test that counts, whichever solver is used.

Further reading