Research Programme and Scientific Requirements

Document role

This page preserves the owner's integrated programme direction, scientific requirements, reasoning, terminology, and methodological detail. These requirements describe intended research, not completed evidence or approval of every proposed method. Current decision status is maintained in Scope and Decisions; the ordered procedure is in Methods, Verification, and Validation.

Research motivation

OWNER INTENT: The group needs a rapid dose predictor for shielding analysis of industrial electron-beam and converted-X-ray accelerators. RAMAL-EBX is intended to be a reusable GUI research application for three-dimensional radiation-protection field prediction, inspection, and independent verification.

The predictor should accelerate preliminary shielding calculations and design-space exploration. It will not replace a verified radiation-transport calculation, professional safety assessment, commissioning measurement, or regulatory approval. PHITS remains the reference Monte Carlo method used to generate and evaluate surrogate-model data. Intended contributions are not completed results.

Programme and Paper 1 scope

Broader programme

OWNER INTENT: The broader programme may eventually cover:

direct electron-beam operation;
electron beams converted to bremsstrahlung X-rays;
incident-electron energies from approximately 0.1 to 10 MeV;
multiple source-head and shielding families;
three-dimensional radiation-protection fields;
particle-resolved fluence and dose quantities; and
later product-dose, surface-dose, and other industrial applications.

This is a programme horizon, not the claim of one paper or one surrogate.

Paper 1

APPROVED DECISION: Paper 1 is restricted to conventional industrial-LINAC converted-X-ray operation at 5-7.5 MeV, one bounded converter/source family, one family of closed ordinary-concrete rooms, PHITS reference calculations, and surrogate prediction of a predefined three-dimensional radiation-protection field for complete configurations excluded from model fitting.

Direct-EB cases may support component verification but are outside the Paper 1 prediction claim. Doors, mazes, ducts, penetrations, activation, post-shutdown dose, arbitrary source heads, product dose, and facility-specific compliance are outside scope unless an explicit later amendment brings them in.

The current two-paper review is narrative and bundle-limited. It does not establish a systematic review or universal novelty. A broader reproducible evidence review may motivate an explicit amendment; it cannot silently reset an approved decision.

Research statement

OWNER INTENT: Paper 1 asks whether a geometry-, energy-, and source-conditioned surrogate can reproduce a predefined three-dimensional radiation-protection field from PHITS reference calculations for 5-7.5 MeV industrial converted-X-ray LINAC shielding configurations within a declared range of applicability.

Evaluation must use complete shielding configurations excluded from model fitting and expose local and subgroup failures, especially underprediction, rather than relying on one global average.

Intended contributions

Paper 1 should aim to produce:

a reproducible PHITS source-to-shield model with documented physical and numerical assumptions;
component, limiting-case, and applicable measurement evidence that separates verification from physical validation;
a traceable PHITS simulation Dataset containing complete configurations, particle-resolved fields, units, normalization, and statistical uncertainty;
a surrogate evaluated against simpler reference methods and an untouched final-test partition;
independent PHITS comparisons for selected Predictions; and
a declared range of applicability and limitations.

RAMAL-EBX should expose the complete researcher workflow through its GUI. The application enables the method; it does not select scientific criteria or establish their validity.

Current application capability and implementation-documentation mismatches affecting Project work are maintained only in Current Project Evidence. Capability does not establish scientific acceptance or completed Paper 1 evidence.

Intended users

Accelerator-shielding researchers and designers.
Radiation-protection and safety researchers.

Physical model

Source definition

Every Case and Prediction must identify its source mode. Paper 1 reference calculations should begin with incident electrons and represent the finite converter and every transport-relevant layer.

A reusable converter phase-space source is acceptable only after its particle types, energy, position, direction, normalization, and converter conditions are independently verified against the coupled electron-to-bremsstrahlung calculation.

The source definition must record:

incident-electron energy distribution;
beam footprint and incidence angle;
source position and orientation;
converter material and dimensions;
cooling, backing, and residual-electron-control layers;
scan or spatial distribution; and
absolute and per-source normalization.

A converted-X-ray source does not emit monoenergetic photons at the nominal electron energy. Its spectrum and angular distribution are part of the model.

Energy interval

APPROVED DECISION: 5 MeV <= E_e <= 7.5 MeV.

If interpolation across this interval is claimed, the simulation design must include justified intermediate energies, with at least one complete intermediate-energy condition excluded from fitting. Endpoint-only calculations support only endpoint comparisons. The broader 0.1-10 MeV programme range contains materially different electron-transport, converter, shielding, neutron, and activation regimes and is not an accepted Paper 1 domain.

Source-head archetype and scan

APPROVED DECISION — amendment-open: Use a broad planar tantalum-water-Type-304-stainless scanned-plate archetype in the Paper 1 frame: +Y upward, electrons toward -Y, scan along X, and the remaining transverse coordinate along Z.

APPROVED DECISION — amendment-open: Use a declared normalized one-dimensional scan with a local Gaussian spot.

UNRESOLVED: Plate extent, layer dimensions, local spot definition, scan weighting, quadrature, Gaussian support, finite-edge interception, transport-response measures, tolerances, statistical precision, and escalation or stopping rules remain open.

Current scan-authoring capability is maintained only in Current Project Evidence. These implementation capabilities do not approve D5 details.

Geometry

APPROVED DECISION: Use a closed rectangular ordinary-concrete room with width 4-10 m, length 6-16 m, height 3-6 m, primary-floor concrete 100-180 cm, shared secondary barriers 50-150 cm, a centred source, +Y upward, and the beam vertically downward. Openings and streaming paths are excluded. These ranges define the study family, not safe construction dimensions.

The model should include the converter and major nearby scattering structures, room air, complete walls, floor and roof, primary and secondary barrier regions, exterior evaluation regions, and explicit compositions and densities. Small details may be omitted only after sensitivity calculations demonstrate that omission does not materially alter the declared output.

Radiation-protection quantities

APPROVED DECISION: The primary field is ambient dose equivalent H*(10), PHITS multiplier -200, normalized as pSv/primary-source-electron. Combined, photon, electron, positron, and neutron pages remain identifiable. Operational dose rate is derived separately from declared beam current, scan utilization, duty factor, workload, and schedule. Beam current must not become a learned input while transport remains linear in source strength; keep it as a separate operational scaling factor.

The predictor must reproduce the exact quantity, unit, normalization, coordinate system, grid, conversion coefficients, and particle definitions of its PHITS Results. H*(10) and ICRU Report 95 ambient dose H* are not interchangeable. Fluence may be retained as conversion-independent transport information; any H* calculation remains separately named.

An unscored voxel is missing information, not zero. Statistical uncertainty, reportable coverage, unscored regions, and particle contributions must remain inspectable.

Photonuclear transport

APPROVED DECISION: Use JENDL-5 conditionally as the photonuclear reference, compare it with ENDF/B-VIII.1 and PHITS built-in modelling, and retain neutron reporting in the combined D2 target unless later approved evidence supports another rule.

Converted-X-ray configurations from 5 to 7.5 MeV must assess neutron generation and transport. The Project must record the photonuclear library, reaction model, cutoffs, and transport settings. Sensitivity calculations are required near threshold. A zero neutron page in one calculation or nominal electron energy alone cannot justify exclusion. Activation and post-shutdown dose remain separate problems.

Methodological sequence

PROPOSED METHOD: Proceed incrementally:

complete a reproducible evidence review and confirm the evidence gap;
approve physical scope, quantity, source, geometry, and parameter domain;
verify component transport and limiting cases;
perform preliminary PHITS calculations to establish numerical behaviour;
validate the physical model against applicable measurements where available;
predefine Study inclusion criteria and Dataset partitions;
generate immutable PHITS reference calculations;
construct a Dataset only from Results meeting every applicable criterion;
compare simple reference methods and surrogate candidates;
finalize the selected Model before final-test access;
evaluate the untouched final test once; and
independently verify selected Predictions with new PHITS calculations.

Failed and inconclusive Paper 1 calculations remain traceable. Disposable application-verification records belong in a separate Project.

Verification and validation

Verification asks whether implementation and numerical behaviour match their specification. Validation asks whether the physical model represents reality adequately for its declared purpose and domain.

PROPOSED METHOD: Verification should cover:

Case and Result integrity and reconstruction;
electron stopping power and range against applicable NIST ESTAR data;
photon interactions and attenuation against applicable NIST XCOM data;
converter yield, residual-electron, spectral, and angular comparisons;
simple concrete attenuation;
grid and source-discretization convergence;
repeated independent seeds;
photonuclear library and model sensitivity; and
selected inter-code comparisons where practical.

Physical validation requires applicable measurements and their uncertainties. Agreement with another Monte Carlo code is verification, not physical validation. If applicable measurements cannot be found, report that limitation.

Preliminary PHITS calculations

Use “preliminary PHITS calculations” or “preliminary simulation study.” They should establish spatial and source-discretization convergence, histories and batches by region and field level, unscored-voxel behaviour, RSE and reportable coverage, repeated-run stability, useful variance reduction, photonuclear sensitivity, neutron contribution, runtime, and storage requirements. Criteria must be approved before final Dataset generation, not selected retrospectively to retain convenient Results.

Study inclusion and Dataset construction

PROPOSED METHOD: Admit a Result only if its complete provenance and artifacts pass integrity checks; it lies inside the approved domain; required transport completed; version, libraries, controls, seed, and variance reduction are declared; quantity, units, normalization, grid, particles, nominal field, and uncertainty field match; approved reportability, convergence, repeated-run, verification, and declared validation requirements are met; and an Assessment records the outcome and exclusion reasons.

Complete physical configurations are indivisible partition groups. All voxels, profiles, replicas, and defined paired variants remain together. Use training, model-selection—currently serialized as validation—and untouched final-test partitions. Predefine grouping keys, split proportions, seed, stratification, paired conditions, interpolation tests, and replica handling.

The Dataset should retain Case and Result traceability, features and units, target and uncertainty fields, reportable masks, coordinates and material representation, particle pages, partition membership, and governing criteria.

Surrogate strategy and evaluation

PROPOSED METHOD: Increase complexity only as evidence requires: nearest configuration; interpolation or fitted attenuation; regularized regression; a compact nonlinear method such as a Gaussian process; and a spatial neural model only if simpler approaches fail and the Dataset supports it.

Architecture, transformations, loss, hyperparameters, and stopping rules use only training and model-selection data. Any post-final-test model change creates a new Model and requires a new untouched final test.

Report performance overall and by energy, geometry boundary, barrier role, region, field level, particle, interpolation condition, domain boundary, worst case, and underprediction. Measures should include signed error, bias, absolute error, reportable relative error, logarithmic field error, underprediction and overprediction frequency and magnitude, high-quantile and maximum error, spatial maps and profiles, comparison with PHITS uncertainty, interval coverage where applicable, computation time, and end-to-end speedup. A favourable average cannot compensate for failure in a declared subdomain. When the Project declares a design dose criterion, report incorrect pass/fail classifications as well as continuous errors. Do not claim regulatory compliance unless the jurisdiction, radiation-protection quantity, occupancy, workload, averaging period, and legal basis are established.

“Rapid” must be measured. Report PHITS wall-clock and CPU time, hardware and process count, histories, batches, variance reduction and uncertainty, Dataset preparation, training and selection, final-test evaluation, single and batch inference, and comparable end-to-end speedup. Do not hide reference-data generation time.

Prediction verification

PROPOSED METHOD: Independently recalculate selected Predictions, including largest underpredictions and overpredictions, high-uncertainty predictions, representative low, middle and high energies, domain boundaries, and conditions near any approved design criterion. Statistical independence requires an independent stochastic calculation. A Prediction remains a surrogate record and cannot be promoted to a PHITS Result.

Terminology

Preliminary PHITS calculation: early calculation used to establish numerical behaviour and finalize a method.
Immutable Case: non-overwriting physical and numerical input record.
PHITS reference calculation: Monte Carlo calculation used to generate or evaluate surrogate data; not automatically physical truth.
Surrogate model: model approximating the declared PHITS mapping.
Verification: evidence that implementation and numerical behaviour match specification.
Validation: evidence that the physical model represents reality adequately.
Reportable voxel: scored voxel satisfying approved criteria.
Unscored voxel: missing Monte Carlo information, not zero.
Model-selection partition: data used for model and hyperparameter selection, distinct from physical validation.
Finalized Model: non-overwriting Model fixed before final-test evaluation.
Range of applicability: physical and numerical domain supported by evidence.

Capitalized Design, Case, Job, Attempt, Result, Study, Assessment, Dataset, Model, Prediction, and Prediction Verification refer to RAMAL-EBX record types. Prefer “procedure,” “criteria,” “rule,” and “settings” to unexplained scientific-facing uses of “policy,” “contract,” “gate,” “freeze,” “ground truth,” or “production Dataset.”