Possible reactant/input-order dependence in Geant4-DNA IRT-sync — clarification requested

Mehmet_Karabuga · September 21, 2026, 7:02am

Geant4 version: 11.4.2
Chemistry model: IRT_syn
Platform: macOS 12.7.6, Intel x86_64

Hello Geant4-DNA developers,

We are investigating reactant-order invariance in the Geant4-DNA IRT-sync chemistry model.

As a baseline control, the unmodified official UHDR example configured with IRT_syn successfully reaches the genuine scheduler path, including:

G4ITModelProcessor::CalculateMinTimeStep
G4DNAIndependentReactionTimeStepper::CalculateStep

We then constructed a minimal scheduler-managed bimolecular test using supported Geant4-DNA molecular-track handoff paths.

The validation considers two independent ordering factors:

Reactant input order:
A then B
versus
B then A
Reaction-registration order:
A+B
versus
B+A

The four combinations were therefore:

AB input / AB registration
BA input / AB registration
AB input / BA registration
BA input / BA registration

Before the confirmatory run, we validated that true reactant-order reversal was actually occurring while preserving the same scheduler lifecycle, geometry, molecular identities, diffusion parameters, and reaction configuration.

A corrected 2×2 factorial validation was then run using predefined criteria and new independent random-seed blocks.

Multiple ordering-dependent effects were observed in the confirmatory analysis.

However, subsequent diagnostics have not been able to determine whether these observations arise from:

changes in RNG consumption caused by reactant ordering,
reactant-input implementation asymmetry,
reaction-registration asymmetry,
or a genuine ensemble/distribution-level ordering dependence inside IRT-sync.

Therefore, at this stage we are NOT asserting that this represents an implementation defect in IRT-sync.

The observation is also separate from an ordinary-IRT ordering issue that we investigated independently; we are not assuming that the same mechanism applies to IRT-sync.

Questions for Geant4-DNA / IRT-sync developers:

Is reactant input order expected to alter any IRT-sync ensemble-level or distributional output?
Is reaction-registration order (A+B versus B+A) expected to alter any IRT-sync distributional output?
Can reversing reactant or reaction-registration order change the random-number consumption sequence while preserving the underlying ensemble distributions?
Are there known reactant-identity or ordering assumptions inside
G4DNAIndependentReactionTimeStepper
or its associated diffusion-controlled reaction model?
Is there an official or recommended regression test for verifying reactant-order and reaction-registration-order invariance in IRT-sync?

We would appreciate clarification on the intended ordering behaviour of IRT-sync and on the recommended diagnostic/regression-testing procedure.

We have a minimal NanoRT-independent reproducer, build/run instructions, preserved logs, and checksums available if they would be useful.

Thank you.

ngoc · September 26, 2026, 8:06pm

Dear,

Thanks for your report! Can you try the chem6 example (examples/extended/medical/dna/chem6)? In the macro file beam.in. You can choose the IRT-syn and one reaction, for example: e_aq + H → H2 + OHm

Could you also change the order of the reactants and check whether this affects the results?

Mehmet_Karabuga · September 27, 2026, 3:52am

Dear,

Thank you very much for the guidance.

Yes, we will repeat the test using the official chem6 example with IRT-syn and the reaction:

e_aq + H → H2 + OHm

We will then reverse only the reactant order:

H + e_aq → H2 + OHm

while keeping all other settings unchanged, and compare the results.

We will report the outcome here.

Thank you again for your help.

Best regards,
Mehmet Karabuga

Mehmet_Karabuga · September 27, 2026, 4:49am

Dear,

Thank you for the suggestion.

We repeated the test using the official Geant4 11.4.2 chem6 example with IRT-syn and the reaction:

e_aq + H → H2 + OHm

We then reversed only the stored reactant order:

H + e_aq → H2 + OHm

while keeping all other settings identical.

We first performed a pilot test and then a reproducibility study with 20 independent matched seed pairs. Within each pair, A and B used exactly the same random seeds.

For all 20 matched pairs:

Species.txt was identical.
Retained scorer-event counts were identical.
Target-reaction counts were identical.
Total reaction counts were identical.
All reported G-values were identical.
ROOT ntuple payloads were identical.
The only ROOT file differences were two metadata bytes associated with timestamps.

Therefore, under this official chem6 + IRT-syn configuration, reversing the reactant order produced no detectable physical or scoring difference.

This suggests that the order-dependent effects we previously observed in our minimal custom test are likely related to some other aspect of that test configuration rather than the official chem6 IRT-syn reaction itself.

Would you recommend any additional official IRT-syn test, or a specific code path we should inspect next to understand why our custom minimal test showed ordering-dependent behaviour?

Thank you again for your guidance.

Best regards,
Mehmet Karabuga

ngoc · September 28, 2026, 8:08am

Thanks for your test! The chemical reactions should be added in the reaction table : G4DNAMolecularReactionTable

Please check if your reaction list is fully added in the G4DNAMolecularReactionTable ?

Best regards,

Mehmet_Karabuga · September 29, 2026, 5:30am

Dear,

Thank you. I checked G4DNAMolecularReactionTable explicitly in a clean Geant4 11.4.2 build.

For this minimal reproducer, the complete reaction list intentionally consists of a single synthetic channel:

e_aq + H3O+ -> H

The table is reset and this channel is added through G4DNAMolecularReactionTable. I then queried the table at runtime in both reactant orders.

GetReactionData(A,B) and GetReactionData(B,A) both resolve to the same G4DNAMolecularReactionData object.

I also repeated the test as a 2×2 factorial check:

reaction registration: A+B or B+A
reactant input: A+B or B+A

Five independent seeds were tested, giving 20 runs in total.

In all 20 runs, the A+B/B+A reaction-table lookup was symmetric. For each fixed seed, all four registration/input-order combinations also produced the same H product, the same final chemical outcome, and the same product position.

Therefore, in this clean minimal reproducer, the intended reaction is present in G4DNAMolecularReactionTable, and I am currently unable to reproduce the ordering-dependent behaviour that I observed in the earlier custom diagnostic.

I also identified a provenance inconsistency in my earlier diagnostic package. Therefore, I do not want to attribute that earlier observation to Geant4 until its origin is understood.

I have preserved the clean reproducer, the 20-run results, build information, and SHA-256 checksums and can upload the reproducibility package if it would be useful.

Would you recommend checking another part of the custom chemistry setup next, or should I now compare this reproducer directly against the reaction-table construction used by the official chem6 IRT-syn configuration?

Thank you again for your guidance.

Best regards,
Mehmet Karabuga

Mehmet_Karabuga · October 14, 2026, 9:34pm

Dear,

Thank you. I checked G4DNAMolecularReactionTableexplicitly in a clean Geant4 11.4.2 build.

For this minimal reproducer, the complete reaction list intentionally consists of a single synthetic channel:

e_aq + H3O+ -> H

The table is reset and this channel is added through G4DNAMolecularReactionTable. I then queried the table at runtime in both reactant orders.

GetReactionData(A,B) and GetReactionData(B,A)both resolve to the same G4DNAMolecularReactionDataobject.

I also repeated the test as a 2×2 factorial check:

reaction registration: A+B or B+A
reactant input: A+B or B+A

Five independent seeds were tested, giving 20 runs in total.

In all 20 runs, the A+B/B+A reaction-table lookup was symmetric. For each fixed seed, all four registration/input-order combinations also produced the same H product, the same final chemical outcome, and the same product position.

Therefore, in this clean minimal reproducer, the intended reaction is present in G4DNAMolecularReactionTable, and I am currently unable to reproduce the ordering-dependent behaviour that I observed in the earlier custom diagnostic.

I also identified a provenance inconsistency in my earlier diagnostic package. Therefore, I do not want to attribute that earlier observation to Geant4 until its origin is understood.

I have preserved the clean reproducer, the 20-run results, build information, and SHA-256 checksums and can upload the reproducibility package if it would be useful.

Would you recommend checking another part of the custom chemistry setup next, or should I now compare this reproducer directly against the reaction-table construction used by the official chem6 IRT-syn configuration?

Thank you again for your guidance.

Best regards,
Mehmet Karabuga

Mehmet_Karabuga · October 19, 2026, 7:56am

Dear,

Just a brief follow-up regarding my previous post.

I would be grateful for any guidance on the next diagnostic step, particularly whether you recommend comparing the reaction-table construction of our minimal reproducer with the official chem6 IRT-syn configuration, or inspecting another part of the IRT-syn chemistry path.

Thank you again for your time and help.

Best regards,
Mehmet Karabuga