Speaker
Description
Monte Carlo neutron transport and instrument simulations are widely used to study moderators, reflectors, beamlines, and other neutron-system components in pulsed neutron sources. Yet coupling such models to formal optimization and uncertainty quantification workflows often remains dependent on problem-specific scripts. This work presents McDakDriver, a Python framework that couples Dakota (Design Analysis Kit for Optimization and Terascale Applications) with Monte Carlo simulation tools for neutron applications. McDakDriver employs a driver-based architecture, currently implemented for MCNP and McStas, in which the workflow is organized through centralized configuration files rather than hard-coded study logic. Because McDakDriver's main configuration file is itself a Python module, it can host static settings, such as simulation code selection or template-to-input mappings and placeholder conventions, alongside executable logic such as user-defined processor and aggregator functions for response construction. Because the configuration is a Python module rather than a static input format, users can import external libraries, define derived quantities, and implement arbitrarily complex response logic directly within the study definition.
In this two-stage scheme a processor function extracts a scalar quantity of interest from raw simulation output (e.g., from tally arrays parsed via our developed mctal_tools library built on top of the mcnptools official MCNP library), and an aggregator function combines one or more such quantities into a single response value returned to Dakota. This design allows the objective formulation to be modified entirely within the configuration file, without changes to the driver code. The main driver handles Dakota parameter ingestion, substitution into parameterized templates, dependent-parameter generation, simulation execution, post-processing of tallies or monitors, and return of responses. For MCNP studies the framework additionally supports pstudy-enabled input-file templates for preprocessing parameterized inputs as well as evaluation logging that preserves parameter and response histories across runs. We also implemented an automated failure recovery in which the driver writes a Dakota-recognized failure marker when an evaluation does not complete, allowing Dakota to apply its failure-capture policy and continue the study. The post-processing layer supports mctal, meshtal, and kcode file types, so the framework can serve both fixed-source and criticality-mode studies.
As a first MCNP demonstration, McDakDriver is applied to a model representative of the Spallation Neutron Source (SNS) coupled hydrogen moderator assembly. The model incorporates the proton target region, the surrounding moderator and reflector system, inner reflector plug features relevant to the viewed beamline, and a parameterized description of the pre-moderator geometry. Beginning from a baseline water pre-moderator configuration, the framework replaces manual parameter sweeps with Dakota-driven design optimziation. The study considers BeH₂ as a candidate solid pre-moderator material, selected for its high hydrogen number density and distinct elastic and inelastic scattering characteristics. Design variables include pre-moderator thickness and temperature, the latter offering a degree of freedom not available with the baseline water pre-moderator. Thermal scattering data at each temperature are generated using the developed tslforge library, which employs NCrystal and NJOY to produce the ACE files used in the MCNP calculations. The objective is to maximize cold-neutron brightness integrated over a representative energy band, constructed directly from MCNP tally data at a beamline viewing position. Because all metric definitions, reference baselines, and objective formulations reside in the configuration file, alternative figures of merit can be explored without modifying the driver.