Models that remain useful when the microscopic picture is incomplete.

01 · Reaction Pathway Intelligence

Chemical manufacturing relies on chemical reactions to convert raw materials into desired products. Many conventional computational approaches start from reaction pathways that scientists already have in mind and then test whether those pathways are likely to occur. Our approach starts one step earlier: it explores how molecular systems evolve under realistic conditions, revealing competing or unexpected reaction pathways and unwanted byproducts without requiring them to be specified in advance. This can give research and development teams a broader view of what may actually happen, help identify potential problems earlier, and focus experiments on the most relevant conditions and reaction mechanisms.

Problem. For well-defined elementary reactions, transition-state calculations, rate theory and kinetic modelling already provide mature and effective approaches. In complex reactive systems, the difficulty can arise earlier. Important intermediates, short-lived complexes, products and competing transitions may not yet be fully determined; finite temperature can also change the stability and interconversion of different structures, bringing states that are not obvious in static analysis into the reaction. When reaction topology itself remains incomplete, the first task is to broaden exploration of microscopic state space and network connectivity.

Approach. We use finite-temperature atomistic dynamics to extend this search. Compared with studying only preselected structures and elementary processes, trajectory-based modelling can further capture metastable states, short-lived intermediates and structural transitions that are actually visited and incorporate them into the reaction network. Depending on the problem, these dynamic data can be used to analyse state populations, reaction pathways, intermediate lifetimes and rates, and can also be connected to experimental observables such as product distributions, kinetic energies, thermalization and spectroscopy. This approach is particularly useful when reaction topology is not yet fully established, or when structural dynamics itself changes reaction behaviour.

Representative systems. Prior work spans finite-temperature formation and dissociation networks of small molecules, formation and collision chemistry of laser-cooling-relevant molecular species, formation and lifetimes of transient molecular complexes, and molecular activation on noble-metal clusters. The chemistry differs, but the systems face a similar problem: the microscopic states and transitions relevant to the final outcome cannot always be fully specified before computation begins. Across different cases, finite-temperature dynamics can reveal new intermediates, condition-dependent dominant pathways, short-lived molecular complexes, or reaction contributions that are difficult to infer from static structure comparison.

Applications. Reaction Pathway Intelligence addresses reaction-mechanism discovery, dynamic catalysis, formation and degradation chemistry, molecular-source chemistry, reactive materials, and complex reactive systems with metastable states, structural rearrangements or competing channels. It is used not only to calculate established reaction steps, but also to expand exploration of possible reaction space and determine which microscopic processes are genuinely worth further study under specific conditions.

02 · Composition-First Predictive Modelling

Problem. In unexplored molecular and multicomponent chemical space, a complete microscopic structure may be unavailable, expensive to obtain or scientifically inappropriate as the first model input. Yet composition and component identity are often known much earlier.

Approach. We ask how far prediction can proceed from composition, atomic information and progressively richer molecular descriptors before a detailed microscopic description becomes necessary. Different observables may require different levels of information, so the boundary of a simplified representation is part of the result rather than hidden.

Representative work. Prior work spans both molecular and materials systems. Diatomic studies show how far spectroscopic properties can be predicted from composition and progressively richer descriptors. Work on disordered Group-III oxide alloys uses composition-derived approximate structures to screen stability before precise atomic structures are available and tests how parent-material data can support related doped systems. Ongoing work extends the same logic to atmospheric cluster formation, where the final model input need not be a unique optimized full-cluster geometry.

Applications. Molecular-property screening, alloy and functional-material screening, atmospheric clusters and other composition-defined systems; the framework is designed to extend toward complex mixtures, electrolytes and multicomponent formulations where structure may not be uniquely known in advance.

03 · Predictive Theory for Precision Experiments

Problem. Precision experiments often require detailed knowledge of the microscopic states and transitions involved before those quantities have been fully measured. Experimental searches therefore depend on reliable theoretical predictions of where relevant states lie, which transitions are accessible, how strong they are and how the system is expected to respond.

Approach. We use high-accuracy electronic-structure and quantum-mechanical calculations to predict experimentally accessible quantities — including states, energies, transitions, lifetimes, couplings and observable signatures. These predictions indicate where relevant signals should appear, how strong they should be and which measurements can most directly test the underlying physical picture. Experimental results then constrain and refine the calculations, creating an iterative prediction → measurement → refinement cycle.

Representative work. Prior work spans high-accuracy electronic-structure calculations, quantitative benchmarking of theoretical predictions, and theory–experiment studies in which microscopic models are tested and refined against spectroscopic measurements.

Applications. Cold and ultracold molecular science, laser cooling and trapping, precision spectroscopy, fundamental-physics measurements, and experiment-guided studies of increasingly complex microscopic systems.

04 · Scientific Software, Data & Research Interfaces

Problem. A scientific result becomes difficult to trust, reproduce or reuse when the final number or figure is separated from the data, assumptions, transformations and parameter choices that produced it. A PDF methods paragraph is not a substitute for machine-readable lineage.

Approach. We build curated scientific databases, reproducible computational workflows, research software and interfaces with provenance designed into the result. The target is an inspectable result record that can trace a derived output back through source data, calculation and processing steps, step parameters, software and model versions, validation checks and uncertainty metadata.

Representative work and current development. Prior work includes experiment-facing spectroscopy resources that combine literature data with live calculations, alongside reusable pipelines for quantum chemistry, atomistic simulation, molecular representations, machine learning and uncertainty analysis. Current software development extends that principle toward traceable-by-construction result records and provenance graphs rather than provenance added only after a workflow has finished.

Applications. Scientific databases, model and data exploration, audit-ready computational R&D, traceable simulation and machine-learning workflows, computational research infrastructure and interactive research interfaces.