Multiscale data-driven modeling

To address the challenges in unraveling data from complex reactions, our group creates and applies informatics-based, multiscale approaches for creating complex reaction models.

Our approach, MultiScale Informatics (MSI), uses multiscale, multiphysics models to bridge molecular interactions (calculable ab initio) to macroscopic observables (measurable experimentally and relevant to engineering design). MSI then uses inverse uncertain quantification to assimilate data from varied sources and scales – including ab initio quantum chemical calculations of molecular properties and measurements of macroscopic observables from multi-reaction systems.

Top: In MSI, theoretical kinetics models relate active molecular properties to rate constants.  Kinetic models, consisting of these rate constants, are then combined with physical models and active physical model parameters for each experiment to predict experimental observables.  Data available at each scale can then be used to impose constraints on the active parameters.
Bottom: MSI unravels complex data and accurately predicts behavior across multiple scales.

MSI’s unique capabilities allow us to extract meaning from complex data that pose challenges for traditional methods.  For example, MSI helped to resolve a two-decade-long debate on the temperature dependence of a reaction important to combustion and atmospheric chemistry – a finding confirmed by later independent experiments.  Furthermore, MSI models can accurately predict behavior at both molecular and macroscopic scales – MSI model predictions were found to be consistent with later higher-level ab initio calculations of molecular properties and later measurements of macroscopic observables.

We are now using MSI to create full multiscale kinetic mechanisms for combustion and pollutant emissions of carbon-free fuels, to test hypotheses, to define and push the limits of data-driven approaches for chemical kinetics, to serve as a platform for automated scientific discovery, and venture into new disciplines.

Columbia Affiliations
Mechanical Engineering
Chemical Engineering
Data Science Institute