Related Concept Videos
Propagation of Uncertainty from Random Error
Propagation of Uncertainty from Systematic Error
Uncertainty: Overview
Mechanistic Models: Compartment Models in Algorithms for Numerical Problem Solving
In individual population analyses, different algorithms are employed, such as Cauchy's method, which uses a...
Uncertainty: Confidence Intervals
Yield Criteria for Ductile Materials under Plane Stress
The Maximum Shearing Stress Criterion, also known as...
You might also read
Related Articles
Articles linked to this work by shared authors, journal, and citation graph.
Extending the accuracy of the SNAP interatomic potential form.
Related Experiment Video
Updated: Dec 11, 2025

Knowledge Based Cloud FE Simulation of Sheet Metal Forming Processes
Published on: December 13, 2016
Multi-fidelity machine-learning with uncertainty quantification and Bayesian optimization for materials design:
Anh Tran1, Julien Tranchida2, Tim Wildey1
1Optimization and Uncertainty Quantification, Center for Computing Research, Sandia National Laboratories, Albuquerque, New Mexico 87123, USA.
This study introduces a machine learning framework that combines high-accuracy (density functional theory) and fast (machine learning interatomic potential) atomistic simulations. This approach efficiently predicts materials properties and quantifies prediction uncertainty for faster materials design.
Area of Science:
- Computational Materials Science
- Machine Learning in Materials Discovery
- Atomistic Simulations
Background:
- Accurate prediction of materials properties often requires computationally expensive high-fidelity methods like density functional theory (DFT).
- Lower-fidelity methods, such as machine learning interatomic potentials (MLIPs), offer speed but lack accuracy and reliable uncertainty quantification.
- Bridging these fidelity gaps is crucial for efficient materials design and discovery.
Purpose of the Study:
- To develop and demonstrate a scale-bridging, multi-fidelity machine learning framework for atomistic materials simulations.
- To integrate predictions from high-fidelity (DFT) and low-fidelity (MLIP) models.
- To enable efficient materials property prediction and uncertainty quantification across alloy composition spaces.
Main Methods:
- A multi-fidelity Gaussian process (MFGP) machine learning framework was developed to fuse predictions from DFT and MLIPs.
- Uncertainty quantification was achieved through the posterior variance of the MFGP predictions.
- The framework was coupled with Bayesian optimization for efficient on-the-fly materials property optimization.
Main Results:
- The MFGP framework successfully reproduced the ternary composition dependence of the bulk modulus for aluminum-niobium-titanium random alloys.
- The approach demonstrated computational efficiency by performing an on-the-fly search for the global optimum of bulk modulus.
- This represents the first application of MFGP to fuse DFT and classical interatomic potential predictions in atomistic materials simulations.
Conclusions:
- The presented multi-fidelity machine learning framework effectively bridges scale and fidelity gaps in atomistic simulations.
- The method provides accurate property predictions with inherent uncertainty quantification, accelerating materials design.
- This approach offers a computationally efficient pathway for exploring complex materials composition spaces.

