Software

  1. Hamiltonian Operator Inference, Anthony Gruber, Irina Tezaur.
    This code reproduces results contained in the papers Gruber and Tezaur 2023, Gruber and Tezaur 2024. Additional functionality is provided for carrying out canonical, noncanonical, and variationally consistent Hamiltonian operator inference, along with example scripts targeting applications to linear elasticity and nonlinear models for water waves.
  2. FFTMatvec: Fast and scalable FFT-based GPU-accelerated algorithms for Hessian actions arising in linear inverse problems governed by autonomous dynamical systems Sreeram Venkat, Milinda Fernando, Stefan Henneking, Omar Ghattas.
    FFTMatvec is a CUDA/C++ library that enables fast matrix vector products (matvecs) involving block triangular Toeplitz matrices. The software is scalable over multi-GPU multi-node systems, and uses NCCL for inter-GPU communication. Though the library is suitable for generic block triangular Toeplitz matvecs, the primary application is to Bayesian inverse problems governed by autonomous dynamical systems.
  3. Multi-fidelity Bayesian Optimization with Multiple Information Sources of Input-dependent Fidelity Mingzhou Fan, Byung-Jun Yoon, Edward R. Dougherty, Nathan M. Urban, Francis J. Alexander, Raymundo Arroyave, Xiaoning Qian.
  4. CuPF: A fast GPU accelerated phase field solver Yigong Qin and George Biros.
    cuPF is a CUDA library that solves phase field equations in both dendrite and grain scales. The library supports 2D dendrite growth simulations and 2D/3D grain growth simulations with specified heat sources. It enables one-way coupling with macro-scale temperature profiles for various melt pool shapes, such as planar, spot-weld, and moving laser configurations, and it includes capabilities for modeling grain nucleation. The solver is optimized for scalability using CUDA-aware MPI, allowing deployment on multi-GPU and multi-node systems. Its accuracy has been validated through multiple convergence studies.
  5. GrainGNN: A dynamic heterogeneous graph neural network for large-scale 3D grain microstructure evolution Yigong Qin and George Biros.
    GrainGNN is a graph neural network designed for 3D grain growth under additive manufacturing conditions. It uses LSTM for grain feature evolution and transformer architecture to learn spatial grain interactions. It supports inference for different numbers of grains, grain size distributions, and thermal conditions. It is about 150-2000 times faster than GPU-accelerated phase field simulations.
  6. OpInf: Data-driven model reduction of dynamical systems Shane McQuarrie, Marco Tezzele, Benjamin Zastrow, Nicole Aretz, and Karen Willcox.
    This package is a Python implementation of Operator Inference (OpInf), a projection-based model reduction technique for learning polynomial reduced-order models of dynamical systems. The procedure is data-driven and non-intrusive, making it a viable candidate for model reduction of “glass-box” systems where the structure of the governing equations is known but intrusive code queries are unavailable.
  7. hIPPYlib: Inverse Problem PYthon library Umberto Villa et. al.
    hIPPYlib implements state-of-the-art scalable adjoint-based algorithms for PDE-based deterministic and Bayesian inverse problems. It builds on FEniCS for the discretization of the PDE and on PETSc for scalable and efficient linear algebra operations and solvers.
  8. Monotone Parameterization Toolkit (MParT): A core library for constructing and using transport maps Daniel Sharp, Matthew Parno, Michael Brennan, Ricardo Baptista, Henning Bonart, Paul-Baptiste Rubio, Youssef Marzouk.
  9. SOUPy: Stochastic Optimization under Uncertainty in Python Dingcheng Luo, Peng Chen, Thomas O’Leary-Roseberry, Umberto Villa and Omar Ghattas. JOSS paper
    SOUPy implements scalable algorithms for the optimization of large-scale complex systems governed by partial differential equations (PDEs) under high-dimensional uncertainty. The library features various risk measures (such as mean, variance, superquantile/condition-value-at-risk and user-defined risk measures), probability/chance constraints, and optimization/state constraints. SOUPy enables efficient PDE-constrained optimization under uncertainty through parallel computation of the risk measures and their derivatives (gradients and Hessians). The library also provides built-in parallel implementations of optimization algorithms (e.g. BFGS, Inexact Newton CG), as well as an interface to the scipy.optimize module.
  10. hippyflow: Dimension reduced surrogate construction for parametric PDE maps Thomas O’Leary-Roseberry et. al.
    hippyflow is a toolkit for the construction of dimension-reduced surrogates for infinite-dimensional parametric PDE maps. This code handles the construction of different reduced bases (e.g., KLE, POD and active subspace) via generalized eigenvalue problems. This code also automates the construction of training data including Fréchet derivatives of PDE outputs with respect to input parameters to be used in derivative-informed operator learning. This includes formulations of parametric PDEs with uncertain parameters and control parameters.
  11. Multifidelity-UQ-Greenland Nicole Aretz, Karen Willcox.
    Supplementary material.
  12. Nested-OpInf Nicole Aretz, Karen Willcox.
    Supplementary material.