SciML/DifferentialEquations.jl

Features

• Differential Equation Solvers - Provides a comprehensive suite of high-performance numerical solvers for ordinary, stochastic, delay, and algebraic differential equations.
• Automatic Differentiation Frameworks - Provides a comprehensive engine for computing exact gradients and Jacobians using forward and adjoint sensitivity analysis.
• Adjoint Gradient Calculation - Calculates gradients of simulation outputs by solving auxiliary equations backward in time for efficient optimization.
• Universal Differential Equation Trainers - Embeds neural networks within scientific simulators to learn unknown model components while maintaining physical constraints.
• Differentiable ODE Integrators - Integrates with differentiation libraries to compute gradients of differential equation solvers using forward and adjoint sensitivity analysis.
• Ensemble Simulation Engines - Runs large-scale parallel simulations across distributed or GPU-accelerated hardware to analyze model behavior across diverse parameter sets.
• Scientific Computing - Implements advanced parameter optimization techniques like adjoint methods for scientific models on CPUs and GPUs.
• Scientific Machine Learning - Integrates neural networks with differential equations to learn physical dynamics and perform automated model discovery.
• Sensitivity Derivative Computations - Scientific computing library calculating fast derivatives of differential equation solutions to analyze model sensitivities, calibrate parameters to data, and perform automated model discovery.
• Symbolic-Numeric Integration Engines - Combines symbolic expression manipulation with numerical solvers to automatically generate optimized code and compute exact derivatives.
• Composable Solver Interfaces - Provides a unified framework where disparate numerical methods and machine learning components interoperate through standardized functional interfaces.
• Multiple Dispatch Abstractions - Uses language-level type specialization to select optimal numerical algorithms and hardware backends without sacrificing code readability.
• Ensemble Simulation Runners - Enables the execution of multiple variations of differential equation problems in parallel by modifying initial conditions or parameters.
• Parameter Estimation Methods - Implements advanced parameter estimation and calibration tools, including Bayesian inference and sensitivity analysis for scientific models.
• Simulation Result Analyzers - Provides standardized interfaces for interpolating, differentiating, and visualizing numerical simulation results.
• Graph Coloring Algorithms - Utilizes graph coloring algorithms to identify independent components and minimize function evaluations during Jacobian computations.
• Hardware-Agnostic Parallelism - Dispatches simulation trajectories across CPUs, GPUs, and distributed clusters by abstracting execution logic from numerical integration routines.
• Batched Sensitivity Computations - Scientific computing library processing multiple parameter sets in parallel by passing matrix-based inputs to the solver to improve performance during large-scale sensitivity analysis.
• Bayesian Probabilistic Programming Frameworks - Provides a collection of methods for quantifying uncertainty through Bayesian inference and Monte Carlo propagation.
• GPU-Accelerated Solvers - Executes numerical solvers in parallel across graphics hardware to accelerate large-scale simulations.
• Symbolic Equation Discoverers - Identifies the symbolic structure or functional form of differential equations that best describe input data.
• Global Sensitivity Analyses - Scientific computing library quantifying how uncertainty in model inputs influences output variability by applying statistical methods to parameter ranges or pre-defined design matrices.
• Parallel Execution Backends - Enables selection of serial, multi-threaded, distributed, or GPU-accelerated backends to optimize performance.
• Matrix-Vector Products - Calculates Jacobian-vector and Hessian-vector products without explicit matrix construction to enable efficient iterative solving.
• Interval-Based Numerical Solvers - Performs numerical integration using floating-point intervals to provide mathematically guaranteed bounds on solution errors.
• Interval Representations - Executes rigorous computations using finite-precision floating-point intervals to guarantee solution bounds.
• Jacobian Matrix Computations - Scientific computing library reducing the number of function evaluations needed to compute sparse matrices by using graph coloring to group independent columns or rows together.
• Linear System Solvers - Provides high-performance solvers for linear systems ranging from small matrices to large-scale sparse systems.
• Monte Carlo Sampling - Quantifies uncertainty by sampling input distributions and propagating parameter sets through solvers.
• Performance Accelerators - Accelerates complex computations through automated sparsity detection, matrix coloring, and hardware offloading.
• Symbolic Code Generators - Automatically produces high-performance code for solving linear systems based on symbolic problem definitions.
• Polynomial Chaos Expansion Solvers - Provides automated computation of output probability distributions using polynomial chaos expansion techniques.
• Simulation Metrics Analysis - Computes mean, variance, and covariance bounds across ensemble trajectories to quantify uncertainty in numerical experiments.
• Coloring Algorithms - Scientific computing library assigning colors to sparse matrix columns or rows based on sparsity patterns to minimize the total number of partial derivatives required.
• Symbolic Differentiation Tools - Integrates symbolic-numeric techniques to compute precise gradients and derivatives for complex mathematical models.

Star history