Senior Applied Research Scientist – AI Native Numerical Methods

At NVIDIA, we are using accelerated computing and machine learning to create a new generation of AI-native scientific, engineering, and industrial simulation. We are seeking an applied researcher who can connect machine learning with numerical algorithms to make solvers faster, more reliable, and more efficient. This work focuses on ML methods inside and around numerical solvers, including AI-guided multigrid, solver-in-the-loop learning, differentiable simulation integrated with AI algorithms, and hybrid numerical / ML methods. We work across NVIDIA solver, simulation, and computational engineering platforms, with collaborators NVIDIA Research, universities, and industrial simulation teams.

The primary focus of this role is to invent AI-native numerical algorithms that combine machine learning with classical scientific and engineering solvers on modern GPU architectures.

What you'll be doing

Research AI-assisted numerical methods that improve convergence, stability, accuracy, robustness, and wall-clock performance for large-scale scientific, engineering, and industrial simulations.

Invent learned coarse spaces, learned preconditioners, AI-guided multigrid methods, sequence-aware solver strategies, solver-control policies, differentiable solver components, and hybrid numerical / ML algorithms.

Build solver-in-the-loop pipelines using residual histories, discretized operators, meshes, geometry, simulation outputs, performance counters, and physics constraints.

Define evaluation methods that measure numerical impact, including convergence rate, failure rate, conservation, memory footprint, correctness, and end-to-end speedup.

Collaborate with teams across numerical methods, CUDA-X, Warp, PhysicsNeMo, NVIDIA Research, CAE, EDA, semiconductor, electronics, thermal-fluid, electromagnetics, and digital twin workflows.

Help define NVIDIA's applied research agenda for AI-native numerical methods and solver intelligence.

What we need to see

PhD or equivalent experience in computer science, machine learning, scientific computing, applied mathematics, computational engineering, physics, or a related field.

5 years of relevant work/research experience

Background in machine learning and scientific computing, with evidence of connecting ML methods to numerical algorithms.

Experience with PyTorch, JAX, or comparable deep learning frameworks, plus Python and GPU computing.

Working knowledge of PDEs, sparse linear algebra, iterative solvers, preconditioning, finite element / finite volume methods, optimization, or differentiable programming.

Research record in scientific machine learning, numerical linear algebra, or differentiable simulation integrated with numerical solvers.

Communication skills that support collaboration across AI research, numerical methods, product, and production software teams.

Ways to stand out from the crowd

Evidence that ML methods improved real numerical solvers through faster convergence, fewer failures, improved robustness, or lower cost on industrial-scale simulations.

Experience with AI-guided multigrid, reduced-order components inside solver algorithms, neural operators connected to solver workflows, or automated solver control.

Experience with differentiable simulation, PDE-constrained learning, inverse design, uncertainty quantification, Bayesian methods, reinforcement learning for solver control, or automated algorithm selection.

Publications, software contributions, or collaborations in scientific computing, matrix computations, industrial simulation, CAE, EDA, semiconductor simulation, electronic build, thermal-fluid simulation, electromagnetics, or digital twins.