Numerical Correctness
Performance and Numerical Reliability of Floating-Point Computation
Floating-point computation is fundamental to scientific computing, yet achieving both high performance and numerical reliability remains challenging. Our research investigates the accuracy, efficiency, and implementation of floating-point computations across numerical algorithms, precision formats, and hardware architectures.
Interval Arithmetic
Interval computing provides reliable bounds on numerical results, but its practical adoption is challenged by computational overhead, enclosure tightness, and programming complexity. We explore the trade-offs between numerical reliability and performance, including efficient parallel execution, interval-aware control flow, and compiler optimization. Our goal is to make interval computing both efficient and accessible for demanding scientific and engineering applications.