FPEAnalysis: Automating Floating Point Error Analysis [2016]

Manual floating-point error analysis of scientific computing code is tedious and error-prone. Under the supervision of Professor David Bindel of computer science at Cornell University, I conducted my research to automated these analyses using the “1+ δ” error model and developed the FPEAnalysis software system. FPEAnalysis analyzes a piece of code, automatically reports any numerically unstable subexpressions, and gives insights for resolving these instabilities. The system can detect numerical instabilities in many textbook examples, leading Professor Bindel to use the system as an educational tool for his introductory scientific computing course.