Topology-Preserving Scientific Data Compression

Scientific simulations and instruments generate data at volumes that make lossy compression essential for practical storage and transfer. Yet conventional compressors, even those with strict pointwise error bounds, can alter the geometric and topological features that scientists rely on for analysis, including isosurfaces, merge trees, and Morse–Smale complexes. This talk presents general frameworks for topology-preserving scientific data compression across scalar, vector, and tensor fields. These approaches preserve a broad class of geometric and topological features while achieving compression ratios comparable to, and in some cases better than, specialized methods. Together, they provide a flexible path toward more reliable compressed scientific data and more trustworthy downstream analysis. This talk is based on joint works with Nathaniel Gorski, Xin Liang, Hanqi Guo, and Lin Yan.