MSz: An Efficient Parallel Algorithm for Correcting Morse-Smale Segmentations in Error-Bounded Lossy Compressors.

Area of Science:

• Data compression
• Scientific visualization
• Computational topology

Background:

• Lossy data compression often alters topological features, leading to potential misinterpretations in scientific data.
• Existing compressors do not typically preserve complex topological structures like Morse-Smale complexes.
• Topological discrepancies can compromise the accuracy of scientific conclusions derived from decompressed data.

Purpose of the Study:

• To develop a novel method for preserving topological segmentations within error-bounded lossy compressors.
• To ensure the accurate reconstruction of Morse-Smale segmentations in 2D/3D scalar fields.
• To maintain data integrity and prevent erroneous scientific conclusions caused by compression artifacts.

Main Methods:

• Deriving compression-time edits to ensure accurate topological reconstruction.
• Implementing an iterative workflow to fix extrema and integral lines until convergence.
• Utilizing shared-memory and GPU parallelism for accelerated processing.
• Applying the method to piecewise linear scalar fields.

Main Results:

• Successful preservation of Morse-Smale segmentations within specified error bounds.
• Accurate reconstruction of minimum/maximum labels induced by integral lines.
• Demonstrated effectiveness on diverse datasets from fluid dynamics, oceanography, and cosmology.
• Significant performance acceleration achieved using NVIDIA A100 GPU.

Conclusions:

• The proposed method offers a viable solution for preserving critical topological information during lossy compression.
• This approach enhances the reliability of scientific data analysis by preventing topological distortions.
• Parallel implementation makes the method practical for integration with existing compression workflows.