Compression of turbulent flow data through an extended space-time correlation: certification against the modal energy budget
Abstract
The fidelity of direct numerical simulation now outpaces the storage and data-movement capacity available for the resulting fields. We present an open, distributed, GPU-accelerated low-rank compression framework for large turbulent flow databases. We assess it using two direct numerical simulations of a controlled diffusion compressor cascade under laminar and turbulent inflow conditions. Compression is formulated using a correlation operator over space and time, which is two orders of magnitude larger than that of snapshot POD. Its dominant range is extracted with a randomized range finder based on communication-avoiding hierarchical QR. The compressed databases are evaluated through a hierarchy of increasingly strict diagnostics, from instantaneous reconstruction and second-order statistics to the modal turbulent kinetic energy budget and its nonlinear interscale transfer. The budget terms are far more sensitive to compression than fieldwise or one-point errors. They remain accurate at conservative compression and degrade sharply once the retained energy is relaxed. The near-wall dissipation and the interscale transfer degrade the most and set the admissible compression level. An aggressive reduction preserves every one-point statistic to within a few per cent while corrupting both by an order of magnitude more. The dissipation error is confined to the viscous sublayer and grows with the development of the boundary layer. It is carried by the high-order modes. For both inflows, the linear budget terms scale with the discarded energy to the power $0.8$ and the triadic transfer to a power above one.
Related
- Extended proper orthogonal decomposition of non-homogeneous thermal fields in a turbulent pipe flow
- A Priori Estimation of the Non-Linear Compression Potential and Quantification of Autoencoder Non-Linearity in Turbulent Flows
- Certification cost of quantum models: measurement correlation, not parameter count
- Analytical model for structured light propagation through a turbulent atmosphere
- Genuine certification of incompatible quantum instruments through sequential communication tasks
BibTeX
@misc{lopes2026compression,
title = {{Compression of turbulent flow data through an extended space-time correlation: certification against the modal energy budget}},
author = {Gustavo Lopes and Davide Lengani and Dan Henningson},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.32427},
eprint = {2610.32427},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.32427},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Compression of turbulent flow data through an extended space-time correlation: certification against the modal energy budget
AU - Gustavo Lopes
AU - Davide Lengani
AU - Dan Henningson
PY - 2026
DA - 2026/10/23
PB - arXiv
DO - 10.48550/arxiv.2610.32427
UR - https://arxiv.org/abs/2610.32427
AB - The fidelity of direct numerical simulation now outpaces the storage and data-movement capacity available for the resulting fields. We present an open, distributed, GPU-accelerated low-rank compression framework for large turbulent flow databases. We assess it using two direct numerical simulations of a controlled diffusion compressor cascade under laminar and turbulent inflow conditions. Compression is formulated using a correlation operator over space and time, which is two orders of magnitude larger than that of snapshot POD. Its dominant range is extracted with a randomized range finder based on communication-avoiding hierarchical QR. The compressed databases are evaluated through a hierarchy of increasingly strict diagnostics, from instantaneous reconstruction and second-order statistics to the modal turbulent kinetic energy budget and its nonlinear interscale transfer. The budget terms are far more sensitive to compression than fieldwise or one-point errors. They remain accurate at conservative compression and degrade sharply once the retained energy is relaxed. The near-wall dissipation and the interscale transfer degrade the most and set the admissible compression level. An aggressive reduction preserves every one-point statistic to within a few per cent while corrupting both by an order of magnitude more. The dissipation error is confined to the viscous sublayer and grows with the development of the boundary layer. It is carried by the high-order modes. For both inflows, the linear budget terms scale with the discarded energy to the power $0.8$ and the triadic transfer to a power above one.
ER -