Quasi-matrix-free hybrid multigrid on dynamically adaptive Cartesian grids
- Submitting institution
-
University of Durham
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 113330
- Type
- D - Journal article
- DOI
-
10.1145/3165280
- Title of journal
- ACM Transactions on Mathematical Software
- Article number
- 32
- First page
- 32:1
- Volume
- 44
- Issue
- 3
- ISSN
- 00983500
- Open access status
- Compliant
- Month of publication
- -
- Year of publication
- 2018
- URL
-
https://doi.org/10.1145/3165280
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
1
- Research group(s)
-
A - Innovative Computing
- Citation count
- 5
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- A new notion of algebraic-geometric multigrid is introduced. It combines the low memory footprint of geometric multigrid with the robustness of algebraic multigrid and thus allows to solve problems which so-far required the setup of large (matrix) data structures without any helper data structure. This becomes possible through an in-situ data compression and yields an algorithm with lower memory footprint than recent Gordon-Bell prizes (the world-cup in supercomputing). For setups which require frequent updates to the matrices, it outperforms PETSc – a de-facto standard toolbox in scientific computing. So far, neither other compression approaches nor algorithms have beaten this approach.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -