Dual Numbers for Arbitrary Order Automatic Differentiation
F. Peñuñuri, K. B. Cantún-Avila, R. Peón-Escalante · 2025
Read Dual Numbers for Arbitrary Order Automatic Differentiation by F. Peñuñuri, K. B. Cantún-Avila, R. Peón-Escalante (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-04159.txt.
Description
Peñuñuri, Cantún-Avila, and Peón-Escalante build DNAOAD, in a 2025 preprint, a Fortran tool that computes derivatives of any order without the usual memory blowup. Dual numbers, a classic trick, extend arithmetic so derivatives fall out of plain evaluation. Stacking them for higher orders works in theory. In practice, nested implementations balloon in memory and crawl. The authors store the dual parts directly, in one flat representation. No recursive nesting is used at any order. Memory use drops sharply as a result.
Very high-order derivatives become practical on ordinary hardware. The framework targets high-performance scientific computing, where Fortran still rules. Problems like sensitivity analysis and series solutions need such derivatives in bulk. The authors present DNAOAD as the scalable base those jobs lacked. Its design keeps the elegance of dual numbers while shedding the cost that kept them a low-order tool.
Catalog record
| CREATOR | F. Peñuñuri, K. B. Cantún-Avila, R. Peón-Escalante |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-04159