We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.NA

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Physics > Computational Physics

Title: Not-Quite Transcendental Functions and their Applications

Abstract: Transcendental functions, such as exponentials and logarithms, appear in a broad array of computational domains: from simulations in curvilinear coordinates, to interpolation, to machine learning. Unfortunately they are typically expensive to compute accurately. In this note, we argue that in many cases, the properties of the function matters more than the exact functional form. We present new functions, which are not transcendental, that can be used as drop-in replacements for the exponential and logarithm in many settings for a significant performance boost. We show that for certain applications using these functions result in no drop in the accuracy at all, as they are perfectly accurate representations of themselves, if not the original transcendental functions.
Comments: Submitted as a short note to the journal of computational physics
Subjects: Computational Physics (physics.comp-ph); Performance (cs.PF); Numerical Analysis (math.NA)
Report number: LA-UR-22-25573
Cite as: arXiv:2206.08957 [physics.comp-ph]
  (or arXiv:2206.08957v1 [physics.comp-ph] for this version)

Submission history

From: Jonah Miller [view email]
[v1] Fri, 17 Jun 2022 18:14:10 GMT (66kb,D)

Link back to: arXiv, form interface, contact.