A faster hafnian formula for complex matrices and its benchmarking on a supercomputer
pith:V4YY6MET Add to your LaTeX paper
What is a Pith Number?\usepackage{pith}
\pithnumber{V4YY6MET}
Prints a linked pith:V4YY6MET badge after your title and writes the identifier into PDF metadata. Compiles on arXiv with no extra files. Learn more
read the original abstract
We introduce new and simple algorithms for the calculation of the number of perfect matchings of complex weighted, undirected graphs with and without loops. Our compact formulas for the hafnian and loop hafnian of $n \times n $ complex matrices run in $O(n^3 2^{n/2})$ time, are embarrassingly parallelizable and, to the best of our knowledge, are the fastest exact algorithms to compute these quantities. Despite our highly optimized algorithm, numerical benchmarks on the Titan supercomputer with matrices up to size $56 \times 56$ indicate that one would require the 288000 CPUs of this machine for about a month and a half to compute the hafnian of a $100 \times 100$ matrix.
This paper has not been read by Pith yet.
Forward citations
Cited by 1 Pith paper
-
Exponentially-improved effective descriptions of physical bosonic systems
A natural energy condition satisfied by most physical bosonic states, including outputs of universal bosonic circuits, allows the effective dimension for ε-approximations to scale as log(1/ε) instead of 1/ε², enabling...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.