### Resumé

Originalsprog | Engelsk |
---|---|

Publikationsdato | 4. dec. 2014 |

Antal sider | 20 |

Status | Udgivet - 4. dec. 2014 |

Udgivet eksternt | Ja |

### Fingeraftryk

### Citer dette

*Faster Exact Algorithms for Computing Steiner Trees in Higher Dimensional Euclidean Spaces*.

}

**Faster Exact Algorithms for Computing Steiner Trees in Higher Dimensional Euclidean Spaces.** / Fonseca, Rasmus; Brazil, Marcus; Winter, Pawel; Zachariasen, Martin.

Publikation: Konferencebidrag uden forlag/tidsskrift › Paper › Forskning › peer review

TY - CONF

T1 - Faster Exact Algorithms for Computing Steiner Trees in Higher Dimensional Euclidean Spaces

AU - Fonseca, Rasmus

AU - Brazil, Marcus

AU - Winter, Pawel

AU - Zachariasen, Martin

PY - 2014/12/4

Y1 - 2014/12/4

N2 - The Euclidean Steiner tree problem asks for a network of minimum total length interconnecting a finite set of points in d-dimensional space. For d ≥ 3, only one practical algorithmic approach exists for this problem --- proposed by Smith in 1992. A number of refinements of Smith's algorithm have increased the range of solvable problems a little, but it is still infeasible to solve problem instances with more than around 17 terminals. In this paper we firstly propose some additional improvements to Smith's algorithm. Secondly, we propose a new algorithmic paradigm called branch enumeration. Our experiments show that branch enumeration has similar performance as an optimized version of Smith's algorithm; furthermore, we argue that branch enumeration has the potential to push the boundary of solvable problems further.

AB - The Euclidean Steiner tree problem asks for a network of minimum total length interconnecting a finite set of points in d-dimensional space. For d ≥ 3, only one practical algorithmic approach exists for this problem --- proposed by Smith in 1992. A number of refinements of Smith's algorithm have increased the range of solvable problems a little, but it is still infeasible to solve problem instances with more than around 17 terminals. In this paper we firstly propose some additional improvements to Smith's algorithm. Secondly, we propose a new algorithmic paradigm called branch enumeration. Our experiments show that branch enumeration has similar performance as an optimized version of Smith's algorithm; furthermore, we argue that branch enumeration has the potential to push the boundary of solvable problems further.

M3 - Paper

ER -