Tommaso Mannelli Mazzoli

AI Architect at PORR AG, Vienna, Austria

Quadratic Assignment Problem

An up-to-date, verified table of best known solutions for QAPLIB, the standard benchmark library for the Quadratic Assignment Problem (QAP) by Burkard, Karisch and Rendl. The official QAPLIB tables were last updated in April 2012. Since then, several instances have been proved optimal and the record for tai100a has been improved three times. This page brings those results together, with a permutation for every value.

This is an unofficial update, not affiliated with the QAPLIB authors or maintainers. The instance data are the official files, unchanged.

136 instances · 108 proven optimal · 28 open · every best known value backed by a permutation recomputed on the official data.

What Changed Since the Official Tables

Loading…

The lower bounds of the open instances are the best ones published, mostly from H.D. Mittelmann’s improved QAPLIB lower bounds (August 2026). A 2026 paper by Sergienko et al. reports another improvement on tai100a; it will be listed once its permutation can be checked.

Conventions

An instance file (.dat) contains n, then the n×n matrix a, then the n×n matrix b. A solution is a permutation p of 1..n, where p(i) is the location of facility i, and its cost is

cost(p) = sum over i, j of  a[i][j] * b[p(i)][p(j)]

This is the QAPLIB convention. A solution file (.sln) has a first line n value, followed by the permutation. The gap of an open instance is (BKS − lower bound) / BKS.

How the Values Were Checked

Known errors in other sources

How to Cite

Please cite QAPLIB itself, and the paper behind any best known value or proof you use:

@article{burkard1997qaplib,
  author  = {Burkard, Rainer E. and Karisch, Stefan E. and Rendl, Franz},
  title   = {{QAPLIB} -- A Quadratic Assignment Problem Library},
  journal = {Journal of Global Optimization},
  volume  = {10},
  pages   = {391--403},
  year    = {1997},
  doi     = {10.1023/A:1008293323270}
}

Contact

New best known solutions, proofs, lower bounds and corrections are welcome at tommanmaz.math@gmail.com.