This file contains explanations to the columns of fulleval.csv.
In case you have any further questions, please contact the authors.
Contact information is provided on the project homepage.

name  		-	The name of the problem instance: First three letters contain the author of the test case followed by the number of items and a unique identifier
			Example: bur26a = Burkhardt et al., 26 Items, Test case a
Kaufman-Broeckx	-	Contains the best lower bound produced by the Kaufman-Broeckx linearization plugged in the Gurobi ILP solver
RLT1		-	Contains the best lower bound produced by the RLT1 linearization plugged in the Gurobi ILP solver
Mosek		-	Describes the performance of our own branch-and-bound framework with Mosek as the underlying SDP solver, average objective function is chosen
Gurobi(quad)	-	Desribes the performance of our SDP relaxation in the Gurobi ILP solver, max objective is used, the quadratic QAP objective is added as a constraint
Gurobi(kb)	-	Same as above; however, the quadratic constraint is replaced by the (linear) Kaufman-Broeckx formulation

Note that the conference paper only describes the Mosek framework (see Section 2 of the paper linked at the project page).
The journal version of this paper contains the description of the Gurobi approach, too, and will be added upon approval.