Optimized Schedule for Large Crystallography Meetings

International crystallography meetings like the International Union of Crystallography, European Crystallographic Meetings and American Crystallographic Association ones currently offer very comprehensive programs with four parallel sessions or more over a week or so. Proper scheduling of these sessions is nearly as important for the scientific success of these meetings as the selection of topics. In order to tackle this problem objectively, it is proposed here to poll crystallographers about the sessions they most wish to attend, and to accept that schedule which minimizes the sum of residual conflicts calculated from the results of the poll and also respects a number of constraints. A novel semi-automated optimization in four steps, based on the existence of major interest profiles among participants, was applied to the scheduling of the American Crystallographic Association 1995 Annual Meeting: a simple rearrangement of a `conflict matrix' to produce clusters of sessions with large conflicts, followed by application of constraints within clusters in step 2, then a small number of permutations within clusters only, each of them lowering the sum of conflicts while respecting constraints, and finally assignment of sessions to rooms. Without necessarily producing the absolute best solution, solutions very superior to those from intuition and respecting constraints can be built in this way with limited computing power. The experimental data were taken from 102 replies to an e-mail poll of crystallographers at large. Only 20 permutations were performed in step 3. The solution respected all constraints but one, which was relaxed in order to avoid a bad conflict. At the meeting, the room assignments turned out to be mostly verified, with just a few surprises, all having to do with outstanding contributed talks at sessions with no invited speakers. The total attendance at sessions remained high for the whole duration of the meeting. Calculations were programmed in Fortran and performed on a 386 PC clone.