Supercomputer-scale scheduling optimisation for employer-coordinated commuting (15mC Topic 2)

Long experience of building optimisation systems for airline operations: large-scale network design, robust schedule construction, crew pairing and rostering, and tail assignment — problems with millions of variables, hard legal and collective-agreement rule sets, and plans that must survive disruption rather than merely look good on paper. We also build the parallel numerical methods underneath, and we run on national and European supercomputing infrastructure (CalcUA, LUMI/EuroHPC), contributed in kind at no cost to the project.

Four transfers to urban mobility:

- Staggering at scale: coordinating start and end times across co-located employers to flatten peak-hour spikes, with each employer's own constraints intact.

- Compute that changes the question: optimising one business park fits on a laptop; optimising every large employer in a metropolitan area at once, across thousands of disruption scenarios, does not. That version tells you what the city-wide coordinated optimum is — and therefore how much is lost when coordination is left to individual sites. That gap is the empirical case for public intervention, and nobody has measured it because nobody has been able to compute it.

- Robustness, not just optimality: a commuting plan that only works on a good day changes no one's behaviour.

- Fleets and charging: tail assignment maps directly onto electrified shared and corporate fleets with charging and bidirectional-charging windows.

Open research question we want in the project: cross-employer coordination needs optimisation over data employers cannot pool (rosters, personnel, home addresses). Privacy-preserving and federated formulations decide whether such coalitions are viable at all.

We are aware that rescheduling shifts can be regressive. Off-peak work is unequally distributed, and any optimisation objective encodes a judgement about whose time gets moved. We want partners who treat that as a core research question, not a dissemination task.