Public transport network design is essential for sustainable, accessible urban mobility, yet exact mathematical optimization methods for the transit network design problem (TNDP) remain largely confined to small, artificial benchmark networks due to their computational complexity. This project develops and validates methods to scale exact optimization approaches to real-world transit networks. The project builds on a two-step sequential mixed-integer linear programming (MILP) decomposition of the TNDP, in which a strategic step selects infrastructure and routes passengers along shortest paths, and a tactical step determines transit line routes, frequencies, and passenger assignment. This decomposition, together with formally proven symmetry-breaking constraints, has been shown to substantially improve computational tractability relative to a simultaneous formulation, validated on Mandl's benchmark network. The current phase of the project extends this method toward real-world applicability through an intermediate network contraction technique applied between the two model steps. Because passenger routing costs are already fixed after the first step, the underlying road network can be reduced in size before the more computationally demanding second step is solved, without compromising the accuracy of line design decisions. Origin-destination demand is conservatively remapped onto the contracted network, and the resulting overestimation of passenger loads is formally characterized and empirically quantified. The method is validated on a real-world case study: the inner-city bus network of Södertälje, Sweden, using automatic vehicle location and smart card data. The uncontracted network (100 nodes, 216 edges, 2,690 OD pairs) is reduced to a contracted network (33 nodes, 39 edges, 555 OD pairs) for the tactical optimization step, substantially reducing the size of the resulting MILP. Solving these models to proven or near-proven optimality with Gurobi requires substantial computation. Preliminary local testing indicates that the tactical step alone, applied to the real-world Södertälje network, involves MILP instances with hundreds of thousands of variables and only reach decent solutions after 12+ hours of running time, indicating that meaningful experiments require both longer per-run time limits and many independent solves across multiple scenarios (varying the operator-passenger cost trade-off and network contraction settings). This project therefore requests computational resources to conduct a systematic experimental evaluation of the network contraction method's computational efficiency gains and any associated loss of solution quality, benchmarked against the uncontracted formulation. The resulting methodology aims to extend exact, tractable optimization for transit network design from small artificial test cases to real-world city-scale networks, directly supporting the practical applicability of formally grounded network design methods for public transport authorities. Main supervisor: Prof. Erik Jenelius jenelius@kth.se Erik Jenelius is a full Professor in Public Transport Systems and Head of the Division of Transport Planning within the Department of Civil and Architectural Engineering at KTH.