The FixedHub Single Allocation Problem A Geometric Rounding固定轮毂的单配位问题.pptVIP

* 3 Then, for each city I, we divide the simplex into k polytopes. Then we check which polytope u falls into. And make the assignment accordingly. * * * * * * * These columns represent the extra percentages comparing the rounded solutions to the lower bounds generated by the first linear program relaxation In most cases LP1 already generates optimal integer assignments. * In this case, Lp1 and LP3 are both hard to solve in real time. The light brown ones are lower bounds generated by LP2 * * Now we are done with our main results. For the future research, there are two main interesting directions * Worst Case Analysis of GRA: II Theorem 3 Assume that for all , then, for FHSAP-LP1, And for FHSAP-LP3, Geometric Rounding generates the optimal linear cost and at most twice the quadratic cost in expectation for both formulations General Case Analysis Scale all hub edges. Assume the distortion r: 2r-approximation. r =(the longest hub edge)/(the shortest one) 2-approximation algorithm for 3-hub: either embed into a line or an equilateral. Simulation Results: 10 Hubs; 100 Cities Running time: LP1: hours. LP2: seconds. LP3 minutes Simulation Results: 10 Hubs; 1000 Cities Simulation Results : tight, computationally intractable even for (100, 10). : the much smaller running time comes at the expense of marginally large gaps. At most 1% worse than optimal assignments on 22 out of 32 instances comparing to . delivers high-quality solutions for large-sized problem in a reasonable amount of time. At most 1% worse than optimal ones on 28 out of 32 instances. Outperforms on 15 out of 32 instances. Performs extremely well when the graph of hubs has a (near) equilateral structure. improves on 19 out of 32 instances. Future Research Geometric Rounding and its applications Quadratic semi-assign