Spectral graph theory studies graph Laplacians, which are calculated from a graph's degree matrix and adjacency matrix. The Laplacian is symmetric and positive-semidefinite. Spectral graph theory has applications in graph sparsification, finding isoperimetric cuts in graphs, and solving the maximum flow minimum cut problem. At Carnegie Mellon University, Gary Miller and his student Timothy Chu study spectral graph theory applications like graph sparsifiers and short cycle decomposition. Challenges include applying the theory practically and improving algorithms by changing paradigms and potentially using randomization and approximations. Future directions include improving the time complexity of algorithms for maximum flow minimum cut and isoperimetric problems.
Spectral graph theory studies graph Laplacians, which are calculated from a graph's degree matrix and adjacency matrix. The Laplacian is symmetric and positive-semidefinite. Spectral graph theory has applications in graph sparsification, finding isoperimetric cuts in graphs, and solving the maximum flow minimum cut problem. At Carnegie Mellon University, Gary Miller and his student Timothy Chu study spectral graph theory applications like graph sparsifiers and short cycle decomposition. Challenges include applying the theory practically and improving algorithms by changing paradigms and potentially using randomization and approximations. Future directions include improving the time complexity of algorithms for maximum flow minimum cut and isoperimetric problems.
Spectral graph theory studies graph Laplacians, which are calculated from a graph's degree matrix and adjacency matrix. The Laplacian is symmetric and positive-semidefinite. Spectral graph theory has applications in graph sparsification, finding isoperimetric cuts in graphs, and solving the maximum flow minimum cut problem. At Carnegie Mellon University, Gary Miller and his student Timothy Chu study spectral graph theory applications like graph sparsifiers and short cycle decomposition. Challenges include applying the theory practically and improving algorithms by changing paradigms and potentially using randomization and approximations. Future directions include improving the time complexity of algorithms for maximum flow minimum cut and isoperimetric problems.
▦ Study Graph Laplacians: Degree Matrix - Adjacency Matrix
More about the Laplacian ▦ 𝐿 = 𝑀𝑀𝑇 where 𝑀 is the incidence matrix (a 𝑉 × 𝐸 matrix where for each edge 𝑒𝑖 = 𝑢, 𝑣 , 𝑀𝑖,𝑢 = −1 and 𝑀𝑖,𝑣 = 1) ▦ Symmetric, positive-semidefinite ▦ Number of connected components is the dimension of the nullspace ▦ Can be used to count the number of spanning trees Why is Spectral Graph Theory Interesting? ▦ Graph Sparsification ▦ Isoperimetric Cuts on Graphs ▦ Max-flow Min-Cut Graph Sparsification ▦ Remove edges so that graph maintains a certain property ▦ Randomly sample edges ▦ Spanning trees have produced terrible results 𝑛 log 𝑛 ▦ Sample edges based on effective resistance, Ο edges, runs in 𝜖2 near linear time, previously Ο(𝑛 log 𝑐 𝑛) for some large c Isoperimetric Cuts ▦ Finds a minimum-cut such that the two partitions have sizes of bounded ratio ▦ Cheeger’s Inequality: a lower bound for the second smallest eigenvalue of the Laplacian ▦ Approximate sparsest cut by the second smallest eigenvalue Maximum Flow Minimum Cut ▦ Used in image segmentation and various reductions ▦ Common algorithms are exponential or at best Ο(𝑚2 ) (Ford-Fulkerson, Dinic’s, Edmond-Karp) ▦ Max-Flow = Min-Cut Recent Advances ▦ Electrical Flows, Laplacian Systems, and Faster Approximations of Maximum Flow in Undirected Graphs (Christiano et al., 2010) ▦ Computes an (1 − 𝜖)-approximation of the Max-Flow Min-Cut problem 1 11 − in Õ(𝑚𝑛 𝜖 3 3 ) time. Who at CMU Studies Spectral Graph Theory? ▦ Gary Miller ▦ Timothy Chu (PhD Student) ▦ Graph sparsifiers ▦ Spectral Sketches ▦ Short cycle decomposition Challenges ▦ Applying theory to practice ▦ Asymptotics vs. Reality ▦ Improvement in algorithms requires change in paradigms of thinking ▦ Randomization/Approximations are often necessary Future Directions for Research
Max-Flow Min-Cut: Isoperimetric:
▦ Shave down poly-log factors ▦ De-randomize
▦ Shave down ϵ factors ▦ Apply theory to practice ▦ Lower constants ▦ Prove that Max-Flow Min-Cut has the same hardness as sorting Questions
Moylett, D. J., Linden, N., & Montanaro, A. (2017) - Quantum Speedup of The Traveling-Salesman Problem For Bounded-Degree Graphs. Physical Review A, 95 (3), (032323)