The Boundary Method and General Auction for Optimal Mass Transportation and Wasserstein Distance Computation
Series: Dissertation Defense
Dissertation advisor: Luca Dieci
Numerical optimal transport is an important area of research, but most problems are too large and complex for easy computation. Because continuous transport problems are generally solved by conversion to either discrete or semi-discrete forms, I focused on methods for those two. I developed a discrete algorithm specifically for fast approximation with controlled error bounds: the general auction method. It works directly on real-valued transport problems, with guaranteed termination and a priori error bounds. I also developed the boundary method for semi-discrete transport. It works on unaltered ground cost functions, rapidly identifying locations in the continuous space where transport destinations change. Because the method computes over region boundaries, rather than the entire continuous space, it reduces the effective dimension of the discretization. The general auction is the first relaxation method designed for compatibility with real-valued costs and weights. The boundary method is the first transport technique designed explicitly around the semi-discrete problem and the first to use the shift characterization to reduce dimensionality. No truly comparable methods exist. The general auction and boundary method are able to solve many transport problems that are intractible using other approaches. Even where other solution methods exist, in testing it appears that the general auction and boundary method outperform them.
Monday, April 3, 2017 - 16:30 , Location: Skiles 006 , Sarah Rasmussen , University of Cambridge , Organizer: Caitlin Leverson
Exploring when a closed oriented 3-manifold has vanishing reduced Heegaard Floer homology---hence is a so-called L-space---lends insight into the deeper question of how Heegaard Floer homology can be used to enumerate and classify interesting geometric structures. Two years ago, J. Rasmussen and I developed a tool to classify the L-space Dehn surgery slopes for knots in 3-manifolds, and I later built on these methods to classify all graph manifold L-spaces. After briefly discussing these tools, I will describe my more recent computation of the region of rational L-space surgeries on any torus-link satellite of an L-space knot, with a result that precisely extends Hedden’s and Hom’s analogous result for cables. More generally, I will discuss the region of L-space surgeries on iterated torus-link satellites and algebraic link satellites, along with implications for conjectures involving co-oriented taut foliations and left-orderable fundamental groups.
Monday, April 3, 2017 - 15:15 , Location: Skiles 006 , Jo Nelson , Barnard College, Columbia University , Organizer: Caitlin Leverson
I will discuss joint work with Hutchings which gives a rigorousconstruction of cylindrical contact homology via geometric methods. Thistalk will highlight our use of non-equivariant constructions, automatictransversality, and obstruction bundle gluing. Together these yield anonequivariant homological contact invariant which is expected to beisomorphic to SH^+ under suitable assumptions. By making use of familyFloer theory we obtain an S^1-equivariant theory defined with coefficientsin Z, which when tensored with Q recovers the classical cylindrical contacthomology, now with the guarantee of well-definedness and invariance. Thisintegral lift of contact homology also contains interesting torsioninformation.
Monday, April 3, 2017 - 14:00 , Location: Skiles 005 , Prof. Michael Muskulus , NTNU: Norwegian University of Science and Technology , email@example.com , Organizer: Joseph Walsh
This talk addresses an important problem in arctic engineering due to interesting dynamic phenomena in a forced linear system. The nonautonomous system considered is representative of a whole class of engineering problems that are not approachable by standard techniques from dynamical system theory.The background are ice-induced vibrations of structures (e.g. wind turbines or measurement masts) in regions with active sea ice. Ice is a complex material and the mechanism for ice-induced vibrations is not fully clear at present. In particular, the conditions under which the observed, qualitatively different vibration regimes are active cannot be predicted accurately so far. A recent mathematical model developed by Delft University of Technology assumes that a number of parallel ice strips are pushing with a constant velocity against a flexible structure. The structure is modelled as a single degree of freedom harmonic oscillator. The contact force acts on the structure, but at the same time slows down the advancement of the ice, thereby introducing a dynamic nonlinearity in the otherwise linear system. When the local contact force becomes large enough, the ice crushes and the corresponding strip is reset to a random offset in front of the structure.This is the first mathematical model that exhibits all three different dynamic regimes that are observed in reality: for slow ice velocities the structure undergoes quasi-static sawtooth responses where all ice strips fail at the same time (a kind of synchronization phenomenon), for large ice velocities the structure response appears random, and for intermediate ice velocities the system exhibits vibrations at the structure eigenfrequency, commonly called frequency lock-in behavior. The latter type of vibrations causes a lot of damage to the structure and poses a safety and economic risk, so its occurrence needs to be predicted accurately.As I will show in this talk, the descriptive terms for the three vibration regimes are slightly misleading, as the mechanisms behind the observed behaviors are somewhat different than intuition suggests. I will present first results in analyzing the system and offer some explanations of the observed behaviors, as well as some simple criteria for the switch between the different vibration regimes.
Monday, April 3, 2017 - 14:00 , Location: Skiles 006 , Josh Greene , Boston College , Organizer: John Etnyre
I will describe a diagrammatic classification of (1,1) knots in S^3 and lens spaces that admit non-trivial L-space surgeries. A corollary of the classification is that 1-bridge braids in these manifolds admit non-trivial L-space surgeries. This is joint work with Sam Lewallen and Faramarz Vafaee.
Friday, March 31, 2017 - 15:05 , Location: Skiles 254 , Lei Zhang , School of Mathematics, GT , Organizer: Jiaqi Yang
In this talk, we will give an introduction to the variational approach to dynamical systems. Specifically, we will discuss twist maps and prove the classical results that area-preserving twist map has Birkhoff periodic orbits for each rational rotation number.
Series: Professional Development Seminar
A conversation with Dan Margalit, GT math professor and inaugural CoS Leddy Family Faculty Fellow, who was a tenure-track assistant professor at Tufts University for two years prior to coming to Tech.
Series: ACO Student Seminar
Using some classical results of invariant theory of finite reflection groups, and Lagrange multipliers, we prove that low degree or sparse real homogeneous polynomials which are invariant under the action of a finite reflection group $G$ are nonnegative if they are nonnegative on the hyperplane arrangement $H$ associated to $G$. That makes $H$ a test set for the above kind of polynomials. We also prove that under stronger sparsity conditions, for the symmetric group and other reflection groups, the test set can be much smaller. One of the main questions is deciding if certain intersections of some simply constructed real $G$-invariant varieties are empty or not.
Series: Algebra Seminar
Networks, or graphs, can represent a great variety of systems in the real world including neural networks, power grid, the Internet, and our social networks. Mathematical models for such systems naturally reflect the graph theoretical information of the underlying network. This talk explores some common themes in such models from the point of view of systems of nonlinear equations.
Series: Stochastics Seminar
We consider the problem of studying the limiting distribution of the number of monochromatic two stars and triangles for a growing sequence of graphs, where the vertices are colored uniformly at random. We show that the limit distribution of the number of monochromatic two stars is a sum of mutually independent components, each term of which is a polynomial of a single Poisson random variable of degree 1 or 2. Further, we show that any limit distribution for the number of monochromatic two stars has an expansion of this form. In the triangle case the problem is more challenging, as in this case the class of limit distributions can involve terms with products of Poisson random variables. In this case, we deduce a necessary and sufficient condition on the sequence of graphs such that the number of monochromatic triangles is asymptotically Poisson in distribution and in the first two moments. This work is joint with Bhaswar B. Bhattacharya at University of Pennsylvania.