SIAM Undergraduate Research Online (SIURO)
Click on title to view PDF of paper or right click to download paper.
Undergraduate Student Research
Published electronically January 19, 2018
Authors: Vinit Ranjan, Junmo Ryang, and Kelly Zhang (Duke University)
Sponsor: David Kraines (Duke University)
Abstract: The goal of this paper was to assess the effect of self driving cars on traffic conditions in the Greater Seattle Area using data from the Consortium for Mathematics and Its Applications in the 2017 Mathematical Contest in Modeling. We built a computational, agent based, mathematical model by which we could vary parameters such as number of lanes and capacity of cars. After polling a sufficient sample space, we ran our model and used curve fitting techniques to create functions to model the system. We used the model to calculate average speeds for various highways in Seattle. After creating our model, we adapted the computational model to allow for self driving cars to show increase in average car speed in various conditions. Our results show the increase in average speed with an increasing percentage of self driving cars in terms of increased average speed. The advantages of our model are the agent based aspects, which allow us to observe and model the system's behavior. Using this data to interpolate surfaces allows for more analytic techniques as well. The computational model is also flexible enough to poll data for different traffic conditions in other cities.
Published electronically January 23, 2018
Author: Shankar Sivarajan (Indian Institute of Science)
Sponsor: Y. Narahari (Indian Institute of Science)
Abstract: In this paper, we propose a family of approval voting-schemes for electing committees based on the preferences of voters. In our schemes, we calculate the vector of distances of the possible committees from each of the ballots and, for a given p-norm, choose the one that minimizes the magnitude of the distance vector under that norm. The minisum and minimax methods suggested by previous authors and analyzed extensively in the literature naturally appear as special cases corresponding to p = 1 and p = 1; respectively. Supported by examples, we suggest that using a small value of p; such as 2 or 3, provides a good compromise between the minisum and minimax voting methods with regard to the weightage given to approvals and disapprovals. For large but finite p; our method reduces to finding the committee that covers the maximum number of voters, and this is far superior to the minimax method which is prone to ties. We also discuss extensions of our methods to ternary voting.
Published electronically January 24, 2018
Author: Kira Parker (University of Utah)
Sponsor: Braxton Osting (University of Utah)
Abstract: Ranking methods are used in all aspects of life, from Google searches to sports tournaments. Because all ranking methods necessarily have advantages and disadvantages, USA Climbing, the organizer of national climbing competitions in the United States, changed their ranking method three times between 2009 and 2016. The combined rank method employed in 2015 marked a drastic step away from the previous two in that it failed to meet the independence of irrelevant alternatives (IIA) criterion and was almost impossible for spectators to use to calculate ranks on their own. We compare this more recent rank aggregation method with older USA Climbing score aggregation methods as well as other methods from the literature. Three particularly important methods we consider are (i) the combined rank method, (ii) a combination of the previous two USA Climbing score aggregation methods (the merged method), and (iii) a linear programming (LP)-based rank aggregation method from the literature. Using data from the 2016 Bouldering Youth National Championships, we perform leave-one-out cross validation and the Friedman hypothesis test to conclude that at the 99% confidence level, the LP-based rank aggregation method has significantly more predictive power than the other two methods, while there was insufficient evidence to distinguish between the predictive power of the combined rank method and the merged method. However, due to the desirable properties, such as the IIA criterion, satisfied by the merged method, we recommend this method for use in competitive climbing.
An Alternating Minimization Method to Train Neural Network Models for Brain Wave Classification
Published electronically February 21, 2018
Author: Grant Sheen (Sage Hill High School)
Sponsor: Knut Solna (UC Irvine)
Abstract: An alternating minimization (AM) method, which updates variables one-by-one while fixing the rest, is developed to train a neural network with low rank weights for brainwave classification. The training involves minimizing a non-smooth and non-convex cross entropy loss function. The neural network model does a projection inside a hidden layer for low dimensional feature extraction. The sub-problem for each variable is shown to be either convex or piece-wise convex with a finite number of minima. The sub-problems are solved using the bisection method with bisection intervals analytically derived. The overall iterative AM method is descending and convergent, free of step size (learning parameter) in the standard gradient descent method. Experiments consist of noninvasive multiple electrode recordings and the classification of brain wave data. The AM method significantly outperforms the standard neural network model trained by the gradient descent method in classifying four thoughts for both normal and Alzheimer subjects.
Published electronically February 21, 2018
Authors: William Noland (North Central College), Dylan Smith (University of Connecticut), and Seth Selken (Iowa State University)
Sponsor: Sergei Fomin (California State University, Chico)
Abstract: Previous literature has investigated the run-up and draw-down of tsunami waves on a one-dimensional, constant-sloped beach, but the existing solutions are complex and computationally unwieldy. Our research aims to establish a simpler model while still obtaining accurate results. We do so by using a quasi-linear theory derived from the nonlinear shallow-water wave equations. These equations are considered over a linear beach with properly imposed initial and boundary conditions. The main difficulty in solving this problem is the moving boundary associated with the shoreline motion. To eliminate this difficulty, we apply an appropriate substitution to the spatial variable, and thus replace the moving boundary of the computational domain with a stationary boundary. A key feature of our tsunami problem is the presence of the small parameter ε = η0/h0, where η0 is the characteristic amplitude of the wave and h0 is the characteristic depth of the ocean. Due to the presence of this small parameter, the problem can be essentially linearized using the method of perturbations and then solved analytically via an integral transformation. Our explicit solution enables us to swiftly predict the behavior of the wave using an essentially linear model. We test the accuracy of our model against the numerical solution obtained using Mathematica, and find minimal discrepancies. Finally, we extend our results to a modified beach configuration that more accurately reflects real-world shoreline topography.
Published electronically March 5, 2018
Author: Kirsten Giesbrecht (Centre College)
Sponsor: Ellen Swanson (Centre College)
Abstract: If an expectant mother develops a splenic artery aneurysm that ruptures, there is a 90% fetal mortality rate. Splenic artery aneurysms account for 46-60% of visceral aneurysms in the human abdomen. The cause for the propensity of aneurysms to develop and rupture along the splenic artery is unknown. A distinguishing characteristic of this artery is its tortuous shape. We investigate how the unique geometries of the artery may lead to unusual patterns in blood flow. Dramatic changes in blood flow properties such as blood pressure, velocity and wall shear are conducive for aneurysm formation, development, and rupture. Using Ansys Fluent computational fluid dynamics software, the influence of these elements of blood flow through both straight and curved arteries is explored. Under identical initial and boundary conditions, the curved artery reaches both higher dynamic pressure and wall shear stress values than the straight artery. The curved artery also has a greater range of dynamic pressure and wall shear stress values compared to the range of values in the straight artery. Additionally, blood flow changes associated with pregnancy, including increased peak velocity, are shown to increase the risk of aneurysm development and rupture. These results indicate that the curved geometry of the splenic artery predisposes it to promote aneurysm rupture and growth.
L1 Regularization for Compact Support
Published electronically March 14, 2018
Author: Timothy Li (Carnegie Mellon University)
Sponsor: Giovanni Leoni (Carnegie Mellon University)
Abstract: In this paper, we solve a conjecture of Osher and Yin that L1 regularization of certain minimization problems ensures compact support of minimizers. In particular, we consider a class of minimization problems and show that adding an L1 term forces compact support of solutions given certain sufficient conditions namely that the given forcing function sufficiently decays toward -∞ and ∞. It turns out that the condition found is sharp.
A Two-Stage Vehicle Routing Algorithm Applied to Disaster Relief Logistics after the 2015 Nepal Earthquake
Published electronically March 15, 2018
Author: Stephanie Allen (SUNY Geneseo)
Sponsor: Caroline Haddad (SUNY Geneseo)
Abstract: After the April 2015 Nepal Earthquake, the Himalayan Disaster Relief Volunteer Group distributed supplies to affected areas. We model the organization’s delivery of supplies as a vehicle routing problem using Fisher and Jaikumar’s two-stage method, which allocates locations to vehicles via an integer program and then uses heuristics to route the vehicles. In the allocation stage, we use the assignment problem formulation to assign locations to vehicles. In the routing stage, we implement the greedy, sub-tour reversal, and simulated annealing heuristics for the sake of comparison. Our results illustrate the necessity of heuristics and the power that comes from using multiple heuristics for the vehicle routing problem.
Calibration of the Ross Recovery Theorem to Real-world Data, and Tests of its Practical Value
Published electronically March 19, 2018
Authors: Ling Lan and Zhengxu Li (Courant Institute of Mathematical Sciences)
Sponsor: Robert Kohn (Courant Institute of Mathematical Sciences)
Abstract: The Ross Recovery Theorem  challenges the commonly held thought that derivative prices do not contain useful predictive information about the distribution of financial variables (such as an equity index). The theorem recovers, under certain hypotheses on the character of the market, the subjective probability distribution of an equity index from current derivative prices. In this paper, building on the method of Backwell  for extracting state prices from option prices, we develop a strategy for combining option data with the Recovery Theorem to estimate the subjective distribution. Using real-world data, we then investigate whether the Recovery Theorem yields predictive information and has practical value, concluding that the answer might be no.
Published electronically March 23, 2018
Authors: Irena Chen, Jessica Fay, and Melissa Stadt (University of Washington)
Sponsor: Sara Billey (University of Washington)
Abstract: United Way of King County's Free Tax Campaign offers free tax return preparation for low-income residents in the greater Seattle area. For many years, United Way has been using the same work ow model, but is now considering switching to a new strategy that could potentially serve more clients. In order to evaluate the which work flow would be more effective, we created Monte Carlo simulations to predict the maximum number of clients a given tax site could serve, given varying numbers of personnel and staff. By analyzing results from both simulations, we concluded it would be beneficial for United Way to change to the proposed new service model. As the Free Tax Campaign grows, the new service model will continue to outperform the previous model. These simulations can also be used to assist in allocating staff across tax sites, in order to maximize the number of people the Free Tax Campaign can serve.
Published electronically March 28, 2018
Author: Zane Huttinga (Montana State University)
Sponsor: Bree Cummins and Tomáš Gedeon (Montana State University)
Abstract: We present a novel combinatorial problem that arises from mathematical biology. In order to understand the dynamics of models of gene regulatory networks over a parameter space, a problem of constructing linear extensions of a partial order with algebraic constraints arises naturally. We formulate the problem for a class of algebraic constraints related to the form of nonlinearities in the gene regulation model. We provide an algorithm that partially solves the problem. We formulate a conjecture on the special role of additive constraints in the class of all considered constraints. We present several examples where we show that the number of solutions is much smaller than the number of unconstrained linear extensions.
Published electronically May 2, 2018
Authors: Lily Wittle (University of Miami) and Theresa Thimons (Saint Vincent College)
Sponsor: Comlan de Souza (California State University, Fresno)
Abstract: Phase information of light and electron waves cannot be physically measured, but it can be estimated. Physicists R. W. Gerchberg and W. O. Saxton wrote an algorithm that estimates the phases of these waves when only their amplitudes are known. We sought to improve this algorithm. We observed the algorithm's performance by experimentation using a numerical implementation of the algorithm that we wrote in MATLAB. We found that functions of the form f x g, where g is a Gaussian function, have better success than those of the corresponding f. We also found that a constant initial phase estimate produces more consistent and efficient results for non-centrosymmetric input than the random initial phase estimate used in the original algorithm. Our paper includes a proof of error convergence, description of the implementation of a low-frequency filter, and full MATLAB scripts.