Bulletin of the American Physical Society
APS March Meeting 2019
Volume 64, Number 2
Monday–Friday, March 4–8, 2019; Boston, Massachusetts
Session F18: Machine Learning Quantum States IIFocus

Hide Abstracts 
Sponsoring Units: DCOMP DCMP DAMOP Chair: Yizhuang You, Harvard University Room: BCEC 156B 
Tuesday, March 5, 2019 11:15AM  11:51AM 
F18.00001: Machine Learning Physics: From Quantum Mechanics to Holographic Geometry Invited Speaker: Yizhuang You Inspired by the "third wave" of artificial intelligence (AI), machine learning has found rapid applications in various topics of physics research. Perhaps one of the most ambitious goals of machine learning physics is to develop novel approaches that ultimately allows AI to discover new concepts and governing equations of physics from experimental observations. In this talk, I will present our progress in applying machine learning technique to reveal the quantum wave function of BoseEinstein condensate (BEC) and the holographic geometry of conformal field theories. In the first part, we apply machine translation to learn the mapping between potential and density profiles of BEC and show how the concept of quantum wave function can emerge in the latent space of the translator and how the Schrodinger equation is formulated as a recurrent neural network. In the second part, we design a generative model to learn the field theory configuration of the XY model and show how the machine can identify the holographic bulk degrees of freedom and use them to probe the emergent holographic geometry. 
Tuesday, March 5, 2019 11:51AM  12:03PM 
F18.00002: Comparing deep reinforcementlearning techniques: applications to quantum memory Petru Tighineanu, Thomas Foesel, Talitha Weiss, Florian Marquardt In our recent work [1] we showed how reinforcement learning with artificial neural networks (ANNs) can be a powerful tool to discover quantumerrorcorrection strategies fully adapted to the quantum hardware of a quantum memory. We employed a reinforcementlearning technique called natural policy gradient, in which the policy of the ANN is updated and improved according to the secondorder gradient of the return (the cumulative sum of the reward) in the parameter space of the ANN. 
Tuesday, March 5, 2019 12:03PM  12:15PM 
F18.00003: Structural Predictors for Machine Learning Modeling of Superconductivity in Ironbased Materials Valentin Stanev, Jack Flowers, Ichiro Takeuchi Superconductivity in ironbased materials continues to be a focus of intense research effort a decade after its discovery. In particular, the interplay between structure and charge doping as drivers of superconductivity is still a matter of active debate. To address this question we use Machine Learning (ML) approach. Based on published data we created a database covering the available structural information of the 122 family of materials. Using the lattice parameters, pnictogen height and charge doping we trained several ML models designed to predict the critical temperature T_{c} over the entire parameter space. The analysis of the models suggests that no single variable can fully explain and predict the evolution of T_{c}, and a combination of at least two parameters are needed. The ML predictions can be used to guide further exploration of the phase diagram of ironbased superconductors. 
Tuesday, March 5, 2019 12:15PM  12:27PM 
F18.00004: Predicting physical properties of alkanes with neural networks Pavao Santak, Gareth Conduit The physical properties of many alkanes are unknown, which prevents engineers from optimally deploying them in base oil lubricants. In order to address this issue, we train neural networks that can work with fragmented data, enabling us to exploit the propertyproperty correlations and increase the accuracy of our models. We encode molecular structure into five nonnegative integers, which enables us to exploit the structureproperty correlations. We establish correlations between branching and the boiling point, heat capacity and vapor pressure as a function of temperature. Furthermore, we explore the connection between the symmetry and the melting point and identify erroneous data entries in the flash point of linear alkanes. Finally, we predict linear alkanes’ kinematic viscosity by exploiting the temperature and pressure dependence of shear viscosity and density. 
Tuesday, March 5, 2019 12:27PM  12:39PM 
F18.00005: Understanding Magnetic Properties of UraniumBased Binary Compounds from Machine Learning Ayana Ghosh, Serge M Nakhmanson, JianXin Zhu Actinide and lanthanidebased materials constitute an important playground for exploring exotic properties stemming from the presence of itinerant or localized felectrons. For example, uraniumbased compounds have exhibited emergent phenomena, including magnetism, unconventional superconductivity, hidden order, and heavy fermion behavior. Among them, the magnetic properties with varying ordered states and moment size are sensitively dependent on pressure, chemical doping and magnetic field due to the strongcorrelation effects on 5felectrons. So far, there have even been no reports of systematic studies to map out connections between structures and properties of these compounds. In order to investigate such links and bridge between theoreotical and experimental learnings, we have constructed two databases combining results of highthroughput DFT simulations and experimental measurements, respectively, for a family of uraniumbased binary compounds. We then utilized different machine learning models to identify related accessible attributes (features) and predict magnetic moments and ordering in these compounds. 
Tuesday, March 5, 2019 12:39PM  12:51PM 
F18.00006: Machine learningassisted search for high performance plasmonic metals Ethan Shapera, Andre Schleife Plasmonics aims to manipulate light through choice of materials and nanoscale structure. Finding materials which exhibit lowloss responses to applied optical fields while remaining feasible for widespread use is an outstanding challenge. Online databases compiled computational data for numerous properties of tens of thousands of materials. Due to the large number of materials and high computational cost, it is not viable to compute optical properties for all materials from first principles. Geometrydependent plasmonic quality factors for a training set of ~1,000 materials are computed using density functional theory and the Drude model. We train then apply randomforest regressors to rapidly screen Materials Project to identify potential new plasmonic metals. Descriptors are limited to symmetry and quantities obtained using the chemical formula and the Mendeleev database. The machine learning models filter through 7,445 metals in Materials Project. We iteratively improve the model with active learning by adding DFT results for predicted high quality factor metals into the training set. From this we predict Cu_{3}Au, MgAg_{3}, and CaN_{2} as candidates and verify their quality factors with DFT. 
Tuesday, March 5, 2019 12:51PM  1:03PM 
F18.00007: Machine Learning and Crystal Structure Prediction of Molecular Crystals Ruggero Lot, Franco Pellegrini, Yusuf Shaidu, Emine Kucukbenli There is a natural synergy between datahungry machine learning methods and crystal structure prediction of molecular crystals that requires a careful search in a vast potential energy landscape. In our previous study we demonstrated how taking advantage of machine learning methods can enable novel predictions even for wellstudied molecular crystals [1]. In order to leverage this synergy further, we have been developing a deep neural network training tool, PANNA (Potentials from Artificial Neural Network Architectures), based on TensorFlow framework [2]. In creating transferable machinelearned potentials, the key step is the nonlinear process of training the network model. We will demonstrate a variety of network training techniques that can be explored within PANNA, from ones that are commonly used in machine learning community to the ones that are specific to atomistic simulations. We will report the effect of data selection, input representation and training methods on the training dynamics and on the resulting potentials in the difficult case of molecular crystals. 
Tuesday, March 5, 2019 1:03PM  1:15PM 
F18.00008: Fitting effective models using QMC parameter derivatives William Wheeler, Shivesh Pathak, Lucas Wagner Effective models are fundamental to our understanding of complex materials, but selecting the right model and parameters to accurately describe a particular material can be challenging. The recently developed density matrix downfolding method (DMD) [1] uses an ensemble of quantum Monte Carlo calculations to both select and fit parameters to effective models for materials. However, this method is computationally extremely demanding. In a similar spirit to force matching in classical model fitting [2], we improve the efficiency of DMD by computing derivatives of the energy and density matrix with respect to parameters of each trial wave function. We demonstrate the new technique by computing a band structure for silicon using first principles quantum Monte Carlo. 
Tuesday, March 5, 2019 1:15PM  1:27PM 
F18.00009: Detection of Phase Transitions in Quantum Spin Chains via Unsupervised Machine Learning Yutaka Akagi, Nobuyuki Yoshioka, Hosho Katsura In the field of machine learning, there has been an important breakthrough in recent years. What was at the center of it is the deep learning by artificial neural networks mimicking human brains. By deepening a process part which repeats extracting feature quantities through nonlinear transformations, socalled hidden layer, data/class separability and the discriminability have dramatically improved. Recently, the machine learning techniques have found a wide variety of applications in condensed matter and statistical physics. Examples include detection of phase transition and acceleration of Monte Carlo simulation. Meanwhile, most of these studies are based on supervised learning under wellknown results, and there are only a few previous examples applying unsupervised learning. In this research, we investigate quantum phase transitions in various quantum spin chains by using an autoencoder which is one of unsupervised learning methods. In particular, we will show that the autoencoder whose input is only shortrange correlators is capable of detecting even topological phase transition from the largeD phase to the Haldane phase without using either topological invariants or entanglement spectra. 
Tuesday, March 5, 2019 1:27PM  1:39PM 
F18.00010: Supervised learning of phase transitions in classical and quantum systems Nicholas Walker, KaMing Tam, Mark Jarrell Supervised machine learning methods are used to identify transitions in physical systems using the classical solidliquid transition of a LennardJones system as well as the strong couplinglocal moment quantum transition in the softgap Anderson model as examples. Monte Carlo sampling was used to achieve a uniform sampling of configurational data across a large range of the relevant transition parameter for each system. Hyperbolic feature scaling is applied to the features followed by training a 1dimensional convolutional neural network with the samples corresponding to the extreme parameters of each phase as training data. The rest of the configurational data is assigned phase classification probabilities by the neural network, allowing for the prediction of the transition point with respect to the chosen varied parameter. This is done by fitting the mean classification probabilities for each set of configurational data with respect to the varied parameter with a logistic function and taking the transition to be at the value of the parameter corresponding to the midpoint of the sigmoid. The results obtained are comparable to results using contemporary methods for each system. 
Follow Us 
Engage
Become an APS Member 
My APS
Renew Membership 
Information for 
About APSThe American Physical Society (APS) is a nonprofit membership organization working to advance the knowledge of physics. 
© 2022 American Physical Society
 All rights reserved  Terms of Use
 Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 207403844
(301) 2093200
Editorial Office
1 Research Road, Ridge, NY 119612701
(631) 5914000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 200452001
(202) 6628700