Bulletin of the American Physical Society
APS March Meeting 2020
Volume 65, Number 1
Monday–Friday, March 2–6, 2020; Denver, Colorado
Session G34: Machine Learning and Data in Polymer Physics IIFocus
|
Hide Abstracts |
Sponsoring Units: DPOLY DBIO DCOMP Chair: Tyler Martin, National Institute of Standards and Technology Room: 506 |
Tuesday, March 3, 2020 11:15AM - 11:51AM |
G34.00001: Machine Learning and Data in Polymer Physics Research - Interpretation of Experiments, Model Development, and Enhanced Sampling Invited Speaker: Juan De Pablo Advances in molecular modeling algorithms, optimization strategies, and machine learning techniques, are ushering a new era of materials science and engineering in which computational tools are routinely used to probe, design, and interrogate matter and functional materials systems. The way in which problems and questions are formulated is rapidly changing, and it is important to rethink the role of research scientists and engineers in the context of these advances. In this presentation I will illustrate some of these ideas by relying on a variety of examples taken from polymer physics. In the first, I will discuss the coupling of experiments and molecular models, and how that coupling can be used to extract additional information from experiments that would otherwise be difficult to generate. In the second I will present models of biological systems – DNA and chromatin - that use machine learning to integrate experimental and computational information form a wide range of sources, and explain how the resulting information can be used to address important questions in epigenetics. In the third, I will discuss how machine learning can be used to design polymer structures or architectures for specific target properties. |
Tuesday, March 3, 2020 11:51AM - 12:03PM |
G34.00002: Neural Network Accelerated Self-Consistent Field Theory Hejin Huang, Karim Gadelrab, Alfredo Alexander-Katz Self-consistent field theory (SCFT) has demonstrated excellent capabilities in predicting self-assembled structures of block copolymers (BCP) at equilibrium. This method has been widely implemented in BCP self-assembly to understand its assembly process. Although SCFT gives accurate results, it is a relatively time-consuming method: SCFT involves solving differential equation numerically in matrix form at each time step and normally takes thousands of steps to reach the final structure at equilibrium, which makes it difficult to be applied to large 3D systems. In this work, we train a neural network (NN) to predict the evolving field during SCFT free energy minimization. After training the NN, we implement a hybrid algorithm combining SCFT with the trained NN. This NN-SCFT model helps to shorten the SCFT simulation time significantly (approximately x10 speedup), with convergence being achieved in all different cases (different volume ratio, and the chemical incompatibility between blocks). The NN-SCFT hybrid system, thus, provides a powerful tool for further exploration of larger BCP directed self-assembly systems and inverse self-assembly. |
Tuesday, March 3, 2020 12:03PM - 12:15PM |
G34.00003: Neural network for phase diagrams of polymer-containing liquid mixtures Issei Nakamura We develop deep neural networks (DNNs) that consider the phase separation of polymeric liquids. In this talk, we discuss our new hidden layer that is constructed through coarse-grained mean-field theory and the scaling laws in polymer physics. This characteristic hidden layer enables us to perform the learning process efficiently with relatively small numbers of artificial neurons and hidden layers and provides the DNNs with reasonable predictive power. To demonstrate the efficacy of our DNNs, we will discuss the phase diagrams of polymer solutions, and the salt-free and salt-doped diblock copolymer melts. Moreover, we will show the predictive power of the DNNs by considering some experiments for the lithium salt-doped diblock copolymers such as PEO-b-PS.<div class="grammarly-disable-indicator"> </div> |
Tuesday, March 3, 2020 12:15PM - 12:27PM |
G34.00004: Predicting the glass transition behaviors of polymers via integration of molecular simulations, theory, and machine learning Wenjie Xia, Amirhadi Alesadi Understanding and predicting the glass transition behavior of glass-forming polymers are of critical importance from both physical and practical standpoints. The substantial change in polymer relaxation and dynamics upon cooling causes major change in most physical properties, including mechanical modulus, density, specific heat, damping characteristics, dielectric properties of a polymer. The cheminformatics-based approach based on machine learning (ML) algorithms is often applied to draw the quantitative relationships between key molecular parameters/descriptors and properties of polymers. In this work, we develop an innovative framework by integrating cheminformatics and coarse-grained molecular dynamics (MD) simulations to predict the glass transition temperature of diverse sets of hundred polymers. Moreover, the use of generalized entropy theory in conjunction with ML uncovers the critical roles of key molecular parameters (i.e., cohesive interactions, chain stiffness, and branching) in influencing the glass transition temperature as well as other characteristic temperatures associated with glass formation of polymers. |
Tuesday, March 3, 2020 12:27PM - 12:39PM |
G34.00005: Extracting molecular mechanisms of shear-thinning of liquids at high strain rates using machine learning Vikram Jadhao, JCS Kadupitiya Recent nonequilibrium molecular dynamics simulations have shown that shear-thinning of molecular liquids such as squalane at high strain rates (over 10^5 per second) exhibit a transition from a low-Newtonian-viscosity regime described well by power-law models to a high-Newtonian-viscosity regime where the flow properties are consistent with thermally-activated flow models. This talk explores the use of machine learning to probe the molecular origins of this rheological transition. Molecular trajectory data from simulations of small-molecular liquids sheared over a broad range of pressures and rates are used to design a 3D feature matrix. Using this matrix as input, several linear and nonlinear dimension reduction techniques are used to reduce the dimensionality of the feature space. We find that t-distributed stochastic neighbor embedding (t-SNE) can rapidly and effectively cluster trajectory data enabling the identification of molecular features (sets of atom pairs) diagnostic of the evolution in molecular order. Subsequent calculations of the order parameter and its linking with macroscopic rheological properties enable the determination of the amount of shear-thinning that comes from the evolution in order. |
Tuesday, March 3, 2020 12:39PM - 12:51PM |
G34.00006: Hybrid machine learning/materials science modeling for semi-crystalline polymer during film fabrication process Jian Yang, Teresa Karjala, Jonathan Mendenhall, Valeriy Ginzburg, Rajen Patel, Fawzi Hamad, Elva Lugo, Pavan Valavala For semi-crystalline polymer like polyethylene (PE), it is well known that PE film physical properties is heavily dependent on the morphology of both the crystalline phase and amorphous chains, which can be largely influence by the film processing conditions. A clear understanding of the relationships of polymer molecular fingerprint, formulation, fabrication conditions and physical properties is important for future materials design, which can be traced back to polymerization process. However, this is generally considered to be a very complicated problem due to the large parameter space. In this report, we developed a new hybrid approach to combine the power of machine learning and fundamental materials science to characterize semi-crystalline PE, develop structure-property relationship and study the effect of fabrication conditions on physical properties during blown film fabrication process and to inform the design of new polymer structures. |
Tuesday, March 3, 2020 12:51PM - 1:03PM |
G34.00007: Developing Databases for Polymer Informatics Roselyne Tchoua, Zhi Hong, Debra Audus, Shrayesh Patel, Logan Ward, Kyle Chard, Juan De Pablo, Ian Foster One significant barrier to the adoption of polymer informatics is a lack of large FAIR (Findable, Accessible, Interoperable, Reusable) databases. In an effort to overcome this barrier, we developed pipelines to harness the vast quantities of valuable experimental polymer data trapped in the literature. In our first effort, we developed the largest Flory-Huggins chi parameter database using crowdsourcing and found that the burden to review papers could be lessened by training a classifier to identify promising articles. To further reduce human input, we turned to natural language processing software coupled with specially designed software modules to extract grass transition temperatures with minimal human input; ultimately, we extracted over 250 glass transition temperatures. All of the resulting data is freely available at the Polymer Property Predictor and Database website (http://pppdb.uchicago.edu). During this process, we found that identification of the polymer names within the literature was a key problem as polymers are referred to by common names, sample names, labels, etc. and subsequently explored named entity recognition to tackle this problem. To further extend our databases, we are working on allowing them to accept user submitted data. |
Tuesday, March 3, 2020 1:03PM - 1:39PM |
G34.00008: Data Science and Machine Learning for polymer films and beyond Invited Speaker: Daniela Ushizima As a powerful example of how machine learning (ML) algorithms can streamline discovery from experimental data, scientists at the LBNL Advanced Light Source have employed Convolutional Neural Networks (CNN) [1, 2] to enable lattice structure classification using diffraction patterns, and Gaussian process regression to construct surrogate models and error functions based on the limited experimental data. Diffraction patterns have come from Grazing Incidence Small Angle X-ray Scattering (GISAXS), a surface sensitive technique with increasingly usage growth in probing complex morphologies, such as conductive polymers. GISAXS allows electron density correlation analyses at surfaces by combining features from small-angle X-ray scattering and diffuse X-ray reflectivity. Resulting scattering patterns work as signatures, which depend on the size, shape, and arrangement of the nano-structured components. We will illustrate some of the advantages of using ML methods over traditional ways of searching for configurations in large materials databases. We will also discuss scaling analysis to high-throughput data to enable quick selection of materials, and benefits of autonomous experiments [3], faster experimental sessions and accelerated scientific discovery from materials science samples. |
Tuesday, March 3, 2020 1:39PM - 1:51PM |
G34.00009: Parameter Estimation for Spatio-Temporal Models using Bayesian Optimisation and Gaussian Processes Nigel Clarke, Joao Cabral, Richard Wilkinson, Wil Ward, Sebastian Pont With many physical models having only numerical solutions, it can be challenging to fit predictions to data. We show how accurate joint estimation of parameters can be achieved efficiently with Bayesian optimization and Gaussian processes, even for spatio-temporal models, using the Cahn-Hilliard equation of phase separation as an exemplar. |
Tuesday, March 3, 2020 1:51PM - 2:03PM |
G34.00010: Evolutionary couplings detect side-chain interactions in protein structures Adam J. Hockenberry, Claus Wilke Patterns of amino acid covariation in large protein sequence alignments can inform the prediction of de novo protein structures, binding interfaces, and mutational effects. While algorithms that detect these so-called evolutionary couplings between residues have proven useful for practical applications, less is known about how and why these methods perform so well, and what insights into biological processes can be gained from their application. Here, we show that evolutionary coupling analyses are significantly more likely to identify structural contacts between side-chain atoms than between backbone atoms. We use both simulations and empirical analyses to highlight that purely backbone-based definitions of true residue–residue contacts may underestimate the accuracy of evolutionary coupling algorithms by as much as 40% and that a commonly used reference point (Cβ atoms) underestimates the accuracy by 10–15%. These findings show that co-evolutionary outcomes differ according to which atoms participate in residue–residue interactions and suggest that accounting for different interaction types may lead to further improvements to contact-prediction methods. |
Tuesday, March 3, 2020 2:03PM - 2:15PM |
G34.00011: Tracking Accelerated Aging of Cross-Linked Polyethylene Pipes by Applying Machine Learning Concepts to Infrared Spectra Melanie Hiles, Joseph D'Amico, Benjamin Morling, Fatemeh Abbasi, Michael Grossutti, John Dutcher Cross-linked polyethylene (PEX) pipes are promising replacements for metal or concrete pipes used for water, gas and sewage transport. Characterizing changes to the polymer and additive compounds with in-service use is paramount to predicting pipe failure. Infrared (IR) microscopy combines the chemical specificity of IR spectroscopy with the high spatial resolution of light microscopy, and we have used this technique to track variations in the degree of crystallinity and additive concentration across the wall thickness of PEX pipes. We have shown that principal component analysis of IR absorbance peaks can be used to differentiate and classify different pipe formulations [1]. We have used this methodology to characterize changes to pipes that have been subjected to accelerated aging involving heating in water and air, and exposure to ultraviolet radiation. This has allowed us to identify and track IR peaks that are most relevant to pipe degradation. We have used these results, together with machine learning techniques, to identify and classify different modes of pipe degradation. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700