Showing papers by "Bobby G. Sumpter published in 2022"

Learning in continuous action space for developing high dimensional potential energy models

Abstract: Reinforcement learning (RL) approaches that combine a tree search with deep learning have found remarkable success in searching exorbitantly large, albeit discrete action spaces, as in chess, Shogi and Go. Many real-world materials discovery and design applications, however, involve multi-dimensional search problems and learning domains that have continuous action spaces. Exploring high-dimensional potential energy models of materials is an example. Traditionally, these searches are time consuming (often several years for a single bulk system) and driven by human intuition and/or expertise and more recently by global/local optimization searches that have issues with convergence and/or do not scale well with the search dimensionality. Here, in a departure from discrete action and other gradient-based approaches, we introduce a RL strategy based on decision trees that incorporates modified rewards for improved exploration, efficient sampling during playouts and a "window scaling scheme" for enhanced exploitation, to enable efficient and scalable search for continuous action space problems. Using high-dimensional artificial landscapes and control RL problems, we successfully benchmark our approach against popular global optimization schemes and state of the art policy gradient methods, respectively. We demonstrate its efficacy to parameterize potential models (physics based and high-dimensional neural networks) for 54 different elemental systems across the periodic table as well as alloys. We analyze error trends across different elements in the latent space and trace their origin to elemental structural diversity and the smoothness of the element energy surface. Broadly, our RL strategy will be applicable to many other physical science problems involving search over continuous action spaces.

Deep Generative Models for Materials Discovery and Machine Learning-Accelerated Innovation

Abstract: Machine learning and artificial intelligence (AI/ML) methods are beginning to have significant impact in chemistry and condensed matter physics. For example, deep learning methods have demonstrated new capabilities for high-throughput virtual screening, and global optimization approaches for inverse design of materials. Recently, a relatively new branch of AI/ML, deep generative models (GMs), provide additional promise as they encode material structure and/or properties into a latent space, and through exploration and manipulation of the latent space can generate new materials. These approaches learn representations of a material structure and its corresponding chemistry or physics to accelerate materials discovery, which differs from traditional AI/ML methods that use statistical and combinatorial screening of existing materials via distinct structure-property relationships. However, application of GMs to inorganic materials has been notably harder than organic molecules because inorganic structure is often more complex to encode. In this work we review recent innovations that have enabled GMs to accelerate inorganic materials discovery. We focus on different representations of material structure, their impact on inverse design strategies using variational autoencoders or generative adversarial networks, and highlight the potential of these approaches for discovering materials with targeted properties needed for technological innovation.

Learning in continuous action space for developing high dimensional potential energy models

Abstract: Reinforcement learning (RL) approaches that combine a tree search with deep learning have found remarkable success in searching exorbitantly large, albeit discrete action spaces, as in chess, Shogi and Go. Many real-world materials discovery and design applications, however, involve multi-dimensional search problems and learning domains that have continuous action spaces. Exploring high-dimensional potential energy models of materials is an example. Traditionally, these searches are time consuming (often several years for a single bulk system) and driven by human intuition and/or expertise and more recently by global/local optimization searches that have issues with convergence and/or do not scale well with the search dimensionality. Here, in a departure from discrete action and other gradient-based approaches, we introduce a RL strategy based on decision trees that incorporates modified rewards for improved exploration, efficient sampling during playouts and a "window scaling scheme" for enhanced exploitation, to enable efficient and scalable search for continuous action space problems. Using high-dimensional artificial landscapes and control RL problems, we successfully benchmark our approach against popular global optimization schemes and state of the art policy gradient methods, respectively. We demonstrate its efficacy to parameterize potential models (physics based and high-dimensional neural networks) for 54 different elemental systems across the periodic table as well as alloys. We analyze error trends across different elements in the latent space and trace their origin to elemental structural diversity and the smoothness of the element energy surface. Broadly, our RL strategy will be applicable to many other physical science problems involving search over continuous action spaces.

CO2-Assisted Oxidative Dehydrogenation of Propane over VOx/In2O3 Catalysts: Interplay between Redox Property and Acid–Base Interactions

Abstract: In this work, a series of VOx-loaded In2O3 catalysts were prepared, and their catalytic performance was evaluated for CO2-assisted oxidative dehydrogenation of propane (CO2-ODHP) and compared with In2O3 alone. The optimal composition is obtained on 3.4V/In2O3 (surface V density of 3.4V nm–2), which exhibited not only a higher C3H6 selectivity than other V/In catalysts and In2O3 under isoconversion conditions but also an improved reaction stability. To elucidate the catalyst structure–activity relationship, the VOx/In2O3 catalysts were characterized by chemisorption [NH3-temperature-programmed desorption (TPD), NH3-diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS), CO2-TPD, and CO2-DRIFTS], H2-temperature-programmed reduction (TPR), in situ Raman spectroscopy, UV–vis diffuse reflectance spectroscopy, near-ambient pressure X-ray photoelectron spectroscopy, X-ray absorption spectroscopy, and further examined using density functional theory. The In–O–V structure and the extent of oligomerization, which play a crucial role in improving selectivity and stability, were identified in the VOx/In2O3 catalysts. In particular, the presence of surface VOx (i) inhibits the deep reduction of In2O3, thereby preserving the activity, (ii) neutralizes the excess basicity on In2O3, thus suppressing propane dry reforming and achieving a higher propylene selectivity, and (iii) introduces additional redox sites that participate in the dehydrogenation reaction by utilizing CO2 as a soft oxidant. The present work provides insights into developing selective, stable, and robust metal-oxide catalysts for CO2-ODHP by controlling the conversion of reagents via desired pathways through the interplay between acid–base interactions and redox properties.

Quantum theory of electronic excitation and sputtering by transmission electron microscopy.

Abstract: Many computational models have been developed to predict the rates of atomic displacements in two-dimensional (2D) materials under electron beam irradiation. However, these models often drastically underestimate the displacement rates in 2D insulators, in which beam-induced electronic excitations can reduce the binding energies of the irradiated atoms. This bond softening leads to a qualitative disagreement between theory and experiment, in that substantial sputtering is experimentally observed at beam energies deemed far too small to drive atomic dislocation by many current models. To address these theoretical shortcomings, this paper develops a first-principles method to calculate the probability of beam-induced electronic excitations by coupling quantum electrodynamics (QED) scattering amplitudes to density functional theory (DFT) single-particle orbitals. The presented theory then explicitly considers the effect of these electronic excitations on the sputtering cross section. Applying this method to 2D hexagonal BN and MoS2 significantly increases their calculated sputtering cross sections and correctly yields appreciable sputtering rates at beam energies previously predicted to leave the crystals intact. The proposed QED-DFT approach can be easily extended to describe a rich variety of beam-driven phenomena in any crystalline material.

A Novel Dynamic Polymer Synthesis via Chlorinated Solvent Quenched Depolymerization

Abstract: Open AccessCCS ChemistryRESEARCH ARTICLES18 Oct 2022A Novel Dynamic Polymer Synthesis via Chlorinated Solvent Quenched Depolymerization Jiadeng Zhu, Sheng Zhao, Jiancheng Luo, Wei Niu, Joshua T. Damron, Zhen Zhang, Md Anisur Rahman, Mark A. Arnould, Tomonori Saito, Rigoberto Advincula, Alexei P. Sokolov, Bobby G. Sumpter and Peng-Fei Cao Jiadeng Zhu Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Google Scholar More articles by this author , Sheng Zhao Departmnet of Chemistry, University of Tennessee, Knoxville, Tennessee 37996 Google Scholar More articles by this author , Jiancheng Luo Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Google Scholar More articles by this author , Wei Niu Department of Chemical and Biomolecular Engineering, University of Tennessee, Knoxville, Tennessee 37996 Google Scholar More articles by this author , Joshua T. Damron Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Google Scholar More articles by this author , Zhen Zhang Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Google Scholar More articles by this author , Md Anisur Rahman Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Google Scholar More articles by this author , Mark A. Arnould Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Google Scholar More articles by this author , Tomonori Saito Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Google Scholar More articles by this author , Rigoberto Advincula Department of Chemical and Biomolecular Engineering, University of Tennessee, Knoxville, Tennessee 37996 Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Google Scholar More articles by this author , Alexei P. Sokolov Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Departmnet of Chemistry, University of Tennessee, Knoxville, Tennessee 37996 Google Scholar More articles by this author , Bobby G. Sumpter Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 Google Scholar More articles by this author and Peng-Fei Cao *Corresponding author: E-mail Address: [email protected] State Key Lab of Organic-Inorganic Composites, Beijing University of Chemical Technology, Beijing 100029 Google Scholar More articles by this author https://doi.org/10.31635/ccschem.022.202202362 SectionsSupplemental MaterialAboutAbstractPDF ToolsAdd to favoritesDownload CitationsTrack Citations ShareFacebookTwitterLinked InEmail Dynamic polymers with both physical interactions and dynamic covalent bonds exhibit superior performance, but achieving such dry polymers in an efficient manner remains a challenge. Herein, we report a novel organic solvent quenched polymer synthesis using the natural molecule thioctic acid (TA), which has both a dynamic disulfide bond and carboxylic acid. The effects of the solvent type and concentration along with reaction times on the proposed reaction were thoroughly explored for polymer synthesis. Solid-state proton nuclear magnetic resonance (1H NMR) and first-principles simulations were carried out to investigate the reaction mechanism. They show that the chlorinated solvent can efficiently stabilize and mediate the depolymerization of poly(TA), which is more kinetically favorable upon lowering the temperature. Attributed to the numerous dynamic covalent disulfide bonds and noncovalent hydrogen bonds, the obtained poly(TA) shows high extensibility, self-healing, and reprocessable properties. It can also be employed as an efficient adhesive even on a Teflon surface and 3D printed using the fused deposition modeling technique. This new polymer synthesis approach of using organic solvents as catalysts along with the unique reaction mechanism provides a new pathway for efficient polymer synthesis, especially for multifunctional dynamic polymers. Download figure Download PowerPoint Introduction Dynamic polymers have been attracting significant attention due to their reprocessability and self-recovery after damage, which can prolong the lifetime of these materials.1–6 This is especially true for dynamic polymers with intrinsic self-healing ability that do not require a sequestered additive or trigger, allowing for multi-cycle healing.7–12 Generally, intrinsic self-healing polymers can be divided into three major categories based on the types of interaction: (1) physical bonds, such as ionic,13,14 metal-coordination,15,16 hydrogen bonding,17,18 host–guest,19 and van der Waals;20 (2) reversible chemical bonds, such as those formed by Diels–Alder reaction,21,22 free-radical reaction by acrylic monomers,23 disulfide exchange,24 and boronic ester bond;25,26 and (3) the combination of physical and chemical bonds.27 Due to the combined advantages from both physical and chemical interactions, the last category typically exhibits superior performance in terms of mechanical strength and extensibility, recyclability, self-healing rate, and efficiency. These advantages have recently led to intensive research; however, tedious, multistep synthetic routes under harsh conditions are generally required to create dynamic polymers with combined physical and chemical interactions. Preparation of such dynamic polymers via a low-cost, high-efficient approach from readily available raw materials is highly desired for sustainable materials development.28–33 With a carboxylic-acid tail and a dynamic covalent disulfide bond, the naturally derived organic molecule, thioctic acid (TA), is a promising candidate for the construction of robust dynamic polymers with combined physical and chemical interactions.34–41 By raising the temperature of TA above its melting point, the five-member ring containing disulfide bonds undergoes ring-opening polymerization, forming linear polymer chains with sulfur radicals at both termini.39 The terminal sulfur radicals initiate the reverse ring-closing depolymerization and revert to monomers after cooling, attesting to a more thermodynamically stable TA monomer.39 In recent years, several strategies report the synthesis of stabilized poly(TA) by quenching the terminal sulfur radicals.37–40 For example, Zhang et al.39 reacted 1,3-diisopropenylbenzene (DIB) with the terminal diradicals to form a chemically cross-linked polymer network. The molar ratio of Fe3+ to TA and reaction temperatures were tuned to enhance the Fe3+-carboxylate complexes, achieving a maximal tension strength of <60 kPa. Wang et al.40 recently prepared poly(TA)-based ionomer gel in the presence of 1-ethyl-3-methylimidazolium ethyl sulfate ([EMI][ES]), which stabilized the as-prepared poly(TA) with the aid of hydrogen bonds between [ES] and carboxylic acids. As an alternative approach, the deprotonated TA monomer, sodium thioctate (ST), was reported to self-organize and ring-opening polymerize during water evaporation process in a structurally ordered fashion. However, it exhibited a relatively low strain at low humidity (<10% elongation before breaks with relative humidity <10%).37 Very recently, the same group found that the poly(TA) could be stabilized with the addition of metal ions (i.e., Fe3+, Cu2+, etc.) as alternative ionic cross-linkers for polymer film formation in the absence of DIB, and the corresponding poly(TA) with metal ions can be recycled.38 According to these studies, extra cross-linkers (i.e., double bonds, ion-based, etc.) are normally required for dynamic elastic polymer poly(TA) synthesis, and although poly(ST) can be obtained without additional cross-linkers, they are a normally brittle, dry polymer that has lost the multifunctionalities including high stretchability and self-healing at ambient condition.42,43 Herein, we demonstrate a novel synthesis for a dynamic polymer that has the combined advantages of high stretchability, autonomous self-healing, excellent recyclability, high adhesion property, and 3D printability. Such design is inspired by some organic solvents with high chain transfer constant (e.g., chlorinated solvents) that react with and terminate the radical species.44–46 In contrast to regular reactions that require huge amounts of organic solvents (>90 wt. % solvent), a small amount of chlorinated solvent (<10 wt. %) is utilized here as a catalyst to produce dynamic polymers with a high monomer conversion. Different than previous studies on poly(TA) synthesis that use a double bond or ionic based cross-linker to interrupt the middle of the polymer chain, the chlorinated solvent herein acts as a “catalyst” that can stabilize and react with the sulfur radical terminated polymer chains at elevated temperatures, thereby preventing the kinetically favorable depolymerization of poly(TA) upon lowering the temperature. The insight of the solvent quenched depolymerization mechanism of the novel polymer synthesis approach, unraveled by solid-state NMR and first-principles simulations, may inspire other researchers to design different polymer synthetic routes via this unique reaction mechanism. Experimental Methods Synthesis of poly(TA) The TA (≥99%, Sigma-Aldrich, USA) was placed in a glass vial and melted at 120 °C in an oil bath. Then, a certain amount of dichloromethane (DCM) with a molar ratio of 0.25:1 to melted TA (anhydrous, ≥99.8%, Sigma-Aldrich, USA) was added during stirring. After 30 min reaction at 120 °C and cooling to room temperature, the obtained sample was dried in a vacuum oven. A sample without DCM was also obtained as a control following the same procedure. Meanwhile, other solvents, including MeOH, chloroform, and so on, with a molar ratio to TA of 0.25:1 were studied, and the corresponding results are summarized in Supporting Information Table S1. Characterizations Structure All NMR data were collected using a Bruker Avance III spectrometer (Bruker) operating at 400.3028 MHz 1H frequency. The monomer conversion was calculated by comparing integral ratios between vinyl peaks of poly(TA), described as follows: Monomer conversion ( % ) = [ Peak ( a ′ + c ′ ) / Peak ( a + c + a ′ + c ′ ) ] × 100 % Reaction mechanism investigation Magic angle spinning (MAS) experiments were performed in a 3.2 mm triple resonance Bruker probe. All experiments were performed at 14 kHz except for the DCM treated material, which became gel-like making the faster MAS unstable. This sample was spun at 8 kHz MAS. 1H T1 relaxation times were estimated from the null point of an inversion recovery curve and ranged from 0.75 to 11 s depending on the sample treatment. Cross polarization (CP) experiments were performed using a 10% ramp CP under Hartmann–Hahn matching conditions and a contact time of 4 ms. SPINAL-64 decoupling was used for decoupling during 13C acquisition.47 The 13C multi-CP experiments were performed using 18–20 CP blocks lasting 400–500 μs with 1–1.5 s 1H repolarization times and a relaxation delay of 1 s.48 For the DCM treated material, an excess of DCM was introduced to the NMR rotor with TA powder. The rotor was then heated in an oven to 120 °C for 30 min. Afterward, the rotor was kept in the oven at 70 °C for 1.5 days to drive off excess DCM. A control experiment where CP spectra were collected on as received TA, which was subsequently heated to 120 °C and cooled without DCM, was performed. Ab initio simulations Geometry optimizations of the TA monomer and poly(TA) were carried out in a vacuum and a model aqueous solution phase without imposing geometrical restrictions by using the NWChem suite of programs (version 7.0.2).49 All calculations were optimized using hybrid meta-functionals50 at the M06-2X/6-311++G** level. The solvent effects were accounted for by using both an explicit model (single solvent shell) and implicitly using the Solvation Model Based on Density.51 Following geometry optimization, ab initio molecular dynamics was performed at the same level of theory using a stochastic velocity rescaling thermostat to control the temperature at 400 K to examine TA polymerization or at room temperature to study depolymerization of poly(TA).52 Simulations were performed up to 1 ns and the trajectories were analyzed for geometric evolution, for example, if bonds cleaved or formed. Molecular weight The molecular weight (MW) and polydispersity (PDI) of each sample were measured with a Malvern OMNISEC GPC system (Malvern Panalytical Ltd.) equipped with OMNISEC RESOLVE and OMNISEC REVEAL (Malvern Panalytical Ltd.). The sample concentrations were 5 mg mL−1 and each was filtered through 30 mm 0.2 μm polytetrafluoroethylene (PTFE) filters prior to analysis. The analysis was carried out using two PLgel 5 μm mixed-C columns (7.5 mm ID × 300 mm) and one PLgel 5 μm guard column (7.5 mm ID × 50 mm) in series. Tetrahydrofuran (THF) was utilized as the eluent (1 mL min−1, with the entire system at 30 °C). Absolute molar masses were calculated relative to polystyrene standards in the OMNISEC software v11.21. The chromatogram of each sample was split into three regions during analysis, the main distribution at higher MW, a lower MW tail, and low mass peaks at the end of the separation. The approximate ratio in each section was calculated based on peak area. Matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectra were acquired on a Bruker autoflex speed in positive ion reflection mode. The sample solutions were prepared at 10 mg mL−1, the sodium trifluoroacetate salt (NaTFA) at 3 mg mL−1, and the matrix (Super-DHB) at 60 mg mL−1. The solutions were mixed in a 10:1:10 (matrix:salt:sample) ratio and 1 μL was deposited on the ground stainless steel target for analysis. The samples, matrix, and salt were dissolved in 50:50 THF:chloroform after repeated vortex mixing to ensure dissolution. The spectra were obtained by summing at least 5000 shots into the sum buffer prior to analysis. The mass spectra were analyzed in Bruker FlexAnalysis software. The absolute MW could not be provided due to the high PDI of the obtained sample; therefore, the lower MW species in the “ski-slope” distribution were targeted. Morphology and thermal properties The scanning electron microscopy (SEM) images were recorded in a Carl Zeiss Merlin scanning electron microscope. The energy-dispersive X-ray spectroscopy (EDS) results were obtained with a system from Bruker Nano GmbH using a XFlash detector 5030 (Bruker, Berlin, Germany). Differential scanning calorimetry (DSC) measurements were performed on a DSC2500 (TA instruments) with ca. 10 mg sample at a scan rate of 5 °C min−1. Thermogravimetric analysis (TGA) was measured on a TA instrument Q-50 TGA for thermal stability of the samples. Samples of 5 –10 mg were placed in a platinum pan. The measurements were conducted from room temperature to 800 °C with a heating rate of 10 °C min−1 rate in a N2 atmosphere. Mechanical tests Dynamic mechanical analysis (DMA) was performed on a TA Discovery DMA850 (TA Instruments, New Castle, Delaware, U.S.A.). Specimens with a dimension of 20 mm length, 10 mm width, and 1 mm thickness were used for temperature ramp measurements. The amplitude and frequency in oscillation mode were set to 20 μm and 1 Hz, respectively. The samples were ramped from −50 to 50 °C, with a ramp rate of 3.0 °C min−1. Small-amplitude oscillatory shear measurements were performed on an AR2000ex rheometer (TA Instruments, New Castle, Delaware, U.S.A.). The experiments were performed between two 4 mm parallel steel plates. The temperature was controlled by a system using nitrogen as the gas source. All the samples were loaded between the plates, equilibrated at 60 °C for 30 min before the measurement, and measured at different temperatures with the angular frequency sweep from 100 to 0.1 rad s−1. The master curve at reference temperature could be built by applying time temperature superposition. An Instron 3343 universal testing system following the ASTM D1708 standard was used to test the mechanical properties, and at least three specimens were prepared for each test. Hysteresis curves were measured from 0% to 100% strain at 0.1 mm s−1 rate. Lap shear adhesion measurements for aluminum, steel, and PTFE were measured using Instron 3343 universal testing system equipped with 1 kN cell at 2 mm min−1 crosshead speed rate. Samples were spread onto the substrates and hot-pressed at 80 °C for 30 min. The adherends were overlapped (12 × 12 mm) in a single lap-shear configuration. After cooling to room temperature, the lap shear strength was measured and average results with standard deviation of three specimens were reported. Lap shear adhesion is defined as the maximum force (in N) of the adhesive joint obtained from the lap shear test divided by the overlap area (in mm2) of adhesives. 3D printing 3D printing of the synthesized poly(TA) was conducted with a Hyrel Engine SR. A Hyrel Tambora printing head composed of heating devices and a stainless-steel cartridge was employed in the printing, where the material was melted, extruded, and deposited on the build plate. Teflon paper and plates were chosen as the build plate for easy removal of the printed parts. The original synthesized materials were loaded in the printing cartridge with 0.5 mm nozzle and heated reversely to remove the bubbles. The printing head was then heated to 110 °C for 3D printing. Printing speed was set at 15 mm s−1. Consistent ink extrusion was achieved during printing. Details of the printing parameters are shown in Supporting Information Table S2. Results and Discussion Synthesis of poly(TA) TA has a unique chemical structure containing two types of dynamic bonds: covalent disulfide bonds and noncovalent hydrogen bonds.38–40 The thermodynamic controlled ring-opening polymerization of TA (Tm = 63 °C, Supporting Information Figure S1) is initiated by thermal treatment to form a linear polymer chain with two sulfur radicals, one at each end (Scheme 1a). However, as demonstrated by the digital images (Scheme 1b) and proton nuclear magnetic resonance (1H NMR, Figure 1b), upon cooling the reaction mixture, the semicrystalline TA monomer is reformed spontaneously due to the reverse ring-closing depolymerization process initiated by the terminal radicals.39 This is consistent with our ab initio quantum calculations showing the dissociation of disulfide bond in the sulfur radical-terminated poly(TA) in the absence of a chlorinated solvent, resulting in the reformation of TA monomers upon cooling. Scheme 1 | Illustration (a) and digital images (b) of reversible reaction between TA)monomer and poly(TA) upon heating and cooling. Illustration (c) and digital images (d) of DCM)solvent-enabled synthesis of poly(TA), a transparent film. Snapshots from ab initio quantum calculations results showing: bond dissociation of disulfide in poly(TA) without DCM (e); firm association of DCM with the sulfur radical end at 400 K over 1 s (f) and 5 s (g); a stable structure of poly(TA) with bonded DCM fragment (h). Download figure Download PowerPoint The key knowledge gained is that the kinetic-controlled depolymerization of poly(TA) is initiated by the terminal sulfur radicals.39 Therefore, we hypothesized that addition of small molecules that react with the terminal radicals may prevent the depolymerization of poly(TA) and afford a linear polymer even after cooling (Scheme 1c and Figure 1a). Drawing inspiration from the termination of radical polymerization, low-cost, normally unfavorable in polymer synthesis, chlorinated solvents,44,45 such as DCM, chloroform, and benzyl chloride, were with a molar ratio of solvent to TA being 0.25:1 employed to investigate their effect on stabilizing the reactive polymer chains. To better understand the effect, 1H NMR characterization of the unpurified reaction mixture was performed to calculate the monomer conversion by comparing the integral ratios of proton peaks adjacent to the disulfide bond in poly(TA) and TA, and the results are summarized in Supporting Information Table S1 and Figures S2–S11. As expected, many organic solvents can act as a “catalyst” to assist in the formation of poly(TA) after cooling the mixture as evidenced by the 1H NMR spectra analysis. For example, with DCM, the appearance of a new broad peak at ∼2.75 ppm corresponds to the polymeric structure, indicating efficient polymerization with a high monomer conversion of ∼77% (Figure 1c). As can be seen in Supporting Information Table S1, the chemical reaction with a small amount of DCM has relatively higher monomer conversion compared to other solvents, which produces a transparent polymer film. Based on the low cost and common use of DCM, dynamic polymer poly(TA) synthesized from the DCM “catalyzed” route was selected as the basic recipe for the remainder of the work. Figure 1 | (a) Schematic illustration of the reaction mechanism in the presence of DCM. 1H NMR spectra of heated and cooled TA sample without DCM (b), and with DCM (c). (d) MALDI results of the prepared poly(TA). (e) Expansion of the mass spectrum of poly(TA) from approximately 1050–1060 Da. Download figure Download PowerPoint The resultant poly(TA) film is highly transparent (Scheme 1d), whereas the sample produced under the same conditions without DCM is opaque because of the reversible reaction that forms crystalline TA monomers (Scheme 1b). The formation of poly(TA) is further confirmed by MALDI-TOF mass spectrometry (MS, Figure 1d,e). The sample generates an overall ski-slope distribution indicating a relatively broad PDI typical for free radical and condensation polymerizations. The spacing between the peaks in the mass spectrum is 206.1 Da, on average, consistent with the theoretical repeat unit mass, that is, 206.0 Da. The MALDI spectrum suggests that the products detected at low MWs are a cyclic species and/or linear chain possessing a double bond to a sulfur on one end and a proton on the other observed at 1053 Da and a second linear species 2 Da higher in mass with hydrogen on both ends (Figure 1e). The formation of these end groups can be explained by either dehydrochlorination (double bond formation) or radical loss of the original terminal groups and the substitution of hydrogen during the MALDI process. It is believed that both products are present due to the isotopic enrichment of the peak at 1055 Da over the naturally occurring isotopic envelope. It can also be seen that the low-mass oligomers have a bimodal distribution with an apex at approximately 1500 Da. The spectrum shows the disappearance of the cyclic species and the dominance of linear species (+2 Da) as mass increases. This suggests that the linear species will dominate chain structure at high MWs which is consistent with previous observations for the low MW oligomers of condensation polymers. MS/MS data seems to support this interpretation of a mixture of a cyclic and linear species and a second linear species 2 Da higher in mass. Other minor peaks in the spectrum arise from H and Na exchange on the carboxylic acid functionality of the repeat unit or matrix cluster formation. Meanwhile, gel permeation chromatography (GPC) measurements also demonstrate the formation of polymers, and the peaks at low retention volume can be attributed to the physical aggregation of polymer chains ( Supporting Information Figure S12). To optimize the reaction conditions, we explored the dynamic polymer synthesis at different reaction times and chlorinated solvent concentrations. As summarized in Figure 2a,b, the monomer conversion of poly(TA) in the presence of DCM is comparable with the different reaction times (from 30 min to 8 h, Supporting Information Figures S13–S16) and various molar ratios between DCM and TA (from 0.125 to 8, Supporting Information Figures S17–S22). Therefore, we simplified the reaction by using a short reaction time and decent “catalyst” quantity, that is, 120 °C for 30 min with a molar ratio of DCM to TA of 0.25:1 (around 9.3 wt. % DCM). The solid-state 13C NMR characterization of the as-received TA monomers (top), heated TA (middle), and heated TA with DCM (bottom) were also used to monitor the polymerization process (Figure 2c,d). No chemical shift changes were observed for the original TA and heated TA sample, indicating that the heating process without DCM does not render chemical changes in the system. The proton T1 relaxation time, however, drops from ∼11 to 1 s for heated TA, indicating decreased crystallinity of TA after the thermal process ( Supporting Information Figure S23),48 because the sample after the same heating process in the presence of DCM transforms the powder to a highly viscous liquid restricting the MAS rate of the semisolid-like sample.47 Therefore, the product spectrum generated in the presence of DCM was spun at a lower MAS rate as compared to the as-received TA (8 kHz vs 14 kHz). Nevertheless, the clear differences of 13C chemical shift observed for the DCM added TA polymerization indicates the chemical transformation. Figure 2 | Summary of the effect of (a) different reaction time and (b) DCM ratios on the monomer conversions of dynamic polymer synthesis. The inserts are the corresponding digital images. (c and d) Solid-state 13C NMR spectra of TA monomer (top), heated TA sample (middle), and heated TA with DCM (bottom). (e) Cl 2p XPS spectra of the TA monomer and prepared poly(TA). Download figure Download PowerPoint The reaction mechanism of DCM-assist poly(TA) synthesis was further investigated by physical analysis, ab initio calculations, and molecular dynamics simulations. The ring-opening polymerization of a disulfide containing five-member ring creates the reactive polymer chain as a fluidic liquid at an elevated temperature. The free-radical terminated polymer chains are stabilized and react with DCM molecules leading to a transparent yellowish poly(TA) (Scheme 1d). The presence of chlorine detected by X-ray photoelectron spectroscopy (XPS) (Figure 2e and Supporting Information Figure S24) and EDS ( Supporting Information Figure S25) confirms the existence of DCM fragments in the resultant polymers after the purification and drying process. Also, a smooth surface morphology is observed from the SEM image, and the corresponding elemental mapping indicates the homogeneity of poly(TA). However, it is difficult to determine the location of attachment of ·CH2Cl to the polymer chain from the 1H NMR spectrum since the proton signal on the DCM fragment is hard to detect. Therefore, two model molecules with a benzene ring, C6H5CH2Cl and ClCH2C6H4CH2Cl ( Supporting Information Figures S5 and S6), were also selected as the “catalyst” to perform the same reaction to help elucidate the structure of the terminated chains. The corresponding 1H NMR results of the obtained polymer indicate the presence of benzyl ring at ∼7.5 ppm after the reaction and purification, supporting the proposed mechanism. Ab initio calculations and molecular dynamics simulations were also employed to validate the proposed mechanism as illustrated in Scheme 1e–h. The results show that, without DCM, the disulfide in poly(TA) tends to dissociate to sulfur radicals, which initiates the depolymerization process. At an elevated temperature around 125 °C, DCM molecules tend to strongly associate and stabilize the sulfur radicals at the ends of polymer chains as exhibited in Scheme 1f,g. Moreover, the reactive polymer chains can also be terminated with a DCM fragment, generating a stable poly(TA) product. These experimental and computational results support the mechanism where the polymer chains with terminal sulfur radicals can react with DCM molecules, preventing the kinetically controlled depolymerization of poly(TA) after cooling to ambient temperature. Thermomechanical property evaluation The thermal properties of the prepared dynamic polymer were evaluated due to their significant influence on the polymeric materials’ processability and other physical properties. The poly(TA) exhibits good thermal stability with no significant weight loss until 190 °C as shown in the TGA curve ( Supporting Information Figu

Dynamic Aspects of Graphene Deformation and Fracture from Approximate Density Functional Theory

Abstract: Graphene is one of the most intriguing two-dimensional carbon materials. Its mechanical strength and failure are key concerns for materials engineering and applications. Despite the success of fracture mechanics, the mechanism of how pristine materials fail remains an elusive problem. While many theoretical studies based on molecular dynamics using empirical forcefields have tried to address this question, atomic-scale mechanics are not clearly understood. Especially, a widely employed bond-breaking approach based on the critical bond length has not been rigorously tested. Here, utilizing molecular dynamics simulations with density functional based tight binding, we investigate how the failure of the material initiates. The Wiberg bond order (WBO) to estimate the change of chemical bonds shows a transition from sp2 (WBO ~ 1.33) to sp3 (WBO < 1.0) during the deformation under loadings. However, it reveals that a single threshold value for either the WBO or bond length is insufficient to decide material failure. Instead, collective behaviors of the local atomic group govern the fracture initiation of pristine graphene. Our study provides dynamic mechanical responses based on quantum mechanics, which have not been captured by empirical forcefields, opening opportunities to design properties by precisely coupling the mechanics and quantum mechanics.

Understanding the Impacts of Support-Polymer Interactions on the Dynamics of Poly(ethyleneimine) Confined in Mesoporous SBA-15.

Abstract: Supported amines are a promising class of CO2 sorbents offering large uptake capacities and fast uptake rates. Among supported amines, poly(ethyleneimine) (PEI) physically impregnated in the mesopores of SBA-15 silica is widely used. Within these composite materials, the chain dynamics and morphologies of PEI strongly influence the CO2 capture performance, yet little is known about chain and macromolecule mobility in confined pores. Here, we probe the impact of the support-PEI interactions on the dynamics and structures of PEI at the support interface and the corresponding impact on CO2 uptake performance, which yields critical structure-property relationships. The pore walls of the support are grafted with organosilanes with different chemical end groups to differentiate interaction modes (spanning from strong attraction to repulsion) between the pore surface and PEI. Combinations of techniques, such as quasi-elastic neutron scattering (QENS), 1H T1-T2 relaxation correlation solid-state NMR, and molecular dynamics (MD) simulations, are used to comprehensively assess the physical properties of confined PEI. We hypothesized that PEI would have faster dynamics when subjected to less attractive or repulsive interactions. However, we discover that complex interfacial interactions resulted in complex structure-property relationships. Indeed, both the chain conformation of the surface-grafted chains and of the PEI around the surface influenced the chain mobility and CO2 uptake performance. By coupling knowledge of the dynamics and distributions of PEI with CO2 sorption performance and other characteristics, we determine that the macroscopic structures of the hybrid materials dictate the first rapid CO2 uptake, and the rate of CO2 sorption during the subsequent gradual uptake stage is determined by PEI chain motions that promote diffusive jumps of CO2 through PEI-packed domains.

A machine learning inversion scheme for determining interaction from scattering

Abstract: Abstract Small angle scattering techniques have now been routinely used to quantitatively determine the potential of mean force in colloidal suspensions. However the numerical accuracy of data interpretation is often compounded by the approximations adopted by liquid state analytical theories. To circumvent this long standing issue, here we outline a machine learning strategy for determining the effective interaction in the condensed phases of matter using scattering. Via a case study of colloidal suspensions, we show that the effective potential can be probabilistically inferred from the scattering spectra without any restriction imposed by model assumptions. Comparisons to existing parametric approaches demonstrate the superior performance of this method in accuracy, efficiency, and applicability. This method can effectively enable quantification of interaction in highly correlated systems using scattering and diffraction experiments.

A Deep-learning Model for Fast Prediction of Vacancy Formation in Diverse Materials

Abstract: The presence of point defects such as vacancies plays an important role in material design. Here, we demonstrate that a graph neural network (GNN) model trained only on perfect materials can also be used to predict vacancy formation energies ( E vac ) of defect structures without the need for additional training data. Such GNN-based predictions are considerably faster than density functional theory (DFT) calculations with reasonable accuracy and show the potential that GNNs are able to capture a functional form for energy predictions. To test this strategy, we developed a DFT dataset of 508 E vac consisting of 3D elemental solids, alloys, oxides, nitrides, and 2D monolayer materials. We analyzed and discussed the applicability of such direct and fast predictions. We applied the model to predict 192494 E vac for 55723 materials in the JARVIS-DFT database.

Fingerprinting Brownian Motions of Polymers under Flow.

Abstract: We present a quantitative approach to the self-dynamics of polymers under steady flow by employing a set of complementary reference frames and extending the spherical harmonic expansion technique to dynamic density correlations. Application of this method to nonequilibrium molecular dynamics simulations of polymer melts reveals a number of universal features. For both unentangled and entangled melts, the center-of-mass motions in the flow frame are described by superdiffusive, anisotropic Gaussian distributions, whereas the isotropic component of monomer self-dynamics in the center-of-mass frame is strongly suppressed. Spatial correlation analysis shows that the heterogeneity of monomer self-dynamics increases significantly under flow.

Strain-Induced Asymmetry and On-Site Dynamics of Silicon Defects in Graphene

Abstract: • DFT calculations predict two distinct isomers for Si defects in strained graphene. • STEM images capture transitions between distinct Si structures with the same bonding. • Anisotropic strain can be generated and also exposed by Si defects in graphene. In the last decade, the atomically-focused electron beams utilized in scanning transmission electron microscopes (STEMs) have been shown to induce a broad set of local structural transformations in materials, opening pathways for directing material synthesis and modification atom-by-atom. The mechanisms underlying these transformations remain largely unknown, due to the intractability of modeling the myriad of reaction pathways that can be accessed through high-energy electron scattering. The information on materials’ structure and dynamics that can be extracted from STEM images is similarly left underexplored. Here, we report the observation of anomalous on-site dynamics of individual silicon impurity atoms in graphene during STEM imaging. Density functional theory-based structural optimizations of anisotropically-strained molecular nanographenes reveal two distinct (but nearly degenerate) stable structures for four-fold coordinated silicon impurities, where interconversion between the two structures manifests slight changes of the silicon position within the lattice site. Implications for defect-based strain engineering in graphene are discussed.

Autonomous Continuous Flow Reactor Synthesis for Scalable Atom-Precision

Abstract: With new instrumentation design, robotics, and in-operando hyphenated analytical tool automation, the intelligent discovery of synthesis pathways is becoming feasible. It can potentially bridge the gap for the scale-up of new materials. We review current progress and describe a new system that uses an autonomous continuous flow chemistry framework to translate high-quality lead molecules and materials to quantities that can meet scalability demands. At the core is a continuous flow synthesis platform that can design its viable synthesis pathway to a particular molecule or material and then autonomously carry it out. This is realized by integrating: (1) A workflow/architecture for multimode chemical/materials characterization in-line. The in-line characterization modes are NMR, ESR, IR, Raman, UV-Vis, GC-MS, and HPLC, along with ex-situ modes for X-Ray and neutron scattering; (2) Integration for feedback/analysis/data storage of the control variables; (3) A core software stack that includes deep learning and reinforcement learning alongside quantum chemistry and molecular dynamics; (4) On-demand compute architectures that parse calculations to compute resources needed which include light-weight edge, mid-level edge (NVIDA DGX-2), and high-performance computing. We demonstrate preliminary results on how this autonomous reactor system can enhance our ability to deliver deuterated materials, copolymers, and site-substituted molecules.

Physics is the New Data

Abstract: The rapid development of machine learning (ML) methods has fundamentally affected numerous applications ranging from computer vision, biology, and medicine to accounting and text analytics. 1-3 Until now, it was the availability of large and often labeled data sets that enabled significant breakthroughs. However, the adoption of these methods in classical physical disciplines has been relatively slow, a tendency that can be traced to the intrinsic differences between correlative approaches of purely data-based ML and the causal hypothesis-driven nature of physical sciences. Furthermore, anomalous behaviors of classical ML necessitate addressing issues such as explainability and fairness of ML. We also note the sequence in which deep learning became mainstream in different scientific disciplines – starting from medicine and biology and then towards theoretical chemistry, and only after that, physics – is rooted in the progressively more complex level of descriptors, constraints, and causal structures available for incorporation in ML architectures. Here we put forth that over the next decade, physics will become a new data, and this will continue the transition from dot-coms and scientific computing concepts of 90ies to big data of 2000-2010 to deep learning of 2010-2020 to physics-enabled scientific ML. Neural networks and machine learning (ML) have since the early work on perceptrons in the 1950s. The introduction of the backpropagation algorithm in the 1980s laid the foundation for modern ML. However, the computational capabilities and lack of large labeled data sets limited the community to relatively shallow networks in applications such as molecular dynamics simulations, theory-experiment matching, and experimental data analysis. 4 The situation started to radically change after the when the availability of high-performance computing capabilities enabled large computational models and early experimental applications of neural networks were demonstrated. Similarly, the growth

Small angle scattering of diblock copolymers profiled by machine learning.

Abstract: We outline a machine learning strategy for quantitively determining the conformation of AB-type diblock copolymers with excluded volume effects using small angle scattering. Complemented by computer simulations, a correlation matrix connecting conformations of different copolymers according to their scattering features is established on the mathematical framework of a Gaussian process, a multivariate extension of the familiar univariate Gaussian distribution. We show that the relevant conformational characteristics of copolymers can be probabilistically inferred from their coherent scattering cross sections without any restriction imposed by model assumptions. This work not only facilitates the quantitative structural analysis of copolymer solutions but also provides the reliable benchmarking for the related theoretical development of scattering functions.

Decoding polymer self-dynamics using a two-step approach.

Abstract: The self-correlation function and corresponding self-intermediate scattering function in Fourier space are important quantities for describing the molecular motions of liquids. This work draws attention to a largely overlooked issue concerning the analysis of these space-time density-density correlation functions of polymers. We show that the interpretation of non-Gaussian behavior of polymers is generally complicated by intrachain averaging of distinct self-dynamics of different segments. By the very nature of the mathematics involved, the averaging process not only conceals critical dynamical information, but also contributes to the observed non-Gaussian dynamics. To fully expose this issue and provide a thorough benchmark of polymer self-dynamics, we perform analyses of coarse-grained molecular dynamics simulations of linear and ring polymer melts as well as several theoretical models using a "two-step" approach, where interchain and intrachain averagings of segmental self-dynamics are separated. While past investigations primarily focused on the average behavior, our results indicate that a more nuanced approach to polymer self-dynamics is clearly required.

Finding Features from Microscopes to Simulations Via Ensemble Learning and Atomic Manipulation

Abstract: Electron and scanning probe microscopies have become one of the primary techniques to investigate systems in the domain of physical and life sciences at the atomic to mesoscopic length scales [1-5]. From the perspective of computational simulations performed at different length scales, there has also been significant advancements with accessibility to faster, parallelizable CPU/GPU architectures. This means there is a heap of information from the simulations performed on a variety of materials, particularly investigating corresponding structures and functionalities. Hence, structural, and spectral data resulting from these experiments along with simulations create an enormous exploratory platform to utilize, design deep learning (DL), machine learning (ML) workflows for feature finding along with atomic manipulation. However, systematic studies combining all three avenues to go beyond comparing endpoint properties, performing DL/ML analyses on static data, assessing structure-property relationships are still in its infancy.

Machine Intelligence-Centered System for Automated Characterization of Functional Materials and Interfaces.

Abstract: Classic design of experiment relies on a time-intensive workflow that requires planning, data interpretation, and hypothesis building by experienced researchers. Here, we describe an integrated, machine-intelligent experimental system which enables simultaneous dynamic tests of electrical, optical, gravimetric, and viscoelastic properties of materials under a programmable dynamic environment. Specially designed software controls the experiment and performs on-the-fly extensive data analysis and dynamic modeling, real-time iterative feedback for dynamic control of experimental conditions, and rapid visualization of experimental results. The system operates with minimal human intervention and enables time-efficient characterization of complex dynamic multifunctional environmental responses of materials with simultaneous data processing and analytics. The system provides a viable platform for artificial intelligence (AI)-centered material characterization, which, when coupled with an AI-controlled synthesis system, could lead to accelerated discovery of multifunctional materials.

Assembly of polyelectrolyte star block copolymers at the oil-water interface.

Abstract: To understand and resolve adsorption, reconfiguration, and equilibrium conformations of charged star copolymers, we carried out an integrated experimental and coarse-grained molecular dynamics simulation study of the assembly process at the oil-water interface. This is important to guide development of novel surfactants or amphiphiles for chemical transformations and separations. The star block copolymer consisted of arms that are comprised of hydrophilic-hydrophobic block copolymers that are covalently tethered via the hydrophobic blocks to one point. The hydrophobic core represents polystyrene (PS) chains, while the hydrophilic corona represents quaternized poly(2-vinylpyridine) (P2VP) chains. The P2VP is modeled to become protonated when in contact with an acidic aqueous phase, thereby massively increasing the hydrophilicity of this block, and changing the nature of the star at the oil-water interface. This results in a configurational change whereby the chains comprising the hydrophilic corona are significantly stretched into the aqueous phase, while the hydrophobic core remains solubilized in the oil phase. In the simulations, we followed the kinetics of the anchoring and assembly of the star block copolymer at the interface, monitoring the lateral assembly, and the subsequent reconfiguration of the star via changes in the interfacial tension that varies as the degree-of-protonation increases. At low fractions of protonation, the arm cannot fully partition into the aqueous side of the interface and instead interacts with other arms in the oil phase forming a network near the interface. These insights were used to interpret the non-monotonic dependence of pH with the asymptotic interfacial tension from pendant drop tensiometry experiments and spectral signatures of aromatic stretches seen in vibrational sum frequency generation (SFG) spectroscopy. We describe the relationship of interfacial tension to the star assembly via the Frumkin isotherm, which phenomenologically describes anti-cooperativity in adsorbing stars to the interface due to crowding. Although our model explicitly considers long-range electrostatics, the contribution of electrostatics to interfacial tension is small and brought about by strong counterion condensation at the interface. These results provide key insights into resolving the adsorption, reconfiguration, and equilibrium conformations of charged star block copolymers as surfactants.

Extracting Inelastic Scattering Cross Sections for Finite and Aperiodic Materials from Electronic Dynamics Simulations.

Abstract: Explicit time-dependent electronic structure theory methods are increasingly prevalent in the areas of condensed matter physics and quantum chemistry, with the broad-band optical absorptivity of molecular and small condensed-phase systems nowadays routinely studied with such approaches. In this paper, it is demonstrated that electronic dynamics simulations can similarly be employed to study cross sections for the scattering-induced electronic excitations probed in nonresonant inelastic X-ray scattering and momentum-resolved electron energy loss spectroscopies. A method is put forth for evaluating the electronic dynamic structure factor, which involves the application of a momentum boost-type perturbation and transformation of the resulting reciprocal space density fluctuations into the frequency domain. Good agreement is first demonstrated between the dynamic structure factor extracted from these electronic dynamics simulations and the corresponding transition matrix elements from linear response theory. The method is then applied to some extended (quasi)one-dimensional systems, for which the wave vector becomes a good quantum number in the thermodynamic limit. Finally, the dispersion of many-body excitations in a series of hydrogen-terminated graphene flakes (and twisted bilayers thereof) is investigated to highlight the utility of the presented approach for capturing morphology-dependent effects in the inelastic scattering cross sections of nanostructured and/or noncrystalline materials.