2024

---

Friday, December 13th, Llyod Building Viz Room, 3pm

From centralized to federated learning of neural operators: Accuracy, efficiency, and reliability

Prof. Lu Lu (Yale University, US)

Lu Lu is an Assistant Professor in the Department of Statistics and Data Science at Yale University.Prior to joining Yale, he was an Assistant Professor in the Department of Chemical and Biomolecular Engineering at University of Pennsylvania from 2021 to 2023, and an Applied Mathematics Instructor in the Department of Mathematics at Massachusetts Institute of Technology from 2020 to 2021. He obtained his Ph.D. degree in Applied Mathematics at Brown University in 2020, master’s degrees in Engineering, Applied Mathematics, and Computer Science at Brown University, and bachelor’s degrees in Mechanical Engineering, Economics, and Computer Science at Tsinghua University in 2013. His current research interest lies in scientific machine learning, including theory, algorithms, software, and its applications to engineering, physical, and biological problems. His broad research interests focus on multiscale modeling and high performance computing for physical and biological systems. He has received the 2022 U.S. Department of Energy Early Career Award, and 2020 Joukowsky Family Foundation Outstanding Dissertation Award of Brown University.

 

 

 

---

Friday, November 29th, Llyod Building Viz Room, 3pm

DFT-FE: Fast and accurate finite-element based methods for density functional theory calculations in the exascale era

Dr. Phani Motamarri (Indian Institute of Science – IISc, Bangalore, India)

Kohn-Sham density functional theory (DFT) calculations have been instrumental in providing many crucial insights into materials behaviour and occupy a sizable fraction of the world’s computational resources today. However, the stringent accuracy requirements necessary to compute meaningful material properties, in conjunction with the asymptotic cubic-scaling computational complexity of the underlying eigenvalue problem, demand enormous computational resources for these calculations. Thus, these methods are routinely limited to high-throughput calculations with a maximum of a few thousand electrons. This talk discusses the recent advancements in real-space DFT calculations through DFT-FE, an open-source finite-element (FE) based DFT code that employs adaptive FE discretisation to handle norm-conserving pseudopotential and all-electron calculations while accommodating generic  boundary conditions. It integrates scalable and efficient solvers for the solution of the Kohn-Sham equations on hybrid CPU-GPU architectures significantly delaying the onset of cubic scaling computational complexity to large system sizes reaching up to 30,000 electrons. Furthermore, DFT-FE employs a configurational force approach to compute atomic forces and unit-cell stresses in a unified framework accounting for Pulay forces and stresses within the same computational framework. The talk will elaborate these recent HPC-centric methodological developments and capabilities in the DFT-FE code, the workhorse behind the ACM Gordon Bell Prize 2023. Finally, we will briefly highlight our group’s very recent efforts in developing a fast and scalable approach combining the efficiency of projector-augmented wave (PAW) formalism involving smooth electronic fields with the ability of systematically improvable higher-order finite-element basis facilitating substantial reduction in degrees of freedom to achieve significant computational gains (~8x-10x) compared to the current DFT-FE calculations for medium to large-scale material systems, with wide-ranging implications for tackling diverse scientific challenges involving large-scale material systems in energy storage, device materials, catalysis, and alloy design.

---

Friday, November 15th, Llyod Building Viz Room, 3pm

Computational tools for the ab-initio design of high-temperature alloys

Anirudh Raju Natarajan (EPFL, Switzerland)

Materials for high-temperature applications require careful engineering to achieve optimal mechanical properties and thermal stability. However, relationships between alloy chemistry, processing conditions, and material properties remain poorly understood. This talk presents recent advances in computational materials science that enable the rigorous and rational design of high-performance engineering alloys. First-principles calculations will be used with machine-learning and statistical mechanics techniques to rigorously describe the thermodynamic and kinetic properties of multicomponent refractory alloys. We will apply these techniques to develop design rules for high-temperature refractory alloys containing elements from groups 4, 5, and 6 of the periodic table.

---

Friday, November 8th, Llyod Building Viz Room, 3pm

Stretching, breaking, and dissolution of polymeric nanofibre bundles

Astrid de Wijn (Norwegian University of Science and Technology)

Bundles of polymeric materials are ubiquitous and play essential roles in
biological systems, and often display remarkable mechanical properties. With
the never-ending experimental advances in control and manipulation of molecular
properties on the nanometric level follows an increasing demand for a
theoretical description that is valid at this scale. This regime of nano-scale
bundles of small numbers of molecules has not been investigated much
theoretically; here chain–chain interactions, surface effects, entropy,
nonlinearities, and thermal fluctuations all play important roles.

I will present an exploration by molecular-dynamics simulations of single
chains and bundles under different types of external loading.  Stretching,
breaking, and rearrangements of chains are investigated, as well as their
nano-scale thermodynamics, breaking, and dissolution.

---

Friday, October 18th, Llyod Building Viz Room, 3pm

Electrically Tunable Picosecond Scale Octupole Fluctuations for Chiral Antiferromagnet-Based Probabilistic Computing

Pramey Upadhyaya (Purdue University, USA)

Magnetic octupole order in chiral antiferromagnets combines the advantages of ferro- and antiferromagnets: octupoles do not produce stray fields and can be effectively integrated with charge-based electronics due to the spin-split Fermi surface induced by octupole order. This makes octupole order a promising candidate for encoding information in next-generation spintronic devices. However, to realize this potential, it is crucial to understand how long octupole-encoded information remains correlated when coupled to thermal baths. Through analytical and numerical calculations, we uncover the mechanisms governing octupole autocorrelation times in chiral antiferromagnets. We show that the strong exchange fields in these materials enable significantly faster autocorrelation times compared to the dipole-based nanomagnets in current spintronic devices. Additionally, drawing on an analogy with current-biased Josephson junctions, we propose and demonstrate a new method to electrically control autocorrelation times via spin-orbit torques. With this understanding, we discuss how chiral antiferromagnets could be used to create advanced building blocks for probabilistic computing.

---

Friday, September 27th, Llyod Building Viz Room, 3pm

Atomistic modelling of moiré materials: from excitons to phasons

Indrajit Maity (Max Plank Institute, Hamburg – Germany)

If one places a regularly ruled transparent plastic sheet on top of another identical sheet and rotates the top one while holding the bottom fixed, a beautiful moiré pattern emerges. Since 2018, researchers have created similar moiré patterns using atomically thin 2D materials like graphene or transition-metal dichalcogenides (TMDs) by precisely controlling the rotation or twist angles between layers. These moiré materials exhibit fascinating electronic and optical properties, such as superconductivity, correlated states, and trapped excitons (electrically bound pairs of holes and electrons), all tunable with the twist angle. This has generated tremendous excitement in the physics, materials science, and chemistry communities, and moiré materials are now regarded as condensed matter quantum simulators.

First principles atomistic approaches to modelling moiré materials remain a major computational challenge due to the large unit cells of the moiré superlattices. Despite significant progress in experiments, atomistic studies are few and far between. In this talk, I will describe our efforts to enable detailed atomistic calculations for phonons, low-energy electrons, and excitons. Specifically, I will discuss the emergence of new sound-wave-like modes called phasons, their impact on localized electrons, and the emergence of trapped Wannier and charge-transfer

---

Friday, September 20th, Llyod Building Viz Room, 3pm

Machine learning for the characterisation and design of battery electrodes

Sam Cooper (Dyson School of Design Engineering, Imperial College London)

Battery companies want to know the relationship between their manufacturing parameters and the performance of the resulting cells, so that they can optimise their products for particular applications, reduce costs, and improve yield. The literature contains many examples of physics-based models of the various manufacturing processes (including mixing, coating, drying and calendaring), but these systems are hugely complex, and as a result they are expensive to simulate and hard to validate.

Recent advances in generative machine learning (ML) methods have allowed the relationship from manufacturing parameters to microstructure to be directly learned from data. In this talk I will present a modular approach to the cell optimisation cycle that makes use of these ML methods, in combination with GPU accelerated metric extraction (TauFactor 2), electrochemical cell simulation (PyBaMM), and Bayesian optimisation. In addition, I will be introducing a new kintsugi SEM imaging method for accurately observing the nanostructure of the carbon binder domain; “VoxCel” an open-source, voxel-based, GPU-accelerated, multi-physics cell simulation; ML methods for generating 3D data from 2D images, as well as, inpainting artefacts in image data; and a data fusion method for combining multi-modal datasets using GANs. Lastly, I’ll present a webapp that normalises the data obtained from testing cells in a lab for easy comparison to commercial cells: cell-normaliser

We are always looking for new collaborations and new data so please get in touch! If you’d like to use any of our suite of open-source tools, then head to our website: https://tldr-group.github.io

We’ve also just spun-out a company from Imperial, called Polaron AI, to bring these tools to market. Check out our website (www.polaron.ai) and get in touch: info@polaron.ai

• Kench, S., & Cooper, S. J. (2021). Generating three-dimensional structures from a two-dimensional slice with generative adversarial network-based dimensionality expansion. Nature Machine Intelligence, 3(4), 299-305.
• Squires, I., Dahari, A., Cooper, S. J., & Kench, S. (2023). Artefact removal from micrographs with deep learning based inpainting. Digital Discovery.
• Dahari, A., Kench, S., Squires, I., & Cooper, S. J. (2023). Fusion of complementary 2D and 3D mesostructural datasets using generative adversarial networks. Advanced Energy Materials, 13(2), 2202407.
• Finegan, D. P., Squires, I., Dahari, A., Kench, S., Jungjohann, K. L., & Cooper, S. J. (2022). Machine-Learning-Driven Advanced Characterization of Battery Electrodes. ACS Energy Letters, 7(12), 4368-4378.
• Kench, S., Squires, I., Dahari, A., & Cooper, S. J. (2022). MicroLib: A library of 3D microstructures generated from 2D micrographs using SliceGAN. Scientific Data, 9(1), 645.
• Cooper, S. J., Roberts, S. A., Liu, Z., & Winiarski, B. (2022). Kintsugi Imaging of Battery Electrodes: Distinguishing Pores from the Carbon Binder Domain using PT Deposition. Journal of the Electrochemical Society, 169(7), 070512.
• Mistry, A., Franco, A. A., Cooper, S. J., Roberts, S. A., & Viswanathan, V. (2021). How machine learning will revolutionize electrochemical sciences. ACS energy letters, 6(4), 1422-1431.
• Nguyen, T. T., Demortière, A., Fleutot, B., Delobel, B., Delacourt, C., & Cooper, S. J. (2020). The electrode tortuosity factor: why the conventional tortuosity factor is not well suited for quantifying transport in porous Li-ion battery electrodes and what to use instead. npj Computational Materials, 6(1), 123.

---

Friday, June 28th, Llyod Building Viz Room, 3pm

Real space density functional theory for large length and time scales

Phanish Suryanarayana (Georgia Institute of Technology, USA)

Over the course of the past few decades, quantum mechanical calculations based on Kohn-Sham density functional theory (DFT) have become a cornerstone of materials research by virtue of the predictive power and fundamental insights they provide. However, the length and time scales accessible to such a rigorous first principles investigation is limited by the large computational expense associated with such calculations. In this talk, previous and current efforts of the speaker to develop efficient real-space formulations and massively parallel implementations for DFT will be discussed. These include (i) SPARC: A general purpose framework for performing large-scale electronic structure calculations based on DFT; (ii) Cyclic+Helical DFT: A framework for studying systems possessing cyclic and/or helical symmetry, with application to the bending and torsional deformations in nanostructures; and (iii) SQDFT: A linear-scaling framework for studying materials under extreme conditions. Overall, the speaker will discuss how the above developments enable electronic structure simulations at length and time scales that were previously intractable.

---

Friday, June 21st, Llyod Building Viz Room, 3pm

Machine learning applications in metal forming

Lukas Morand (Fraunhofer Institute for Mechanics of Materials)

The talk will provide insights into how machine learning is advancing the field of materials manufacturing, with a particular focus on metal forming. The presentation will begin with an overview of the challenges faced in materials manufacturing and metal forming that can be addressed using data-based approaches. The talk will then delve into various topics in the intersection of machine learning and materials manufacturing, which are: the use of machine learning in materials and process design along the process-structure-properties chain, the significance of active learning for efficient model training and design space exploration, the identification of material model parameters, and the challenges associated with solving inverse problems. Finally, the talk will address semantic data and workflow management techniques as the foundation for machine learning applications and concludes with a future outlook on the potential of automated data processing through machine learning applications using a dataspace management system.

---

Friday, May 31th, Llyod Building Viz Room, 3pm

Exploring Next-Generation Quantum Chemistry Calculations: Leveraging a Toolbox with Grassmannians, Fragmentation, and Machine Learning

Ka Un Lao (Virginia Commonwealth University)

Quantum chemistry (QC) has significantly advanced research in modern chemistry, allowing the study of chemical properties and processes at the quantum scale. However, accurate QC methods often require substantial computational resources and time, particularly for large and complex systems. Our group has developed three innovative QC tools to address this computational bottleneck, enabling faster research progress than ever before. The first tool integrates Grassmannian mathematics from differential geometry with electronic structure theory, enabling the efficient calculation of high-quality density matrices at each point on potential energy surfaces. Consequently, this approach substantially accelerates or even eliminates the time-consuming self-consistent field iterative procedure commonly used in density functional theory. Another tool to overcome the nonlinear scaling of QC methods is the fragmentation approach. Our group has devised an accurate and efficient GMBE-DM fragmentation scheme based on set theory. It demonstrates highly accurate absolute and relative energies across a wide range of systems, outperforming other fragmentation approaches by an order of magnitude. Moreover, GMBE-DM achieves these results with faster computational speed, without requiring significant parallelization. Lastly, we have developed accurate and efficient machine learning (ML) models for predicting ab initio dispersion potentials, utilizing only Cartesian coordinates as input. This ML model can be integrated with MP2 to create the most accurate N5 scaling wavefunction method, suitable for modelling even large noncovalent complexes. Together, these three QC tools enable accurate and efficient calculations, significantly expanding the scope of systems amenable to electronic structure calculations. They make large systems and long-time scale simulations accessible within a fraction of the original computing time, thereby accelerating scientific discovery in quantum chemistry.

---

Friday, May 24th, Llyod Building Viz Room, 12am

Electron charge density is a fundamental physical quantity, determining various properties of matter. In this talk, I will introduce a deep learning model for accurate charge-density prediction. Firstly, the Deep Potential framework, based on which the Deep Charge was created, will be discussed and exhibit why the model naturally preserves physical symmetries. Following this, I will demonstrate the model’s accuracy and efficiency in capturing detailed atomic environment information, enabling precise predictions of charge density across bulk, surface, molecular, and amorphous structures. The implementation of Deep Charge is scalable and facilitates efficient analysis of material properties in large-scale condensed matter systems. Finally, I will address the current limitations of the model and potential areas for improvement.

---

Friday, May 17th, Llyod Building Viz Room, 3pm

Theoretical Approaches to study Degradation in Li-ion Batteries

Dr. Hrishit Banerjee (School of Science and Engineering, University of Dundee, Dundee, and Yusuf Hamied Department of Chemistry, University of Cambridge, Cambridge, UK)

Layered Li-ion battery cathode materials are strongly correlated transition metal oxides, where static and dynamic correlations play an important role in the degradation and stability of these materials. In this talk, we examine the effect of strong correlations on the structural and electronic stability of the most relevant Li-ion battery cathodes, like LiNiO₂, LiMnO₂, Li-NMC cathodes.

LiNiO₂ has the highest voltages and capacities in layered Li-ion batteries but is prone to oxygen loss via the formation of singlet oxygen. We observe spontaneous O₂ loss from the (012) surface of delithiated LiNiO₂, singlet oxygen forming in the process. We find that the origin of the instability lies in the pronounced oxidation of O during delithiation, i.e., O plays a central role in Ni-O redox in LiNiO₂. For LiNiO₂, NiO₂, & NiO dynamical mean-field theory (DMFT) calculations yield a Ni charge state of ca.+2, with O varying between –2 (NiO), –1.5 (LiNiO₂) and –1 (NiO₂). Calculated XAS Ni K and O K-edge spectra show excellent with experimental XAS, confirming the predicted charge states and that a shift in Ni K-edge energy primarily arises from changes in Ni-O bond lengths rather than the Ni charge state. Singlet oxygen formation is caused by the singlet ground state of the peroxide ion, with spin conservation dictating the preferential release of ¹O₂ rather than ³O₂ [1]. Moreover, DMFT calculations on rhombohedral LiNiO₂ show for the first time a gap of combined Mott and charge-transfer character, and show the ligand hole state. The paramagnetic insulating state has a band gap of ∼0.6 eV, in excellent agreement with experiments and is in sharp contrast to DFT calculations that require the presence of an extra structural symmetry breaking in the form of Jahn-Teller distortions to open a gap. We further show that whereas DFT shows the presence of an unphysical metallic Drude peak in optical absorption spectra, DMFT calculations capture the correct form of the optical absorption spectra and have an excellent match with the calculated band gap as well [2].

LiMnO₂ is a low-cost, low-toxicity, high safety, and environment-friendly cathode however it undergoes degradation in the form of irreversible transformation from layered to spinel structure causing loss of capacity and voltage. Using ab initio DMFT we explore the electronic and magnetic states of layered Li_xMnO₂ as a function of x, the state-of-charge. Upon delithiation the system proceeds through several states: ferrimagnetic correlated metals at x=0.92, 0.83; multiple charge ordered ferromagnetic correlated metals with large quasiparticle peaks at x=0.67, 0.50, 0.33; ferromagnetic metals with small quasiparticle peaks at x=0.17, 0.08. At moderate x, between 0.67- 0.33, a mix of +3/+4 oxidation states of Mn is observed. The charge ordering in the system is correlated to different quasiparticle weights on different sites, giving rise to the origin of the pathways leading to the observed layered to spinel structural transformation in LiMnO₂ [3].

Finally, we examine O loss in commercially relevant Ni-rich LiNi_aMn_bCo_cO₂ cathodes. Experimentally it is known that O loss becomes more pronounced as Ni content is increased and at high voltages. We find that despite Ni, Mn, and Co K-edges showing an excellent match with experimentally obtained XAS, the ionic model of ascribing shifts in the XAS spectra to changes in metal oxidation states is inappropriate. We show that in these cases, characterised by strong covalency between the transition metal and oxygen, DMFT calculations are essential to calculate charges and hence assign oxidation states accurately. Due to the corresponding charge transfer from O p to Ni d, a ligand hole forms on O in Ni-rich regions. The individual Ni charge remains fairly constant throughout the charging/discharging process, particularly in Ni-rich environments in the material. In contrast, O has dual redox behavior, showing greater involvement in redox in Ni-rich regions while showing negligible redox involvement in Ni-poor regions. The Ni-O covalent system starts participating in redox around a state of delithiation of ~17%, which represents the beginning of charge. Contrary to previous DFT calculations, we show that Co oxidation does not occur at the very end of charge but rather starts at an earlier state of delithiation of ~67%. The dual behaviour of O in terms of participation in the redox process helps explain the overall higher relative stability of lower Ni content NMCs compared to Ni-rich NMCs or LiNiO₂ in terms of O loss and evolution of singlet oxygen. [4] [1] A. R. G-Schriever, H. Banerjee, A. K Menon, E. N. Bassey, L. F. J. Piper C. P. Grey*, A. J. Morris*, Joule 7, 1623-1640 (2023)

[2] H. Banerjee*, M. Aichhorn, C. P. Grey, A. J. Morris, 10.26434/chemrxiv-2024-3n05d (2024).

[3] H. Banerjee*, C. P. Grey, A. J. Morris*, Phys. Rev. B. 108, 165124 (2023).

[4] H. Banerjee*, C. P. Grey*, A. J. Morris*, 10.26434/chemrxiv-2024-bjzbt (2024).

---

Friday, April 26th, Llyod Building Viz Room, 3pm

Adapting Computational Materials Science to Exascales 

Logan Ward (Argonne National Laboratory, US)

The simultaneous arrival of Exascale computing resources and major advances in Artificial Intelligence (AI) algorithms presents a wonderful opportunity. Scientists can both acquire new data and learn from it faster than with smaller computers or with less sophisticated empirical models. In this talk, we will discuss the intersection of AI and supercomputing with an emphasis on how they are natural fits for each other. The talk will cover how AI can augment human expertise in the design and evaluation of materials for both energy and CO2 storage materials; and how AI can become even more integrated into computational materials science. 

---

Friday, April 19th, Llyod Building Viz Room, 3pm

---

Friday, April 5th, Llyod Building Viz Room, 3pm

Matilda Sipilä (1), Farrokh Mehryary (2), Sampo Pyysalo (2), Filip Ginter (2), and Milica Todorović (1)

1 Department of Mechanical and Materials Engineering, University of Turku; 2 Department of Computing, University of Turku

Scientific text is a promising source of data in materials science, and there is ongoing research on how to utilise textual data in materials discovery. In addition to the more established approaches like named entity recognition or dictionary-based methods, new machine learning tools such as question answering (QA) are becoming available. The advantages of this method are that it is easy to scale and that it does not require manual text labeling or annotating work, but there may be some loss in precision compared to other methods.

We tested the performance of the QA method on the well-known task of information extraction. We extracted bandgap values of halide perovskite materials from scientific literature. Large language models (BERT models) were tuned towards a specific QA task and then used to select the correct answer for the question about materials properties. In comparison to more established methods, the QA method performed well, and we were able to extract correct information from text. This information can be used to map the space of materials properties and find promising new materials solutions. The potential in QA method lies in versatility, accessibility and scalability, since it is easy to use even for researchers with no previous knowledge of language technology and can be easily scaled to extract different materials and properties.

---

Friday, March 21st, Llyod Building Viz Room, 3pm

Magnon spin transport in multiferroic materials

Tianxiang Nan (Tsinghua University, China)

Magnons, bosonic quasiparticles carrying angular momentum, can flow through insulators for information transmission with minimal power dissipation. However, it remains challenging to develop a magnon-based logic due to the lack of efficient electrical manipulation of magnon transport. In this talk, I will first review the recent advances in magnon spin transport in ferrimagnetic and antiferromagnetic materials. Then I will present our strategies to achieve the voltage control of magnon transport in multiferroic materials. We show the electric excitation and control of multiferroic magnon modes in a spin-source/multiferroic/ferromagnet structure. We demonstrate that the ferroelectric polarization can electrically modulate the magnon-mediated spin-orbit torque by controlling the non-collinear antiferromagnetic structure in multiferroic bismuth ferrite thin films with coupled antiferromagnetic and ferroelectric orders. In this multiferroic magnon torque device, magnon information is encoded to ferromagnetic bits by the magnon-mediated spin torque. By manipulating the two coupled non-volatile state variables—ferroelectric polarization and magnetization—we further present reconfigurable logic operations in a single device.

---

Friday, March 1st, Llyod Building Viz Room, 3pm

A quantum engine in the BEC-BCS crossover

Jennifer Koch (University of Kaiserslautern-Landau, Germany)

Heat engines convert thermal energy into mechanical work both in the classical and quantum regimes. However, quantum theory offers genuine nonclassical forms of energy, different from heat, which so far have not been exploited in cyclic engines to produce useful work. In this talk, I will discuss a recently realized quantum many-body engine fuelled by the energy difference between fermionic and bosonic ensembles of ultracold particles that follows from the Pauli exclusion principle [1]. We employ a harmonically trapped superfluid gas of Lithium-6 atoms close to a magnetic Feshbach resonance, which allows us to effectively change the quantum statistics from Bose-Einstein to Fermi-Dirac by tuning the gas between a Bose-Einstein condensate of bosonic molecules and a unitary Fermi gas (and back) through a magnetic field. The talk will focus on the quantum nature of such a Pauli engine. Additionally, I will present the pressure-volume diagram of the new kind of engine and show how the engine behaves after multiple cycles. Our findings establish quantum statistics as a useful thermodynamic resource for work production.

[1] J. Koch et al., Nature 621, 723 (2023)

---

Friday, February 16th, Llyod Building Viz Room, 3pm

Vibrational thermal transport modeling: background and insights

Lucas Lindsay (Oak Ridge National Laboratory)

This talk will provide an introduction into first principles phonon thermal transport calculations highlighting a variety of insights and predictions that they have enabled. In this context, we will explore the underpinnings of density functional theory derived lattice dynamics and phonon Boltzmann transport models and how these translate into observables and functionalities of various semiconducting and insulating materials. More specifically, we will examine the roles of symmetry, chirality, and conservation conditions in determining inelastic neutron scattering spectra and lattice thermal transport.

L.L. acknowledges support from the U. S. Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division.

---

Friday, February 9th, Llyod Building Viz Room, 3pm

Self-Driving Fluidic Labs: Accelerated Materials Discovery with Autonomous Experimentation in Flow

Milad Abolhasani (Department of Chemical & Biomolecular Engineering, North Carolina State University)

Accelerating the discovery of new advanced functional materials, as well as green and sustainable ways to synthesize and manufacture them, will have a profound impact on the worldwide challenges in renewable energy, sustainability, and healthcare. The current human-dependent paradigm of experimental research in chemical and materials sciences fails to identify technological solutions for worldwide challenges in a short timeframe. The time-, resource-, and labor-intensive nature of current experimental sciences necessitates the development and implementation of new strategies to accelerate the pace of discovery. Recent advances in reaction miniaturization, lab automation, and data science provide an exciting opportunity to reshape the discovery and manufacturing of new materials related to energy transition and sustainability. In this talk, I will present an overview of our recent efforts to establish a ‘self-driving fluidic lab (SDFL)’ through integration of flow chemistry, robotics, online characterization, and machine learning (ML) for autonomous discovery and manufacturing of emerging advanced functional materials with multi-step chemistries.^1-5 I will discuss how modularization of different material synthesis and processing stages in tandem with a constantly evolving ML modeling and decision-making under uncertainty can enable a resource-efficient navigation through high dimensional experimental design spaces (>10²⁰ possible experimental conditions). Example applications of SDFL for the autonomous synthesis of colloidal quantum dots will be presented to illustrate the potential of autonomous robotic experimentation in reducing synthetic route discovery timeframe from >10 years to a few months. Finally, I will present the unique reconfigurability aspect of flow chemistry to close the scale gap in materials research through facile switching from the reaction exploration/exploitation to smart manufacturing mode.

References.

[1] Abolhasani, M.; Kumacheva, E. The rise of self-driving labs in chemical and materials sciences. Nature Synthesis, 2, 483–492, 2023.

[2] Volk, A. A.; Epps, R. W.; Yonemoto, D. T.; Masters, B. S.; Castellano, F. N.; Reyes, K. G.; Abolhasani, M. AlphaFlow: autonomous discovery and optimization of multi-step chemistry using a self-driven fluidic lab guided by reinforcement learning. Nature Communications, 14 (1), 1403, 2023.

[3] Volk, A. A.; Abolhasani, M. Autonomous flow reactors for discovery and invention. Trends in Chemistry, 3 (7), 519-522, 2021.

[4] Delgado-Licona, F.; Abolhasani, M. Research Acceleration in Self-Driving Labs: Technological Roadmap toward Accelerated Materials and Molecular Discovery. Advanced Intelligent Systems, 5, 2200331, 2023.

[5] Epps, R. W.; Bowen, M. S.; Volk, A. A.; Abdel-Latif, K.; Han, S.; Reyes, K. G.; Amassian, A.; Abolhasani, M. Artificial Chemist: An Autonomous Quantum Dot Synthesis Bot. Advanced Materials, 32 (30), 2001626, 2020.

---

Friday, February 2nd, 2024, Lloyd Building Viz Room, 3pm

Antiferromagnetic Tunnel Junctions for Spintronics

Evgeny Y. Tsymbal (Department of Physics and Astronomy, University of Nebraska, Lincoln, NE 68588, USA)

Antiferromagnetic (AFM) spintronics has emerged as a subfield of spintronics, where an AFM Néel vector is used as a state variable. Due to being robust against magnetic perturbations, producing no stray fields, and exhibiting ultrafast dynamics, antiferromagnets can serve as promising functional materials for spintronic applications. To realize this potential, efficient electric control and detection of the AFM Néel vector are required. This talk features fundamental properties of AFM tunnel junctions (AFMTJs) as spintronic devices where such electric control and detection can be realized [1]. We emphasize critical requirements for observing a large tunneling magnetoresistance (TMR) effect in AFMTJs with collinear [2] and noncollinear [3,4] AFM electrodes, such as crystallinity of the junction, AFM metals exhibiting momentum-dependent spin polarization [2,3], and/or AFM metals supporting Néel spin currents [5]. We further discuss the unique property of non-collinear antiferromagnets to sustain virtually 100% spin polarization [4], the possibility of magnetic tunnel junctions (MTJs) with a single ferromagnetic electrode [6], and spin torques that are capable of Néel vector switching [5]. Overall, AFMTJs have potential to become a new standard for spintronics providing larger magnetoresistive effects, few orders of magnitude faster switching speed, and much higher packing density than conventional MTJs.

1. D.-F. Shao and E. Y. Tsymbal, Antiferromagnetic tunnel junctions for spintronics, arXiv: 2312.13507 (2023).
2. D.-F. Shao, S.-H. Zhang, M. Li, C.-B. Eom, and E. Y. Tsymbal, Spin-neutral currents for spintronics, Nat. Commun. 12, 7061 (2021).
3. J. Dong, X. Li, G. Gurung, M. Zhu, P. Zhang, F. Zheng, E. Y. Tsymbal, and J. Zhang, Tunneling magnetoresistance in noncollinear antiferromagnetic tunnel junctions, Phys. Rev. Lett. 128, 197201 (2022).
4. G. Gurung, D.-F. Shao, and E. Y. Tsymbal, Extraordinary tunneling magnetoresistance in antiferromagnetic tunnel junctions with antiperovskite electrodes, arXiv:2306.03026 (2023).
5. D.-F. Shao, Y.-Y. Jiang, J. Ding, S.-H. Zhang, Z.-A. Wang, R.-Ch. Xiao, G. Gurung, W. J. Lu, Y. P. Sun, and E. Y. Tsymbal, Néel spin currents in antiferromagnets, Phys. Rev. Lett. 130, 216702 (2023).
6. K. Samanta, Y.-Y. Jiang, T. R. Paudel, D.-F. Shao, and E. Y. Tsymbal, Tunneling magnetoresistance in magnetic tunnel junctions with a single ferromagnetic electrode, arXiv:2310.02139 (2023).

---

Friday, January 26th, 2024, Lloyd Building Viz Room, 3pm

Machine learning interatomic potentials for phase change compounds

Marco Bernasconi (Department of Materials Science, University of Milano-Bicocca, Milano, Italy)

Phase change materials such as the flagship Ge₂Sb₂Te₅ (GST) compound are exploited in key enabling technologies including non-volatile electronic memories and neuromorphic computing [1]. In the phase change electronic memory, the two digital states are encoded in the amorphous and crystalline phases of GST that feature a difference in the electrical resistivity by about three orders of magnitude. Readout of the memory consists of the measurement of the resistance at low bias while the set/reset processes consist of a reversible transformation between the crystalline and amorphous phases induced by Joule heating.

Atomistic simulations based on density functional theory (DFT) have provided useful insights on the structural and functional properties of phase change materials over the years. However, several key issues such as the effect of confinement and nanostructuring on the crystallization kinetics, just to name a few, are presently beyond the reach of DFT simulations. A route to overcome the limitations in system size and time scale and enlarge the scope of DFT methods is the exploitation of machine learning techniques trained on a DFT database to generate interatomic potentials for large scale molecular dynamics simulations. The first example of the application of such an approach to the study of phase change compounds dates to 2012 when we devised an interatomic potential for GeTe [2] within the neural network (NN) scheme proposed by Behler and Parrinello [3]. The NN potential was then used to address several issues such as the crystallization in bulk and nanowires, and the thermal conductivity and aging of the amorphous phase [4].

In this talk, we report on the generation of an interatomic potential for the Ge₂Sb₂Te₅ compound within the NN framework implemented in the DeePMD-kit package [6]. The interatomic potential allows simulating several tens of thousands of atoms for tens of ns at a modest computational cost. The validation of the potential and its application to the study of the crystallization kinetics of the amorphous phase will be discussed.

[1] W. Zhang, R. Mazzarello, M. Wuttig, E. Ma, Nat. Rev. Mater. 4, 150 (2019)

[2] G. C. Sosso, G. Miceli, S. Caravati, J. Behler, and M. Bernasconi, Phys. Rev. B 85, 174103 (2012).

[3] J. Behler and M. Parrinello, Phys. Rev. Lett. 98, 146401 (2007).

[4] G. C. Sosso and M. Bernasconi, MRS Bulletin 44, 705 (2019).

[6] H. Wang, L. Zhang, J. Han, and W. E, Comp. Phys. Commun. 228, 178 (2018); L. Zhang, J. Han, H. Wang, R. Car, W. E, Phys. Rev. Lett. Phys. Rev. Lett. 120, 143001 (2018).

[7] O. Abou El Kheir, L. Bonati, M. Parrinello, M. Bernasconi, arXiv: 304.03109; npj Comp. Mater., in press.

---

Friday, January 19th, 2024, Lloyd Building Viz Room, 3pm

Artificial intelligence (AI) is aiding the discovery of sustainable materials in every step along the computational workflow. Machine learning (ML) supports the automated extraction of data from the literature, the creation of large simulation data sets, the generative design of new materials, as well as the computational validation of discovery outcomes. In this talk, I will present our team’s research in the computational discovery of polymers [1,2] and nanopores [3,4,5] for carbon dioxide capture applications. I will discuss some of challenges in the AI/ML design of complex materials and provide examples of how discovery outcomes could be computationally validated, prior to lab synthesis and characterization. In view of global challenges such as climate change, open-science strategies with publicly shared data and models are needed for accelerating computational materials discovery.

[1] Giro, R., Hsu, H., Kishimoto, A. et al. AI powered, automated discovery of polymer membranes for carbon capture. npj Comput Mater 9, 133 (2023). https://doi.org/10.1038/s41524-023-01088-3.

[2] Ferrari, B.S., Manica, M., Giro, R. et al. Predicting polymerization reactions via transfer learning using chemical language models. Preprint (2023). https://arxiv.org/abs/2310.11423.

[3] Oliveira, F.L., Cleeton, C., Neumann Barros Ferreira, R. et al. CRAFTED: An exploratory database of simulated adsorption isotherms of metal-organic frameworks. Sci Data 10, 230 (2023). https://doi.org/10.1038/s41597-023-02116-z.

[4] Zheng, B., Lopes Oliveira, F., Neumann Barros Ferreira, R. et al. Quantum Informed Machine-Learning Potentials for Molecular Dynamics Simulations of CO2’s Chemisorption and Diffusion in Mg-MOF-74. ACS Nano 17 (6), 5579-5587 (2023). https://doi.org/10.1021/acsnano.2c11102.

[5] Cipcigan, F., Booth, J., Neumann Barros Ferreira, R. et al. Discovery of Novel Reticular Materials for Carbon Dioxide Capture using GflowNets. Preprint (2023). https://arxiv.org/abs/2310.07671.

---

Old Seminars

---

Friday, March 9th, 2018, Lloyd Building Viz Room, 3pm

Friday, January 26th, 2018, Lloyd Building Viz Room, 3pm

Friday, January 19th, 2018, Lloyd Building Viz Room, 3pm