The CTCI launched a seminar series in 2024 to discuss the latest scientific findings in computational chemistry, foster collaboration amongst scientists, and gain inspiration for future work. The CTCI's seminar series features renowned experts in computational chemistry, artificial intelligence, chemical theory, quantum chemistry, and more. 

Previous seminars:

AI Computations Using DNA Molecules for Medical Therapy Design in a Lab-on-Chip Device  Amlan Ganguly, Rochester Institute of Technology In this talk, Dr. Ganguly discussed the design and evaluation of a microfluidic Lab-on-Chip (LoC) system capable of performing neural network computations using DNA molecules. Recent research has demonstrated DNA as a potential medium for digital data storage. Inspired by this breakthrough, computer scientists have been researching ways to compute with DNA molecules resulting in designs capable of performing logical operations as well as complex arithmetic such as vector-matrix operations. In this research, Ganguly and his team explored a novel method for implementing deep learning operations such as deep neural networks (DNNs), with molecular reactions on DNA in a microfluidic device. Their proposed system implementing a DNA based neural network processor leveraging its bio-compatibility can provide intelligence to an Organ-on-Chip (OoC) which is a device with natural or engineered miniature tissues in a LoC. Artificial Intelligence (AI) in OoCs can solve difficult problems in health care such as drug evaluation. With the progress in DNA based therapeutics, Ganguly believes that there will be a need to evaluate the effectiveness of the effect of such therapy in real-time and in-situ, that is, as the drug interacts with living tissue or cells.
Efficiency and Accuracy Challenges in AI-Driven Atomistic Simulation Justin Smith, NVIDIA This presentation delved into advancements and challenges in atomistic simulation for chemistry and materials science, focusing on three key themes. First, Smith explored batching techniques to improve simulation efficiency for high-throughput atomistic modeling. Second, he discussed issues in scaling Graph Neural Network (GNN)-based atomistic machine learning models for large-scale simulations, examining limitations and potential solutions. Third, he addressed the impact of training data selection and targets on improving model accuracy. By covering these topics, this presentation provided an overview of his recent publications and the status of software tools development for AI-driven atomistic simulation, offering valuable insights and future directions for researchers and practitioners.
Rapid quantum ground state preparation via dissipative dynamics Lin Lin, University of California, Berkeley and Lawrence Berkeley National Laboratory Inspired by natural cooling processes, dissipation has become a promising approach for preparing low-energy states of quantum systems. However, the potential of dissipative protocols remains unclear beyond certain commuting Hamiltonians. Dr. Lin and his team provided significant analytical and numerical insights into the power of dissipation for preparing the ground state of non-commuting Hamiltonians. For quasi-free dissipative dynamics, which includes certain 1D spin systems with boundary dissipation, we prove an explicit and sharp bound on the mixing time, which scales polynomially in system size. Their results revealed a new connection between the mixing time in trace distance and spectral properties of a non-Hermitian Hamiltonian, and highlight the role of the coherent term in the dissipative dynamics that is often not amenable to previous analysis. They also proved rapid mixing for certain weakly interacting spin and fermionic models in arbitrary dimensions, extending recent results for high-temperature quantum Gibbs samplers to the zero-temperature regime. Their theoretical approaches are applicable to systems with singular stationary states, and are thus expected to have more general applications.
Excess Electron Density at the Metal or Ligand in Actinide Compounds: Insights into Metal-Ligand Bonding and Reactivity Ivan A. Popov, Assistant Professor, Department of Chemistry, Washington State University Dr. Popov discussed how separating heavy elements is one of the key challenges in nuclear waste management. Gaining a deeper understanding of actinide (An)-ligand interactions—across both low and high oxidation states— from an electronic structure perspective is crucial for designing more efficient ligands for nuclear separation chemistry.
Recent Developments in the Quantum Fragmentation Methodology for Studying Large Molecules and Condensed-phase Systems Xiao He, East China Normal University        Honghui Shang, University of Science and Technology of China The primary computational challenge with traditional ab initio methods stems from the scaling issue, wherein the expense of calculations scales as nth power or even worse with the size of the system. Over the last two decades, the fragmentation approach has emerged as a promising avenue for advancing quantum mechanical (QM) methodologies and their utilization in studying large molecules. In this presentation, Drs. He and Shang outlined their recent development, the Electrostatically Embedded Generalized Molecular Fractionation with Conjugate Caps (EE-GMFCC) method, tailored for QM investigations of biomolecular systems.  Through the incorporation of an electrostatic embedding field in each fragment calculation and streamlining the computational burden associated with long-range two-body QM interactions, the EE-GMFCC strategy accurately reproduces molecular properties (such as dipole moment, electron density, and electrostatic potential), total energy, and electrostatic solvation energy obtained from full-system QM computations for proteins.
Foundational models for materials chemistry Gábor Csányi, University of Cambridge The latest and most successful computing architectures leverage many-body symmetric descriptions of local geometry and equivariant message passing networks. Perhaps the most surprising recent result is the stability of models trained on very diverse training sets across the whole periodic table. Dr. Csányi and his team discovered that the MACE-MP-0 model, which was trained on just ~150,000 real and hypothetical small inorganic crystals, is capable of stable molecular dynamics at ambient conditions on any system tested so far. The astounding generalization performance of such foundation models opens the possibility to creating a universally applicable interatomic potential with useful accuracy for materials (especially when fine-tuned with a little bit of domain-specific data) and democratizes quantum-accurate large scale molecular simulations by lowering the barrier to entry into the field.
Recent Advances in Electronic Structure Calculations via Tensor Network State Methods on HPC Infrastructures Örs Legeza, Wigner Research Centre for Physics and Technical University of Munich Finding an optimal representation of a quantum many-body wave function, i.e., a parametrization with the minimum number of parameters for a given error margin is a task of utmost importance in modern quantum physics and chemistry. Recently, Dr. Legaza and his team proposed a general approach to achieve this via global fermionic mode optimization. They demonstrated that altogether several orders of magnitude in computational time can be saved by performing calculations on an optimized basis and by utilizing AI accelerator-based hybrid CPU-multiGPU parallelization. Scaling analysis on NVIDIA DGX-H100 is also presented advertising that quarter petaflops performance can be reached on a single node. This means a factor of 80 speedup compared to a 128-core CPU node or equivalently to a factor of 10 000 with respect to a single CPU thread.