Editorial Advisory Board, Journal of Physical Chemistry Letters, 2013–15
Chemist of the Year, South Carolina Section of the ACS, 2014
USC Educational Fund Research Award, 2007
Fellow of the American Physical Society, 2000
Sloan Foundation Fellowship, 1992
National Science Foundation Presidential Young Investigator, 1990
Camille and Henry Dreyfus Distinguished Young Faculty Award, 1987
Department of Chemistry and Biochemistry
Students, Researchers and Emeritus
Mark A. Berg
| Title: | Distinguished Professor Emeritus Biophysical / Chemical Physics / Materials / Nano / Polymer / Spectroscopy |
| Department: | Chemistry and Biochemistry Department of Chemistry and Biochemistry |
| Email: | berg@sc.edu |
| Phone: | 803-777-1514 |
| Fax: | 803-777-9521 |
| Office: | Office: JONES 005A Lab: JONES 016, 803-777-1926 Lab 2: JONES 012 |
| Resources: | CV [pdf] All Publications Department of Chemistry and Biochemistry |
Education
B.S., 1979, University of Minnesota
Ph.D., 1985, University of California at Berkeley
Honors and Awards
Research Interests
We aim to make fundamental advances in the way chemical and biological dynamics are measured and analyzed. Specifically, we are combining techniques from machine learning with nonequilibrium statistical mechanics to improve the analysis of single-molecule experiments and computer simulations. The ultimate goal is a deeper understanding of molecular dynamics in soft, condensed-phase materials, such as biomolecules, polymers, supercooled liquids, and ionic liquids.
Initial results of this work have shown orders-of-magnitude improvement in resolution in both time and state-space over existing methods. In addition, new abilities to measure rate heterogeneity and rate exchange have been demonstrated. Work is ongoing to optimize these methods, to expand them to more experiments, and to create broadly accessible tools for studying complex molecular dynamics. This theoretical and computational work takes inspiration from my earlier, experimental research in ultrafast laser spectroscopy.
Complex Dynamics. Conventionally, experimental kinetics has focused on experiments that measure linear relaxation over a single time period (1D kinetics). However, strong interactions with the environment often lead to complex dynamics, even for elementary molecular processes. The resulting nonexponential decays cannot be assigned to a unique mechanism using only linear, 1D kinetics. We previously pioneered multidimensional and nonlinear perturbation–response experiments to address this problem.
Single-Molecule Kinetics. In perturbation–response experiments, a system at equilibrium is perturbed and the response is measured at a later time. This type of measurement has long dominated kinetics. More recently, single-molecule measurements have made a fundamentally different type of measurement common: The molecule remains at equilibrium, and the dynamics of its thermal fluctuations are measured. Using classical nonequilibrium statistical mechanics, a single-time correlation function of the observable itself extracts the equivalent of a linear, 1D perturbation–response experiment. However, there is no real advance in understanding complex kinetics.
High-Order Correlation Analysis. Our methods rely on using the full range of high-order correlation functions. These include functions that are nonlinear in the experimental observable and/or ones that use multiple time dimensions. This idea is poorly explored in both statistical mechanics and in the broader field of statistical analysis. We have shown that a complete set of these functions achieve the full potential of single-molecule measurements. It extracts all the information available in the data without requiring an initial model or imposing other restrictions on the underlying mechanism.
Machine Learning. Despite its conceptual advantages, high-order correlation analysis of realistic data presents challenges. Single-molecule datasets can be very large, but each point is noisy and low quality—a big data problem. The noise is non-Gaussian and well outside the perturbation limit, obviating many standard noise-averaging methods. Learning about mechanism from kinetics data is an ill-posed, inverse problem that can lead to unstable calculations. We are addressing these challenges with modern machine-learning methods, such as regularization, dimensionality reduction, and optimization theory, as well as more familiar techniques, such as tensor analysis, linear algebra and eigenvalue theory.
Selected Publications
"Using High-Order Correlation Functions on Multivariable, Time-Series Data: A Single-Molecule, FRET Example," M. Dhar and M. A. Berg, J. Chem. Phys. 163, 184113 (19 pp.), 2025. https://doi.org/10.1063/5.0284658.
"Efficient, Nonparametric Removal of Noise and Recovery of Probability Distributions from Time Series Using Nonlinear-Correlation Functions: Photon and Photon-Counting Noise," M. Dhar and M. A. Berg, J. Chem. Phys. 161, 034116 (22 pp.) 2024. http://doi.org/10.1063/5.0212157.
"Efficient, Nonparametric Removal of Noise and Recovery of Probability Distributions from Time Series Using Nonlinear-Correlation Functions: Additive Noise," M. Dhar, J. A. Dickinson, and M. A. Berg, J. Chem. Phys. 159, 054110 (22 pp.) 2023. http://doi.org/10.1063/5.0158199.
"Jump-Precursor State Emerges Below the Crossover Temperature in Supercooled o-Terphenyl," H. Kaur and M. A. Berg, Phys. Rev. E (Letter) 103, L050601 (7 pp.) 2021. http://doi.org/10.1103/PhysRevE.103.L050601.
"Nonlinear Measurements of Kinetics and Generalized Dynamical Modes: II. Application to a Simulation of Solvation Dynamics in an Ionic Liquid," S. R. Hodge, S. A. Corcelli, and M. A. Berg, J. Chem. Phys. 155, 024123 (11 pp.) 2021. http://doi.org/10.1063/5.0053424.
"Nonlinear Measurements of Kinetics and Generalized Dynamical Modes: I. Extracting the One-Dimensional Green's Function from a Time Series," S. R. Hodge and M. A. Berg, J. Chem. Phys. 155, 024122 (20 pp.) 2021. http://doi.org/10.1063/5.0053422.