It doesn’t matter how beautiful your theory is, it doesn’t matter how smart you are. If it doesn’t agree with experiment, it’s wrong. In that simple statement is the key to science.

-- Richard Feynman

My current research is on computational and data-driven study of nanoscale/microscale materials. I have developed a highly efficient and storage/memory cheap computational method to resolve spatially-dependent dynamics of complex fluids spanning multiple length and time scales, with good agreements to experiments (X-ray photon correlation spectroscopy, X-ray analog of dynamic light scattering). This code was used for liquid-liquid extraction study of multi-component organic systems (surfactant and organic solvents) and shear thickening behavior of dense colloidal particles. Additionally, I’m working on the development of reinforcement learning algorithms to guide the discovery of kinetically driven synthesis pathways of metastable materials, e.g., carbon and 2-dimensional phosphorene.

Past research

My PhD research focused on studying polymer behaviors using statistical mechanics and polymer theories. In particular, the polymer involved in my PhD research was polyethylene oxide (PEO) and/or polypropylene oxide (PPO). The behaviors and properties in solutions and nanostructures (polymer brushes), including PEO-grafted nanopores, PEO/PPO block copolymers (BCP) in diblock or triblock, highly grafted polymers (bottlebrush polymers, e.g., PVA-g-PEO) were elucidated through a series of publications in Macromolecules, the premier journal in polymer science.

Previous research also includes computational mechanics modeling at the continuum level. For example, the viscoelastic modeling of 3D printable concrete using a meshfree method, i.e., the reproducing kernel particle method (RKPM). The main difficulty was in the numerical implementation of material (hyperelasticity, viscoelasticity, and/or viscoplasticity depending on the rheology states of the mixtures) and geometrical nonlinearity associated with the large strain model. I also have rich experience in using machine learning (ML) for scientific problems such as drug-membrane interactions and computational viscoelasticity problems.

Computational design of polymer-grafted hybrid systems

Polymer-grafted hybrid materials have been ubiquitously employed in various engineering applications. The design of these hybrid materials with superior performances requires a molecularly detailed understanding of the structure and dynamics of the polymer brushes and their interactions with the grafting substrate. Molecular dynamics (MD) simulations are very well suited for the study of these materials which can provide molecular insights into the effects of polymer composition and length, grafting density, substrate composition and curvatures, and nanoconfinement. However, few existing tools are available to generate such systems, which would otherwise reduce the barrier of preparation for such systems to enable high throughput simulations. To that end, I have developed a general, flexible, and easy to use Python program named polyGraft, for automated generation of molecular structure and topology of polymer grafted hybrid materials for MD simulations purposes, ranging from polymer brushes grafted to hard substrates, to densely grafted bottlebrush polymers. polyGraft is openly accessible on GitHub (https://github.com/nanogchen/polyGraft) and also listed on LAMMPS official website (here).

Self-assembly of nonlinear topological polymers

self-assembly of topological polymers has very board applications, due to the nanometer-sized nanostructures. Conventional linear block copolymers have enjoyed great success, attributing to their diverse self-assembled nanostructures, such as lamellar (perforated lamellar), cylinders, spheres, and gyroids etc. Novel nonlinear topological polymers are less examined. Our research has discovered broad self-assembled nanostructures, similar to conventional linear block copolymer, but with more design freedom (design parameters) to control the system.

Bottlebrush polymers: architecture-property relationships

Bottlebrush polymers (BBPs) are densely grafted cylindrical brush polymers, which has unique properties than conventional polymers endowed by its broad architectures enabled by the backbone polymerization, side chain polymerization and type (homopolymer vs block copolymer), and the grafting density. The backbone chain rigidity can be tuned by the grafts (grafting density and grafts length), and demonstrates backbone strengthening upon side chain length increasing. Because of the steric effect brought by grafts, BBPs have reduced entanglements and therefore can be made as super soft elastomers.
  BBPs are good candidates for responsive control by external stimuli, such as temperature and pH, when responsive polymers are grafted. In these cases, the BBPs can serve as a carrier used for, e.g., drug delivery. Other associative interaction can also be exploited for responsive control, such as DNA base pairs. There are much more possibilities with BBPs. For example, both hydrophobic and hydrophilic polymers can be grafted to the backbone.
  I have developed multiscale MD models for bottlebrush polymers, namely the all-atom (AA) model based on the OPLS-AA force field, coarse-grained models based on the Martini force field (the top snapshot shown left) and the bead-spring model or the Kremer–Grest model (the bottom snapshot shown left). See more: [AA model] [Bead-spring model]

Polymer-grafted nanopore: hard-soft interface and nanoconfinement effect

Polymer hydration is very important for practical applications of hydrophilic polymers, such as PEO. The functionality of PEO in solution relies on its hydration in water, namely the ability to form hydrogen bonds with water. PEO has been extensively applied to promote anti-biofouling (due to its biocompatibility), surface modification as a means to reduce surface frictions, and filtration membranes. One typical example of use is the PEO-grafted nanopore for gated control of material transportation, which demand control of structural and hydration properties of the grafted PEO.
  The polymer chain length, grafting density, and pore radii can influence the structural and hydration properties. Dependent on these parameters, polymers can be adsorbed on the gold surface, undergo mushroom-like states, to overlap chain states, and to the brush states. We find that if the pore size exceeds the polymer length, increasing the grafting density or chain length for a given pore results in conformational changes of the polymer from pancake-like shapes to well-hydrated overlapping mushrooms to a denser less hydrated polymer brush. An increase of pore curvature results in considerable polymer crowding within the pore, which translates into an increase of the effective grafting density, that has to be taken into consideration when concave nanopores and planar and convex surfaces are compared. For a given polymer grafting density and chain length, a decrease in the pore radius R results in an increased radial orientation, and for a low grafting density, there is a slight increase of the grafted layer height H until the polymer chains fill the pore (R ∼ H) and start folding near the pore surface, conforming into a cone-shape with a further brush height decrease until the high-density limit is reached. The properties of polymer-grafted nanopores, such as water exchange and gating capability, depend on the polymer conformation and hydration, which are strongly influenced by the polymer grafting density and differ for polymer-filled and open nanopores. See more: [how to build the model]

Polymer-grafted nanopores: co-solvent/temperature induced gated control; nanofluidic flow and/or polymer translocation

Polymer grafted nanopores have found various nanotechnological applications, endowed by the surface modifications by the grafted polymers. One application is the gated control of materials transportation through the nanopore driven by external triggers. For example, with addition of co-solvents leads to pore coverage change (opening/closing). The interactions between the co-solvent and the grafted polymers brings an active control element into the gated nanopore control due to the competitive interactions between polymer and co-solvents/water (left figure). Temperature can also play a role for nanopore control when grafted with temperature-responsive polymers, for example, the PEO-diblock-PPO grafted nanopores (bottom left figure). There are many more possibilities for gated control, as long as additional forces are introduced to change the morphology of the grafted polymers, such as external flow which tends to shear the polymers and make it more aligned (e.g., flow-induced crystallization) and results in pore opening with different degrees (bottom right figure).

Artificial Intelligence for Materials Science/Computational Mechanics

Artificial intelligence (AI) has demonstrated strong capabilities to advance scientific discoveries in the last decade, such as AlphaGo and AlphaFold, both of which solved long-standing challenges that conventional methods cannot address. Machine learning as one subset of AI, has become more and more accepted and popular across many research fields. Two types of problems can be solved by ML, namely forward problems and inverse problems. I have developed research experience in cheminformatics of small drug-like molecules and polymers as well as in computational mechanics (constitutive modeling).
  Using DNN and LASSO regression, we can find important descriptors (with physical and chemical insights) among many of them in determining drug permeation through lipid membranes, combined with physics-based simulations. With a DNN/CNN hybrid model, the potential of mean force can be predicted with high accuracy. Using the RNN, we show that the surrogate RNN model can have high prediction accuracy compared to physics-based computational models. See more: [Me] [ML in MS] [RNN-molecular-generation]

Viscoelastic modeling towards 3D printable concrete

To numerically model the behavior of concrete at fresh or solidified state, a viscoelastic constitutive law for both small-strain and finite-strain were implemented based on the generalized Maxwell model. In the solidified state, a small-strain version of the viscoelastic constitutive relation can be used to model phenomena like stress relaxation and creep. While for fresh state mixture, the large strain version can model the finite deformation present in the deposition process, and interface evolution in 3D printing. However, careful selection of a numerical framework is necessary, since in the traditional finite element approach, large deformations can cause mesh distortion and entanglement, and topological changes in the domain (free surface formation, closure) require computationally intensive remeshing. Meshfree methods on the other hand, do not require mesh and can deal with these aforementioned problems easily. A numerical framework based on the meshfree reproducing kernel particle method (RKPM) was developed in this work to model the deposition process. It is observed that the numerical results agree well with reference results, which indicates a strong potential for the effectiveness of the numerical framework for viscoelastic modeling of extremely large deformation problems such as deposition of concrete. See more: [finite-strain] [small-strain] [the code]