Research

research lines | freeware | publications

Two aspects of our research actives are, in a nutshell, developing codes and algorithms and applying them to model systems of molecules, surfaces, interfaces and bulks. We are in close collaboration with experimentalist researches on surface physics. Our research interests cover various topics as detailed below.

Machine Learning in Physics

Machine Learning (ML), a subset of artificial intelligence, refers to a broad set of computational methods to implement statistical models. In contrast to physical models that are based on physical insight or empirical laws, ML is a data-driven approach to find (not necessarily physically meaningful) relations or patters in a set of data-points obtained in experiment or by simulations. We use the supervised techniques (artificial neural network and kernel ridge regression) to train a flexible mathematical model on the set of training samples, and then make predictions for new samples. For instance, the charge on an atom or its contribution to the electron density in a large assembly of atoms is predicted by comparing its environment with that of already known samples.

Atomic-scale energy dissipation

Macroscopic and microscopic pictures of friction differ in several aspects. For instance, a frictionless and nondissipative motion, known as superlubricity, emerges at the nanoscale between flat surfaces even under ambient conditions. Using atomistic modeling and quantum mechanical simulations, we attack the problem of energy dissipation to reveal some novel physics of the phenomenon at the atomic scale. Joint computational-experimental-statistical projects are performed in close collaboration with the Nanolino Lab and Center For Complex Networks & Social Cognition.
Related publications: [23, 26, 27, 28]

Code for fast simulation of nanoscale friction within the framework of the Frenkel-Kontrova-Tomlinson model.

Scanning probe microscopy

Scanning probe methods provide high resolution images of various atomic surfaces. With a theoretical modelling, the images are interpreted and related to physical and chemical properties of the probed surface. For instance, Kelvin probe force microscopy (KPFM) is a variant of the non-contact atomic force microscopy (NC-AFM) for imaging surface potential even with an atomic resolution. Other operation modes like friction force microscopy are also at our focus. Classical electrostatics, ab initio calculations and molecular dynamics simulations are our tools in the research. Related publications: [1, 3, 6, 12, 23]

Molecular adsorption on surfaces

In the field of condensed matter and surface science, joint computational-experimental studies have been the common practice over the past decade. In collaboration with the Nanolino Lab, we study electronic and structural properties of various organic molecules adsorbed at surfaces as involved in many areas of nanotechnology and surface physics from molecular electronics to molecular self-assembly. Related publications: [13, 17, 19, 21, 22, 24]

Structural properties of materials at nanoscale

Many properties of nanostructures depend on trusty determination of their geometries. This, in turn, depends not only on the efficiency of the method used to explore their configuration space, but also on the accuracy of the description of the interatomic interactions. The density functional theory is nowadays a unique, practically feasible approach for studying structural and electronic properties of nanostructures having tens to a few hundred of atoms on an ab initio level. Related publications: [9, 14, 15, 18]

Electronic & mechanical properties of sp² materials

We use quantum mechanical and classical simulations to investigate physical properties of virtually 2D materials like graphene, silicene, h-BN sheets, etc. In such layered materials, each layer is a one atom thick sheet and has a variety of interesting electronic and mechanical properties. For instance, graphene has several unusual electronic properties which originate from the massless Dirac fermions, which can be tuned e.g. by doping with external atoms, applying electric fields, introducing structural defects and so on. Related publications: [2, 4, 5, 11, 16]

Developing computational methods

Computational physics, the amazing scientific mixture made out of physics, mathematics and computer science, is nowadays a critical element in research. The progress in the field of computational condensed matter physics, for instance, has made it possible to perform accurate quantum mechanical calculations on realistic model systems. We contribute to the development of electronic structure calculation packages, and developing efficient algorithms for exploring the configuration spaces of nanostructures in collaboration with the Computational Physics Group at Basel University. Related publications: [7, 8, 10]

Freeware

Charges from Kernel Ridge Regression

CKRR is an implementation of the ridge regression with Gaussian kernel for predicting atomic quantities (charge, e.g.) for molecules or extended systems with periodic cells. The code can be used in three modes: training, k-fold cross validation, or prediction. The code is still under development and sample input files and detailed instructions can be provided by communication with the developer.

Fast Integrator for FKT model

FKT.f90 is an implementation of the Frenkel-Kontrova-Tomlinson model for simulation of friction between a long chain of atoms sliding on an ideal atomic surface. The fast integrator is based on the fourth order Runge-Kutta algorithm.

Electrostatics of SPM

CapSol.f90 is a finite difference capacitance solver specialized for simulating the electrostatics of a scanning probe microscope. The probe is modeled as a conducting cone terminated at a spherical cap and supported by a disk cantilever. The probed sample can be conducing or a multilayer dielectric slab.

Fingerprinting nano-clusters

fingerprint.f90 calculates the eigenvalues of an overlap matrix for atom-centered Gaussian type orbitals (GTO) as a structural fingerprint vector. fpdriver.f90 is a wrapper program to test and learn how to use the subroutines. The wrapper reads any arbitrary number of structures in xyz formated input file and gives out the corresponding fingerprints and their pairwise distances. See also the Q&A page. Subroutines with additional functionalities (arbitrary number of GTOs, derivatives of the fingerprints with respect to atomic positions, etc.) are also available upon request.

Generating charge-like density

gausscube.f90 is a small program to show how one can generates cube files. It reads in not only atoms coordinates, but also their electronic charge and radius form some input file called input.xyz. Then puts a Gaussian on top of each atom according to the given charge and radius. The resulting charge density is written in file charge.cube and can be visualized by e.g. VESTA package. You may adapt this toy program so that it suits your purpose which can be generating any kind of volumetric field.