Virtual Material Design

 

Software

  • software package Tremolo-X/ATK-ForceField  for numerical simulation in molecular dynamics
  • software package HCFFT for efficient treatment of high-dimensional multivariate functions

Our Services

  • data-driven virtual material design
  • mathematical modelling
  • developing algorithms
  • individual solutions

Projects

A survey of application cases

Progress is largely driven by innovative materials with properties  customized to the point.  Hence, materials science, chemistry and nanotechnology  are key technologies  of our century .

The essential basis for designing novel materials and molecules is the understanding of their properties on the nanoscale and their effect on the macroscale.  So multi-scale models in space and time  are necessary to describe a material accurately and to predict its long-term behavior.

Our goal is to optimize, to create and to study novel materials and molecules in the virtual computer lab in order to assist the practical development process by substantially reducing its cost in time and money.

Applied research and development at VMD@SCAI  focuses on data driven design, modern multi-scale modeling, and efficient simulation techniques. To this end we use state-of-the-art numerical methods for high dimensional problems, optimization, machine learning and (big) data analysis.

By interlinking these techniques with in-house knowledge, empirical  data and newly-made own software we pave the way to a generic and comprehensive design for new materials.

 

Innovative data-driven design

 

Traditionally, empiricism (e.g. experiments) and model-based theories (e.g. quantum mechanics) were the two dominant scientific paradigms. Over the last few decades, computational science has established itself as a third paradigm.  Meanwhile, based on the abundance of experimental and complementing simulation data, even so called data-driven techniques have emerged, as the all-embracing, fourth paradigm of science.

Today’s approach to a given question at hand may either proceed via a “white-box” model, an exact physical description of the modeled system (by,  e.g. PDEs). Or alternatively, we might use a “black-box” model reproducing just the input-output behavior of a system (by, e.g., deep learning neural networks).  If not yet satisfactory, hybrid approaches   i.e. “gray-box” models, are  the way to go.