Home | CV | Research | People | Teaching | Blog |

Schmidt | - Perspective: Machine Learning |

We present

......................................................

With
this paper, we can push beyond the limitations of dynamical
density functional theory. And due to the use of
**neural networks**, we can do this much faster.

Superadiabatic
forces occur in **nonequilibrium** and they can be very
strong.

To get some training data for the neural network, we had to run a
bunch of nonequilibrium many-body simulations. For this we
used **adaptive Brownian dynamics**, which finds just the right
pace for the numerical
time evolution of the systems.

Our neural network also satisfies **exact Noether
identities**. These identities
determine how the different force contributions balance each
other.

The adiabatic approximation in dynamical density functional theory
is uncontrolled. If you plan to ride the **exciting trail of
nonequilibrium soft matter physics** then be on the safe side and
use
power functional theory. Yeah!

**Update** 13.12.2023: Treating equilibrium via ** Neural functional theory for soft matter**
is based on machine learning deep functional relationships of statistical mechanics; see
Press Release for
PNAS,
Review, and
Tutorial.

Daniel de las Heras, Toni Zimmermann, Florian Sammüller, Sophie Hermann, and Matthias Schmidt,

J. Phys.: Condens. Matter

Florian Sammüller, Sophie Hermann, Daniel de las Heras, and Matthias Schmidt,

Proc. Nat. Acad. Sci.

Florian Sammüller, Sophie Hermann, and Matthias Schmidt (submitted). arxiv, tutorial, Press Release, pdf.

Watch the

Copyright © MS 29 March 2022