Mixed-Gen Season 3 – Session 3: Soft matter and machine learning
Location: On-line, hosted by CECAM-HQ
The Mixed-Gen on-line series is aimed mainly at PhD students and researchers in their first post-doc. Our goal is to continue providing a venue for these young scientists to share their work, get expert feedback and have an opportunity to strengthen scientific relations within the CECAM community and beyond.
Sessions consist of two parts. In the first, publicly available on Zoom, an experienced speaker and two/three young scientists present talks. In the second, accessible only to registered participants, posters are presented in a GatherTown room.
More detailed information on the program will appear on this page closer to the date of the event.
Links for the session:
To register use the Participate tab on this page
If you do not have a CECAM account register by clicking here...and welcome to CECAM!
Submission of posters
(Please note that - at least for the time being - we shall accept posters only from PhD students or researchers in their first two post-docs)
Register for the session as described above.
After your application is accepted, you will be able to submit a poster. In the CECAM page for this event, go to “My participation” tab and click on “Add a poster”, providing title and abstract following the recommended format. On the same form you can upload your poster file in png or jpg as soon as it is ready. These formats are strict to enable showing of the poster in the Gather session. If the poster file is not ready at the moment of submitting your abstract, you can upload it later by editing your submission (Go to “My participation” tab and click three vertical dots on “Actions” column on table “My posters”). Please upload your poster as soon as possible to enable a decision from the selection committee - see below.
Please note that posters will be visible in the Gather room associated with this session until the end of the series (June 2023) unless otherwise requested.
DEADLINE FOR SUBMISSION: TEN DAYS BEFORE THE EVENT
Selection of posters
Posters will be selected by the event organisers with the support of our main speaker and experts who will take part in the poster session.
Selection of the talks by PhD or first year postdocs
These contributions, to be broadcasted in the Zoom webinar in the first part of the event, will be selected, after a preliminary screening by the organisers, the main speaker and guest experts, from the posters selected for the Gather session. Please tick “No” to the question “Upgrade to talk?” in your application if you DO NOT WANT your poster to be considered for upgrade to a talk.
THE DECISION ON THE POSTER AND THE OUTCOME OF THE SELECTION OF THE TALKS WILL BE COMMUNICATED AT THE LATEST FOUR DAYS BEFORE THE EVENT
POSTER SUBMISSIONS BEYOND THIS DEADLINE WILL BE ACCEPTED BUT NOT CONSIDERED FOR UPGRADE TO TALK. SUBMISSION WILL BE DEFINITELY CLOSED FOUR DAYS BEFORE THE EVENT.
SESSION 3. Title and abstract of talks
Soft matter and machine learning
Laura Filion, Utrecht University
Developments in machine learning (ML) have opened the door to fully new methods for studying phase transitions due to their ability to extremely efficiently identify complex patterns in systems of many particles. Applications of machine learning techniques vary from the use of developing new ML-based order parameters for complex crystal structures, to locating phase transitions, to speeding up simulations. The rapid emergence of multiple applications of machine learning to statistical mechanics and materials science demonstrates that these techniques are becoming an important tool for studying soft matter systems. In this talk, I will briefly present an overview of the work my group is doing on using machine learning to study soft matter systems, with a focus on new strategies to connect the dynamics and structure in glassy fluids.
Learning mappings between equilibrium states of liquid systems using normalizing flows
Alessandro Coretti, Sebastian Falkner, Phillip Geissler, Christoph Dellago, University of Vienna
In the framework of Deep Learning, generative models are getting increasing attention due to their ability of generating completely new samples starting from a set of given examples. The use of these tools in statistical mechanics is really tempting considering the difficulty of generating decorrelated samples from the Boltzmann distribution, notwithstanding the knowledge of its analytical expression. This path of thoughts has been followed in recent publications and physical configurations [1,2] as well as reactive trajectories  have been generated using a particular network architecture called normalizing flows, with very interesting results for biophysical and many-body systems.
The standard approach of these models consits in training a network to transform samples from an easy-to-sample distribution (e.g a Gaussian or a uniform distribution) to samples obtained from the correct Boltzmann probability. In particular, normalizing flows are build in such a way so that it is possible to reweight the network-generated density to obtain samples that are exactly distributed according to the Boltzmann distribution. Yet, the training process of such networks can be painful, in particular for disordered systems, where prior and target distributions are very different from each other.
In this view, this study  aims at exploring different choices for the prior distribution which are closer, from a physical point of view, to the target one. This can lead on the one hand to an efficient exploration of the space of thermodynamics variables for a given model and, on the other, to the possibility of transforming between configurations obtained using different representations of the same physical system. Proof-of-principles calculations are presented for disordered systems of particles at different thermodynamics states and with different potential energy functions.
F. Noé, S. Olsson, J. Köhler, H. Wu, Science, 365, (2019)
P. Wirnsberger, A. Ballard, G. Papamakarios, S. Abercrombie, S. Racanière, A. Pritzel, D. Jimenez Rezende, C. Blundell, J. Chem. Phys., 153, 144112 (2020)
S. Falkner, A. Coretti, S. Romano, P. Geissler and C. Dellago, arXiv preprint arXiv:2207.14530 (2022)
A. Coretti, S. Falkner, P. Geissler and C. Dellago, arXiv preprint arXiv:2208.10420 (2022)
Glass Transition in Polymer Melts using Data-driven Methods
Atreyee Banerjee, Hsiao-Ping Hsu, Kurt Kremer, Oleksandra Kukharekno, Max Planck Institute for Polymer Research
On cooling, the dynamical properties of many polymer melts slow down exponentially, leading to a glassy state without any drastic change in static structure. Understanding the nature of glass transition, as well as precise estimation of the glass transition temperature (Tg) for polymeric materials, remain open questions in both experimental and theoretical polymer sciences. We propose a data-driven approach, which utilises the high-resolution details accessible through the molecular dynamics simulation and considers the structural information of individual chains. It clearly identifies the glass transition temperature of polymer melts of semiflexible chains. By combining principal component analysis (PCA) and clustering (shown in the schematic), we identify glass transition temperature at the asymptotic limit even from relatively short-time trajectories, which just reach into the Rouse-like monomer displacement regime. We observe that fluctuations captured by the principal component analysis reflect the change in a chain's behaviour: from conformational rearrangement above to small vibrations below the glass transition temperature. We demonstrate the generality of the approach by using different dimensionality reduction and clustering approaches. The method can be applied to a wide range of systems with microscopic/atomistic information. More recently we applied this methodology to all-atom acrylic paint systems.
Ignacio Pagonabarraga (University of Barcelona) - Organiser
Sara Bonella (CECAM HQ) - Organiser
Andrea Cavalli (CECAM HQ) - Organiser