SOft MAtter dynamics with Dlaunay-based Neighbours Search – SOMA-DNS

In all simulations of colloidal suspensions, the main bottleneck related to computational time is the neighborhood search, which requires special attention. Most of the time, this search is performed by grid-based methods, Verlet lists or combinations of both methods. The XLIM and IRCER teams have been collaborating on this topic since 2015 with several publications.Recently, a new technique based on 2D Delaunay triangulations on GPU has been proposed by Chilean participants in the present project, which seems promising especially for mesoscopic techniques. To the best of our knowledge, the Delaunay triangulation approach for 2D simulations has never been applied to the simulation of soft matter dynamics in 3D.

The first step of the project was the purchase of GPU processors for scientific computing, which were delivered and installed in spring 2021 within a computing cluster accessible to all participants. This cluster is managed and maintained by the DSI of the University of Limoges: www.unilim.fr/dsi/

The scientific part has really started from September 1st 2021, with the recruitment of a PhD student and a research engineer, and the results obtained so far are in accordance with the forecasts:
- The project website is operational and should be enriched as the project progresses: gitlab.xlim.fr/soma-dns/project
- The first scientific deliverable concerns the simulation code for the Delaunay 2D approach. The aim was to present a proof of concept to demonstrate the interest of this approach for simulations based on particles of different types, which was the subject of a publication in the form of a poster and a calculation code available online
- Regarding the simulation code for colloidal suspensions with Brownian dynamics, several versions have been developed and none is yet published but this should be the case in the upcoming weeks.

For the rest of the project, we believe that the delivery of the computational codes described in the document initially submitted to the ANR should be as expected:
- Simulation code for the 3D Delaunay approach expected at T0+27
- Simulation code for complex colloidal suspensions (Brownian dynamics) planned for T0+21
- Simulation code for colloidal suspensions (SRD-MD) planned for T0+27
- Simulation code for colloidal suspensions with constraints planned at T0+36
- Simulation code for colloidal suspensions integrating the Delaunay 3D approach planned for T0+45

The part of the project related to the Delaunay approach has been published as a poster presented at the Eurographics 2022 conference (Reims): diglib.eg.org/handle/10.2312/egp20221002
The corresponding code is available online: gitlab.com/hporro01/mcleap

For the part related to the simulation of colloidal suspensions, a poster will be presented at the Journées de la Matière Condensée (JMC) which will be held in Lyon from August 22 to 26, 2022: «SRD-MD simulations of colloidal suspensions in confined environments« (H. Semaan, M. Cerbelaud, B. Crespin and A. Videcoq)
Conference website: jmc2022.univ-lyon1.fr/fr

During the last few decades in Material Science, numerous computer simulations have been developed to study soft matter (including complex fluids, polymers in suspension, colloidal suspensions, etc). The main bottleneck is related to neighborhood search, which allows to determine the set of neighbouring particles within a given radius, in order to compute their interactions. The present project aims at studying this specific problem using a Delaunay-based approach in 3 dimensions on the GPU, thanks to a collaboration between two french groups at the U. of Limoges and a team in Chile. We wish to extend an existing GPU parallel Delaunay model from 2D to 3D, which presents several challenging issues, and provide a general framework, easy to use and to adapt to different simulations setups. To achieve these goals we will rely on a co-supervised PhD student for 36 month and an engineer for 24 months. The total financial support asked is 331?825 €.