Unsupervised topological learning for identification of atomic structures

Becker, Sandrine; Devijver, Émilie; Molinier, Rémi; Jakse, N.

doi:10.1103/physreve.105.045304

Cited by 12 publications

(4 citation statements)

References 55 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To identify translational and orientational orderings during homogeneous nucleation in large-scale MD simulations, an unsupervised learning approach was developed through persistent homology (PH) within the topological data analysis (TDA). In a recently published article [24] we successfully assessed the accuracy of PH against classical methods, such as the common neighbour analysis (CNA) [25], the Steinhardt parameters [26] or the polyhedra template matching (PTM) [27] on known structures. CNA is a very robust method which detects small structural building blocks (bonded pairs) and gives access to a characterization in terms of global proportions of these pairs in the simulation box.…”

Section: Toward Unsupervised ML For Topological Characterisation and ...mentioning

confidence: 99%

Medium Range Ordering in liquid Al-based alloys: towards a machine learning approach of solidification

Jarry

Jakse

2023

IOP Conf. Ser.: Mater. Sci. Eng.

View full text Add to dashboard Cite

Ab initio molecular dynamic simulations of liquid Al93Cr7 and Al83Zn10Cr7 alloy have evidenced the presence of an icosahedral short range order (iSRO) which develops into icosahedral medium range order (iMRO) as the melt is undercooled. This atomic arrangement accounts for the presence of Dynamic Heterogeneities characterized by Al fast-dynamics regions and Cr-rich slow-dynamics regions. Characterisation of the medium range order was carried out by a direct connectivity approach. However, given the small size of the simulation (256 atoms), such characterisation remains partial. In order to better describe both iMRO formation and more dilute alloys closer to industrial compositions, a new modelling strategy has been initiated to allow in the long term for large-scale atomic-level simulations. Molecular Dynamics (MD) of million to billion atoms may indeed lead to meaningful results. Exploitation of such large amounts of MD-generated big data can be carried out by means of Machine Learning (ML) tools which provide relevant and powerful analysis methods. An unsupervised ML approach based on topological descriptors using persistent homology concepts is proposed to reveal the structural features of atomic arrangements without a priori knowledge on the studied system. This approach has been applied so far to pure Al melts. Both translational and orientational orderings are thus evidenced together with nucleation pathways, whose revealed features are beyond the hypotheses of the Classical Nucleation Theory.

show abstract

Section: Toward Unsupervised ML For Topological Characterisation and ...mentioning

confidence: 99%

Medium Range Ordering in liquid Al-based alloys: towards a machine learning approach of solidification

Jarry

Jakse

2023

IOP Conf. Ser.: Mater. Sci. Eng.

View full text Add to dashboard Cite

show abstract

“…As such, they are often application-and/or structure-specific, and are not always easy to generalize beyond their original scope of applicability. More recently, data-driven machine-learning (ML) approaches are being developed for performing ordered phase classification and sometimes defect detection [10][11][12][13][14][15][16][17], often employing existing tools such as Steinhardt order parameters [1] for featurization. While comparatively more straightforward to develop with modern ML pipelines, these emerging methods require considerable amounts of carefully curated training data and are often informed by material-specific physics and domain knowledge which limit transferability of the trained models.…”

Section: Introductionmentioning

confidence: 99%

Score-based denoising for atomic structure identification

Hsu

Sadigh

Bertin

et al. 2023

Preprint

View full text Add to dashboard Cite

We propose an accurate method for removing thermal vibrations that complicate the task of analyzing complex dynamics in atomistic simulation of condensed matter. Our method iteratively subtracts thermal noises or perturbations in atomic positions using a denoising score function trained on synthetically noised but otherwise perfect crystal lattices. The resulting denoised structures clearly reveal underlying crystal order while retaining disorder associated with crystal defects. Purely geometric, agnostic to interatomic potentials, and trained without inputs from explicit simulations, our denoiser can be applied to simulation data generated from vastly different interatomic interactions. Followed by a simple phase classification tool such as the Common Neighbor Analysis, the denoiser outperforms other existing methods and reaches perfect classification accuracy on a recently proposed benchmark dataset consisting of perturbed crystal structures (DC3). Demonstrated here in a wide variety of atomistic simulation contexts, the denoiser is general, robust, and readily extendable to delineate order from disorder in structurally and chemically complex materials.

show abstract

“…Interestingly, new unsupervised machine learning (UML) algorithms also give new avenues to probe structure (see, e.g., Refs. [20][21][22][23][24][25][26][27][28][29]. Recent studies have even demonstrated that simple, UML-based approaches are able to extract variations in disorder in the structure of supercooled fluids from, e.g., a vector of bond order parameters.…”

Section: Introductionmentioning

confidence: 99%

In search of a precursor for crystal nucleation of hard and charged colloids

de Jager,

Smallenburg,

Filion

2023

The Journal of Chemical Physics

View full text Add to dashboard Cite

The interplay between crystal nucleation and the structure of the metastable fluid has been a topic of significant debate over recent years. In particular, it has been suggested that even in simple model systems such as hard or charged colloids, crystal nucleation might be foreshadowed by significant fluctuations in local structure around the location where the nucleus first arises. We investigate this using computer simulations of spontaneous nucleation events in both hard and charged colloidal systems. To detect local structural variations, we use both standard and unsupervised machine learning methods capable of finding hidden structures in the metastable fluid phase. We track numerous nucleation events for the face-centered cubic and body-centered cubic crystals on a local level and demonstrate that all signs of crystallinity emerge simultaneously from the very start of the nucleation process. We thus conclude that we observe no precursor for the crystal nucleation of hard and charged colloids.

show abstract

Unsupervised topological learning for identification of atomic structures

Cited by 12 publications

References 55 publications

Medium Range Ordering in liquid Al-based alloys: towards a machine learning approach of solidification

Medium Range Ordering in liquid Al-based alloys: towards a machine learning approach of solidification

Score-based denoising for atomic structure identification

In search of a precursor for crystal nucleation of hard and charged colloids

Contact Info

Product

Resources

About