We present an accurate machine learning (ML) model for atomistic simulations of carbon, constructed using the Gaussian approximation potential (GAP) methodology. The potential, named GAP-20, describes the properties of the bulk crystalline and amorphous phases, crystal surfaces, and defect structures with an accuracy approaching that of direct ab initio simulation, but at a significantly reduced cost. We combine structural databases for amorphous carbon and graphene, which we extend substantially by adding suitable configurations, for example, for defects in graphene and other nanostructures. The final potential is fitted to reference data computed using the optB88-vdW density functional theory (DFT) functional. Dispersion interactions, which are crucial to describe multilayer carbonaceous materials, are therefore implicitly included. We additionally account for long-range dispersion interactions using a semianalytical two-body term and show that an improved model can be obtained through an optimization of the many-body smooth overlap of atomic positions descriptor. We rigorously test the potential on lattice parameters, bond lengths, formation energies, and phonon dispersions of numerous carbon allotropes. We compare the formation energies of an extensive set of defect structures, surfaces, and surface reconstructions to DFT reference calculations. The present work demonstrates the ability to combine, in the same ML model, the previously attained flexibility required for amorphous carbon [V. L. Deringer and G. Csányi, Phys. Rev. B 95, 094203 (2017)] with the high numerical accuracy necessary for crystalline graphene [Rowe et al., Phys. Rev. B 97, 054303 (2018)], thereby providing an interatomic potential that will be applicable to a wide range of applications concerning diverse forms of bulk and nanostructured carbon.

A central paradigm of self-assembly is to create ordered structures starting from molecular monomers that spontaneously recognize and interact with each other via noncovalent interactions. In the recent years, great efforts have been directed towards perfecting the design of a variety of supramolecular polymers and materials with different architectures. The resulting structures are often thought of as ideally perfect, defect-free supramolecular fibers, micelles, vesicles, etc., having an intrinsic dynamic character, which are typically studied at the level of statistical ensembles to assess their average properties. However, molecular simulations recently demonstrated that local defects that may be present or may form in these assemblies, and which are poorly captured by conventional approaches, are key to controlling their dynamic behavior and properties. The study of these defects poses considerable challenges, as the flexible/dynamic nature of these soft systems makes it difficult to identify what effectively constitutes a defect, and to characterize its stability and evolution. Here, we demonstrate the power of unsupervised machine learning techniques to systematically identify and compare defects in supramolecular polymer variants in different conditions, using as a benchmark 5Å-resolution coarse-grained molecular simulations of a family of supramolecular polymers. We show that this approach allows a complete data-driven characterization of the internal structure and dynamics of these complex assemblies and of the dynamic pathways for defects formation and resorption. This provides a useful, generally applicable approach to unambiguously identify defects in these dynamic self-assembled materials and to classify them based on their structure, stability and dynamics.