Capabilities and limits of autoencoders for extracting collective variables in atomistic materials science.

Free energy calculations in materials science are routinely hindered by the need to provide reaction coordinates that can meaningfully partition atomic configuration space, a prerequisite for most enhanced sampling approaches. Recent studies on molecular systems have highlighted the possibility of constructing appropriate collective variables directly from atomic motions through deep learning techniques. Here we extend this class of approaches to condensed matter problems, for which we encode the finite temperature collective variable by an iterative procedure starting from 0 K features of the energy landscape i.e. activation events or migration mechanisms given by a minimum - saddle point - minimum sequence. We employ the autoencoder neural networks in order to build a scalar collective variable for use with the adaptive biasing force method. Particular attention is given to design choices required for application to crystalline systems with defects, including the filtering of thermal motions which otherwise dominate the autoencoder input. The machine-learning workflow is tested on body-centered cubic iron and its common defects, such as small vacancy or self-interstitial clusters and screw dislocations. For localized defects, excellent collective variables as well as derivatives, necessary for free energy sampling, are systematically obtained. However, the approach has a limited accuracy when dealing with reaction coordinates that include atomic displacements of a magnitude comparable to thermal motions, e.g. the ones produced by the long-range elastic field of dislocations. We then combine the extraction of collective variables by autoencoders with an adaptive biasing force free energy method based on Bayesian inference. Using a vacancy migration as an example, we demonstrate the performance of coupling these two approaches for simultaneous discovery of reaction coordinates and free energy sampling in systems with localized defects.

Gate-Tunable Magnetism and Giant Magnetoresistance in Suspended Rhombohedral-Stacked Few-Layer Graphene.

Conventionally, magnetism arises from the strong exchange interaction among the magnetic moments of d- or f-shell electrons. It can also emerge in perfect lattices from nonmagnetic elements, such as that exemplified by the Stoner criterion. Here we report tunable magnetism in suspended rhombohedral-stacked few-layer graphene (r-FLG) devices with flat bands. At small doping levels (n ∼ 1011 cm-2), we observe prominent conductance hysteresis and giant magnetoconductance that exceeds 1000% as a function of magnetic fields. Both phenomena are tunable by density and temperature and disappear at n > 1012 cm-2 or T > 5 K. These results are confirmed by first-principles calculations, which indicate the formation of a half-metallic state in doped r-FLG, in which the magnetization is tunable by electric field. Our combined experimental and theoretical work demonstrate that magnetism and spin polarization, arising from the strong electronic interactions in flat bands, emerge in a system composed entirely of carbon atoms.

Theoretical study of metal/silica interfaces: Ti, Fe, Cr and Ni on ß-cristobalite.

The understanding of interfacial effects and adhesion at oxide-metal contacts is of key importance in modern technology. Metal-silica interfaces specifically are relevant in electronics, catalysis and nanotechnology. However, adhesion at these interfaces is hindered by a formation of siloxane rings on the silica surface which saturate the dangling bonds at stoichiometric terminations. In this context, we report a thorough density functional theory study of the interaction between ß-cristobalite and selected 3d transition metals under different oxygen conditions. For any given interface stoichiometry, we find a progressive decrease of the metal/silica interaction along the series, following the increase of metal electronegativity. Crucially, in presence of early transition metals (Ti or Cr) the surface siloxane rings are spontaneously broken, allowing for strong adhesion. Late transition metals interact only weakly with reconstructed surfaces, similarly to what was found for zinc. In absence of reconstruction, stoichiometric silica/metal contacts behave similarly to alumina/metal contacts, but display larger interactions across the interface. Based on these results, we show that early transition metal or stainless steel buffers can significantly improve the weak adhesion between silica and zinc, responsible for a poor performance of anti-corrosive galvanic zinc coatings on modern advanced high strength steels.

The CRYSTAL code, 1976-2020 and beyond, a long story.

CRYSTAL is a periodic ab initio code that uses a Gaussian-type basis set to express crystalline orbitals (i.e., Bloch functions). The use of atom-centered basis functions allows treating 3D (crystals), 2D (slabs), 1D (polymers), and 0D (molecules) systems on the same grounds. In turn, all-electron calculations are inherently permitted along with pseudopotential strategies. A variety of density functionals are implemented, including global and range-separated hybrids of various natures and, as an extreme case, Hartree-Fock (HF). The cost for HF or hybrids is only about 3-5 times higher than when using the local density approximation or the generalized gradient approximation. Symmetry is fully exploited at all steps of the calculation. Many tools are available to modify the structure as given in input and simplify the construction of complicated objects, such as slabs, nanotubes, molecules, and clusters. Many tensorial properties can be evaluated by using a single input keyword: elastic, piezoelectric, photoelastic, dielectric, first and second hyperpolarizabilities, etc. The calculation of infrared and Raman spectra is available, and the intensities are computed analytically. Automated tools are available for the generation of the relevant configurations of solid solutions and/or disordered systems. Three versions of the code exist: serial, parallel, and massive-parallel. In the second one, the most relevant matrices are duplicated on each core, whereas in the third one, the Fock matrix is distributed for diagonalization. All the relevant vectors are dynamically allocated and deallocated after use, making the code very agile. CRYSTAL can be used efficiently on high performance computing machines up to thousands of cores.

Surface thermodynamics of silicate compounds: the case of Zn2SiO4(001) surfaces and thin films.

Silicate compounds are ubiquitous in nature and display a vast variety of structures and properties. Thin silicate films may also form under specific conditions at interfaces between metals and silica. In the present study, we focus on zinc silicate and present a thorough density functional theory-based study of polar and non-polar (001) surfaces of various stoichiometry of its tetragonal polymorph t-Zn2SiO4. At the non-polar surfaces, the main features are the existence of the chain reconstruction at the ZnO termination, and the presence of unsaturated surface silanols at the SiO2 termination. Stabilization of polar surfaces is provided by the formation of O22- peroxo groups, reduction of the surface or subsurface Si atoms or formation of Zn22+ groups, depending upon the surface stoichiometry. While the non-polar stoichiometric and ZnO rich terminations are the most stable in a large part of the accessible phase diagram, the SiO2 termination is less stable due to the absence of siloxane group formation. We show that, while bulk Zn2SiO4 is stable with respect to decomposition into the ZnO and SiO2 oxides, the same is not true for ultra-thin films due to the fundamental difference of silicate and silica surface energies. Preliminary results show that a similar conclusion could be drawn for Fe2SiO4. This study opens the way towards a deeper understanding and possible improvement of zinc adhesion at silica surfaces, of crucial industrial importance.

Critical Role of the Exchange Interaction for the Electronic Structure and Charge-Density-Wave Formation in TiSe_{2}.

We show that the inclusion of screened exchange via hybrid functionals provides a unified description of the electronic and vibrational properties of TiSe_{2}. In contrast to local approximations in density functional theory, the explicit inclusion of exact, nonlocal exchange captures the effects of the electron-electron interaction needed to both separate the Ti-d states from the Se-p states and stabilize the charge-density-wave (CDW) (or low-T) phase through the formation of a p-d hybridized state. We further show that this leads to an enhanced electron-phonon coupling that can drive the transition even if a small gap opens in the high-T phase. Finally, we demonstrate that the hybrid functionals can generate a CDW phase where the electronic bands, the geometry, and the phonon frequencies are in agreement with experiments.

Raman spectroscopic features of the neutral vacancy in diamond from ab initio quantum-mechanical calculations.

Quantum-mechanical ab initio calculations are performed to elucidate the vibrational spectroscopic features of a common irradiation-induced defect in diamond, i.e. the neutral vacancy. Raman spectra are computed analytically through a Coupled-Perturbed-Hartree-Fock/Kohn-Sham approach as a function of both different defect spin states and defect concentration. The experimental Raman features of defective diamond located in the 400-1300 cm(-1) spectral range, i.e. below the first-order line of pristine diamond at 1332 cm(-1), are well reproduced, thus corroborating the picture according to which, at low damage densities, this spectral region is mostly affected by non-graphitic sp(3) defects. No peaks above 1332 cm(-1) are found, thus ruling out previous tentative assignments of different spectral features (at 1450 and 1490 cm(-1)) to the neutral vacancy. The perturbation introduced by the vacancy to the thermal nuclear motion of carbon atoms in the defective lattice is discussed in terms of atomic anisotropic displacement parameters (ADPs), computed from converged lattice dynamics calculations.

Electron density analysis of large (molecular and periodic) systems: A parallel implementation.

A parallel implementation is presented of a series of algorithms for the evaluation of several one-electron properties of large molecular and periodic (of any dimensionality) systems. The electron charge and momentum densities of the system, the electrostatic potential, X-ray structure factors, directional Compton profiles can be effectively evaluated at low computational cost along with a full topological analysis of the electron charge density (ECD) of the system according to Bader's quantum theory of atoms in molecules. The speedup of the parallelization of the different algorithms is presented. The search of all symmetry-irreducible critical points of the ECD of the crystallized crambin protein and the evaluation of all the corresponding bond paths, for instance, would require about 32 days if run in serial mode and reduces to less than 2 days when run in parallel mode over 32 processors.