اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدNeural network representation of electronic structure from ab initio molecular dynamics
110
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
Despite their rich information content, electronic structure data amassed at high volumes in ab initio molecular dynamics simulations are generally under-utilized. We introduce a transferable high-fidelity neural network representation of such data in the form of tight-binding Hamiltonians for crystalline materials. This predictive representation of ab initio electronic structure, combined with machine-learning boosted molecular dynamics, enables efficient and accurate electronic evolution and sampling. When applied to a one-dimension charge-density wave material, carbyne, we are able to compute the spectral function and optical conductivity in the canonical ensemble. The spectral functions evaluated during soliton-antisoliton pair annihilation process reveal significant renormalization of low-energy edge modes due to retarded electron-lattice coupling beyond the Born-Oppenheimer limit. The availability of an efficient and reusable surrogate model for the electronic structure dynamical system will enable calculating many interesting physical properties, paving way to previously inaccessible or challenging avenues in materials modeling.
قيم البحث
اقرأ أيضاً
The marriage of density functional theory (DFT) and deep learning methods has the potential to revolutionize modern research of material science. Here we study the crucial problem of representing DFT Hamiltonian for crystalline materials of arbitrary
configurations via deep neural network. A general framework is proposed to deal with the infinite dimensionality and covariance transformation of DFT Hamiltonian matrix in virtue of locality and use message passing neural network together with graph representation for deep learning. Our example study on graphene-based systems demonstrates that high accuracy ($sim$meV) and good transferability can be obtained for DFT Hamiltonian, ensuring accurate predictions of materials properties without DFT. The Deep Hamiltonian method provides a solution to the accuracy-efficiency dilemma of DFT and opens new opportunities to explore large-scale materials and physics.
The effects of tetragonal strain on electronic and magnetic properties of strontium-doped lanthanum manganite, La_{2/3}Sr_{1/3}MnO_3 (LSMO), are investigated by means of density-functional methods. As far as the structural properties are concerned, t
he comparison between theory and experiments for LSMO strained on the most commonly used substrates, shows an overall good agreement: the slight overestimate (at most of 1-1.5 %) for the equilibrium out-of-plane lattice constants points to possible defects in real samples. The inclusion of a Hubbard-like contribution on the Mn d states, according to the so-called LSDA+U approach, is rather ineffective from the structural point of view, but much more important from the electronic and magnetic point of view. In particular, full half-metallicity, which is missed within a bare density-functional approach, is recovered within LSDA+U, in agreement with experiments. Moreover, the half-metallic behavior, particularly relevant for spin-injection purposes, is independent on the chosen substrate and is achieved for all the considered in-plane lattice constants. More generally, strain effects are not seen to crucially affect the electronic structure: within the considered tetragonalization range, the minority gap is only slightly (i.e. by about 0.1-0.2 eV) affected by a tensile or compressive strain. Nevertheless, we show that the growth on a smaller in-plane lattice constant can stabilize the out-of-plane vs in-plane e_g orbital and significatively change their relative occupancy. Since e_g orbitals are key quantities for the double-exchange mechanism, strain effects are confirmed to be crucial for the resulting magnetic coupling.
We investigated the structural and dynamical properties of a tetrahedrally coordinated crystalline ice from first principles based on density functional theory within the generalized gradient approximation with the projected augmented wave method. Fi
rst, we report the structural behaviour of ice at finite temperatures based on the analysis of radial distribution functions obtained by molecular dynamics simulations. The results show how the ordering of the hydrogen bonding breaks down in the tetrahedral network of ice with entropy increase in agreement with the neutron diffraction data. We also calculated the phonon spectra of ice in a 3x1x1 supercell by using the direct method. So far, due to the direct method used in this calculation, the phonon spectra is obtained without taking into account the effect of polarization arising from dipole-dipole interactions of water molecules which is expected to yield the splitting of longitudinal and transverse optic modes at the Gamma-point. The calculated longitudinal acoustic velocities from the initial slopes of the acoustic mode is in a reasonable agreement with the neutron scatering data. The analysis of the vibrational density of states shows the existence of a boson peak at low energy of translational region a characteristic common to amorphous systems.
We present a scheme for the improved description of the long-range interatomic force constants in a more accurate way than the procedure which is commonly used within plane-wave based density-functional perturbation-theory calculations. Our scheme is
based on the inclusion of a q point grid which is denser in a restricted area around the center of the Brillouin Zone than in the remaining parts, even though the method is not limited to an area around Gamma. We have tested the validity of our procedure in the case of high-pressure phases of bulk silicon considering the bct and sh structure.
We develop a Python-based open-source package to analyze the results stemming from ab initio molecular-dynamics simulations of fluids. The package is best suited for applications on natural systems, like silicate and oxide melts, water-based fluids,
various supercritical fluids. The package is a collection of Python scripts that include two major libraries dealing with file formats and with crystallography. All the scripts are run at the command line. We propose a simplified format to store the atomic trajectories and relevant thermodynamic information of the simulations, which is saved in UMD files, standing for Universal Molecular Dynamics. The UMD package allows the computation of a series of structural, transport and thermodynamic properties. Starting with the pair-distribution function it defines bond lengths, builds an interatomic connectivity matrix, and eventually determines the chemical speciation. Determining the lifetime of the chemical species allows running a full statistical analysis. Then dedicated scripts compute the mean-square displacements for the atoms as well as for the chemical species. The implemented self-correlation analysis of the atomic velocities yields the diffusion coefficients and the vibrational spectrum. The same analysis applied on the stresses yields the viscosity. The package is available via the GitHub website and via its own dedicated page of the ERC IMPACT project as open-access package.
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
