Searcharxiv⌕ Search

arXiv subjects

Georg K. H. Madsen

Publications and source records attributed to Georg K. H. Madsen.

At least 37 records · Page 2Linked to original sources

Deep Ensembles vs. Committees for Uncertainty Estimation in Neural-Network Force Fields: Comparison and Application to Active Learning

A reliable uncertainty estimator is a key ingredient in the successful use of machine-learning force fields for predictive calculations. Important considerations are correlation with error, overhead during training and inference, and efficient workflows to systematically improve the force field. However, in the case of neural-network force fields, simple committees are often the only option considered due to their easy implementation. Here we present a generalization of the deep-ensemble design, based on multiheaded neural networks and a heteroscedastic loss, that can efficiently deal with uncertainties in both the energy and the forces. We compare uncertainty metrics based on deep ensembles, committees and bootstrap-aggregation ensembles using data for an ionic liquid and a perovskite surface. We demonstrate an adversarial approach to active learning to efficiently and progressively refine the force fields. That active learning workflow is realistically possible thanks to exceptionally fast training enabled by residual learning and a nonlinear learned optimizer.

physics.comp-ph↗

Chemical trends in the high thermoelectric performance of the pyrite-type dichalcogenides: ZnS2, CdS2 and CdSe2

The thermoelectric properties of the three pyrite-type IIB-VIA2 dichalcogenides (ZnS2, CdS2 and CdSe2) are systematically investigated and compared with those of the prototype ZnSe2 in order to optimize their thermoelectric properties. Using the phonon Boltzmann transport equation, we find that they all have ultralow lattice thermal conductivities. By analyzing their vibrational properties, these are attributed to soft phonon modes derived from the loosely bound rattling-like metal atoms and to strong anharmonicities caused by the vibrations of all atoms perpendicular to the strongly bound nonmetallic dimers. Additionally, by correlating those properties along the series, we elucidate a number of chemical trends. We find that heavier atom masses, larger atomic displacement parameters and longer bond lengths between metal and nonmetal atoms can be beneficial to the looser rattling of the metal atoms and therefore lead to softer phonon modes, and that stronger nonmetallic dimer bonds can boost the anharmonicities, both leading to lower thermal conductivities. Furthermore, we find that all three compounds have complex energy isosurfaces at valence and conduction band edges that simultaneously allow for large density-of-states effective masses and small conductivity effective masses for both p-type and n-type carriers. Consequently, the calculated thermoelectric figures of merit (ZT), can reach large values both for p-type and n-type doping. Our study illustrates the effects of rattling-like metal atoms and localized nonmetallic dimers on the thermal transport properties and the importance of different carrier effective masses to electrical transport properties in these pyrite-type dichalcogenides, which can be used to predict and optimize the thermoelectric properties of other thermoelectric compounds in the future.

physics.app-ph↗

A Differentiable Neural-Network Force Field for Ionic Liquids

We present NeuralIL, a model for the potential energy of an ionic liquid that accurately reproduces first-principles results with orders-of-magnitude savings in computational cost. Based on a multilayer perceptron and spherical Bessel descriptors of the atomic environments, NeuralIL is implemented in such a way as to be fully automatically differentiable. It can thus be trained on ab-initio forces instead of just energies, to make the most out of the available data, and can efficiently predict arbitrary derivatives of the potential energy. Using ethylammonium nitrate as the test system, we obtain out-of-sample accuracies better than 2 meV/atom (<0.05 kcal/mol) in the energies and 70 meV/Å in the forces. We show that encoding the element specific density in the spherical Bessel descriptors is key to achieving this. Harnessing the information provided by the forces drastically reduces the amount of atomic configurations required to train a neural network force field based on atom-centered descriptors. We choose the Swish-1 activation function and discuss the role of this choice in keeping the neural network differentiable. Furthermore, the possibility of training on small data sets allows for an ensemble-learning approach to the detection of extrapolation. Finally, we find that a separate treatment of long-range interactions is not required to achieve a high-quality representation of the potential energy surface of these dense ionic systems.

physics.comp-ph↗

Accurate effective harmonic potential treatment of the high-temperature cubic phase of Hafnia

HfO$_2$ is an important high-$κ$ dielectric and ferroelectric, exhibiting a complex potential energy landscape with several phases close in energy. It is, however, a strongly anharmonic solid, and thus describing its temperature-dependent behavior is methodologically challenging. We propose an approach based on self-consistent, effective harmonic potentials and higher-order corrections to study the potential energy surface of anharmonic materials. The introduction of a reweighting procedure enables the usage of unregularized regression methods and efficiently harnesses the information contained in every data point obtained from density functional theory. This renders the approach highly efficient and a promising candidate for large-scale studies of materials and phase transitions. We detail the approach and test it on the example of the high-temperature cubic phase of HfO$_2$. Our results for the thermal expansion coefficient, $α_V \approx 3.3\times 10^{-5}$ K$^{-1}$, are in agreement with existing experimental ($α_V \approx 4\pm1\times 10^{-5}$ K$^{-1}$) and theoretical ($α_V \approx 5\pm1\times 10^{-5}$ K$^{-1}$) work. Likewise, the bulk modulus agrees well with experiment. We show the detailed temperature dependence of these quantities.

cond-mat.mtrl-sci↗

Evolutionary computing and machine learning for the discovering of low-energy defect configurations

Density functional theory (DFT) has become a standard tool for the study of point defects in materials. However, finding the most stable defective structures remains a very challenging task as it involves the solution of a multimodal optimization problem with a high-dimensional objective function. Hitherto, the approaches most commonly used to tackle this problem have been mostly empirical, heuristic and/or based on domain knowledge. In this contribution, we describe an approach for exploring the potential energy surface based on the covariance matrix adaption evolution strategy (CMA-ES) and supervised and unsupervised machine learning models. We show how the original CMA-ES can be modified to suit the specific problem of DFT studies of point defects in the dilute limit. The resulting algorithm depends only on a limited set of physically interpretable hyperparameters. The approach offers a robust and systematic way for finding low-energy configurations of point defects in solids. We demonstrate the applicability and moderate computational cost on the intrinsic defects in silicon. We also apply the methodology to the neutral oxygen vacancy oxygen vacancy in TiO$_2$ anatase and reproduce the known defect structures. Furthermore, a new defect structure, stable at the level of hybrid density functional theory and characterized by a delocalized electronic structure, is found for this system.

cond-mat.mtrl-sci↗

How do defects limit the ultrahigh thermal conductivity of BAs? A first principles study

The promise enabled by BAs high thermal conductivity in power electronics cannot be assessed without taking into account the reduction incurred when doping the material. Using first principles calculations, we determine the thermal conductivity reduction induced by different group IV impurities in BAs as a function of concentration and charge state. We unveil a general trend, where neutral impurities scatter phonons more strongly than the charged ones. $\text{C}_{\text{B}}$ and $\text{Ge}_{\text{As}}$ impurities show by far the weakest phonon scattering and retain BAs $κ$ values of over $\sim$ 1000 $\text{W}\cdot\text{K}^{-1}\cdot\text{m}^{-1}$ even up to high densities making them ideal n-type and p-type dopants. Furthermore, going beyond the doping compensation threshold associated to Fermi level pinning triggers observable changes in the thermal conductivity. This informs design considerations on the doping of BAs, and it also suggests a direct way to determine the onset of compensation doping in experimental samples.

cond-mat.mtrl-sci↗

Machine-learning prediction of infrared spectra of interstellar polycyclic aromatic hydrocarbons

We design and train a neural network (NN) model to efficiently predict the infrared spectra of interstellar polycyclic aromatic hydrocarbons (PAHs) with a computational cost many orders of magnitude lower than what a first-principles calculation would demand. The input to the NN is based on the Morgan fingerprints extracted from the skeletal formulas of the molecules and does not require precise geometrical information such as interatomic distances. The model shows excellent predictive skill for out-of-sample inputs, making it suitable for improving the mixture models currently used for understanding the chemical composition and evolution of the interstellar medium. We also identify the constraints to its applicability caused by the limited diversity of the training data and estimate the prediction errors using a ensemble of NNs trained on subsets of the data. With help from other machine-learning methods like random forests, we dissect the role of different chemical features in this prediction. The power of these topological descriptors is demonstrated by the limited effect of including detailed geometrical information in the form of Coulomb matrix eigenvalues.

astro-ph.GA↗

Spinney: post-processing of first-principles calculations of point defects in semiconductors with Python

Understanding and predicting the thermodynamic properties of point defects in semiconductors and insulators would greatly aid in the design of novel materials and allow tuning the properties of existing ones. As a matter of fact, first-principles calculations based on density functional theory (DFT) and the supercell approach have become a standard tool for the study of point defects in solids. However, in the dilute limit, of most interest for the design of novel semiconductor materials, the raw DFT calculations require an extensive post-processing. Spinney is an open-source Python package developed with the aim of processing first-principles calculations to obtain several quantities of interest, such as the chemical potential limits that assure the thermodynamic stability of the defectladen system, defect charge transition levels, defect formation energies, including electrostatic corrections for finite-size effects, and defect and carrier concentrations. In this paper we demonstrate the capabilities of the Spinney code using c-BN, Mg-doped GaN, TiO2 and ZnO as examples.

cond-mat.mtrl-sci↗

Shortcomings of meta-GGA functionals when describing magnetism

Several recent studies have shown that SCAN, a functional belonging to the meta-generalized gradient approximation (MGGA) family, leads to significantly overestimated magnetic moments in itinerant ferromagnetic metals. However, this behavior is not inherent to the MGGA level of approximation since TPSS, for instance, does not lead to such severe overestimations. In order to provide a broader view of the accuracy of MGGA functionals for magnetism, we extend the assessment to more functionals, but also to antiferromagnetic solids. The results show that to describe magnetism there is overall no real advantage in using a MGGA functional compared to GGAs. For both types of approximation, an improvement in ferromagnetic metals is necessarily accompanied by a deterioration (underestimation) in antiferromagnetic insulators, and vice-versa. We also provide some analysis in order to understand in more detail the relation between the mathematical form of the functionals and the results.

cond-mat.mtrl-sci↗

Growth, charge and thermal transport of flowered graphene

We report on the structural and transport properties of the smallest dislocation loop in graphene, known as a flower defect. First, by means of advanced experimental imaging techniques, we deduce how flower defects are formed during recrystallization of chemical vapor deposited graphene. We propose that the flower defects arise from a bulge type mechanism in which the flower domains are the grains left over by dynamic recrystallisation. Next, in order to evaluate the use of such defects as possible building blocks for all-graphene electronics, we combine multiscale modeling tools to investigate the structure and the electron and phonon transport properties of large monolayer graphene samples with a random distribution of flower defects. For large enough flower densities, we find that electron transport is strongly suppressed while, surprisingly, hole transport remains almost unaffected. These results suggest possible applications of flowered graphene for electron energy filtering. For the same defect densities, phonon transport is reduced by orders of magnitude as elastic scattering by defects becomes dominant. Heat transport by flexural phonons, key in graphene, is largely suppressed even for very low concentrations.

cond-mat.mtrl-sci↗

Excellent Thermoelectric Performances of Pressure Synthesized ZnSe2

We calculate the lattice thermal conductivities of the pyrite-type ZnSe2 at pressures of 0 and 10 GPa using the linearized phonon Boltzmann transport equation. We obtain a very low value [0.69 W/(mK) at room temperature at 0 GPa], comparable to the best thermoelectric materials. The vibrational spectrum is characterized by the isolated high-frequency optical phonon modes due to the stretching of Se-Se dimers and low-frequency optical phonon modes due to the rotation of Zn atoms around these dimers. The low-frequency optical phonon modes are characterized by a strong anharmonicity and will substantially increase the three-phonon scattering space which suppress the thermal conductivity. Interestingly, two transverse acoustic phonon modes with similar frequencies and wave vectors have very different degrees of anharmonicity depending on their polarization. We relate this to the low thermal conductivity and show that the anharmonicities of the transverse acoustic phonon modes are connected to the corresponding change in the pyrite parameter, which can be interpreted as a descriptor for the local volume change. To determine the thermoelectric performance of ZnSe2, we also investigate its electrical transport properties. The results show that both p-type or n-type ZnSe2 can show promising electrical transport properties. We trace this back to the complex energy isosurfaces of both valence and conduction bands. The low thermal conductivities and promising electrical transport properties lead to a large thermoelectric figure of merit of ZnSe2 for both p-type and n-type doping.

cond-mat.mtrl-sci↗

A comparative first-principles investigation on the defect chemistry of TiO$_2$ anatase

Understanding native point defects is fundamental in order to comprehend the properties of TiO$_2$ anatase in technological applications. Several first-principles studies have been performed in order to investigate the defect chemistry of this material. The reported values are, however, scattered over a wide range. In this manuscript we perform a comparative study employing different approaches based on semilocal, DFT+$U$ and screened hybrid functionals in order to investigate the dependence of defect properties, such as formation energies and charge transition levels, on the employed computational method. While the defects in anatase, like in most transition-metal oxides, generally induce the localization of electrons or holes on atomic sites, we notice that, provided an alignment of the valence bands has been performed, the calculated defect formation energies and transition levels using semi-local functionals are in a fair agreement with those obtained using hybrid functionals. A similar conclusion can be reached for the thermochemistry of the Ti-O system and the limit values of the elemental chemical potentials. We interpret this as a cancellation of error between the self-interaction error and the overbinding of the O$_2$ molecule in semi-local functionals. Inclusion of a $U$ term in the electron Hamiltonian offers a convenient way for obtaining more precise geometric and electronic configurations of the defective systems.

cond-mat.mtrl-sci↗

Combined treatment of phonon scattering by electrons and point defects explains the thermal conductivity reduction in highly-doped Si

The mechanisms causing the reduction in lattice thermal conductivity in highly P- and B-doped Si are looked into in detail. Scattering rates of phonons by point defects, as well as by electrons, are calculated from first principles. Lattice thermal conductivities are calculated considering these scattering mechanisms both individually and together. It is found that at low carrier concentrations and temperatures phonon scattering by electrons is dominant and can reproduce the experimental thermal conductivity reduction. However, at higher doping concentrations the scattering rates of phonons by point defects dominate the ones by electrons except for the lowest phonon frequencies. Consequently, phonon scattering by point defects contributes substantially to the thermal conductivity reduction in Si at defect concentrations above $10^{19}$ cm$^{-3}$ even at room temperature. Only when, phonon scattering by both point defects and electrons are taken into account, excellent agreement is obtained with the experimental values at all temperatures.

cond-mat.mtrl-sci↗

Effect of local chemistry and structure on thermal transport in doped GaAs

Using a first-principles approach, we analyze the impact of \textit{DX} centers formed by S, Se, and Te dopant atoms on the thermal conductivity of GaAs. Our results are in good agreement with experiments and unveil the physics behind the drastically different effect of each kind of defect. We establish a causal chain linking the electronic structure of the dopants to the thermal conductivity of the bulk solid, a macroscopic transport coefficient. Specifically, the presence of lone pairs leads to the formation of structurally asymmetric \textit{DX} centers that cause resonant scattering of incident phonons. The effect of such resonances is magnified when they affect the part of the spectrum most relevant for thermal transport. We show that these resonances are associated with localized vibrational modes in the perturbed phonon spectrum. Finally, we illustrate the connection between flat adjacent minima in the energy landscape and resonant phonon scattering through detailed analyses of the energy landscape of the defective structures.

cond-mat.mtrl-sci↗

Comparing the performance of LDA and GGA functionals in predicting the lattice thermal conductivity of semiconductor materials: the case of AlAs

In this contribution we assess the performance of two different exchange-correlation functionals in the first-principle prediction of the lattice thermal conductivity of bulk semiconductors, namely the local density approximation (LDA) and the Perdew-Burke-Ernzerhof implementation of the generalized gradient approximation (GGA). Both functionals are shown to give results in good agreement with experimental measurements. Such a consistency between the two functionals may seem a bit surprising, as the LDA is known to overbind and the GGA to soften the interatomic bonds. Such features ought to greatly affect the value of the system interatomic force constants (IFCs) which are necessary for the first-principle prediction of the lattice thermal conductivity. In this study we show that the errors introduced by such approximations tend to cancel themselves. In the case of LDA, the overbinding generates larger absolute third-order IFCs, which tend to increase the three-phonon scattering rates. On the other hand, larger absolute second-order IFCs lead to a a larger acoustic-optical phonon band gap which in turns decrease the available phase space for three-phonon scattering, compensating the increase in the scattering rates due to stiffer IFCs.

cond-mat.mtrl-sci↗

Orbital-free approximations to the kinetic-energy density in exchange-correlation MGGA functionals: tests on solids

A recent study of Mejia-Rodriguez and Trickey [Phys. Rev. A 96, 052512 (2017)] showed that the deorbitalization procedure (replacing the exact Kohn-Sham kinetic-energy density by an approximate orbital-free expression) applied to exchange-correlation functionals of the meta-generalized gradient approximation (MGGA) can lead to important changes in the results for molecular properties. For the present work, the deorbitalization of MGGA functionals is further investigated by considering various properties of solids. It is shown that depending on the MGGA, common orbital-free approximations to the kinetic-energy density can be sufficiently accurate for the lattice constant, bulk modulus, and cohesive energy. For the band gap, calculated with the modified Becke-Johnson MGGA potential, the deorbitalization has a larger impact on the results.

cond-mat.mtrl-sci↗

Vibrational properties of metastable polymorph structures by machine learning

Despite vibrational properties being critical for the ab initio prediction of the finite temperature stability and transport properties of solids, their inclusion in ab initio materials repositories has been hindered by expensive computational requirements. Here we tackle the challenge, by showing that a good estimation of force constants and vibrational properties can be quickly achieved from the knowledge of atomic equilibrium positions using machine learning. A random-forest algorithm trained on only 121 metastable structures of KZnF$_3$ reaches a maximum absolute error of 0.17 eV/$\textrmÅ^2$ for the interatomic force constants, and it is much less expensive than training the complete force field for such compound. The predicted force constants are then used to estimate phonon spectral features, heat capacities, vibrational entropies, and vibrational free energies, which compare well with the ab initio ones. The approach can be used for the rapid estimation of stability at finite temperatures.

cond-mat.mtrl-sci↗

BoltzTraP2, a program for interpolating band structures and calculating semi-classical transport coefficients

BoltzTraP2 is a software package for calculating a smoothed Fourier expression of periodic functions and the Onsager transport coefficients for extended systems using the linearized Boltzmann transport equation. It uses only the band and $k$-dependent quasi-particle energies, as well as the intra-band optical matrix elements and scattering rates, as input. The code can be used via a command-line interface and/or as a Python module. It is tested and illustrated on a simple parabolic band example as well as silicon. The positive Seebeck coefficient of lithium is reproduced in an example of going beyond the constant relaxation time approximation.

cond-mat.mtrl-sci↗