| 1. |
- Heinen, Stefan, et al.
(författare)
-
Reducing training data needs with minimal multilevel machine learning (M3L)
- 2024
-
Ingår i: Machine Learning. - : Institute of Physics Publishing (IOPP). - 2632-2153. ; 5:2
-
Tidskriftsartikel (refereegranskat)abstract
- For many machine learning applications in science, data acquisition, not training, is the bottleneck even when avoiding experiments and relying on computation and simulation. Correspondingly, and in order to reduce cost and carbon footprint, training data efficiency is key. We introduce minimal multilevel machine learning (M3L) which optimizes training data set sizes using a loss function at multiple levels of reference data in order to minimize a combination of prediction error with overall training data acquisition costs (as measured by computational wall-times). Numerical evidence has been obtained for calculated atomization energies and electron affinities of thousands of organic molecules at various levels of theory including HF, MP2, DLPNO-CCSD(T), DFHFCABS, PNOMP2F12, and PNOCCSD(T)F12, and treating them with basis sets TZ, cc-pVTZ, and AVTZ-F12. Our M3L benchmarks for reaching chemical accuracy in distinct chemical compound sub-spaces indicate substantial computational cost reductions by factors of ∼1.01, 1.1, 3.8, 13.8, and 25.8 when compared to heuristic sub-optimal multilevel machine learning (M2L) for the data sets QM7b, QM9LCCSD (T), Electrolyte Genome Project, QM9CACESD(T), and QM9CECASD(T), respectively. Furthermore, we use M2L to investigate the performance for 76 density functionals when used within multilevel learning and building on the following levels drawn from the hierarchy of Jacobs Ladder: LDA, GGA, mGGA, and hybrid functionals. Within M2L and the molecules considered, mGGAs do not provide any noticeable advantage over GGAs. Among the functionals considered and in combination with LDA, the three on average top performing GGA and Hybrid levels for atomization energies on QM9 using M3L correspond respectively to PW91, KT2, B97D, and τ-HCTH, B3LYP*(VWN5), and TPSSH.
|
|
| 2. |
- Karandashev, Konstantin, et al.
(författare)
-
Evolutionary Monte Carlo of QM Properties in Chemical Space : Electrolyte Design
- 2023
-
Ingår i: Journal of Chemical Theory and Computation. - : American Chemical Society (ACS). - 1549-9618 .- 1549-9626. ; 19:23, s. 8861-8870
-
Tidskriftsartikel (refereegranskat)abstract
- Optimizing a target function over the space of organic molecules is an important problem appearing in many fields of applied science but also a very difficult one due to the vast number of possible molecular systems. We propose an evolutionary Monte Carlo algorithm for solving such problems which is capable of straightforwardly tuning both exploration and exploitation characteristics of an optimization procedure while retaining favorable properties of genetic algorithms. The method, dubbed MOSAiCS (Metropolis Optimization by Sampling Adaptively in Chemical Space), is tested on problems related to optimizing components of battery electrolytes, namely, minimizing solvation energy in water or maximizing dipole moment while enforcing a lower bound on the HOMO–LUMO gap; optimization was carried out over sets of molecular graphs inspired by QM9 and Electrolyte Genome Project (EGP) data sets. MOSAiCS reliably generated molecular candidates with good target quantity values, which were in most cases better than the ones found in QM9 or EGP. While the optimization results presented in this work sometimes required up to 106 QM calculations and were thus feasible only thanks to computationally efficient ab initio approximations of properties of interest, we discuss possible strategies for accelerating MOSAiCS using machine learning approaches.
|
|
| 3. |
- Faber, Felix A., et al.
(författare)
-
Machine Learning Energies of 2 Million Elpasolite (AB2D6) Crystals
- 2016
-
Ingår i: Physical Review Letters. - : American Physical Society. - 0031-9007 .- 1079-7114. ; 117:13
-
Tidskriftsartikel (refereegranskat)abstract
- Elpasolite is the predominant quaternary crystal structure (AlNaK2F6 prototype) reported in the Inorganic Crystal Structure Database. We develop a machine learning model to calculate density functional theory quality formation energies of all ∼2×106 pristine ABC2D6 elpasolite crystals that can be made up from main-group elements (up to bismuth). Our model’s accuracy can be improved systematically, reaching a mean absolute error of 0.1 eV/atom for a training set consisting of 10×103 crystals. Important bonding trends are revealed: fluoride is best suited to fit the coordination of the D site, which lowers the formation energy whereas the opposite is found for carbon. The bonding contribution of the elements A and B is very small on average. Low formation energies result from A and B being late elements from group II, C being a late (group I) element, and D being fluoride. Out of 2×106 crystals, 90 unique structures are predicted to be on the convex hull—among which is NFAl2Ca6, with a peculiar stoichiometry and a negative atomic oxidation state for Al.
|
|
| 4. |
- Faber, Felix, et al.
(författare)
-
Crystal structure representations for machine learning models of formation energies
- 2015
-
Ingår i: International Journal of Quantum Chemistry. - : Wiley. - 0020-7608 .- 1097-461X. ; 115:16, s. 1094-1101
-
Tidskriftsartikel (refereegranskat)abstract
- We introduce and evaluate a set of feature vector representations of crystal structures for machine learning (ML) models of formation energies of solids. ML models of atomization energies of organic molecules have been successful using a Coulomb matrix representation of the molecule. We consider three ways to generalize such representations to periodic systems: (i) a matrix where each element is related to the Ewald sum of the electrostatic interaction between two different atoms in the unit cell repeated over the lattice; (ii) an extended Coulomb-like matrix that takes into account a number of neighboring unit cells; and (iii) an ansatz that mimics the periodicity and the basic features of the elements in the Ewald sum matrix using a sine function of the crystal coordinates of the atoms. The representations are compared for a Laplacian kernel with Manhattan norm, trained to reproduce formation energies using a dataset of 3938 crystal structures obtained from the Materials Project. For training sets consisting of 3000 crystals, the generalization error in predicting formation energies of new structures corresponds to (i) 0.49, (ii) 0.64, and (iii) 0.37eV/atom for the respective representations.
|
|
| 5. |
- French, Roger H., et al.
(författare)
-
Long range interactions in nanoscale science
- 2010
-
Ingår i: Reviews of Modern Physics. - 0034-6861. ; 82:2, s. 1887-1944
-
Forskningsöversikt (refereegranskat)abstract
- Our understanding of the “long range” electrodynamic, electrostatic, and polar interactions that dominate the organization of small objects at separations beyond an interatomic bond length is reviewed. From this basic-forces perspective, a large number of systems are described from which one can learn about these organizing forces and how to modulate them. The many practical systems that harness these nanoscale forces are then surveyed. The survey reveals not only the promise of new devices and materials, but also the possibility of designing them more effectively.
|
|
| 6. |
- Olsthoorn, Bart
(författare)
-
Homology and machine learning for materials informatics
- 2023
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Materials informatics is the field of study where materials science is combined with modern data science. This data-driven approach is powered by the growing availability of computational power and storage capability. The development and application of these methods accelerates materials science and represents an effective way to study and model material properties. This thesis is a compilation of theoretical and computational works that can be divided into three key areas: materials databases, machine learning for materials, and homology for materials.Machine learning and data mining rely on the availability of materials databases to test methods and models. The Organic Materials Database (OMDB), for example, contains a large number of organic crystals and their corresponding electronic structures. The electronic properties of the organic crystals are computed using atomic scale materials modelling, which is computationally expensive because organic crystals typically contain many atoms in the unit cell. However, the resulting data can be used in a variety of materials informatics applications. We demonstrate data mining for dark matter sensors as an example application.Accurate machine learning models can capture the structure-property relationship of materials and accelerate the discovery of new materials with desired properties. This is explored by investigating the properties of the organic crystals in the OMDB. For example, we employ supervised learning on the electronic band gap, an important material property for technological applications. Unsupervised learning is used to construct a dimensionality-reduced chemical space that reveals interesting clusters of materials.Finally, persistent homology is a relatively new method from the field of algebraic topology that studies the shapes that are present in data at different length scales. In this thesis, the method is used to study magnetic materials and their phase transitions. More specifically, in the case of classical models, we use persistent homology to detect the phase transition directly from sampled spin configurations. For quantum spin models, the shapes in the entanglement structure are captured and a sudden change reveals a quantum phase transition.In summary, these three topics provide an overview on how to study material properties with modern data science methods. The tools can be used in combination with the traditional methods in materials science and accelerate materials design.
|
|
