Data-driven prediction of in situ CO2 foam strength for enhanced oil recovery and carbon sequestration.

Carbon dioxide foam injection is a promising enhanced oil recovery (EOR) method, being at the same time an efficient carbon storage technology. The strength of CO2 foam under reservoir conditions plays a crucial role in predicting the EOR and sequestration performance, yet, controlling the strength of the foam is challenging due to the complex physics of foams and their sensitivity to operational conditions and reservoir parameters. Data-driven approaches for complex fluids such as foams can be an alternative method to the time-consuming experimental and conventional modeling techniques, which often fail to accurately describe the effect of all important related parameters. In this study, machine learning (ML) models were constructed to predict the oil-free CO2 foam apparent viscosity in the bulk phase and sandstone formations. Based on previous experimental data on various operational and reservoir conditions, predictive models were developed by employing six ML algorithms. Among the applied algorithms, neural network algorithms provided the most precise predictions for bulk and porous media. The established models were then used to compute the critical foam quality under different conditions and determine the maximum apparent foam viscosity, effectively controlling CO2 mobility to co-optimize EOR and CO2 sequestration.

Introducing artificial MOFs for improved machine learning predictions: Identification of top-performing materials for methane storage.

In the field of materials science, the main objective of predictive models is to provide scientists with reliable tools for fast and accurate identification of new materials with exceptional properties. Over the last few years, machine learning methods have been extensively used for the study of the gas-adsorption in nanoporous materials as an efficient alternative of molecular simulations and experiments. In several cases, the accuracy of the constructed predictive models for unknown materials is extremely high. In this study, we explored the adsorption of methane by metal organic frameworks (MOFs) and concluded that many top-performing materials often deviate significantly from the known materials used for the training of the machine learning algorithms. In such cases, the predictions of the machine learning algorithms may not be adequately accurate. For lack of the required appropriate data, we put forth a simple approach for the construction of artificial MOFs with the desired superior properties. Incorporation of such data during the training phase of the machine learning algorithms improves the predictions outstandingly. In some cases, over 96% of the unknown top-performing materials are successfully identified.

A Universal Machine Learning Algorithm for Large-Scale Screening of Materials.

Application of machine learning (ML) methods for the determination of the gas adsorption capacities of nanomaterials, such as metal-organic frameworks (MOF), has been extensively investigated over the past few years as a computationally efficient alternative to time-consuming and computationally demanding molecular simulations. Depending on the thermodynamic conditions and the adsorbed gas, ML has been found to provide very accurate results. In this work, we go one step further and we introduce chemical intuition in our descriptors by using the "type" of the atoms in the structure, instead of the previously used building blocks, to account for the chemical character of the MOF. ML predictions for the methane and carbon dioxide adsorption capacities of several tens of thousands of hypothetical MOFs are evaluated at various thermodynamic conditions using the random forest algorithm. For all cases examined, the use of atom types instead of building blocks leads to significantly more accurate predictions, while the number of MOFs needed for the training of the ML algorithm in order to achieve a specified accuracy can be reduced by an order of magnitude. More importantly, since practically there are an unlimited number of building blocks that materials can be made of but a limited number of atom types, the proposed approach is more general and can be considered as universal. The universality and transferability was proved by predicting the adsorption properties of a completely different family of materials after the training of the ML algorithm in MOFs.

A Robust Machine Learning Algorithm for the Prediction of Methane Adsorption in Nanoporous Materials.

In the present study, we propose a new set of descriptors that, along with a few structural features of nanoporous materials, can be used by machine learning algorithms for accurate predictions of the gas uptake capacities of these materials. All new descriptors closely resemble the helium atom void fraction of the material framework. However, instead of a helium atom, a particle with an appropriately defined van der Waals radius is used. The set of void fractions of a small number of these particles is found to be sufficient to characterize uniquely the structure of each material and to account for the most important topological features. We assess the accuracy of our approach by examining the predictions of the random forest algorithm in the relative small dataset of the computation-ready, experimental (CoRE) MOFs (∼4700 structures) that have been experimentally synthesized and whose geometrical/structural features have been accurately calculated before. We first performed grand canonical Monte Carlo simulations to accurately determine their methane uptake capacities at two different temperatures (280 and 298 K) and three different pressures (1, 5.8, and 65 bar). Despite the high chemical and structural diversity of the CoRE MOFs, it was found that the use of the proposed descriptors significantly improves the accuracy of the machine learning algorithm, particularly at low pressures, compared to the predictions made based solely on the rest structural features. More importantly, the algorithm can be easily adapted for other types of nanoporous materials beyond MOFs. Convergence of the predictions was reached even for small training set sizes compared to what was found in previous works using the hypothetical MOF database.

Evaluation of global simulations of aerosol particle and cloud condensation nuclei number, with implications for cloud droplet formation.

A total of 16 global chemistry transport models and general circulation models have participated in this study; 14 models have been evaluated with regard to their ability to reproduce the near-surface observed number concentration of aerosol particles and cloud condensation nuclei (CCN), as well as derived cloud droplet number concentration (CDNC). Model results for the period 2011-2015 are compared with aerosol measurements (aerosol particle number, CCN and aerosol particle composition in the submicron fraction) from nine surface stations located in Europe and Japan. The evaluation focuses on the ability of models to simulate the average across time state in diverse environments and on the seasonal and short-term variability in the aerosol properties. There is no single model that systematically performs best across all environments represented by the observations. Models tend to underestimate the observed aerosol particle and CCN number concentrations, with average normalized mean bias (NMB) of all models and for all stations, where data are available, of -24% and -35% for particles with dry diameters > 50 and > 120nm, as well as -36% and -34% for CCN at supersaturations of 0.2% and 1.0%, respectively. However, they seem to behave differently for particles activating at very low supersaturations (< 0.1 %) than at higher ones. A total of 15 models have been used to produce ensemble annual median distributions of relevant parameters. The model diversity (defined as the ratio of standard deviation to mean) is up to about 3 for simulated N3 (number concentration of particles with dry diameters larger than 3 nm) and up to about 1 for simulated CCN in the extra-polar regions. A global mean reduction of a factor of about 2 is found in the model diversity for CCN at a supersaturation of 0.2% (CCN0.2) compared to that for N3, maximizing over regions where new particle formation is important. An additional model has been used to investigate potential causes of model diversity in CCN and bias compared to the observations by performing a perturbed parameter ensemble (PPE) accounting for uncertainties in 26 aerosol-related model input parameters. This PPE suggests that biogenic secondary organic aerosol formation and the hygroscopic properties of the organic material are likely to be the major sources of CCN uncertainty in summer, with dry deposition and cloud processing being dominant in winter. Models capture the relative amplitude of the seasonal variability of the aerosol particle number concentration for all studied particle sizes with available observations (dry diameters larger than 50, 80 and 120 nm). The short-term persistence time (on the order of a few days) of CCN concentrations, which is a measure of aerosol dynamic behavior in the models, is underestimated on average by the models by 40% during winter and 20% in summer. In contrast to the large spread in simulated aerosol particle and CCN number concentrations, the CDNC derived from simulated CCN spectra is less diverse and in better agreement with CDNC estimates consistently derived from the observations (average NMB -13% and -22% for updraft velocities 0.3 and 0.6 ms-1, respectively). In addition, simulated CDNC is in slightly better agreement with observationally derived values at lower than at higher updraft velocities (index of agreement 0.64 vs. 0.65). The reduced spread of CDNC compared to that of CCN is attributed to the sublinear response of CDNC to aerosol particle number variations and the negative correlation between the sensitivities of CDNC to aerosol particle number concentration (∂N d/∂N a) and to updraft velocity (∂N d/∂w). Overall, we find that while CCN is controlled by both aerosol particle number and composition, CDNC is sensitive to CCN at low and moderate CCN concentrations and to the updraft velocity when CCN levels are high. Discrepancies are found in sensitivities ∂N d/∂N a and ∂N d/∂w; models may be predisposed to be too "aerosol sensitive" or "aerosol insensitive" in aerosol-cloud-climate interaction studies, even if they may capture average droplet numbers well. This is a subtle but profound finding that only the sensitivities can clearly reveal and may explain inter-model biases on the aerosol indirect effect.

A surprisingly simple correlation between the classical and quantum structural networks in liquid water.

Nuclear quantum effects in liquid water have profound implications for several of its macroscopic properties related to the structure, dynamics, spectroscopy, and transport. Although several of water's macroscopic properties can be reproduced by classical descriptions of the nuclei using interaction potentials effectively parameterized for a narrow range of its phase diagram, a proper account of the nuclear quantum effects is required to ensure that the underlying molecular interactions are transferable across a wide temperature range covering different regions of that diagram. When performing an analysis of the hydrogen-bonded structural networks in liquid water resulting from the classical (class) and quantum (qm) descriptions of the nuclei with two interaction potentials that are at the two opposite ends of the range in describing quantum effects, namely the flexible, pair-wise additive q-TIP4P/F, and the flexible, polarizable TTM3-F, we found that the (class) and (qm) results can be superimposed over the temperature range T = 250-350 K using a surprisingly simple, linear scaling of the two temperatures according to T(qm) = α T(class) + ΔT, where α = 0.99 and ΔT = -6 K for q-TIP4P/F and α = 1.24 and ΔT = -64 K for TTM3-F. This simple relationship suggests that the structural networks resulting from the quantum and classical treatment of the nuclei with those two very different interaction potentials are essentially similar to each other over this extended temperature range once a model-dependent linear temperature scaling law is applied.

An extension of Wolf's method for the treatment of electrostatic interactions: application to liquid water and aqueous solutions.

We report the results of molecular dynamics simulations, which indicate that, by carefully splitting the electrostatic interactions in short- and long-range contributions and employing the charge-neutralization method of Wolf, accurate predictions of various properties of liquid water and aqueous solutions can be achieved without the need for the Ewald summation. In order to assess the accuracy of the proposed approach, several molecular dynamics simulations under different thermodynamic conditions are performed, employing various rigid, flexible, pairwise additive, and many-body polarizable water models. The predictions of the new approach are compared to the benchmark results obtained with the Ewald summation. It is found that while in the new approach there are no adjustable parameters, such as a damping parameter, the obtained results are more accurate than the results of similar approaches that are based on the Wolf method, while at the same time less or no additional computational effort is required. It is also concluded that the error of the results is smaller or at least comparable to the statistical error of a typical molecular dynamics simulation.

Simulating liquid water for determining its structural and transport properties.

Molecular dynamics simulations are carried out for calculating structural and transport properties of pure liquid water, such as radial distribution functions and self-diffusion and viscosity coefficients, respectively. We employed reparameterized versions of the ab initio water potential by Niesar, Clementi and Corongiu (NCC). In order to investigate the role of the electrostatic contribution, the partial charges of the NCC model are adjusted so that to reproduce the dipole moment values of the SPC/E, SPC/Fw and TIP4P/2005 water models. The single and collective transport coefficients are obtained by employing the Green-Kubo relations at various temperatures. Additionally, in order to overcome convergence difficulties arising from the long correlation times of the stress-tensor autocorrelation functions, a previously reported fitting scheme was employed. The present results indicate that there is a significant relationship between the dipole moment value of the model, and the calculated transport coefficients. We found that by adjusting the molecular dipole moment of the NCC to the value of the TIP4P/2005, the obtained values for the self-diffusion and viscosity coefficients are in better agreement with experiment, compared to the values obtained with the original NCC model. Even though the predictions of the present model exhibits an overall correct behavior, we conclude that further improvements are still required. In order to achieve that, a careful reparameterization of the repulsion-dispersion terms of the potential model is proposed. Also, the effect of the inclusion of many-body effects such as polarizability, should also be investigated.

Determining the bulk viscosity of rigid water models.

We use equilibrium molecular dynamics methods to compute the shear and bulk viscosities of the pairwise additive and rigid SPC/E, TIP4P, and TIP4P/2005 water models. For the latter model it was found in a recent study (J. Chem. Phys. 2009, 131, 246101) an excellent agreement with experiment in the prediction of the shear viscosity over a range of different thermodynamic conditions. Here, we examine, for a wide range of temperatures, whether this remarkable accuracy of the TIP4P/2005 model remains in the prediction of the bulk viscosity. Moreover, we examine whether equilibrium molecular dynamics methods provide reasonable accuracy in the calculation of the bulk viscosity, as it was previously found for the shear viscosity (J. Chem. Phys. 2010, 132, 096101). We concluded that, by performing the appropriate data analysis, accurate estimates of the bulk viscosity can be obtained, while, compared to the other simple rigid/pairwise additive water models, the predictions of the TIP4P/2005 model for the bulk viscosity are significantly closer to the experiment.

Insights in quantum dynamical effects in the infrared spectroscopy of liquid water from a semiclassical study with an ab initio-based flexible and polarizable force field.

The dynamical properties of liquid water play an important role in many processes in nature. In this paper, we focus on the infrared (IR) absorption spectrum of liquid water based on the linearized semiclassical initial value representation (LSC-IVR) with the local Gaussian approximation (LGA) [J. Liu and W. H. Miller, J. Chem. Phys. 131, 074113 (2009)] and an ab initio based, flexible, polarizable Thole-type model (TTM3-F) [G. S. Fanourgakis and S. S. Xantheas, J. Chem. Phys. 128, 074506 (2008)]. Although the LSC-IVR (LGA) gives the exact result for the isolated three-dimensional shifted harmonic stretching model, it yields a blueshifted peak position for the more realistic anharmonic stretching potential. By using the short-time information of the LSC-IVR correlation function; however, it is shown how one can obtain more accurate results for the position of the stretching peak. Due to the physical decay in the condensed phase system, the LSC-IVR (LGA) is a good and practical approximate quantum approach for the IR spectrum of liquid water. The present results offer valuable insight into future attempts to improve the accuracy of the TTM3-F potential or other ab initio-based models in reproducing the IR spectrum of liquid water.

A fast path integral method for polarizable force fields.

A quantum simulation of an imaginary time path integral typically requires around n times more computational effort than the corresponding classical simulation, where n is the number of ring polymer beads (or imaginary time slices) used in the calculation. It is however possible to improve on this estimate by decomposing the potential into a sum of slowly and rapidly varying contributions. If the slowly varying contribution changes only slightly over the length scale of the ring polymer, it can be evaluated on a contracted ring polymer with fewer than the full n beads (or equivalently on a lower order Fourier decomposition of the imaginary time path). Here we develop and test this idea for systems with polarizable force fields. The development consists of iterating the induction on the contracted ring polymer and applying an appropriate transformation to obtain the forces on the original n beads. In combination with a splitting of the Coulomb potential into its short- and long-range parts, this results in a method with little more than classical computational effort in the limit of large system size. The method is illustrated with simulations of liquid water at 300 K and hexagonal ice at 100 K using a recently developed flexible and polarizable Thole-type potential energy model.

Comparison of path integral molecular dynamics methods for the infrared absorption spectrum of liquid water.

The ring polymer molecular dynamics (RPMD) and partially adiabatic centroid molecular dynamics (PA-CMD) methods are compared and contrasted in an application to the infrared absorption spectrum of a recently parametrized flexible, polarizable, Thole-type potential energy model for liquid water. Both methods predict very similar spectra in the low-frequency librational and intramolecular bending region at wavenumbers below 2500 cm(-1). However, the RPMD spectrum is contaminated in the high-frequency O-H stretching region by contributions from the internal vibrational modes of the ring polymer. This problem is avoided in the PA-CMD method, which adjusts the elements of the Parrinello-Rahman mass matrix so as to shift the frequencies of these vibrational modes beyond the spectral range of interest. PA-CMD does not require any more computational effort than RPMD and it is clearly the better of the two methods for simulating vibrational spectra.

Development of transferable interaction potentials for water. V. Extension of the flexible, polarizable, Thole-type model potential (TTM3-F, v. 3.0) to describe the vibrational spectra of water clusters and liquid water.

We present a new parametrization of the flexible, polarizable Thole-type model for water [J. Chem. Phys. 116, 5115 (2002); J. Phys. Chem. A 110, 4100 (2006)], with emphasis in describing the vibrational spectra of both water clusters and liquid water. The new model is able to produce results of similar quality with the previous versions for the structures and energetics of water clusters as well as structural and thermodynamic properties of liquid water evaluated with classical and converged quantum statistical mechanical atomistic simulations. At the same time it yields accurate redshifts for the OH vibrational stretches of both water clusters and liquid water.

The bend angle of water in ice Ih and liquid water: The significance of implementing the nonlinear monomer dipole moment surface in classical interaction potentials.

The implementation of the physically accurate nonlinear dipole moment surface of the water monomer in the context of the Thole-type, polarizable, flexible interaction potential results in the only classical potential, which, starting from the gas phase value for the bend angle (104.52 degrees), reproduces its experimentally observed increase in the ice Ih lattice and in liquid water. This is in contrast to all other classical potentials to date, which predict a decrease of the monomer bend angle in ice Ih and in liquid water with respect to the gas phase monomer value. Simulations under periodic boundary conditions of several supercells consisting of up to 288 molecules of water used to sample the proton disorder in the ice Ih lattice yield an average value of vartheta(HOH)(I(h))=108.4 degrees +/-0.2 degrees for the minimized structures (T=0 K) and 108.1 degrees +/-2.8 degrees at T=100 K. Analogous simulations for liquid water predict an average value of vartheta(HOH)(liquid)=106.3 degrees +/-4.9 degrees at T=300 K. The increase of the monomer bend angle of water in condensed environments is attributed to the use of geometry-dependent charges that are used to describe the nonlinear character of the monomer's dipole moment surface. Our results suggest a new paradigm in the development of classical interaction potential models of water that can be used to describe condensed aqueous environments.

The flexible, polarizable, thole-type interaction potential for water (TTM2-F) revisited.

We present a revision of the flexible, polarizable, Thole-type interaction potential for water [J. Chem. Phys.2002, 116, 5115], which allows for condensed-phase simulations. The revised version (TTM2.1-F) of the potential correctly describes the individual water molecular dipole moment and alleviates problems arising at short intermolecular separations that can be sampled in the course of molecular dynamics and Monte Carlo simulations of condensed environments. Furthermore, its parallel implementation under periodic boundary conditions enables the efficient calculation of the macroscopic structural and thermodynamic properties of liquid water, as its performance scales superlinearly with up to a number of 64 processors for a simulation box of 512 molecules. We report the radial distribution functions, average energy, internal geometry, and dipole moment in the liquid as well as the density, dielectric constant, and self-diffusion coefficient at T = 300 K from (NVT) and (NPT) classical molecular dynamics simulations by using the revised version of the potential.

The spectroscopic signature of the "all-surface" to "internally solvated" structural transition in water clusters in the n = 17-21 size regime.

The existence of a transitional size regime where preferential stabilization alternates between "all-surface" (all atoms on the surface of a cluster) and "internally solvated" (one water molecule at the center of the cluster, fully solvated) configurations with the addition or the removal of a single water molecule, predicted earlier with the flexible, polarizable (many-body) Thole-type model interaction potential (TTM2-F), has been confirmed from electronic structure calculations for (H2O)n, n = 17-21. The onset of the appearance of the first "interior" configuration in water clusters occurs for n = 17. The observed structural alternation between interior (n = 17, 19, 21) and all-surface (n = 18, 20) global minima in the n = 17-21 cluster regime is accompanied by a corresponding spectroscopic signature, namely, the undulation in the position of the most redshifted OH stretching vibrations according to the trend: interior configurations exhibit more redshifted OH stretching vibrations than all-surface ones. These most redshifted OH stretching vibrations form distinct groups in the intramolecular region of the spectra and correspond to localized vibrations of donor OH stretches that are connected to neighbors via "strong" (water dimer-like) hydrogen bonds and belong to a water molecule with a "free" OH stretch.

High-level ab initio calculations for the four low-lying families of minima of (H2O)20. II. Spectroscopic signatures of the dodecahedron, fused cubes, face-sharing pentagonal prisms, and edge-sharing pentagonal prisms hydrogen bonding networks.

We report the first harmonic vibrational spectra for each of the lowest lying isomers within the four major families of minima of (H2O)20, namely, the dodecahedron, fused cubes, face-sharing pentagonal prisms, and edge-sharing pentagonal prisms. These were obtained at the second-order Moller-Plesset perturbation level of theory (MP2) with the augmented correlation consistent basis set of double zeta quality (aug-cc-pVDZ) at the corresponding minimum energy geometries. The computed infrared (IR) spectra are the first ones obtained from first principles for these clusters. They were found to contain spectral features, which can be directly mapped onto the distinctive spectroscopic signatures of their constituent tetramer, pentamer, and octamer fragments. The dodecahedron spectra show the richest structure in the OH stretching region and are associated with the most redshifted OH vibrations with respect to the monomer. The lowest lying edge-sharing pentagonal prism isomer displays intense IR active vibrations that are redshifted by approximately 600 cm(-1) with respect to the water monomer. Furthermore the most redshifted, IR-active OH stretching vibrations for all four networks correspond to hydrogen bonded OH groups, which exhibit the following two common characteristics: (i) they belong to fragments which have a "free" OH stretch and (ii) they act as donors to a neighboring water molecule along a "dimerlike" (strong) hydrogen bond. The zero-point energy corrected MP2/CBS (complete basis set) limit binding energies D(0) for the four isomers are -163.1 kcal/mol (edge-sharing pentagonal prism), -160.1 kcal/mol (face-sharing pentagonal prism), -157.5 kcal/mol (fused cubes), and -148.1 kcal/mol (dodecahedron).

High-level ab initio calculations for the four low-lying families of minima of (H2O)20. I. Estimates of MP2/CBS binding energies and comparison with empirical potentials.

We report estimates of complete basis set (CBS) limits at the second-order Møller-Plesset perturbation level of theory (MP2) for the binding energies of the lowest-lying isomers within each of the four major families of minima of (H(2)O)(20). These were obtained by performing MP2 calculations with the family of correlation-consistent basis sets up to quadruple zeta quality, augmented with additional diffuse functions (aug-cc-pVnZ, n=D, T, Q). The MP2/CPS estimates are -200.1 (dodecahedron, 30 hydrogen bonds), -212.6 (fused cubes, 36 hydrogen bonds), -215.0 (face-sharing pentagonal prisms, 35 hydrogen bonds), and -217.9 kcal/mol (edge-sharing pentagonal prisms, 34 hydrogen bonds). The energetic ordering of the various (H(2)O)(20) isomers does not follow monotonically the number of hydrogen bonds as in the case of smaller clusters such as the different isomers of the water hexamer. The dodecahedron lies ca. 18 kcal/mol higher in energy than the most stable edge-sharing pentagonal prism isomer. The TIP4P, ASP-W4, TTM2-R, AMOEBA, and TTM2-F empirical potentials also predict the energetic stabilization of the edge-sharing pentagonal prisms with respect to the dodecahedron, albeit they universally underestimate the cluster binding energies with respect to the MP2/CBS result. Among them, the TTM2-F potential was found to predict the absolute cluster binding energies to within <1% from the corresponding MP2/CBS values, whereas the error for the rest of the potentials considered in this study ranges from 3% to 5%.