Steep Scaling Research Articles

Local Energy Decomposition Analysis of London Dispersion Effects: From Simple Model Dimers to Complex Biomolecular Assemblies.

ConspectusLondon dispersion (LD) forces are ubiquitous in chemistry, playing a pivotal role in a wide range of chemical processes. For example, they influence the structure of molecular crystals, the selectivity of organocatalytic transformations, and the formation of biomolecular assemblies. Harnessing these forces for chemical applications requires consistent quantification of the LD energy across a broad and diverse spectrum of chemical scenarios. Despite the great progress made in recent years in the development of experimental strategies for LD quantification, quantum chemical methods remain one of the most useful tools in the hand of chemists for the study of these weak interactions. Unfortunately, the accurate quantification of LD effects in complex systems poses many challenges for electronic structure theories. One of the problems stems from the fact that LD forces originate from long-range electronic dynamic correlation, and hence, their rigorous description requires the use of complex, highly correlated wave function-based methods. These methods typically feature a steep scaling with the system size, limiting their applicability to small model systems. Another core challenge lies in disentangling short-range from long-range dynamic correlation, which from a rigorous quantum mechanical perspective is not possible.In this Account, we describe our research endeavors in the development of broadly applicable computational methods for LD quantification in molecular chemistry as well as challenging applications of these schemes in various domains of chemical research. Our strategy lies in the use of local correlation theories to reduce the computational cost associated with complex electronic structure methods while providing at the same time a simple means of decomposition of dynamic correlation into its long-range and short-range components. In particular, the local energy decomposition (LED) scheme at the domain-based local pair natural orbital coupled cluster (DLPNO-CCSD(T)) level has emerged as a powerful tool in our research, offering a clear-cut quantitative definition of the LD energy that remains valid across a plethora of different chemical scenarios. Typical applications of this scheme are examined, encompassing protein-ligand interactions and reactivity studies involving many fragments and complex electronic structures. In addition, our research also involves the development of novel cost-effective methodologies, which exploit the LED definition of the LD energy, for LD energy quantification that are, in principle, applicable to systems with thousands of atoms. The Hartree-Fock plus London Dispersion (HFLD) scheme, correcting the HF interaction energy using an approximate CCSD(T)-based LD energy, is a useful, parameter-free electronic structure method for the study of LD effects in systems with hundreds of molecular fragments. However, the usefulness of the LED scheme reaches beyond providing an interpretation of the calculated DLPNO-CCSD(T) or DLPNO-MP2 interaction energies. For example, the dispersion energies obtained from the LED can be fruitfully used in order to parametrize semiempirical dispersion models. We will demonstrate this in the context of an emerging semiempirical method, namely, the Natural Orbital Tied Constructed Hamiltonian (NOTCH) method. NOTCH incorporates LED-derived LD energies and shows promising accuracy at a minimum amount of empiricism. Thus, it holds substantial promise for large and complex systems.

Building an ab initio solvated DNA model using Euclidean neural networks.

Accurately modeling large biomolecules such as DNA from first principles is fundamentally challenging due to the steep computational scaling of ab initio quantum chemistry methods. This limitation becomes even more prominent when modeling biomolecules in solution due to the need to include large numbers of solvent molecules. We present a machine-learned electron density model based on a Euclidean neural network framework that includes a built-in understanding of equivariance to model explicitly solvated double-stranded DNA. By training the machine learning model using molecular fragments that sample the key DNA and solvent interactions, we show that the model predicts electron densities of arbitrary systems of solvated DNA accurately, resolves polarization effects that are neglected by classical force fields, and captures the physics of the DNA-solvent interaction at the ab initio level.

Active sparse Bayesian committee machine potential for isothermal-isobaric molecular dynamics simulations.

Recent advancements in machine learning potentials (MLPs) have significantly impacted the fields of chemistry, physics, and biology by enabling large-scale first-principles simulations. Among different machine learning approaches, kernel-based MLPs distinguish themselves through their ability to handle small datasets, quantify uncertainties, and minimize over-fitting. Nevertheless, their extensive computational requirements present considerable challenges. To alleviate these, sparsification methods have been developed, aiming to reduce computational scaling without compromising accuracy. In the context of isothermal and isobaric ML molecular dynamics (MD) simulations, achieving precise pressure estimation is crucial for reproducing reliable system behavior under constant pressure. Despite progress, sparse kernel MLPs struggle with precise pressure prediction. Here, we introduce a virial kernel function that significantly enhances the pressure estimation accuracy of MLPs. Additionally, we propose the active sparse Bayesian committee machine (BCM) potential, an on-the-fly MLP architecture that aggregates local sparse Gaussian process regression (SGPR) MLPs. The sparse BCM potential overcomes the steep computational scaling with the kernel size, and a predefined restriction on the size of kernel allows for fast and efficient on-the-fly training. Our advancements facilitate accurate and computationally efficient machine learning-enhanced MD (MLMD) simulations across diverse systems, including ice-liquid coexisting phases, Li10Ge(PS6)2 lithium solid electrolyte, and high-pressure liquid boron nitride.

A 3D Numerical Study of Anisotropies in Supernova Remnants

We develop a suite of 3D hydrodynamic models of supernova remnants (SNRs) expanding against the circumstellar medium (CSM). We study the Rayleigh–Taylor instability forming at the expansion interface by calculating an angular power spectrum for each of these models. The power spectra of young SNRs are seen to exhibit a dominant angular mode, which is a diagnostic of their ejecta density profile as found by previous studies. The steep scaling of power at smaller modes and the time evolution of the spectra are indicative of the absence of a turbulent cascade. Instead, as the time evolution of the spectra suggests, they may be governed by an angular mode-dependent net growth rate. We also study the impact of anisotropies in the ejecta and in the CSM on the power spectra of velocity and density. We confirm that perturbations in the density field (whether imposed on the ejecta or the CSM) do not influence the anisotropy of the remnant significantly unless they have a very large amplitude and form large-scale coherent structures. In any case, these clumps can only affect structures on large angular scales. The power spectrum on small angular scales is completely independent of the initial clumpiness and governed only by the growth and saturation of the Rayleigh–Taylor instability.

Tensor-Reduced Atomic Density Representations.

Density-based representations of atomic environments that are invariant under Euclidean symmetries have become a widely used tool in the machine learning of interatomic potentials, broader data-driven atomistic modeling, and the visualization and analysis of material datasets. The standard mechanism used to incorporate chemical element information is to create separate densities for each element and form tensor products between them. This leads to a steep scaling in the size of the representation as the number of elements increases. Graph neural networks, which do not explicitly use density representations, escape this scaling by mapping the chemical element information into a fixed dimensional space in a learnable way. By exploiting symmetry, we recast this approach as tensor factorization of the standard neighbour-density-based descriptors and, using a new notation, identify connections to existing compression algorithms. In doing so, we form compact tensor-reduced representation of the local atomic environment whose size does not depend on the number of chemical elements, is systematically convergable, and therefore remains applicable to a wide range of data analysis and regression tasks.

Open-Shell Tensor Hypercontraction

The extension of least-squares tensor hypercontracted second- and third-order Møller-Plesset perturbation theory (LS-THC-MP2 and LS-THC-MP3) to open-shell systems is an important development due to the scaling reduction afforded by THC and the ubiquity of molecular ions, radicals, and other open-shell reactive species. The complexity of wavefunction-based quantum-chemical methods such as Møller-Plesset and coupled cluster theory is reflected in the steep scaling of the computational costs with the molecular size. The least-squares tensor hypercontraction (LS-THC) method is an efficient, single-step factorization for the two-electron integral tensor but can also be used to factorize the double excitation amplitudes, leading to significant scaling reduction. Here we extend this promising method to open-shell variants of LS-THC-MP2 and -MP3 by using diagrammatic techniques and explicit spin summation. The accuracy of the resulting methods for open-shell species is benchmarked on standard test systems such as regular alkanes as well as realistic systems involving bond breaking, radical stabilization, and other effects. We find that open-shell LS-THC-MPn methods exhibit errors highly comparable to those produced by closed-shell LS-THC-MPn and are highly insensitive to particular chemical interactions, geometries, or even moderate spin contamination.

Natural virtual orbitals for the GW method in the random-phase approximation and beyond.

The increasingly popular GW method is becoming a convenient tool to determine vertical ionization energies in molecular systems. However, depending on the formalism used and the range of orbitals investigated, it may be hampered by a steep computational scaling. To alleviate this issue, correlated natural virtual orbitals (NVOs) based on second-order Møller-Plesset (MP2) and direct MP2 correlation energies are implemented, and the resulting correlated NVOs are tested on GW quasiparticle energies. Test cases include the popular GW variants G0W0 and evGW0 as well as more elaborate vertex corrections. We find that for increasingly larger molecular systems and basis sets, NVOs considerably improve efficiency. Furthermore, we test the performance of the truncated (frozen) NVO ansatz on the GW100 test set. For the latter, it is demonstrated that, using a carefully chosen truncation threshold, NVOs lead to a negligible loss in accuracy while providing speedups of one order of magnitude. Furthermore, we compare the resulting quasiparticle energies to very accurate vertical ionization energies obtained from coupled-cluster theory with singles, doubles, and noniterative triples [CCSD(T)], confirming that the loss in accuracy introduced by truncating the NVOs is negligible compared to the methodical errors in the GW approximation. It is also demonstrated that the choice of basis set impacts results far more than using a suitably truncated NVO space. Therefore, at the same computational expense, more accurate results can be obtained using NVOs. Finally, we provide improved reference CCSD(T) values for the GW100 test set, which have been obtained using the def2-QZVPP basis set.

Multi-site reaction dynamics through multi-fragment density matrix embedding.

The practical description of disordered chemical reactions, where the reactions involve multiple species at multiple sites, is presently a challenge using correlated electronic structure methods due to their high computational cost and steep scaling. Here, we describe the gradient theory of multi-fragment density matrix embedding theory, which potentially provides a minimal computational framework to model such processes at the correlated electron level. We present the derivation and implementation of the gradient theory, its validation on model systems and chemical reactions using density matrix embedding, and its application to a molecular dynamics simulation of proton transport in a small water cluster, a simple example of multi-site reaction dynamics.

Quantitative Prediction of Vertical Ionization Potentials from DFT via a Graph-Network-Based Delta Machine Learning Model Incorporating Electronic Descriptors.

While accurate wave function theories like CCSD(T) are capable of modeling molecular chemical processes, the associated steep computational scaling renders them intractable for treating large systems or extensive databases. In contrast, density functional theory (DFT) is much more computationally feasible yet often fails to quantitatively describe electronic changes in chemical processes. Herein, we report an efficient delta machine learning (ΔML) model that builds on the Connectivity-Based Hierarchy (CBH) scheme─an error correction approach based on systematic molecular fragmentation protocols─and achieves coupled cluster accuracy on vertical ionization potentials by correcting for deficiencies in DFT. The present study integrates concepts from molecular fragmentation, systematic error cancellation, and machine learning. First, we show that by using an electron population difference map, ionization sites within a molecule may be readily identified, and CBH correction schemes for ionization processes may be automated. As a central feature of our work, we employ a graph-based QM/ML model, which embeds atom-centered features describing CBH fragments into a computational graph to further increase accuracy for the prediction of vertical ionization potentials. In addition, we show that the incorporation of electronic descriptors from DFT, namely electron population difference features, improves model performance well beyond chemical accuracy (1 kcal/mol) to approach benchmark accuracy. While the raw DFT results are strongly dependent on the underlying functional used, for our best models, the performance is robust and much less dependent on the functional.

Using Anisotropies as a Forensic Tool for Decoding Supernova Remnants

We present a method for analyzing supernova remnants (SNRs) by diagnosing the drivers responsible for structure at different angular scales. First, we perform a suite of hydrodynamic models of the Rayleigh–Taylor instability (RTI) as a supernova (SN) collides with its surrounding medium. Using these models we demonstrate how power spectral analysis can be used to attribute which scales in an SNR are driven by RTI and which must be caused by intrinsic asymmetries in the initial explosion. We predict the power spectrum of turbulence driven by RTI and identify a dominant angular mode that represents the largest scale that efficiently grows via RTI. We find that this dominant mode relates to the density scale height in the ejecta, and therefore reveals the density profile of the SN ejecta. If there is significant structure in an SNR on angular scales larger than this mode, then it is likely caused by anisotropies in the explosion. Structure on angular scales smaller than the dominant mode exhibits a steep scaling with wavenumber, possibly too steep to be consistent with a turbulent cascade, and therefore might be determined by the saturation of RTI at different length scales (although systematic 3D studies are needed to investigate this). We also demonstrate, consistent with previous studies, that this power spectrum is independent of the magnitude and length scales of perturbations in the surrounding medium and therefore this diagnostic is unaffected by “clumpiness” in the circumstellar medium.

Pushing the limits: Efficient wavefunction methods for excited states in complex systems using frozen-density embedding.

Frozen density embedding (FDE) is an embedding method for complex environments that is simple for users to set up. It reduces the computation time by dividing the total system into small subsystems and approximating the interaction by a functional of their densities. Its combination with wavefunction methods is, however, limited to small- or medium-sized molecules because of the steep scaling in computation time of these methods. To mitigate this limitation, we present a combination of the FDE approach with pair natural orbitals (PNOs) in the TURBOMOLE software package. It combines the uncoupled FDE (FDEu) approach for excitation energy calculations with efficient implementations of second-order correlation methods in the ricc2 and pnoccsd programs. The performance of this combination is tested for tetraazaperopyrene (TAPP) molecular crystals. It is shown that the PNO truncation error on environment-induced shifts is significantly smaller than the shifts themselves and, thus, that the local approximations of PNO-based wavefunction methods can without the loss of relevant digits be combined with the FDE method. Computational wall times are presented for two TAPP systems. The scaling of the wall times is compared to conventional supermolecular calculations and demonstrates large computational savings for the combination of FDE- and PNO-based methods. Additionally, the behavior of excitation energies with the system size is investigated. It is found that the excitation energies converge quickly with the size of the embedding environment for the TAPPs investigated in the current study.

Exchanging Replicas with Unequal Cost, Infinitely and Permanently.

We developed a replica exchange method that is effectively parallelizable even if the computational cost of the Monte Carlo moves in the parallel replicas are considerably different, for instance, because the replicas run on different types of processor units or because of the algorithmic complexity. To prove detailed-balance, we make a paradigm shift from the common conceptual viewpoint in which the set of parallel replicas represents a high-dimensional superstate, to an ensemble-based criterion in which the other ensembles represent an environment that might or might not participate in the Monte Carlo move. In addition, based on a recent algorithm for computing permanents, we effectively increase the exchange rate to infinite without the steep factorial scaling as a function of the number of replicas. We illustrate the effectiveness of this replica exchange methodology by combining it with a quantitative path sampling method, replica exchange transition interface sampling (RETIS), in which the costs for a Monte Carlo move can vary enormously as paths in a RETIS algorithm do not have the same length and the average path lengths tend to vary considerably for the different path ensembles that run in parallel. This combination, coined ∞RETIS, was tested on three model systems.

Bubble-driven gas uplift in galaxy clusters and its velocity features

ABSTRACT Buoyant bubbles of relativistic plasma are essential for active galactic nucleus feedback in galaxy clusters, stirring and heating the intracluster medium (ICM). Observations suggest that these rising bubbles maintain their integrity and sharp edges much longer than predicted by hydrodynamic simulations. In this study, we assume that bubbles can be modelled as rigid bodies and demonstrate that intact bubbles and their long-term interactions with the ambient ICM play an important role in shaping gas kinematics, forming thin gaseous structures (e.g. H α filaments), and generating internal waves in cluster cores. We find that well-developed eddies are formed in the wake of a buoyantly rising bubble, and it is these eddies, rather than the Darwin drift, that are responsible for most of the gas mass uplift. The eddies gradually elongate along the bubble’s direction of motion due to the strong density stratification of the atmosphere and eventually detach from the bubble, quickly evolving into a high-speed jet-like stream propagating towards the cluster center in our model. This picture naturally explains the presence of long straight and horseshoe-shaped H α filaments in the Perseus cluster, inward and outward motions of the gas, and the X-ray-weighted gas velocity distributions near the northwestern bubble observed by Hitomi. Our model reproduces the observed H α velocity structure function of filaments, providing a simple interpretation for its steep scaling and normalization: laminar gas flows and large eddies within filaments driven by the intact bubbles, rather than spatially homogeneous small-scale turbulence, are sufficient to produce a structure function consistent with observations.

Representation of Diabatic Potential Energy Matrices for Multiconfiguration Time-Dependent Hartree Treatments of High-Dimensional Nonadiabatic Photodissociation Dynamics.

Conventional quantum mechanical characterization of photodissociation dynamics is restricted by steep scaling laws with respect to the dimensionality of the system. In this work, we examine the applicability of the multi-configurational time-dependent Hartree (MCTDH) method in treating nonadiabatic photodissociation dynamics in two prototypical systems, taking advantage of its favorable scaling laws. To conform to the sum-of-product form, elements of the ab initio diabatic potential energy matrix (DPEM) are re-expressed using the recently proposed Monte Carlo canonical polyadic decomposition method, with enforcement of proper symmetry. The MCTDH absorption spectra and product branching ratios are shown to compare well with those calculated using conventional grid-based methods, demonstrating its promise for treating high-dimensional nonadiabatic photodissociation problems.

Fast Predictions of Reaction Barrier Heights: Toward Coupled-Cluster Accuracy.

Quantitative estimates of reaction barriers are essential for developing kinetic mechanisms and predicting reaction outcomes. However, the lack of experimental data and the steep scaling of accurate quantum calculations often hinder the ability to obtain reliable kinetic values. Here, we train a directed message passing neural network on nearly 24,000 diverse gas-phase reactions calculated at CCSD(T)-F12a/cc-pVDZ-F12//ωB97X-D3/def2-TZVP. Our model uses 75% fewer parameters than previous studies, an improved reaction representation, and proper data splits to accurately estimate performance on unseen reactions. Using information from only the reactant and product, our model quickly predicts barrier heights with a testing MAE of 2.6 kcal mol-1 relative to the coupled-cluster data, making it more accurate than a good density functional theory calculation. Furthermore, our results show that future modeling efforts to estimate reaction properties would significantly benefit from fine-tuning calibration using a transfer learning technique. We anticipate this model will accelerate and improve kinetic predictions for small molecule chemistry.

Growth and Mortality as Causes of Variation in Metabolic Scaling Among Taxa and Taxonomic Levels.

Metabolic rate (MR) usually changes (scales) out of proportion to body mass (BM) as MR=aBMb, where a is a normalisation constant and b is the scaling exponent that reflects how steep this change is. This scaling relationship is fundamental to biology, but over a century of research has provided little consensus on the value of b, and why it appears to vary among taxa and taxonomic levels. By analysing published data on fish and taking an individual-based approach to metabolic scaling, I show that variation in growth of fish under naturally restricted food availability can explain variation in within-individual (ontogenetic) b for standard (maintenance) metabolic rate (SMR) of brown trout (Salmo trutta), with the fastest growers having the steepest metabolic scaling (b≈1). Moreover, I show that within-individual b can vary much more widely than previously assumed from work on different individuals or different species, from -1 to 1 for SMR among individual brown trout. The negative scaling of SMR for some individuals was caused by reductions in metabolic rate in a food limited environment, likely to maintain positive growth. This resulted in a mean within-individual b for SMR that was significantly lower than the across-individual ("static") b, a difference that also existed for another species, cunner (Tautogolabrus adspersus). Interestingly, the wide variation in ontogenetic b for SMR among individual brown trout did not exist for maximum (active) metabolic rate (MMR) of the same fish, showing that these two key metabolic traits (SMR and MMR) can scale independently of one another. I also show that across-species ("evolutionary") b for SMR of 134 fishes is significantly steeper (b approaching 1) than the mean ontogenetic b for the brown trout and cunner. Based on these interesting findings, I hypothesise that evolutionary and static metabolic scaling can be systematically different from ontogenetic scaling, and that the steeper evolutionary than ontogenetic scaling for fishes arises as a by-product of natural selection for fast-growing individuals with steep metabolic scaling (b≈1) early in life, where size-selective mortality is high for fishes. I support this by showing that b for SMR tends to increase with natural mortality rates of fish larvae within taxa.

Quantum Wave Packet Treatment of Cold Nonadiabatic Reactive Scattering at the State-To-State Level.

Cold and ultracold collisions are dominated by quantum effects, such as resonances, tunneling, and nonadiabatic transitions between different electronic states. Due to the extremely long de Broglie wavelength in such processes, quantum reactive scattering is most conveniently characterized using the time-independent close-coupling (TICC) methods. However, the TICC approach is difficult for systems with a large number of channels because of its steep numerical scaling laws. Here, a recently proposed quantum wave packet (WP) approach for solving adiabatic reactive scattering problems at low collision energies is extended to include nonadiabatic transitions. To impose the outgoing boundary conditions, the total scattering wavefunction is split into three parts, the interaction, the asymptotic, and the long-range regions. Each region is associated with a different set of basis functions, which could be optimized separately. In this way, an extremely long grid can be used to accommodate the characteristic long de Broglie wavelengths in the scattering coordinate. The better numerical scaling laws of the WP approach have the potential for handling larger nonadiabatic reactive systems at low temperatures in the future.

Mapping Quantum Chemical Dynamics Problems to Spin-Lattice Simulators.

The accurate computational determination of chemical, materials, biological, and atmospheric properties has a critical impact on a wide range of health and environmental problems, but is deeply limited by the computational scaling of quantum mechanical methods. The complexity of quantum chemical studies arises from the steep algebraic scaling of electron correlation methods and the exponential scaling in studying nuclear dynamics and molecular flexibility. To date, efforts to apply quantum hardware to such quantum chemistry problems have focused primarily on electron correlation. Here, we provide a framework that allows for the solution of quantum chemical nuclear dynamics by mapping these to quantum spin-lattice simulators. Using the example case of a short-strong hydrogen-bonded system, we construct the Hamiltonian for the nuclear degrees of freedom on a single Born-Oppenheimer surface and show how it can be transformed to a generalized Ising model Hamiltonian. We then demonstrate a method to determine the local fields and spin-spin couplings needed to identically match the molecular and spin-lattice Hamiltonians. We describe a protocol to determine the on-site and intersite coupling parameters of this Ising Hamiltonian from the Born-Oppenheimer potential and nuclear kinetic energy operator. Our approach represents a paradigm shift in the methods used to study quantum nuclear dynamics, opening the possibility to solve both electronic structure and nuclear dynamics problems using quantum computing systems.

A DLPNO-CCSD(T) benchmarking study of intermolecular interactions of ionic liquids.

The accuracy of correlation energy recovered by coupled cluster single-, double-, and perturbative triple-excitations, CCSD(T), has led to the method being considered the gold standard of computational chemistry. The application of CCSD(T) has been limited to medium-sized molecular systems due to its steep scaling with molecular size. The recent development of alternative domain-based local pair natural orbital coupled-cluster method, DLPNO-CCSD(T), has significantly broadened the range of chemical systems to which CCSD(T) level calculations can be applied. Condensed systems such as ionic liquids (ILs) have a large contribution from London dispersion forces of up to 150 kJ mol-1 in large-scale clusters. Ionic liquids show appreciable charge transfer effects that result in the increased valence orbital delocalization over the entire ionic network, raising the question whether the application of methods based on localized orbitals is reliable for these semi-Coulombic materials. Here the performance of DLPNO-CCSD(T) is validated for the prediction of correlation interaction energies of two data sets incorporating single-ion pairs of protic and aprotic ILs. DLPNO-CCSD(T) produced results within chemical accuracy with tight parameter settings and a non-iterative treatment of triple excitations. To achieve spectroscopic accuracy of 1 kJ mol-1 , especially for hydrogen-bonded ILs and those containing halides, the DLPNO settings had to be increased by two orders of magnitude and include the iterative treatment of triple excitations, resulting in a 2.5-fold increase in computational cost. Two new sets of parameters are put forward to produce the performance of DLPNO-CCSD(T) within chemical and spectroscopic accuracy.

Heat-transport scaling and transition in geostrophic rotating convection with varying aspect ratio

High-precision experimental and numerical data reveal that the power-law scaling (Nu $\ensuremath{\sim}$ Ra${}^{\ensuremath{\gamma}}$) of heat transport by geostrophic convection depends strongly on the fluid domain aspect ratio. Contrary to general belief, the data show that the very steep power-law scaling near onset ($\ensuremath{\gamma}\ensuremath{\ge}3$), does not exist in realistic experimental systems, owing to enhanced heat transfer by boundary flows formed near the lateral boundary. These results imply that the scaling relationship of heat transport measured in rotating convection cells with finite aspect ratio cannot be extrapolated to most large-scale, laterally unbounded, geophysical and astrophysical flows.