Search

Wavefunction forms based on products of electron pairs are usually constructed as closed-shell singlets, which is insufficient when the molecular state has a nonzero spin or when the chemistry is determined by $d$- or $f-$electrons. A set of two-electron forms are considered as explicit couplings of second-quantized operators to open-shell singlets. Geminal wavefunctions are constructed and their structure is elaborated. Numerical results for small model systems clearly demonstrate improvement over closed-shell singlet pairs.

Richardson-Gaudin states provide a basis of the Hilbert space for strongly correlated electrons. In this study, optimal expressions for the transition density matrix elements between Richardson-Gaudin states are obtained with a cost comparable with the corresponding reduced density matrix elements. Analogues of the Slater-Condon rules are identified based on the number of near-zero singular values of the RG state overlap matrix. Finally, a perturbative approach is shown to be close in quality to a configuration interaction of Richardson-Gaudin states while being feasible to compute.

This chapter gives an overview of Richardson-Gaudin states which represent weakly correlated pairs of electrons. They are parametrized by sets of numbers obtained from non-linear equations. The best method to solve these equations is presented in a straightforward manner with enough detail to implement the method computationally. Optimal expressions for the density matrix elements are discussed. A simple description of 1-dimensional hydrogen chains is presented in terms of Richardson-Gaudin states. Richardson-Gaudin states are placed in the larger context of integrable models and geminal wavefunctions of quantum chemistry.

Richardson-Gaudin (RG) states are employed as a variational wavefunction ansatz for strongly correlated isomers of H$_4$ and H$_{10}$. In each case a single RG state describes the seniority-zero sector quite well. Simple natural orbital functionals offer a cheap and reasonable approximation of the outstanding weak correlation in the seniority-zero sector, while systematic improvement is achieved by performing a configuration interaction (CI) in terms of RG states. Other pair theories (e.g. generalized valence bond and pair-coupled-cluster doubles) can provide a good description of many of the geometries considered, but, at short distances, the wavefunctions for the 2D and 3D structures of H$_{10}$ take the form of an RG state that cannot be described by these other theories.

Recommendation models are very large, requiring terabytes (TB) of memory during training. In pursuit of better quality, the model size and complexity grow over time, which requires additional training data to avoid overfitting. This model growth demands a large number of resources in data centers. Hence, training efficiency is becoming considerably more important to keep the data center power demand manageable. In Deep Learning Recommendation Models (DLRM), sparse features capturing categorical inputs through embedding tables are the major contributors to model size and require high memory bandwidth. In this paper, we study the bandwidth requirement and locality of embedding tables in real-world deployed models. We observe that the bandwidth requirement is not uniform across different tables and that embedding tables show high temporal locality. We then design MTrainS, which leverages heterogeneous memory, including byte and block addressable Storage Class Memory for DLRM hierarchically. MTrainS allows for higher memory capacity per node and increases training efficiency by lowering the need to scale out to multiple hosts in memory capacity bound use cases. By optimizing the platform memory hierarchy, we reduce the number of nodes for training by 4-8X, saving power and cost of training while meeting our target training performance.

We report the simultaneous softening and compaction of a confined dense granular pack in acoustic resonance experiments. Elastic softening is manifested by a reduction of the shear-wave speed, as the wave amplitude increases beyond some threshold. No macroscopic rearrangement of grains or dilatancy is observed; instead, elastic softening is accompanied by a tiny amount of compaction on the scale of grain asperities. We explain these apparent contradictory observations using a theoretical model, based on shear transformation zones (STZs), of soft spots and slipping contacts. It predicts a linear shear stress-strain response with negligible macro-plastic deformation due to the small-amplitude acoustic oscillation. However, these waves reduce the interparticle friction and contact stiffness through the acoustic lubrication of grain contacts, resulting in an increase in the structural disorder or compactivity and softening of dynamic modulus. The compaction associated with this microscopic friction decrease is consistent with the prediction by an Ising-like correlation between STZs in the subyield regime.

Seniority-zero geminal wavefunctions are known to capture bond-breaking correlation. Among this class of wavefunctions, Richardson-Gaudin states stand out as they are eigenvectors of a model Hamiltonian. This provides a clear physical picture, clean expressions for reduced density matrix (RDM) elements, and systematic improvement (with a complete set of eigenvectors). Known expressions for the RDM elements require the computation of rapidities, which are obtained by first solving for the so-called eigenvalue based variables (EBV) then root-finding of a Lagrange interpolation polynomial. In this manuscript we obtain expressions for the RDM elements directly in terms of the EBV. The final expressions can be computed with the same cost as the rapidity expressions. Therefore, except in particular circumstances, it is entirely unnecessary to compute rapidities at all. The RDM elements require numerically inverting a matrix and while this is usually undesirable we demonstrate that it is stable, except when there is degeneracy in the single-particle energies. In such cases a different construction would be required.

We develop a bivariational principle for an antisymmetric product of nonorthogonal geminals. Special cases reduce to the antisymmetric product of strongly-orthogonal geminals (APSG), the generalized valence bond-perfect pairing (GVB-PP), and the antisymmetrized geminal power (AGP) wavefunctions. The presented method employs wavefunctions of the same type as Richardson-Gaudin (RG) states, but which are not eigenvectors of a model Hamiltonian which would allow for more freedom in the mean-field. The general idea is to work with the same state in a primal picture in terms of pairs, and in a dual picture in terms of pair-holes. This leads to an asymmetric energy expression which may be optimized bivariationally, and is strictly variational when the two representations are consistent. The general approach may be useful in other contexts, such as for computationally feasible variational coupled-cluster methods.

Eigenvectors of the reduced Bardeen-Cooper-Schrieffer Hamiltonian, Richardson-Gaudin (RG) states, are used as a variational wavefunction Ansatz for strongly-correlated electronic systems. These states are geminal products whose coefficients are solutions of non-linear equations. Previous results showed un-physical behaviour but in this contribution it is shown that with only the variational solution for the ground state, all the seniority-zero states are quite well approximated. The difficulty is in choosing the correct RG state. The systems studied showed a clear choice and we expect it should always be possible to reason physically which state to choose.

Scalar products and density matrix elements of closed-shell pair geminal wavefunctions are evaluated directly in terms of the pair amplitudes, resulting in an analogue of Wick's theorem for fermions or bosons. This expression is in general intractable, but it is shown how it becomes feasible in three distinct ways for Richardson-Gaudin (RG) states, the antisymmetrized geminal power, and the antisymmetrized product of strongly-orthogonal geminals. Dissociation curves for hydrogen chains are computed with off-shell RG states and the antisymmetrized product of interacting geminals. Both are near exact suggesting that the incorrect results observed with ground state RG states are fixable using a different RG state.

Machine learning models using seismic emissions can predict instantaneous fault characteristics such as displacement in laboratory experiments and slow slip in Earth. Here, we address whether the acoustic emission (AE) from laboratory experiments contains information about near-future frictional behavior. The approach uses a convolutional encoder-decoder containing a transformer layer. We use as input progressively larger AE input time windows and progressively larger output friction time windows. The attention map from the transformer is used to interpret which regions of the AE contain hidden information corresponding to future frictional behavior. We find that very near-term predictive information is indeed contained in the AE signal, but farther into the future the predictions are progressively worse. Notably, information for predicting near future frictional failure and recovery are found to be contained in the AE signal. This first effort predicting future fault frictional behavior with machine learning will guide efforts for applications in Earth., Comment: 12 pages, 4 figures

Rayleigh Schr\"{o}dinger perturbation theory corrections are developed for an algebraic Bethe ansatz of individual electrons. Numerical results are ambiguous and would need either an orbital optimization or a configuration interaction singles solution to be satisfactory. Two different expressions are computed to be certain that results are correct.

Data-driven machine-learning for predicting instantaneous and future fault-slip in laboratory experiments has recently progressed markedly due to large training data sets. In Earth however, earthquake interevent times range from 10's-100's of years and geophysical data typically exist for only a portion of an earthquake cycle. Sparse data presents a serious challenge to training machine learning models. Here we describe a transfer learning approach using numerical simulations to train a convolutional encoder-decoder that predicts fault-slip behavior in laboratory experiments. The model learns a mapping between acoustic emission histories and fault-slip from numerical simulations, and generalizes to produce accurate results using laboratory data. Notably slip-predictions markedly improve using the simulation-data trained-model and training the latent space using a portion of a single laboratory earthquake-cycle. The transfer learning results elucidate the potential of using models trained on numerical simulations and fine-tuned with small geophysical data sets for potential applications to faults in Earth., Comment: 12 pages, 8 figures

Time Reverse Nonlinear Elastic Wave Spectroscopy (TR-NEWS) has been used to focus acoustic energy, and make measurements correlated with damage in a variety of industrial materials. Most studies using TR-NEWS in Earth Science have focused on relatively small objects and may have multiple acoustic sources. In Earth, within energy extraction settings, the structure and scale of wellbores makes acoustic focusing challenging. This paper explores the feasibility of applying TR NEWS for damage detection in wellbores by constructing a laboratory-scale wellbore, and using TR to focus and make dynamic linear and nonlinear elastic measurements. After successive cycles of induced, localized mechanical damage the sample, the hysteretic nonlinear elastic parameter {\alpha}, increased with damage cycle indicating progressive mechanical damage. In addition to these strain-dependent changes, TR peak width and changes to peak amplitude near the damage sites was also observed. To deploy TR in a wellbore, it will be necessary to choose sensors that are suitable for the environment, and that can be distributed along the wellbore. Thus, this paper demonstrates that acoustic TR can be conducted using both an intrinsic Fabry Perot interferometer fiber optic strain sensor, and an intrinsic Michaelson interferometer fiber optic strain sensor, as a first step towards deployable sensing for TR in a wellbore.

Systematic characterization of slip behaviours on active faults is key to unraveling the physics of tectonic faulting and the interplay between slow and fast earthquakes. Interferometric Synthetic Aperture Radar (InSAR), by enabling measurement of ground deformation at a global scale every few days, may hold the key to those interactions. However, atmospheric propagation delays often exceed ground deformation of interest despite state-of-the art processing, and thus InSAR analysis requires expert interpretation and a priori knowledge of fault systems, precluding global investigations of deformation dynamics. Here we show that a deep auto-encoder architecture tailored to untangle ground deformation from noise in InSAR time series autonomously extracts deformation signals, without prior knowledge of a fault's location or slip behaviour. Applied to InSAR data over the North Anatolian Fault, our method reaches 2 mm detection, revealing a slow earthquake twice as extensive as previously recognized. We further explore the generalization of our approach to inflation/deflation-induced deformation, applying the same methodology to the geothermal field of Coso, California.

Active faults release tectonic stress imposed by plate motion through a spectrum of slip modes, from slow, aseismic slip, to dynamic, seismic events. Slow earthquakes are often associated with tectonic tremor, non-impulsive signals that can easily be buried in seismic noise and go undetected. We present a new methodology aimed at improving the detection and location of tremors hidden within seismic noise. After identifying tremors with a classic convolutional neural network, we rely on neural network attribution to extract core tremor signatures and denoise input waveforms. We then use these cleaned waveforms to locate tremors with standard array-based techniques. We apply this method to the Cascadia subduction zone, where we identify tremor patches consistent with existing catalogs. In particular, we show that the cleaned signals resulting from the neural network attribution analysis correspond to a waveform traveling in the Earth's crust and mantle at wavespeeds consistent with local estimates. This approach allows us to extract small signals hidden within the noise, and therefore to locate more tremors than in existing catalogs., Comment: 7 pages, 6 figures

Recently, ground state eigenvectors of the reduced Bardeen-Cooper-Schrieffer Hamiltonian, Richardson-Gaudin (RG) states, have been employed as a wavefunction ansatz for strong correlation. This wavefunction physically represents a mean-field of pairs of electrons (geminals) with a constant pairing strength. To move beyond the mean-field, one must develop the wavefunction in the basis of all the RG states. This requires both practical expressions for transition density matrices and an idea of which states are most important in the expansion. In this contribution, we present expressions for the transition density matrix elements and calculate them numerically for half-filled picket fence models. There are no Slater-Condon rules for RG states, though an analogue of the aufbau principle proves to be useful in choosing which states are important.