OrbNet Denali: A machine learning potential for biological and organic chemistry with semi-empirical cost and DFT accuracy

We present OrbNet Denali, a machine learning model for electronic structure that is designed as a drop-in replacement for ground-state density functional theory (DFT) energy calculations. The model is a message-passing neural network that uses symmetry-adapted atomic orbital features from a low-cost quantum calculation to predict the energy of a molecule. OrbNet Denali is trained on a vast dataset of 2.3 million DFT calculations on molecules and geometries. This dataset covers the most common elements in bio- and organic chemistry (H, Li, B, C, N, O, F, Na, Mg, Si, P, S, Cl, K, Ca, Br, I) as well as charged molecules. OrbNet Denali is demonstrated on several well-established benchmark datasets, and we find that it provides accuracy that is on par with modern DFT methods while offering a speedup of up to three orders of magnitude. For the GMTKN55 benchmark set, OrbNet Denali achieves WTMAD-1 and WTMAD-2 scores of 7.19 and 9.84, on par with modern DFT functionals. For several GMTKN55 subsets, which contain chemical problems that are not present in the training set, OrbNet Denali produces a mean absolute error comparable to those of DFT methods. For the Hutchison conformers benchmark set, OrbNet Denali has a median correlation coefficient of R^2=0.90 compared to the reference DLPNO-CCSD(T) calculation, and R^2=0.97 compared to the method used to generate the training data (wB97X-D3/def2-TZVP), exceeding the performance of any other method with a similar cost. Similarly, the model reaches chemical accuracy for non-covalent interactions in the S66x10 dataset. For torsional profiles, OrbNet Denali reproduces the torsion profiles of wB97X-D3/def2-TZVP with an average MAE of 0.12 kcal/mol for the potential energy surfaces of the diverse fragments in the TorsionNet500 dataset.

Active gyrotactic stability of microswimmers using hydromechanical signals

Many plankton species undergo daily vertical migration to large depths in the turbulent ocean. To do this efficiently, the plankton can use a gyrotactic mechanism, aligning them with gravity to swim downwards, or against gravity to swim upwards. Many species show passive mechanisms for gyrotactic stability. For example, bottom-heavy plankton tend to align upwards. This is efficient for upward migration in quiescent flows, but it is often sensitive to turbulence which upsets the alignment. Here we suggest a simple, robust active mechanism for gyrotactic stability, which is only lightly affected by turbulence and allows alignment both along and against gravity. We use a model for a plankton that swims with a constant speed and can actively steer in response to hydrodynamic signals encountered in simulations of a turbulent flow. Using reinforcement learning, we identify the optimal steering strategy. By using its setae to sense its settling velocity transversal to its swimming direction, the swimmer can deduce information about the direction of gravity, allowing it to actively align upwards. The mechanism leads to a rate of upward migration in a turbulent flow that is of the same order as in quiescent flows, unless the turbulence is very vigorous. In contrast, passive swimmers show much smaller upward velocity in turbulence. Settling may even cause them to migrate downwards in vigorous turbulence.

Pseudo-ISP: Learning Pseudo In-camera Signal Processing Pipeline from A Color Image Denoiser

The success of deep denoisers on real-world color photographs usually relies on the modeling of sensor noise and in-camera signal processing (ISP) pipeline. Performance drop will inevitably happen when the sensor and ISP pipeline of test images are different from those for training the deep denoisers (i.e., noise discrepancy). In this paper, we present an unpaired learning scheme to adapt a color image denoiser for handling test images with noise discrepancy. We consider a practical training setting, i.e., a pre-trained denoiser, a set of test noisy images, and an unpaired set of clean images. To begin with, the pre-trained denoiser is used to generate the pseudo clean images for the test images. Pseudo-ISP is then suggested to jointly learn the pseudo ISP pipeline and signal-dependent rawRGB noise model using the pairs of test and pseudo clean images. We further apply the learned pseudo ISP and rawRGB noise model to clean color images to synthesize realistic noisy images for denoiser adaption. Pseudo-ISP is effective in synthesizing realistic noisy sRGB images, and improved denoising performance can be achieved by alternating between Pseudo-ISP training and denoiser adaption. Experiments show that our Pseudo-ISP not only can boost simple Gaussian blurring-based denoiser to achieve competitive performance against CBDNet, but also is effective in improving state-of-the-art deep denoisers, e.g., CBDNet and RIDNet.

Maximum Energy Growth Rate in Dilute Quantum Gases

In this letter we study how fast the energy density of a quantum gas can increase in time, when the inter-atomic interaction characterized by the $s$-wave scattering length $a_text{s}$ is increased from zero with arbitrary time dependence. We show that, at short time, the energy density can at most increase as $sqrt{t}$, which can be achieved when the time dependence of $a_text{s}$ is also proportional to $sqrt{t}$, and especially, a universal maximum energy growth rate can be reached when $a_text{s}$ varies as $2sqrt{hbar t/(pi m)}$. If $a_text{s}$ varies faster or slower than $sqrt{t}$, it is respectively proximate to the quench process and the adiabatic process, and both result in a slower energy growth rate. These results are obtained by analyzing the short time dynamics of the short-range behavior of the many-body wave function characterized by the contact, and are also confirmed by numerical solving an example of interacting bosons with time-dependent Bogoliubov theory. These results can also be verified experimentally in ultracold atomic gases.

Multi-task learning for electronic structure to predict and explore molecular potential energy surfaces

We refine the OrbNet model to accurately predict energy, forces, and other response properties for molecules using a graph neural-network architecture based on features from low-cost approximated quantum operators in the symmetry-adapted atomic orbital basis. The model is end-to-end differentiable due to the derivation of analytic gradients for all electronic structure terms, and is shown to be transferable across chemical space due to the use of domain-specific features. The learning efficiency is improved by incorporating physically motivated constraints on the electronic structure through multi-task learning. The model outperforms existing methods on energy prediction tasks for the QM9 dataset and for molecular geometry optimizations on conformer datasets, at a computational cost that is thousand-fold or more reduced compared to conventional quantum-chemistry calculations (such as density functional theory) that offer similar accuracy.

Swimming strategy of settling elongated microswimmers by reinforcement learning

Particular types of plankton in aquatic ecosystems can coordinate their motion depending on the local flow environment to reach regions conducive to their growth or reproduction. Investigating their swimming strategies with regard to the local environment is important to obtain in-depth understanding of their behavior in the aquatic environment. In the present research, to examine an impact of the shape and gravity on a swimming strategy, plankton is considered as settling swimming particles of ellipsoidal shape. The Q-learning approach is adopted to obtain swimming strategies for smart particles with a goal of efficiently moving upwards in a two-dimensional steady flow. Strategies obtained from reinforcement learning are compared to those of naive gyrotactic particles that is modeled considering the behavior of realistic plankton. It is found that elongation of particles improves the performance of upward swimming by facilitating particles resistance to the perturbation of vortex. In the case when the settling velocity is included, the strategy obtained by reinforcement learning has similar performance to that of the naive gyrotactic one, and they both align swimmers in upward direction. The similarity between the strategy obtained from machine learning and the biological gyrotactic strategy indicates the relationship between the aspherical shape and settling effect of realistic plankton and their gyrotactic feature.

Directional coupling and focusing surface plasmon polaritons controlled by arbitrary spin of light

Surface plasmon polaritons have attracted varies of interests due to its special properties, especially in the polarization-controlled devices. Typically, the polarization-controlled devices include directional coupling, focusing lens and plasmonic vortex lens, and almost all of them are controlled by the input circularly polarized light or the linearly polarized light. We present a novel device that realize the functions of directional coupling and focusing with high polarization extinction ratio for arbitrary spin of input light. This device offers opportunities for polarization sensing, polarization splitting and polarization-multiplexed near-field images and surface plasmon holography in the future.

Arbitrary spin-controlled directional coupling of surface plasmon polaritons

Controlling the directionality of surface plasmon polaritons (SPPs) has been widely studied, while the direction of SPPs was always switched by orthogonal polarizations in the reported methods. Here, we present a scheme to control the directionality of SPPs by arbitrary spin polarizations. Extremely, the device can split two quite adjacent polarization components to two opposite directions. The versatility of the presented design scheme can offer opportunities for polarization sensing, polarization splitting and polarization-controlled plasmonic devices.

Dynamic Super Efimov Effect

Super Efimov effect is a recently proposed three-body effect characterized by a double-exponential scaling, which has not been observed experimentally yet. Here, we present the general dynamic equations determining the cloud size of a scale invariant quantum gas in a time dependent harmonic trap. We show that a double-log periodicity as the hallmark of the super Efimov effect emerges when the trap frequency is decreased with a specially designed time-dependence. We also demonstrate that this dynamic super Efimov effect can be realized with realistic choices of parameters in current experiments.

Vortex Lattices in the Bose-Fermi Superfluid Mixture

In this letter we show that the vortex lattice structure in the Bose-Fermi superfluid mixture can undergo a sequence of structure transitions when the Fermi superfluid is tuned from the BCS regime to the BEC regime. This is due to different vortex core structure of the Fermi superfluid in the BCS regime and in the BEC regime. In the former the vortex core is nearly filled, while the density at the vortex core gradually decreases until it empties out at the BEC regime. Therefore, with the density-density interaction between the Bose and the Fermi superfluids, the two sets of vortex lattices interact stronger in the BEC regime that yields the structure transition of vortex lattices. In view of recent realization of this superfluid mixture and vortices therein, our theoretical predication can be verified experimentally in near future.