Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userDeep Learning Model for Finding New Superconductors
54
0
0.0
(
0
)
Ask ChatGPT about the research
No Arabic abstract
Exploration of new superconductors still relies on the experience and intuition of experts and is largely a process of experimental trial and error. In one study, only 3% of the candidate materials showed superconductivity. Here, we report the first deep learning model for finding new superconductors. We introduced the method named reading periodic table which represented the periodic table in a way that allows deep learning to learn to read the periodic table and to learn the law of elements for the purpose of discovering novel superconductors that are outside the training data. It is recognized that it is difficult for deep learning to predict something outside the training data. Although we used only the chemical composition of materials as information, we obtained an $R^{2}$ value of 0.92 for predicting $T_text{c}$ for materials in a database of superconductors. We also introduced the method named garbage-in to create synthetic data of non-superconductors that do not exist. Non-superconductors are not reported, but the data must be required for deep learning to distinguish between superconductors and non-superconductors. We obtained three remarkable results. The deep learning can predict superconductivity for a material with a precision of 62%, which shows the usefulness of the model; it found the recently discovered superconductor CaBi2 and another one Hf0.5Nb0.2V2Zr0.3, neither of which is in the superconductor database; and it found Fe-based high-temperature superconductors (discovered in 2008) from the training data before 2008. These results open the way for the discovery of new high-temperature superconductor families. The candidate materials list, data, and method are openly available from the link https://github.com/tomo835g/Deep-Learning-to-find-Superconductors.
rate research
Read More
We introduce novel communication strategies in synchronous distributed Deep Learning consisting of decentralized gradient reduction orchestration and computational graph-aware grouping of gradient tensors. These new techniques produce an optimal overlap between computation and communication and result in near-linear scaling (0.93) of distributed training up to 27,600 NVIDIA V100 GPUs on the Summit Supercomputer. We demonstrate our gradient reduction techniques in the context of training a Fully Convolutional Neural Network to approximate the solution of a longstanding scientific inverse problem in materials imaging. The efficient distributed training on a dataset size of 0.5 PB, produces a model capable of an atomically-accurate reconstruction of materials, and in the process reaching a peak performance of 2.15(4) EFLOPS$_{16}$.
Superconductivity results from a Bose condensate of Cooper-paired electrons with a macroscopic quantum wavefunction. Dramatic effects can occur when the region of the condensate is shaped and confined to the nanometer scale. Recent progress in nanostructured superconductors has revealed a route to topological superconductivity, with possible applications in quantum computing. However, challenges remain in controlling the shape and size of specific superconducting materials. Here, we report a new method to create nanostructured superconductors by partial crystallization of the half-Heusler material, YPtBi. Superconducting islands, with diameters in the range of 100 nm, were reproducibly created by local current annealing of disordered YPtBi in the tunneling junction of a scanning tunneling microscope (STM). We characterize the superconducting island properties by scanning tunneling spectroscopic measurements to determine the gap energy, critical temperature and field, coherence length, and vortex formations. These results show unique properties of a confined superconductor and demonstrate that this new method holds promise to create tailored superconductors for a wide variety of nanometer scale applications.
The Nernst effect is the transverse electric field produced by a longitudinal thermal gradient in presence of magnetic field. In the beginning of this century, Nernst experiments on cuprates were analyzed assuming that: i) The contribution of quasi-particles to the Nernst signal is negligible; and ii) Gaussian superconducting fluctuations cannot produce a Nernst signal well above the critical temperature. Both these assumptions were contradicted by subsequent experiments. This paper reviews experiments documenting multiple sources of a Nernst signal, which, according to the Brigman relation, measures the flow of transverse entropy caused by a longitudinal particle flow. Along the lines of Landauers approach to transport phenomena, the magnitude of the transverse magneto-thermoelectric response is linked to the quantum of thermoelectric conductance and a number of material-dependent length scales: the mean-free-path, the Fermi wavelength, the de Broglie thermal wavelength and the superconducting coherence length. Extremely mobile quasi-particles in dilute metals generate a widely-documented Nernst signal. Fluctuating Cooper pairs in the normal state of superconductors have been found to produce a detectable Nernst signal with an amplitude conform to the Gaussian theory, first conceived by Ussishkin, Sondhi and Huse. In addition to these microscopic sources, mobile Abrikosov vortices, mesoscopic objects carrying simultaneously entropy and magnetic flux, can produce a sizeable Nernst response. Finally, in metals subject to a magnetic field strong enough to truncate the Fermi surface to a few Landau tubes, each exiting tube generates a peak in the Nernst response. The survey of these well-established sources of the Nernst signal is a helpful guide to identify the origin of the Nernst signal in other controversial cases.
Reinforcement learning (RL) is well known for requiring large amounts of data in order for RL agents to learn to perform complex tasks. Recent progress in model-based RL allows agents to be much more data-efficient, as it enables them to learn behaviors of visual environments in imagination by leveraging an internal World Model of the environment. Improved sample efficiency can also be achieved by reusing knowledge from previously learned tasks, but transfer learning is still a challenging topic in RL. Parameter-based transfer learning is generally done using an all-or-nothing approach, where the networks parameters are either fully transferred or randomly initialized. In this work we present a simple alternative approach: fractional transfer learning. The idea is to transfer fractions of knowledge, opposed to discarding potentially useful knowledge as is commonly done with random initialization. Using the World Model-based Dreamer algorithm, we identify which type of components this approach is applicable to, and perform experiments in a new multi-source transfer learning setting. The results show that fractional transfer learning often leads to substantially improved performance and faster learning compared to learning from scratch and random initialization.
Charge-density waves are responsible for symmetry-breaking displacements of atoms and concomitant changes in the electronic structure. Linear response theories, in particular density-functional perturbation theory, provide a way to study the effect of displacements on both the total energy and the electronic structure based on a single ab initio calculation. In downfolding approaches, the electronic system is reduced to a smaller number of bands, allowing for the incorporation of additional correlation and environmental effects on these bands. However, the physical contents of this downfolded model and its potential limitations are not always obvious. Here, we study the potential-energy landscape and electronic structure of the Su-Schrieffer-Heeger (SSH) model, where all relevant quantities can be evaluated analytically. We compare the exact results at arbitrary displacement with diagrammatic perturbation theory both in the full model and in a downfolded effective single-band model, which gives an instructive insight into the properties of downfolding. An exact reconstruction of the potential-energy landscape is possible in a downfolded model, which requires a dynamical electron-biphonon interaction. The dispersion of the bands upon atomic displacement is also found correctly, where the downfolded model by construction only captures spectral weight in the target space. In the SSH model, the electron-phonon coupling mechanism involves exclusively hybridization between the low- and high-energy bands and this limits the computational efficiency gain of downfolded models.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
