基于深度学习的ELM实时识别研究

基于深度学习的方法，在HL-2A装置上开发出了一套边缘局域模(ELM)实时识别算法。算法使用5200次放电数据(约24.19万数据切片)进行学习，得到一个深度为22层的卷积神经网络。为衡量算法的识别能力，识别了HL-2A装置自2009年实现稳定ELMy H模放电以来所有历史数据(约26000次放电数据)，共识别出1665次H模放电，其中误识别35次，误报率为2.10%。在实际的1634次H模放电中，漏识别4次，漏识别率为0.24%。该误报率和漏报率可以满足ELM实时识别的精度要求。识别算法在实时控制环境下，对单个时间点的平均计算时间为0.46ms，可以满足实时控制的计算速度要求。     

Highly Luminescent Red Phosphorescent OLED with Interfacial Layers for Hole Transport and Electron Injection

Four kinds of red phosphorescent organic light-emitting devices were fabricated and compared to investigate the effect of interfacial layers for hole transport and electron injection. 1 nm-thick LiF in the device A and C and 1 nm-thick Cs₂CO₃ in the device B and D were deposited as an electron injection layer between the anode and the electron transport layer, and 5 nm-thick layer of dipyrazion[2,3-f:2′,2′-h]quinoxaline-2,3,6,7,10,11-hexacarbonitrile[HATCN] was inserted as a hole transport interfacial layer between the hole injection layer and the hole transport layer only in the device C and D. Under a luminance of 1000 cd/m², the power efficiencies were 7.6 lm/W and 8.5 lm/W in the device A and B, and 8.6 lm/W and 13.4 lm/W in the device C and D. The quantum efficiency of the device D was 15.8% under 1000 cd/m² which was somewhat lower than those of the device A and C, but a little higher than that of the device B. The luminance of the device D was much higher than those of the other devices at a given votage. The luminance of the device D at 7 V was 23,710 cd/m², which was 13.0, 3.4, and 4.0 times higher than those of the device A, B, and C at the same voltage, respectively.     

Convergence of solutions for the fractional Cahn–Hilliard system

This paper deals with the Cauchy–Dirichlet problem for the fractional Cahn–Hilliard equation. The main results consist of global (in time) existence of weak solutions, characterization of parabolic smoothing effects (implying under proper condition eventual boundedness of trajectories), and convergence of each solution to a (single) equilibrium. In particular, to prove the convergence result, a variant of the so-called ?ojasiewicz–Simon inequality is provided for the fractional Dirichlet Laplacian and (possibly) non-analytic (but $C^1$) nonlinearities.     

Potts models with magnetic field: Arithmetic,geometry, and computation

We give a sheaf theoretic interpretation of Potts models with external magnetic field, in terms of constructible sheaves and their Euler characteristics. We show that the polynomial countability question for the hypersurfaces defined by the vanishing of the partition function is affected by changes in the magnetic field: elementary examples suffice to see non-polynomially countable cases that become polynomially countable after a perturbation of the magnetic field. The same recursive formula for the Grothendieck classes, under edge-doubling operations, holds as in the case without magnetic field, but the closed formulae for specific examples like banana graphs differ in the presence of magnetic field. We give examples of computation of the Euler characteristic with compact support, for the set of real zeros, and find a similar exponential growth with the size of the graph. This can be viewed as a measure of topological and algorithmic complexity. We also consider the computational complexity question for evaluations of the polynomial, and show both tractable and NP-hard examples, using dynamic programming.     

Gas-phase 21Ne NMR studies and the nuclear magnetic dipole moment of neon-21

Gas-phase ²¹Ne nuclear magnetic resonance spectra were measured at the natural abundance of ²¹Ne isotope for samples consisting of pressurized neon up to 60 bar at room temperature and applying the magnetic field of the strength B₀ = 11.7574 T. It showed that the nuclear magnetic resonance frequency is linearly dependent on the density of gaseous neon. The resonance frequency was extrapolated to the zero-density point, and it permitted the determination of the ²¹Ne nuclear magnetic moment, μ(²¹Ne) = 0.6617774(10) μ_N. The present value of μ(²¹Ne) is not influenced by the bulk magnetic susceptibility of neon and interactions between neon atoms; therefore, it is more precise and reliable than the previous result obtained for μ(²¹Ne).     

单层n型MoS2/p型c-Si异质结太阳电池数值模拟

单层二硫化钼(MoS₂)是一种具有优异光电性能的半导体材料,在太阳能能量转换中表现出很大的应用潜力。本文基于AMPS模拟软件,对单层n型MoS₂/p型c-Si异质结太阳电池进行了数值模拟与分析。通过模拟优化,n型MoS₂的电子亲和能为3.75 eV、掺杂浓度为10¹⁸ cm^-3,p型c-Si的掺杂浓度为10¹⁷ cm^-3时,太阳电池能够取得最高22.1%的转换效率。最后模拟了n型MoS₂/p型c-Si异质结界面处的界面态对太阳电池性能的影响,发现界面态密度超过10¹¹ cm^-2·eV^-1时会严重影响太阳电池的光伏性能。