将主动学习策略应用于构建全维神经网络的潜在能量表面:H2O-He光谱计算中的关键测试
You Li1,2, Xiao-Long Zhang3, Hui Li1
1Institute of Theoretical Chemistry, College of Chemistry, Jilin University, 2519 Jiefang Road, Changchun 130023, People's Republic of China.
The Journal of chemical physics
|March 26, 2025
概括
一个以不确定性为驱动的积极学习策略有效地采样了构建潜在能量表面 (PES) 的点. 这种方法可以使用更少的数据点进行准确的 PES 构建,从而降低神经网络模型的计算成本.
科学领域:
- 计算化学的计算化学
- 机器学习在化学中的应用
- 频谱学是一种光谱学.
背景情况:
- 准确的潜在能量表面 (PES) 对于理解分子相互作用和预测光谱性质至关重要.
- 构建高维的PES在计算上要求很高,需要有效的数据采样策略.
研究的目的:
- 开发和验证一个以不确定性为驱动的积极学习策略,以在全维 PES 构建中进行高效的点抽样.
- 使用拟议的采样方法,比较不同神经网络模型 (LS-FI-NN和MLRNet) 的效率和准确性.
主要方法:
- 采用基于两个神经网络模型之间的能量差异的不确定性驱动的积极学习策略.
- 实施了两步抽样程序,以降低双精度神经网络训练的计算成本.
- 使用MLRNet为6-D H2O-He系统构建了参考PES,并配备了长距离功能,并采用了削减基础扩展方法.
主要成果:
- 与MLRNet相比,远程交换基本不变神经网络 (LS-FI-NN) 实现了光谱精确的PES,比MLRNet要少得多.
- 单一精度的LS-FI-NN实现了0.3253厘米-1的测试组RMSE,获得472分;双重精度的LS-FI-NN获得0.0710厘米-1的测试组,获得613分.
- 光谱计算证实了双精度LS-FI-NN的高精度,显示出与参考PES的良好一致.
结论:
- 不确定性驱动的积极学习策略可靠地使用LS-FI-NN构建精确的PES,在数据效率方面表现优于MLRNet.
- 提出的方法有效地降低了计算成本,同时保持了对光谱预测的高精度.
- 通过适当的抽样,MLRNet显示了潜力,显示了较低的培训和初始测试错误.
更多相关视频
相关概念视频
01:02¹H NMR: Interpreting Distorted and Overlapping Signals
965
Spin systems where the difference in chemical shifts of the coupled nuclei is greater than ten times J are called first-order spin systems. These nuclei are weakly coupled, and their chemical shifts and coupling constant can generally be estimated from the well-separated signals in the spectrum.
As Δν decreases and the signals move closer, the doublets appear increasingly distorted. The intensities of the inner lines increase at the cost of those of the outer lines as the signals are...
As Δν decreases and the signals move closer, the doublets appear increasingly distorted. The intensities of the inner lines increase at the cost of those of the outer lines as the signals are...
965
01:192D NMR: Heteronuclear Single-Quantum Correlation Spectroscopy (HSQC)
598
Heteronuclear single-quantum correlation spectroscopy (HSQC) is a 2D NMR technique that reveals one-bond correlations between hydrogen and a heteronucleus. The HSQC experiment is similar to the heteronuclear correlation experiment (HETCOR) but is more sensitive. In the HSQC spectrum, the proton chemical shift is plotted on the horizontal F2 axis, while the 13C chemical shift is plotted on the vertical F1 axis. The corresponding proton and 13C spectra are also shown. The HSQC contour plot does...
598
01:20¹³C NMR: Distortionless Enhancement by Polarization Transfer (DEPT)
983
When proton-coupled carbon-13 spectra are simplified by a broadband proton decoupling technique, structural information about the coupled protons is lost. Distortionless enhancement by polarization transfer (DEPT) is a technique that provides information on the number of hydrogens attached to each carbon in a molecule. While the DEPT experiment utilizes complex pulse sequences, the pulse delay and flip angle are specifically manipulated. The resulting signals have different phases depending on...
983
01:07Nuclear Overhauser Enhancement (NOE)
609
Irradiation of a spin-active nucleus causes an increase or decrease in the signal intensity of neighboring nuclei that are not necessarily chemically bonded or involved in J-coupling. This phenomenon, called the Nuclear Overhauser Enhancement (NOE), results from through-space interactions between the nuclear spins. The NOE effect decreases with increasing internuclear distance and is generally not observed beyond 4 angstroms. In NOE, dipole-dipole interactions between neighboring...
609