论文标题
使用机器学习的动能功能衍生物的直接方案计算
Direct Scheme Calculation of the Kinetic Energy Functional Derivative Using Machine Learning
论文作者
论文摘要
我们报告了使用机器学习对动能功能导数的直接方案计算。支持向量回归和内核脊回归技术被独立地用于估计动能功能及其衍生物。即使准确性应该是建模现实功能的决定性因素，但我们表明，在一定层面上，它会影响模型的普遍性。通过选择正确的正则化项并考虑其与准确性之间的合理相互作用，我们能够从训练的模型中推导出功能性衍生物，该模型经过培训以估计动能。尽管导数计算需要很高的精度来解释动能的较小变化，但开发的估计量能够捕获这些极小的电子密度变化。这项工作涌入高效的无轨道密度功能理论，因为它仅采用直接计算方案
We report a direct scheme calculation of kinetic energy functional derivative using Machine Learning. Support Vector Regression and Kernel Ridge Regression techniques were independently employed to estimate the kinetic energy functional and its derivative. Even though the accuracy should have been a decisive factor in modeling a realistic functional, we show that at a certain level it affects the generalizability of the model. By choosing the right regularization term and by considering a reasonable interplay between it and the accuracy, we were able to deduce the functional derivative from a model that was trained to estimate the kinetic energy. Although the derivative calculations demand very high accuracy to account for small variations of the kinetic energy, the developed estimator was capable of capturing these extremely small changes of the electron density. This work pours into highly effective implementation of the orbital-free density functional theory as it employs only direct calculation scheme
