NucleoNet: A Nuclear-Aware Determinant Ansatz for Multi-Geometry Neural Variational Monte Carlo
School of Artificial Intelligence and Data Science, University of Science and Technology of China(中国科学技术大学人工智能与数据科学学院);University of Science and Technology of China(中国科学技术大学);Suzhou Institute for Advanced Research, University of Science and Technology of China(中国科学技术大学苏州高等研究院);Singapore University of Technology and Design(新加坡科技设计大学);AI for Science Institute(北京智源人工智能研究院);School of Mathematical Sciences, University of Science and Technology of China(中国科学技术大学数学科学学院);National Laboratory of Computational Physics, Institute of Applied Physics and Computational Mathematics(应用物理与计算数学研究部计算物理国家实验室);HEDPS, CAPT, College of Engineering, Peking University(北京大学工学院)
Qiao Gao, Yang Huang, Enze Hou, Yuzhi Liu, Jingrun Chen, Han Wang
Neural-network variational Monte Carlo provides an accurate route to solving the electronic Schrödinger equation, but many neural wavefunctions are optimized for a single nuclear geometry, requiring separate calculations to map out a potential energy surface. We introduce NucleoNet, a nuclear-aware determinant ansatz in which electrons and nuclei form a unified particle set propagated through a shared electron--nucleus backbone that jointly evolves one-particle and two-particle representations. Nuclei carry learned representations that evolve alongside the electronic representations and condition orbital construction, while antisymmetry is imposed only by electronic determinants. On equilibrium small-molecule benchmarks, NucleoNet achieves accuracy comparable to established neural-wavefunction baselines with substantially fewer parameters. For each molecular system, a single model optimized jointly across geometries describes the N$_2$ dissociation curve, Li displacement in LiCN, and the ethane and formamide minimum-energy paths for bond dissociation. Sparse-geometry training generalizes to unseen geometries for N$_2$ and LiCN without further optimization, and N$_2$ ablations show that nuclear streams substantially improve this interpolation. These results show that a compact wavefunction with jointly evolved nuclear representations can describe the geometries of a molecular system with a single set of parameters.
Convergence to the fixed-node limit in deep variational Monte Carlo
FU Berlin(柏林自由大学);TU Berlin(柏林工业大学);Rice University(莱斯大学)
变分量子蒙特卡罗(Variational Quantum Monte Carlo,QMC)是一种求解电子薛定谔方程的从头计算方法,它在原理上是精确的,但在实际应用中受到现有方法灵活性的限制。最近引入的深度QMC方法,特别是两个深度神经网络ansatzes PauliNet和FermiNet,允许变分QMC达到扩散QMC的精度,但对这些ansatzes的收敛行为知之甚少。在这里,我们分析如何深入变QMC方法的固定节点的限制,随着网络规模的增加。首先,我们证明了深度神经网络可以克服小基集的限制,并达到平均场完全基集极限。移动到电子相关性,然后我们进行了广泛的超参数扫描的深Jastrow因子的LiH和H $_4$,并发现在固定节点的限制,可以得到一个足够大的网络的变分能量。最后,我们基准的平均场和多体ansatzes H $_2$ O,增加的分数恢复固定节点的相关能量的单决定斯莱特-Jastrow型ansatzes的半个数量级相比,以前的变QMC的结果,并证明了一个单一的决定斯莱特-Jastrow-回流版本的ansatzes克服了固定节点的限制。这种分析有助于理解与传统的试验波函数相比,深度变分ansatzes在各自的理论水平上的高超精度,并将指导未来改进深度QMC中的神经网络架构。
TurboRVB: a many-body toolkit for {\it ab initio} electronic simulations by quantum Monte Carlo
TurboRVB是一个计算软件包,用于分子和体电子系统的{\it ab initio}量子蒙特卡罗(QMC)模拟。该代码实现了两种类型的成熟的QMC算法:变分蒙特卡罗(VMC)和扩散蒙特卡罗在其强大和有效的晶格正则化的变体。该代码的一个关键特点是使用强相关的多体波函数的可能性。电子波函数(WF)是通过将考虑动力学相关性的Jastrow因子应用于最一般的平均场基态而获得的,该平均场基态被写为具有自旋单重态配对的反对称化的孪积,或者被写为包括单重态相关性和三重态相关性的Pfweian。这种波函数可以看作是由L.鲍林和P.W.安德森在量子化学和凝聚态物理,分别。TurboRVB中实现的RVB计算器具有很大的变分自由度,包括Jastrow相关的Slater行列式作为其最简单但非平凡的情况。此外,它有一个显着的优势,剩余的一个负担得起的计算成本,成比例的一个用于评估一个单一的斯莱特行列式。该代码实现了伴随算法的差异,使一个非常有效的评估能源衍生物,包括离子力。因此,人们可以在VMC水平上进行正则NVT系综的结构优化和分子动力学。对于电子部分,一个完整的WF优化是可能的,由于国家的最先进的随机算法的能量最小化。通过使用混合MPI-OpenMP协议,代码已被高效并行化,这也是利用现代GPU加速器计算能力的理想环境。
Basis-free neural-network geminal and Jastrow factors for variational Monte Carlo
University of Paderborn(帕德博恩大学);Center for Advanced Systems Understanding (CASUS)(先进系统理解中心);Helmholtz Zentrum Dresden-Rossendorf(德累斯顿-罗斯多夫亥姆霍兹中心);Technische Universität Dresden(德累斯顿工业大学)