Mendes-Santos Tiago, Schmitt Markus, Heyl Markus
Theoretical Physics III, Center for Electronic Correlations and Magnetism, Institute of Physics, University of Augsburg, 86135 Augsburg, Germany.
Forschungszentrum Jülich GmbH, Peter Grünberg Institute, Quantum Control (PGI-8), 52425 Jülich, Germany.
Phys Rev Lett. 2023 Jul 28;131(4):046501. doi: 10.1103/PhysRevLett.131.046501.
Spectral functions are central to link experimental probes to theoretical models in condensed matter physics. However, performing exact numerical calculations for interacting quantum matter has remained a key challenge especially beyond one spatial dimension. In this work, we develop a versatile approach using neural quantum states to obtain spectral properties based on simulations of the dynamics of excitations initially localized in real or momentum space. We apply this approach to compute the dynamical structure factor in the vicinity of quantum critical points (QCPs) of different two-dimensional quantum Ising models, including one that describes the complex density wave orders of Rydberg atom arrays. When combined with deep network architectures we find that our method reliably describes dynamical structure factors of arrays with up to 24×24 spins, including the diverging timescales at critical points. Our approach is broadly applicable to interacting quantum lattice models in two dimensions and consequently opens up a route to compute spectral properties of correlated quantum matter in yet inaccessible regimes.
在凝聚态物理中，谱函数是将实验探针与理论模型联系起来的核心。然而，对相互作用的量子物质进行精确的数值计算仍然是一个关键挑战，尤其是在超过一维空间的情况下。在这项工作中，我们开发了一种通用方法，利用神经量子态，基于对最初局域在实空间或动量空间中的激发动力学的模拟来获得谱性质。我们将此方法应用于计算不同二维量子伊辛模型量子临界点（QCP）附近的动态结构因子，其中包括一个描述里德堡原子阵列复杂密度波序的模型。当与深度网络架构相结合时，我们发现我们的方法能够可靠地描述多达24×24个自旋的阵列的动态结构因子，包括临界点处发散的时间尺度。我们的方法广泛适用于二维相互作用量子晶格模型，从而为计算尚未可及的相关量子物质的谱性质开辟了一条途径。
