Towards chemical space exploration for functional nanostructured systems

纳米结构和(超)分子材料为催化、生物医学、热电、电池和传感等应用提供了有趣的解决方案。我们的总体目标是理解和设计功能纳米结构系统的结构模型，其中两个具有代表性的设计目标是长距离铁磁耦合和波动分子结中的窄电导分布。为此，我们将开发和实现一种可靠和有效的理论方法来计算a)双核过渡金属络合物中的交换自旋耦合常数和b)波动分子结的电导。为了提高效率，我们将依靠机器学习，而为了可靠性，我们将使用提供内在误差度量的方法。与主动学习相结合，这有助于我们为探索功能纳米结构系统的化学空间奠定基础。此外，我们的目标是从我们的机器学习模型中获得尽可能多的洞察力。这将使我们不仅能够理解结构-属性关系，而且还可以进行稳健的外推，并理解为什么某个机器学习模型是成功的或失败的。作为一个长期的愿景，向大系统的延伸，如表面的分子层或金属-有机骨架，可以允许与例如催化、传感、气体分离和自旋电子学等技术应用相关的预测。

Nanostructured and (supra-) molecular materials offer intriguing solutions for applications such as catalysis, biomedicine, thermoelectrics, batteries, and sensing. Our overall objective is to understand and design structural models of functional nanostructured systems, with two representative design goals of ferromagnetic coupling over large distances and narrow conductance distributions in fluctuating molecular junctions. For this purpose, we will develop and implement a reliable and efficient theoretical methodology for evaluating a) exchange spin coupling constants in dinuclear transition metal complexes and b) conductances for fluctuating molecular junctions. For efficiency, we will rely on machine learning, and for reliability, we will employ methods that provide intrinsic error measures. Combined with active learning, this with help us lay the groundwork for exploring the chemical space of functional nanostructured systems. Furthermore, our goal is to gain as much insight as possible from our machine learning models. This will allow us not only to understand structure-property relationships but also robust extrapolation and to understand why a certain machine learning model is successful or not. As a long-term vision, extensions towards large systems such as molecular layers on surfaces or metal-organic frameworks can allow for predictions relevant for technological applications in, e.g., catalysis, sensing, gas separation, and spintronics.