Water splitting has emerged as a key process for obtaining clean molecular hydrogen, a primary energy carrier for replacing fossil fuels, and a valuable fine chemical for several industrial applications. In this framework, great research efforts are being devoted to finding appropriate photocatalysts for water oxidation/reduction, thus directly converting solar energy into hydrogen fuel.(1) To date, several materials have been suggested, including metals or transition metal oxides, sulfides, carbides, and amides.(2) However, the grand challenge is achieving pollution-free energy conversion with sustainable resources. This calls for avoiding exotic material compositions and for focusing only on Earth-abundant elements. For example, even though first-row transition metal oxides are less effective than noble metals (Ir, Rh, and Pt), their catalytic activity can be optimized by doping, defect engineering, and exploiting the properties that emerge at the nanoscale. Specifically, nanostructured materials such as nanoparticles, nanoflakes, and nanorods have the advantage of a large surface area for catalysis and for enhanced solar energy absorption, as well as a low diffusion length, thus reducing the chances for unwanted charge recombination. Once a catalyst is found, there are still many features to optimize for producing an effective device. Overall, water splitting involves charge and mass transport across several materials, interfaces, and phases, with these processes spanning several scales in space and time. Thus, a standard design approach based on trial-and-error will not be effective. The development of water splitting technologies based on nanocrystals requires new fundamental scientific understanding at the atomic scale, and it can be achieved only via computational materials science. In order to tackle the ambitious task of multiscale modeling for water splitting photocatalysis, we built a European-based consortium of theoreticians within the framework of Cooperation in Science and Technology (COST) titled “Computational materials sciences for efficient water splitting with nanocrystals from abundant elements” (CA18234).(3) The aim of the COST action is to bridge the different methodologies dedicated to specific materials and process regimes. The computational modeling of water splitting presents many challenges. Photoexcitation and exciton diffusion (from femtosecond to picosecond), complex coupled chemical reactions (from picosecond to nanosecond), solvent reorganization, and electrolyte diffusion (from nanosecond to millisecond) are all intertwined processes with different time scales, but they must all be taken into account. The complex structure of nanoscale materials and interfaces can affect catalysis. Different material defects, such as dopants, vacancies, dislocations, and grain boundaries, can span different scales (from a few angstroms to many nanometers), and several computational approaches are required to account for each regime. These internal structural defects become more involved when the surface of the nanocrystal edges are exposed or if there is an interesting architecture, such as nanorods, nanoflakes, core–shell structures, and pyramidal shapes. Material modeling must account for structural changes at all length scales and their consequences on electronic structure and catalysis. Here, we provide an overview of the current status in this field and an outlook into the future of multiscale modeling for water splitting, following the first scientific conference of this COST action (Figure 1) that was held on February 11–13, 2020 in Naples, Italy, and was organized by the vice chair of the action, Michele Pavone, together with Ana B. Muñoz Garcia. Figure 1. First international conference on Computational Material Science for Nanoscale Modeling. The chair of the action, Maytal Caspary Toroker, presented the main objective of the consortium, which is bridging the gaps of different theoretical methods. The approach of modeling each regime of an electrochemical device (bulk, surface, and back heterojunctions) separately using density functional theory (DFT) needs to be extended for modeling the entire device; therefore, a multiscale mindset is vital for understanding the catalytic performance of new materials. Even modeling the photoanode junctions with the back contact is an obstacle with regard to the current computational limits, such as accounting for orientation relationships, mismatch, and interface defects. A possible solution is a high-throughput screening approach that allows charge transport to be modeled through a junction using the Kohn–Sham potential.(4) Francesc Illas opened the discussion of the working group (WG) on electronic structure methods. In this WG, the need emerged for a pragmatic approach in modeling TiO₂ nanostructures, one of the most studied candidates for water splitting, including ZnO, WO₃, SrTiO₃, and ZnS. Using DFT, TiO₂ nanoparticles with an impressive size of hundreds of atoms without periodic boundaries can be simulated with a scaling ability of up to 4000 cores, and the error in calculating the band gap can be corrected according to a systematic deviation found relative to G₀W₀ calculations.(5,6) Much of the analysis achieved to date using DFT-based methods for modeling heterogeneous catalysis of water splitting is the scaling relationship found among the efficiency, binding energy of the constituting adsorbates, and position of the transition metal d-band center.(7) However, DFT can allow us to model a particle with a limited number of atoms or to model at most a certain termination or small part of a surface edge. We need to account for several possible terminations with different stabilities, such as a method for considering all possible configurational orderings in nanoparticle alloys.(8) This approach is based on calculating the free energy needed for each configuration and implementing a Monte Carlo algorithm for the most stable arrangements. An additional complication is the interaction of nanoparticles with their substrate and the consequences of electronic interaction and charge transport between them. Moreover, photoinduced water splitting requires treating excited states, which has been conventionally achieved with time-dependent DFT but is a distant challenge for large-scale simulations. The vision is that these nanoparticles must also be solvated and treated with an increased length scale.(9) With an increase in scale, there is concern for the balance between the amount of atomic information and the accuracy of the method. Nicu Goga opened the discussion of the WG on molecular dynamics (MD). More chemical information can be included in MD simulations with either reactive or nonreactive force fields, and the force fields can be obtained from DFT. A major practical limitation of MD is the amount of time needed to calculate the parameters of the force field. Obtaining a reliable force field that reproduces accurate DFT calculations needs to be developed each time a new material is considered. An unsatisfying solution is to use already developed force fields for similar materials. Moreover, MD can be pushed to address further space and time scales via a coarse-grain metaparticle parametrization for multiscale simulations. Indeed, the community by far is using DFT more than MD. The upside of a large number of personnel working in DFT is the larger amount of material data that can potentially be obtained through combining efforts. However, bridging the DFT and MD communities is more challenging, and one approach that came out of discussion is to use a smooth transition between the methodologies, such as a tight-binding density functional (DFTB), which makes the calculation more compact through the use of the most important information. Another rapidly developing area is machine learning, which can be trained to predict useful parameters that can be imported between methods, such as parametrizing the force field from DFT data to MD simulations. Once MD and DFT data are in hand, a microkinetic model may be built in a continuum regime, as done by Anja Bieberle for hematite (Fe₂O₃).(10) The microkinetic model for the oxygen evolution reaction (OER) is translated to the current, which is directly comparable to experimental data. Juan Antonio Anta and Sofia Calero presented a kinetic Monte Carlo model with periodic boundary conditions for calculating the diffusion of charge carriers between trapping sites in solar cells.(11) Some of these high-scale methods, such as heat transfer calculations, lack chemical information that is especially important for modeling heterogeneous catalysis. An important caution needs to be invoked to avoid losing quantum mechanical information when upscaling models. A natural, yet advanced approach, for multiscale modeling is to divide the system into small segments. Florian Libisch presented an embedding approach that separates the material into areas; the electronic structure problem is thereby solved in each area separately, and the size of the problem is reduced. The challenges are formulating the electron–electron interactions in the embedding potential and choosing the initial condition for generating the embedding potential. One approach is to solve the sum of electron densities via an optimized effective potential scheme,(12) thereby allowing the formulation of an iterative procedure that can be used to obtain a unique embedding potential. The growing network established within the COST action aims to obtain a unified protocol for modeling nanocatalysts. The COST action provides a framework for researchers from different areas of expertise to join forces and combine efforts for advancing modeling tools at different length scales. An important step toward this ambitious goal is to document how to use novel algorithms and approaches for each modeling scale and their combination with other scales. It will be inspiring to develop new algorithms and ideas that will enable bridging between the different theoretical communities, and as a starting point, we will focus on distributing the knowledge between one scale to the next. Our ultimate goal is to achieve a unified protocol that includes the algorithms needed to treat nanocatalysts from first-principles to a continuum regime by the end of the action period. The majority of the workforce will be theoreticians from chemistry, physics, material science, and computer science. The action will involve a few experimental researchers as well in order to have some vision in mind for an experimental setup we may wish to model. We anticipate that this COST action will encourage projects in this direction and make breakthroughs in broadening the viewpoint of computational materials science toward a common ground among theoretical regimes. Views expressed in this Energy Focus are those of the authors and not necessarily the views of the ACS. The authors declare no competing financial interest. This article is based upon work from COST Action 18234, supported by COST (European Cooperation in Science and Technology). This work was supported by the Nancy and Stephen Grand Technion Energy Program (GTEP) and a grant from the Ministry of Science and Technology (MOST), Israel. M.P. acknowledge support from the University of Naples Federico II, the Department of Physics “E. Pancini”, and the Department of Chemical Sciences. This article references 12 other publications.

中文翻译：

---

迈向雄心勃勃的水分解纳米晶体催化剂的多尺度建模

水分解已成为获取清洁分子氢的关键工艺，替代化石燃料的主要能源载体以及对多种工业应用有价值的精细化学品。在此框架下，人们正在进行大量的研究工作，以寻找合适的光催化剂来进行水的氧化/还原，从而将太阳能直接转化为氢燃料。（1）迄今为止，已经提出了几种材料，包括金属或过渡金属氧化物，硫化物，碳化物和酰胺。（2）然而，巨大的挑战是如何利用可持续资源实现无污染的能源转换。这就要求避免外来物质的成分，只关注地球上丰富的元素。例如，即使第一行过渡金属氧化物的有效性不如贵金属（Ir，Rh和Pt），它们的催化活性可以通过掺杂，缺陷工程和利用纳米级出现的特性来优化。具体地，诸如纳米颗粒，纳米薄片和纳米棒的纳米结构材料具有用于催化和用于增强的太阳能吸收的大表面积以及低扩散长度的优点，因此减少了不想要的电荷复合的机会。一旦找到催化剂，仍然有许多功能可以优化以生产有效的设备。总的来说，水分解涉及跨多种材料，界面和相的电荷和质量传输，这些过程在空间和时间上跨越多个尺度。因此，基于反复试验的标准设计方法将无效。基于纳米晶体的水分解技术的发展需要在原子尺度上有新的基础科学理解，而这只有通过计算材料科学才能实现。为了解决用于水分解光催化的多尺度建模的雄心勃勃的任务，我们在...的框架内建立了一个欧洲的理论家财团。科学技术合作（COST）的标题为“利用丰富元素的纳米晶体有效分解水的计算材料科学” （CA18234）。（3）COST行动的目的是弥合专用于特定材料和工艺制度的不同方法。水分解的计算模型提出了许多挑战。光激发和激子扩散（从飞秒到皮秒），复杂的耦合化学反应（从皮秒到纳秒），溶剂重组和电解质扩散（从纳秒到毫秒）都是时间尺度相互交织的过程，但是必须将它们全部考虑在内帐户。纳米级材料和界面的复杂结构会影响催化作用。不同的材料缺陷（例如掺杂剂，空位，位错和晶界）可以跨越不同的尺度（从几埃到几纳米），并且需要几种计算方法来说明每种情况。当暴露纳米晶体边缘的表面或存在有趣的结构（例如纳米棒，纳米薄片，核-壳结构和金字塔形状）时，这些内部结构缺陷会变得更加复杂。材料建模必须考虑所有长度范围的结构变化及其对电子结构和催化作用的影响。在此，我们于2020年2月11日至13日在COST举行了首次科学会议（图1）之后，概述了该领域的现状，并展望了水分解的多尺度建模的未来。由行动副主席Michele Pavone和Ana B.MuñozGarcia共同组织，意大利那不勒斯。图1.第一次关于纳米级建模的计算材料科学国际会议。行动主席 Maytal Caspary Toroker提出了财团的主要目标，即弥合不同理论方法之间的差距。需要扩展使用密度泛函理论（DFT）分别对电化学设备的每个状态（本体，表面和背面异质结）建模的方法，以对整个设备进行建模。因此，多尺度思维对于理解新材料的催化性能至关重要。就当前的计算限制而言，甚至使用背面触点对光电阳极结进行建模也是一个障碍，例如考虑了方向关系，失配和界面缺陷。一种可能的解决方案是高通量筛选方法，该方法允许使用Kohn-Sham势通过结模拟电荷传输。（4）Francesc Illas开启了关于电子结构方法工作组的讨论。在该工作组中，出现了一种实用的方法来模拟TiO_2个纳米结构是水分解研究最多的候选之一，包括ZnO，WO ₃，SrTiO ₃和ZnS。使用DFT，可以模拟具有多达4000个核的缩放能力的，具有数百个原子且没有周期性边界的令人印象深刻的尺寸的TiO ₂纳米粒子，并且可以根据相对于G的系统偏差来校正计算带隙的误差。₀瓦₀（5,6）迄今为止，使用基于DFT的水分解非均相催化建模方法进行的大部分分析是效率，构成吸附物的结合能和过渡金属d-位置之间的比例关系。 （7）然而，DFT可以使我们对原子数量有限的粒子建模，或者最多对表面边缘的特定终止点或一小部分进行建模。我们需要考虑几种可能具有不同稳定性的末端，例如一种考虑纳米粒子合金中所有可能的构型有序性的方法。（8）该方法基于计算每种构型所需的自由能并为该构想实现蒙特卡罗算法最稳定的安排。另一个复杂因素是纳米粒子与其底物的相互作用以及电子相互作用和它们之间的电荷传输的后果。此外，光致水分解需要处理激发态，这通常是通过与时间有关的DFT实现的，但对于大规模模拟而言，这是遥远的挑战。愿景是这些纳米颗粒还必须被溶剂化并以增加的长度比例进行处理。（9）随着比例的增加，人们需要在原子信息量和方法准确性之间取得平衡。Nicu Goga开启了关于分子动力学（MD）的工作组的讨论。具有反应性或非反应性力场的MD模拟中可以包含更多的化学信息，并且可以从DFT获得力场。MD的主要实际限制是计算力场参数所需的时间量。每次考虑使用新材料时，都需要获得可靠的力场以重现准确的DFT计算结果。一种不令人满意的解决方案是将已经开发的力场用于类似材料。此外，可以通过多粒度模拟的粗粒度超微粒参数化将MD推向其他空间和时间尺度。确实，到目前为止，社区使用DFT的比例要高于MD。在DFT中工作的大量人员的好处是，可以通过综合努力来潜在地获得大量的材料数据。但是，将DFT和MD社区联系起来更具挑战性，一种讨论的方法是在方法之间使用平滑过渡，例如紧密结合密度泛函（DFTB），这可以通过使用最重要的信息来使计算更加紧凑。另一个快速发展的领域是机器学习，可以对其进行训练以预测可在方法之间导入的有用参数，例如对从DFT数据到MD仿真的力场进行参数化。一旦掌握了MD和DFT数据，就可以在连续介质中建立微观动力学模型，就像Anja Bieberle对赤铁矿（Fe 可以训练这些参数以预测可以在方法之间导入的有用参数，例如对从DFT数据到MD仿真的力场进行参数化。一旦掌握了MD和DFT数据，就可以在连续介质中建立微观动力学模型，就像Anja Bieberle对赤铁矿（Fe 可以训练这些参数以预测可以在方法之间导入的有用参数，例如对从DFT数据到MD仿真的力场进行参数化。一旦掌握了MD和DFT数据，就可以在连续体状态下建立微观动力学模型，就像Anja Bieberle对赤铁矿（Fe₂ ö ₃）。（10）将氧释放反应（OER）的微动力学模型转换为电流，可以直接与实验数据进行比较。Juan Antonio Anta和Sofia Calero提出了带有周期边界条件的动力学蒙特卡洛模型，用于计算太阳能电池俘获位点之间的载流子扩散。（11）其中一些大规模方法，例如传热计算，缺乏化学信息这对于建模多相催化尤为重要。需要调用重要的警告，以避免在模型放大时丢失量子力学信息。对于多尺度建模，一种自然而先进的方法是将系统分成小段。Florian Libisch提出了一种嵌入方法，可以将材料分成多个区域。从而在每个区域中分别解决了电子结构问题，并且减小了问题的大小。面临的挑战是在嵌入电势中拟定电子与电子的相互作用，并选择产生嵌入电势的初始条件。一种方法是通过优化的有效电势方案求解电子密度的总和，（12）从而允许制定可用于获得唯一嵌入势的迭代程序。在COST行动中建立的不断发展的网络旨在获得用于建模纳米催化剂的统一协议。COST行动为来自不同专业领域的研究人员提供了一个框架，使他们能够联合起来，并共同努力，以不同的规模发展先进的建模工具。朝着这个宏伟目标迈出的重要一步是，记录如何对每种建模比例及其与其他比例的组合使用新颖的算法和方法。开发能够在不同理论界之间架起桥梁的新算法和思想将鼓舞人心，并且作为起点，我们将专注于在一种规模到另一种规模之间分配知识。我们的最终目标是要实现一个统一的协议，其中包括在行动期结束之前处理从第一原理到连续介质的纳米催化剂所需的算法。大部分劳动力将来自化学，物理，材料科学和计算机科学领域的理论家。该行动还将涉及一些实验研究人员，以便对我们可能希望建模的实验装置有一些构想。我们预计，这项COST行动将鼓励朝这个方向发展的项目，并在扩大计算材料科学的观点方面取得突破，从而朝着理论体系之间的共同点迈进。本“能源焦点”中表达的观点是作者的观点，不一定是ACS的观点。作者宣称没有竞争性的经济利益。本文基于COST行动18234，由COST（欧洲科技合作组织）支持。这项工作得到了南希和斯蒂芬·格兰德Technion能源计划（GTEP）的支持以及以色列科学技术部（MOST）的资助。国会议员感谢那不勒斯费德里科第二大学物理系的支持。Pancini”和化学科学系。本文引用了其他12个出版物。