DATA-DRIVEN PARALLELIZATION OF TIME IN MOLECULAR DYNAMICS SIMULATIONS

这个子项目是许多研究子项目中利用 资源由NIH/NCRR资助的中心拨款提供。子项目和 调查员(PI)可能从NIH的另一个来源获得了主要资金， 并因此可以在其他清晰的条目中表示。列出的机构是 该中心不一定是调查人员的机构。 传统的并行化策略不能很好地扩展计算工作量，因为需要模拟很长的时间跨度，而不是来自大的状态空间。分子动力学模拟构成了一类重要的应用，其中这被证明是一个瓶颈。我们正在开发一种分子动力学并行化的新方法，这是基于这样的观察，即模拟通常发生在来自其他相关模拟的丰富数据的背景下。我们使用这些数据来并行化时间域，从而产生更具可伸缩性的算法。这种方法是基于这样的观察，即在具有多个时间尺度的模拟中经常遇到长时间跨度。细微尺度是造成大量计算工作量的原因。然而，细尺度的重要贡献往往是它们对粗尺度的影响。我们使用降阶模型来识别重要的粗尺度效应。我们使用聚类和机器学习来动态确定正在执行的模拟和先前数据之间的关系。这使我们能够预测未来时间的合理初始状态，这可以用于并行化时间域。

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. Conventional parallelization strategies do not scale well when the computational effort arises from the need to simulate to long time spans, rather than from large state space. Molecular Dynamics simulations constitute an important class of applications where this proves to be a bottleneck. We are developing a new approach to parallelization of Molecular Dynamics, which is based on the observation that simulations typically occur in a context rich in data from other related simulations. We use such data to parallelize the time domain, which yields a more scalable algorithm. This approach is based on the observation that long time-spans are often encountered in simulations with multiple time scales. The fine scales are responsible for the large computational effort. However, the important contribution of the fine-scales is often to the effect they have on the coarse scale. We use reduced order modeling to identify important coarse scale effects. We use clustering and machine learning to dynamically determine the relationship between the simulation being performed and prior data. This enables us to predict reasonable initial states at future times, which can be used to parallelize the time domain.