Collaborative Research: Calibration of Thermochemical Models using Bayesian Methods--Building MELTS 2.0.

熔岩对地球和其他行星的结构和长期演化至关重要。 当加热到熔点以上时，岩石的性质（包括成分、密度、热含量和热膨胀）在大量的地质过程中都起着重要的作用。这些范围很广，从行星形成的最早阶段的深岩浆海洋的形成，一直到板块构造的稳步发展，负责创造，塑造和摧毁地球的海洋和大陆。 不幸的是，这些地质过程发生在行星的长度尺度和时间尺度上，在数千公里和数十亿年的时间里发挥作用。 由于这种复杂性，不可能进行直接探测行星内部演化的实验。相反，科学家们依赖于热力学模型，它可以根据在受控条件下进行的实验室实验结果预测岩石，熔体和流体的物理和能量特性。 因此，这些模型的有用性完全取决于它们的校准程度，包括实验数据的数量和种类以及用于提取热力学建模参数的统计方法。 在这项提案中，该团队将开发新的建模技术，以改进和扩展用于预测行星条件下固体和熔融岩石性质的地质热力学模型，这与我们对太阳系内外岩石行星的理解有关。合作者将设计分析工具，以便使用贝叶斯统计方法为地质界构建灵活而稳健的自洽热力学模型。 目前，热力学数据库（如流行的MELTS模型）的校准程序是一项非常繁重的任务。 更新这些模型非常耗时，而且仅限于极少数具备在不破坏模型准确性和自洽性的情况下整合新信息所需技能的专家。 该提案的目的是大大减少与重新校准相关的挑战，使各种用户能够简单快速地将新的实验数据纳入数据库。 建议的目标是：（1）设计和创建统计校准工具（2）用大量以前被忽略的数据来扩充和扩充校准数据库，包括固相缺失约束、无熔体亚固相线实验和熔体与未测成分固相共存的观察;（3）使用新的方法和数据生成MELTS 2.0，这是一种新的可调硅酸盐熔体模型，提供了模型预测的不确定性，使用户能够快速查明和解决模型的弱点。 由此产生的校准工具将能够生成和可视化模型预测分布-用于地质过程建模，教学和未来的实验规划-并有助于缩小模型使用和模型设计之间仍然存在的巨大差距。

Molten rocks are essential to the structure and long-term evolution of the Earth and other planets. The properties of rocks when heated above their melting point (including composition, density, heat content, and thermal expansion) all play important roles in a vast array of geologic processes. These range widely from the creation of deep magma oceans during the earliest stages of planet formation all the way to the steady progress of plate tectonics, responsible for creating, shaping, and destroying the Earth's oceans and continents. Unfortunately, these geologic processes occur over planetary length-scales and time-scales, playing out over thousands of kilometers and billions of years. Due to this complexity, it is not possible to perform experiments that directly probe the evolution of planetary interiors. Instead, scientist rely upon thermodynamic models, which can predict the physical and energetic properties of rocks, melts, and fluids based on the results of laboratory experiments performed under controlled conditions. The usefulness of these models thus depends entirely upon how well they are calibrated, including the amount and variety of experimental data as well as the statistical methods used to extract the thermodynamic modeling parameters. In this proposal, the team will develop new modeling techniques needed to improve and extend the geologic thermodynamic models used to predict the properties of solid and molten rocks at planetary conditions, relevant to our understanding of rocky planets both within and outside our solar system.To accomplish these overarching goals, the collaborators will design the analytic tools needed to construct flexible and robust self-consistent thermodynamic models for the geologic community using Bayesian statistical methods. Currently, the calibration procedure for thermodynamic databases (like the popular MELTS model) is an unfortunately onerous task. Updating these models is time-consuming and restricted to the very few experts with the skills required to integrate new information without breaking the accuracy and self-consistency of the model. The purpose of this proposal is to dramatically reduce the challenges associated with recalibration, enabling simple and rapid incorporation of new experimental data into the database by a wide variety of users. The proposal objectives are: (1) Design and create statistical calibration tools (using novel Bayesian techniques) to simplify thermochemical model building; (2) Expand and augment the calibration database with the large quantity of previously ignored data, including solid phase-absent constraints, melt-free sub-solidus experiments, and observations of melt coexisting with solid phases of unmeasured-composition; (3) Use the new methods and data to produce MELTS 2.0, a new tunable silicate melts model that provides model prediction uncertainties, enabling users to rapidly pinpoint and address model weaknesses. The resulting calibration tool will thus be able to generate and visualize model prediction distributions - useful for geologic process modeling, teaching, and future experimental planning -and help to close the large gap that still exists between model-use and model-design.