Stable isotope fractionation in minerals as a tool for climate reconstruction: insights from molecular modelling

自然气候档案（如海洋沉积物、极地冰盖或洞穴）中稳定同位素比例的细微变化可以被精确测量，以重建气候历史。例如，通过测量在海洋沉积物中发现的贝壳中氧同位素的比例，可以确定海水温度和全球冰量随时间的变化，因为同位素比例受到这两个因素的影响。同位素记录的解释需要对控制同位素交换和分馏的复杂过程有详细的了解，这推动了理论模型的发展，以预测和合理地解释不同相之间稳定同位素的分馏。这种方法通常基于分子水平的考虑：原子的振动行为取决于所涉及的同位素质量，影响平衡自由能和动力学势垒。基于量子化学或经典力场的分子建模技术可以用来预测相之间稳定同位素的分馏。虽然近年来在这一研究方向上取得了重大进展，但仍然存在一些重要的挑战，因为必须经常进行激烈的近似，以避免原子级模拟的巨大计算成本：例如：1)原子间相互作用被简化；Ii)同位素分馏预测只考虑矿物的体积行为，忽略了矿物-水界面；非平衡（动力学）效应通常被忽略。在这个项目中，我们将利用最近分子建模的主要进展来克服这些限制，实现对稳定同位素分馏的更快、更准确的预测，并有可能改变同位素记录的解释。我们将采用分子动力学模拟来模拟矿物和水溶液之间的同位素交换，考虑到：i)原子间相互作用的精确量子力学表示；Ii)矿物-水界面以及粒径对整体分馏比的影响；Iii)非平衡和动力学效应。为了承担这种改进所需要的计算成本的急剧增加，我们将把量子化学与机器学习（ML）技术结合起来，后者用于在不牺牲显著精度的情况下大大加速前者。

Subtle changes in the ratio of stable isotopes in natural climate archives such as sediments in oceans, polar ice caps or speleothems can be precisely measured to reconstruct climate history. For example, by measuring the ratio of oxygen isotopes in shells found in marine sediments, it is possible to determine changes in seawater temperature and global ice volume over time, because the isotope ratio is affected by both factors. The interpretation of isotope records requires a detailed understanding of the complex processes governing isotope exchange and fractionation, which has motivated the development of theoretical models to predict and rationalise the fractionation of stable isotopes between different phases. Such approaches are typically based on molecular-level considerations: the vibrational behaviour of atoms depends on the isotopic masses involved, affecting equilibrium free energies and kinetic barriers. Molecular modelling techniques based on quantum chemistry or classical forcefields can then be used to predict the fractionation of stable isotopes between phases. While significant progress has been achieved in recent years in this research direction, some important challenges remain, because drastic approximations must often be made to avoid the huge computational cost of the atomistic-level simulations: e.g.: i) interatomic interactions are simplified; ii) isotope fractionations are predicted considering only the mineral bulk behaviour, neglecting the mineral-aqueous interface; and iii) non-equilibrium (kinetic) effects are usually ignored. In this project, we will take advantage of recent major advances in molecular modelling to overcome these limitations and achieve a faster and more accurate prediction of stable isotope fractionation, with the potential to transform the interpretation of isotope records. We will employ molecular dynamics simulations to model isotope exchange between minerals and aqueous solution, considering: i) accurate quantum-mechanical representation of interatomic interactions; ii) the effect of the mineral-aqueous interface and therefore of particle size on the overall fractionation ratios; iii) non-equilibrium and kinetic effects. To afford the drastic increase in computational cost that such improvements require, we will combine quantum chemistry with machine-learning (ML) techniques, where the latter are used to vastly accelerate the former without significant accuracy sacrifice.