Thermal history modelling: HeFTy vs. QTQt
Thermal history modelling: HeFTy vs. QTQt
复制标题
DOI：
10.1016/j.earscirev.2014.09.010
复制
发表时间：
2014-12-01
影响因子：
12.1
通讯作者：
Tian, Yuntao
中科院分区：
文献类型：
--
作者：
Vermeesch, Pieter;Tian, Yuntao
HeFTy是一个流行的热历史建模程序，它以一个垃圾袋品牌命名，提醒人们“垃圾进，垃圾出”的原则。QTQt是一个替代程序，它的名字是指它能够从复杂的热时序数据集中提取具有视觉吸引力（“可爱”）的时间-温度路径。本文将对这两个程序进行比较和对比，并通过一些简单的例子来解释这些“黑盒”的算法基础。这两个代码都包含“正向”和“反向”建模功能。“正演模型”允许用户预测任何给定热历史的预期数据分布。“逆模型”找到与某些输入数据最匹配的热历史。heavy和QTQt基于相同的物理原理，因此它们的前向建模功能几乎相同。相比之下，它们的逆建模算法是根本不同的，具有重要的后果。HeFTy采用了一种“频率主义”方法，即通过形式化的统计假设检验来评估输入数据与热模型预测之间的拟合度。QTQt使用贝叶斯“马尔可夫链蒙特卡罗”（MCMC）算法，其中随机遍历模型空间导致“最可能”热历史的集合。原则上，Frequentist方法的主要优点是它包含一个内置的质量控制机制，可以检测坏数据（“垃圾”），并保护新手用户避免应用不适当的模型。然而，在实践中，由于Frequentist算法对样本大小的不良敏感性，这种质量控制机制不适用于小型或不精确的数据集，当数据集足够大或足够精确时，这会导致HeFTy“中断”。QTQt不会受到这个问题的困扰，因为它的性能会随着样本量的增加而提高。然而，MCMC方法的健壮性也存在风险，因为QTQt将接受物理上不可能的数据集，并为它们提出“最佳拟合”的热历史。这在新手手中可能是危险的。总之，“HeFTy”这个名字更适合QTQt，反之亦然。(C) 2014年作者。Elsevier B.V.出版
HeFTy is a popular thermal history modelling program which is named after a brand of trash bags as a reminder of the 'garbage in, garbage out' principle. QTQt is an alternative program whose name refers to its ability to extract visually appealing ('cute') time-temperature paths from complex thermochronological datasets. This paper compares and contrasts the two programs and aims to explain the algorithmic underpinnings of these 'black boxes' with some simple examples. Both codes consist of 'forward' and 'inverse' modelling functionalities. The 'forward model' allows the user to predict the expected data distribution for any given thermal history. The 'inverse model' finds the thermal history that best matches some input data. HeFTy and QTQt are based on the same physical principles and their forward modelling functionalities are therefore nearly identical. In contrast, their inverse modelling algorithms are fundamentally different, with important consequences. HeFTy uses a 'Frequentist approach, in which formalised statistical hypothesis tests assess the goodness-of-fit between the input data and the thermal model predictions. QTQt uses a Bayesian `Markov Chain Monte Carlo' (MCMC) algorithm, in which a random walk through model space results in an assemblage of 'most likely' thermal histories. In principle, the main advantage of the Frequentist approach is that it contains a built-in quality control mechanism which detects bad data ('garbage') and protects the novice user against applying inappropriate models. In practice, however, this quality-control mechanism does not work for small or imprecise datasets due to an undesirable sensitivity of the Frequentist algorithm to sample size, which causes HeFTy to 'break' when datasets are sufficiently large or precise. QTQt does not suffer from this problem, as its performance improves with increasing sample size in the form of tighter credibility intervals. However, the robustness of the MCMC approach also carries a risk, as QTQt will accept physically impossible datasets and come up with 'best fitting' thermal histories for them. This can be dangerous in the hands of novice users. In conclusion, the name 'HeFTy' would have been more appropriate for QTQt, and vice versa. (C) 2014 The Authors. Published by Elsevier B.V.