论文部分内容阅读
在对MG(1996a,b)的答复中,我们总结了他们的主要缺陷。由于这一缺陷,他们的这次和1992年的计算毫无疑问是错误的。对MG(1996a,b)提出问题的逐一答复见答复之二。 MG(1992)的计算存在以下问题:(1)正如Rhoades和Evison(1996)所指出的,MG的计算存在一个很严重的错误,他们在计算中把概率值定为高达11,这违反了概率的定义。其做法类似于把在n次掷币中至少一次得到正面的概率近似为在n次掷币中得到正面的期望值,而这种概率只是n=1时的期望值。(2)没有考虑到每一例VAN预测都对应一个确定的空间范围这一事实。(3)当用虚拟的理想完美的地震预报方法IPEPM(它被定义为成功的、在一个完整地域内对某一震级以上、如,M_S≥5.3,和没有任何虚假警报发布的所有地震的预报)时,它导致一个谬论,即就置信水平值而言,这些理想的预报可以归因于偶然。 MG在这两篇文章中没有为其在1992年的早期工作提供支持,仍然继承了上述错误。而他们这次宣称发现了VAN预报与地震之间的重要的反向时间相关关系。我们发现,除了他们文中表1的明显错误外,他们的这一结论纯粹是这样产生的,即虽然他们使用了泊松分布,但却违背了这一原理,把依赖于时间的事件引入了他们的计算(例如,对于M_(EQ)≥5.3,当把依赖于时间的事件考虑进去时,有6个正向时间相关,6个反向时间相关;而只考虑独立的事件时,有5个正向时间相关,只有2个反向时间相关)。我们用简单的例子证明了MG(1996a)的算法导致了第二个谬误:它甚至可以从IPEPM中导出反向时间相关,而IPEPM可以成功预报的不只有主震,还有与主震相关的余震。更进一步,MG进行了一个错误假设,不但破坏了试验误差的真实含义,而且人为地将VAN预报的成功率降低了近2倍。我们的主要结论是:既然MG的统计过程导致对理想状态得出错误结果,它就不能作为一个工具去对一个实验方法进行评判。
In our reply to MG (1996a, b), we summarized their main weaknesses. Due to this flaw, their calculations this time and 1992 are undoubtedly wrong. For a one-by-one answer to the questions raised by MG (1996a, b), see answer 2. (1) As noted by Rhoades and Evison (1996), there is a very serious error in the computation of MG. They set the probability value up to 11 in the calculation, which violates the probability Definition. This is done by approximating the probability of getting positive at least once in n twitches to a positive expectation in n tossings, and this probability is only the expectation at n = 1. (2) The fact that each case of VAN prediction corresponds to a certain spatial extent is not considered. (3) When using IPEPM (which is defined as a perfect ideal earthquake prediction method based on a virtual ideal, it is defined as a prediction of all earthquakes issued above a certain magnitude or above in a complete region, such as M_S ≧ 5.3, and without any false alarm ), It leads to the fallacy that these ideal predictions can be attributed to incidental values in terms of confidence levels. MG did not provide support for its early work in 1992 in both articles and still inherited the above error. And this time they claimed to have found significant back-time correlations between VAN forecasts and earthquakes. We find that, except for the obvious mistakes in Table 1 in their text, their conclusion is purely derived from the fact that although they used the Poisson distribution, they violated this principle and introduced time-dependent events into them (For example, for M_EQ ≥ 5.3, there are 6 forward time correlations and 6 reverse time correlations when time dependent events are taken into account, whereas for independent events only 5 Forward time related, only 2 reverse time related). We demonstrate with simple examples that the algorithm of MG (1996a) leads to a second fallacy: it can even derive backward time correlation from IPEPM, while IPEPM can predict not only the main shocks but also the main shocks aftershock. Furthermore, MG made a false assumption that not only undermined the true meaning of the experimental error, but also artificially reduced the success rate of VAN forecast by nearly two times. Our main conclusion is that MG can not be used as a tool to judge an experimental method since the statistical process of MG leads to the wrong result on the ideal state.