指出了（阻尼）最小二乘法能够成功解决反问题的本质，然而它并不总是最优准则。如果误差服从拉普拉斯分布，那么最小绝对值准则最优，如果误差近似均匀分布，则应选极小极大准则。首次提出的两个准则可一般地评价或选择最优模，最优模使后验概率取极大值或使某种差值取极小值，最后的数值例子说明了这些结论 It is pointed out that the (damping) least square method can successfully solve the inverse problem, however, it is not always the optimal criterion. If the error obeys the Laplace distribution, then the minimum absolute value criterion is optimal, and if the errors are approximately uniformly distributed, then a minimum criterion should be chosen. The first two proposed criteria can generally evaluate or select the optimal model, the best model to make the posterior probability take the maximum or make a certain difference between the minimum, the last numerical examples illustrate these conclusions