目前,地球物理反演问题参数化常常有10~4至10~5个未知数,与此相关的约束条件多达10~6至10~7个。由这一线性系统所确定的矩阵问题((?)_m=b)常常用最小方准则(m=(A~+A)~(-1)(?)~tb)来求解。这样大的矩阵很难直接求解,研究人员一般改用共轭梯度一类的迭代法来求取最小平方解。这些迭代法利用了(?)为稀疏矩阵这一特点(它只有2％-3％的非零元素,而无需计算(或存储)矩阵(?)~tA)。尽管数据约束条件常常多于未知数, At present, the parameterization of geophysical inversion problems often has 10 ~ 4 to 10 ~ 5 unknowns, with the related constraints of up to 10 ~ 6 to 10 ~ 7. The matrix problem ((?) _ M = b) determined by this linear system is often solved by the least square criterion (m = (A ~ + A) -1 (?) ~ Tb) Such a large matrix is ​​difficult to solve directly, researchers generally use a class of iterative methods such as conjugate gradient to find the least square solution. These iterative methods take advantage of the (?) Sparse matrix (which has only 2% -3% of nonzero elements without the need to compute (or store) the matrix (t)). Although data constraints are often more than unknown,