it is straightforward to see that e

it is straightforward to see that each dimension may be separately applied. This reduces the cost of applying the operator to O(Mk4 ) and reduces the storage to O(Mk2 ). The computational cost may be further reduced by using singular value decomposition to exploit the low operator rank away from the singularity. The length of the expansion (M ) is referred to as the separation rank. It is critical that this rank be as small as possible, and also that the coefficients in
the expansion be small and of the same sign so as to avoid loss of precision.

結果 (中文) 1:

它很容易看到，每个维度可能分别适用。这将减少到 O (Mk4) 应用运算符的成本并减少存储到 O (mk2 型)。基于奇异值分解，利用低操作员等级从奇点可能进一步减少计算成本。扩展 (M) 的长度称为分离等级。至关重要的是这种等级是一样小的可能，而且作为中的系数扩展会较小，且以避免丢失精度相同的符号。

結果 (中文) 3:

这是直接看到，每个维度可以单独应用。这减少了应用运营商成本O（Mk4）降低了存储的O（MK2）。利用奇异值分解，利用奇异值分解，利用奇异值分解，可以进一步降低计算成本。扩张（米）的长度被称为作为分离等级。这是至关重要的，这等级是尽可能小的，而且，在的系数中，扩大是小的和相同的符号，以避免损失的精度。