数值运算的误差估计
近似数 x i ∗ x_{\mathrm{i}}^{*} xi∗ 与 x 2 ∗ x_{2}^{*} x2∗,误差限 ε ( x 1 ∗ ) (x_1^{*}) (x1∗)及 ε ( x 2 ∗ ) (x_2^{*}) (x2∗)
进行加减运算得到的误差限为
ε ( x 1 ∗ ± x 2 ∗ ) = ε ( x 1 ∗ ) + ε ( x 2 ∗ ) , \varepsilon(x_1^*\pm x_2^*)=\varepsilon(x_1^*)+\varepsilon(x_2^*)\:, ε(x1∗±x2∗)=ε(x1∗)+ε(x2∗),
进行乘法运算得到的误差限为
ε ( x 1 ∗ x 2 ∗ ) ≈ ∣ x 1 ∗ ∣ ε ( x 2 ∗ ) + ∣ x 2 ∗ ∣ ε ( x 1 ∗ ) , \varepsilon(x_1^*x_2^*)\approx\mid x_1^*\mid\varepsilon(x_2^*)+\mid x_2^*\mid\varepsilon(x_1^*), ε(x1∗x2∗)≈∣x1∗∣ε(x2∗)+∣x2∗∣ε(x1∗),
进行除法运算得到的误差限为
ε
(
x
1
∗
x
2
∗
)
≈
∣
x
1
∗
∣
ε
(
x
2
∗
)
+
∣
x
2
∗
∣
ε
(
x
1
∗
)
∣
x
2
∗
∣
2
(
x
2
∗
≠
0
)
.
\varepsilon\left(\frac{x_1^*}{x_2^*}\right)\approx\frac{\mid x_1^*\mid\varepsilon(x_2^*)+\mid x_2^*\mid\varepsilon(x_1^*)}{\mid x_2^*\mid^2}\quad(x_2^*\neq0).
ε(x2∗x1∗)≈∣x2∗∣2∣x1∗∣ε(x2∗)+∣x2∗∣ε(x1∗)(x2∗=0).
更一般的情况是，当自变量有误差时，计算函数值也会产生误差。
这种误差限可利用函数的 Taylor 展开式进行估计。
设 f ( x ) f(x) f(x)是一元函数， x x x 的近似值为 x ∗ x^* x∗，以 f ( x ∗ ) f(x^*) f(x∗)近似 f ( x ) f(x) f(x)，其误差限记作 ε ( f ( x ∗ ) ) \varepsilon(f(x^*)) ε(f(x∗))，可用 Taylor 展开
f ( x ) − f ( x ∗ ) = f ′ ( x ∗ ) ( x − x ∗ ) + f ′ ′ ( ξ ) 2 ( x − x ∗ ) 2 , f(x)-f(x^*)=f'(x^*)(x-x^*)+\frac{f''(\xi)}{2}(x-x^*)^2, f(x)−f(x∗)=f′(x∗)(x−x∗)+2f′′(ξ)(x−x∗)2,
ξ \xi ξ介于 x , x ∗ x,x^* x,x∗之间。
取绝对值得
∣ f ( x ) − f ( x ∗ ) ∣ ⩽ ∣ f ′ ( x ∗ ) ∣ ε ( x ∗ ) + ∣ f ′ ′ ( ξ ) ∣ 2 ε 2 ( x ∗ ) . \mid f(x)-f(x^*)\mid\leqslant\mid f^{\prime}(x^*)\mid\varepsilon(x^*)+\frac{\mid f^{\prime\prime}(\xi)\mid}{2}\varepsilon^2(x^*). ∣f(x)−f(x∗)∣⩽∣f′(x∗)∣ε(x∗)+2∣f′′(ξ)∣ε2(x∗).
假定
f
′
(
x
∗
)
f^\prime(x^*)
f′(x∗)与
f
′
′
(
x
∗
)
f^{\prime\prime}(x^*)
f′′(x∗)的比值不太大，可忽略 ε
(
x
∗
)
(x^*)
(x∗)的高阶项，于是可得计算函数的误差限为
(
f
(
x
∗
)
)
≈
∣
f
′
(
x
∗
)
∣
ε
(
x
∗
)
.
(f(x^{*}))\approx\mid f^{\prime}(x^{*})\mid\varepsilon(x^{*}).
(f(x∗))≈∣f′(x∗)∣ε(x∗).
当 f f f为多元函数时计算 A = f ( x 1 , x 2 , ⋅ ⋅ ⋅ , x n ) A=f(x_1,x_2,\cdotp\cdotp\cdotp,x_n) A=f(x1,x2,⋅⋅⋅,xn)，如果 x 1 , x 2 , ⋅ ⋅ ⋅ , x n x_1,x_2,\cdotp\cdotp\cdotp,x_n x1,x2,⋅⋅⋅,xn的近似值为 x 1 ∗ , x 2 ∗ ⋯ , x n ∗ x_1^*,x_2^*\cdots,x_n^* x1∗,x2∗⋯,xn∗，则 A A A 的近似值为 A ∗ = f ( x 1 ∗ , x 2 ∗ , ⋯ , x n ∗ ) A^*=f(x_1^*,x_2^*,\cdots,x_n^*) A∗=f(x1∗,x2∗,⋯,xn∗)，于是函数值 A ∗ A^* A∗的误差 e ( A ∗ ) e(A^*) e(A∗)由 Taylor 展开，得
e ( A ∗ ) = A ∗ − A = f ( x 1 ∗ , x 2 ∗ , ⋯ , x n ∗ ) − f ( x 1 , x 2 , ⋯ , x n ) ≈ ∑ k = 1 n ( ∂ f ( x 1 ∗ , x 2 ∗ , ⋯ , x n ∗ ) ∂ x k ) ( x k ∗ − x k ) = ∑ k = 1 n ( ∂ f ∂ x k ) ∗ e k ∗ \begin{aligned}&e(A^*)=A^*-A=f(x_1^*,x_2^*,\cdots,x_n^*)-f(x_1,x_2,\cdots,x_n) \\&\approx\sum_{k=1}^n\Big(\frac{\partial f(x_1^*,x_2^*,\cdots,x_n^*)}{\partial x_k}\Big)(x_k^*-x_k)\:=\:\sum_{k=1}^n\Big(\frac{\partial f}{\partial x_k}\Big)^*e_k^*\ \end{aligned} e(A∗)=A∗−A=f(x1∗,x2∗,⋯,xn∗)−f(x1,x2,⋯,xn)≈k=1∑n(∂xk∂f(x1∗,x2∗,⋯,xn∗))(xk∗−xk)=k=1∑n(∂xk∂f)∗ek∗
于是误差限为
ε ( A ∗ ) ≈ ∑ k = 1 n ∣ ( ∂ f ∂ x k ) ∗ ∣ ε ( x k ∗ ) \varepsilon(A^*)\approx\sum_{k=1}^n\left|\left(\frac{\partial f}{\partial x_k}\right)^*\right|\varepsilon(x_k^*) ε(A∗)≈k=1∑n (∂xk∂f)∗ ε(xk∗)
而 A ∗ A^* A∗的相对误差限为
ε r ∗ = ε r ( A ∗ ) = ε ( A ∗ ) ∣ A ∗ ∣ ≈ ∑ k = 1 k ∣ ( ∂ f ∂ x k ) ∗ ∣ ε ( x k ∗ ) ∣ A ∗ ∣ \varepsilon_{\mathrm{r}}^{*}=\varepsilon_{\mathrm{r}}(A^{*})=\frac{\varepsilon(A^{*})}{\mid A^{*}\mid}\approx\sum_{k=1}^{k}\left|\left(\frac{\partial f}{\partial x_{k}}\right)^{*}\right|\frac{\varepsilon(x_{k}^{*})}{\mid A^{*}\mid} εr∗=εr(A∗)=∣A∗∣ε(A∗)≈k=1∑k (∂xk∂f)∗ ∣A∗∣ε(xk∗)
扫一扫
4340
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
