标签:
牛顿迭代法是一个可以求一个任意函数的零点的工具。它比二分法快得多。
公式是:x=a-f(a)/f‘(a)。其中a是猜测值,x是新的猜测值。不断迭代,f(x)就越来越接近0。
我们将f(x)做泰勒一阶展开:f(x)∼f(a)+(x-a)f‘(a)。
令f(x)=0,
∴f(a)+(x-a)f‘(a) = 0
∴f(a)+xf‘(a)-af‘(a) = 0
∴xf‘(a)=af‘(a)-f(a)
∴x=a-f(a)/f‘(a)
∵ | x = √2 |
∴ | x2 = 2 |
∴ | x2 -2 = 0 |
令f(x)=方程左边,则f(x)∼0↔x∼√2。
f‘(x) = 2x。于是可以得到迭代公式:
x | |
= | a-f(a)/f‘(a) |
= | a-(a2-2)/(2a) |
= | a-a/2+1/a |
= | a/2+1/a |
代码如下(要求误差小于1e-6):
#include <stdio.h> #include <math.h> int main(int argc, char const *argv[]) { double a = 2.0; double expect_error = 0.000001; double expect_answer = 1.4142135623731; double x; double actual_error; unsigned iteration_count = 0; do { if (a == 0.0) a = 0.1; /* 避免除0 */ x = a/2 + 1/a; actual_error = fabs(expect_answer - x); a = x; ++iteration_count; printf("%d\t%.9f\t%.9f\n", iteration_count, a, actual_error); } while (actual_error >= expect_error); printf("%d\n", iteration_count); return 0; }
输出:
1 1.500000000 0.085786438 2 1.416666667 0.002453104 3 1.414215686 0.000002124 4 1.414213562 0.000000000 4
迭代了4次。用二分法呢?
#include <stdio.h> #include <math.h> int main(int argc, char const *argv[]) { double high = 2.0; double low = 1.0; double expect_error = 0.000001; double expect_answer = 1.4142135623731; double x; double actual_error; unsigned iteration_count = 0; do { x = (high+low)/2; if (x*x-2 > 0) high = x; else low = x; actual_error = fabs(expect_answer - x); ++iteration_count; printf("%d\t%.9f\t%.9f\n", iteration_count, x, actual_error); } while (actual_error >= expect_error); printf("%d\n", iteration_count); return 0; }
输出:
1 1.500000000 0.085786438 2 1.250000000 0.164213562 3 1.375000000 0.039213562 4 1.437500000 0.023286438 5 1.406250000 0.007963562 6 1.421875000 0.007661438 7 1.414062500 0.000151062 8 1.417968750 0.003755188 9 1.416015625 0.001802063 10 1.415039062 0.000825500 11 1.414550781 0.000337219 12 1.414306641 0.000093078 13 1.414184570 0.000028992 14 1.414245605 0.000032043 15 1.414215088 0.000001526 16 1.414199829 0.000013733 17 1.414207458 0.000006104 18 1.414211273 0.000002289 19 1.414213181 0.000000382 19
迭代了19次。
标签:
原文地址:http://www.cnblogs.com/destro/p/newton-s-method.html