很牛的牛顿迭代法


在MIT公开课《计算机科学与编程导论》的第五讲中,讲到编写求解平方根的函数sqrt时,提到了牛顿迭代法。今天仔细一查,发现这是一个用途很广、很牛的计算方法。

首先,考虑如何编写一个开平方根的函数sqrt(float num, float e)。参数num是要求开平方根的实数,参数e是计算结果可以达到多大误差。这是一个无法得到精确解,只能求出近似解的问题。该如何编写呢?


1. 传统的二分法

我们可以先猜测一个值作为解,看这个值是否在误差范围内。如果没有达到误差要求,就构造一个更好的猜测值,继续迭代。猜测值可以简单地初始化为num/2,但怎样在下一轮迭代前构造一个更好的猜测值呢?我们不妨参照二分查找算法的思路,解的初始范围是[0, num],用二分法逐渐缩小范围。

     private static float sqrt(float num, float e) {
              
              float low = 0F;
              float high = num;
              float guess, e0;
              int count = 0;
              
              do {
                     guess = (low + high) / 2;
                     if (guess*guess > num) {
                           high = guess;
                           e0 = guess*guess - num;
                     } else {
                           low = guess;
                           e0 = num - guess*guess;
                     }
                     
                     count++;
                     System.out.printf("Try %f, e: %f\n", guess, e0);
              } while (e0 > e);

              System.out.printf("Try %d times, result: %f\n", count, guess);
              
              return guess;
       }

在区间[low, high]内取中点(low+high)/2作为猜测值。如果guess*guess大于num,说明猜测值偏大,则在区间[low, guess]进行下一轮迭代,否则在区间[guess, high]中继续。当误差值e0小于能够接受的误差值e时停止迭代,返回结果。

取num=2, e=0.01进行测试,运行结果如下:
Try 1.000000, e: 1.000000
Try 1.500000, e: 0.250000
Try 1.250000, e: 0.437500
Try 1.375000, e: 0.109375
Try 1.437500, e: 0.066406
Try 1.406250, e: 0.022461
Try 1.421875, e: 0.021729
Try 1.414063, e: 0.000427
Try 8 times, result: 1.414063
可见尝试了八次才达到0.01的误差。


2. 神奇的牛顿法

仔细思考一下就能发现,我们需要解决的问题可以简单化理解。
从函数意义上理解:我们是要求函数f(x) = x²,使f(x) = num的近似解,即x² - num = 0的近似解
从几何意义上理解:我们是要求抛物线g(x) = x² - num与x轴交点(g(x) = 0)最接近的点。

我们假设g(x0)=0,即x0是正解,那么我们要做的就是让近似解x不断逼近x0,这是函数导数的定义:



可以由此得到



从几何图形上看,因为导数是切线,通过不断迭代,导数与x轴的交点会不断逼近x0。




3. 牛顿法的实现与测试

     public static void main(String[] args) {
              float num = 2;
              float e = 0.01F;
              sqrt(num, e);
              sqrtNewton(num, e);
              
              num = 2;
              e = 0.0001F;
              sqrt(num, e);
              sqrtNewton(num, e);
              
              num = 2;
              e = 0.00001F;
              sqrt(num, e);
              sqrtNewton(num, e);
       }

     private static float sqrtNewton(float num, float e) {
              
              float guess = num / 2;
              float e0;
              int count = 0;
              
              do {
                     guess = (guess + num / guess) / 2;
                     e0 = guess*guess - num;
                     
                     count++;
                     System.out.printf("Try %f, e: %f\n", guess, e0);
              } while (e0 > e);

              System.out.printf("Try %d times, result: %f\n", count, guess);
              
              return guess;
       }

与二分法的对比测试结果:

Try 1.000000, e: 1.000000
Try 1.500000, e: 0.250000
Try 1.250000, e: 0.437500
Try 1.375000, e: 0.109375
Try 1.437500, e: 0.066406
Try 1.406250, e: 0.022461
Try 1.421875, e: 0.021729
Try 1.414063, e: 0.000427
Try 8 times, result: 1.414063

Try 1.500000, e: 0.250000
Try 1.416667, e: 0.006945
Try 2 times, result: 1.416667

Try 1.000000, e: 1.000000
Try 1.500000, e: 0.250000
Try 1.250000, e: 0.437500
Try 1.375000, e: 0.109375
Try 1.437500, e: 0.066406
Try 1.406250, e: 0.022461
Try 1.421875, e: 0.021729
Try 1.414063, e: 0.000427
Try 1.417969, e: 0.010635
Try 1.416016, e: 0.005100
Try 1.415039, e: 0.002336
Try 1.414551, e: 0.000954
Try 1.414307, e: 0.000263
Try 1.414185, e: 0.000082
Try 14 times, result: 1.414185

Try 1.500000, e: 0.250000
Try 1.416667, e: 0.006945
Try 1.414216, e: 0.000006
Try 3 times, result: 1.414216

Try 1.000000, e: 1.000000
Try 1.500000, e: 0.250000
Try 1.250000, e: 0.437500
Try 1.375000, e: 0.109375
Try 1.437500, e: 0.066406
Try 1.406250, e: 0.022461
Try 1.421875, e: 0.021729
Try 1.414063, e: 0.000427
Try 1.417969, e: 0.010635
Try 1.416016, e: 0.005100
Try 1.415039, e: 0.002336
Try 1.414551, e: 0.000954
Try 1.414307, e: 0.000263
Try 1.414185, e: 0.000082
Try 1.414246, e: 0.000091
Try 1.414215, e: 0.000004
Try 16 times, result: 1.414215

Try 1.500000, e: 0.250000
Try 1.416667, e: 0.006945
Try 1.414216, e: 0.000006
Try 3 times, result: 1.414216

可以看到随着对误差要求的更加精确,二分法的效率很低下,而牛顿法的确非常高效。
可在两三次内得到结果。

如果搞不清牛顿法的具体原理,可能就要像我一样复习下数学知识了。极限、导数、泰勒展开式、单变量微分等。


4. 更快的方法

在Quake源码中有段求sqrt的方法,大概思路是只进行一次牛顿迭代,得到能够接受误差范围内的结果。
因此肯定是更快的。

float Q_rsqrt( float number )
{
  long i;
  float x2, y;
  const float threehalfs = 1.5F;

  x2 = number * 0.5F;
  y  = number;
  i  = * ( long * ) &y;  // evil floating point bit level hacking
  i  = 0x5f3759df - ( i >> 1 ); // what the fuck?
  y  = * ( float * ) &i;
  y  = y * ( threehalfs - ( x2 * y * y ) ); // 1st iteration
  // y  = y * ( threehalfs - ( x2 * y * y ) ); // 2nd iteration, this can be removed

  #ifndef Q3_VM
  #ifdef __linux__
    assert( !isnan(y) ); // bk010122 - FPE?
  #endif
  #endif
  return y;
}


参考文章

牛顿迭代方程的近似解 http://blueve.me/archives/369


posted on 2011-12-26 22:33  毛小娃  阅读(251)  评论(0编辑  收藏  举报

导航