这可以用深度为O(3^(n))的树来求解。其中n是所有平方和。

首先考虑用O(9^n)树来解决问题是很简单的，只需考虑所有可能的爆炸位置。有关示例，请参阅Alfe的实现。

接下来我们意识到，我们可以从下往上轰炸，仍然得到一个最小的轰炸模式。

Start from the bottom left corner. Bomb it to oblivion with the only plays that make sense (up and to the right). Move one square to the right. While the target has a value greater than zero, consider each of the 2 plays that make sense (straight up or up and to the right), reduce the value of the target by one, and make a new branch for each possibility. Move another to the right. While the target has a value greater than zero, consider each of the 3 plays that make sense (up left, up, and up right), reduce the value of the target by one, and make a new branch for each possibility. Repeat steps 5 and 6 until the row is eliminated. Move up a row and repeat steps 1 to 7 until the puzzle is solved.

这个算法是正确的，因为

有必要在某一时刻完成每一行。 完成一行总是需要一个游戏，一个在上面，一个在下面，或者在这一行内。 选择在未清除的最低行之上的玩法总是比选择在该行之上或该行之下的玩法更好。

在实践中，这个算法通常会比它的理论最大值做得更好，因为它会定期轰炸邻居并减少搜索的大小。如果我们假设每次轰炸都会减少4个额外目标的价值，那么我们的算法将运行在O(3^(n/4))或大约O(1.3^n)。

Because this algorithm is still exponential, it would be wise to limit the depth of the search. We might limit the number of branches allowed to some number, X, and once we are this deep we force the algorithm to choose the best path it has identified so far (the one that has the minimum total board sum in one of its terminal leaves). Then our algorithm is guaranteed to run in O(3^X) time, but it is not guaranteed to get the correct answer. However, we can always increase X and test empirically if the trade off between increased computation and better answers is worthwhile.

这是一个广度搜索，通过这个“迷宫”的位置寻找最短路径(一系列轰炸)。不，我不能证明没有更快的算法，抱歉。

#!/usr/bin/env python

M = ((1,2,3,4),
     (2,3,4,5),
     (5,2,7,4),
     (2,3,5,8))

def eachPossibleMove(m):
  for y in range(1, len(m)-1):
    for x in range(1, len(m[0])-1):
      if (0 == m[y-1][x-1] == m[y-1][x] == m[y-1][x+1] ==
               m[y][x-1]   == m[y][x]   == m[y][x+1] ==
               m[y+1][x-1] == m[y+1][x] == m[y+1][x+1]):
        continue
      yield x, y

def bomb(m, (mx, my)):
  return tuple(tuple(max(0, m[y][x]-1)
      if mx-1 <= x <= mx+1 and my-1 <= y <= my+1
      else m[y][x]
      for x in range(len(m[y])))
    for y in range(len(m)))

def findFirstSolution(m, path=[]):
#  print path
#  print m
  if sum(map(sum, m)) == 0:  # empty?
    return path
  for move in eachPossibleMove(m):
    return findFirstSolution(bomb(m, move), path + [ move ])

def findShortestSolution(m):
  black = {}
  nextWhite = { m: [] }
  while nextWhite:
    white = nextWhite
    nextWhite = {}
    for position, path in white.iteritems():
      for move in eachPossibleMove(position):
        nextPosition = bomb(position, move)
        nextPath = path + [ move ]
        if sum(map(sum, nextPosition)) == 0:  # empty?
          return nextPath
        if nextPosition in black or nextPosition in white:
          continue  # ignore, found that one before
        nextWhite[nextPosition] = nextPath

def main(argv):
  if argv[1] == 'first':
    print findFirstSolution(M)
  elif argv[1] == 'shortest':
    print findShortestSolution(M)
  else:
    raise NotImplementedError(argv[1])

if __name__ == '__main__':
  import sys
  sys.exit(main(sys.argv))

你可以把这个问题表示成整数规划问题。(这只是解决这个问题的一种可能的方法)

有分:

a b c d
e f g h
i j k l
m n o p

我们可以写出16个方程其中以点f为例

f <= ai + bi + ci + ei + fi + gi + ii + ji + ki   

最小化所有索引的总和和整数解。

解当然是这些指标的和。

这可以通过将所有xi设置为边界0来进一步简化，因此在本例中最终得到4+1方程。

问题是没有解决这类问题的简单算法。我不是这方面的专家，但解决这个问题作为线性规划是NP困难。

到目前为止，一些答案给出了指数时间，一些涉及动态规划。我怀疑这些是否有必要。

我的解是O(mnS)其中m和n是板子的维度，S是所有整数的和。这个想法相当野蛮:找到每次可以杀死最多的位置，并在0处终止。

对于给定的棋盘，它给出28步棋，并且在每次落子后打印出棋盘。

完整的，不言自明的代码:

import java.util.Arrays;

public class BombMinDrops {

    private static final int[][] BOARD = {{2,3,4,7,1}, {1,5,2,6,2}, {4,3,4,2,1}, {2,1,2,4,1}, {3,1,3,4,1}, {2,1,4,3,2}, {6,9,1,6,4}};
    private static final int ROWS = BOARD.length;
    private static final int COLS = BOARD[0].length;
    private static int remaining = 0;
    private static int dropCount = 0;
    static {
        for (int i = 0; i < ROWS; i++) {
            for (int j = 0; j < COLS; j++) {
                remaining = remaining + BOARD[i][j];
            }
        }
    }

    private static class Point {
        int x, y;
        int kills;

        Point(int x, int y, int kills) {
            this.x = x;
            this.y = y;
            this.kills = kills;
        }

        @Override
        public String toString() {
            return dropCount + "th drop at [" + x + ", " + y + "] , killed " + kills;
        }
    }

    private static int countPossibleKills(int x, int y) {
        int count = 0;
        for (int row = x - 1; row <= x + 1; row++) {
            for (int col = y - 1; col <= y + 1; col++) {
                try {
                    if (BOARD[row][col] > 0) count++;
                } catch (ArrayIndexOutOfBoundsException ex) {/*ignore*/}
            }
        }

        return count;
    }

    private static void drop(Point here) {
        for (int row = here.x - 1; row <= here.x + 1; row++) {
            for (int col = here.y - 1; col <= here.y + 1; col++) {
                try {
                    if (BOARD[row][col] > 0) BOARD[row][col]--;
                } catch (ArrayIndexOutOfBoundsException ex) {/*ignore*/}
            }
        }

        dropCount++;
        remaining = remaining - here.kills;
        print(here);
    }

    public static void solve() {
        while (remaining > 0) {
            Point dropWithMaxKills = new Point(-1, -1, -1);
            for (int i = 0; i < ROWS; i++) {
                for (int j = 0; j < COLS; j++) {
                    int possibleKills = countPossibleKills(i, j);
                    if (possibleKills > dropWithMaxKills.kills) {
                        dropWithMaxKills = new Point(i, j, possibleKills);
                    }
                }
            }

            drop(dropWithMaxKills);
        }

        System.out.println("Total dropped: " + dropCount);
    }

    private static void print(Point drop) {
        System.out.println(drop.toString());
        for (int[] row : BOARD) {
            System.out.println(Arrays.toString(row));
        }

        System.out.println();
    }

    public static void main(String[] args) {
        solve();
    }

}

这是我的解决方案。由于时间有限，我不会用代码写出来，但我相信这应该每次都能产生最优的移动数量——尽管我不确定它在寻找要轰炸的点时是否有效。

首先，正如@Luka Rahne在一条评论中所说的，你轰炸的顺序并不重要，重要的是组合。

其次，正如许多人所说的那样，从角的对角线上轰炸1是最优的，因为它接触的点比角多。

这就生成了我的算法版本的基础: 我们可以在第一个或最后一个炸掉拐角的1-off，这没有关系(理论上) 我们首先破坏这些，因为它可以让后面的决定更容易(在实践中) 我们轰炸影响最大的点，同时轰炸那些角落。

让我们将阻力点定义为棋盘上具有最多不可炸点+周围0数量最多的点

非爆炸点可以定义为在我们正在研究的黑板的当前范围内不存在的点。

我还将定义4个处理范围的边界: 上=0，左=0，下=k，右=j。 (起始值)

最后，我将最优炸弹定义为投掷在与阻力点相邻的点上的炸弹，并接触(1)最高值的阻力点和(2)可能的最大数量的点。

关于方法，很明显我们正在从外到内的工作。我们将能够同时与4架“轰炸机”一起工作。

第一个阻力点显然是我们的弯道。“边界外”的点是不可轰炸的(每个角落的范围外都有5个点)。所以我们先在对角线上炸一个角。

算法:

找到4个最佳炸弹点。 如果一个炸弹点正在轰炸一个接触2个边界(即一个角)的阻力点，则一直轰炸到该点为0。否则，逐个轰炸，直到其中一个触及最佳轰炸点的阻力点为0。 对于每个边界: 如果(sum(bound)==0)前进界

重复以上步骤，直到上=下，左=右

稍后我将尝试编写实际代码

这可以用深度为O(3^(n))的树来求解。其中n是所有平方和。

首先考虑用O(9^n)树来解决问题是很简单的，只需考虑所有可能的爆炸位置。有关示例，请参阅Alfe的实现。

接下来我们意识到，我们可以从下往上轰炸，仍然得到一个最小的轰炸模式。

Start from the bottom left corner. Bomb it to oblivion with the only plays that make sense (up and to the right). Move one square to the right. While the target has a value greater than zero, consider each of the 2 plays that make sense (straight up or up and to the right), reduce the value of the target by one, and make a new branch for each possibility. Move another to the right. While the target has a value greater than zero, consider each of the 3 plays that make sense (up left, up, and up right), reduce the value of the target by one, and make a new branch for each possibility. Repeat steps 5 and 6 until the row is eliminated. Move up a row and repeat steps 1 to 7 until the puzzle is solved.

这个算法是正确的，因为

有必要在某一时刻完成每一行。 完成一行总是需要一个游戏，一个在上面，一个在下面，或者在这一行内。 选择在未清除的最低行之上的玩法总是比选择在该行之上或该行之下的玩法更好。

在实践中，这个算法通常会比它的理论最大值做得更好，因为它会定期轰炸邻居并减少搜索的大小。如果我们假设每次轰炸都会减少4个额外目标的价值，那么我们的算法将运行在O(3^(n/4))或大约O(1.3^n)。

Because this algorithm is still exponential, it would be wise to limit the depth of the search. We might limit the number of branches allowed to some number, X, and once we are this deep we force the algorithm to choose the best path it has identified so far (the one that has the minimum total board sum in one of its terminal leaves). Then our algorithm is guaranteed to run in O(3^X) time, but it is not guaranteed to get the correct answer. However, we can always increase X and test empirically if the trade off between increased computation and better answers is worthwhile.