CodeChef-LECOINS Little Elephant and Colored Coins

Little Elephant and Colored Coins

The Little Elephant from the Zoo of Lviv very likes coins. But most of all he likes colored coins.

He has N types of coins, numbered from 1 to N, inclusive. The coin of the i-th type has the value Vi dollars and the color Ci. Note that he has infinite supply of each type of coins.

The Little Elephant wants to make exactly S dollars using the coins. What is the maximal number of different colors he can use to make exactly S dollars using some of the coins he has? If it's impossible, output -1. Also note that the Little Elephant wants to know this for many values of S.

Input

The first line of the input contains a single integer N, denoting the number of types of coins. Each of the following N lines contains two space-separated integers Vi and Ci, denoting the value and the color of the coin of the i-th type. The next line contains a single integer Q, denoting the number of values of S to process. Each of the following Q lines contains a single integer S, denoting the coinage you should represent via given coins using maximum number of colors.

Output

For each value of S in the input, output the maximum number of different colors in the representation of S or -1 if it is impossible to represent S via given coins.

Constraints

  • 1N30
  • 1Vi200000 (2 * 10^5)
  • 1Ci10^9
  • 1Q200000 (2 * 10^5)
  • 1S10^18

Example

Input:
3
2 1
3 4
4 4
4
1
3
5
7
Output:
-1
1
2
2

Explanation

  • It is not possible to represent S = 1 since every coin has value more than 1.
  • S = 3 can only be represented using one coin of the second type, hence only one color is used in the representation.
  • S = 5 can only be represented as 2 + 3, which leads to two colors used.
  • For S = 7 we have two representations as 2 + 2 + 3 (with two colors used) and 3 + 4 (with one color used). Hence, the answer is 2.

妙题啊。

可达性

如果有个 \(C_i \le 1\) 的SubTask的话会更容易想到正解。

我们先不考虑颜色,只考虑是否能找到一种方案把硬币总面值凑成 \(S\)

有一种非常朴素的想法是从 \(0\)\(S\) 做可达性dp: \(f_{x+V_i} = f_{x+V_i} | f_x\) ,不过因为 \(S \le 10^{18}\) ,显然这是天方夜谭。

可达性dp只有 \(0/1\) 两种状态,非常浪费;发现 \(V_i\) 的值域很小,这启示我们将 \(V_i\) 作为状态减少有效状态数。但究竟如何设置呢?

这里是最巧妙的地方。不失一般性,设 \(V_1 = \min V_i\) 。我们尝试把 \(V_1\) 孤立出来考虑。 假设保证过程中最终选择的总面额永远不会比 \(S\) 大(换句话说,允许选择负数个硬币),那么只需知道硬币集合 \(V - \{ V_1 \}\) 中是否存在一种方案其总面值模 \(V_i\)\(S\) 同余即可。这样就只需对去掉 \(V_1\) 的硬币集合做 \(1\)\(V_1- 1\) 的可达性dp就好了。

考虑把前面那个假设干掉。之前提到 \(f\) 只有 \(0/1\) 两种状态,明显可以再塞点东西进去。于是令 \(f_x\) 为硬币集合 \(V - \{ V_1 \}\) 中所有总面值模 \(V_1\) 等于 \(x\) 的选择方案中最小的总面值。容易写出状态转移

\[ \mathrm{relax} \ f_{(x+V_i)\% V_1} \ \mathrm{by} \ f_x + V_i \]

\(\%\) 代表取模, \(\mathrm{relax} \ A \ \mathrm{by} \ B\)\(A = \min(A,B)\)

这样一来,若 \(f_{S \% V_1} \le S\) ,那么 \(S\) 就是可以被组成的。

发现这个dp就是个最短路,形成了一个 \(|V| = V_1, |E| = (n-1) V_1\) 的图。于是 Dijkstra 一下就可以了。不过这个图还有更好的性质。

首先,路径中边的顺序可以任意调换而不影响可达性和最短路,所以我们可以分开考虑每一种边的松弛。而一分开看就非常明朗了,根据数论常识,所有 \(x\)\((x + V_i) \% m\) 的边会在图上形成 \(\gcd(V_i, m)\) 个大小为 \(\frac{V_1}{\gcd({V_i, m})}\) 的环,于是现在我们又可以分别考虑每一个环。只需从环上当前距离值最小的点绕环一圈就可以做到松弛了。于是现在预处理的时间复杂度是 \(O(|E|) = O(nV_1)\) ,比直接 Dijkstra 少了个 \(\log\) 。询问当然是 \(O(1)\) 的。

考虑颜色

把颜色放到dp状态里面去就好了。设 \(f_{x, c}\)\(c\) 记录了选取的颜色种类数,其余与前述相同。

比较懒,不写代码了。实在需要可以看官方题解。

参考

官方题解

这篇题解基本上是官方题解的简化翻译版本