2019牛客多校第九场
A. The power of Fibonacci
upsolved
求斐波那契数列的\(m\)次方的前\(n\)项的和模1e9的值
\((1<=n<=1e9, 1<=m<=1000)\)
首先要知道斐波那契数列模意义下是有循环节的,它的循环节也必定是它的\(m\)次方的循环节
然后把\(1e9\)拆成\(2^9*5^9\), 对\(2^9\)和\(5^9\)这两个模数分别找循环节,\(2^9\)的循环节是768, \(5^9\)的循环节是7812500
这两个分别求出了然后拼起来就可以得到答案了
ps:为什么快速幂的时候每次把模数(\(2^9\)或\(5^9\))传进去的时间是直接把模数钉死为1e9的三倍啊...
#include <bits/stdc++.h>
using namespace std;
const int N = 8e6 + 10;
const int mod[2] = {512, 125 * 125 * 125};
int f[N];
int ans[2];
int n, m;
int qp(int a, int n, int mod = 1000000000) {
int res = 1;
while(n) {
if(n & 1)
res = 1LL * res * a % mod;
a = 1LL * a * a % mod;
n >>= 1;
}
return res;
}
int main() {
scanf("%d%d", &n, &m);
for(int k = 0; k < 2; ++k) {
f[0] = 0; f[1] = 1;
int j = 2;
for(; ; ++j) {
f[j] = f[j - 1] + f[j - 2];
if(f[j] >= mod[k]) f[j] -= mod[k];
if(f[j] == 0 && f[j - 1] == 1) break;
}
for(int i = 0; i < j; ++i) {
int cnt = n / j + (n % j >= i);
ans[k] = (ans[k] + 1LL * cnt * qp(f[i], m)) % mod[k];
}
}
while(ans[1] % mod[0] != ans[0]) ans[1] += mod[1];
printf("%d\n", ans[1]);
return 0;
}
B. Quadratic equation
solved at 00:53
队友做的,二次剩余,不会
D. Knapsack Cryptosystem
solved at 00:20
\(n\)个物品,每个物品有重量,你要找到一个子集使得物品的重量和恰好为\(s\)\((1<=n<=36)\)输入保证有唯一解
折半搜索,先把考虑前一半,二进制状压把它们能组成的所有情况全部丢到\(map\)里去,然后考虑后一半,每次二进制状压然后去\(map\)里寻找解
#include <bits/stdc++.h>
using namespace std;
int n, p, q;
long long s, a[40];
int to[(1 << 18) + 10];
map<long long, int> pp;
void init() {
int msk = 2;
to[1] = 0;
for(int i = 1; i <= 17; ++i) {
to[msk] = i;
msk *= 2;
}
}
void print(int maskp, int maskq) {
for(int i = 0; i < p; ++i)
printf("%d", 1 & (maskp >> i));
for(int i = 0; i < q; ++i)
printf("%d", 1 & (maskq >> i));
puts("");
}
int main() {
init();
cin >> n >> s;
for(int i = 0; i < n; ++i)
cin >> a[i];
p = n / 2;
q = n - p;
for(int i = 0; i < (1 << p); ++i) {
long long tmp = 0;
for(int j = i; j; j -= j & -j)
tmp += a[to[j & -j]];
pp[tmp] = i;
}
for(int i = 0; i < (1 << q); ++i) {
long long tmp = 0;
for(int j = i; j; j -= j & -j)
tmp += a[to[j & -j] + p];
auto it = pp.lower_bound(s - tmp);
if(it->first + tmp == s) {
print(it->second, i);
return 0;
}
}
puts("-1");
return 0;
}
E. All men are brothers
solved at 00:49
\(n\)个人,一开始互相都不认识,\(m\)次操作,每次会让两个人互相认识,这种认识关系具有传递性和对称性,每次操作后要求输出选择\(4\)个人互相都不认识的方案数
并查集维护连通块大小即可
考虑合并两个连通块,那么答案会比上一次的答案减少一些,减少的是这两个连通块各选一个,再从外面选不在同一个连通块里的两个的方案数,具体就不写了,看代码(sz指所有连通块大小的平方和)
#include <bits/stdc++.h>
using namespace std;
const int N = 1e5 + 10;
int n, m, x, y;
int fa[N], size[N];
long long sz;
unsigned long long ans;
int find(int x) {
return fa[x] == x ? x : find(fa[x]);
}
void unite(int x, int y) {
int fx = find(x), fy = find(y);
if(fx != fy) {
if(size[fx] < size[fy])
swap(fx, fy);
fa[fy] = fx;
size[fx] += size[fy];
}
}
int main() {
scanf("%d%d", &n, &m);
if(n <= 3) {
for(int i = 0; i <= m; ++i)
puts("0");
return 0;
}
for(int i = 1; i <= n; ++i) {
fa[i] = i;
size[i] = 1;
}
ans = 1ULL * n * (n - 1) / 2 * (n - 2) / 3 * (n - 3) / 4;
printf("%llu\n", ans);
sz = n;
for(int i = 1; i <= m; ++i) {
scanf("%d%d", &x, &y);
if(find(x) == find(y)) {
printf("%llu\n", ans);
continue;
}
int sx = size[find(x)], sy = size[find(y)];
ans -= 1ULL * sx * sy * (1ULL * (n - sx - sy) * (n - sx - sy) - (sz - 1LL * sx * sx - 1LL * sy * sy)) / 2;
sz -= 1LL * sx * sx;
sz -= 1LL * sy * sy;
sz += 1LL * (sx + sy) * (sx + sy);
printf("%llu\n", ans);
unite(find(x), find(y));
}
return 0;
}
H. Cutting Bamboos
solved at 03:11(+2)
有\(n\)株竹子,每株有高度,\(q\)次询问,每次询问一个区间\(l, r\),给定\(x, y\)
,指的是水平方向切了\(y\)刀并且每一刀切下来的竹子总量相同且最后一刀高度为\(0\), 求第\(x\)刀的高度
首先每一刀切下来的总量相同,那么可以求出第\(x\)刀到第\(y\)到切下的总量,然后二分答案,查询当前高度下的竹子总量(小于等于刀的高度的就是竹子本身高度,大于的是刀的高度),那么相当于求区间小于\(k\)的数的个数以及它们的和,可以用主席树实现,总复杂度\(O(q\log(n)(\log(h) + 50))\)(50是浮点数二分次数,可能不需要这么多)
#include <bits/stdc++.h>
using namespace std;
const int N = 2e5 + 10;
struct node {
int l, r, cnt;
long long sum;
}Tree[N * 20];
int root[N];
int cnt;
void init() {
cnt = 0;
root[0] = 0;
Tree[0].l = Tree[0].r = Tree[0].cnt = Tree[0].sum = 0;
}
void update(int &rt, int l, int r, int num) {
Tree[++cnt] = Tree[rt];
rt = cnt;
Tree[rt].cnt++;
Tree[rt].sum += num;
if(l == r)
return;
int mid = l + r >> 1;
if(num <= mid)
update(Tree[rt].l, l, mid, num);
else
update(Tree[rt].r, mid + 1, r, num);
}
pair<int, long long> query(int i, int j, int num, int l, int r) {
if(l == r)
return make_pair(Tree[j].cnt - Tree[i].cnt, Tree[j].sum - Tree[i].sum);
int mid = l + r >> 1;
if(num <= mid)
return query(Tree[i].l, Tree[j].l, num, l, mid);
pair<int, long long> tmp = query(Tree[i].r, Tree[j].r, num, mid + 1, r);
return make_pair(Tree[Tree[j].l].cnt - Tree[Tree[i].l].cnt + tmp.first, Tree[Tree[j].l].sum - Tree[Tree[i].l].sum + tmp.second);
}
int n, q, l, r, x, y;
int h[N];
long long sumh[N];
double check(double height) {
if(height < 1)
return (r - l + 1) * height;
int c; long long s;
tie(c, s) = query(root[l - 1], root[r], (int)height, 1, 100000);
return s + (r - l + 1 - c) * height;
}
int main() {
init();
scanf("%d%d", &n, &q);
for(int i = 1; i <= n; ++i) {
scanf("%d", &h[i]);
root[i] = root[i - 1];
update(root[i], 1, 100000, h[i]);
sumh[i] = sumh[i - 1] + h[i];
}
while(q--) {
scanf("%d%d%d%d", &l, &r, &x, &y);
int ll = 0, rr = 1e5, ans;
double tar = (1.0 * y - x) / y * (sumh[r] - sumh[l - 1]);
while(ll <= rr) {
int mid = ll + rr >> 1;
if(check(mid) <= tar) {
ans = mid;
ll = mid + 1;
}
else {
rr = mid - 1;
}
}
double o, lll = ans, rrr = ans + 1;
for(int _ = 0; _ <= 50; ++_) {
o = (lll + rrr) / 2;
if(check(o) <= tar)
lll = o;
else
rrr = o;
}
printf("%.15f\n", o);
}
return 0;
}
I. KM and M
upsolved
给你\(N, M(1<=N<=1e18, 1<=M<=1e11)\),求\[\sum\limits_{k=1}^N((kM)\&M) mod (1e9+7)\]
按二进制位每一位考虑,考虑第\(i\)位,那么一个数\(x\)的贡献就是\((2^i)\&x\), 考虑\(x\)的第\(i\)位是否为\(1\),显然可以用\(\frac x {2^i} - 2 * \frac x {2^{i+1}}\)表示
那么对于\(N\)个数,第\(i\)位为\(1\)的就是\[\sum \limits_{i=1}^{N} \frac {M * i} {2^i} - 2*\sum \limits_{i=1}^{N} \frac {M * i} {2^{i+1}}\]
这两个求和都是求等差数列除以一个数的和,可以用类欧几里得算法求解
#include <bits/stdc++.h>
using namespace std;
using LL = long long;
const int mod = 1e9 + 7, inv2 = (mod + 1) / 2;
LL n, m, ans;
LL calc(LL a, LL b, LL c, LL n) {
if(!a) return 0;
if(a >= c || b >= c) {
LL tmp = calc(a % c, b % c, c, n);
return (tmp + ((n % mod) * ((n + 1) % mod) % mod * inv2 % mod) * ((a / c) % mod) % mod + ((n + 1) % mod) * ((b / c) % mod) % mod) % mod;
}
LL f = ((__int128)a * n + b) / c;
return (((n % mod) * (f % mod) % mod - calc(c, c - b - 1, a, f - 1)) % mod + mod) % mod;
}
int main() {
cin >> n >> m;
for(int i = 0; i < 40; ++i) if(m >> i & 1) {
LL res = ((calc(m, 0, 1LL << i, n) - 2 * calc(m, 0, 2LL << i, n)) % mod + mod) % mod;
ans = (ans + (1LL << i) % mod * res % mod) % mod;
}
printf("%lld\n", ans);
return 0;
}
J. Symmetrical Painting
solved at 01:38(+4)
为什么map一直MLE啊.....
二维平面上有n个矩形,每个矩形左下角是\((i - 1, L_i)\), 右上角是\((i, R_i)\),矩形一开始全是黑色,平面不被矩形覆盖的地方是白色,你要把某些黑色区域涂白(一个矩形可以内部颜色不一样),使得黑色区域是一个轴对称图形并且对称轴平行于x轴,求最大黑色区域面积
首先取得答案时的对称轴纵坐标要么是整数要么是\(x.5\),枚举对称轴,维护两个集合,一个代表当前枚举的对称轴在矩形下半部分,一个表示上半部分,每次对称轴往上移动0.5,第一个集合里每一个元素都会给答案贡献1, 第二个集合贡献-1, 然后可能会有第一个集合的矩形跳到第二个集合(到达中线),也可能有矩形进入第一个集合(到达下底),也可能有矩形从第二个集合出去(到达上底)
实际上你并不需要真的维护集合,只需要知道集合的大小就可以了
#include <bits/stdc++.h>
using namespace std;
const int N = 3e5 + 10;
int n, L[N], R[N], md[N], m;
long long tmp, ans;
int b[N * 3];
int in[N * 3], change[N * 3], out[N * 3];
int main() {
scanf("%d", &n);
for(int i = 1; i <= n; ++i) {
scanf("%d%d", &L[i], &R[i]);
L[i] = 2 * L[i] - 1;
R[i] = 2 * R[i] - 1;
md[i] = (0LL + L[i] + R[i]) / 2;
b[3 * i - 2] = L[i];
b[3 * i - 1] = R[i];
b[3 * i] = md[i];
}
tmp = ans = 0;
sort(b + 1, b + 3 * n + 1);
m = unique(b + 1, b + 3 * n + 1) - b - 1;
for(int i = 1; i <= n; ++i) {
in[lower_bound(b + 1, b + m + 1, L[i]) - b]++;
change[lower_bound(b + 1, b + m + 1, md[i]) - b]++;
out[lower_bound(b + 1, b + m + 1, R[i]) - b]++;
}
int down = 0, up = 0;
down += in[1];
for(int i = 2; i <= m; ++i) {
tmp += 1LL * (down - up) * (b[i] - b[i - 1]);
ans = max(ans, tmp);
down += in[i];
down -= change[i];
up += change[i];
up -= out[i];
}
printf("%lld\n", ans);
return 0;
}