$\text{GYM}$ 链接:Matrix Exponentiation
给一个无向图,求包含 $k$ 条边的最小路径。
考虑 $\text{dp}$,定义 $f[i][j][k]$ 为从第 $i$ 个点出发经过 $k$ 条边到第 $j$ 个点的最短路。则会有如下的转移方程
$$f[i][j][k] = min_u\{f[i][u][k-1] + g[u][j]\}$$
但 $k$ 值很大,所以不能直接$\text{dp}$。因为取最小值的操作和乘法都满足结合律交换律,且上述形式和矩阵乘法类似,所以可以把矩阵乘法改写成上面的转移方程,再用矩阵快速幂加速。
code
code
#include<bits/stdc++.h> #define ALL(x) (x).begin(),(x).end() #define ll long long #define db double #define ull unsigned long long #define pii_ pair<int,int> #define mp_ make_pair #define pb push_back #define fi first #define se second #define rep(i,a,b) for(int i=(a);i<=(b);i++) #define per(i,a,b) for(int i=(a);i>=(b);i--) #define show1(a) cout<<#a<<" = "<<a<<endl #define show2(a,b) cout<<#a<<" = "<<a<<"; "<<#b<<" = "<<b<<endl using namespace std; const ll INF = 1LL<<60; const int inf = 1<<30; const int maxn = 2e5+5; inline void fastio() {ios::sync_with_stdio(false);cin.tie(0);cout.tie(0);} int n = 100; struct Matrix { ll mat[100][100]; //Matrix() {memset(mat,0,sizeof(mat));} /*Matrix operator * (Matrix b) { Matrix res; rep(i,0,n-1) rep(j,0,n-1) rep(k,0,n-1) res.mat[i][j] = (res.mat[i][j] + mat[i][k]*b.mat[k][j]%M)%M; return res; }*/ Matrix() {rep(i,0,n-1) rep(j,0,n-1) mat[i][j] = INF;} Matrix operator * (Matrix b) { Matrix res; rep(k,0,n-1){ rep(i,0,n-1) rep(j,0,n-1) { res.mat[i][j] = min(res.mat[i][j],mat[i][k]+b.mat[k][j]); } } return res; } Matrix operator ^ (ll b) { Matrix res,A=*this; rep(i,0,n-1) rep(j,0,n-1) res.mat[i][j] = (i!=j)*INF; while(b){ if(b&1) res = res*A; A = A*A; b>>=1; }return res; } }; int main() { fastio(); int m,k; cin>>n>>m>>k; Matrix A; while(m--){ int u,v,w; cin>>u>>v>>w; u--,v--; A.mat[u][v] = w; } A = A^k; ll ans = INF; rep(i,0,n-1) rep(j,0,n-1) ans = min(ans,A.mat[i][j]); if(ans>1e18) cout<<"IMPOSSIBLE"<<endl;else cout<<ans<<endl; return 0; }
求序列
$$a_i = c_1\cdot a_{i-1} + c_2\cdot a_{i-2} + \cdots + c_n\cdot a_{i-n} + p + i\cdot q + i^2 \cdot r$$
的第 $k$ 项,也就是归纳一般的线性递推式的矩阵构造方法。
code
code
/* #pragma GCC optimize(2) #pragma GCC optimize(3,"Ofast","inline") */ #include<bits/stdc++.h> #define ALL(x) (x).begin(),(x).end() #define ll long long #define db double #define ull unsigned long long #define pii_ pair<int,int> #define mp_ make_pair #define pb push_back #define fi first #define se second #define rep(i,a,b) for(int i=(a);i<=(b);i++) #define per(i,a,b) for(int i=(a);i>=(b);i--) #define show1(a) cout<<#a<<" = "<<a<<endl #define show2(a,b) cout<<#a<<" = "<<a<<"; "<<#b<<" = "<<b<<endl using namespace std; const ll INF = 1LL<<60; const int inf = 1<<30; const int maxn = 2e5+5; const int M = 1e9+7; inline void fastio() {ios::sync_with_stdio(false);cin.tie(0);cout.tie(0);} int n; ll k; struct Matrix { ll mat[15][15]; Matrix() {memset(mat,0,sizeof(mat));} Matrix operator * (const Matrix other) const { Matrix product; rep(i,1,n+3) rep(j,1,n+3) rep(k,1,n+3) product.mat[i][j] = (product.mat[i][j] + mat[i][k]*other.mat[k][j])%M; return product; } Matrix operator ^ (ll b) const { Matrix res,A=*this; rep(i,1,n+3) res.mat[i][i] = 1; while(b){ if(b&1) res = res*A; A = A*A; b>>=1; }return res; } }; ll a[15]; int main() { fastio(); cin>>n>>k; k -= n-1; rep(i,1,n) cin>>a[n-i+1]; a[n+1] = 1,a[n+2] = n,a[n+3] = n*n; Matrix single; rep(i,1,n+3) cin>>single.mat[1][i]; rep(i,2,n) single.mat[i][i-1] = 1; single.mat[n+1][n+1] = 1; single.mat[n+2][n+1] = single.mat[n+2][n+2] = 1; single.mat[n+3][n+1] = 1,single.mat[n+3][n+2] = 2,single.mat[n+3][n+3] = 1; Matrix total = single^k; ll ans = 0; rep(i,1,n+3) ans = (ans + total.mat[1][i]*a[i])%M; cout<<ans<<endl; return 0; }
给一个只包含 $26$ 个大写字母或者 $?$ 的字符串。一开始 $Limak$ 的心情是好的,接着从左往右遍历,如果遇到 $AEIOU$ 中的字母则心情翻转,如果遇到 $H$ 则心情变好,如果遇到 $S$ 和 $D$ 则心情变差。字符 $?$ 可以是任何一种字母。
问遍历了字符串后,$Limak$ 的心情仍然是好的的情况有多少种。并且会给出 $q$ 次修改,每次修改某个位置的字符,并询问最后的结果。
如果没有修改则可以线性的 $dp$ 求解。在有修改的情况下,考虑 $dp$ 的过程是一个矩阵的乘法,即每个字母对应了一种矩阵,那修改就可以用线段树来维护了。
code
code
/* #pragma GCC optimize(2) #pragma GCC optimize(3,"Ofast","inline") */ #include<bits/stdc++.h> #define ALL(x) (x).begin(),(x).end() #define ll long long #define db double #define ull unsigned long long #define pii_ pair<int,int> #define mp_ make_pair #define pb push_back #define fi first #define se second #define rep(i,a,b) for(int i=(a);i<=(b);i++) #define per(i,a,b) for(int i=(a);i>=(b);i--) #define show1(a) cout<<#a<<" = "<<a<<endl #define show2(a,b) cout<<#a<<" = "<<a<<"; "<<#b<<" = "<<b<<endl using namespace std; const ll INF = 1LL<<60; const int inf = 1<<30; const int maxn = 2e5+5; const int M = 1e9+7; inline void fastio() {ios::sync_with_stdio(false);cin.tie(0);cout.tie(0);} char s[maxn]; int n,q; struct Matrix { ll mat[2][2]; Matrix() {memset(mat,0,sizeof(mat));} Matrix operator * (const Matrix other) const { Matrix product; rep(i,0,1)rep(j,0,1)rep(k,0,1) product.mat[i][j] = (product.mat[i][j] + mat[i][k] * other.mat[k][j])%M; return product; } }tr[maxn<<2]; void push_up(int id) { tr[id] = tr[id<<1] * tr[id<<1|1]; } void build(int id,int l,int r) { if(l==r){ if(s[n-l+1]=='A' || s[n-l+1]=='E' || s[n-l+1]=='I' || s[n-l+1]=='O' || s[n-l+1]=='U'){ tr[id].mat[0][1] = tr[id].mat[1][0] = 1; }else if(s[n-l+1]=='H'){ tr[id].mat[0][0] = tr[id].mat[0][1] = 1; }else if(s[n-l+1]=='S' || s[n-l+1]=='D'){ tr[id].mat[1][0] = tr[id].mat[1][1] = 1; }else if(s[n-l+1]=='?'){ tr[id].mat[0][0] = 19,tr[id].mat[0][1] = 6; tr[id].mat[1][0] = 7,tr[id].mat[1][1] = 20; }else { tr[id].mat[0][0] = tr[id].mat[1][1] = 1; } return ; } int mid = (l+r)>>1; build(id<<1,l,mid);build(id<<1|1,mid+1,r); push_up(id); } void update(int id,int stl,int str,int pos,char o) { if(stl==str){ memset(tr[id].mat,0,sizeof(tr[id].mat)); if(o=='A' || o=='E' || o=='I' || o=='O' || o=='U'){ tr[id].mat[0][1] = tr[id].mat[1][0] = 1; }else if(o=='H'){ tr[id].mat[0][0] = tr[id].mat[0][1] = 1; }else if(o=='S' || o=='D'){ tr[id].mat[1][0] = tr[id].mat[1][1] = 1; }else if(o=='?'){ tr[id].mat[0][0] = 19,tr[id].mat[0][1] = 6; tr[id].mat[1][0] = 7,tr[id].mat[1][1] = 20; }else { tr[id].mat[0][0] = tr[id].mat[1][1] = 1; } return ; } int mid = (stl+str)>>1; if(pos<=mid) update(id<<1,stl,mid,pos,o); else update(id<<1|1,mid+1,str,pos,o); push_up(id); } int main() { fastio(); cin>>n>>q>>s+1; build(1,1,n); cout<<tr[1].mat[0][0]<<endl; while(q--){ int pos;char o; cin>>pos>>o; update(1,1,n,n-pos+1,o); cout<<tr[1].mat[0][0]<<endl; } return 0; }
给一个有向图和 $q$ 次询问,每次询问是求 $s$ 到 $t$ 并经过 $k$ 条边的路径数。
因为 $n,q \le 200$,所以每次都用快速幂求一次是不行的。因为每次询问只要求特定两点之间的路径,所以很多乘法是没有意义的,考虑如何只维护矩阵的第 $s$ 行向量:
可以先倍增求出邻接矩阵的 $2$ 的幂次的幂次。然后每次询问时只需要维护第 $s$ 行的向量即可。
code
code
/* #pragma GCC optimize(2) #pragma GCC optimize(3,"Ofast","inline") */ #include<bits/stdc++.h> #define ALL(x) (x).begin(),(x).end() #define ll long long #define db double #define ull unsigned long long #define pii_ pair<int,int> #define mp_ make_pair #define pb push_back #define fi first #define se second #define rep(i,a,b) for(int i=(a);i<=(b);i++) #define per(i,a,b) for(int i=(a);i>=(b);i--) #define show1(a) cout<<#a<<" = "<<a<<endl #define show2(a,b) cout<<#a<<" = "<<a<<"; "<<#b<<" = "<<b<<endl using namespace std; const ll INF = 1LL<<60; const int inf = 1<<30; const int maxn = 2e5+5; const int M = 1e9+7; inline void fastio() {ios::sync_with_stdio(false);cin.tie(0);cout.tie(0);} int n,m,q; struct Matrix { ll mat[201][201]; Matrix() {memset(mat,0,sizeof(mat));} Matrix operator * (const Matrix other) const { Matrix product; rep(i,1,n) rep(j,1,n) rep(k,1,n) product.mat[i][j] = (product.mat[i][j] + mat[i][k]*other.mat[k][j])%M; return product; } }A[35]; int ans[205],tmp[205]; int main() { fastio(); cin>>n>>m>>q; while(m--){ int u,v; cin>>u>>v; A[0].mat[u][v] = 1; } rep(i,1,30) A[i] = A[i-1]*A[i-1]; while(q--){ int s,t,k; cin>>s>>t>>k; rep(i,1,n) ans[i] = (i==s); rep(b,0,30){ if((k>>b)&1){ rep(i,1,n) tmp[i] = 0; rep(i,1,n) rep(j,1,n) tmp[i] = (tmp[i] + ans[j] * A[b].mat[j][i])%M; rep(i,1,n) ans[i] = tmp[i]; } } cout<<ans[t]<<endl; } return 0; }