這題要用到的是矩陣-樹定理。ios
對於有邊權的圖,其全部生成樹邊權積的和等於基爾霍夫矩陣的任意餘子式\(M(i,i)\)的行列式。ide
其中,基爾霍夫矩陣\(C=D-A\).\(D\)爲發出邊權和矩陣,對於任意\(u \neq v\),有\(D(u,v) = 0\).\(A\)是鄰接矩陣。spa
總而言之,矩陣-樹定理能求出的是\(\sum\limits_{T} \prod\limits_{e \in T} p_e\).
回頭看這道題,這道題讓咱們求的是\(\sum\limits_{T} [\prod\limits_{e \in T} p_e \prod\limits_{e \notin T}(1-p_e)]\).
因此咱們對上式變形,化成能求的形式:string
所以咱們將每一條邊的邊權改爲\(\frac{p}{1-p}\),再構造基爾霍夫矩陣求解餘子式的行列式就好了。
可是若是\(p=1\)或\(0\)怎麼辦?it
這個我想了好一下子都沒出來,只能看題解了:由於題目說精度在\(10^{-4}\)就算對,因此對於上述狀況,咱們能夠令\(p=1-eps\)和\(eps\)就行的通了!這個着實妙。io
#include<cstdio> #include<iostream> #include<cmath> #include<algorithm> #include<cstring> using namespace std; #define In inline typedef double db; const db eps = 1e-8; const int maxn = 55; int n; db p[maxn][maxn]; In db Gauss(db f[][maxn], int n) { db ret = 1; for(int i = 1; i <= n; ++i) { int pos = i; for(int j = i + 1; j <= n; ++j) if(fabs(f[j][i]) > fabs(f[pos][i])) pos = j; if(pos != i) swap(f[pos], f[i]), ret = -ret; if(fabs(f[i][i]) < eps) return 0; for(int j = i + 1; j <= n; ++j) { db tp = f[j][i] / f[i][i]; for(int k = i; k <= n; ++k) f[j][k] -= tp * f[i][k]; } ret *= f[i][i]; } return ret; } int main() { scanf("%d", &n); for(int i = 1; i <= n; ++i) for(int j = 1; j <= n; ++j) scanf("%lf", &p[i][j]); db sum = 1; for(int i = 1; i <= n; ++i) for(int j = 1; j <= n; ++j) { if(i == j) continue; if(1 - p[i][j] < eps) p[i][j] = 1 - eps; else if(p[i][j] < eps) p[i][j] = eps; if(j > i) sum *= (1 - p[i][j]); p[i][j] = -p[i][j] / (1 - p[i][j]); } for(int i = 1; i <= n; ++i) for(int j = 1; j <= n; ++j) if(i ^ j) p[i][i] -= p[i][j]; printf("%.9lf\n", sum * Gauss(p, n - 1)); return 0; }