题面描述
思路
状态转移方程:( s s s为前缀和)
F i = min ( F j + ( s i − s j ) 2 + m ) F_i=\min(F_j+(s_i-s_j)^2+m) Fi=min(Fj+(si−sj)2+m)
决策单调性
F j + ( s i − s j ) 2 + m ≥ F k + ( s i − s j ) 2 + m F_j+(s_i-s_j)^2+m\ge F_k+(s_i-s_j)^2+m Fj+(si−sj)2+m≥Fk+(si−sj)2+m
化简
F j − 2 ∗ s i ∗ s j + s j 2 ≥ F k − 2 ∗ s i ∗ s k + s k 2 F_j-2*s_i*s_j+{s_j}^2\ge F_k-2*s_i*s_k+{s_k}^2 Fj−2∗si∗sj+sj2≥Fk−2∗si∗sk+sk2
对于未来状态 t t t,证明:
F j + ( s t − s j ) 2 + m ≥ F k + ( s t − s j ) 2 + m F_j+(s_t-s_j)^2+m\ge F_k+(s_t-s_j)^2+m Fj+(st−sj)2+m≥Fk+(st−sj)2+m
F j − 2 ∗ s t ∗ s j + s j 2 ≥ F k − 2 ∗ s t ∗ s k + s k 2 F_j-2*s_t*s_j+{s_j}^2\ge F_k-2*s_t*s_k+{s_k}^2 Fj−2∗st∗sj+sj2≥Fk−2∗st∗sk+sk2
由 s t = s i + v a l , s_t=s_i+val, st=si+val,
F j − 2 ∗ ( s i + v a l ) ∗ s j + s j 2 ≥ F k − 2 ∗ ( s i + v a l ) ∗ s k + s k 2 F_j-2*(s_i+val)*s_j+{s_j}^2\ge F_k-2*(s_i+val)*s_k+{s_k}^2 Fj−2∗(si+val)∗sj+sj2≥Fk−2∗(si+val)∗sk+sk2
由
F j − 2 ∗ s i ∗ s j + s j 2 ≥ F k − 2 ∗ s i ∗ s k + s k 2 F_j-2*s_i*s_j+{s_j}^2\ge F_k-2*s_i*s_k+{s_k}^2 Fj−2∗si∗sj+sj2≥Fk−2∗si∗sk+sk2
可知仅需证明:
− 2 ∗ v a l ∗ s j ≥ − 2 ∗ v a l ∗ s k -2*val*{s_j}\ge-2*val*s_k −2∗val∗sj≥−2∗val∗sk
s j ≤ s k s_j\le s_k sj≤sk
得证。
踢队头
F j − 2 ∗ s i ∗ s j + s j 2 ≥ F k − 2 ∗ s i ∗ s k + s k 2 F_j-2*s_i*s_j+{s_j}^2\ge F_k-2*s_i*s_k+{s_k}^2 Fj−2∗si∗sj+sj2≥Fk−2∗si∗sk+sk2
F j − F k + s j 2 − s k 2 ≥ 2 ∗ s i ∗ ( s j − s k ) F_j-F_k+{s_j}^2-{s_k}^2\ge2*s_i*(s_j-s_k) Fj−Fk+sj2−sk2≥2∗si∗(sj−sk)
由 s j − s k < 0 s_j-s_k<0 sj−sk<0
c a l c ( j , k ) = F j − F k + s j 2 − s k 2 s j − s k ≤ 2 ∗ s i calc(j,k)=\frac{F_j-F_k+{s_j}^2-{s_k}^2}{s_j-s_k}\le2*s_i calc(j,k)=sj−skFj−Fk+sj2−sk2≤2∗si
当 c a l c ( j , k ) ≤ 2 ∗ s i calc(j,k)\le 2*s_i calc(j,k)≤2∗si时, k k k优于 j j j.
因此当 c a l c ( q h e a d , q h e a d + 1 ) ≤ 2 ∗ s i calc(q_{head},q_{head+1})\le 2*s_i calc(qhead,qhead+1)≤2∗si
h e a d + 1 head+1 head+1优于 h e a d head head。
由于 s i s_i si随 i i i增大而增大,那么 c a l c ( q h e a d , q h e a d + 1 ) calc(q_{head},q_{head+1}) calc(qhead,qhead+1)随 h e a d head head增大而增大,才符合 h e a d head head为最优解,因此斜率是不断递增的。
踢队尾
根据斜率是不断递增的,仅当
c a l c ( q t a i l , i ) ≥ c a l c ( q t a i l − 1 , q t a i l ) calc(q_{tail},i)\ge calc(q_{tail-1},q_{tail}) calc(qtail,i)≥calc(qtail−1,qtail)
斜率才满足不断递增。
故当
c a l c ( q t a i l , i ) ≤ c a l c ( q t a i l − 1 , q t a i l ) calc(q_{tail},i)\le calc(q_{tail-1},q_{tail}) calc(qtail,i)≤calc(qtail−1,qtail)
删去队尾。
特别提醒
当 c i = 0 c_i=0 ci=0时, s i − 1 = s i s_{i-1}=s_i si−1=si;
不具有单调性,所以要进行特殊操作。
AC code
#include<cstdio>
#include<cstring>
#include<algorithm>
#include<cmath>
#include<cstdlib>
#define gc getchar()
#define ll long long
using namespace std;
const int N=5e5+10;
inline void qr(ll &x)
{
x=0;int f=1;char c=gc;
while(c<'0'||c>'9'){if(c=='-')f=-1;c=gc;}
while(c>='0'&&c<='9'){x=x*10+(c^48);c=gc;}
x*=f;
}
inline void qw(ll x)
{
if(x<0)x=-x,putchar('-');
if(x/10)qw(x/10);
putchar(x%10+48);
}
ll f[N],s[N];int q[N],l,r;
inline double calc(int j,int k)
{
if(s[j]-s[k]==0)return -1e10;
return (f[j]-f[k]+s[j]*s[j]-s[k]*s[k])/(double)(s[j]-s[k]);
}
int main()
{
int n,m;
while(scanf("%d%d",&n,&m)!=EOF)
{
memset(s,0,sizeof(s));memset(f,0,sizeof(f));
for(int i=1;i<=n;i++){qr(s[i]);s[i]+=s[i-1];}
l=1;r=1;q[1]=0;
for(int i=1;i<=n;i++)
{
while(l<r&&calc(q[l],q[l+1])<2.0*s[i])++l;
f[i]=f[q[l]]+(s[i]-s[q[l]])*(s[i]-s[q[l]])+m;
while(l<r&&calc(q[r],i)<calc(q[r-1],q[r]))--r;
q[++r]=i;
}
qw(f[n]);puts("");
}
return 0;
}