令 (n) 为值域。
首先有个 (mathcal O(3^{log_2n}) = mathcal O(n ^ {1.5})) 的子集 DP。
式子很像子集卷积,但如果把每个数依次卷起来复杂度过高,不可接受。
考虑分组,发现有相同一个二进制位的数必然不会被卷到一起。
但如果按照一般二进制位分成 (log n) 组,再依次卷积,复杂度是 (mathcal O(n log^3 n)),其中每次卷积是 (mathcal O(n log^2 n))。
考虑限制每组元素的个数,按照最高位分组,这样每组的二进制长度受到限制 ,复杂度就会为:
[sum_{i=1}^{log n} 2^i cdot i^2 = mathcal O(n log^2 n)
]
其中 (i) 为每组的长度。
Code
当然可以继续压常数
#include <cstdio>
#include <algorithm>
using namespace std;
#define File(s) freopen(s".in", "r", stdin), freopen(s".out", "w", stdout)
typedef long long ll;
namespace io {
const int SIZE = (1 << 21) + 1;
char ibuf[SIZE], *iS, *iT, obuf[SIZE], *oS = obuf, *oT = oS + SIZE - 1, c, qu[55]; int f, qr;
#define gc() (iS == iT ? (iT = (iS = ibuf) + fread (ibuf, 1, SIZE, stdin), (iS == iT ? EOF : *iS ++)) : *iS ++)
char getc () {return gc();}
inline void flush () {fwrite (obuf, 1, oS - obuf, stdout); oS = obuf;}
inline void putc (char x) {*oS ++ = x; if (oS == oT) flush ();}
template <class I> inline void gi (I &x) {for (f = 1, c = gc(); c < '0' || c > '9'; c = gc()) if (c == '-') f = -1;for (x = 0; c <= '9' && c >= '0'; c = gc()) x = x * 10 + (c & 15); x *= f;}
template <class I> inline void print (I x) {if (!x) putc ('0'); if (x < 0) putc ('-'), x = -x;while (x) qu[++ qr] = x % 10 + '0', x /= 10;while (qr) putc (qu[qr --]);}
struct Flusher_ {~Flusher_(){flush();}}io_flusher_;
}
using io :: gi; using io :: putc; using io :: print; using io :: getc;
template<class T> void upmax(T &x, T y){x = x>y ? x : y;}
template<class T> void upmin(T &x, T y){x = x<y ? x : y;}
const int N = 262144;
const int mod = 1000000007;
int phi[N + 1], p[N + 1], pc = 0;
bool np[N + 1];
inline int add(int x, int y) {return x+y>=mod ? x+y-mod : x+y;}
inline int sub(int x, int y) {return x-y<0 ? x-y+mod : x-y;}
inline int mul(int x, int y) {return 1LL * x * y % mod;}
inline void inc(int &x, int y=1) {x += y; if(x >= mod) x -= mod;}
inline void dec(int &x, int y=1) {x -= y; if(x < 0) x += mod;}
inline int power(int x, int y){
int res = 1;
for(; y; y>>=1, x = mul(x, x)) if(y & 1) res = mul(res, x);
return res;
}
inline int inv(int x){return power(x, mod - 2);}
void getPhi(int n){
phi[1] = 1;
for(int i=2; i<=n; i++){
if(!np[i]) p[++pc] = i, phi[i] = i - 1;
for(int j=1; j<=pc && i * p[j] <= n; j++){
np[i * p[j]] = true;
if(i % p[j] == 0){
phi[i * p[j]] = phi[i] * p[j];
break;
}
phi[i * p[j]] = phi[i] * (p[j] - 1);
}
}
}
void FWT(int a[], int len){
for(int w=2, k=1; w<=len; w<<=1, k<<=1)
for(int i=0; i!=len; i+=w)
for(int j=i, li=i+k; j!=li; ++j)
inc(a[j + k], a[j]);
}
void IFWT(int a[], int len){
for(int w=2, k=1; w<=len; w<<=1, k<<=1)
for(int i=0; i!=len; i+=w)
for(int j=i, li=i+k; j!=li; ++j)
dec(a[j + k], a[j]);
}
void subsetConv(int a[], int b[], int c[], int len){
static int popc[N];
static int A[19][N], B[19][N], C[19][N];
for(int i=0; i<len; i++){
popc[i] = popc[i >> 1] + (i & 1);
A[popc[i]][i] = a[i];
B[popc[i]][i] = b[i];
}
int n = popc[len - 1];
for(int i=0; i<n; i++){
FWT(A[i], len);
FWT(B[i], len);
}
for(int k=0; k<=n; k++)
for(int i=0; i<=k; i++)
for(int p=0; p<len; p++)
inc(C[k][p], mul(A[i][p], B[k - i][p]));
for(int i=0; i<=n; i++) IFWT(C[i], len);
for(int i=0; i<len; i++) c[i] = C[popc[i]][i];
for(int i=0; i<=n; i++){
fill_n(A[i], len, 0);
fill_n(B[i], len, 0);
fill_n(C[i], len, 0);
}
}
int cnt[N];
int a[19][N], num[N];
int main(){
getPhi(262144);
int n;
gi(n);
static int high[N];
for(int i=1; i<262144; i++)
high[i] = high[i >> 1] + 1;
int bit = 0;
for(int i=0; i<n; i++){
int x; gi(x);
++a[high[x]][x];
upmax(bit, high[x]);
}
static int s[N];
s[0] = power(2, a[0][0]);
s[1] = mul(s[0], a[1][1]);
for(int i=2; i<=bit; i++){
a[i][0] = 1;
subsetConv(s, a[i], s, 1 << i);
}
int sum = 0;
for(int i=0; i<(1 << bit); i++)
inc(sum, mul(phi[i + 1], s[i]));
print(sum); putc('
');
return 0;
}