SOS DP

learn from blog of 海边拾贝的言

definition

sos: sum over subsets
target: $F(mask)={\sum }_{i\in mask}A(i)$
linked to FWT (need a complement), fellow are definition of FWT
$FWTor$ : $FWT(k)={\sum }_{iorj=k}A(i)\ast B(j)$
$FWTand$ : $FWT(k)={\sum }_{iandj=k}A(i)\ast B(j)$
$FWTxor$ : $FWT(k)={\sum }_{ixorj=k}A(i)\ast B(j)$

you can ignore FWT, because I don’t understand well and in this blog, it will not appear again.

solution

One obvious way to solve this is brute force, whose time complexity is $O(n^4)$.

for i = 1 to mask
	for j = 1 to mask
		if i in j
			F[j] += A[i]

It’s obvious that for a mask, we iterate all range is low efficient.
A method to imporve is to iterate the subset of mask.

for i = 1 to mask
	for j = i; j > 0; j = (j - 1) & i
		F[i] += A[j];

Time complexity $={\sum }_i{2}^{icontains1}=O(n^3)$

sos dp

$dp[mask][i]$ present the sum of subsets of mask that the former $i$ index has been fixed, and the value of left index can be changed so here the subsets mean the subset of latter.
$dp[mask][i]=\{\begin{array}{cc}dp[mask][i-1],& (mask\&(2^i)==0)\\ dp[mask][i-1]+dp[mask⨁2^i][i-1],& (mask\&(2^i)==1)\end{array}$

code:

for i = 1 to N
	for j = 0 to log(N)
		if (i & (2^j))
			dp[i][j] = dp[i][j - 1] + dp[i - (2^j)][j - 1];
		else
			dp[i][j] = dp[i][j - 1];

$dp[i][-1]=A[i]$
The below is dp of $i\in mask$, and the fellow if dp of $mask\in i$.
$dp[mask][i]=\{\begin{array}{cc}dp[mask][i-1]+dp[mask⨁2^i][i-1],& (mask\&(2^i)==0)\\ dp[mask][i-1],& (mask\&(2^i)==1)\end{array}$

code:

for i = 1 to N
	for j = 0 to log(N)
		if (i & (2^i))
			dp[i][j] = dp[i][j - 1];
		else
			dp[i][j] = dp[i][j - 1] + dp[i + (2^j)][j - 1];

The difference of $or$、$and$ can be thought of bits.
When $i\in mask$, the bit 1 of mask can change to 0 or 1, 0 of mask still remain 0, which is equivalent to $bit(!iormask)=1$;
When $mask\in i$, the bit 1 of $mask$ still remain 1, 1 of mask can change to 0 or 1, which is equivalent to $bit(ior!mask)=1$.