Complexity class of sensitivity

Question

Let $f:\{0,1\}^n\rightarrow \{0,1\}$ be a Boolean function.

Let $[n]=\{1,2,\dots,n\}$.

If $i\in[n]$, let $\Bbb 1_i$ be length $n$ vector with all $0$s except $1$ at $i$th position.

If $B\subseteq [n]$, then $\Bbb 1_B$ be the length $n$ vector with $1$s only in positions marked by $B$.

If $i\in[n]$ and $x\in\{0,1\}^n$, let $x^i=x\oplus\Bbb 1_i$ where $\oplus$ is $XOR$ operation.

If $B\subseteq [n]$ and $x\in\{0,1\}^n$, let $x^{B}=x\oplus\Bbb 1_B$ where $\oplus$ is $XOR$ operation.

Sensitivity of $f$ at input $x$ is $$S_x(f) = |\{i:f(x)\neq f(x^i)\}|$$

Sensitivity of $f$ is $$S(f)=\max_xS_x(f)$$

Block Sensitivity of $f$ at input $x$, $BS_x(f)$ is maximum $r$ such that there is a set of disjoint subsets $\{B_i\}_{i=1}^r$($\forall i\neq j$, $B_i\cap B_j=\emptyset$) such that $$\forall j\mbox{, }f(x)\neq f(x^{B_j})$$

Block Sensitivity of $f$ is $$BS(f)=\max_xBS_x(f)$$

Given $f$ in $n$-variables, would it be reasonable to ask the complexity class of deciding $S(f)>c$ and $BS(f)>c$ for a fixed $c>0$? I understand that describing a Boolean function takes $2^n$ bits fully.

Does computing $BS(f)$ also take $O(2^{n+\epsilon})$ time as in answers below?

Answer 1

For fixed $c$ the complexity is polynomial time. In fact, you can calculate $S(f)$ in polynomial time by just going over all inputs and calculating the pointwise sensitivity. This takes $O(n2^n)$ on a RAM machine. Given $c > 0$, there are at most $(2+c)^n = (2^n)^{\log_2 (2+c)}$ ways to choose $c+1$ disjoint subsets, and you can check whether there is a point sensitive to all of them in time $O(2^n)$. In total, this gives an $O((2^n)^{\log_2 (2+c)+1})$ algorithm for determining whether $bs(f) > c$.

This leaves open the question of calculating the block sensitivity of a function $f$.

Answer 2

[An alternative answer for $BS(f)$]

There is a paper related to this problem.

Scott Aaronson: Algorithms for Boolean Function Query Properties. SIAM J. Comput. 32(5): 1140-1157 (2003)

The paper gives an $O(N^{\log_2 5} \log N)$ algorithm for computing $BS(f)$ if $f$ is given by a truth table of size $N=2^n$. Thus, even if $c$ is not fixed, deciding $BS(f) > c$ is in P.