Sensitivity conjecture
For every Boolean function , the sensitivity and the block sensitivity are polynomially related; specifically, for absolute constants (and in fact , where is the real polynomial degree of ).
In July 2019, Hao Huang proved that every induced subgraph of the -dimensional hypercube with vertices has maximum degree . Huang inductively constructed a signed adjacency matrix of satisfying , whose eigenvalues are each with multiplicity . By Cauchy's interlace theorem, any principal submatrix of has largest eigenvalue at least , forcing . Via the Gotsman–Linial equivalence (1992) and Tal's bound (2013), this immediately implies and .
Prior to Huang's theorem, decision-tree depth , certificate complexity , block sensitivity , polynomial degree , and quantum query complexity were all known to be polynomially equivalent, with sensitivity being the sole holdout. While is tight, the optimal exponent between and —known to lie between (Rubinstein's quadratic separation) and (Huang + Tal)—remains open.
References
- Hao Huang (2019). Induced subgraphs of hypercubes and a proof of the Sensitivity Conjecture · DOI:10.4007/annals.2019.190.3.6 · arXiv:1907.00847
- Craig Gotsman, Nathan Linial (1992). The equivalence of two problems on the cube · DOI:10.1016/0097-3165(92)90060-8