Junta approximation of monotone Boolean functions
DefinitionThe Servedio–Tan–Verbin conjecture#✓fully proved
Order \(\{0,1\}^n\) coordinatewise with \(0 < 1\), and write \(\deg f\) for the Boolean degree of a function \(f : \{0,1\}^n \to \{0,1\}\). This is the statement that for every real \(\varepsilon > 0\) there is a natural number \(C\) such that for every \(n \in \mathbb{N}\) and every monotone \(f : \{0,1\}^n \to \{0,1\}\) there are a set \(J\) of at most \((\deg f + 2)^C\) coordinates and a function \(g : \{0,1\}^n \to \{0,1\}\) that depends only on the coordinates in \(J\) such that \(f(x) \ne g(x)\) for at most \(\varepsilon 2^n\) points \(x \in \{0,1\}^n\).
The junta size \((\deg f + 2)^C\) expresses a polynomial bound in \(\deg f\): every polynomial in \(d \ge 0\) is at most \((d + 2)^C\) for some \(C\).
def servedioTanVerbinConjecture : Prop :=
∀ ε : ℝ, 0 < ε → ∃ C : ℕ, ∀ (n : ℕ) (f : (Fin n → Bool) → Bool), Monotone f →
∃ (J : Finset (Fin n)) (g : (Fin n → Bool) → Bool),
J.card ≤ (BooleanCube.booleanDegree f + 2) ^ C ∧ DependsOn g (J : Set (Fin n)) ∧
((Finset.univ.filter fun x => f x ≠ g x).card : ℝ) ≤ ε * 2 ^ n
noncomputable def booleanDegree {n : ℕ}
(f : BoolCube n → Bool) : ℕ :=
(multilinearExtension f).totalDegree
MathlibExt/Analysis/Fourier/BooleanCube.lean:246
Uses 1 · Used by 1
TheoremThe Servedio–Tan–Verbin conjecture is false#●ready to prove
Order \(\{0,1\}^n\) coordinatewise with \(0 < 1\), and write \(\deg f\) for the Boolean degree of a function \(f : \{0,1\}^n \to \{0,1\}\). It is not the case that for every real \(\varepsilon > 0\) there is a natural number \(C\) such that for every \(n \in \mathbb{N}\) and every monotone \(f : \{0,1\}^n \to \{0,1\}\) there are a set \(J\) of at most \((\deg f + 2)^C\) coordinates and a function \(g : \{0,1\}^n \to \{0,1\}\) that depends only on the coordinates in \(J\) such that \(f(x) \ne g(x)\) for at most \(\varepsilon 2^n\) points \(x \in \{0,1\}^n\).
Kane constructed, for a fixed \(\varepsilon > 0\), a constant \(c > 0\) and every \(d\), monotone functions of decision-tree depth \(d\) (hence degree at most \(d\)) that are \(\varepsilon\)-far from every junta on \(\exp(c \sqrt{d})\) coordinates; since \(\exp(c \sqrt{d})\) eventually exceeds \((d + 2)^C\), no polynomial junta size works for \(\varepsilon / 2\). O'Donnell's list records the conjecture as resolved in the negative by Kane.
public theorem_wanted servedioTanVerbinConjecture_refuted : ¬ servedioTanVerbinConjecture
def servedioTanVerbinConjecture : Prop :=
∀ ε : ℝ, 0 < ε → ∃ C : ℕ, ∀ (n : ℕ) (f : (Fin n → Bool) → Bool), Monotone f →
∃ (J : Finset (Fin n)) (g : (Fin n → Bool) → Bool),
J.card ≤ (BooleanCube.booleanDegree f + 2) ^ C ∧ DependsOn g (J : Set (Fin n)) ∧
((Finset.univ.filter fun x => f x ≠ g x).card : ℝ) ≤ ε * 2 ^ n
WantedExt/Combinatorics/MonotoneJuntaApproximationAMR127Problem0022Wanted.lean:46
Sources
- R. O'Donnell, Open Problems in Analysis of Boolean Functions (2012), arXiv:1204.6447, Problem 24 (Servedio–Tan–Verbin Conjecture); resolution: D. M. Kane, A low-depth monotone function that is not an approximate junta, arXiv:1206.6541 (2012); recorded as "RESOLVED (in the negative) by Daniel Kane, 2012" in O'Donnell's list, Problem 24, Kane, Theorem 2: there is a constant ε > 0 such that for every d there are k = exp(Ω(√d)) and a monotone f : {0,1}ⁿ → {0,1} computed by a depth-d decision tree (so deg f ≤ d) that disagrees with every k-junta on at least an ε-fraction of inputs, https://arxiv.org/abs/1206.6541.
Uses 1
Loading dependency explorer…
The dependency explorer needs JavaScript and its graph data.