Decision trees with CNF or DNF leaves
DefinitionParityBarrier563_Trilinear_Supplement_DecisionTreeboolean-circuitscomplexity-theoryp2o-batch-stoc27sp2o-gran-per-chapterp2o-plan-paperp2o-v1
A finite binary decision tree queries one input variable at an internal node and has a Boolean function on the whole input cube at each leaf. Evaluation follows queried bits and evaluates the reached leaf on the same input. Define
GoodLeaves means that every leaf function is a k-CNF or a k-DNF; the alternatives may be mixed. Equality of tree evaluation with F requires agreement only on each leaf's path subcube. This is the representation used in Lemma 6.1, including depth zero.
Definition code
import Mathlib
import Definitions.Def_ParityBarrier563_Trilinear_CNF
set_option autoImplicit false
open scoped Classical
namespace ParityBarrier563.Trilinear.Supplement
/-- A finite decision tree with Boolean functions at its leaves, not necessarily constants. -/
inductive LeafTree (V : Type)
| leaf (F : BoolFn V)
| query (v : V) (zero one : LeafTree V)
namespace LeafTree
variable {V : Type}
/-- Evaluation follows the queried bit and evaluates the reached leaf function on that input. -/
def eval : LeafTree V → BoolFn V
| .leaf F => F
| .query v zero one => fun x => if x v then one.eval x else zero.eval x
/-- Depth counts queries, so a leaf has depth zero. -/
def depth : LeafTree V → ℕ
| .leaf _ => 0
| .query _ zero one => 1 + max zero.depth one.depth
/-- Leaf types may be mixed; agreement with the target is required only along reached paths
by equality with eval, not by a global equality of every leaf with the target. -/
def GoodLeaves [DecidableEq V] (k : ℕ) : LeafTree V → Prop
| .leaf F => IsCNF k F ∨ IsDNF k F
| .query _ zero one => zero.GoodLeaves k ∧ one.GoodLeaves k
end LeafTree
end ParityBarrier563.Trilinear.Supplement
Source
Nikolai Chukhin, Xin Li, Ivan Mihajlin, Yan Zhong, Breaking the Parity Barrier for Constant-Depth Circuits, STOC 2027 submission #563 (October 2026), §2, printed p.11 / PDF 13; Lemma 6.1, printed p.26 / PDF 28