Cost of all layers of the dynamic program

Proved

OAI.ThreeMachine.StackCompiler.Costs.fullLayers

by wurtle · Nov 4, 2026 · Mathlib 0df444a (Lean v4.33.1)

complexityschedulingthree-machine

Throughout, a routine R : Uniform f is a program in the compiler's while-language Code, with a charge function, that computes the function family $f_i$ on encoded data. R.time i x is the charge of $R$ on the encoding of $x$ , a natural number that bounds the number of evaluation steps of the program on that input. For $s,f:I\to\mathbb{N}$ , the predicate Poly s f d means that there is a constant $C$ with $f(i)\le C\,(s(i)+2)^d$ for all $i\in I$ . For $K\le 10000$ , the routine Uniform.fullLayers K, computing Algorithm.layers over effectiveFamily with formula size $K$ , has running time satisfying Poly of degree $150010$ in the number $n$ of jobs.

Formalization Note The statement is reproduced verbatim from upstream SchedulingCosts.lean (openai/math, Apache-2.0, credit OpenAI). It feeds the bound requestRawBound_poly and the running-time statement of main_theorem.

Preamble

import Mathlib
import Definitions.Def_OAIThreeMachineSched02

set_option maxHeartbeats 2000000
set_option maxRecDepth 100000
set_option linter.all false
set_option Elab.async false

namespace OAI

section
namespace ThreeMachine.StackCompiler.Costs
variable {J : Type} (n : J → ℕ) (U : ∀ j, Universe (n j)) (M : ∀ j, Matrix (n j))
variable (tab : ∀ j, Table (n j)) (fs zs : ∀ j, List (Finset (Fin (n j))))

end ThreeMachine.StackCompiler.Costs
end

section
namespace ThreeMachine.StackCompiler.Uniform
open ThreeMachine.Structure

end ThreeMachine.StackCompiler.Uniform
namespace ThreeMachine.StackCompiler.Costs
open ThreeMachine.Structure

Formal statement

theorem fullLayers (K : ℕ) (hk : K ≤ 10000) : Poly (fun x : Σ n, Universe n × Matrix n => x.1)
    (fun x => (Uniform.fullLayers K).time x.1 x.2) 150010 := by
  sorry

end ThreeMachine.StackCompiler.Costs
end
end OAI

Source

OpenAI, "A Polynomial-Time Algorithm for Three-Machine Unit-Job Scheduling" (September 24, 2026), openai/math, cost bounds of the stack compiler. Lean: https://github.com/openai/math/blob/adc7f1241b42e322a6451854ab7e4b4c146bf78a/lean/OAI/Computability/Scheduling/SchedulingCosts.lean#L106

View graph