Online Load Balancing on Unrelated Machines
ProvedPrimalDual.buchbinder_naor_unrelated_machines_exactOnline load balancing on unrelated machines (exact form of Theorem 8.1). Consider a normalized instance with machines, jobs arriving in order, and nonnegative normalized loads , where machine is eligible for job exactly when . If the packing dual admits a feasible solution assigning unit total mass to every job, then the deterministic Buchbinder–Naor primal–dual algorithm never fails, assigns every job irrevocably, and every machine ends with normalized load
The source states the guarantee as ; the displayed constant is the explicit bound derived in its proof. In the survey's doubling scheme this normalized phase is invoked with a guessed makespan, and a feasible dual assigning all jobs exists precisely when the guess is at least the optimal makespan, which is how the phase yields the -competitive online algorithm.
Formalization Note The theorem covers the single normalized phase; the doubling wrapper that produces the competitive algorithm is outside the mission's scope. The algorithm, its state, and the dual-feasibility premise all refer to the mission's shared definitions module.
import Definitions.Def_pd_unrelated_machines
namespace PrimalDual
/--
**Online load balancing on unrelated machines (exact form).** If the normalized
instance admits a feasible dual assigning unit mass to every job, the
Buchbinder--Naor deterministic online algorithm assigns all jobs, never fails,
and puts normalized load at most `ln(3m) / ln(3/2)` on every machine.
-/
theorem buchbinder_naor_unrelated_machines_exact
{jobs machines : ℕ} (I : UnrelatedMachines.Instance jobs machines)
(y : I.EligiblePair → ℝ) (hy : I.DualAssignsAll y) :
I.Succeeded ∧ I.AssignsAll ∧
∀ j, I.assignedLoad j ≤ I.logarithmicLoadBound := by
sorry
end PrimalDual