Guarantee
Returns an integer schedule of makespan at most (1+\varepsilon)\mathrm{OPT}, or detects infeasibility even without a load bound.
Objective and output
Assign every original type total while preserving one exact assignment-cost budget.
Randomness
Deterministic profile enumeration and exact integral recovery, including the compressed-slot-surplus branch, with paid deterministic primitives.
Time
With \delta=\varepsilon/2, K=\lfloor1/\delta\rfloor, and L=\left\lceil(1-\delta)/\delta^2\right\rceil, time is
Space
Polynomial in the complete binary input and paid oracle workspace; numerical resource multiplicities are not expanded into individual units. The rational accuracy is never expanded into unary coordinates.
Parameters
q is the physical-machine count; 0<\varepsilon\le1 is rational; H_\varepsilon=H+\mathrm{size}(\varepsilon); all original totals n_j=f_j(P) are mandatory. H includes the explicit ground, all rational and integer encodings, and paid rank, submodular-minimization, LP, intersection, and decomposition work.
Model and representation
Arbitrary nonnegative rational processing times and assignment costs, integral job polymatroids, eligibility masks, and one exact budget. Every original resource base must be recovered.
Lower-bound scope
No lower bound is claimed for this algorithm.
Review status
Ready for review. The complete exact-budget arbitrary-processing Theorem PS: original mandatory job-polymatroid bases, arbitrary nonnegative rational pair times/costs, exact bin boundaries, sparse occupied-bin grouping, load-pruned profile streaming with both composition and partition count bounds, one joint minimum-cost upper-prefix recovery, zero feasibility and one-sided grid optimization. The separate common-type-size slot-surplus Theorem S is also audited, including the fixed full-machine workload dual, binary scaled decomposition, guarded compressed count recursion, large-corank enumeration for polynomial space and the supplied-surplus-budget target search. The reference implements the PS profile and shared joint recovery semantics with explicit-table solver costs.
Dependencies and scope. The main approximation scheme uses the stated paid exact rank/SFM, rational LP and integral polymatroid-intersection primitives; its accuracy encoding and occupied-bin polynomial-space ledger remain explicit. The shipped reference shares canonical LP-vertex and numeric count enumeration, whose costs and potentially exponential vertex storage are charged separately from the fast theorem. The slot-surplus companion has common positive type sizes and no assignment budget. Its stronger bound retains the slot-surplus promise at OPT and the additive largest-smaller-job term, alongside the unconditional additive-largest-job fallback. That distinct compressed-slot algorithm is mathematically proved but not implemented by the profile reference. Finite reference tests, complete mathematical proofs, formal verification, external review and publication priority are distinct; the latter three are not claimed.
The verification note identifies executable coverage. This status does not claim external peer review or proof-assistant verification.
Source
Manuscript and proof, with dependencies and verification sources.
The compressed slot-surplus theorem handles common
processing sizes and mandatory original job-polymatroid bases, with
arbitrary binary multiplicities and no assignment-cost budget.
Let a be the maximum size, h its mandatory total, u the largest
smaller size, and
s_p(T)=\min{f_b({p}),\lfloor T/a\rfloor}, c_{\mathrm{slot}}(T)=\sum_p s_p(T)-h.
C_{\max}\le\mathrm{OPT}+a
Always retained as a polynomial fallback
C_{\max}\le\min{\mathrm{OPT}+a,(1+\varepsilon)\mathrm{OPT}+u}
\Gamma(1/\varepsilon)^k\mathrm{poly}(H) when c_{\mathrm{slot}}(\mathrm{OPT})\le k, for rational 0<\varepsilon\le1/2
Factor 1+\eta
\Gamma(2/\eta)^k\mathrm{poly}(H) when u\le(\eta/2)\mathrm{OPT} and c_{\mathrm{slot}}(\mathrm{OPT})\le k, for rational 0<\eta\le1
Here \Gamma(R)=R^R/(R-1)^{R-1}. The supplied slot-surplus budget k
can itself be binary encoded. A hybrid count enumeration guarantees
polynomial workspace even when its numerical value exceeds the number
of machines. This theorem does not impose at most one largest unit per
machine, and its additive smaller-unit loss is explicit.
Attribution. icekylinx · GPT-6 Astra Pro