On the approximation of separable non-convex optimization programs to an arbitrary numerical precision
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
We consider the problem of minimizing the sum of a series of univariate (possibly non-convex) functions on a polyhedral domain. We introduce an iterative method with optimality guarantees to approximate this problem to an arbitrary numerical tolerance. At every iteration, our method replaces the objective by a piecewise linear relaxation to compute a dual bound. Since the polyhedral domain in our method remains unchanged, a primal bound is computed by evaluating the cost function on the solution provided by the relaxation. If the difference between these two values is deemed as not satisfactory, the relaxation is locally tightened with an objective-driven refinement procedure, that computes an optimal domain partitioning and the process repeated. By keeping the scope of the update local, the computational burden is only slightly increased from iteration to iteration. The convergence of the method is assured under very mild assumptions, and no NLP nor MINLP solver/oracle is required to ever be invoked to do so. As a consequence, our method presents very nice scalability properties and is little sensitive to the desired tolerance. We provide a formal proof of the convergence of our method, and assess its efficiency in approximating the non-linear variants of five problems: the transportation problem, the uncapacitated facility location problem, the multicommodity flow problem, the multi-commodity network design problem, and the continuous knapsack problem. Our results indicate that the overall performance of our method is competitive to three state-of-the-art mixed-integer nonlinear solvers, often performing better. It also scales better than a naive variant of the method that avoids performing successive iterations in exchange of solving a much larger mixed-integer linear program.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it