Writing study: Query Lower Bounds for Diffusion Sampling
← NeurIPS 2026 oral writing corpus
Zhiyang Xun, Eric Price · Official NeurIPS 2026 oral record
Figure, table and section ledger
Editorial contribution type: Theory & formal analysis (primary).
Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.
- Canonical PDF
- 29 physical pages
- Before references / appendix
- ≈ 12.39 page-equivalents
- Main figures / tables
- 3 / 0
- Attached appendix figures / tables
- 0 / 0
- Attached appendix span
- ≈ 11.91 page-equivalents
Source-bound correction: Top-level heading and reference/appendix boundaries checked against source PDF text. Bibliography titles/author initials, numbered contribution/property bullets, diagram steps, equations and TOC entries are not substantive headings. Section spans are page-coordinate estimates, not ink area or prose-only length. This is an author manuscript mirror, not a verified accepted-version template. Source layout uses physical single-column y/page-height, including equation-heavy pages.
Section-by-section lengths and source pages
The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.
| Top-level heading | Starts on PDF page | Estimated span | Share of entire PDF |
|---|---|---|---|
| 1 Introduction | 1 | ≈ 3.78 pp. | 13.03% |
| 2 Proof Overview | 5 | ≈ 4.92 pp. | 16.97% |
| 3 Related Work | 10 | ≈ 1.08 pp. | 3.72% |
| 4 Applicability to Practical Algorithms | 11 | ≈ 1.18 pp. | 4.07% |
| 5 Conclusion | 12 | ≈ 0.84 pp. | 2.9% |
| Acknowledgments | 13 | ≈ 0.09 pp. | 0.31% |
| References | 13 | ≈ 4.7 pp. | 16.21% |
| A Setup | 18 | ≈ 0.4 pp. | 1.38% |
| B The target–oracle family | 18 | ≈ 3.6 pp. | 12.41% |
| C The informative window is narrow | 22 | ≈ 1.37 pp. | 4.72% |
| D From oracle indistinguishability to sampling lower bounds | 23 | ≈ 2.27 pp. | 7.83% |
| E An extension to non-uniform weights | 25 | ≈ 0.86 pp. | 2.97% |
| F Subexponential score errors | 26 | ≈ 1.13 pp. | 3.9% |
| G A Wasserstein extension | 27 | ≈ 0.65 pp. | 2.24% |
| H A synthetic illustration of the informative window | 28 | ≈ 1.62 pp. | 5.59% |
All detected figure and table locations
Attached appendix
No recognized caption labels.
Argument and organization
- Problem and contribution
States a query-complexity question for score-based sampling under an explicit oracle model, then separates number of queries from number of distinct smoothing levels.
- Argument sequence
Formal upper/lower-bound context is followed by a proof overview that first uses an intuitive counting story, then explains why Gaussian concentration changes the rate.
- Evidence and verification
The result statements retain target classes, error notions and oracle assumptions; synthetic proxies illustrate the shrinking informative window separately from the proof of a lower bound.
Problem, claim and contribution
States a query-complexity question for score-based sampling under an explicit oracle model, then separates number of queries from number of distinct smoothing levels.
Section 1.1, pp. 2–4 · Main paper
Argument and section order
Formal upper/lower-bound context is followed by a proof overview that first uses an intuitive counting story, then explains why Gaussian concentration changes the rate.
Section 2, pp. 5–9 · Main paper
Figures, captions and visual explanation
Figure 2 draws off-distribution, informative and averaging-out regimes with the same query marker and changing mixture density. Brackets and labels, not color alone, identify the narrow informative regime. Figure 1 is a query-complexity ordering, not a measured-speed curve. It places the proposed lower bound alongside prior upper bounds and marks the ruled-out polylogarithmic region; its caption binds the comparison to two assumptions stated immediately below.
Figure 2, p. 6 (render inspected); Figure 1 and Assumptions 1–2, p. 3 (additional rendered inspection) · Main paper
Tables, comparisons and evidence
The result statements retain target classes, error notions and oracle assumptions; synthetic proxies illustrate the shrinking informative window separately from the proof of a lower bound.
Section 1.1, pp. 2–4; Figure 3, p. 7; Appendix H, pp. 28–29 · Main paper
Appendix and reproducibility
The appendix decomposes the lower-bound argument into score control, window analysis and conversion to sampling hardness, followed by extensions with changed assumptions.
Appendices B–G, pp. 18–27 · Main paper
Clarity and writing conventions
The proof overview explicitly flags where its first intuition fails before repairing it; the practical discussion distinguishes worst-case oracle limitations from typical trained diffusion models.
Section 2, p. 6; Section 4, pp. 11–12 · Main paper
What to reuse—and what not to infer
Transferable lesson: A theory figure can visualize the hard instance and the informative regime, while theorem prose specifies exactly which algorithms and distributions the bound covers.
Boundary: Do not present a worst-case score-oracle lower bound as a universal empirical barrier for all diffusion models or special structured distributions.
Apply this lesson using the argument and evidence worksheet.
Inspection record and sources
- Checked
- 2026-10-02
- Reading status
- Writing-focused manuscript scan + selected visuals inspected
- Main text
- Writing-oriented scan across manuscript framing, section sequence, figure/table captions, selected argument/evaluation prose and conclusion; not exhaustive technical or proof verification. Sections 1–5, pp. 1–12, including result statements, proof overview, related work, practical implications and conclusion. Source is the exact public author version listed below, not a verified accepted/camera-ready revision.
- Appendix / supplement
- Writing-oriented appendix/documentation scan: Appendix material on pp. 17–29: score/oracle bounds, informative windows, indistinguishability, nonuniform mixtures, error regimes, Wasserstein guarantees and synthetic illustrations. Source code, data, model outputs and external video assets were not executed or independently validated.
- Visually inspected PDF pages
- 3, 6
This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.