imaging-paper-style

name	imaging-paper-style
description	Writing-style guide for this project's papers, distilled from the author's own work (Salehi et al., arXiv:2412.06436) and the Bath/Cambridge mathematical- imaging school (M. J. Ehrhardt, C.-B. Schönlieb; e.g. arXiv:2510.01755). Apply to every abstract, introduction, contribution list, theorem remark, experimental description, and limitation section written for this project.

Mathematical-imaging paper style (Salehi / Ehrhardt / Schönlieb blend)

Voice and stance

• First-person plural, active, unadorned: "We consider", "We derive", "We show", "We propose". Never "novel", "remarkably", "surprisingly", "to the best of our knowledge" as filler (one "to our knowledge" is allowed for the single central novelty claim, nowhere else).
• Authoritative but never promotional. The result carries the sentence; the adjective never does. Delete every intensifier on sight ("very", "highly", "significantly" outside its statistical meaning, "extremely").
• No rhetorical questions in body text. (A question is permitted in the title.)

The signature sentence patterns

1. Constraint → consequence. State the mathematical fact, then its implication, in one sentence:

   "As A is ill-conditioned or non-invertible, the inverse problem is ill-posed in the sense of Hadamard." Use this to motivate every design decision: the reader should feel each choice is forced by mathematics, not preferred by the authors.

2. Purpose-driven contribution clauses. Each contribution opens with an infinitive of purpose:

   "To estimate the accuracy of the computed hypergradient, we derive an a-posteriori error bound…" / "To efficiently solve the upper-level problem, we also propose…"

3. Inexactness as fact, not confession. Limitations of the machinery are stated declaratively as properties of the setting:

   "Given that the lower-level problem is solved numerically, the loss and thus its gradient can only be computed inexactly." Never apologise; never frame as weakness; state, then handle.

Claim calibration

• Every theoretical claim carries its conditions in-sentence: "Under Assumptions 1–3, …", "for linear forward operators with known A, …".
• Empirical claims name the protocol in the same breath: "on n = 100 patient-stratified slices with block-clustered intervals".
• Distinguish verbs strictly: prove (theorems), show (empirics with CIs), observe (empirics without), expect (conjecture). Never upgrade.
• Negative and null results are reported in main text with the same sentence structure as positive ones. Comparative framing over dominance framing: "strengths and limitations", never "outperforms all baselines".

Structure conventions

• Motivate from the classical inverse-problems tradition (Hadamard, variational regularization, Morozov) before any learning content; the paper is problem-driven, not application-driven.
• Formal definitions arrive after one sentence of motivating context, not before and not pages later.
• Connectives of lineage and contrast: "In contrast to…", "Extending the idea of…", "To address this…", "Parallel to these developments…".
• Related work positions by assumption differences, not by adjectives: what each prior work assumes/requires, what we assume/require.

Experimental writing

• Experiments are investigations, not victory laps: state the question each experiment answers before its result.
• Pre-registered criteria and failed predictions are reported with the same prominence as confirmations, in the same tense and voice.
• Every number in prose carries its uncertainty and its n; every table caption states protocol, statistic, and what bold/underline mean.

Micro-rules

• Sentences medium-length; one idea each; embedded clauses only to attach a condition to its claim.
• Symbols defined at first use, reused consistently; no synonym drift (pick "adaptation depth" once; never alternate with "adaptation level").
• British-neutral spelling acceptable ("regularisation" or "regularization" — pick one per document; this project uses -ization in tex).
• Acknowledge funding/compute plainly at the end; no flourishes.