handbook / Research Practices

Writing Papers

Motivation

Manuscripts get written for all sorts of reasons, from a fairly weak one (“we need a publication for the grant report”) to a genuine need to share a finding with the community. Whatever the reason, once you sit down to write, the goal should be the same: the reader should feel their time was well spent. Write for the reader, not just to document that the work happened.

Outline before you write

Draft the figures and the outline before full paragraphs. If you can’t outline it, you probably don’t yet know what the paper is about.

George Whitesides makes this case well in his short, classic essay Writing a Paper: go through several rounds of outlining before writing full text, because most of the thinking happens in deciding what the story is, not in the sentences. If co-authors don’t agree on the outline, no amount of polishing the prose will fix the paper.

Ten Simple Rules for Structuring Papers (Mensh & Kording, 2017) is a good companion to this: it gives a concrete way to build every paragraph and section around context, content, and conclusion, so the reader never loses track of why they’re reading a given sentence.

Content

Use simple, precise scientific language. If an explanation doesn’t feel convincing to you while writing it, it won’t convince anyone else either.
Structure your argument around three questions, and make sure a reader can answer all three from your intro and conclusion alone:
1. Why is the problem interesting?
2. How did you try to address it?
3. What are the key results?
Every paragraph and figure should be doing work for the reader, not just recording events in the order you did them.
Draw the figures early. They usually carry the core argument, so the text should support and interpret them rather than repeat what’s already visible in the plot.

Length and paper type

Length is mostly set by the target journal, but it helps to think about which format actually fits your result:

Letters / Communications are a short format, best suited to a focused, high-impact finding: a genuinely novel observation or a result that doesn’t need much methodological scaffolding to be convincing.
Full/regular articles (like a standard Physical Review B paper) suit a workflow, a method, or a systematic investigation of a phenomenon, where part of the contribution is the methodological detail and breadth of analysis itself.

In both cases the target is the same: say what the reader needs to trust and reuse your result, and nothing more.

Send a rough draft to your advisor and co-authors earlier than feels comfortable. Early feedback is cheap to act on. Late feedback usually means redoing figures too.

Authorship

Discuss authorship order early in a project, not at submission time. When in doubt, ask your advisor.

As a reference point, the ICMJE authorship criteria (widely used outside biomedicine too) say an author should meet all of the following:

Substantial contribution to the conception/design of the work, or to the acquisition, analysis, or interpretation of data;
Drafting the work or revising it critically for important intellectual content;
Final approval of the version to be published;
Agreement to be accountable for the accuracy and integrity of the whole work.

Funding, general supervision, or providing materials on their own usually don’t meet this bar. They’re real contributions and belong in the acknowledgments instead. Many journals now also ask for a structured author contribution statement using the CRediT taxonomy (14 standard roles such as conceptualization, methodology, software, writing, supervision). It’s worth running through with co-authors even when a journal doesn’t require it, since it surfaces “who did what” disagreements before they turn into disputes.

Use of AI

There’s nothing wrong with using AI tools to write from scratch, revise, polish, or just clean up your English. It would be a little ironic for a group that works on AI to then discourage its use elsewhere.

That said, it’s worth having a writing style of your own, one that’s recognizably yours. So for at least the first manuscript you write in the group, aim to have your own contribution outweigh the AI’s. Use AI to sharpen what you’ve written, not to write it for you from a blank page. If a journal or your co-authors ask, disclose AI use in the acknowledgments (most major publishers, including APS and RSC, now ask for this explicitly).

Figure making

Good figures are worth the time they take. Some general guidelines:

A decent paper (if it’s not a Letter/Communication) usually shouldn’t need more than about 7 main-text figures. If you find yourself past that, consider whether some belong in supplementary material.
Journal formatting/production teams often resize figures, change aspect ratios, or recompress them during typesetting. Submit figures at higher resolution and quality than you think you need, so they still hold up after that processing.
A common LaTeX mistake: preparing a figure as several separate subpanel files (e.g., with subfigure/subcaption) when the journal wants each figure submitted as a single combined image file. Flatten multi-panel figures into one file before submission using a vector editor like Inkscape (free) or Adobe Illustrator, or by building the whole panel layout directly in your plotting script (e.g., matplotlib’s GridSpec or subplots) so it’s already a single figure when exported. This avoids last-minute reformatting.
Read ACS’s “Mastering the Art of Scientific Publication” collection for general figure and writing advice that applies well beyond ACS journals.

Tools for plotting figures from scratch

Matplotlib (Python) — the default choice for most of us; flexible and scriptable, good for reproducibility.
Origin — common in experimental physics/chemistry labs, good for quick polished plots and error analysis.
Veusz — free, LaTeX-friendly scientific plotting, a lighter alternative to Origin.
Gnuplot — lightweight, scriptable, still widely used for quick plots from simulation output.
Plotly — useful when you want interactive figures, e.g. for a webpage or SI.

Tools for aligning and assembling panels

Keynote (if you’re on a Mac) — surprisingly good for quick, clean figure assembly.
Inkscape — free, cross-platform, vector editor; handles combining SVG/PDF panels into one figure well.
Adobe Illustrator — the standard in many labs, but paid.
PowerPoint — works in a pinch, less precise for exact alignment/spacing.

(add any group-specific templates or shared style files here, e.g., a shared matplotlib style sheet or Illustrator template)

Journal-specific figure guidelines

Formatting requirements (panel labeling, resolution, color, sizing) vary a lot by publisher. Click a journal below to go to its author guidelines for figures.

(this covers the journals we publish in most often; add more slices if the group starts targeting other publishers regularly)

Iteration between advisor and group member

Ideally, we discuss the direction of the draft together from the very beginning. In practice this is flexible and depends on how the project goes.

There can, and usually should, be several rounds of revision. Early rounds are mostly about content: is the argument right, is the evidence sufficient, is anything missing. Later rounds shift toward language and phrasing (with current AI tools, this stage usually doesn’t take long). Keep in mind that this order isn’t strict: some comments will only come up at the very end, even though they could in principle have been raised earlier.

If there’s a disagreement about how an idea or figure should be presented, and it’s a matter of presentation rather than scientific or technical correctness, the final call goes to whoever is leading the work. Disagreements on the science or technical content are a different matter and should be worked through together.

Further resources on scientific writing

(add any journal- or field-specific style guides your group commonly targets here.)

Literature Study →