In short
A famous machine-learning paper set again from its open-access source, with numbered equations you can refer to, boxed theorems and lemmas, proofs, and author–year references.
What you'll build
Direct Preference Optimization by Rafailov, Sharma, Mitchell, Ermon, Manning and Finn (NeurIPS 2023), abridged from its arXiv source and set as a ten-page preprint on a 7 × 10 in page. A plum band carries the title over a fan of logistic curves, the shape of the weight that the DPO gradient puts on each example. Below it, one generous column of Spectral holds thirteen numbered equations, the definition, lemmas and theorem of section 5 as boxes with run-in labels, a proof sketch and a full proof that end in a square, a pipeline figure redrawn in code and twenty works cited author–year. Every equation, statement and section reference in the text is a link, and every number comes from a label, as in LaTeX.
This recipe answers
- How do I number equations, state theorems and set proofs in a mathematical paper, and refer to them?
- How do I typeset maths (inline, display, equations) and keep it vector in the PDF?
- How do I re-set an open-access paper from arXiv or PubMed Central with Postext, keeping its citations, figures and licence line?
The short answer
// \label{eq:x} in a display formula numbers it, on its row of an align; a box opened as
// :::callout{type="lemma" #lem:x} counts as a statement. \eqref{eq:x}, \ref{lem:x} and
// :ref{id="lem:x"} print the number and link to it.
const equationNumbering = { // (1) to (13): one sequence through the paper and its appendix
numberingTemplate: '{n}', resetOn: 'never', format: '({n})' };
// A proof: an italic label with no number, and a □ flush right on its last line, drawn
// from TeX's \square because maths runs (Spectral has no □ of its own).
const proofOf = (label) => ({ endMark: '□',
numbering: { label, counter: false, bold: false, italic: true } });
const statements = [ // a counter per kind, as the paper has it: Definition 1, Lemma 1, Theorem 1
{ id: 'definition', numbering: { label: 'Definition' } },
{ id: 'lemma', numbering: { label: 'Lemma' } }, // counter: 'theorem' would share one sequence
{ id: 'theorem', numbering: { label: 'Theorem' } },
{ id: 'proof', ...proofOf('Proof') },
{ id: 'sketch', ...proofOf('Proof Sketch') },
];
Amsthm in Markdown: numbered equations, theorems and proofs
Ingredients
- Features
- MathematicsCallout boxesCross-referencesCitations in a citation styleBibliography from the referencesNumbered headingsHeading stylesHeading attributesDesigned openersRunning heads and foliosHeads by page roleFigures and tables as resourcesNumbered captionsCaption styleTable styleSemantic colour paletteSource and credit linesFigure placementPDF export
- Also uses
- Boxes that split or stay wholeColumn balancingCitations that place figuresFigures exactly hereFigure and Table in your languageLeaving the grid on purposePage and column breaksParagraph stylesFonts embedded in the PDFCustom resource typesSuperscripts and subscriptsLine breaks in titlesUnnumbered chapters
- Type
- Spectral, Work Sans, JetBrains Mono (SIL OFL 1.1)
- Assets
- None: every picture is drawn in code
Method
#1 · Label equations and statements, refer to them by key
The code is the short answer above. Since postext 1.19 the engine keeps LaTeX's counters. A \label{eq:dpo} in a display formula numbers it in reading order, one number for each labelled row of an align and none for a row marked \notag; resetOn: 'never' carries the count from (1) through the appendix to (13), as this abridged paper is renumbered. A box opened as :::callout{type="lemma" #lem:same-policy} starts with “Lemma 2.”, because its style has numbering: { label: 'Lemma' } and a counter of its own. In the text, \eqref{eq:dpo} prints (7), \ref{lem:same-policy} prints 2 and :ref{id="lem:same-policy"} prints Lemma 2, each a link to its target; Eq.~\eqref{…} keeps LaTeX's tie. The restated lemma of the appendix opens with ***:ref{id="lem:same-preference"} Restated.***, and the reference takes the bold of the label around it. Sections need no help: their numbers come from the heading templates, {startAt=3} keeps the paper's own numbering after the cut section 2, and crossRefs.section prints “Section 5” with a capital, as the paper does.
#2 · Give each environment its own box
const box = ({ id, ...numbered }, extra) => ({ id, marginTop: pt(LEAD * 0.6),
marginBottom: pt(LEAD * 0.6), backgroundEnabled: false,
snapToGrid: false, // exact space round a statement, as amsthm's \topsep; one column, no grid
padding: { top: mm(1.8), right: mm(4), bottom: mm(1.8), left: mm(4) },
body: { italic: true, firstLineIndent: pt(0), boldColor: col('ink') }, ...extra, ...numbered });
const stripe = (color) => ({ enabled: true, side: 'left', width: pt(3), color: col(color) });
const proof = { keepTogether: false, // a long proof runs on to the next page
// A proof is upright text with an italic run-in label, set off by space alone.
padding: { top: pt(0), right: pt(0), bottom: pt(0), left: pt(0) },
body: { firstLineIndent: pt(0), italicColor: col('ink') } };
const look = { definition: { stripe: stripe('rule') }, lemma: { stripe: stripe('accent') },
theorem: { backgroundEnabled: true, background: col('tint') }, proof, sketch: proof };
const theoremStyles = [...statements.map((s) => box(s, look[s.id])),
box({ id: 'restated' }, look.lemma)]; // Lemma 1 again in the appendix, with no new number
Each environment is a callout style with one device: a grey stripe for the definition, a plum stripe for the lemmas, a pale plum tint for the theorem. box() lays that look over the numbering of the short answer. body.italic sets the statement in italics as amsthm's plain style does, and the run-in label stays bold and upright inside it. A proof has no box at all: its label is italic and has no number (counter: false), endMark: '□' sets a square flush right on its last line, drawn from TeX's \square because maths runs, and keepTogether: false lets a long proof run on to the next page. snapToGrid: false keeps the same space around every statement; in one column with no neighbour to line up with, an exact gap reads better than one rounded to the grid.
#3 · Cite author–year from the paper's own reference list
registerCitationEngine(createCiteprocEngine({ styles: STYLES, locales: LOCALES }));
const citations = { style: 'harvard-cite-them-right', link: true,
bibliography: { fontSize: pt(8.2), lineHeight: pt(10.8), hangingIndent: mm(5),
entrySpacing: pt(1.6), doi: 'link' } };
The references are BibTeX in a :::references{format=bibtex} block, rebuilt from the paper's .bbl with only the twenty works the abridged text still cites. Cite Them Right Harvard sets them as “(Bong and Rinaldo, 2022)” in the text and “Ouyang, L. et al. (2022)” in the list, cutting author lists of four or more to the first name. @ziegler2020finetuning in the text prints the narrative form, Ziegler et al. (2020). The list stands where :::bibliography is, after section 7 and before the appendix, as in the paper. A :::pagebreak before it starts the appendix on a page of its own, as conference papers do; the four-row display of A.1 would otherwise leave five lines empty at the foot of the last page of references.
#4 · Put the title on a band, the source at the foot
const text = (id, content, family, size, color, placement, extra) => ({ kind: 'text', id,
content, fontFamily: family, fontSize: pt(size), color: col(color), align: 'left',
overflow: 'wrap', placement, ...extra });
const at = (to, edge, x, y, width) => ({ anchor: { to, edge }, offset: { x: mm(x), y: mm(y) },
...(width && { size: { width: mm(width), height: 'auto' } }) });
const caps = (size, fontWeight = 600) => ({ fontWeight, letterSpacing: pt(size * 0.18),
textTransform: 'uppercase' });
const BAND = 80; // mm from the trim's top to the foot of the band
const titleBlock = {
enabled: true,
minHeight: mm(BAND + 13 - TOP), // the band, then the byline: the abstract starts under it
slot: { elements: [
{ kind: 'box', id: 'band', style: { backgroundColor: col('accent') },
placement: { anchor: { to: 'bleed', edge: 'top-left' },
size: { width: 'fill', height: mm(BAND + 3) } } }, // + 3 mm of bleed
{ kind: 'image', id: 'curves', resourceId: 'sigmoids', // drawn in code: σ(βu)
placement: at('page', 'top-left', 70, 3, 108) },
text('kicker', '{attr.kicker}', SANS, 7.5, 'mist',
at('page', 'top-left', INNER, 30, 120), caps(7.5)),
text('title', '{titleText}', SERIF, 25, 'paper', at('#kicker', 'below', 0, 3.5, MEASURE),
{ fontWeight: 600, lineHeight: 1.08 }),
text('venue', '{attr.venue}', MONO, 7, 'mist',
at('page', 'top-left', INNER, BAND - 7, 120), { overflow: 'clip' }),
text('authors', '{attr.authors}', SANS, 9.5, 'ink', at('page', 'top-left', INNER, BAND + 6,
MEASURE), { fontWeight: 500, lineHeight: 1.45 }),
text('affiliations', '{attr.affiliations}', SANS, 7.5, 'muted',
at('#authors', 'below', 0, 1.4, MEASURE)),
] },
};
const head = (id, content, parity, edge, x, extra) => text(id, content, SANS, 7.5, 'muted',
at('page', `top-${edge}`, x, 13), { overflow: 'clip', ...caps(7.5, 500), align: edge,
parity, pages: 'body', ...extra });
const folio = { fontWeight: 700, color: col('accent') };
// The running heads are the paper's short forms: checks.contentInConfig allows them (#849).
const RUNNING = ['Rafailov, Sharma, Mitchell, Ermon, Manning & Finn',
'Direct Preference Optimization'];
const header = { elements: [
head('v-folio', '{pageNumber}', 'even', 'left', OUTER, folio),
head('v-title', RUNNING[0], 'even', 'left', OUTER + 9),
head('r-title', RUNNING[1], 'odd', 'right', -(OUTER + 9)),
head('r-folio', '{pageNumber}', 'odd', 'right', -OUTER, folio),
] };
const footer = { elements: [ // the first page: where the text comes from, and its licence
text('source', '{attr.source}', MONO, 6.2, // the H1's attribute, as kicker and venue are
'muted', at('page', 'bottom-left', INNER, -13, MEASURE - 10), { pages: 'opener' }),
text('drop-folio', '{pageNumber}', SANS, 7.5, 'accent', at('page', 'bottom-right', -OUTER, -13),
{ ...folio, align: 'right', pages: 'opener', overflow: 'clip' }),
] };
The title is the only level-1 heading, and its paper style draws a slot of design elements instead: the band bleeds off the top of the page, the title wraps at the measure, and the authors and affiliations come from heading attributes, with ¹ ² and * written as characters. The footer element with pages: 'opener' prints the source, the licence and the changes made on page 1 only, so the first page says where the text comes from even when it travels alone.
#5 · Redraw the figure in code
const R = (x) => Math.round(x * 100) / 100;
const svg = (w, h, body) => `<svg xmlns="http://www.w3.org/2000/svg" width="${w * PX_PER_MM}" `
+ `height="${h * PX_PER_MM}" viewBox="0 0 ${w} ${h}">${body}</svg>`;
function sigmoids() { // 108 × 24 mm over the kicker: σ(βu) for β from 0.25 to 4
let out = '';
[0.25, 0.4, 0.6, 1, 1.6, 2.5, 4].forEach((beta, i) => {
const pts = Array.from({ length: 105 }, (_, k) => {
const u = (k - 52) / 8; // u from −6.5 to 6.5
return `${k ? 'L' : 'M'}${R(2 + k)} ${R(22 - 20 / (1 + Math.exp(-beta * u)))}`;
}).join('');
out += `<path d="${pts}" fill="none" stroke="${palette.mist}" stroke-width="${R(0.3
+ i * 0.06)}" stroke-opacity="${R(0.25 + i * 0.1)}"/>`;
});
return svg(108, 24, out);
}
function tex(markup, x, y, size, color = 'ink', anchor = 0) { // maths as MathJax paths
const r = renderMath(markup, false, 100);
const k = size / 1000;
return `<g transform="translate(${R(x - anchor * r.viewBox.width * k)} ${R(y)}) scale(${k})" `
+ `fill="${palette[color]}">${r.paths.map((p) => `<path d="${p.d}"/>`).join('')}</g>`;
}
// Each label names its family and weight: loadSvg embeds the Work Sans files they set.
const words = (x, y, s, { size = 2.6, weight = 400, color = 'ink', anchor = 'start' } = {}) =>
`<text x="${R(x)}" y="${R(y)}" font-family="${SANS}" font-weight="${weight}" `
+ `font-size="${size}" fill="${palette[color]}" text-anchor="${anchor}">${s}</text>`;
const rect = (x, y, w, h, stroke, fill = 'paper', r = 1.2) => `<rect x="${x}" y="${y}" `
+ `width="${w}" height="${h}" rx="${r}" fill="${palette[fill]}" stroke="${palette[stroke]}" `
+ 'stroke-width="0.35"/>';
function arrow(points, color) { // a line with a drawn head, never an SVG marker
const [[x1, y1], [x2, y2]] = points.slice(-2);
const a = Math.atan2(y2 - y1, x2 - x1);
const head = [a - 0.45, a + 0.45].map((t) => `${R(x2 - 2 * Math.cos(t))} ${R(y2 - 2
* Math.sin(t))}`);
return `<path d="M${points.map(([x, y]) => `${R(x)} ${R(y)}`).join('L')}" fill="none" `
+ `stroke="${palette[color]}" stroke-width="0.45"/><path d="M${head[0]}L${R(x2)} ${R(y2)}`
+ `L${head[1]}Z" fill="${palette[color]}"/>`;
}
function preferences(x, y, w) { // a prompt and two answers, the preferred one first
const page = (dx) => rect(x + dx, y + 15.5, 6, 7.2, 'rule', 'paper', 0.6)
+ [0, 1, 2].map((l) => `<path d="M${x + dx + 1.2} ${y + 17.6 + l * 1.6}h3.6" `
+ `stroke="${palette.rule}" stroke-width="0.4"/>`).join('');
return rect(x, y, w, 25, 'rule', 'tint')
+ words(x + 2.4, y + 4.6, 'PREFERENCE DATA', { size: 2.1, weight: 600, color: 'muted' })
+ tex('x', x + 2.4, y + 9.8, 3) + words(x + 4.3, y + 9.8, ': “write me a poem about',
{ size: 2.25 }) + words(x + 5.2, y + 13, 'the history of jazz”', { size: 2.25 })
+ page(2.4) + tex('y_w', x + 9.4, y + 20.6, 3) + tex('\\succ', x + 14, y + 20.4, 3, 'accent')
+ page(18.4) + tex('y_l', x + 25.4, y + 20.6, 3);
}
function pipeline() { // 130 × 64 mm: RLHF on the left, DPO on the right
const title = (x, name, sub, color) => words(x, 4, name, { size: 3.2, weight: 600, color })
+ words(x, 8, sub, { size: 2.3, color: 'muted' });
const model = (x, y, w, label, color) => rect(x, y, w, 9, color)
+ words(x + w / 2, y + 5.8, label, { size: 2.6, weight: 600, color, anchor: 'middle' });
const note = (x, y, lines, color, anchor) => lines.map((line, i) => words(x, y + i * 3, line,
{ size: 2.3, color, anchor })).join('');
return svg(MEASURE, 64, title(0, 'RLHF', 'Reinforcement learning from human feedback', 'ochre')
+ preferences(0, 14, 35) // 35 mm: the prompt's first line ends 32.6 mm into the box
+ arrow([[36, 26.5], [48, 26.5]], 'ochre') + note(42, 22.4, ['maximum'], 'ochre', 'middle')
+ note(42, 31.4, ['likelihood'], 'ochre', 'middle')
+ model(49, 22, 25, 'reward model', 'ochre') + model(49, 50, 25, 'LM policy', 'ochre')
+ arrow([[55, 31.5], [55, 49]], 'ochre')
+ note(53.4, 41.6, ['label rewards,', 'reinforcement', 'learning'], 'ochre', 'end')
+ arrow([[68, 49], [68, 31.5]], 'ochre') + note(69.6, 39, ['sample', 'completions'], 'ochre')
+ `<path d="M88 2V62" stroke="${palette.rule}" stroke-width="0.3"/>`
+ title(93, 'DPO', 'Direct preference optimization', 'accent')
+ preferences(93, 14, 37)
+ arrow([[111.5, 39.5], [111.5, 49]], 'accent')
+ note(113.2, 43.4, ['maximum', 'likelihood'], 'accent', 'start')
+ model(99, 50, 25, 'final LM', 'accent'));
}
Figure 1 is drawn as SVG from the boxes and words of the original. Its labels name Work Sans in font-family, and loadSvg passes the markup through the engine's inlineSvgFonts, which embeds the two Work Sans files they use; tex() writes the y_w ≻ y_l labels as MathJax paths, the same glyphs the equations use. Arrowheads are drawn paths rather than SVG markers, so the figure stays vector in the PDF.
The whole recipe
// ═══ Postext Cookbook · Nº 138 · Machine-learning paper with theorems and proofs ═══
// https://postext.dev/en/cookbook/ml-paper-theorems-proofs
// Code: MIT · Text: Rafailov et al. 2023, arXiv:2305.18290 (CC BY 4.0), abridged · Art: code
// Fonts: Spectral, Work Sans, JetBrains Mono (SIL OFL 1.1) · Needs postext ≥ 1.65.0
// DPO (NeurIPS 2023) re-set as a preprint: numbered equations with labels and references,
// definition, lemma and theorem boxes, proofs that end in a square, author–year citations.
import {
buildDocumentWithFonts, renderPageToCanvas, registerResourceImage, registerCitationEngine,
defaultResourceTypes, initMathEngine, renderMath, inlineSvgFonts,
} from 'https://esm.sh/postext';
import { renderToPdf, decompressWoff2 } from 'https://esm.sh/postext-pdf';
import { createCiteprocEngine, STYLES, LOCALES } from 'https://esm.sh/postext-citeproc';
const LANG = 'en'; // @lang: the language of the sample document ('en' | 'es')
const RECIPE = 'ml-paper-theorems-proofs';
// ─── 1 · Design ─────────────────────────────────────────────────────────────
const palette = {
ink: '#1d1a22', accent: '#6b2a5f', // text; plum: the band, labels, theorem stripes
ochre: '#a8621c', tint: '#f4edf2', // the RLHF route in Figure 1; the theorem fill
rule: '#cfc3cc', muted: '#675f6b', // hairlines; running heads, notes
mist: '#e3c9dc', paper: '#ffffff', // type and curves on the band; white
};
const col = (id) => ({ hex: palette[id], model: 'hex', paletteId: id });
const colorPalette = Object.entries({ ...palette, 'main-color': palette.accent })
.map(([id, hex]) => ({ id, name: id, value: { hex, model: 'hex' } }));
const [SERIF, SANS, MONO] = ['Spectral', 'Work Sans', 'JetBrains Mono'];
const [BODY, LEAD] = [9.6, 13.4]; // pt: one column of 130 mm, about 74 characters a line
// mm: a 7 × 10 in trim, as technical books and many preprint series print
const [TRIM_W, TRIM_H, TOP, BOTTOM, INNER, OUTER] = [178, 254, 22, 22, 21, 27];
const MEASURE = TRIM_W - INNER - OUTER;
// #region answer: amsthm in Markdown: numbered equations, theorems and proofs
// \label{eq:x} in a display formula numbers it, on its row of an align; a box opened as
// :::callout{type="lemma" #lem:x} counts as a statement. \eqref{eq:x}, \ref{lem:x} and
// :ref{id="lem:x"} print the number and link to it.
const equationNumbering = { // (1) to (13): one sequence through the paper and its appendix
numberingTemplate: '{n}', resetOn: 'never', format: '({n})' };
// A proof: an italic label with no number, and a □ flush right on its last line, drawn
// from TeX's \square because maths runs (Spectral has no □ of its own).
const proofOf = (label) => ({ endMark: '□',
numbering: { label, counter: false, bold: false, italic: true } });
const statements = [ // a counter per kind, as the paper has it: Definition 1, Lemma 1, Theorem 1
{ id: 'definition', numbering: { label: 'Definition' } },
{ id: 'lemma', numbering: { label: 'Lemma' } }, // counter: 'theorem' would share one sequence
{ id: 'theorem', numbering: { label: 'Theorem' } },
{ id: 'proof', ...proofOf('Proof') },
{ id: 'sketch', ...proofOf('Proof Sketch') },
The composed script.js runs as it is: paste it into any page’s module script, or open the recipe on CodePen. Recipe folder on GitHub ↗ (opens in a new tab)
Variations
#Cite with numbers, as the NeurIPS template does
The original paper uses natbib's numbered citations; IEEE prints [1] in the text and numbers the list in order of citation.
-const citations = { style: 'harvard-cite-them-right', link: true,
+const citations = { style: 'ieee', link: true,#Count lemmas and theorems together
Some journals number every statement in one sequence (Lemma 1, Lemma 2, Theorem 3), as \newtheorem{lemma}[theorem] does. Point the lemmas at the theorem counter.
- { id: 'lemma', numbering: { label: 'Lemma' } }, // counter: 'theorem' would share one sequence
+ { id: 'lemma', numbering: { label: 'Lemma', counter: 'theorem' } },Pitfalls
Pitfall
Maths needs await initMathEngine() before the first build
Formulas laid out before the maths engine is running are grey placeholder boxes. Import initMathEngine from https://esm.sh/postext with the other symbols and await it before the first build; a document with maths built without it prints one console warning. Since postext 1.5 MathJax ships inside the package as one module that initMathEngine fetches, so the plain import works from a CDN and ?bundle is not needed (up to 1.4 every symbol had to come from https://esm.sh/postext?bundle). Import from one URL only: two URLs load two engines. Mathematics →
Pitfall
A bare $ opens maths: write \$
A dollar sign opens inline maths, so a price such as $40 starts a formula. Write \$40. Escapes and literal characters →
endMarkdraws its squares (□ ∎ ■) from TeX only when maths runs. In a document withoutinitMathEngine()the mark is set in the box's body face, and Spectral, like every Fontsource latin file, has no □: pick a mark the face can set there.- Column balancing fills a page that a tall display leaves short by adding grid lines above its headings.
balancing: { maxLinesPerHeading: 1 }keeps a paper's heads close to their own spacing; such a page may then end a few lines up. - A display with an
\underbracelabel wider than the brace, such as the gradient in section 4, needs analignedbreak; set wider than the measure, it overflows to the right.
Credits
- Recipe
- Ignacio Ferro
- Text
- Abridged from Rafailov, R., Sharma, A., Mitchell, E., Ermon, S., Manning, C. D. and Finn, C. (2023) “Direct Preference Optimization: Your Language Model is Secretly a Reward Model”, Advances in Neural Information Processing Systems 36 (NeurIPS 2023), arXiv:2305.18290v3. Sections 2 and 5.2, most of 6 and appendices A.2–A.4, A.6 and B–E cut, other paragraphs and sentences shortened; section numbers kept, equations renumbered in the abridged order; citations re-set author–year; Figure 1 redrawn · Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, Chelsea Finn · CC BY 4.0
- Fonts
- Spectral (SIL OFL 1.1) · Work Sans (SIL OFL 1.1) · JetBrains Mono (SIL OFL 1.1)
Edit this write-up ↗ (opens in a new tab)Recipe folder on GitHub ↗ (opens in a new tab)


