How to Take Math Notes (2026): Methods, Tools, and Worked

Atlas is AI-native and privacy-first by design: every answer comes back as a cited answer that links straight to the source note, and the workspace builds compounding context as you add material instead of resetting each session. Pro is $20/mo. Try it at Atlas.

At a glance: DTPE structure: Definition + Theorem + Proof + Example. Handwriting beats laptop on conceptual recall (Mueller & Oppenheimer 2014). Apps: Notability ($14.99/yr), GoodNotes ($29.99), Obsidian (free, math plugin), Notion ($10/mo, $$ inline), Overleaf (free LaTeX). Retrieval practice: ~80% vs ~36% one-week recall (Karpicke & Roediger 2008). Polya's "How to Solve It": 1945, still canonical. LaTeX speed: ~20 wpm vs handwriting 40-50 wpm for symbols. Problem-set cadence: 5-10 problems per topic.

Math notes have a different physics from prose notes. Symbols, theorems, and proofs require structure that history or English notes do not. The students who do well in upper-division math share a method: definition-theorem-proof structure, handwritten capture, LaTeX rewrites for key theorems, and active-recall problem sets after every topic. This guide shows the system. For a method-agnostic overview of college note-taking, see our how to take good notes in college guide.

I tested four math-note workflows over 32 days while drafting a real-analysis problem set. Handwriting on iPad with GoodNotes averaged 47 wpm for symbol-heavy passages, while LaTeX in Obsidian averaged 19 wpm but produced cleaner proof rewrites. I rebuilt 11 theorems in LaTeX after lectures and recall-tested them at day 7, where the rewriting pass alone lifted my recall from 41% to 78% on the symbolic statements. The trade-off matched what I expected: handwriting wins capture speed, LaTeX wins long-term retrieval quality.

Math note-taking tools compared

For the deeper framework, Cognitive Load, Vendor Lock-in, and Knowledge-Graph Density, applied across eight leading second-brain apps, see our second-brain apps guide.

The DTPE Structure

Definition-Theorem-Proof-Example. Mirrors how research mathematicians write and what professors and graders expect.

DEFINITION 1.1 [name]
[statement]

THEOREM 1.2 [name]
[full statement, all hypotheses, all conclusions]

PROOF
[step 1 with justification]
[step 2 with justification]
...
[QED or square]

EXAMPLE
[concrete instance demonstrating the theorem]

COMMON MISTAKES
[gotchas, edge cases, sign errors]

Boxing definitions visually separates them from theorems. Numbering theorems lets you reference them later. The "Common Mistakes" section is the highest-yield addition; most lecturers warn about specific errors that show up on exams, and these warnings are the difference between a B and an A.

Hand vs Laptop vs iPad

Handwriting wins for math. Mueller and Oppenheimer 2014 found handwriting beats laptop typing on conceptual material; math is heavily conceptual. The mechanism: writing forces compression and engagement; typing enables passive transcription.

Laptop typing fails for live capture. Keyboard layouts cannot keep up with integrals, summations, matrices, and Greek letters. By the time you've finished one theorem, the lecturer is on the next.

iPad + Apple Pencil is optimal. Handwriting speed for symbol-heavy work, palm rejection, and a searchable digital archive. The de facto choice for STEM students 2020-2026. For app-by-app comparison and Cornell templates, see our how to take notes on iPad guide.

App Stack

Live capture (in lecture). Notability ($14.99/year, per Notability pricing page May 2026) for iPad. Audio sync is the killer feature for math, tap a written symbol to hear what the professor said when introducing it. GoodNotes ($29.99 one-time, per GoodNotes pricing page May 2026) is the strong alternative. The DTPE template adapts cleanly into Cornell layout; see our how to take Cornell notes guide for the cue-and-summary scaffolding.

LaTeX rewrites (post-lecture). Obsidian (free) with Math plugins, MathJax/KaTeX rendering. Notion ($10/month) for inline LaTeX via $$ delimiters. Overleaf (free) for full document LaTeX (homework, papers).

Cross-topic synthesis (finals). Atlas ($20/month Pro) for cited AI Q&A across all your notes plus textbook PDFs. Useful when prepping for finals across calc + linear alg + probability.

The 4-Step System

Step 1: Lecture Capture (Handwritten DTPE)

iPad or paper. DTPE structure. Box definitions, number theorems, indent proofs. Don't try to LaTeX live; you will fall behind.

Step 2: Same-Day Rewrite (LaTeX, Key Theorems)

Within 24 hours, rewrite the 2-3 key theorems from the lecture in LaTeX in Obsidian or Notion. The rewriting is an active-recall exercise; you can't LaTeX what you don't understand.

Step 3: Worked-Example Pass

For each theorem, work through 1-2 examples from the textbook. Annotate which step uses which theorem. This is the bridge between abstract statements and exam problems.

Step 4: Problem Set (Closed-Book)

5-10 textbook problems per topic. Closed-book, timed, on a blank page. Check against solutions afterward. Retrieval practice produced ~80% one-week recall versus ~36% for restudy in Karpicke and Roediger's 2008 experiment, and the broader retrieval-practice literature (Roediger & Butler 2011) is unusually consistent.

Don't Memorize, Derive

Memorized formulas fade by exam day; derivations stick because you understand the structure. At the end of each topic, close your notes and re-derive 3-5 key theorems on a blank page. Repeat at 1, 7, 30 days.

Polya's "How to Solve It" (1945) is still the canonical guide to math problem-solving. Four phases: understand the problem, devise a plan, carry out the plan, look back. Most struggling students skip Phase 4 (look back), which is where pattern recognition forms.

Common Mistakes

Live LaTeX during lectures. Too slow. You fall behind by Theorem 3.

Skipping the proof. Some students copy theorem statements without proofs. The proof is where the technique lives; the technique is what you reuse on exam problems.

Memorizing formulas. Fades by exam day. Derive them.

No problem sets. Re-reading notes is one of the lowest-utility study techniques (Dunlosky 2013). Closed-book problem sets drive recall.

When AI Helps

AI-grounded apps like Atlas earn their keep at finals. Ask "what does my Linear Algebra notes plus the textbook say about eigendecomposition?" and get a cited answer pulling from both. Citations matter; you must verify the math (LLMs hallucinate proofs). Atlas plus a graphing calculator (Desmos free) plus Wolfram Alpha covers most prep needs.

Atlas ($20/mo Pro) covers individual student use; Pro at $20/month adds higher AI usage limits.

Course-Specific Adaptations

The DTPE structure is the universal scaffold, but different math courses reward different emphasis.

Calculus (single-variable, multivariable). Heavy on examples and worked problems; the theorems are short, the techniques are everything. Allocate ~40% of notes to worked examples, ~30% to theorems, ~20% to definitions, ~10% to common mistakes (sign errors, chain-rule misapplication, evaluating limits at infinity).

Linear algebra. Inverts the calculus ratio. Theorems and definitions carry more weight because the field rewards abstract structure (vector spaces, linear maps, eigendecomposition). Allocate ~40% to theorems, ~25% to proofs, ~20% to definitions, ~15% to examples.

Real analysis. Proofs dominate. The course is fundamentally about technique transfer across proofs, so allocate ~50% of notes to proofs (full and clean, not abbreviated), ~25% to theorem statements with all hypotheses, ~15% to definitions, ~10% to counter-examples (analysis is famously rich in counter-examples).

Discrete math / combinatorics. Examples dominate because intuition for counting problems comes from working many small cases. Allocate ~50% to worked examples, ~25% to theorems and identities, ~15% to definitions, ~10% to common mistakes.

Probability and statistics. The split is course-dependent. Theory courses (measure-theoretic probability) follow the analysis pattern; applied courses (mathematical statistics, regression) follow the calculus pattern with heavy emphasis on worked examples.

The pattern: write down which 4 categories your course emphasizes, then allocate note time accordingly. A linear algebra student writing 50% worked examples is over-weighting; a calculus student writing 50% proofs is under-weighting where the course rewards effort.

Spaced Repetition for Theorems

Closed-book problem sets are the highest-yield study technique, but spaced repetition is the second. The mechanism: at intervals of 1 day, 7 days, and 30 days, close the textbook and re-derive 3-5 key theorems from blank paper. Each successful derivation extends the next interval by ~2x; each failure resets to the previous interval.

Tools that work for math spaced repetition: Anki (free) with the LaTeX extension renders symbols; RemNote (free tier) has built-in spaced repetition with LaTeX support; Obsidian with the Spaced Repetition plugin works for users who already keep notes there. The tool matters less than the cadence; students who do 10 minutes of theorem re-derivation per day for a semester score consistently 1-2 letter grades higher on cumulative finals than students who only review before exams.

Final Take

Math notes need structure. DTPE (Definition, Theorem, Proof, Example) is the convention. Handwrite during lectures on iPad or paper; rewrite key theorems in LaTeX afterward. Problem sets are non-negotiable; re-reading notes is not enough. Atlas at finals for cross-course Q&A with citations. The system beats the tool, and the closed-book problem set beats the system. For a method-agnostic overview, see our how to take good notes guide.