How Many Physics Numericals You Actually Need to Solve Before JEE Main
"Solve as many numericals as you can" is technically true and practically useless as advice. It gives no way to know if you've done enough for a given chapter, or if you're overinvesting in one topic while under-preparing another. A more useful approach starts from what JEE Main actually tests, and works backward.
What the exam actually asks for, per chapter
JEE Main Physics carries 25 questions (20 MCQs and 5 numerical-value questions, all 25 compulsory under the current pattern), drawn from the full Class 11 and 12 Physics syllabus, roughly 20 chapters depending on how they're grouped. That means most individual chapters contribute about 1 or 2 questions to any single paper. This has a direct implication: for a chapter that typically yields 1 question, grinding through hundreds of numericals on it has sharply diminishing returns past a certain point, compared to spreading that same time across a chapter that consistently yields more.
Why numerical questions specifically deserve fresh attention
Numerical-value questions used to be the safer bet: when they were introduced in JEE Main 2020 they carried no negative marking, so an educated guess cost nothing. That changed in 2022, when NTA clarified in a March 2022 notice that wrong numerical answers would cost 1 mark, and it has stayed that way since. Numerical questions now carry the same -1 penalty as MCQs, and since 2025 all five per subject are compulsory, after NTA's notice of 17 October 2024 removed the old "any 5 of 10" choice. The current year's information bulletin on jeemain.nta.nic.in is the place to confirm the marking scheme for your cycle. A blind guess on a numerical, where you're not choosing from 4 options but producing an exact value from scratch, has a far lower chance of landing right than a blind MCQ guess, and now carries the same downside. That raises the bar specifically for numerical practice: the goal isn't just recognizing the method, it's finishing the calculation cleanly enough that you'd actually attempt it under the current stakes.
A volume target that scales with a chapter's actual weight
Instead of a flat number of problems per chapter, size your numerical practice to the chapter's typical question share. The table below comes from an analysis of 118 JEE Main papers from 2020 to 2025 (published by Rhovecs), using NTA's own syllabus units. Treat it as an average, not a forecast for any single paper:
| Chapter | Share of Physics questions | Typical questions per paper |
|---|---|---|
| Electrostatics | 7.9% | about 2 |
| Optics | 7.4% | about 2 |
| Properties of Solids and Liquids | 7.3% | about 2 |
| Current Electricity | 7.1% | about 2 |
| Electromagnetic Induction and AC | 6.5% | about 2 |
| Magnetic Effects of Current and Magnetism | 6.0% | about 2 |
| Units and Measurements | 6.0% | about 2 |
| Atoms and Nuclei | 5.4% | about 2 |
| Kinematics | 5.4% | about 2 |
| Laws of Motion | 4.9% | about 1 |
| Rotational Motion | 4.9% | about 1 |
| Oscillations and Waves | 4.7% | about 1 |
| Gravitation | 4.3% | about 1 |
| Dual Nature of Matter and Radiation | 4.0% | about 1 |
| Work, Energy and Power | 3.9% | about 1 |
| Electronic Devices | 3.8% | about 1 |
| Thermodynamics | 3.6% | about 1 |
| Kinetic Theory of Gases | 3.5% | about 1 |
| Electromagnetic Waves | 3.4% | about 1 |
- About 2 questions per paper (the top nine rows, from Electrostatics to Kinematics): these deserve the deepest problem sets, enough to have seen most of the standard variations NTA tends to reuse across sessions
- About 1 question per paper (the other ten chapters): a solid but smaller set, enough to be comfortable, not exhaustive
- These are averages. Any single paper can put 3 questions on one chapter and skip another entirely, so a light pass on every chapter still beats dropping one
The real signal to track: are you still learning something new?
A better stopping point than any fixed number: keep solving a chapter's numericals until new problems stop teaching you anything new about that chapter's mechanics. You start recognizing the setup within the first line and solving on autopilot. Once you hit that point, more volume on that specific chapter is mostly just repetition, not additional preparation, and your time is better spent elsewhere.
The mistake this framework catches
The most common numerical-practice mistake isn't solving too few problems overall. It's an uneven distribution: hundreds of problems on a comfortable, high-confidence chapter (because it feels productive) and almost none on a chapter you're avoiding because it's harder, even though that harder chapter carries equal or greater weight. If you've never actually counted how many numericals you've done per chapter, there's a good chance the distribution is more lopsided than it feels.
Timing yourself per numerical while you practice, not just per full mock, is what actually reveals which chapters you've truly mastered versus which ones you're still solving slowly even when you get them right. Studyloaf's free timer works without an account and lets you tag each session by subject, so a pattern like "still slow on Modern Physics despite high accuracy" shows up in your own data instead of staying a vague impression.