How to Learn Mathematics
A complete mathematical education — proof and algebra through calculus, linear algebra, discrete math, probability, and modeling — runs about 1,075 hours. That is a multi-year project, and pretending otherwise is how people quit in month two. Most adults do not need all of it: they need one destination, and the honest version of this page tells you which branches you can leave standing. What stops nearly everyone who tries is not the difficulty of the top of the stack but an unrepaired layer near the bottom, so the first thing to do is find out where your foundation actually ends rather than where you assume it does.
Updated
Why Learn Mathematics?
It is the entry requirement for the fields that are currently expanding
Machine learning, quantitative finance, data science, statistics, engineering, and research all screen on mathematical maturity before they screen on anything else. You can learn the tools of those fields from documentation; you cannot learn to reason about why a model, a portfolio, or a study design is wrong without the mathematics underneath it.
The structures outlast every technology built on them
Frameworks and languages churn every few years. Groups, vector spaces, measures, and graphs do not — the linear algebra that powers today's models is the same linear algebra that powered signal processing decades ago. Time spent here has an unusually long half-life compared to anything else technical you could study.
It is the most reliable way to fix a broken relationship with your own competence
A large share of adults carry a specific memory of being told they were bad at math, and it quietly shapes what careers they consider. Repairing the actual gap tends to dissolve the belief, because the belief was always downstream of a curriculum that moved on before you were ready.
It gives you a defense against confident nonsense
Base rates, expected value, exponential growth, correlation versus causation, and what a confidence interval does and does not claim are the tools for evaluating a claim rather than a claimant. Probability and statistics in particular are where most public argument goes wrong, and where a modest amount of real training changes what you can see.
Your Learning Path
Find your real starting line and name one destination
~10 hoursTake cold diagnostics on arithmetic, algebra, and function notation, and write down the specific thing you want math for — machine learning, a degree prerequisite, quantitative finance, or curiosity. Without a destination you cannot tell what is optional, so everything feels mandatory and the project has no end. Without the diagnostic you will start one or two levels above where your foundation actually stops.
Build the language: algebra, trigonometry, logic, sets, and proof
~140 hoursSymbolic fluency to the point of automaticity, then the part school skipped — logical statements, quantifiers, sets, and the standard proof techniques including induction. This is the non-negotiable pillar regardless of destination, and the proof material is what converts you from someone who computes answers into someone who can evaluate a claim.
Work through the continuous branch: limits to differential equations
~280 hoursLimits, derivatives, integrals, the Fundamental Theorem, series, multivariable and vector calculus, then differential equations and the beginnings of analysis. This is the largest single pillar, and if your target is machine learning you can take a deliberately narrower slice of it — partial derivatives, the chain rule, and gradients — rather than the full sequence.
Take the structural branch: linear algebra and abstract algebra
~200 hoursVectors, matrices as transformations, rank, orthogonality, eigenvalues, and the singular value decomposition, then groups, rings, and fields. Start here rather than with calculus if your destination is data, AI, graphics, or cryptography — it needs less prerequisite repair and converts to working tools much sooner.
Take the discrete branch: counting, graphs, number theory, computation
~190 hoursCombinatorics, graph theory and algorithms, modular arithmetic and primes, automata, and computability. This is the pillar computer science runs on, and it is the one that most rewards the proof work from step two, since nearly every result here is an argument rather than a calculation.
Take the uncertain branch: probability, statistics, optimization, modeling
~265 hoursProbability, distributions, stochastic processes, statistical inference, optimization, game theory, and mathematical modeling. This is the branch with the widest everyday application and the one where trained intuition is most reliably wrong, so expect to spend real time unlearning rather than only accumulating. Practitioners in data and finance often need this one before the calculus pillar is finished.
Convert study into a permanent practice
~ongoingPick one field, read its actual papers or textbooks, and work the problems in them. Mathematics decays fast when unused and re-derives quickly when exercised, so the endgame is not finishing a syllabus but keeping a small amount of live contact with the material indefinitely. This is also where you find out which branch you want to go deep in.
Common Mistakes to Avoid
Starting at the level you believe you should be at
Take cold diagnostics before choosing a starting point, and begin one full level below your worst result. Adults consistently overestimate their retained algebra and then interpret the resulting confusion as difficulty with the new subject. The two hours a diagnostic costs routinely saves a hundred hours of studying the wrong thing.
Studying all five branches at once
Go depth-first: one branch at a time, in an order chosen by your destination. Breadth-first study across calculus, linear algebra, and probability simultaneously means nothing reaches the fluency threshold where it becomes useful, and the constant context switching prevents the consolidation these subjects specifically require.
Substituting visual intuition for the ability to do it
Excellent explainer videos build intuition, which is necessary and not sufficient. Cap passive consumption at roughly a quarter of your study time; the rest belongs to problems attempted cold with the solution covered. Being able to follow why a result is true is a different capacity from being able to produce it, and only the second one transfers.
Never reviewing, then discovering last quarter's material is gone
Schedule spaced retrieval: once a week, work three problems from material you finished a month or more ago, from memory. Mathematics unused decays quickly, but re-derivation is fast, so short deliberate review beats relearning. Without this, a multi-year arc turns into repeatedly rebuilding the same first year.
Treating the whole 1,075 hours as mandatory
Write your destination down, then cut. Machine learning needs the language pillar, linear algebra, probability, and a slice of multivariable calculus — closer to 450 hours than 1,075. Computer science needs the language pillar and the discrete branch. Only a mathematics degree or genuine polymath ambition requires all five, and confusing those cases is the main source of overwhelm.
Structured Roadmaps
Follow a guided learning path on Mochivia:
Frequently Asked Questions
Can I really learn math from scratch as an adult?
How long does it take to learn math?
Where should I start if I do not know where my gaps are?
What order should I learn math subjects in?
Do I need all of math, or only the parts for my goal?
Is it too late to switch into a quantitative career?
Learn Mathematics in short lessons
Interactive lessons, a few minutes each, in the Mochivia app for iPhone.
See the roadmap →