You can learn linear algebra alone from a textbook. Here is the checking stack that makes it work:
- Pick a book that ships worked solutions. Jim Hefferon’s Linear Algebra is free, written for a first course, and its solutions book “gives the answer to every one, completely worked.”
- Verify computations mechanically whenever the math allows it. Solved a linear system? Substitute the solution back in. Found an inverse? Multiply it out. Claimed an eigenvector? Apply the matrix. Linear algebra is unusually generous with self-checks — use every one.
- Write your work out in full, every step. An error cannot be found in work that was never written down, and “I can see it in my head” is where wrong answers hide.
- Get the work read, not the answer matched. Something — or someone — has to read what you actually wrote and point at the step where it went wrong.
- Come back to each topic on a schedule. Row reduction you can’t do anymore is row reduction you no longer know.
Step 4 is the one every study guide skips, and it is the entire subject of this post. Because the honest answer to “how do I check my own work?” is that for the interesting half of linear algebra — the proofs, the arguments, the why — you mostly can’t. Not alone. And the standard advice quietly assumes you aren’t alone.
Why checking the answer isn’t checking your work
Start with the good news: computational linear algebra is one of the most self-checkable subjects in mathematics. A candidate solution to a system either satisfies the equations or it doesn’t. That back-substitution habit will catch a large fraction of your errors, and you should build it early.
Now the trap. Suppose the check fails — your solution doesn’t satisfy the system. Somewhere in fifteen lines of row operations there is a sign error. Which line? The answer key can’t tell you: even a fully worked model solution shows one correct path, and yours may differ from it legitimately (there are many valid sequences of row operations) or differ from it at exactly the step you got wrong. Comparing the two, you are the grader of your own work — and the divergence you’re hunting for is precisely the one you couldn’t see when you wrote it. If you could see it, you wouldn’t have made it.
This is not a personal failing; it is the documented weak point of all self-study. When researchers surveyed 177 students on their real study habits, most reread and few self-tested, and the authors attributed it to “illusions of competence” — a distorted internal signal of what you actually know (Karpicke, Butler & Roediger, 2009). Grading your own proof relies on exactly the signal that study shows is broken. It’s the same illusion of competence that lets readers mistake familiarity for understanding, transposed into mathematics — where it doesn’t feel like vagueness, it feels like a finished answer that happens to be wrong.
What the standard advice actually offers
Search for how self-learners check linear algebra work and the most useful thing you will find is a writeup by Slava Akhmechet, who worked through Axler’s Linear Algebra Done Right alone, doing the assigned problems from two university syllabi — 197 problems assigned, 167 completed. His checking stack, in his own words: “I extensively used ChatGPT 4 to check my homework problems,” with prompts like “critique the following proof” — which “usually produce very good results,” though “one edge case where ChatGPT performed poorly was proofs by contradiction.” And when he needed a hint rather than a verdict, he “texted a friend with a math degree”: “I never found a prompt that would get ChatGPT to give good hints; my friend’s hints were always dramatically better.”
Read that list again as infrastructure. The state of the art for a solo learner is: a general-purpose chatbot with known failure modes, plus a personal friend with a mathematics degree. Both halves are doing real work — and one of them you cannot download.
The rest of the advice landscape doesn’t engage the question at all. The standard textbook roadmaps tell you what to read — Physics Forums recommends Friedberg, Insel and Spence, with a proof-writing book first if proofs are new to you — and say nothing about how you’d know your solutions are right. The encouragement posts say to find “someone who can answer your questions.” That someone is the thing the self-learner doesn’t have. It is the same hole we found in reading Kant without a professor: every resource explains the material or condenses it; almost nothing reads what you produced and tells you where it breaks. In philosophy the silent failure happens while you read. In mathematics it happens while you write.
How do you check a proof when studying alone?
Computations have back-substitution. Proofs have nothing like it. There is no mechanical test you can run on “Suppose v₁, …, vₙ are linearly dependent…” to see whether the argument holds — which is why proof-based courses run on graders and office hours, two more things the shelf can’t give you.
What you can do alone:
- Write the proof for a hostile reader. Every “clearly” and “it follows that” is a place to demand the justification from yourself. If you can’t cite the definition or theorem that licenses a step, the step is a guess.
- Attack your own proof. Does the argument use every hypothesis? A proof of a statement about finite-dimensional spaces that never uses finiteness is almost certainly wrong — or proving something false. Try the claim on a small concrete case and on the ugliest example you know.
- Let it cool. Rereading your own argument immediately, you reread your intention, not your text. A day later you can sometimes see what is actually on the page.
These habits raise your hit rate. None of them closes the loop, because they all still end with you grading you. The loop closes only when the proof gets read by something that didn’t write it — a friend with the degree, if you have one, or a tool that can genuinely read your argument against the book’s definitions rather than pattern-match your final line against a stored answer.
Checking your work with Study Junkie
This claim shouldn’t be taken on faith, so we recorded it: watch Study Junkie working on Hefferon’s Linear Algebra — the same free textbook from step 1. The book becomes a course, the course sets a linear-systems problem, and when the written answer comes back it finds the one sign error and shows where it happened. Not “incorrect, try again” — the clause that was wrong, and why. Then it schedules the topic back at one, three, and seven days.
In practice: you upload the PDF, DOCX or text file of the linear algebra book you’re actually using, and it becomes a chaptered course built from the full text — never a summary of it. Assignments make you write the work out; the feedback reads what you wrote. The tutor is grounded in your upload, so “what does this book mean by rank?” is answered from your book’s definitions and notation, not the internet’s. Scheduled recall brings chapter two back while you’re in chapter five, which is when it starts silently decaying. Signup comes with 1,500 free Study Credits and no card; after that it’s one-time packs from $10 that never expire — no subscription running while you spend three weeks on eigenvalues.
The honest limits. An AI grounded in a textbook is not a mathematician. On a subtle proof it can miss, and the right stance is the one Akhmechet took with ChatGPT: a first reader whose verdicts you check against the text, not an oracle — with the difference that grounding in your book means its answers come from the definitions you’re actually working with. A friend with a math degree, a study group, or a real course is better wherever you can get one; his friend’s hints beat every prompt he tried, and that result will generalize. And the free stack is genuinely complete for the computational half: Hefferon plus its worked solutions plus back-substitution costs nothing. Where the software earns its credits is the part of the subject where no mechanical check exists and the manual can only show you one path: reading your work. If linear algebra has defeated you before, it was probably there — upload your book and let something check the work this time.
FAQ
Which linear algebra textbook has solutions for self-study? Jim Hefferon’s Linear Algebra is the standout: free to download, aimed at a first undergraduate course with one semester of calculus as the stated prerequisite, used — in its own words — “by thousands of people for independent study,” with a solutions book that works every single exercise. Whatever book you choose, confirm the solutions situation before committing; a beautiful text with no accessible solutions is a much harder solo road.
Can I use ChatGPT to check my linear algebra proofs? Yes — one careful writeup of self-studying a proof-based linear algebra text did exactly that — Akhmechet reports “very good results” from prompts like “critique the following proof,” with proofs by contradiction as his reported weak spot, and hints as the place it clearly lost to a human. Two cautions: a general chatbot answers from the whole internet, so its terminology and conventions may not match your book’s, and its confidence doesn’t track its correctness — verify its verdicts against your text’s definitions.
Do I need calculus before linear algebra? Depends on the book. Hefferon states one semester of calculus as its prerequisite; proof-based texts like Friedberg or Axler need comfort with proofs more than calculus — Physics Forums’ roadmap recommends a proof-writing book first, such as Hammack’s free Book of Proof. The real prerequisite is the willingness to write your work out in full.
How long does it take to self-study linear algebra? One documented account on Medium covered the computational core in a month at 1–3 hours a day — roughly 15–25 hours a week. A proof-based text at a sustainable adult pace is a project measured in months, not weeks, and the schedule matters less than the ratio: time spent doing and checking problems should rival time spent reading. A chapter you read but couldn’t solve problems from is a chapter you haven’t learned yet.