What This Book Actually Is
Gilbert Strang's Introduction to Linear Algebra 5th Edition is the textbook that accompanies his MIT OpenCourseWare courses 18.06 and 18.065. It's not a theoretical math text. It's written for people who need to use linear algebra in engineering, data science, physics, or computer science, and who would rather understand what they're doing than prove theorems for their own sake. The 5th edition, published around 2016, added material on deep learning and the singular value decomposition as a central organizing idea rather than an endpoint. The book covers the standard syllabus: systems of equations, vector spaces, determinants, eigenvalues and eigenvectors, positive definite matrices, and the SVD. What makes it useful is that Strang rearranges the order of topics from what most textbooks do. He introduces linear transformations and matrix multiplication conceptually early on, before grinding through row reduction algorithms for forty pages. Most other books make you compute by hand for weeks before telling you what a matrix actually represents. I worked through this book while teaching myself numerical methods for fluid dynamics. The chapter on least squares and the pseudo-inverse saved me from reinventing a fitting routine from scratch. Strang derives the normal equations, explains why they can be numerically unstable, and then points you toward QR factorization. That sequence alone is worth the price of the book.
The companion lectures on MIT's website are free. They match the chapters fairly closely, though they lag behind the 5th edition in a few places because the lectures were originally recorded for earlier editions. Chapters 5 through 7 on eigenvalues move at a pace that feels natural if you work along with the video. Chapter 4 on determinants is the section I always skip. You need to know what a determinant is, and you need to know it's multiplicative and that it tells you whether a matrix is invertible. Strang covers this in about three lectures. After that you're done with it for almost every application.
How to Use It Without Wasting Time
Don't read it cover to cover like a novel. The exercises are where the actual learning happens, and they range from computational drills to conceptual questions that force you to think about what the math means. Do the early exercises in each section. If you can do them without looking at the solutions, move on. If you can't, re-read the section and watch the corresponding lecture again. The solution manual exists and is widely available. The 5th edition includes new exercises on machine learning applications. These are not trivial. They assume you already understand the core material from chapters 1 through 4. Don't jump into them expecting them to teach you linear algebra. They apply it. One thing I learned the hard way: do not skip the section on the four fundamental subspaces. Almost everyone skips it. It appears in chapter 3, right after you think you already understand what's going on. The column space, the null space, the row space, and the left null space. Getting these straight changes how you read every subsequent chapter. Eigenvalues, SVD, least squares—they all map cleanly onto these four spaces. I spent a week confused about why rank deficient matrices behaved badly until I actually drew the subspaces out on paper. It took maybe two hours once I realized I'd been treating matrices as arrays of numbers instead of geometric objects.
Get the Full Details
Common Pitfalls
The biggest mistake students make is treating matrix multiplication as a mechanical procedure. Strang insists repeatedly that A times x is a combination of the columns of A. This is not a cute phrasing trick. It is the single most important way to think about what a matrix does. If you compute products by rows instead, you will get the right answer, but you will miss the structure that makes everything else in the book make sense. Another issue: the book assumes you are comfortable with basic arithmetic and can handle algebra. It does not teach you what a function is. If your algebra is rusty, you will struggle with the later chapters regardless of how well you understand the linear algebra itself. There is also a computational trap. The book teaches you to derive formulas by hand. That is good for understanding. It is bad for actually running calculations on real data. The normal equations A transpose A x equals A transpose b look fine on paper. In practice, forming A transpose A squares the condition number of your matrix. If your original problem was ill-conditioned, squaring that condition number will make your solution garbage. I ran into this when fitting a model with nearly collinear predictors. The analytical solution gave me coefficients that were numerically correct but practically useless. Switching to a QR-based solver in numpy reduced the error from something on the order of 10 to minus 3 down to 10 to minus 10. Same problem. Same data. Different numerical path.
What It Does Not Cover Well
This is not a rigorous proof-based text. If you are a mathematics major who needs epsilon-delta arguments and formal vector space axioms proven from the ground up, this is the wrong book. Strang's approach is intuitive and computational. The proofs are sketches at best. For that, you want a text like Axler's Linear Algebra Done Right or Friedberg, Insel, and Spence. It also does not go deep enough into sparse matrix computation. If you are working with large-scale problems where sparsity matters, you will need supplementary material on iterative methods, preconditioners, and sparse factorization. The 5th edition mentions these topics but does not develop them. For that, look at Trefethen and Bau or Saad's iterative methods books. The coverage of complex vector spaces is minimal. If your work involves quantum mechanics or signal processing, you will need to supplement this with a text that treats complex inner product spaces properly.
Where to Find It
The official publisher is Wellesley-Cambridge Press. The MIT OpenCourseWare site hosts all the course materials for free, including lecture videos, notes, and problem sets. The textbook itself is a commercial product, though copies circulate widely. I am not going to link to an unofficial source. If you are on a tight budget, the OCW materials alone will get you most of the way there. The book is valuable mainly for the exercises and the reference sections. You can live without it if you are disciplined about doing the problem sets.
The Bottom Line
Strang's Introduction to Linear Algebra is the best bridge between computational practice and conceptual understanding that I have found. It will not make you a proof writer. It will not teach you sparse linear algebra or numerical linear algebra in sufficient depth for production work. But it will teach you to think about matrices the way people who actually use them think about them. That foundation makes everything else you learn later easier to absorb. The 5th edition tightens things up compared to earlier versions, particularly in how it positions the SVD as a unifying concept rather than a separate chapter tacked on at the end. If you are starting out, this is the book to begin with.
