Chapter 9 — Further Reading

Annotated pointers for going deeper on the inverse matrix, invertibility, and why you rarely compute it. Each entry says what to read and why it complements this chapter. If you watch only one thing, watch the 3Blue1Brown inverse video — it animates the "undo" picture this chapter is built on.

Watch first (the geometry, animated)

• 3Blue1Brown, Essence of Linear Algebra, Chapter 6/7: "Inverse matrices, column space and null space" (free, YouTube). Grant Sanderson animates exactly the picture of §9.1–9.4: $A^{-1}$ as the transformation that plays $A$ in reverse, and a singular matrix as one that squashes space so that no reverse exists. His framing of "when does a solution exist?" in terms of dimension lost is the visual companion to our Invertible Matrix Theorem. Watch it alongside the chapter, not after.

Core textbooks (the standard references for this book)

• Gilbert Strang, Introduction to Linear Algebra (5th/6th ed.), §2.5 ("Inverse Matrices") and §2.6 ("Elimination = Factorization: A = LU"). Strang's §2.5 is the closest match to this chapter: the Gauss-Jordan method on $[A\,|\,I]$, the conditions for invertibility, and the order-reversal rule $(AB)^{-1} = B^{-1}A^{-1}$, all in his geometry-forward voice. His matching MIT 18.06 video Lecture 3 ("Multiplication and inverse matrices") covers the same ground. Crucially, Strang is emphatic about not inverting to solve — §2.6 immediately pivots to LU (our Chapter 10), making the §9.8 lesson concrete. Best for: the algorithm, the invertibility conditions, and the "factor, don't invert" philosophy.

• Sheldon Axler, Linear Algebra Done Right (4th ed.), §3.D ("Invertibility and Isomorphic Vector Spaces"). The abstract counterpart to our §9.6–9.7. Axler defines invertibility for linear maps and proves the equivalence of injective, surjective, and invertible for maps between finite-dimensional spaces of equal dimension — exactly the square-matrix miracle of §9.7, done at the level of operators. Best for: math majors who want the operator-theoretic proof that one-sided inverses are two-sided; pairs with our A-track sidebars. (Axler famously delays determinants, so the "$\det \neq 0$" face comes from Strang.)

• Stephen Boyd & Lieven Vandenberghe, Introduction to Applied Linear Algebra (VMLS) (free PDF), Chapter 11 ("Matrix inverses"). The applied/data-science angle: inverses, left and right inverses, pseudoinverses, and — importantly — a clear-eyed discussion of when you actually need an inverse versus when you should solve. Best for: CS/data-science readers; its left/right-inverse treatment is the perfect follow-up to our Math-Major Sidebar, and it leads naturally into least squares (our Chapter 19).

On specific topics in this chapter

• Why you should not invert to solve (§9.8). The canonical reference is Nicholas Higham, Accuracy and Stability of Numerical Algorithms (2nd ed.), whose discussion of "the dangers of matrix inversion" gives the rigorous numerical-analysis backing for our maxim; see also Cleve Moler's Numerical Computing with MATLAB, freely available, for an accessible version. The slogan "never invert a matrix to solve a system" traces to numerical-analysis folklore around Forsythe and Moler [verify]. We make the efficiency side rigorous in Chapter 10 (LU) and the accuracy side in Chapter 38 (conditioning).

• The Invertible Matrix Theorem (§9.6). David C. Lay, Linear Algebra and Its Applications, §2.3 states and proves the theorem as an explicit list of equivalent conditions (Lay's is the most cited textbook formulation, and our list follows it closely); a fine reference for seeing all the equivalences chained together with proofs.

• The condition number (the §9.3 Warning). Deferred to our Chapter 38; for a preview, Strang §9.2 and Boyd–Vandenberghe's chapters on least squares discuss why nearness-to-singular is measured by conditioning, not by the determinant's magnitude.

For the applications in this chapter

• Cryptography (Case Study 1): the Hill cipher is treated in most introductory cryptography texts; see Trappe & Washington, Introduction to Cryptography with Coding Theory for the matrix-over-$\mathbb{Z}_{26}$ development, including modular matrix inverses and the known-plaintext attack that breaks it. For the modular-arithmetic background (modular inverses, coprimality), any discrete-math text — including this series' modular arithmetic and number theory material — covers why $\gcd(\det K, 26) = 1$ is the invertibility condition.

• Economics (Case Study 2): Wassily Leontief's original input-output analysis is surveyed in any intermediate macroeconomics or mathematical-economics text; for the linear-algebra-first treatment, Lay §2.6 ("The Leontief Input-Output Model") works the productive-economy condition and the $(I-C)^{-1} = I + C + C^2 + \cdots$ series in exactly our notation. The connection between invertibility of $I - C$ and the spectral radius of $C$ is a preview of our Chapter 23 and the Perron–Frobenius theory behind it.

• Graphics / color (the §9.5 application): color-space conversions as invertible $3\times 3$ matrices appear in any computer-graphics or image-processing text (e.g. Fundamentals of Computer Graphics, Marschner & Shirley); the inverse-transform-for-screen-picking idea is developed in the transformations in video game design material referenced in the chapter.

Free, interactive, and visual

• MIT 18.06 (Strang), OpenCourseWare — free lecture videos, problem sets, and exams; Lecture 3 covers inverses and Lecture 4 covers $A = LU$, mirroring Chapters 9 and 10.
• The recurring toolkit/visualizer.py in this book's repository — run it on a matrix and its np.linalg.inv, then on their product, to watch $A^{-1}A$ restore the unit square (Figure 9.1). Then try a singular matrix and watch the square collapse with no way back (Figure 9.2). Type a matrix and look.

Where to go next in this book

Chapter 10 (LU and PLU decomposition — the efficient way to solve systems, and the real reason §9.8 says "factor, don't invert"), then Chapter 11 (the determinant as signed volume scaling, making "$\det = 0 \iff$ singular" rigorous), and Chapter 13 (the null space — the vectors a singular matrix kills — placed inside the four fundamental subspaces).