Chapter 14 — Further Reading

DataField.Dev

Chapter 14 — Further Reading

Annotated pointers for going deeper on the four fundamental subspaces, rank, and the rank-nullity theorem. The three "anchor" textbooks below are referenced throughout this book; we map each chapter to the relevant sections so you can read in parallel. Section numbers follow the most widely circulated editions and may shift slightly between printings. This is the most Strang-centric chapter of the book — the four-subspaces picture is his signature contribution, so his treatment is the primary recommendation.

The three anchor textbooks

Gilbert Strang, Introduction to Linear Algebra (5th ed.), §3.5 and §4.1 — and really all of Chapters 3–4. This is the source for this chapter, because the four-fundamental-subspaces framework is Strang's own. §3.5 ("Dimensions of the Four Subspaces") gives the dimension count $r,\ n-r,\ r,\ m-r$ and the big diagram exactly as we present it; §4.1 ("Orthogonality of the Four Subspaces") develops the right angles we only previewed in §14.10. Strang's "big picture of linear algebra" figure — the two boxes with the four spaces and the arrows between them — is worth taping to your wall. If you read one outside source for this chapter, read this; his MIT lectures (below) narrate the same diagram on the blackboard. Strang has called the four subspaces and their orthogonality the heart of the whole subject, and this chapter is built on that conviction.
Sheldon Axler, Linear Algebra Done Right (4th ed.), §3B (Null Spaces and Ranges) and the Fundamental Theorem of Linear Maps. Axler is the rigorous, proof-first complement, and he proves rank-nullity in its cleanest, coordinate-free form: for a linear map $T$, $\dim V = \dim(\operatorname{null} T) + \dim(\operatorname{range} T)$ — exactly the abstract version in our §14.9 sidebar, which he calls the Fundamental Theorem of Linear Maps. He deliberately avoids matrices and row reduction, so reading him alongside our pivot-counting proof shows you the same theorem from opposite ends. Math majors should read §3B in parallel with our §14.9, and note that Axler treats "rank" as a derived notion rather than a starting point — a bracing inversion of Strang's matrix-first view.
Stephen Boyd & Lieven Vandenberghe, Introduction to Applied Linear Algebra (VMLS), Chapters 5, 8, and 11. The applied, data-oriented view. Their treatment of independence, the matrix interpretations, and least squares connects directly to this chapter's regression case study; the idea that the column space is "the set of achievable outputs" and that residuals are orthogonal to it (our Case Study 1) is exactly their framing of least squares. Freely and legally downloadable as a PDF — see below. Best matched to the CS/data-science learning path.

Free online resources

MIT OpenCourseWare, 18.06 Linear Algebra (Gilbert Strang), Lectures 9–11. Strang's lectures on independence/basis/dimension (Lecture 9), the four fundamental subspaces (Lecture 10), and matrix spaces / rank-one matrices (Lecture 11). Lecture 10 is this chapter, delivered by the person who named the subspaces — watch it. Full video, transcripts, and problem sets, free. His later 18.065 and the short "big picture" videos restate the four-subspaces diagram with the orthogonality made explicit.
3Blue1Brown, Essence of Linear Algebra, "Inverse matrices, column space and null space" (Chapter 7 of the series). Grant Sanderson animates the column space and null space geometrically; while he does not draw all four subspaces, his visual of "rank = dimension of the output space the transformation lands on" and "null space = what gets squished to zero" is the perfect intuition pump for our §14.6 diagram. Watch it before re-reading the rank-nullity proof if the dimension count has not yet clicked.
Khan Academy, Linear Algebra, "Alternate coordinate systems" and "Matrix transformations" units. Gentler, exercise-rich coverage of column space, null space, rank, and the rank-nullity theorem (which Khan states as "rank + nullity = number of columns"), with immediate auto-graded practice. Good for shoring up the ⭐ and ⭐⭐ exercises before the harder tiers.
Boyd & Vandenberghe, VMLS free PDF and Python companion. The full textbook and its companion notebooks (with numpy-friendly examples) are posted by the authors at no cost. The companion shows rank, least squares, and the four-subspaces computations in code, reinforcing the C-track exercises and Case Study 1.

On the applications in this chapter

Degrees of freedom in regression (Case Study 1). For the linear-algebra core, see Strang §4.3 (least squares) and Chapter 17 of this book. Any mathematical-statistics text frames residual degrees of freedom as $n - p$; the geometric reading — that residuals lie in a subspace orthogonal to the column space of the design matrix — is exactly the four-subspaces picture, and is developed in regression-via-projection treatments. For the broader framing of degrees of freedom as a count of independent directions, follow the chapter's link on degrees of freedom.
Structural trusses, mechanisms, and self-stress (Case Study 2). The linear-algebra-of-structures viewpoint is the subject of C. R. Calladine's work extending Maxwell's rule (1864) into the modern statement that the numbers of mechanisms and states of self-stress are the dimensions of the left null space and null space of the equilibrium matrix [verify]. Pellegrino and Calladine's papers on the matrix analysis of statically and kinematically indeterminate frameworks are the classic references; any matrix-methods structural-analysis text covers the equilibrium matrix and its rank.
Incidence matrices and electrical networks (§14.6). Strang devotes a full section of Introduction to Linear Algebra to graphs and networks (often §10.1 or the "Applications" chapter), showing the four subspaces of an incidence matrix as potentials, currents, loops (the left null space), and Kirchhoff's laws. This is the canonical concrete example of the four subspaces and is worth reading alongside our §14.6 application.
Rank and low-rank approximation (the data-compression reading). The idea that rank measures information content and that low-rank matrices compress is developed fully in Chapters 30–32 of this book; for the broader data-science framing, follow the link on dimensionality reduction.

A note on where this is going

The four fundamental subspaces are the organizing diagram for the rest of the book, and this chapter's biggest unfinished promise is orthogonality. We checked numerically that the row space is perpendicular to the null space and the column space to the left null space, but we have not yet built the dot-product machinery that makes "perpendicular" rigorous and computational. That is the whole of Part IV: Strang's §4.1–§4.4 (orthogonality, projections, least squares) and Axler's Chapter 6 (inner-product spaces) are the parallel reading. The cleanest single statement to carry forward is Strang's: the input space splits as $\mathbb{R}^n = C(A^{\mathsf{T}}) \oplus N(A)$ and the output space as $\mathbb{R}^m = C(A) \oplus N(A^{\mathsf{T}})$, both orthogonal direct sums. Hold that picture; Part IV turns it into the most powerful computational tool you will learn.