Chapter 36 — Further Reading

The Jordan canonical form is where the "abstract algebra" flavor of linear algebra comes fully into view, so the proof-oriented texts (Axler, Hoffman–Kunze) are the natural companions, while the applied texts treat it more briefly and pivot quickly to the numerically stable substitutes. Read for the structure theorem and its proof; reserve the numerical caveat for the computing references.

Historical note. The form is named for Camille Jordan, who is generally credited with its introduction in his Traité des substitutions et des équations algébriques around 1870 [verify] (the precise priority is tangled — Weierstrass developed an equivalent theory of elementary divisors in the same era, and some sources attribute the canonical-form idea jointly). Treat the "Jordan 1870" attribution as the conventional shorthand, not a settled date.

Core textbooks

  • Sheldon Axler, Linear Algebra Done Right (4th ed.), Chapters 8 ("Operators on Complex Vector Spaces") and the generalized-eigenvector material. This is the primary companion to the chapter. Axler builds the Jordan form the way we sketched it in §36.6 — through generalized eigenspaces and nilpotent operators, without determinants — and his proof that every operator on a complex vector space has a basis of generalized eigenvectors is exactly the existence argument we deferred. His treatment of why the characteristic polynomial splits over $\mathbb{C}$ (the condition in §36.5) is unusually clean. If you read one source on this chapter, read this; §36.6's Math-Major Sidebar follows his approach directly.

  • Gilbert Strang, Introduction to Linear Algebra (5th/6th ed.) and Linear Algebra and Its Applications. Strang treats the Jordan form more briefly and pragmatically, with the emphasis the working scientist needs: the block structure, similarity, and — characteristically — an honest acknowledgment that it is "not for computing." His sections on diagonalization (our Chapter 25) and on $e^{At}$ for differential equations (our Chapter 37) bracket this material and show why the Jordan form sits between them. Best read as the applied counterpoint to Axler's structural treatment.

  • Hoffman & Kunze, Linear Algebra (2nd ed.), Chapters 6–7 ("Elementary Canonical Forms," "The Rational and Jordan Forms"). The rigorous classical reference, building the Jordan form from the primary decomposition theorem and the structure of modules over a PID (implicitly). This is the source for the rational canonical form mentioned in §36.6 — the form that always exists even over fields that are not algebraically closed — and for the full proof of the rank formula for block sizes. Graduate-level; read after Axler if you want every detail of the existence and uniqueness proofs.

On powers, the matrix exponential, and the bridge to Chapter 37

  • Boyd & Vandenberghe, Introduction to Applied Linear Algebra (VMLS). The most application-forward of the standard texts; light on the Jordan form itself but excellent on the linear dynamical systems $\mathbf{x}_{m+1} = A\mathbf{x}_m$ and $\mathbf{x}' = A\mathbf{x}$ that motivate it (the cohort model of Case Study 2 and the oscillator of Case Study 1). Freely available online. Use it to see why you want powers and exponentials of a matrix before Chapter 37 computes them.

  • Cleve Moler & Charles Van Loan, "Nineteen Dubious Ways to Compute the Exponential of a Matrix" (SIAM Review, 1978; updated 2003). The classic, readable paper on computing $e^{At}$ — and a perfect companion to §36.8. It explains in detail why methods based on the Jordan form (or on eigendecomposition of a near-defective matrix) are numerically treacherous, and what robust algorithms (scaling-and-squaring, Padé) do instead. The single best reference for the "honest note on numerical fragility."

On the numerical fragility (Case for Chapter 38)

  • Trefethen & Bau, Numerical Linear Algebra. The standard modern text on stable matrix computation. Its lectures on the Schur decomposition and on conditioning make precise the §36.8 claims: why the Schur form ($A = QTQ^*$) is the stable stand-in for the Jordan form, and why eigenvalue sensitivity explodes near a defective matrix. This is the natural next read after this chapter and the direct lead-in to our Chapter 38.

  • Golub & Van Loan, Matrix Computations (4th ed.). The encyclopedic reference. Its sections on the sensitivity of eigenvalues and on the (rarely used) numerical computation of the Jordan form document exactly the $\sqrt{\epsilon}$ perturbation sensitivity we demonstrated in code, and explain why production libraries omit the Jordan form entirely.

Free online resources

  • MIT OpenCourseWare 18.06 (Strang), lectures on diagonalization, similarity, and $e^{At}$. Video lectures that connect the Jordan form to the differential-equations payoff of Chapter 37; Strang's geometric framing of similar matrices as "the same transformation in different coordinates" is the Chapter 16 idea reused here.
  • 3Blue1Brown, Essence of Linear Algebra, the eigenvalue and change-of-basis episodes. For rebuilding the geometric intuition (eigenvectors as invariant directions, similarity as a change of viewpoint) that makes the shear picture of a Jordan block land.
  • sympy documentation, Matrix.jordan_form. The reference for the exact-arithmetic Jordan form we used to verify the worked examples; the docs note (as we did) that it is for symbolic/exact matrices, not floating-point ones.
  • The matrix exponential as the solution operator of a linear system is developed wherever differential equations are taught; see differential equations for the scalar $e^{at}$ that $e^{At}$ generalizes, and numerical methods for the stable computations that replace the Jordan form in practice.

Where to go next in this book

Chapter 37 is the immediate payoff: the matrix exponential $e^{At}$ solves $\mathbf{x}' = A\mathbf{x}$, and the $t\,e^{\lambda t}$ term derived in §36.7 is exactly what a defective matrix contributes — the degenerate node of a phase portrait, the critically-damped creep of Case Study 1. Chapter 38 then confronts the numerical fragility head-on with the language of condition number and stability, explaining why the Jordan form is a theorem to understand rather than a routine to call.