Chapter 12 · Introduction

Chapter 12 — Canonical Forms — Diagonal, Triangular, Jordan

One operator, many matrices — which is the simplest? Diagonal when the minimal polynomial has distinct linear factors, triangular (even unitarily) whenever the characteristic polynomial splits, and in general the Jordan form, which classifies matrices up to similarity completely.

If $A=PDP^{-1}$ with $D$ diagonal, every computation with $A$ becomes a computation with numbers: $A^k=PD^kP^{-1}$, $f(A)=Pf(D)P^{-1}$, $e^{tA}=Pe^{tD}P^{-1}$, and the system $x'=Ax$ splits into independent equations $y_i'=\lambda_iy_i$. Engineers separate the vibration modes of a structure this way, and the long-run behaviour of a Markov chain is read off the same way. But not every matrix can be diagonalised. The shear $\begin{pmatrix}1&1\\0&1\end{pmatrix}$ has only one eigen-direction; differentiation on polynomials and every nonzero nilpotent matrix have too few eigenvectors; and the rotation $\begin{pmatrix}0&-1\\1&0\end{pmatrix}$ is diagonalisable over $\mathbb{C}$ but has no real eigenvector at all, so the answer can depend on the field.

This chapter finds the best possible matrix in every case. Chapter 11 produced the minimal polynomial; its factorisation decides everything. Distinct linear factors mean the matrix is diagonalisable; the primary decomposition splits the space into invariant pieces, one for each factor; whenever the characteristic polynomial splits the matrix is triangular in a suitable basis — even an orthonormal one, by Schur's theorem — and commuting matrices can be diagonalised or triangularised together. Refining the triangular form gives the Jordan canonical form: a block-diagonal matrix of Jordan blocks, read off from the ranks of the powers $(A-\lambda I)^k$ and unique up to the order of the blocks, so that classifying matrices up to similarity becomes counting partitions of integers. The rational canonical form does the same job over any field. For CSIR NET this is one of the most examined parts of linear algebra.

Diagonalisation→Minimal-polynomial test→Invariant subspaces→Primary decomposition→Triangular & Schur form→Commuting matrices→Generalised eigenvectors→Jordan form→Partitions & similarity classes→Rational canonical form
🎯 By the end of this chapter you can…
  • decide whether a matrix is diagonalisable over $\mathbb{R}$, $\mathbb{C}$ or a finite field — by multiplicities or by the minimal-polynomial test — and diagonalise it, including matrices known only through an identity such as $A^2=A$ or $A^k=I$;
  • use invariant subspaces, block matrices and the primary decomposition to split an operator into pieces with one eigenvalue each;
  • triangularise a matrix whose characteristic polynomial splits, apply Schur's unitary triangularisation, and diagonalise or triangularise commuting matrices simultaneously;
  • compute the Jordan form from the ranks of powers of $A-\lambda I$, find a Jordan basis, and use the form to compute $A^m$ and decide when $A^m\to0$;
  • count the similarity classes, or the possible Jordan forms, with a prescribed characteristic and minimal polynomial, and state the rational canonical form.
Teaser — Ranks of Powers Tell You the Jordan Blocks

Take an $n\times n$ matrix with the single eigenvalue $\lambda$ and choose how its Jordan form splits into blocks — a partition of $n$. On the left is the Jordan matrix $J$; top right is its dot diagram (row $i$ has $k_i$ dots for a block of size $k_i$); bottom right are the ranks of $(J-\lambda I)^k$. The drops between consecutive ranks are the column lengths of the dot diagram, so the ranks determine the blocks. Change $n$, pick a partition, or press Play to run through them all.

4
block sizes—
g(λ) = number of blocks—
minimal polynomial—
rank (J − λI)ᵏ, k = 1, 2, …—
diagonalisable?—

How this is examined

This chapter supplies many of the linear-algebra questions of a CSIR NET paper. In Part B a typical question asks whether a given matrix — a triangular matrix with repeated diagonal entries, a matrix with all entries equal, a permutation or rotation matrix, a matrix over a finite field, a matrix satisfying a polynomial identity — is diagonalisable; or it gives the characteristic polynomial together with the ranks of $A-\lambda I$ and $(A-\lambda I)^2$ and asks for the number or sizes of the Jordan blocks, the minimal polynomial, or the number of similarity classes of matrices with prescribed characteristic and minimal polynomials. Part C, where several options may be correct, prefers statements to be judged: whether $A^k=I$ forces diagonalisability over $\mathbb{C}$ or over $\mathbb{R}$, whether a nilpotent or idempotent matrix can be diagonalisable, whether two matrices with the same characteristic and minimal polynomials must be similar, whether commuting matrices share an eigenvector or can be diagonalised together, whether a matrix is similar to its transpose. The false options are the counterexamples collected in the ⚠️ boxes of §12.1 and §12.2, and almost every question needs the tools of Chapter 11 first. GATE MA asks the same material in single-answer and multiple-select form.

Sections in this chapter

✅ Before you start

Chapter 11 is essential: eigenvalues and eigenspaces, the characteristic and minimal polynomials, algebraic and geometric multiplicity, and the Cayley–Hamilton theorem (§11.1, introduced in Chapter 11). From Chapters 9–10 you need bases, dimension, direct sums and quotient spaces (§9.1); matrices of linear maps, change of basis and similarity, projections and nilpotent maps (§9.2); and determinants and rank (§10.1). From algebra you need the division algorithm, greatest common divisors and Bézout's identity $\gcd(f,g)=uf+vg$ (used in the primary decomposition). Unitary matrices and Gram–Schmidt are used once, in Schur's theorem, and are developed fully in §13.1.