This sheet develops the rational canonical form and Jordan canonical form—the definitive structural classifications of linear transformations up to similarity.

Commentary. The characteristic polynomial tells us the eigenvalues; the minimal polynomial tells us the largest Jordan blocks. But to fully understand a linear transformation, we need to know the precise block structure. The rational canonical form works over any field, using invariant factors. The Jordan canonical form requires the characteristic polynomial to split completely (e.g., over C\mathbb{C}). Together, they provide a complete invariant for similarity of matrices.

Prerequisites. This sheet uses the following results from earlier sheets:

  • Eigenvalues, characteristic polynomial, minimal polynomial (Sheet 10)
  • PIDs, UFDs, polynomial rings (Sheet 9)
  • Cayley-Hamilton theorem (Sheet 10)
  • Linear transformations, matrices, vector spaces (Sheet 10)

Review: Similarity and Invariants#

Definition 11.1. Two matrices A,B∈Mn(F)A, B \in M_n(F) are similar, denoted A∼BA \sim B, if there exists an invertible matrix P∈Mn(F)P \in M_n(F) such that:B=P−1AP.B = P^{-1}AP.

Theorem 11.2. Similar matrices have the same:

  1. Characteristic polynomial
  2. Minimal polynomial
  3. Eigenvalues (with algebraic multiplicities)
  4. Rank
  5. Trace
  6. Determinant

Theorem 11.3. The converse is false: matrices with the same characteristic and minimal polynomials need not be similar.

Example 11.4. A=(2102)A = \begin{pmatrix} 2 & 1 \\ 0 & 2 \end{pmatrix} and B=(2002)B = \begin{pmatrix} 2 & 0 \\ 0 & 2 \end{pmatrix} have the same characteristic polynomial (t−2)2(t-2)^2 but different minimal polynomials: mA(t)=(t−2)2m_A(t) = (t-2)^2, mB(t)=t−2m_B(t) = t-2. They are not similar.

Exercise 11.5. Show that similar matrices have the same trace and determinant.

Exercise 11.6. Find two non-similar matrices with the same characteristic polynomial (t−1)2(t-1)^2.


Cyclic Subspaces and Cyclic Vectors#

Definition 11.7. Let T:V→VT: V \to V be a linear transformation and v∈Vv \in V. The cyclic subspace generated by vv is:Z(v;T)=span⁡{v,T(v),T2(v),…}.Z(v; T) = \operatorname{span}\{v, T(v), T^2(v), \ldots\}.

Definition 11.8. A vector vv is a cyclic vector for TT if Z(v;T)=VZ(v; T) = V.

Theorem 11.9. If vv is a cyclic vector for TT, then the set {v,T(v),…,Tn−1(v)}\{v, T(v), \ldots, T^{n-1}(v)\} is a basis for VV.

Theorem 11.10. If VV has a cyclic vector vv, then the matrix of TT with respect to the basis {v,T(v),…,Tn−1(v)}\{v, T(v), \ldots, T^{n-1}(v)\} is the companion matrix of the minimal polynomial of TT.

Example 11.11. Let T:R3→R3T: \mathbb{R}^3 \to \mathbb{R}^3 have matrix A=(00−210−301−4)A = \begin{pmatrix} 0 & 0 & -2 \\ 1 & 0 & -3 \\ 0 & 1 & -4 \end{pmatrix}. The vector e1e_1 is cyclic, and the minimal polynomial is t3+4t2+3t+2t^3 + 4t^2 + 3t + 2.

Exercise 11.12. Show that e1e_1 is a cyclic vector for the companion matrix C(p)C(p).

Exercise 11.13. Let TT be defined by T(x,y,z)=(y,z,0)T(x, y, z) = (y, z, 0). Find a cyclic vector for TT.

Exercise 11.14. Prove Theorem 11.9.


Companion Matrices#

Definition 11.15. The companion matrix of a monic polynomialp(t)=tn+an−1tn−1+⋯+a1t+a0p(t) = t^n + a_{n-1}t^{n-1} + \cdots + a_1 t + a_0is:C(p)=(00⋯0−a010⋯0−a101⋯0−a2⋮⋮⋱⋮⋮00⋯1−an−1).C(p) = \begin{pmatrix} 0 & 0 & \cdots & 0 & -a_0 \\ 1 & 0 & \cdots & 0 & -a_1 \\ 0 & 1 & \cdots & 0 & -a_2 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \cdots & 1 & -a_{n-1} \end{pmatrix}.

Theorem 11.16. The characteristic and minimal polynomials of the companion matrix C(p)C(p) are both equal to p(t)p(t).

Example 11.17. The companion matrix of p(t)=t2+bt+cp(t) = t^2 + bt + c is:C(p)=(0−c1−b).C(p) = \begin{pmatrix} 0 & -c \\ 1 & -b \end{pmatrix}.

Exercise 11.18. Write the companion matrix of p(t)=t3−2t2+3t−4p(t) = t^3 - 2t^2 + 3t - 4.

Exercise 11.19. Show that C(p)C(p) has cyclic vector e1=(1,0,…,0)Te_1 = (1, 0, \ldots, 0)^T.

Exercise 11.20. Prove Theorem 11.16.


Rational Canonical Form#

Definition 11.21. Let T:V→VT: V \to V be a linear transformation. A decomposition V=W1⊕⋯⊕WkV = W_1 \oplus \cdots \oplus W_k into TT-invariant subspaces is a cyclic decomposition if each WiW_i is cyclic (i.e., has a cyclic vector).

Theorem 11.22 (Cyclic Decomposition Theorem). Every linear transformation T:V→VT: V \to V on a finite-dimensional vector space over a field FF has a cyclic decomposition.

Theorem 11.23 (Rational Canonical Form). Every matrix A∈Mn(F)A \in M_n(F) is similar to a block diagonal matrix:R=(C(p1)C(p2)⋱C(pk))R = \begin{pmatrix} C(p_1) & & & \\ & C(p_2) & & \\ & & \ddots & \\ & & & C(p_k) \end{pmatrix}where each pi∈F[t]p_i \in F[t] is monic and p1∣p2∣⋯∣pkp_1 \mid p_2 \mid \cdots \mid p_k. The polynomials p1,…,pkp_1, \ldots, p_k are the invariant factors of AA, and they are uniquely determined.

Theorem 11.24. The last invariant factor pkp_k is the minimal polynomial of AA.

Theorem 11.25. The product of all invariant factors is the characteristic polynomial:χA(t)=p1(t)p2(t)⋯pk(t).\chi_A(t) = p_1(t) p_2(t) \cdots p_k(t).

Example 11.26. If AA has invariant factors t−1t-1, (t−1)(t−2)(t-1)(t-2), then χA(t)=(t−1)2(t−2)\chi_A(t) = (t-1)^2(t-2) and mA(t)=(t−1)(t−2)m_A(t) = (t-1)(t-2). The rational form is:([1]C((t−1)(t−2)))=(10000−2013).\begin{pmatrix} [1] & & \\ & C((t-1)(t-2)) & \end{pmatrix} = \begin{pmatrix} 1 & 0 & 0 \\ 0 & 0 & -2 \\ 0 & 1 & 3 \end{pmatrix}.

Exercise 11.27. Find the rational canonical form of a matrix with invariant factors t2+1t^2 + 1, t2+1t^2 + 1, t2+1t^2 + 1.

Exercise 11.28. Show that the rational canonical form of a diagonalizable matrix is diagonal.

Exercise 11.29. Show that the invariant factors determine the matrix up to similarity.


Jordan Canonical Form#

Definition 11.30. A Jordan block of size mm with eigenvalue λ\lambda is the m×mm \times m matrix:Jm(λ)=(λ10⋯00λ1⋯000λ⋯0⋮⋮⋮⋱1000⋯λ).J_m(\lambda) = \begin{pmatrix} \lambda & 1 & 0 & \cdots & 0 \\ 0 & \lambda & 1 & \cdots & 0 \\ 0 & 0 & \lambda & \cdots & 0 \\ \vdots & \vdots & \vdots & \ddots & 1 \\ 0 & 0 & 0 & \cdots & \lambda \end{pmatrix}.

Theorem 11.31 (Jordan Canonical Form). Let A∈Mn(F)A \in M_n(F) have characteristic polynomial that splits completely over FF (i.e., all eigenvalues lie in FF). Then AA is similar to a block diagonal matrix:J=(Jm1(λ1)Jm2(λ2)⋱Jmk(λk)).J = \begin{pmatrix} J_{m_1}(\lambda_1) & & & \\ & J_{m_2}(\lambda_2) & & \\ & & \ddots & \\ & & & J_{m_k}(\lambda_k) \end{pmatrix}.The Jordan blocks are uniquely determined up to order.

Theorem 11.32. The number of Jordan blocks for eigenvalue λ\lambda equals the geometric multiplicity of λ\lambda, i.e., dim⁡ker⁡(A−λI)\dim \ker(A - \lambda I).

Theorem 11.33. The size of the largest Jordan block for eigenvalue λ\lambda is the multiplicity of λ\lambda as a root of the minimal polynomial.

Theorem 11.34. The sum of the sizes of all Jordan blocks for eigenvalue λ\lambda equals the algebraic multiplicity of λ\lambda.

Example 11.35. A matrix with characteristic polynomial (t−2)4(t-2)^4 and minimal polynomial (t−2)2(t-2)^2 could have Jordan blocks: J2(2)⊕J2(2)J_2(2) \oplus J_2(2) or J2(2)⊕J1(2)⊕J1(2)J_2(2) \oplus J_1(2) \oplus J_1(2) or J3(2)⊕J1(2)J_3(2) \oplus J_1(2). Which one occurs depends on the geometric multiplicity.

Exercise 11.36. Find all possible Jordan canonical forms for a matrix with characteristic polynomial (t−1)3(t-1)^3 and minimal polynomial (t−1)2(t-1)^2.

Exercise 11.37. Let AA be a 4×44 \times 4 matrix with χA(t)=(t−2)4\chi_A(t) = (t-2)^4 and dim⁡ker⁡(A−2I)=2\dim \ker(A-2I) = 2. What are the possible Jordan forms?

Exercise 11.38. Prove Theorem 11.32.


Computing the Jordan Form#

Definition 11.39. For a matrix AA and eigenvalue λ\lambda, the generalized eigenspace is:Eλ(k)=ker⁡(A−λI)k.E_\lambda^{(k)} = \ker(A - \lambda I)^k.

Theorem 11.40. The number of Jordan blocks of size at least kk for eigenvalue λ\lambda is:dk=dim⁡ker⁡(A−λI)k−dim⁡ker⁡(A−λI)k−1.d_k = \dim \ker(A - \lambda I)^k - \dim \ker(A - \lambda I)^{k-1}.

Theorem 11.41. The number of Jordan blocks of size exactly kk for eigenvalue λ\lambda is:dk−dk+1.d_k - d_{k+1}.

Example 11.42. Suppose for λ=2\lambda = 2:

  • dim⁡ker⁡(A−2I)=2\dim \ker(A-2I) = 2
  • dim⁡ker⁡(A−2I)2=4\dim \ker(A-2I)^2 = 4
  • dim⁡ker⁡(A−2I)3=4\dim \ker(A-2I)^3 = 4

Then d1=2d_1 = 2, d2=2d_2 = 2, d3=0d_3 = 0. So there are 2 blocks of size 1 and 2 blocks of size 2.

Exercise 11.43. Let AA have dim⁡ker⁡(A−λI)=1\dim \ker(A-\lambda I) = 1, dim⁡ker⁡(A−λI)2=3\dim \ker(A-\lambda I)^2 = 3, dim⁡ker⁡(A−λI)3=3\dim \ker(A-\lambda I)^3 = 3. Determine the Jordan blocks for λ\lambda.

Exercise 11.44. Find the Jordan canonical form of:A=(2100020000310003).A = \begin{pmatrix} 2 & 1 & 0 & 0 \\ 0 & 2 & 0 & 0 \\ 0 & 0 & 3 & 1 \\ 0 & 0 & 0 & 3 \end{pmatrix}.


Applications of Jordan Form#

Theorem 11.45. The Jordan canonical form allows computation of functions of matrices. If A=PJP−1A = PJP^{-1}, then:f(A)=Pf(J)P−1,f(A) = P f(J) P^{-1},where f(J)f(J) is computed blockwise.

Theorem 11.46. For a Jordan block Jm(λ)J_m(\lambda):Jm(λ)k=(λk(k1)λk−1(k2)λk−2⋯(km−1)λk−m+10λk(k1)λk−1⋯(km−2)λk−m+2⋮⋮⋮⋱⋮000⋯λk).J_m(\lambda)^k = \begin{pmatrix} \lambda^k & \binom{k}{1}\lambda^{k-1} & \binom{k}{2}\lambda^{k-2} & \cdots & \binom{k}{m-1}\lambda^{k-m+1} \\ 0 & \lambda^k & \binom{k}{1}\lambda^{k-1} & \cdots & \binom{k}{m-2}\lambda^{k-m+2} \\ \vdots & \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & 0 & \cdots & \lambda^k \end{pmatrix}.

Example 11.47. The exponential of a Jordan block:eJm(λ)t=eλt(1tt2/2!⋯tm−1/(m−1)!01t⋯tm−2/(m−2)!⋮⋮⋮⋱⋮000⋯1).e^{J_m(\lambda)t} = e^{\lambda t} \begin{pmatrix} 1 & t & t^2/2! & \cdots & t^{m-1}/(m-1)! \\ 0 & 1 & t & \cdots & t^{m-2}/(m-2)! \\ \vdots & \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & 0 & \cdots & 1 \end{pmatrix}.

Exercise 11.48. Compute eAte^{At} for A=(2102)A = \begin{pmatrix} 2 & 1 \\ 0 & 2 \end{pmatrix}.

Exercise 11.49. Use the Jordan form to compute A10A^{10} for A=(1101)A = \begin{pmatrix} 1 & 1 \\ 0 & 1 \end{pmatrix}.


Relationship to Rational Canonical Form#

Theorem 11.50. If the characteristic polynomial splits completely over FF, the rational canonical form can be refined to the Jordan canonical form.

Theorem 11.51. The invariant factors determine the Jordan form and vice versa, when the field is algebraically closed.

Example 11.52. Over C\mathbb{C}, invariant factors (t−1)(t-1), (t−1)(t−2)(t-1)(t-2) correspond to Jordan form J1(1)⊕J1(2)J_1(1) \oplus J_1(2).

Exercise 11.53. Convert the invariant factors t2+1t^2 + 1, t2+1t^2 + 1 to Jordan form over C\mathbb{C}.

Exercise 11.54. Prove that two matrices are similar over C\mathbb{C} iff they have the same Jordan canonical form (up to block order).


The Minimal Polynomial and Jordan Blocks#

Theorem 11.55. The minimal polynomial of a Jordan block Jm(λ)J_m(\lambda) is (t−λ)m(t - \lambda)^m.

Theorem 11.56. For a matrix in Jordan canonical form, the minimal polynomial is:mA(t)=∏λ∈Spec⁡(A)(t−λ)mλ,m_A(t) = \prod_{\lambda \in \operatorname{Spec}(A)} (t - \lambda)^{m_\lambda},where mλm_\lambda is the size of the largest Jordan block for eigenvalue λ\lambda.

Theorem 11.57. The characteristic polynomial is:χA(t)=∏λ∈Spec⁡(A)(t−λ)aλ,\chi_A(t) = \prod_{\lambda \in \operatorname{Spec}(A)} (t - \lambda)^{a_\lambda},where aλa_\lambda is the algebraic multiplicity of λ\lambda (sum of sizes of all Jordan blocks for λ\lambda).

Exercise 11.58. Let AA have Jordan form J2(1)⊕J1(1)⊕J3(2)J_2(1) \oplus J_1(1) \oplus J_3(2). Find χA(t)\chi_A(t) and mA(t)m_A(t).

Exercise 11.59. Let AA have χA(t)=(t−1)4(t−2)3\chi_A(t) = (t-1)^4(t-2)^3 and mA(t)=(t−1)2(t−2)3m_A(t) = (t-1)^2(t-2)^3. What are the possible Jordan forms?


Summary#

We have introduced:

  • Similarity invariants: characteristic and minimal polynomials
  • Cyclic subspaces and cyclic vectors
  • Companion matrices
  • The Cyclic Decomposition Theorem
  • Rational Canonical Form via invariant factors
  • Jordan blocks and Jordan Canonical Form
  • Computing Jordan form using generalized eigenspaces
  • Applications: computing powers, exponentials, and functions of matrices
  • The relationship between rational and Jordan forms
  • The minimal polynomial of a Jordan block