ProofComplete

Proof Sketch: Existence of Jordan Form

We outline the proof that every matrix over an algebraically closed field has a Jordan canonical form. The full proof requires careful induction and the primary decomposition theorem.

ProofExistence of Jordan Canonical Form (Sketch)

Theorem: Every n×nn \times n complex matrix AA is similar to a Jordan matrix.

Proof outline:

Step 1: Primary decomposition.

By the fundamental theorem of algebra, the characteristic polynomial factors completely: pA(λ)=(λλ1)a1(λλk)akp_A(\lambda) = (\lambda - \lambda_1)^{a_1} \cdots (\lambda - \lambda_k)^{a_k}

By the Primary Decomposition Theorem: Cn=Gλ1Gλk\mathbb{C}^n = G_{\lambda_1} \oplus \cdots \oplus G_{\lambda_k}

where Gλi=ker((AλiI)ai)G_{\lambda_i} = \ker((A - \lambda_i I)^{a_i}) has dimension aia_i.

Step 2: Reduce to single eigenvalue case.

Since each GλiG_{\lambda_i} is AA-invariant, we can choose a basis for Cn\mathbb{C}^n by combining bases for each GλiG_{\lambda_i}. In this basis, AA is block diagonal: A=[A1Ak]A = \begin{bmatrix} A_1 & & \\ & \ddots & \\ & & A_k \end{bmatrix}

where Ai=AGλiA_i = A|_{G_{\lambda_i}} acts on GλiG_{\lambda_i} and has only eigenvalue λi\lambda_i.

Step 3: Handle single eigenvalue case.

Focus on A1A_1 acting on Gλ1G_{\lambda_1} with only eigenvalue λ1\lambda_1. Write A1=λ1I+NA_1 = \lambda_1 I + N where N=A1λ1IN = A_1 - \lambda_1 I.

Since A1A_1 has only eigenvalue λ1\lambda_1, and (A1λ1I)a1=Na1=0(A_1 - \lambda_1 I)^{a_1} = N^{a_1} = 0 on Gλ1G_{\lambda_1}, the operator NN is nilpotent.

Step 4: Structure of nilpotent operators.

For nilpotent NN with Nm=0N^m = 0, we can find Jordan chains. Consider the filtration: {0}ker(N)ker(N2)ker(Nm)=Gλ1\{0\} \subset \ker(N) \subset \ker(N^2) \subset \cdots \subset \ker(N^m) = G_{\lambda_1}

Choose vectors to form Jordan chains by working backwards from higher kernels:

  • Pick vkker(Nk)ker(Nk1)\mathbf{v}_k \in \ker(N^k) \setminus \ker(N^{k-1})
  • Form chain: vk,Nvk,N2vk,,Nk1vk\mathbf{v}_k, N\mathbf{v}_k, N^2\mathbf{v}_k, \ldots, N^{k-1}\mathbf{v}_k

These chains are linearly independent and span Gλ1G_{\lambda_1}.

Step 5: Convert chains to Jordan blocks.

For Jordan chain {vk,Nvk,,Nk1vk}\{\mathbf{v}_k, N\mathbf{v}_k, \ldots, N^{k-1}\mathbf{v}_k\}, we have: A1(vj)=(λ1I+N)(vj)=λ1vj+NvjA_1(\mathbf{v}_j) = (\lambda_1 I + N)(\mathbf{v}_j) = \lambda_1\mathbf{v}_j + N\mathbf{v}_j

The action of A1A_1 on this chain is:

  • A1(Nk1vk)=λ1Nk1vkA_1(N^{k-1}\mathbf{v}_k) = \lambda_1 N^{k-1}\mathbf{v}_k (since Nkvk=0N^k\mathbf{v}_k = 0)
  • A1(Njvk)=λ1Njvk+Nj+1vkA_1(N^{j}\mathbf{v}_k) = \lambda_1 N^j\mathbf{v}_k + N^{j+1}\mathbf{v}_k

In the ordered basis (Nk1vk,,Nvk,vk)(N^{k-1}\mathbf{v}_k, \ldots, N\mathbf{v}_k, \mathbf{v}_k), this is exactly a Jordan block Jk(λ1)J_k(\lambda_1).

Step 6: Combine all blocks.

Performing this for each generalized eigenspace and each Jordan chain within, we obtain a basis where AA is in Jordan form. ∎

Remark

The proof's key ideas are: (1) decompose by eigenvalue using primary decomposition, (2) recognize that single-eigenvalue operators are nilpotent shifts, (3) construct Jordan chains systematically by analyzing kernel filtrations. The existence is guaranteed by working in algebraically closed fields where all eigenvalues exist. Over R\mathbb{R}, real matrices may not have Jordan form (e.g., rotation matrices with complex eigenvalues), requiring rational canonical form instead.