From Similarity to Eigenvalues

I. The Core Philosophy: Operators vs. Matrices

To understand linear algebra deeply, we must distinguish between the Linear Operator (the “Soul”) and the Matrix (the “Body”).

• The Operator ($T$): An abstract, physical transformation of a vector space $V$. $T \in \mathcal{L}(V)$ It exists independently of any coordinate system (e.g., rotation, stretching, shearing).
• The Matrix ($A$): A numerical snapshot of operator $T$ observed from a specific basis (coordinate system).

II. Similarity: The Change of Perspective

Two matrices $A$ and $B$ are called similar ($A \sim B$) if they represent the same linear operator $T$ but viewed through different bases. Mathematically, this is expressed as:

$A = P B P^{-1}$

The “Translation” Mechanism

The conjugation $P B P^{-1}$ can be understood as a three-step process, often visualized via a commutative diagram:

1. $P^{-1}$ (Translate): Convert a vector from our standard basis into the “new” basis (where $B$ lives).
2. $B$ (Process): Apply the transformation in that new coordinate system.
3. $P$ (Translate Back): Convert the result back to our standard basis.

Key Insight: Since $A$ and $B$ are just different descriptions of the same underlying operator, they share invariant properties: $\det(A) = \det(B), \quad \operatorname{tr}(A) = \operatorname{tr}(B), \quad \operatorname{rank}(A) = \operatorname{rank}(B)$

III. Diagonalization: The Search for Simplicity

If similarity is about changing perspectives, diagonalization is the search for the perfect perspective. We seek a basis $\mathcal{B}$ in which the operator $T$ behaves in the simplest possible way: pure scaling along the axes.

• In this basis, the matrix representation $D$ is diagonal.
• The action of the operator becomes decoupled: dimensions do not interfere with one another.
• This is only possible if we can find a “Change of Basis” matrix $P$ such that: $P^{-1}AP = D = \operatorname{diag}(\lambda_1, \dots, \lambda_n)$

IV. Eigenvectors: The Perfect Basis

To achieve a diagonal matrix, our new basis vectors $\{v_1, v_2, \dots, v_n\}$ must satisfy a strict condition: The operator must not rotate or shear them; it must only stretch them.

If $v_i$ is a basis vector and the matrix is diagonal, then the operator’s action on $v_i$ must be: $T(v_i) = \lambda_i v_i$

This is exactly the definition of an Eigenvector.

The Geometric Intuition

• Eigenvectors ($v$) are the “preferred directions” of the operator—the axes of the universe that remain stable (invariant 1D subspaces) during the transformation.
• Eigenvalues ($\lambda$) are simply the scaling factors along those stable axes.

V. The Null Space Connection

To find these stable vectors, we solve the characteristic equation. But geometrically, why does the eigenvector $x$ reside in the null space of $(\lambda I - A)$?

We rewrite $Ax = \lambda x$ as: $(\lambda I - A)x = 0$

The “Cancellation of Forces”

Imagine two competing transformations acting on vector $x$:

1. $Ax$: The complex action of the matrix $A$ trying to transform $x$.
2. $\lambda I x$: A pure, uniform scaling action.

For $x$ to be an eigenvector, the action of $A$ must be identical to the action of the scalar $\lambda$. Therefore, the difference between them must be zero.

• The matrix $(\lambda I - A)$ measures the “deviation” between the operator $A$ and pure scaling.
• Vectors in the Null Space ($\mathcal{N}$) of this difference matrix have “zero deviation.” They are the directions where $A$ behaves exactly like a scalar multiplication.

VI. The Condition: Algebraic vs. Geometric Multiplicity

For an operator to be diagonalizable, we need enough eigenvectors to form a complete basis (a full coordinate system).

• Algebraic Multiplicity ($n_a$): The number of times a root $\lambda$ appears in the characteristic polynomial $\det(\lambda I - A) = 0$. This is the “promised” dimensional space.
• Geometric Multiplicity ($n_g$): The actual dimension of the null space $\dim \mathcal{N}(\lambda I - A)$. This is the “actual” number of independent directions found.

The Theorem

A matrix is diagonalizable if and only if Geometric Multiplicity = Algebraic Multiplicity for every eigenvalue ($n_g = n_a$). This state is often called “Semi-simple”.

$n_g < n_a \implies \text{Defective Matrix (Non-diagonalizable)}$

If $n_g < n_a$, the space “collapses,” and we cannot form a complete basis of eigenvectors to diagonalize the matrix.