Deep Connections: Invertibility, Null Space, Independence, Rank, and Pivots

linear algebra
MIT 18.06
matrix theory
fundamental concepts
Exploring how invertibility, null space, linear independence, rank, and pivots are all different perspectives on the same underlying mathematical structure
Author

Chao Ma

Published

October 18, 2024

Introduction

In linear algebra, concepts like invertibility, null space, linear independence, rank, and pivots often seem like separate topics. However, they are deeply interconnected - different views of the same mathematical reality. This post synthesizes these fundamental concepts and reveals how they all tell the same story about whether a linear transformation preserves information.

Invertibility: The Foundation

Reference: Lecture 3: Matrix Multiplication and Inverse

Definition

A square matrix \(A\) is invertible if there exists a matrix \(A^{-1}\) such that:

\[ AA^{-1} = A^{-1}A = I_n \]

Properties of Invertible Matrices

An invertible matrix \(A\) must satisfy:

  • Square: \(m = n\) (same number of rows and columns)
  • Full rank: \(\text{rank}(A) = n\)
  • All pivot variables: \(r = n\) pivot positions
  • No free variables: \(n - r = 0\) free variables
  • Independent rows: All rows are linearly independent

Solving Systems with Invertible Matrices

When \(A\) is invertible, solving \(Ax = b\) becomes straightforward:

\[ \begin{aligned} Ax &= b \\ A^{-1}Ax &= A^{-1}b \\ x &= A^{-1}b \end{aligned} \]

The solution is unique and exists for every \(b\).

Null Space: A Direct Test for Invertibility

Reference: Lecture 7: Solving Ax=0

The Connection

The null space \(N(A)\) provides a direct test for invertibility:

  • If \(N(A) = \{\mathbf{0}\}\): Matrix is invertible
  • If \(N(A) \neq \{\mathbf{0}\}\): Matrix is NOT invertible

Why Non-Trivial Null Space Prevents Invertibility

Proof by contradiction:

Suppose \(N(A) \neq \{\mathbf{0}\}\) and \(A^{-1}\) exists. Let \(v_1 \in N(A)\) with \(v_1 \neq \mathbf{0}\).

Then:

\[ \begin{aligned} Av_1 &= \mathbf{0} \quad \text{(by definition of null space)} \\ A^{-1}(Av_1) &= A^{-1}\mathbf{0} \quad \text{(multiply both sides by } A^{-1}\text{)} \\ (A^{-1}A)v_1 &= \mathbf{0} \\ I_n v_1 &= \mathbf{0} \\ v_1 &= \mathbf{0} \end{aligned} \]

Contradiction! We assumed \(v_1 \neq \mathbf{0}\), but our logic forces \(v_1 = \mathbf{0}\).

Therefore, \(A^{-1}\) cannot exist when \(N(A)\) contains non-zero vectors.

TipKey Insight

A non-trivial null space means information is lost in the transformation \(A\), making it impossible to uniquely reverse.

Linear Independence: The Geometric View

Reference: Lecture 8: Solving Ax=b

Connection to Invertibility

If the rows (or columns) of a matrix are not all linearly independent, the matrix cannot be invertible.

For a square matrix:

  • Independent rows/columns ⟺ \(\text{rank}(A) = n\)
  • Dependent rows/columns ⟺ \(\text{rank}(A) < n\)

Why Dependence Prevents Invertibility

When rows are dependent:

  • Rank: \(r < n\) (not full rank)
  • Pivots: Only \(r < n\) pivot positions
  • Free variables: \(n - r > 0\) free variables exist

Two Perspectives on Dependence

1. Algebraic Perspective

When free variables exist:

\[ R_{\text{rref}} = [I_r \mid F] \]

where \(F\) is the \((n-r)\)-dimensional free variable matrix.

The null space \(N(A)\) has dimension \(n - r > 0\), containing infinitely many vectors. By our earlier proof, this means \(A\) is not invertible.

2. Geometric Perspective

If rows are dependent, the transformation \(A\) collapses the \(n\)-dimensional space down to an \(r\)-dimensional subspace (where \(r < n\)).

  • Information loss: The transformation maps multiple distinct inputs to the same output
  • Cannot invert: We cannot uniquely recover the original \(n\)-dimensional vector from its \(r\)-dimensional image
  • Missing dimensions: The \((n-r)\) dimensions of information are permanently lost
NoteExample: Dimensional Collapse

A \(3 \times 3\) matrix with rank 2 maps all of \(\mathbb{R}^3\) onto a 2-dimensional plane. Infinitely many points in 3D space map to each point on the plane. There’s no way to uniquely invert this mapping.

The Big Picture: Everything is Connected

All these concepts are different views of the same mathematical reality:

Perspective Invertible (\(A^{-1}\) exists) Not Invertible (\(A^{-1}\) doesn’t exist)
Null Space \(N(A) = \{\mathbf{0}\}\) \(N(A) \neq \{\mathbf{0}\}\)
Rank \(\text{rank}(A) = n\) (full rank) \(\text{rank}(A) < n\) (rank deficient)
Pivots \(n\) pivots (all columns) \(r < n\) pivots
Free Variables 0 free variables \(n - r > 0\) free variables
Independence Rows/columns independent Rows/columns dependent
Dimension \(\dim(N(A)) = 0\) \(\dim(N(A)) = n - r > 0\)
Solutions to \(Ax = 0\) Only \(x = \mathbf{0}\) Infinitely many solutions
Determinant \(\det(A) \neq 0\) \(\det(A) = 0\)

Fundamental Theorem of Invertible Matrices

For a square \(n \times n\) matrix \(A\), the following are equivalent (all true or all false):

  1. \(A\) is invertible
  2. \(A^{-1}\) exists
  3. \(\text{rank}(A) = n\)
  4. \(N(A) = \{\mathbf{0}\}\)
  5. Columns of \(A\) are linearly independent
  6. Rows of \(A\) are linearly independent
  7. \(\det(A) \neq 0\)
  8. \(Ax = 0\) has only the trivial solution
  9. \(Ax = b\) has a unique solution for every \(b\)
  10. \(A\) has \(n\) pivot positions
ImportantThe Core Insight

All these conditions are testing whether the linear transformation preserves information. If any information is lost (through dimensional collapse, non-trivial null space, or linear dependence), the transformation cannot be inverted.