Linear Algebra: Dimensional Analysis and Unit Consistency

Abstract

This article examines how dimensional analysis—the practice of tracking the "shape" and compatibility of matrices—functions as a foundational principle in linear algebra. By treating matrix dimensions as constraints analogous to physical units, we clarify why certain operations succeed while others fail, and how this perspective unifies seemingly disparate topics including matrix multiplication, determinants, and diagonalization.

Background

Linear algebra is often taught as a collection of mechanical procedures: multiply matrices, compute determinants, find eigenvalues. Students learn how to perform these operations but may struggle to understand why certain combinations are valid and others are not. The answer lies in dimensional consistency—a principle borrowed from physics that applies equally to abstract matrix algebra.

In physics, dimensional analysis prevents nonsensical operations: you cannot add meters to seconds. Similarly, in linear algebra, the dimensions of matrices constrain which operations are permissible. A $3 \times 5$ matrix cannot be multiplied by a $3 \times 5$ matrix in that order; a $2 \times 2$ determinant is a scalar, not a matrix. These are not arbitrary rules but consequences of how linear transformations compose.

This article develops the thesis that understanding matrix dimensions as a form of dimensional analysis clarifies the structure of linear algebra and makes seemingly abstract concepts concrete.

Key Results

Matrix Multiplication and Dimensional Compatibility

The definition of matrix multiplication ^{[matrix-multiplication]} specifies that for matrices $A$ (size $m \times n$) and $B$ (size $n \times p$), the product $AB$ is defined as:

$$(AB)_{ij} = \sum_{k=1}^{n} A_{ik} B_{kj}$$

The critical observation is the dimensional requirement: the number of columns in $A$ must equal the number of rows in $B$. This is not a convenience; it is a necessity. The summation index $k$ ranges from $1$ to $n$, the shared dimension. Without this match, the sum is undefined.

This dimensional constraint mirrors physical unit analysis. Just as multiplying a distance (meters) by a time (seconds) requires a rate (meters per second) to make dimensional sense, composing linear transformations requires their intermediate dimensions to align. The result $AB$ has dimensions $m \times p$: the "input dimension" of $A$ and the "output dimension" of $B$.

The non-commutativity of matrix multiplication ^{[matrix-multiplication]} follows naturally from dimensional analysis. If $A$ is $m \times n$ and $B$ is $n \times p$, then $AB$ is defined and has shape $m \times p$. But $BA$ requires $B$ to have as many columns as $A$ has rows—that is, $p = m$. Even when both products exist, they have different shapes unless $m = p$, and even then they generally differ in value. Dimensional analysis explains why commutativity cannot be assumed.

Determinants and Dimensional Reduction

The determinant ^{[determinant-of-a-matrix]} is defined only for square matrices. This restriction is not arbitrary: the determinant measures how a linear transformation scales volumes in $n$-dimensional space. A non-square matrix maps between spaces of different dimensions and cannot have a well-defined volume scaling factor in the same sense.

For a $2 \times 2$ matrix $A = \begin{pmatrix} a & b \\ c & d \end{pmatrix}$, the determinant is:

$$\det(A) = ad - bc$$

This scalar encodes whether the transformation is invertible. A zero determinant indicates that the transformation collapses the 2D space into a lower dimension—the rows or columns are linearly dependent ^{[determinant-of-a-matrix]}.

The properties of determinants ^{[determinant-properties]} reinforce this dimensional perspective:

1. Swapping two rows multiplies the determinant by $-1$: the volume scaling factor changes sign.
2. Multiplying a row by scalar $k$ multiplies the determinant by $k$: scaling one dimension scales the volume by that factor.
3. Adding a multiple of one row to another does not change the determinant: this operation preserves volume.
4. $\det(A^T) = \det(A)$: the volume scaling is intrinsic to the transformation, independent of how we represent it.

These properties are not isolated facts but consequences of the determinant's role as a dimensional invariant.

Column Space, Null Space, and Rank

The column space of a matrix $A$ ^{[basis-of-column-space]} is the set of all vectors that can be expressed as $Ax$ for some input vector $x$. Its dimension equals the number of pivot columns in row echelon form—a quantity called the rank of $A$.

The null space ^{[basis-of-null-space]} is the set of all vectors $x$ satisfying $Ax = 0$. Its dimension equals the number of free variables in the solution.

These two spaces are complementary: they partition the input space. If $A$ is $m \times n$, then:

$$\text{rank}(A) + \text{dim}(\text{Null}(A)) = n$$

This is the rank-nullity theorem, a direct consequence of dimensional analysis. The $n$ input dimensions are partitioned into those that affect the output (rank) and those that vanish (null space dimension).

Eigenvalues and Diagonalization

Eigenvalues ^{[eigenvalues-of-a-matrix]} are found by solving:

$$\det(A - \lambda I) = 0$$

This equation is dimensional in a subtle way: we subtract $\lambda I$ (a scalar times the identity) from $A$ (a matrix). This operation is valid because $I$ has the same dimensions as $A$, and scalar multiplication preserves dimensions. The determinant of the resulting matrix is a polynomial in $\lambda$, whose roots are the eigenvalues.

A matrix is diagonalizable ^{[diagonalizable-matrix]} if it can be written as:

$$A = PDP^{-1}$$

where $D$ is diagonal and $P$ is invertible. The columns of $P$ are eigenvectors. This decomposition is possible if and only if the eigenvectors form a basis—that is, if there are enough linearly independent eigenvectors to span the entire space. Dimensionally, $P$ must be $n \times n$ (to have $n$ columns) and invertible (so $P^{-1}$ exists and has the same dimensions).

Diagonalization simplifies computation because diagonal matrices are easy to manipulate. Raising $A$ to a power becomes:

$$A^k = PD^kP^{-1}$$

and $D^k$ is simply the diagonal matrix with each diagonal entry raised to the $k$-th power. This is a dimensional argument: the structure of the decomposition guarantees that the dimensions work out at every step.

Worked Example

Consider the matrix equation ^{[matrix-equation-solution]}:

$$B^{-1}(A - X) = AX$$

where $A$, $B$, and $BA + I$ are invertible $n \times n$ matrices. We wish to solve for $X$.

Step 1: Dimensional check. All matrices are $n \times n$, so all operations are dimensionally valid.

Step 2: Expand and rearrange.

$$B^{-1}A - B^{-1}X = AX$$ $$B^{-1}A = AX + B^{-1}X$$ $$B^{-1}A = (A + B^{-1})X$$

Step 3: Dimensional reasoning. We need to isolate $X$. The term $A + B^{-1}$ is $n \times n$, so if it is invertible, we can multiply both sides on the right by its inverse.

Step 4: Manipulate to standard form.

$$B^{-1}A = (A + B^{-1})X$$

Multiply both sides on the left by $B$:

$$A = B(A + B^{-1})X = (BA + I)X$$

Step 5: Solve for $X$.

$$X = (BA + I)^{-1}A$$

The solution is valid provided $BA + I$ is invertible. Dimensionally, $(BA + I)$ is $n \times n$ (since $BA$ is $n \times n$ and $I$ is $n \times n$), and $(BA + I)^{-1}$ has the same dimensions. Multiplying by $A$ (which is $n \times n$) yields an $n \times n$ result, consistent with $X$.

This example illustrates how dimensional analysis guides algebraic manipulation: at each step, we verify that operations are dimensionally valid, and this verification ensures the solution is correct.

References

• ^{[matrix-multiplication]}
• ^{[determinant-of-a-matrix]}
• ^{[determinant-properties]}
• ^{[determinant-properties]}
• ^{[determinant-of-a-matrix]}
• ^{[basis-of-column-space]}
• ^{[basis-of-null-space]}
• ^{[eigenvalues-of-a-matrix]}
• ^{[diagonalizable-matrix]}
• ^{[diagonalizable-matrix]}
• ^{[matrix-equation-solution]}
• ^{[matrix-inversion-formula]}

AI Disclosure

This article was drafted with the assistance of an AI language model based on personal class notes (Zettelkasten). The mathematical claims and worked examples are derived from the cited notes and verified for technical accuracy. The framing of dimensional analysis as a unifying principle is an original synthesis intended to clarify connections between standard linear algebra topics. All factual claims are attributed to source notes via citation.