Types of matrices

Shape-based types. A row matrix has order $1 \times n$ — a single row, e.g. $\begin{bmatrix} 2 & 5 & -1 \end{bmatrix}$. A column matrix has order $m \times 1$ — a single column, e.g. $\begin{bmatrix} 3 \\ 0 \\ 4 \end{bmatrix}$. A square matrix has order $n \times n$: equal rows and columns. These three are distinguished purely by shape, not by what the entries are.

Entry-pattern types (all built on top of "square"). A diagonal matrix is a square matrix whose off-diagonal entries are all zero:

where: $a_{ij}$ is the entry in row $i$, column $j$; the condition says nothing about the diagonal entries $a_{ii}$ themselves, which may be anything, including zero. It is written $\operatorname{diag}(d_1, d_2, \dots, d_n)$. A scalar matrix tightens this further: it is a diagonal matrix whose diagonal entries are all equal to one constant $k$:

where: $k$ is any fixed real number, the same one used at every diagonal position. An identity matrix $I$ is the special scalar matrix with $k = 1$:

A zero (null) matrix $O$ has every entry equal to $0$, and — unlike diagonal, scalar, and identity matrices — it need not be square; a zero matrix can have any order $m \times n$.

These containments only run one way: every identity matrix is scalar, every scalar matrix is diagonal, every diagonal matrix is square — but a diagonal matrix need not be scalar, and a scalar matrix need not be the identity. It helps to picture the chain as four boxes, each sitting strictly inside the one before it: square matrices are the outermost box (any equal number of rows and columns qualifies), diagonal matrices are a smaller box inside it (off-diagonal entries pinned to zero), scalar matrices a smaller box still (the surviving diagonal entries also forced equal), and the identity matrix a single point inside all of them (the equal diagonal value pinned to exactly $1$). A matrix can sit in the outer box without reaching any inner one — most square matrices are not diagonal at all — but nothing inside an inner box can escape the boxes surrounding it. A quick way to test membership without redrawing the diagram: given any matrix, first ask whether rows equal columns (square?); if yes, scan the off-diagonal positions for a nonzero entry (if none, diagonal); if diagonal, compare the diagonal entries to each other (all equal, scalar); if scalar, compare that common value to $1$ (identity only if it matches). Failing any one check simply stops the matrix at that box — it still legitimately belongs to every box already passed.

Understanding why these types matter beats memorising their definitions. The identity matrix behaves under matrix multiplication exactly the way $1$ behaves under ordinary multiplication: for any matrix $A$ compatible in size, $AI = IA = A$. That single fact is why $I$ shows up in the defining condition for an inverse, $AB = BA = I$, and why it is the natural "do nothing" matrix. The zero matrix plays the same role under addition that $0$ plays for numbers: $A + O = A$ for any $A$ of matching order, so it is the additive identity of matrix algebra. Diagonal and scalar matrices earn their special status computationally — when you multiply a diagonal matrix into another matrix, most of the row-by-column products collapse because one factor in each term is zero, leaving only a single surviving term per entry. That is precisely what makes diagonal and scalar matrices the fastest matrices to multiply by hand, and why textbook problems and exam questions lean on them for quick, checkable arithmetic.

Classify each matrix below with every applicable type, then build a related scalar matrix and confirm it is not the identity.

Step 1. Classify $A$. It is square (order $2 \times 2$), all off-diagonal entries are $0$ (diagonal), and its diagonal entries are equal (scalar, $k = 5$). It is not identity, since $5 \neq 1$.

Step 2. Classify $B$. Every entry is $0$, so $B$ is a zero matrix. Because it also happens to be $2 \times 2$ it technically satisfies the diagonal/scalar pattern with $k = 0$, but by convention a matrix of all zeros is simply called the zero matrix, not a scalar matrix.

Step 3. Classify $C$. It is square (order $3 \times 3$) with off-diagonal entries all $0$, so it is diagonal — but its diagonal entries ($3, -1, 7$) are not equal, so it is not scalar.

Step 4. Build a related $3 \times 3$ scalar matrix using diagonal constant $k = -4$:

Step 5. Check whether $S$ is the identity. The identity condition requires every diagonal entry to equal exactly $1$; here every diagonal entry is $-4 \neq 1$. So $S$ is a valid scalar matrix but not the identity matrix — identity is only the special case $k = 1$ of the scalar pattern.

Competitive stretch

If a scalar matrix $S = kI_n$ is multiplied by any square matrix $M$ of the same order, the result is simply $kM$ — every entry of $M$ scaled by $k$. This is because $S$ picks off exactly one term ($k \cdot m_{ij}$) from each row-by-column sum in the product, since all other entries in that row of $S$ are zero. It is the fastest way to verify a scalar-matrix multiplication without doing full row-by-column arithmetic.

• Calling every diagonal matrix a scalar matrix — scalar matrices additionally require the diagonal entries to be equal.
• Assuming a zero matrix must be square — it can be any order $m \times n$, e.g. a $2 \times 3$ zero matrix is perfectly valid, and it is not called diagonal or scalar even when it happens to be square.
• Thinking a diagonal matrix cannot contain a zero on the diagonal — it can; the definition only restricts the off-diagonal entries.
• Confusing "square matrix" (a shape condition: equal rows and columns) with "scalar matrix" (an entry-pattern condition) just because both are square-related — every scalar matrix is square, but very few square matrices are scalar.

Ask the tutor: If you keep mixing up scalar and diagonal matrices, or want the AI tutor to quiz you rapid-fire on classifying random matrices, just ask.