Transpose of Matrix

Take a railway timetable. Suppose three trains — Rajdhani, Shatabdi, and Duronto — run between two cities, and you record their departure times in a table:

This is a 3 \times 2 matrix — three rows (trains), two columns (cities). Now suppose someone asks you to reorganise the same data so that the cities are the rows and the trains are the columns:

No data was created or destroyed. Every entry is exactly where it was — but the 3 \times 2 table has become a 2 \times 3 table. Rows became columns. Columns became rows. The first row of the new table is the first column of the old one.

This operation — swapping rows and columns — is the transpose. It is one of the most basic operations in matrix algebra, and it has surprisingly deep consequences.

Definition

The transpose does two things simultaneously: it swaps the roles of rows and columns, and it reflects the matrix across its principal diagonal (the line from the top-left to the bottom-right corner).

A concrete example. Take

This is a 2 \times 3 matrix. Its transpose is

The first row of A — (1, 4, 7) — became the first column of A^T. The second row — (2, 5, 8) — became the second column. The 2 \times 3 matrix has become a 3 \times 2 matrix.

For a square matrix, the transpose is a reflection across the principal diagonal. The diagonal entries a_{11}, a_{22}, \ldots stay where they are (since swapping i and j does nothing when i = j). The entries above the diagonal swap with their mirrors below.

The diagonal entries 1, 5, 9 did not move. The entry 4 (position (1,2)) swapped with 2 (position (2,1)). The entry 7 (position (1,3)) swapped with 3 (position (3,1)). And so on.

Properties (with proofs)

The transpose interacts with the other matrix operations in clean, predictable ways. Each property is worth proving — the proofs are all short, and they show you exactly where the rule comes from.

Property 1. (A^T)^T = A (double transpose is the original).

Proof. The (i,j)-entry of (A^T)^T is the (j,i)-entry of A^T, which is the (i,j)-entry of A. In symbols: ((A^T)^T)_{ij} = (A^T)_{ji} = a_{ij}. Since every entry matches, (A^T)^T = A. \square

This says that transposing is its own inverse: do it twice and you are back where you started.

Property 2. (A + B)^T = A^T + B^T (transpose of a sum is the sum of transposes).

Proof. Both A and B are m \times n, so A + B is m \times n and (A + B)^T is n \times m. The (i,j)-entry of (A + B)^T is the (j,i)-entry of A + B, which is a_{ji} + b_{ji}. The (i,j)-entry of A^T + B^T is (A^T)_{ij} + (B^T)_{ij} = a_{ji} + b_{ji}. Equal. \square

Property 3. (kA)^T = k A^T (scalars pass through the transpose).

Proof. The (i,j)-entry of (kA)^T is the (j,i)-entry of kA, which is k \cdot a_{ji}. The (i,j)-entry of kA^T is k \cdot (A^T)_{ij} = k \cdot a_{ji}. Equal. \square

Property 4. (AB)^T = B^T A^T (transpose of a product reverses the order).

This is the most important property and the one most likely to trip you up. The transpose of a product is the product of the transposes in reverse order. Not A^T B^T — that is wrong. B^T A^T.

Proof. Let A be m \times p and B be p \times n, so AB is m \times n and (AB)^T is n \times m. On the other side, B^T is n \times p and A^T is p \times m, so B^T A^T is n \times m. The orders match.

Now compare entries. The (i,j)-entry of (AB)^T is the (j,i)-entry of AB:

The (i,j)-entry of B^T A^T is:

The two expressions are identical (the last step uses commutativity of real-number multiplication). So (AB)^T = B^T A^T. \square

The reversal of order has a natural interpretation. If A and B represent two sequential operations (first B, then A — remember, AB applies B first), then transposing reverses the sequence. This is the same pattern as reversing a sequence of steps: to undo "put on socks, then put on shoes," you "take off shoes, then take off socks." The transpose of a product reverses the factors for the same structural reason.

Extension to three matrices. Applying the rule twice:

The pattern extends to any number of factors: the transpose of a product is the product of the transposes in reversed order.

Symmetric and skew-symmetric matrices

The transpose gives rise to two special types of square matrices, each defined by how the matrix relates to its own transpose.

In a symmetric matrix, the entry in row i, column j is the same as the entry in row j, column i. The matrix is a mirror image of itself across the principal diagonal.

Check: s_{12} = 4 = s_{21}. s_{13} = 7 = s_{31}. s_{23} = 5 = s_{32}. The matrix is symmetric.

Symmetric matrices appear everywhere in applications. In physics, the moment of inertia tensor of a rigid body is symmetric. In statistics, every covariance matrix is symmetric. In graph theory, the adjacency matrix of an undirected graph is symmetric (because if there is an edge from vertex i to vertex j, there is also one from j to i). Distance tables — like the road distance chart printed on the back of Indian road atlases — are symmetric: the distance from Jaipur to Udaipur is the same as from Udaipur to Jaipur.

Setting i = j in the condition: a_{ii} = -a_{ii}, which forces 2a_{ii} = 0, so a_{ii} = 0. Every diagonal entry of a skew-symmetric matrix is zero.

Check: k_{12} = 3 and k_{21} = -3. k_{13} = -5 and k_{31} = 5. k_{23} = 7 and k_{32} = -7. Each off-diagonal pair sums to zero. The diagonal is all zeros. The matrix is skew-symmetric.

The decomposition theorem

Here is the deepest result in this article: every square matrix can be split into a symmetric part and a skew-symmetric part. Not approximately, not sometimes — always, uniquely.

Before proving it, see it in action. Take

Compute A^T = \begin{bmatrix} 2 & 5 \\ 3 & 4 \end{bmatrix}.

Symmetric part: P = \frac{1}{2}(A + A^T) = \frac{1}{2}\begin{bmatrix} 4 & 8 \\ 8 & 8 \end{bmatrix} = \begin{bmatrix} 2 & 4 \\ 4 & 4 \end{bmatrix}

Skew-symmetric part: Q = \frac{1}{2}(A - A^T) = \frac{1}{2}\begin{bmatrix} 0 & -2 \\ 2 & 0 \end{bmatrix} = \begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix}

Check: P + Q = \begin{bmatrix} 2 & 4 \\ 4 & 4 \end{bmatrix} + \begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix} = \begin{bmatrix} 2 & 3 \\ 5 & 4 \end{bmatrix} = A. It works.

Now the proof.

Proof that the decomposition works.

Let P = \frac{1}{2}(A + A^T) and Q = \frac{1}{2}(A - A^T).

Step 1: P is symmetric.

Why: use Property 3 (scalar passes through transpose), Property 2 (transpose of sum), and Property 1 (double transpose is the original). The last step uses commutativity of addition.

Step 2: Q is skew-symmetric.

Why: the same properties as before. The key moment is A^T - A = -(A - A^T), which flips the sign and gives Q^T = -Q.

Step 3: P + Q = A.

Why: the A^T terms cancel, leaving \frac{1}{2} \cdot 2A = A.

Step 4: The decomposition is unique.

Suppose A = P_1 + Q_1 where P_1 is symmetric and Q_1 is skew-symmetric. Then A^T = P_1^T + Q_1^T = P_1 - Q_1. Adding the two equations: A + A^T = 2P_1, so P_1 = \frac{1}{2}(A + A^T). Subtracting: A - A^T = 2Q_1, so Q_1 = \frac{1}{2}(A - A^T). Both are uniquely determined by A. \square

The decomposition theorem is not just a formal exercise. In physics, when you decompose the velocity gradient tensor of a fluid into symmetric and skew-symmetric parts, the symmetric part gives the strain rate (how fast the fluid is being stretched) and the skew-symmetric part gives the rotation rate (how fast it is spinning). The same matrix, split into two pieces, tells you two fundamentally different things about the fluid's motion. The decomposition separates the "stretching" from the "twisting."

Worked examples

Common confusions

• "(AB)^T = A^T B^T." The most common mistake. The correct rule is (AB)^T = B^T A^T — the order reverses. The proof shows exactly why: the row-column pairings in the product force the reversal.

• "Every matrix with zeros on the diagonal is skew-symmetric." Having zeros on the diagonal is necessary for skew-symmetry, but not sufficient. You also need a_{ij} = -a_{ji} for all off-diagonal entries. The matrix \begin{bmatrix} 0 & 3 \\ 3 & 0 \end{bmatrix} has zeros on the diagonal but is symmetric, not skew-symmetric.

• "Symmetric matrices must have equal entries everywhere." No — only mirror pairs across the diagonal must be equal. The diagonal entries can all be different. The matrix \begin{bmatrix} 1 & 4 \\ 4 & 9 \end{bmatrix} is symmetric even though 1 \neq 9.

• "The decomposition theorem works only for special matrices." It works for every square matrix, with no exceptions. The matrix A does not need to have any special properties. The decomposition always exists and is always unique.

Going deeper

Transpose and inner products

If you think of column vectors as n \times 1 matrices, then the dot product of two vectors \mathbf{u} and \mathbf{v} can be written as a matrix product:

The transpose turns the column \mathbf{u} into a row, and then the matrix product of a 1 \times n row and an n \times 1 column gives a 1 \times 1 matrix — a scalar — which is exactly the dot product. This is why the transpose is fundamental to linear algebra: it connects matrix multiplication to geometry through the inner product.

Symmetric matrices in spectral theory

Symmetric matrices have a remarkable property that non-symmetric matrices generally lack: they can always be diagonalised, and their eigenvalues are always real. This is the spectral theorem, one of the central results of linear algebra. It says that every symmetric matrix A can be written as A = Q \Lambda Q^T, where \Lambda is a diagonal matrix of eigenvalues and Q is an orthogonal matrix (one whose transpose is its inverse: Q^T Q = I). The spectral theorem is the mathematical foundation of principal component analysis (PCA), a technique used in data science, image processing, and the analysis of large datasets — including the statistical methods used in India's census surveys.

Skew-symmetric matrices and rotations

Every 3 \times 3 skew-symmetric matrix corresponds to a rotation axis in three-dimensional space. Given a vector \mathbf{w} = (w_1, w_2, w_3), the matrix

satisfies W\mathbf{v} = \mathbf{w} \times \mathbf{v} for any vector \mathbf{v} — the matrix-vector product is the cross product. This connection between skew-symmetric matrices and the cross product is used throughout mechanics and robotics. The ISRO Mars Orbiter Mission (Mangalyaan) used rotation matrices built from skew-symmetric components to compute the spacecraft's orientation in space.

Where this leads next

You now know the transpose, its algebraic properties, symmetric and skew-symmetric matrices, and the decomposition theorem that connects them. Here is where these ideas are used next:

• Determinants — Introduction — one key property of determinants is that \det(A^T) = \det(A), which follows directly from the transpose.
• Inverse of Matrix — for invertible matrices, (A^{-1})^T = (A^T)^{-1}: the transpose and the inverse commute.
• Special Matrices — orthogonal matrices satisfy A^T A = I; Hermitian matrices generalise symmetric matrices to the complex numbers.
• Eigenvalues and Eigenvectors — the spectral theorem guarantees that symmetric matrices have real eigenvalues and orthogonal eigenvectors.
• Systems of Linear Equations — the transpose appears in the normal equations A^T A \mathbf{x} = A^T \mathbf{b} for least-squares solutions.