Does a transpose change the numerical entries?

Ordinary transposition preserves their values and changes their positions.

Can a rectangular matrix be transposed?

Yes. Its row and column counts exchange.

When does a real matrix equal its transpose?

When it is square and symmetric, meaning entries on opposite sides of the main diagonal match.

AI Glossary / Mathematics and optimization / Beginner

Transpose

Definition

The transpose of a matrix is formed by swapping its row and column positions. An entry originally in row i, column j moves to row j, column i. The transpose of A is written A-transpose or Aᵀ. A matrix with m rows and n columns becomes a matrix with n rows and m columns, with its entry values preserved.

What it means

Transposing changes how a table of numbers is oriented. If rows originally represent customers and columns represent products, the transpose has products as rows and customers as columns. The quantities themselves do not become different measurements, but the meaning assigned to each axis changes.
This operation is fundamental in machine-learning formulas because different products require compatible dimensions. A column vector becomes a row vector, making inner products and other constructions possible.
Transposition is not the same as reversing row order, rotating an image, or computing an inverse. It changes index positions according to one precise rule. For a square matrix, entries on the main diagonal stay in place and off-diagonal entries exchange positions across it.

How it works

For every entry A[i,j], set Aᵀ[j,i] to that same value. Applying the operation twice restores A. Transposition distributes over addition and scalar multiplication: (A+B)ᵀ = Aᵀ+Bᵀ and (cA)ᵀ = cAᵀ.
For a product, the order reverses: (AB)ᵀ = BᵀAᵀ. Checking dimensions helps explain why. If A is m-by-n and B is n-by-p, AB is m-by-p, so its transpose is p-by-m. BᵀAᵀ has exactly that shape.
With complex entries, ordinary transposition still only exchanges indices. A conjugate transpose additionally replaces each complex number with its conjugate. Those operations agree for real matrices but differ in general for complex matrices.

Three practical examples

A data analyst can switch a customer-by-product table into a product-by-customer view while preserving the relationship between each label and measurement.
A regression calculation can use Xᵀr to combine residuals across observations into one quantity per feature.
A similarity calculation can form XᵀX to compare columns through inner products, provided the chosen scaling and interpretation make those comparisons meaningful.

Think of it this way

Think of a seating chart whose row labels are families and whose column labels are events. Transposing lets events run down the page and families run across it, with each attendance count staying attached to the same family-event pair. The analogy captures reorientation, but it does not imply sorting, aggregation, or combining duplicate records.

Worked example

Suppose X records two features for three observations, with rows [1, 2], [3, 4], and [5, 6]. Its transpose has rows [1, 3, 5] and [2, 4, 6]. The original entry in row 3, column 2 is 6; after transposition it is in row 2, column 3.
Let the residual column vector be r = [2, −1, 0]. Then Xᵀr has first entry 1×2 + 3×(−1) + 5×0 = −1 and second entry 2×2 + 4×(−1) + 6×0 = 0. The result has two entries, one per feature.
The expression Xr is not a defined matrix product for these shapes: X has two columns, while r has three rows. This dimension mismatch is a useful diagnostic, not something to fix by randomly rearranging data.
Also, XᵀX has rows [35, 44] and [44, 56]. The off-diagonal value 44 is the inner product of the two feature columns: 1×2 + 3×4 + 5×6. The matrix is symmetric, as expected from this construction.

Illustrative case study

In a fictional reporting project, one tool exports products in columns, while another expects products in rows. An analyst transposes the numbers but initially forgets to transpose the associated labels. The receiving tool accepts the shape yet assigns sales figures to the wrong products.
The team checks a tiny table with distinct values and follows one named product-customer pair through the transformation. It then moves the labels and values together.
The lesson is that mathematical transposition preserves position relationships only when metadata is transformed consistently. A file that loads successfully has not necessarily preserved the intended business meaning.

Common misconceptions and mistakes

Do not confuse a transpose with an inverse. For instance, the 1-by-1 matrix [2] has transpose [2] but inverse [0.5]. Only special square matrices have their transpose as their inverse.
Do not forget to reverse multiplication order when transposing a product. Nor should XᵀX automatically be described as a covariance matrix: centering and normalization are needed for the usual sample covariance construction.
Finally, a spreadsheet pivot operation is not necessarily a simple transpose. Pivoting may aggregate repeated categories, introduce missing combinations, or group records. Verify what the tool actually does before treating the two operations as interchangeable.

Business use

Transposition often appears in data exchange and analytical calculations rather than as a visible business feature. Its value lies in aligning the representation with the intended operation while preserving meaning.
Document what each axis represents and check named sample entries after transformation. When an analysis changes from comparing observations to comparing features, communicate that change explicitly. Correct dimensions are necessary, but business labels and units determine whether the resulting calculation answers the right question.

Check your understanding

A matrix has rows [7, 8, 9] and [1, 2, 3]. What are the rows of its transpose?

Answer: They are [7, 1], [8, 2], and [9, 3]. The original 2-by-3 matrix becomes 3-by-2, and transposing again restores the original rows.

Related terms

Learn next

Transpose a small labeled table by hand and track one entry using both its labels and indices. Then verify the product-order rule with two compatible nonsquare matrices. Study conjugate transposition next if your work involves complex-valued signals or unitary matrices.

Sources

Aalto Dictionary of Machine Learning, transpose, symmetric matrix, and unitary matrix entries: index exchange and the distinction involving conjugate transposition. https://aaltodictionaryofml.github.io/ADictML_Main.pdf

Previous: Transparency Next: Transposition table