Transpose

In linear algebra, a matrix is a rectangular grid of numbers. The transpose is a fundamental operation that transforms this grid. To transpose a matrix, you essentially flip it over its main diagonal. This diagonal runs from the top left corner to the bottom right corner.

This operation switches the row and column indices of every element. If an element is in the third row and first column, it moves to the first row and third column. This process is an involution, which means it is its own inverse. If you transpose a matrix twice, you return to the original grid.

There are several ways to construct a transpose. You can reflect the elements across the main diagonal. Alternatively, you can write the original rows as the new columns. You can also write the original columns as the new rows. Formally, if the element in the $i$-th row and $j$-th column of matrix $A$ is $a_{ij}$, then the element in the $j$-th row and $i$-th column of the transpose $A^T$ is $a_{ji}$. If matrix $A$ is an $m \times n$ matrix, its transpose will be an $n \times m$ matrix.

Transposition allows mathematicians to define several special types of square matrices. A symmetric matrix is one where the transpose is equal to the original matrix. A skew-symmetric matrix is one where the transpose equals the negative of the original matrix. In complex numbers, we find even more specific types. A Hermitian matrix is a square complex matrix whose transpose equals its complex conjugate. A skew-Hermitian matrix is one where the transpose equals the negative of the complex conjugate.

Other types include orthogonal matrices, where the transpose equals the inverse, and unitary matrices, where the transpose equals the conjugate inverse.

This mathematical concept was introduced by the British mathematician Arthur Cayley. He presented his findings in 1858. His work appeared in a memoir titled "A memoir on the theory of matrices." This paper was published in the Philosophical Transactions of the Royal Society of London. Cayley's contributions helped formalize the study of matrix theory. His work provided a foundation for how we manipulate these numerical grids today.

Transposition has many important mathematical properties. The operation respects addition, meaning the transpose of a sum is the sum of the transposes. When multiplying multiple matrices, the order of the factors reverses: $(AB)^T = B^T A^T$. This rule also applies to any number of matrices. Furthermore, the determinant of a square matrix remains the same after transposition. The eigenvalues of a square matrix are also identical to the eigenvalues of its transpose. This is because they share the same characteristic polynomial.

In computer science, transposition is a matter of how data is stored in memory. Computers often use row-major order, where rows are stored in contiguous memory blocks.

In this setup, columns are discontiguous, which can slow down certain calculations. To improve performance, software libraries like BLAS may interpret matrices in a transposed order without moving the data. However, some tasks, such as the fast Fourier transform, require physically reordering the data. This makes columns contiguous to increase memory locality. Researchers have studied efficient "in-place" transposition since the late 1950s. This aims to reorder data using minimal extra storage.

Beyond simple grids, the transpose relates to higher-level concepts like linear maps. A matrix can represent a linear map between vector spaces. The transpose of a matrix represents the transpose of that linear map. This works even for infinite-dimensional vector spaces. There is also a closely related concept called the adjoint. The adjoint is defined using bilinear forms on vector spaces. While the terms are sometimes used interchangeably, the matrix of an adjoint is only the transpose if the bases are orthonormal. This connection links matrix algebra to the broader study of geometry and functional analysis.