Can a rectangular matrix have a conjugate transpose?

Yes; its dimensions simply reverse.

Does the operation require solving equations?

No. It rearranges entries and conjugates them.

Is Aᴴ always Hermitian?

No. Being a conjugate transpose and being equal to one's own conjugate transpose are different properties.

AI Glossary / Mathematics and optimization / Advanced

Conjugate transpose

Definition

The conjugate transpose of a matrix is obtained by exchanging its rows and columns and taking the complex conjugate of every entry. Complex conjugation changes a + bi to a − bi, where i² = −1. For a matrix A, the result is commonly written Aᴴ or A*. An m-by-n matrix becomes an n-by-m matrix.

What it means

Transposing reorganizes entries; conjugating reverses their imaginary parts. Both operations matter when numbers encode phase as well as magnitude. With real entries, conjugation changes nothing, so the conjugate transpose equals the ordinary transpose.
The operation is central to measuring complex-vector lengths and comparing complex signals. It is not a procedure for discarding imaginary components. Those components carry information, and removing them would change the original problem. Nor does applying the operation necessarily undo a transformation: a conjugate transpose generally differs from a matrix inverse.

How it works

Move entry A at row j, column k to row k, column j, replacing it by its complex conjugate. Transposing before conjugating gives the same result as conjugating before transposing. Applying both operations twice returns the original matrix. For compatible matrices, (AB)ᴴ = BᴴAᴴ: the order reverses.
Using the convention adopted here, the complex inner product is xᴴy. In particular, xᴴx is the sum of squared magnitudes of the entries, and is positive for a nonzero vector. A square matrix is Hermitian when Aᴴ = A. A square unitary matrix satisfies AᴴA = I, and in that special case Aᴴ is its inverse.

Three practical examples

An audio analyst compares complex frequency coefficients from a recorded sound with a reference. Conjugating the reference in the inner product allows matching phase information to contribute correctly. Taking only real parts would throw away part of the comparison.
A sensor team models measurements using a complex matrix. Constructing AᴴA gives a Hermitian positive semidefinite Gram matrix. The team can check that its quadratic forms equal squared output norms before trusting a larger calculation.
A developer tests a complex transformation intended to preserve vector lengths. Checking UᴴU against the identity tests the defining unitary condition. Checking UᵀU instead tests a different property and can reject valid transformations or accept inappropriate ones.

Think of it this way

Think of turning a spreadsheet's columns into rows while also reversing every imaginary sign. Doing only the first job leaves the layout correct but the numbers wrong for complex inner products. A correct shape is therefore a necessary check, but it is not sufficient evidence that the operation was performed correctly.

Worked example

Take an original example with A = [[1 + i, 2], [3i, 4 − i]]. Its conjugate transpose is Aᴴ = [[1 − i, −3i], [2, 4 + i]]. Notice that 3i moves from the lower-left position to the upper-right position and becomes −3i.
Multiplication gives AᴴA = [[11, −1 − 14i], [−1 + 14i, 21]]. Its diagonal entries are real, while the off-diagonal entries are conjugates. For instance, the upper-right entry is (1 − i) × 2 + (−3i) × (4 − i) = −1 − 14i.
Now let x be the column vector (1, i). Then Ax = (1 + 3i, 1 + 7i). Its squared norm is |1 + 3i|² + |1 + 7i|² = 10 + 50 = 60. Computing xᴴ(AᴴA)x also gives 60, providing an independent consistency check.
For x alone, xᴴx = 2. But xᵀx = 1 + i² = 0. A nonzero vector has not acquired zero length; the ordinary transpose was used in a formula that requires conjugation.

Illustrative case study

In a fictional acoustic-monitoring project, a developer builds a two-coefficient reference s = (1, i). A synthetic test signal is y = 2s = (2, 2i), deliberately representing a perfect match at twice the amplitude. The intended comparison sᴴy gives 4. The implementation accidentally uses sᵀy and returns zero.
The team catches the error because the perfect-match test contradicts the expected result. After correcting the operation, it also tests a phase-reversed signal and a real-only signal. It records whether the matching score uses the complex value, magnitude or normalized magnitude; these are separate choices. This case illustrates a calculation failure, not a claim about measured detection performance in a deployed system.

Common misconceptions and mistakes

Conjugating each entry without transposing is not the conjugate transpose. Transposing without conjugating works for real matrices but fails in general for complex matrices. Replacing every complex entry with its magnitude is another operation entirely.
Do not assume AᴴA is always invertible: dependent columns can make it singular. Likewise, a Hermitian matrix need not be positive definite. In software, symbols such as a star or an apostrophe have language-specific meanings, so notation in a textbook is not sufficient to establish what an expression executes.

Business use

The concept matters when AI systems incorporate signal processing, communications, imaging or complex-valued models. A small operation error can produce misleading similarity scores or break numerical calculations while leaving dimensions apparently correct. Review teams can request examples with nonzero imaginary components, checks against directly computed squared magnitudes, and documented normalization conventions. Physical interpretations such as energy also require appropriate units and calibration; algebra alone does not supply them.

Check your understanding

What happens to 2 + 5i?

Answer: Its conjugate is 2 − 5i, and its matrix position is transposed.

Why does (1, i) have squared norm two?

Answer: Conjugation produces 1 × 1 + (−i) × i = 2.

When does a square matrix's conjugate transpose equal its inverse?

Answer: When the matrix is unitary.

Related terms

Unitary matrix
Inner product
Gram matrix
Singular value decomposition
Complex number
matrix transpose
Hermitian matrix
positive semidefinite matrix

Learn next

Practice complex inner products and distinguish Hermitian, unitary and positive definite matrices. Then study complex least squares and how QR or singular value decompositions support stable numerical solutions.

Sources

Justin Solomon, Stanford CS 205A, Mathematical Methods for Computer Vision, Robotics, and Graphics, definitions 5.5–5.7: complex conjugation, conjugate transpose and Hermitian matrices. https://graphics.stanford.edu/courses/cs205a-13-fall/assets/notes/cs205a_notes.pdf MIT OpenCourseWare, Calculus with Applications, section 32.7: complex inner products and unitary transformations. https://ocw.mit.edu/ans7870/18/18.013a/textbook/chapter32/section07.html The matrix calculation and fictional acoustic case are original teaching examples.

Previous: Conjugate prior Next: Conjunct ordering