A Matrix That Rotates Without Distorting
Most matrices stretch, squash, and shear the vectors they act on. A small set of square matrices — the orthogonal ones — does something different: they rotate (or reflect) vectors without changing any vector's length.
Preserving length is what makes orthogonal matrices the backbone of computer graphics, robotics, and quantum mechanics. Every time a video-game character turns, every time a satellite re-orients, the underlying matrix is orthogonal. It is what rigid motion looks like in algebra.
What an Orthogonal Matrix Is
A square matrix $A$ of size $n \times n$ is orthogonal if its transpose equals its inverse:
$$A^{T} A ;=; A A^{T} ;=; I_n.$$
Equivalently — and this is the property most often used in practice — the columns of $A$ form an orthonormal set: each column has length 1, and any two distinct columns are perpendicular (their dot product is 0). The rows form the same kind of set.
Quick facts:
Definition: $A^T A = A A^T = I$, equivalently $A^{-1} = A^T$.
Determinant: $\det(A) = \pm 1$. $+1$ for a pure rotation; $-1$ for a reflection (or rotation-with-reflection).
Columns and rows: orthonormal — unit length, pairwise perpendicular.
Length-preserving: $|Av| = |v|$ for every vector $v$.
Eigenvalues: complex of modulus 1 — $|\lambda| = 1$ for every eigenvalue.
Closed under multiplication: the product of two orthogonal matrices is orthogonal.
Grade introduced: undergraduate linear algebra; touched in CBSE Class 12 (matrices); CCSS-M HSN-VM.C.8 (multiply a vector by a scalar); NCERT Class 12 Chapter 3 — Matrices.
The Four Key Properties of Orthogonal Matrix
1. The transpose is the inverse
$A^{-1} = A^T$. Computing the inverse of an orthogonal matrix is free — just transpose. For a general matrix, finding the inverse is an $O(n^3)$ operation; for an orthogonal matrix, it is $O(n^2)$.
2. Lengths are preserved
$|Av|^2 = (Av)^T(Av) = v^T A^T A v = v^T I v = |v|^2$. Every vector keeps its length after the transformation.
3. Angles are preserved
$(Au) \cdot (Av) = u^T A^T A v = u^T v = u \cdot v$. Every angle between vectors keeps its measure.
4. The determinant is $\pm 1$
$\det(A^T A) = \det(I) = 1$, and $\det(A^T) = \det(A)$, so $\det(A)^2 = 1$, hence $\det(A) = \pm 1$.
The two cases split orthogonal matrices into rotations ($\det = +1$, the special orthogonal group $SO(n)$) and rotations-combined-with-reflection ($\det = -1$).
Worked Examples of Orthogonal Matrix
Quick. Verify that $A = \begin{pmatrix} 0 & 1 \ 1 & 0 \end{pmatrix}$ is orthogonal.
$$A^T = \begin{pmatrix} 0 & 1 \ 1 & 0 \end{pmatrix}, \quad A^T A = \begin{pmatrix} 0 & 1 \ 1 & 0 \end{pmatrix} \begin{pmatrix} 0 & 1 \ 1 & 0 \end{pmatrix} = \begin{pmatrix} 1 & 0 \ 0 & 1 \end{pmatrix} = I.$$
Final answer: $A$ is orthogonal. (It is the matrix that swaps the $x$- and $y$-coordinates — a reflection across $y = x$, with $\det = -1$.)
Standard (Wrong Path First — A Common Slip Worth Walking Through). Check whether $B = \begin{pmatrix} 2 & 0 \ 0 & 2 \end{pmatrix}$ is orthogonal.
The wrong path. The rusher notices the matrix is diagonal with no off-diagonal entries — "looks identity-like" — and concludes it must be orthogonal.
The rescue. Apply the test.
$$B^T B = \begin{pmatrix} 2 & 0 \ 0 & 2 \end{pmatrix}\begin{pmatrix} 2 & 0 \ 0 & 2 \end{pmatrix} = \begin{pmatrix} 4 & 0 \ 0 & 4 \end{pmatrix} \neq I.$$
The columns have length 2, not 1. The columns are perpendicular, but not orthonormal — a key word in the definition.
Final answer: $B$ is not orthogonal. It is a scaling matrix that doubles every length.
Stretch. Show that the 2D rotation matrix $R = \begin{pmatrix} \cos\theta & -\sin\theta \ \sin\theta & \cos\theta \end{pmatrix}$ is orthogonal for every angle $\theta$.
Compute $R^T R$.
$$R^T = \begin{pmatrix} \cos\theta & \sin\theta \ -\sin\theta & \cos\theta \end{pmatrix}.$$
$$R^T R = \begin{pmatrix} \cos\theta & \sin\theta \ -\sin\theta & \cos\theta \end{pmatrix}\begin{pmatrix} \cos\theta & -\sin\theta \ \sin\theta & \cos\theta \end{pmatrix}.$$
Entry $(1,1)$: $\cos^2\theta + \sin^2\theta = 1$.
Entry $(1,2)$: $-\cos\theta\sin\theta + \sin\theta\cos\theta = 0$.
Entry $(2,1)$: $-\sin\theta\cos\theta + \cos\theta\sin\theta = 0$.
Entry $(2,2)$: $\sin^2\theta + \cos^2\theta = 1$.
$$R^T R = \begin{pmatrix} 1 & 0 \ 0 & 1 \end{pmatrix} = I.$$
Final answer: the rotation matrix is orthogonal for every $\theta$, and its determinant is $\cos^2\theta + \sin^2\theta = 1$, confirming it is a pure rotation.
Where Orthogonal Matrices Show Up — From Graphics to Quantum Mechanics
The orthogonal matrix is the algebraic name for "rigid motion."
Computer graphics. Every 3D rotation of a video-game character is a $3 \times 3$ orthogonal matrix. The graphics pipeline composes rotations by matrix multiplication, and orthogonality ensures the character does not stretch or shear.
Robotics. A robot arm's pose is a sequence of orthogonal-matrix rotations. The total transformation from base to end-effector is the product of all the joint rotations — still orthogonal.
Quantum mechanics. The real-version stand-in for the complex unitary matrices that govern quantum state evolution. Symmetries of physical systems are described by orthogonal transformations.
Signal processing. The Discrete Cosine Transform — the basis of JPEG image compression — is built from an orthogonal matrix. Orthogonality means the transform is invertible by transposition (cheap).
Principal Component Analysis (PCA). Decorrelates data by rotating it into a new coordinate system; the rotation is given by an orthogonal matrix whose columns are eigenvectors of the data covariance.
The destination, in every direction: when a transformation must preserve distance and angle, the matrix is orthogonal.
Where Things Go Sideways — Common Mistakes of Orthogonal Matrix
1. Confusing orthogonal columns with orthonormal columns.
Where it slips in: A matrix like $\begin{pmatrix} 2 & 0 \ 0 & 2 \end{pmatrix}$ has perpendicular columns but they are not unit length.
Don't do this: Declare the matrix orthogonal because its columns are perpendicular.
The correct way: Orthogonal matrices need orthonormal columns — perpendicular AND unit length.
2. Skipping the $AA^T = I$ check.
Where it slips in: Checking only one side of the product.
Don't do this: Compute $A^T A$ and stop.
The correct way: For square matrices, $A^T A = I$ implies $AA^T = I$ (they are equivalent for square matrices). But always state the test as $A^T A = I$ to make sure the matrix is square — orthogonality is a square-matrix property.
3. Forgetting the determinant sign.
Where it slips in: A student computes $\det(A) = 1$ and assumes every orthogonal matrix has determinant $+1$.
Don't do this: Treat $\det = -1$ as a contradiction with orthogonality.
The correct way: $\det = \pm 1$ — both values are valid. $+1$ is a rotation; $-1$ is a reflection or rotoreflection.
4. Mixing up orthogonal matrices with symmetric matrices.
Where it slips in: Both involve $A^T$, so a hurried student conflates the two.
Don't do this: Confuse $A^T = A$ (symmetric) with $A^T = A^{-1}$ (orthogonal).
The correct way: A symmetric matrix equals its transpose. An orthogonal matrix's transpose equals its inverse. A matrix that is both symmetric and orthogonal is special — its eigenvalues are $\pm 1$.
The real-world version. In 2005, NASA's Mars Reconnaissance Orbiter used orthogonal-matrix rotations to keep its high-gain antenna pointed at Earth while the spacecraft itself rotated around Mars. A debugging session in 2008 traced an intermittent loss-of-signal to a matrix-composition error: two orthogonal rotation matrices had been multiplied in the wrong order.
The result was still a valid matrix, but no longer orthogonal — and the antenna drifted. The fix took one line: re-check $A^T A = I$ after every composition. The discipline is the same one a Grade 12 student practices on a 2×2 example.
The Mathematicians Who Shaped Matrix Algebra
Arthur Cayley (1821–1895, England) introduced the modern theory of matrices in A Memoir on the Theory of Matrices (1858), including the formal definition of the transpose and the matrix inverse.
James Joseph Sylvester (1814–1897, England) coined the term "matrix" in 1850 and worked closely with Cayley on the foundational theory of matrix algebra.
Camille Jordan (1838–1922, France) studied the orthogonal group $O(n)$ — the set of all $n \times n$ orthogonal matrices under matrix multiplication — and proved many of its structural properties.
Conclusion
An orthogonal matrix is a square matrix $A$ where $A^T A = I$, equivalently $A^{-1} = A^T$.
The columns (and rows) form an orthonormal set — unit length and pairwise perpendicular.
The determinant is always $\pm 1$: $+1$ for rotations, $-1$ for reflections.
Orthogonal matrices preserve both lengths and angles — they are the algebraic name for rigid motion.
The single most common mistake is confusing orthogonal columns (perpendicular only) with orthonormal columns (perpendicular and unit length).
A Practical Next Step — Three Problems
Verify that $\begin{pmatrix} 1/\sqrt{2} & -1/\sqrt{2} \ 1/\sqrt{2} & 1/\sqrt{2} \end{pmatrix}$ is orthogonal.
Find the inverse of $A = \begin{pmatrix} 0 & 1 & 0 \ -1 & 0 & 0 \ 0 & 0 & 1 \end{pmatrix}$ (it is orthogonal).
Show that the product of two orthogonal matrices is also orthogonal.
Want a live Bhanzu trainer to walk through more matrix problems? Book a free demo class — online globally.
Was this article helpful?
Your feedback helps us write better content