4.5.37 · D5Linear Algebra (Full)

Question bank — Orthogonal matrices — properties, det = ±1

1,778 words8 min readBack to topic

True or false — justify

A real matrix whose columns are perpendicular is orthogonal.
False — perpendicular is only half of it; the columns must also each have length 1. Columns are perpendicular but .
Every orthogonal matrix has .
True — taking of gives , so . This is a genuine property of all orthogonal matrices.
Every matrix with is orthogonal.
False — is necessary, not sufficient. The shear has yet skews space; its columns aren't orthonormal.
An orthogonal matrix can be non-square, say , as long as its columns are orthonormal.
False — "orthogonal matrix" is defined only for square ; a matrix with orthonormal columns is called having orthonormal columns, but it has no inverse or determinant, so it isn't an orthogonal matrix.
For an orthogonal matrix, .
True — that is the definition rearranged: says undoes . This is why orthogonal maps are so cheap to invert.
The transpose trick works for any invertible matrix.
False — it holds exactly when is orthogonal. Always test before trusting it; see Orthogonal matrices — properties, det = ±1.
Orthogonal matrices preserve the angle between any two vectors.
True — since and lengths are unchanged, the ratio stays the same, so the angle survives — exactly the rigidity shown in the figure above.
An orthogonal matrix can stretch one direction if it shrinks another to compensate.
False — it preserves the length of every vector individually (), so no direction is stretched at all. Volume preservation alone would allow stretch-and-shrink; length preservation forbids it.
The product of two orthogonal matrices is always orthogonal.
True — . This closure is why forms a group.
The product of two orthogonal matrices always has .
False — signs multiply: two reflections ( each) give , but a rotation times a reflection gives . Only "same sign" pairs land in .
Every orthogonal matrix has real eigenvalues.
False — a genuine rotation like has complex eigenvalues (no real arrow keeps its direction), see Eigenvalues and eigenvectors.

Spot the error

" is orthogonal, so ."
The determinant can also be . Writing "" silently assumes a rotation; reflections are orthogonal too and have .
", therefore need not hold."
For a square matrix the two are equivalent — a one-sided inverse of a square matrix is automatically two-sided. Both automatically hold together.
"The columns of are orthonormal, so its rows might not be."
For a square orthogonal matrix rows are orthonormal too, because says exactly that about the rows. Columns orthonormal rows orthonormal.
" means shrinks area, since ."
The magnitude is the area factor, so area is preserved. The minus sign is about orientation flip (handedness), not size — the flip shown in the second panel of the figure.
"Since , we have for all ."
Preserving length does not mean fixing the vector — a rotation moves every nonzero vector while keeping its length. Equal length equal vector.
" has orthonormal columns because ."
The determinant says nothing about column length or perpendicularity here: column has length and isn't perpendicular to .
" preserves length, so it preserves the dot product of any two vectors — hence it must be the identity."
Preserving dot products makes an isometry, not the identity. Every rotation and reflection preserves all dot products yet moves vectors.

Why questions

Why does force the columns to be orthonormal?
The entry of is the dot product of columns and ; setting it equal to (1 on the diagonal, 0 off it) means each column has length 1 () and distinct columns are perpendicular ().
Why can the determinant only be and nothing in between, like ?
Because has only the two real solutions ; geometrically length preservation kills all scaling, so must be exactly .
Why must all eigenvalues of have magnitude ?
An eigenvalue satisfies for some nonzero arrow ; then , forcing — the map can't change the length of an eigenvector, so its stretch factor lies on the unit circle.
Why do complex eigenvalues of an orthogonal matrix come in conjugate pairs ?
is real, so its characteristic polynomial has real coefficients; complex roots of a real polynomial always pair with their conjugates, and pins them to .
Why is called "rotation" and called "reflection"?
is the signed volume factor: keeps orientation (spin, right panel of the figure stays a spun frame), reverses it (mirror flip), linking to [[Rotations and reflections in and ]].
Why is the transpose, not a full matrix inversion, enough to invert an orthogonal matrix?
Because orthonormal columns make directly, so is the inverse — no elimination needed, which is exactly what makes QR decomposition so efficient.
Why does Gram–Schmidt naturally produce the columns of an orthogonal matrix?
It outputs mutually perpendicular unit vectors — precisely the cyan arrows in the figure — and stacking them as columns is exactly the condition ; see Orthonormal bases & Gram–Schmidt.

Edge cases

Is the identity matrix orthogonal, and what is its determinant?
Yes — its columns are the standard orthonormal basis (the grey arrows in the figure), so , and (the trivial rotation that moves nothing).
Is the negative identity in orthogonal, and is it a rotation or reflection?
Yes, orthogonal; in , so it's a rotation by — not a reflection, despite the minus sign.
Is in a rotation or a reflection?
, so in odd dimensions reverses orientation and counts as a reflection-type map — the parity of the dimension decides.
Can a orthogonal matrix exist, and what are the only options?
Yes: with , so . The only orthogonal matrices are (identity) and (reflection through the origin on the line).
Is the zero matrix orthogonal?
No — , and its columns have length , not . An orthogonal matrix must be invertible, but the zero matrix is not.
Can an orthogonal matrix have a zero determinant or be singular?
Never — , so every orthogonal matrix is invertible (with inverse ). Singularity is impossible.
Is a permutation matrix (one per row and column) always orthogonal?
Yes — its columns are the standard basis vectors in some order, hence orthonormal; the determinant is depending on whether the permutation is an even or odd number of swaps.
Does the set of orthogonal matrices with form a group?
No — it lacks the identity () and isn't closed (two of them multiply to ). Only with is a subgroup.