Geometric Transformations: Matrices in R² & R³

How to Read a Transformation Matrix

The geometric effect of a matrix

A

is revealed by two pieces of information: what it does to the standard basis vectors (the columns) and what its determinant is.

Column

1

is the image of

\mathbf{e}_1 = (1, 0)

in

\mathbb{R}^2

or

(1, 0, 0)

in

\mathbb{R}^3

. Column

2

is the image of

\mathbf{e}_2

. The matrix maps the standard grid to the parallelogram (or parallelepiped) spanned by these column vectors.

RULE

The absolute value

|\det(A)|

measures how the transformation scales areas (in

\mathbb{R}^2

) or volumes (in

\mathbb{R}^3

).

The sign of

\det(A)

indicates orientation: positive means the transformation preserves handedness, negative means it reverses it. An orthogonal matrix (

\det = \pm 1

) preserves all lengths and angles — it is a rigid motion.

Pitfall: Mistaking area-preserving for rigid

Treating

\det = \pm 1

as proof of a rigid motion fails for

\begin{pmatrix} 1 & 1 \\ 0 & 1 \end{pmatrix}

: its determinant is

1

, yet it sends

(0, 1)

to

(1, 1)

, of length

\sqrt{2}

. A rigid motion needs

A^T A = I

; here

A^T A = \begin{pmatrix} 1 & 1 \\ 1 & 2 \end{pmatrix}

.

Both readings are needed. The determinant is a single number, and it records only what happens to area and to orientation. The columns carry the rest of the picture: their lengths and the angle between them show whether each square of the grid keeps its shape or comes out stretched and tilted.

Geometric Transformation Notation

Notation

Geometric Transformation Notation

Named matrices with their defining parameter hung in the subscript — an angle, an axis, a factor — so the letter says what kind of motion it is and the subscript says which one. Two sign conventions ride along, and one composition order that reads backwards. All catalogued among the linear algebra symbols.

The matrix bracket anatomy and

AB

juxtaposition come from matrix operation notation;

\det

and

|\det|

as a scale factor from determinant and determinant geometry notation;

[T]

and its columns from matrix representation notation;

\theta

,

\cos

and

\sin

from trigonometric notation.

R_\theta

,

S_k

,

P_W

rotation by theta, scaling by k, projection onto W

A naming scheme rather than three separate symbols: the letter names the kind of motion and the subscript supplies the parameter that pins down which one. The subscript's type varies with the letter — an angle for

R

, a number for

S

, a whole subspace for the projection matrix — and reading it wrongly is the usual way these get confused.

CasesIn

\mathbb{R}^3

the rotation needs an axis as well, so the subscript grows:

R_{z,\theta}

or

R_{\mathbf{n}}(\theta)

name the axis first and the angle second — the convention varies, and Rotations in R³ below fixes one.

Do not confuse

P

at its other jobs. The same letter is a projection here, a change-of-basis matrix elsewhere, and a permutation matrix in factorisations — three unrelated tenants distinguished only by what sits in the subscript.

R_\theta = \begin{pmatrix} \cos\theta & -\sin\theta \\ \sin\theta & \cos\theta \end{pmatrix}

the minus sign sits above the diagonal — counterclockwise

The placement of the single minus is the entire convention: with

-\sin\theta

in the upper right, positive

\theta

turns counterclockwise, the standard orientation this page uses. Transposing the matrix — equivalently negating

\theta

— reverses the direction.

CasesComputer graphics often uses the opposite sign because screen coordinates run

y

downward, so the same matrix appears with the minus below the diagonal and still called "rotation by

\theta

"; the convention is a property of the coordinate system, not of the formula.

Do not confuseThe rotation's inverse as something to be computed.

R_\theta^{-1} = R_{-\theta} = R_\theta^{T}

— the transpose already is the inverse, since a rotation is orthogonal, and inverting one by elimination is wasted effort.

BA

means "do

A

first"

the rightmost factor acts first

Composing transformations writes the matrices in the reverse of the order they happen — Combining Transformations below relies on it — because each acts on the column to its right. The notation inherits its direction from function composition, not from reading order.

CasesThe consequence is that

AB \neq BA

carries geometric meaning here: rotating then reflecting is a different motion from reflecting then rotating, and the notation's order is what records which was meant.

Do not confuseLeft-to-right execution. Writing

R_\theta S_k

and reading it as "rotate, then scale" reverses the actual sequence — the scaling happens first, and the mistake is invisible until the two operations fail to commute.

All linear algebra symbols →↑ Linear Transformations

Scaling

Uniform scaling multiplies every coordinate by the same factor:

T(\mathbf{x}) = c\mathbf{x}

, with matrix

cI

. When

c > 1

the transformation enlarges, when

0 < c < 1

it shrinks, and when

c < 0

it reflects through the origin and scales.

Non-uniform scaling stretches each axis independently. In

\mathbb{R}^2

,

T(x, y) = (c_1 x, c_2 y)

has matrix

\text{diag}(c_1, c_2)

. The horizontal axis is scaled by

c_1

and the vertical by

c_2

. A unit square maps to a rectangle with side lengths

|c_1|

and

|c_2|

.

The determinant is

c^n

for uniform scaling and

c_1 c_2

(or

c_1 c_2 c_3

) for non-uniform. When any scaling factor is zero, the transformation collapses that axis entirely and the determinant is zero.

Pitfall: Forgetting the power of c

Writing

\det(cI) = c

gives

\det(3I) = 3

in

\mathbb{R}^3

, although the unit cube becomes a cube of side

3

. Each of the three axes contributes a factor

3

, so

\det(3I) = 3^3 = 27

.

In

\mathbb{R}^3

,

\text{diag}(c_1, c_2, c_3)

scales each coordinate axis independently. The unit cube maps to a rectangular box with side lengths

|c_1|

,

|c_2|

,

|c_3|

and volume

|c_1 c_2 c_3|

.

Scaling the unit square by 1.5

The dashed unit square and its image. With A = 1.5I, e₁ lands at (1.5, 0) and e₂ at (0, 1.5), so the square becomes a square of side 1.5 — every direction stretched by the same factor, the area multiplied by 1.5² = 2.25, the shape unchanged.

Scaling by c multiplies every length by |c| and every area by c².

Rotations in R²

Rotation by angle

\theta

counterclockwise about the origin has matrix

Rotation Matrix 2D

R_\theta = \begin{pmatrix} \cos\theta & -\sin\theta \\ \sin\theta & \cos\theta \end{pmatrix}

Learn more about this formula: Rotation Matrix 2D →

Browse all linear algebra formulas →

The first column

R_\theta \mathbf{e}_1 = (\cos\theta, \sin\theta)

is the image of

(1, 0)

— the point on the unit circle at angle

\theta

. The second column

R_\theta \mathbf{e}_2 = (-\sin\theta, \cos\theta)

is the image of

(0, 1)

— the point at angle

\theta + 90°

.

The determinant is

\cos^2\theta + \sin^2\theta = 1

for every

\theta

: rotations preserve areas and orientation. The matrix is orthogonal:

R_\theta^T R_\theta = I

, so lengths and angles are preserved. The inverse is

R_\theta^{-1} = R_{-\theta} = R_\theta^T

— rotating backward by the same angle.

RULE

Rotations compose by adding angles:

R_\alpha R_\beta = R_{\alpha + \beta}

.

This follows from the trigonometric addition formulas and corresponds to the fact that rotating by

\beta

then by

\alpha

is the same as rotating by

\alpha + \beta

.

Common cases:

R_{90°} = \begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix}

,

R_{180°} = \begin{pmatrix} -1 & 0 \\ 0 & -1 \end{pmatrix}

,

R_{45°} = \frac{1}{\sqrt{2}}\begin{pmatrix} 1 & -1 \\ 1 & 1 \end{pmatrix}

.

Rotating the unit square by 30°

Under a rotation by θ, e₁ lands at (cos θ, sin θ) and e₂ at (−sin θ, cos θ) — exactly the two columns of the rotation matrix. The image is the same square turned about the origin: lengths, angles and area all survive, which is why the determinant is 1.

A rotation matrix is written down by asking one question: where do e₁ and e₂ go?

Rotations in R³

In three dimensions, a rotation is specified by an axis and an angle. The three coordinate-axis rotation matrices are:

Rotation Matrices 3D

R_x(\theta) = \begin{pmatrix} 1 & 0 & 0 \\ 0 & \cos\theta & -\sin\theta \\ 0 & \sin\theta & \cos\theta \end{pmatrix}, \;\; R_y(\theta) = \begin{pmatrix} \cos\theta & 0 & \sin\theta \\ 0 & 1 & 0 \\ -\sin\theta & 0 & \cos\theta \end{pmatrix}, \;\; R_z(\theta) = \begin{pmatrix} \cos\theta & -\sin\theta & 0 \\ \sin\theta & \cos\theta & 0 \\ 0 & 0 & 1 \end{pmatrix}

Learn more about this formula: Rotation Matrices 3D →

Browse all linear algebra formulas →

Rotation by

\theta

about the

z

-axis leaves the

z

-coordinate unchanged and rotates the

xy

-plane:

Rotations in R³

R_z(\theta) = \begin{pmatrix} \cos\theta & -\sin\theta & 0 \\ \sin\theta & \cos\theta & 0 \\ 0 & 0 & 1 \end{pmatrix}

Rotations about the

x

-axis and

y

-axis have the same

2 \times 2

rotation block embedded in different positions:

Rotations in R³

R_x(\theta) = \begin{pmatrix} 1 & 0 & 0 \\ 0 & \cos\theta & -\sin\theta \\ 0 & \sin\theta & \cos\theta \end{pmatrix}, \quad R_y(\theta) = \begin{pmatrix} \cos\theta & 0 & \sin\theta \\ 0 & 1 & 0 \\ -\sin\theta & 0 & \cos\theta \end{pmatrix}

RULE

Every

3 \times 3

rotation matrix is orthogonal with determinant

+1

.

Pitfall: Misplacing the sign for the y-axis

Copying the top-right

-\sin\theta

of

R_x

and

R_z

into

R_y

gives

R_y(-\theta)

: at

\theta = 90°

it sends

(0, 0, 1)

to

(-1, 0, 0)

. In the correct

R_y(\theta)

the

-\sin\theta

sits bottom left, and

R_y(90°)

sends

(0, 0, 1)

to

(1, 0, 0)

.

Each of these matrices can be checked one column at a time, exactly as in the plane: column

j

is the image of

\mathbf{e}_j

. The direction of turn comes from the right-hand rule. With the thumb along the positive

y

-axis, the fingers curl from the positive

z

-axis toward the positive

x

-axis, so in the

xz

-plane the turn runs from

z

to

x

rather than from

x

to

z

.

The axis of rotation is the eigenvector with eigenvalue

1

— the direction that remains fixed. Any rotation in

\mathbb{R}^3

can be decomposed into rotations about the coordinate axes (Euler angles), though the decomposition is not unique.

Fact: Two quarter turns, one third of a turn

The product

R_z(90°)\,R_x(90°)

is a single turn by

120°

about the diagonal line through

(1, 1, 1)

. Every rotation of space has such an axis (Euler's rotation theorem): for a rotation

\det(R - I) = \det(I - R)

, and in odd dimension that forces the value

0

.

Products of these matrices are where the templates stop helping. Turning by

90°

about the

x

-axis and then by

90°

about the

z

-axis gives

R_z(90°)\,R_x(90°) = \begin{pmatrix} 0 & 0 & 1 \\ 1 & 0 & 0 \\ 0 & 1 & 0 \end{pmatrix}

. None of the coordinate axes is left in place, since the columns send

\mathbf{e}_1

to

\mathbf{e}_2

,

\mathbf{e}_2

to

\mathbf{e}_3

and

\mathbf{e}_3

to

\mathbf{e}_1

. The matrix is still orthogonal with determinant

+1

, yet none of the three templates above describes the motion it performs.

Axis of rotation	Matrix	Fixed direction	Plane that rotates
x-axis: Rₓ(θ)	[1 0 0; 0 cosθ −sinθ; 0 sinθ cosθ]	x-coordinate unchanged	yz-plane
y-axis: R_y(θ)	[cosθ 0 sinθ; 0 1 0; −sinθ 0 cosθ]	y-coordinate unchanged	xz-plane
z-axis: R_z(θ)	[cosθ −sinθ 0; sinθ cosθ 0; 0 0 1]	z-coordinate unchanged	xy-plane

A rotation of space is a plane rotation around a fixed axis: the axis stays put while everything turns around it.

Reflections in R²

Reflection across the

x

-axis flips the

y

-coordinate:

T(x, y) = (x, -y)

, with matrix

\begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}

.

Reflection across the

y

-axis flips the

x

-coordinate: matrix

\begin{pmatrix} -1 & 0 \\ 0 & 1 \end{pmatrix}

.

Reflection across the line

y = x

swaps coordinates: matrix

\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}

.

Reflection across an arbitrary line through the origin at angle

\alpha

has matrix

Reflection Across Line 2D

H_\alpha = \begin{pmatrix} \cos 2\alpha & \sin 2\alpha \\ \sin 2\alpha & -\cos 2\alpha \end{pmatrix}

Learn more about this formula: Reflection Across Line 2D →

Browse all linear algebra formulas →

RULE

All reflection matrices share the same properties: the determinant is

-1

(orientation-reversing, area-preserving), the matrix is orthogonal (lengths and angles preserved), and the matrix is involutory (

H^2 = I

) — reflecting twice returns every vector to its starting point.

The eigenvalues are

+1

(vectors on the mirror line) and

-1

(vectors perpendicular to it).

Download

PNGRaster · 2×PDFPrint quality

Transformations · reflections

Reflections in the plane

Four mirrors, four matrices. The first three are special cases of the fourth — substituting $\alpha = 0$ , $\tfrac{\pi}{2}$ and $\tfrac{\pi}{4}$ into the general form recovers them exactly.

4mirrors

Mirrors along the axes2

The x-axis§ 5

\alpha = 0

matrix

\begin{bmatrix} 1 & 0 \\ 0 & -1 \end{bmatrix}

effect on (x, y)

(x, y) \mapsto (x, -y)

fixed directionthe

x

-axis

Diagonal, so its action is entrywise: the first coordinate survives and the second flips. The eigenvalues sit on the diagonal,

+1

and

-1

, exactly as the general result predicts.

The y-axis§ 5

\alpha = \tfrac{\pi}{2}

matrix

\begin{bmatrix} -1 & 0 \\ 0 & 1 \end{bmatrix}

effect on (x, y)

(x, y) \mapsto (-x, y)

fixed directionthe

y

-axis

The same matrix with the signs exchanged, which makes it the negative of the row above. Negating both diagonal entries of

I

instead gives

-I

, a rotation by

\pi

, which reverses both coordinates and has

\det = +1

.

Mirror at 45 degrees1

The line

y = x

§ 5

\alpha = \tfrac{\pi}{4}

matrix

\begin{bmatrix} 0 & 1 \\ 1 & 0 \end{bmatrix}

effect on (x, y)

(x, y) \mapsto (y, x)

fixed directionthe line

y = x

Swapping coordinates is a reflection, which is worth noticing because it also describes the transpose at the level of a

2 \times 2

permutation. Its own inverse, as every reflection is.

The general mirror1

Line at angle

\alpha

§ 5any line through the origin

matrix

\begin{bmatrix} \cos 2\alpha & \sin 2\alpha \\ \sin 2\alpha & -\cos 2\alpha \end{bmatrix}

effect on (x, y)reflects across that line

fixed directionthe mirror line itself

The

2\alpha

is not a typo: rotating the mirror by

\alpha

turns the image by

2\alpha

, since the incoming and outgoing rays each make angle

\alpha

with it. Symmetric and orthogonal at once, which is unusual and is why

A^{-1} = A^{\mathsf{T}} = A

.

Every reflection has

\det = -1

, eigenvalues

+1

along the mirror and

-1

across it, and equals its own inverse. Those three facts hold whatever

\alpha

is, which is what makes reflection a family rather than a list — and why the general row is the only one that has to be remembered.

Reflections in the plane·/linear-algebra/transformations/geometricLearn Math Class

The doubled angle in the general matrix is the detail worth pausing on. Rotating the mirror by

\alpha

moves the image by

2\alpha

, because the incoming ray and the outgoing ray each make angle

\alpha

with the line — so the total turn is twice the tilt. It also explains why composing two reflections gives a rotation by twice the angle between their mirrors, which is the fastest route to that result.

Reflection in an axis, then in a diagonal

Points on the mirror line stay exactly where they are while everything else swaps to the far side at equal distance. Changing the line changes the matrix but not the behaviour, and applying either matrix twice returns every point to its start. Rotate the mirror line and watch the entries change on the reflection explorer.

Any orthogonal matrix with determinant −1 in two dimensions is a reflection of this kind, and such a matrix is automatically its own inverse.

Reflections in R³

Reflection across a coordinate plane negates the coordinate perpendicular to that plane. Reflection across the

xy

-plane:

\text{diag}(1, 1, -1)

. Across the

xz

-plane:

\text{diag}(1, -1, 1)

. Across the

yz

-plane:

\text{diag}(-1, 1, 1)

.

Reflection across an arbitrary plane through the origin with unit normal

\mathbf{n}

is given by the Householder matrix:

Householder Reflection

H = I - 2\,\mathbf{n}\mathbf{n}^T

Learn more about this formula: Householder Reflection →

Browse all linear algebra formulas →

This matrix subtracts twice the component of each vector in the direction of

\mathbf{n}

, effectively mirroring across the plane perpendicular to

\mathbf{n}

. Householder reflections are orthogonal, have determinant

-1

, and satisfy

H^2 = I

. They are the building blocks of the QR decomposition and are widely used in numerical algorithms.

A reflection keeps its mirror fixed and reverses only the one direction perpendicular to it.

Projections

Orthogonal projection onto a subspace collapses each vector onto its nearest point in the subspace, discarding the perpendicular component.

Projection onto the

x

-axis in

\mathbb{R}^2

:

T(x, y) = (x, 0)

, with matrix

\begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix}

.

Projection onto a line through the origin in direction

\mathbf{u}

:

Projection onto Line

P = \frac{\mathbf{u}\mathbf{u}^T}{\mathbf{u}^T\mathbf{u}}

Learn more about this formula: Projection onto Line →

Browse all linear algebra formulas →

Projection onto a plane with unit normal

\mathbf{n}

in

\mathbb{R}^3

:

Projection onto Plane

P = I - \frac{\mathbf{n}\mathbf{n}^T}{\mathbf{n}^T\mathbf{n}}

Learn more about this formula: Projection onto Plane →

Browse all linear algebra formulas →

Pitfall: Dropping the denominator of P

Using

\mathbf{u}\mathbf{u}^T

alone for

\mathbf{u} = (1, 2)

gives

\begin{pmatrix} 1 & 2 \\ 2 & 4 \end{pmatrix}

, which sends

(1, 2)

to

(5, 10)

instead of leaving it in place. Dividing by

\mathbf{u}^T\mathbf{u} = 5

gives

P = \frac{1}{5}\begin{pmatrix} 1 & 2 \\ 2 & 4 \end{pmatrix}

, and

P(1, 2) = (1, 2)

.

All orthogonal projection matrices share the same algebraic signature:

P^2 = P

(idempotent — projecting twice is the same as projecting once),

P^T = P

(symmetric), eigenvalues are

0

or

1

, and

\text{rank}(P) = \text{tr}(P)

. The determinant is

0

unless the projection is onto the full space — any projection onto a proper subspace collapses at least one dimension.

The plane dropped onto a line

Unlike a reflection, this one destroys information: two different points can land on the same place and there is no way back. The determinant is zero and the matrix is idempotent, which are the algebraic signatures of exactly that loss. Compare it against the reflection above on the 2D projection explorer.

Reflections and rotations preserve information; projections and their relatives are where it starts being lost.

Shears

A shear displaces each point in proportion to its distance from a fixed line or plane. The two

2 \times 2

shear matrices are:

Shear Matrix

\text{Shear}_x = \begin{pmatrix} 1 & k \\ 0 & 1 \end{pmatrix}, \qquad \text{Shear}_y = \begin{pmatrix} 1 & 0 \\ k & 1 \end{pmatrix}

Learn more about this formula: Shear Matrix →

Browse all linear algebra formulas →

In

\mathbb{R}^2

, a horizontal shear shifts the

x

-coordinate by

k

times the

y

-coordinate:

Shears

T(x, y) = (x + ky, \; y), \quad \text{matrix } \begin{pmatrix} 1 & k \\ 0 & 1 \end{pmatrix}

A vertical shear shifts the

y

-coordinate by

k

times the

x

-coordinate:

Shears

T(x, y) = (x, \; kx + y), \quad \text{matrix } \begin{pmatrix} 1 & 0 \\ k & 1 \end{pmatrix}

Fact: Three shears make a rotation

For

\theta \neq 180°

,

R_\theta

equals a horizontal shear by

-\tan\frac{\theta}{2}

, then a vertical shear by

\sin\theta

, then the first shear again. Paeth's 1986 method rotates digital images this way, since each shear only slides whole rows or columns of pixels.

Both are triangular matrices with determinant

1

: shears are area-preserving and orientation-preserving. They are not orthogonal — angles are distorted. A square sheared horizontally becomes a parallelogram of the same area but with tilted sides.

In

\mathbb{R}^3

, there are six possible shear directions (each coordinate shifted by a multiple of each other coordinate). Each is represented by an identity matrix with one off-diagonal entry replaced by

k

.

Shears and rotations look like opposites. A rotation keeps every length and every angle, while a shear tilts the grid and changes the angle at the corners of every square. Yet both have determinant

1

, and the tilt that one shear puts in can be taken out again by a shear acting in the other direction.

THEOREM

Shears are fundamental building blocks — any invertible matrix with determinant

1

can be written as a product of shears.

Shearing the unit square

The horizontal shear [1 1; 0 1] leaves e₁ where it is and pushes e₂ sideways to (1, 1). Every point slides parallel to the x-axis by its own height, so the square leans into a parallelogram with the same base and the same height — and therefore the same area, 1.

A shear changes shape but never area, which is why its determinant is 1.

Combining Transformations

RULE

Composing transformations corresponds to multiplying their matrices.

A rotation followed by a scaling is

SA

where

S

is the scaling matrix and

A

is the rotation matrix. The product is applied right-to-left:

A

acts first, then

S

.

Order matters. Rotating then shearing produces a different result from shearing then rotating:

\text{Shear} \cdot \text{Rotation} \neq \text{Rotation} \cdot \text{Shear}

in general.

Pitfall: Reading a product left to right

Writing rotate by

90°

, then reflect across the

x

-axis, as

R_{90°} H

with

H = \text{diag}(1, -1)

gives

\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}

, which sends

(1, 0)

to

(0, 1)

. The rotation acts first in

H R_{90°} = \begin{pmatrix} 0 & -1 \\ -1 & 0 \end{pmatrix}

, which sends

(1, 0)

to

(0, -1)

.

Turning a description into a matrix is therefore a matter of placement. The step named first in the description stands at the far right, next to the vector it acts on, and each later step is written to the left of the steps before it. A three-step description such as scale, then rotate, then shear becomes

\text{Shear} \cdot R_\theta \cdot S_k

.

The singular value decomposition reveals the hidden geometric structure of any matrix:

A = U\Sigma V^T

, where

V^T

is a rotation (or rotation-reflection),

\Sigma

is a coordinate-axis scaling, and

U

is another rotation (or rotation-reflection). Every linear transformation is a rotation or reflection, followed by a scaling along the coordinate axes, followed by another rotation or reflection. The singular values in

\Sigma

measure the maximum stretching in each orthogonal direction.

This decomposition means no linear transformation is truly exotic — even the most complex-looking matrix is just three simple geometric operations composed together.

Shear after rotation, rotation after shear

A shear and a rotation applied to the same vector in both orders. The solid two-step trail is one order, ending at ABv; the dashed trail is the other order, ending somewhere else — and the unit square is carried to a different parallelogram by each. The matrices are the same two; only the order of multiplication changed. Swap the order yourself on the matrix composition visualizer.

Because matrix products do not commute, the order in which transformations are applied is part of the transformation.

Determinant as Geometric Signature

The determinant classifies every linear transformation by its effect on size and orientation.

|\det(A)|

is the factor by which the transformation scales areas (in

\mathbb{R}^2

) or volumes (in

\mathbb{R}^3

). A unit square maps to a parallelogram of area

|\det(A)|

. A unit cube maps to a parallelepiped of volume

|\det(A)|

.

\det(A) > 0

: the transformation preserves orientation. Counterclockwise stays counterclockwise in

\mathbb{R}^2

; right-handed stays right-handed in

\mathbb{R}^3

. Rotations and shears fall in this category.

\det(A) < 0

: the transformation reverses orientation. Counterclockwise becomes clockwise; right-handed becomes left-handed. Reflections are the canonical example.

\det(A) = 0

: the transformation collapses at least one dimension. The image is a proper subspace — a line or point in

\mathbb{R}^2

, a plane, line, or point in

\mathbb{R}^3

. Projections onto proper subspaces and singular matrices fall here.

|\det(A)| = 1

: the transformation preserves area or volume. Rotations (

\det = +1

) and reflections (

\det = -1

) are the area-preserving and volume-preserving transformations. Shears also have

\det = 1

, preserving area despite distorting angles.

det(A)	Effect on area / volume	Effect on orientation	Typical transformations
\|det(A)\| > 1	expands by factor \|det(A)\|	preserves if det > 0, reverses if det < 0	uniform scaling with \|c\| > 1
0 < \|det(A)\| < 1	shrinks by factor \|det(A)\|	preserves if det > 0, reverses if det < 0	uniform scaling with 0 < \|c\| < 1
det(A) = +1	preserves area / volume	preserves orientation	rotation, shear
det(A) = −1	preserves area / volume	reverses orientation (handedness flip)	reflection, improper rotation
det(A) = 0	collapses at least one dimension (image is a proper subspace)	undefined / degenerate	projection onto a proper subspace, any singular matrix

Area scaled, then area destroyed

The determinant reports what the transformation does to area: above, the unit square has become a parallelogram of some definite size; below, it has been flattened to nothing. A negative value would mean the square had been turned over as well. One number, carrying both the scaling and the orientation. Change the matrix and watch it respond on the linear transformation explorer.

This is why the determinant can classify a transformation family before any of its entries are examined individually.

Summary: Algebraic Signature of Each Family

Each family of geometric transformations has been treated in its own section, with its own matrix templates and its own algebraic identities. The table below sets all five side by side across the diagnostic signatures that distinguish them — orthogonality, determinant, eigenvalues, the characteristic algebraic identity, and what each family does to length and to area. It is the recognition card to keep nearby when a matrix is in hand and the question is which geometric family it belongs to.

Read the other way round, this is a recognition problem: given a matrix and no picture, which family is it? The two columns below answer that without any geometry — an identity the matrix satisfies, and the spectrum it must have. Both are checkable by computation, which matters because in dimensions above three the picture is unavailable anyway.

Download

PNGRaster · 2×PDFPrint quality

Transformations · signatures

Recognising a family from its matrix

Each family has an algebraic fingerprint — an identity its matrix satisfies and a spectrum it must have. Given an unfamiliar matrix, these are what identify it without drawing anything.

5families

Rigid — lengths preserved2

1

Rotation§ 3

eigenvalues

e^{\pm i\theta}

— complex pair

R^{\mathsf{T}}R = I, \quad \det R = 1

Orthogonal with positive determinant. Over

\mathbb{R}

a plane rotation has no eigenvectors at all, since no direction survives unturned — which is exactly what the complex pair records. In

\mathbb{R}^3

there is one real eigenvalue

1

, and its eigenvector is the axis.

2

Reflection§ 5

eigenvalues

+1

on the mirror,

-1

across it

H^2 = I, \quad H^{\mathsf{T}} = H, \quad \det H = -1

Involutory and symmetric. Applying it twice returns the original, so

H

is its own inverse. The eigenvalue

-1

is what the negative determinant reports — orientation reverses, and no rotation can do that.

Non-rigid — lengths change2

3

Uniform scaling§ 2

eigenvalue

c

, repeated

n

times

A = cI, \quad \det A = c^n

The only family commuting with everything, since

cI

passes through any product. Every direction is an eigenvector, so the matrix is diagonal in every basis — the extreme opposite of a shear.

4

Shear§ 8

eigenvalue

1

repeated — defective

triangular,

1

s on the diagonal,

\det = 1

Area is preserved yet nothing is rigid. The repeated eigenvalue has only one eigenvector rather than two, so a shear cannot be diagonalised — the standard example of algebraic multiplicity exceeding geometric.

Not invertible1

5

Orthogonal projection§ 7

eigenvalues only

0

or

1

P^2 = P, \quad P^{\mathsf{T}} = P, \quad \det P = 0

Idempotent and symmetric. The zero determinant is not incidental — a projection collapses the perpendicular directions and nothing recovers them, so it is the only family here with no inverse. See projection properties.

The determinant sorts them at a glance: rotations and shears preserve area, reflections flip it, projections destroy it, and scaling multiplies it by

c^n

. Two families share

\det = 1

— rotation and shear — and the eigenvalues separate them: a rotation has a complex pair, a shear has a repeated real one and too few eigenvectors to diagonalise.

Recognising a family from its matrix·/linear-algebra/transformations/geometricLearn Math Class

Notice which pairs are hard to tell apart. Rotation and shear both have

\det = 1

, so area alone will not separate them — the eigenvalues do, a complex pair against a defective repeated root. Reflection and projection are both symmetric, and there the determinant separates them:

-1

against

0

, orientation reversed against information destroyed. No single test identifies a family; the identity and the spectrum together do.

Common Pitfalls

Five mistakes come up again and again when a transformation is read from its matrix or a matrix is written down from a description. Each is marked with a box where it happens in the sections above; here each is worked through in full, with the wrong and the right working side by side.

Mistaking area-preserving for rigid

A matrix with determinant

1

or

-1

is taken to be a rotation or a reflection, because it preserves area. The determinant measures only area and orientation; lengths and angles are never tested.

Wrong working:

A = \begin{pmatrix} 1 & 1 \\ 0 & 1 \end{pmatrix}

has

\det A = 1 \cdot 1 - 1 \cdot 0 = 1

, so it is declared a rotation, and every length is assumed to stay the same.

Right working: test orthogonality directly.

A^T A = \begin{pmatrix} 1 & 0 \\ 1 & 1 \end{pmatrix}\begin{pmatrix} 1 & 1 \\ 0 & 1 \end{pmatrix} = \begin{pmatrix} 1 & 1 \\ 1 & 2 \end{pmatrix} \neq I

The matrix is a horizontal shear. It sends

(0, 1)

to

(1, 1)

, of length

\sqrt{2}

, and the right angle between the two axes becomes

45°

.

Why it happens: every rotation and every reflection has

\det = \pm 1

, so the reverse statement feels natural. It is false: area can be kept while shape is lost.

How to check: look at the columns. A rigid motion has columns of length

1

that are perpendicular to each other. Here the second column

(1, 1)

has length

\sqrt{2}

, and the dot product of the two columns is

1

, not

0

.

Forgetting the power of c

The determinant of the uniform scaling

cI

is taken to be

c

, as if one factor of

c

were enough. Each of the

n

axes is stretched by

c

, so the factor appears

n

times.

Wrong working: in

\mathbb{R}^3

,

\det(3I) = 3

, so the unit cube would map to a solid of volume

3

.

Right working:

3I = \text{diag}(3, 3, 3)

, so

\det(3I) = 3 \cdot 3 \cdot 3 = 27

The unit cube becomes a cube of side

3

and volume

27

. In

\mathbb{R}^2

the same matrix has

\det(3I) = 9

.

Why it happens:

cI

reads as

c

times

I

, and

\det I = 1

, so taking the

c

outside once seems natural. The general rule for an

n \times n

matrix is

\det(cA) = c^n \det(A)

.

How to check: compute the area or volume of the image of the unit square or cube directly. Its sides have lengths

|c|

, one per axis, and their product is

|c|^n

.

Misplacing the sign for the y-axis

R_x

and

R_z

both have

-\sin\theta

above the diagonal, and that pattern is carried over to

R_y

. In

R_y

the

-\sin\theta

sits below the diagonal, in the bottom-left corner.

Wrong working: the copied pattern at

\theta = 90°

is

\begin{pmatrix} \cos\theta & 0 & -\sin\theta \\ 0 & 1 & 0 \\ \sin\theta & 0 & \cos\theta \end{pmatrix} = \begin{pmatrix} 0 & 0 & -1 \\ 0 & 1 & 0 \\ 1 & 0 & 0 \end{pmatrix}

It sends

(0, 0, 1)

to

(-1, 0, 0)

and

(1, 0, 0)

to

(0, 0, 1)

. This is

R_y(-\theta)

, a turn the wrong way.

Right working:

R_y(90°) = \begin{pmatrix} 0 & 0 & 1 \\ 0 & 1 & 0 \\ -1 & 0 & 0 \end{pmatrix}

It sends

(0, 0, 1)

to

(1, 0, 0)

and

(1, 0, 0)

to

(0, 0, -1)

, as the right-hand rule about the positive

y

-axis requires.

Why it happens: the plane that turns is the

xz

-plane, and the matrix lists its coordinates in the order

x

,

z

. The cyclic order

x \to y \to z \to x

puts

z

before

x

, so reading the

2 \times 2

block in the matrix's own order reverses the turn.

How to check: look at column

3

, the image of

(0, 0, 1)

. For a small positive angle about the

y

-axis, the

z

-axis tips toward the positive

x

-axis, so the top entry of column

3

must be

+\sin\theta

.

Dropping the denominator of P

The projection onto the line through

\mathbf{u}

is written as

\mathbf{u}\mathbf{u}^T

, without the division by

\mathbf{u}^T\mathbf{u}

. This is correct only when

\mathbf{u}

is a unit vector.

Wrong working: with

\mathbf{u} = (1, 2)

,

\mathbf{u}\mathbf{u}^T = \begin{pmatrix} 1 & 2 \\ 2 & 4 \end{pmatrix}, \qquad \begin{pmatrix} 1 & 2 \\ 2 & 4 \end{pmatrix}\begin{pmatrix} 1 \\ 2 \end{pmatrix} = \begin{pmatrix} 5 \\ 10 \end{pmatrix}

A vector already on the line is moved and made five times as long.

Right working:

\mathbf{u}^T\mathbf{u} = 1 + 4 = 5

, so

P = \frac{1}{5}\begin{pmatrix} 1 & 2 \\ 2 & 4 \end{pmatrix}, \qquad P\begin{pmatrix} 1 \\ 2 \end{pmatrix} = \begin{pmatrix} 1 \\ 2 \end{pmatrix}, \qquad P\begin{pmatrix} 1 \\ 0 \end{pmatrix} = \begin{pmatrix} 1/5 \\ 2/5 \end{pmatrix}

Why it happens: many formulas are stated for unit vectors, such as the Householder matrix

I - 2\,\mathbf{n}\mathbf{n}^T

on this page. There the denominator equals

1

and is left out, and the shorter form is then used for vectors that are not unit vectors.

How to check: a projection satisfies

P^2 = P

, and its trace equals its rank. Here

(\mathbf{u}\mathbf{u}^T)^2 = \begin{pmatrix} 5 & 10 \\ 10 & 20 \end{pmatrix} = 5\,\mathbf{u}\mathbf{u}^T

, and the trace of

\mathbf{u}\mathbf{u}^T

is

5

, while a projection onto a line has rank

1

.

Reading a product left to right

A sequence of transformations is written down in the order the steps happen, from left to right. The product then performs them in the reverse order, because the rightmost factor acts first.

Wrong working: rotate by

90°

, then reflect across the

x

-axis, with

H = \text{diag}(1, -1)

, written as

R_{90°} H = \begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix}\begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix} = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}

This is the reflection across the line

y = x

, and it sends

(1, 0)

to

(0, 1)

.

Right working: the first step goes on the right.

H R_{90°} = \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}\begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix} = \begin{pmatrix} 0 & -1 \\ -1 & 0 \end{pmatrix}

The rotation takes

(1, 0)

to

(0, 1)

and the reflection takes that to

(0, -1)

. The result is the reflection across the line

y = -x

.

Why it happens: text is read from left to right, and the description lists the steps in that order. But each matrix acts on the vector to its right:

(H R_{90°})\mathbf{x} = H(R_{90°}\mathbf{x})

.

How to check: follow one basis vector through the steps by hand, then compare the result with the matching column of the product. Here

(1, 0)

must end at

(0, -1)

, the first column of

H R_{90°}

.

Geometric Transformations FAQ

In $R_{\theta}S_k$ , does the rotation or the scaling happen first?

+

The scaling, because the rightmost factor acts first. Reading the product left to right as "rotate, then scale" reverses the actual sequence, and the mistake stays invisible until the two operations fail to commute. Here

AB \neq BA

carries real geometric weight: rotating then reflecting is a different motion from reflecting then rotating.Read more →

How do you invert a rotation matrix?

+

You do not need to compute anything:

R_{\theta}^{-1} = R_{-\theta} = R_{\theta}^{T}

. Because a rotation is orthogonal, its transpose already is its inverse, so running elimination on one is wasted effort. Negating the angle and transposing the matrix are two descriptions of the same operation.Read more →

Why does $P$ mean different things in different places?

+

It has three unrelated tenants in linear algebra: a projection matrix here, a change-of-basis matrix in coordinate work, and a permutation matrix in factorisations such as

PA = LU

. Only the subscript and the surrounding claim distinguish them, so the letter alone never settles which object is meant.Read more →

Geometric Transformations

Rotations, Reflections, Projections, and More

How to Read a Transformation Matrix

Geometric Transformation Notation

Geometric Transformation Notation

Scaling

Rotations in R²

Rotations in R³

Reflections in R²

Reflections in the plane

Reflections in R³

Projections

Shears

Combining Transformations

Determinant as Geometric Signature

Summary: Algebraic Signature of Each Family

Recognising a family from its matrix

Common Pitfalls

Geometric Transformations FAQ

In RθSkR_{\theta}S_kRθ​Sk​, does the rotation or the scaling happen first?

How do you invert a rotation matrix?

Why does PPP mean different things in different places?

In $R_{\theta}S_k$ , does the rotation or the scaling happen first?

Why does $P$ mean different things in different places?