ISOMETRIES OF THE PLANE AND LINEAR ALGEBRA

KEITH CONRAD

1. Introduction
An isometry of R2 is a function h : R2 → R2 that preserves the distance between vectors:
||h(v) − h(w)|| = ||v − w||
p
for all v and w in R2 , where ||(x, y)|| = x2 + y 2 .
Example 1.1. The identity transformation: id(v) = v for all v ∈ R2 .
Example 1.2. Negation: − id(v) = −v for all v ∈ R2 .
Example 1.3. Translation: fixing u ∈ R2 , let tu (v) = v + u. Easily ||tu (v) − tu (w)|| =
||v − w||.
Example 1.4. Rotations around points and reflections across lines in the plane are isome-
tries of R2 . Formulas for these isometries will be given in Example 3.3 and Section 4.
The effects of a translation, rotation (around the origin) and reflection across a line in
R2 are pictured below on sample line segments.

The composition of two isometries of R2 is an isometry. Is every isometry invertible? It is

clear that the three kinds of isometries pictured above (translations, rotations, reflections)
are each invertible (translate by the negative vector, rotate by the opposite angle, reflect a
second time across the same line).
In Section 2, we show the close link between isometries and the dot product on R2 ,
which is more convenient to use than distances due to its algebraic properties. Section 3 is
about the matrices that act as isometries on on R2 , called orthogonal matrices. Section 4
describes the isometries of R2 geometrically.

2. Isometries, dot products, and linearity

Using translations, we can reduce the study of isometries of R2 to the case of isometries
fixing 0.
1
2 KEITH CONRAD

Theorem 2.1. Every isometry of R2 can be uniquely written as the composition t ◦ k where
t is a translation and k is an isometry fixing the origin.
Proof. Let h : R2 → R2 be an isometry. If h = tw ◦ k, where tw is translation by a vector w
and k is an isometry fixing 0, then for all v in R2 we have h(v) = tw (k(v)) = k(v)+w. Setting
v = 0 we get w = h(0), so w is determined by h. Then k(v) = h(v)−w = h(v)−h(0), so k is
determined by h. Turning this around, if we define t(v) = v + h(0) and k(v) = h(v) − h(0),
then t is a translation, k is an isometry fixing 0, and h(v) = k(v) + h(0) = tw ◦ k, where
w = h(0). 
Theorem 2.2. For a function h : R2 → R2 , the following are equivalent:
(1) h is an isometry and h(0) = 0,
(2) h preserves dot products: h(v) · h(w) = v · w for all v, w ∈ R2 .
Proof. The link between length and dot product is the formula
||v||2 = v · v.
Suppose h satisfies (1). Then for all vectors v and w in R2 ,
(2.1) ||h(v) − h(w)|| = ||v − w||.
As a special case, when w = 0 in (2.1) we get ||h(v)|| = ||v|| for all v ∈ R2 . Squaring both
sides of (2.1) and writing the result in terms of dot products makes it
(h(v) − h(w)) · (h(v) − h(w)) = (v − w) · (v − w).
Carrying out the multiplication,
(2.2) h(v) · h(v) − 2h(v) · h(w) + h(w) · h(w) = v · v − 2v · w + w · w.
The first term on the left side of (2.2) equals ||h(v)||2 = ||v||2 = v · v and the last term on
the left side of (2.2) equals ||h(w)||2 = ||w||2 = w · w. Canceling equal terms on both sides
of (2.2), we obtain −2h(v) · h(w) = −2v · w, so h(v) · h(w) = v · w.
Now assume h satisfies (2), so
(2.3) h(v) · h(w) = v · w
for all v and w in R2 . Therefore
||h(v) − h(w)||2 = (h(v) − h(w)) · (h(v) − h(w))
= h(v) · h(v) − 2h(v) · h(w) + h(w) · h(w)
= v · v − 2v · w + w · w by (2.3)
= (v − w) · (v − w)
= ||v − w||2 ,
so ||h(v) − h(w)|| = ||v − w||. Thus h is an isometry. Setting v = w = 0 in (2.3), we get
||h(0)||2 = 0, so h(0) = 0. 
Corollary 2.3. The only isometry of R2 fixing 0 and the standard basis is the identity.
Proof. Let h : R2 → R2 be an isometry that satisfies
       
1 1 0 0
h(0) = 0, h = , h = .
0 0 1 1
Theorem 2.2 says
h(v) · h(w) = v · w
 ISOMETRIES OF THE PLANE AND LINEAR ALGEBRA 3

for all v and w in R2 . Fix v ∈ R2 and let w run over the standard basis vectors e1 = 10

and e2 = 01 , so we see

h(v) · h(ei ) = v · ei .
Since h fixes each ei ,
h(v) · ei = v · ei .
Writing v = c1 e1 + c2 e2 , we get
h(v) · ei = ci
for i = 1, 2, so h(v) = c1 e1 + c2 e2 = v. As v was arbitrary, h is the identity on R2 . 
It is essential in Corollary 2.3 that the isometry fixes 0. An isometry of R2 fixing the
standard basis without fixing 0 need not be the identity! For example, reflection across the
line x + y = 1 in R2 is an isometry of R2 fixing (1, 0) and (0, 1) but not 0 = (0, 0). See
below.

Theorem 2.4. For a function h : R2 → R2 , the following are equivalent:

(1) h is an isometry and h(0) = 0,
(2) h is linear, and the matrix A such that h(v) = Av for all v ∈ R2 satisfies AA> = I2 .
Proof. Suppose h is an isometry and h(0) = 0. We want to prove linearity: h(v + w) =
h(v)+h(w) and h(cv) = ch(v) for all v and w in R2 and all c ∈ R. The mapping h preserves
dot products by Theorem 2.2:
h(v) · h(w) = v · w
for all v and w in R . For the standard basis e1 , e2 of R2 this says h(ei )·h(ej ) = ei ·ej = δij ,
2

so h(e1 ), h(e2 ) is an orthonormal basis of R2 . Thus two vectors in R2 are equal if they have
the same dot product with each of h(e1 ) and h(e2 ).
For all u in R2 we have
h(v + w) · h(u) = (v + w) · u
and
(h(v) + h(w)) · h(u) = h(v) · h(u) + h(w) · h(u) = v · u + w · u = (v + w) · u,
so h(v + w) · h(u) = (h(v) + h(w)) · h(u) for all u. Letting u = e1 , e2 shows h(v + w) =
h(v) + h(w). Similarly,
h(cv) · h(u) = (cv) · u = c(v · u) = c(h(v) · h(u)) = (ch(v)) · h(u),
so again letting u be e1 and e2 tells us h(cv) = ch(v). Thus h is linear.
Let A be the matrix for h: h(v) = Av for all v ∈ R2 , where A has jth column h(ej ). We
want to show AA> = I2 . Since h preserves dot products, the condition h(v) · h(w) = v · w
for all v, w ∈ R2 says Av · Aw = v · w. The fundamental link between the dot product and
4 KEITH CONRAD

matrix transposes, which you should check, is that we can move a matrix to the other side
of a dot product by using its transpose:
(2.4) v · M w = M >v · w
for every 2 × 2 matrix M and v, w ∈ R2 . Using M = A and Av in place of v in (2.4),
Av · Aw = A> (Av) · w = (A> A)v · w.
This is equal to v · w for all v and w, so (A> A)v · w = v · w for all v and w in R2 . Since
the (i, j) entry of a matrix M is M ej · ei , letting v and w run through the standard basis of
R2 tells us A> A = I2 , so A is invertible. An invertible matrix commutes with its inverse,
so A> A = I2 ⇒ AA> = I2 .
For the converse, assume h(v) = Av for v ∈ R2 where AA> = I2 . Trivially h fixes 0. To
show h is an isometry, by Theorem 2.2 it suffices to show
(2.5) Av · Aw = v · w
for all v, w ∈ Since A and its inverse A> commute, we have A> A = I2 , so Av · Aw =
R2 .
A> (Av) · w = (A> A)v · w = v · w. 
Remark 2.5. Linearity of isometries fixing 0 on finite-dimensional (not just 2-dimensional)
vector spaces is due to A. Vogt [1, Lemma 1.5, Theorem 2.4]. The proof above is from [1].
Corollary 2.6. Isometries of R2 are invertible, the inverse of an isometry is an isometry,
and two isometries on R2 that have the same values at 0 and any basis of R2 are equal.
This gives a second proof of Corollary 2.3 as a special case.
Proof. Let h : R2 → R2 be an isometry. By Theorem 2.1, h = k + h(0) where k is an
isometry of R2 fixing 0. Theorem 2.4 tells us there is an invertible matrix A such that
k(v) = Av for all v ∈ R2 , so
h(v) = Av + h(0).
−1 −1
This has inverse h (v) = A (v − h(0)). In particular, h is surjective.
The isometry condition ||h(v) − h(w)|| = ||v − w|| for all v and w in R2 implies ||v − w|| =
||h−1 (v) − h−1 (w)|| for all v and w in R2 by replacing v and w in the isometry condition
with h−1 (v) and h−1 (w). Thus h−1 is an isometry of R2 .
If h1 and h2 are isometries of R2 that are equal on 0 and a basis then the functions
k1 (v) = h1 (v) − h1 (0) and k2 (v) = h2 (v) − h2 (0) are linear and are equal on that basis, so
by linearity k1 = k2 on R2 . That is, h1 (v) − h1 (0) = h2 (v) − h2 (0) for all v in R2 . Since
h1 (0) = h2 (0) we get h1 = h2 on R2 . 
Corollary 2.7. Let P0 , P1 , P2 be 3 points in R2 in “general position”, i.e., they don’t all
lie on a line. Two isometries of R2 that are equal at P0 , P1 , P2 are the same.
In the definition of “general position”, the lines in R2 need not contain the origin.
Proof. We know isometries of R2 are invertible. If h1 and h2 are isometries of R2 with the
same values at each Pi then h2−1 ◦ h1 is an isometry that fixes each Pi . Therefore to prove
h1 = h2 it suffices to show an isometry of R2 that fixes P0 , P1 , and P2 is the identity.
Let h be an isometry of R2 such that h(Pi ) = Pi for i = 0, 1, 2. Set t(v) = v − P0 , which
is a translation. Then tht−1 is an isometry with formula
(tht−1 )(v) = h(v + P0 ) − P0 .
 ISOMETRIES OF THE PLANE AND LINEAR ALGEBRA 5

Thus (tht−1 )(0) = h(P0 )−P0 = 0, so tht−1 is linear by Theorem 2.4. Also (tht−1 )(Pi −P0 ) =
h(Pi ) − P0 = Pi − P0 .
Upon subtracting P0 from P0 , P1 , P2 , the points 0, P1 −P0 , , P2 −P0 are in general position.
That means no hyperplane can contain them all, so there is no nontrivial linear relation
among P1 − P0 and Pn − P0 (a nontrivial linear relation would place these 2 points, along
with 0, on a common line), and thus P1 − P0 , , P2 − P0 is a basis of R2 . By Corollary 2.6,
tht−1 is the identity, so h is the identity. 

3. Orthogonal matrices
We have seen that the isometries of R2 that fix 0 come from matrices A such that
AA> = I2 . These matrices have a name.

Definition 3.1. A 2 × 2 matrix A is called orthogonal if AA> = I2 , or equivalently if

A> A = I2 .

A matrix is orthogonal when its transpose is its inverse. Since det(A> ) = det A, an or-
thogonal matrix A satisfies (det A)2 = 1, so det A = ±1. (Not all matrices with determinant
±1 are orthogonal, such as ( 35 12 ).)

Example 3.2. Negation on R2 (Example 1.2) is an isometry that is described by the matrix
−I2 , which is orthogonal: (−I2 )(−I2 )> = (−I2 )(−I2 ) = I2 .

Example 3.3. By algebra, AA> = I2 if and only if A = ( ab −εb εa

), where a2 + b2 = 1 and
ε = ±1. Writing a = cos θ and b = sin θ, we get the matrices ( cos θ − sin θ ) and ( cos θ sin θ ).
sin θ cos θ sin θ − cos θ
Algebraically, these types of matrices are distinguished by their determinants: the first type
has determinant 1 and the second type has determinant −1.
The geometric effects of these two types of matrices differ. Below on the left, ( cos θ − sin θ )
sin θ cos θ
is a counterclockwise rotation by angle θ around the origin. Below on the right, ( cos θ sin θ
sin θ − cos θ )
is a reflection across the line through the origin at angle θ/2 with respect to the positive
x-axis. (Check ( cos θ sin θ
sin θ − cos θ ) squares to the identity, as any reflection should.)

A = ( cos θ − sin θ ) A = ( cos θ sin θ

sin θ cos θ A(v) sin θ − cos θ )
v v
w w
A(w)

A(v)
A(w)

Let’s explain why ( cos θ sin θ

sin θ − cos θ ) is a reflection at angle θ/2. See the figure below. Pick a
line L through the origin, say at an angle ϕ with respect to the positive x-axis. To find a
formula for reflection across L, we’ll use a basis of R2 with
one vector on L and the other
vector perpendicular to L. The unit vector u1 = cos sin
ϕ
ϕ lies on L and the unit vector
− sin ϕ 2


u2 = cos ϕ is perpendicular to L. For v ∈ R , write v = c1 u1 + c2 u2 with c1 , c2 ∈ R.
6 KEITH CONRAD

v
u2 L
c2 u2
c1 u1
ϕ u1
s(v)
−c2 u2

The reflection of v across L is s(v) = c1 u1 − c2 u2 . Writing a = cos ϕ and b = sin ϕ (so

a2 + b2 = 1), in standard coordinates this becomes
        
a −b c1 a − c2 b a −b c1
(3.1) v = c1 u1 + c2 u2 = c1 + c2 = =
b a c1 b + c2 a b a c2
and in a similar way
s(v) = c1 u1 − c2 u2
  
a b c1
=
b −a c2
  −1
a b a −b
= v by (3.1)
b −a b a
  
a b a b
= v
b −a −b a
 2
a − b2

2ab
= v.
2ab −(a2 − b2 )
cos(2ϕ) sin(2ϕ)
By the sine and cosine duplication formulas, the last matrix is ( sin(2ϕ) − cos(2ϕ)
). Therefore
( cos θ sin θ
sin θ − cos θ ) is a reflection across the line through the origin at angle θ/2.

The geometric meaning of the condition A> A = I2 is that the columns of A are mutually
perpendicular unit vectors (check!). From this we see how to create orthogonal matrices:
starting with an orthonormal basis of R2 , a 2 × 2 matrix having this basis as its columns
(in any order) is an orthogonal matrix, and all 2 × 2 orthogonal matrices arise in this way.
Let O2 (R) denote the set of 2 × 2 orthogonal matrices:
(3.2) O2 (R) = {A ∈ GL2 (R) : AA> = I2 }.
Theorem 3.4. The set O2 (R) is a group under matrix multiplication.
Proof. Clearly I2 ∈ O2 (R). If A and B are in O2 (R), then
(AB)(AB)> = ABB > A> = AA> = I2 ,
so AB ∈ O2 (R). For A ∈ O2 (R), we have A−1 = A> and
(A−1 )(A−1 )> = A> (A> )> = A> A = I2 .
Therefore A−1 ∈ O2 (R). 
 ISOMETRIES OF THE PLANE AND LINEAR ALGEBRA 7

The link between isometries and dot products (Theorem 2.2) gives us a more geometric
description of O2 (R) than (3.2):
(3.3) O2 (R) = {A ∈ GL2 (R) : Av · Aw = v · w for all v, w ∈ R2 }.
The label “orthogonal matrix” is very unfortunate. It suggests that such matrices should
be the ones that preserves orthogonality of vectors:
(3.4) v · w = 0 =⇒ Av · Aw = 0
for all v and w in R2 . While orthogonal matrices do satisfy (3.4), since (3.4) is a special
case of the condition Av · Aw = v · w in (3.3), many matrices satisfy (3.4) and are not
orthogonal matrices! That is, orthogonal matrices (which, by definition, preserve all dot
products) are not the only matrices that preserve orthogonality of vectors (dot products
equal to 0). A simple example of a nonorthogonal matrix satisfying (3.4) is a scalar matrix
cI2 , where c 6= ±1 in R. Since (cv) · (cw) = c2 (v · w), cI2 does not preserve dot products
in general but it does preserve dot products equal to 0. It’s natural to ask which 2 × 2
matrices besides orthogonal matrices preserve orthogonality. Here is the complete answer,
which shows they are not that far from being orthogonal.
Theorem 3.5. A 2 × 2 real matrix A satisfies (3.4) if and only if A is a scalar multiple of
an orthogonal matrix.
Proof. If A = cA0 where A0 is orthogonal and c ∈ R, then Av ·Aw = c2 (A0 v ·A0 w) = c2 (v ·w),
so if v · w = 0 then Av · Aw = 0.
Now assume A satisfies (3.4). Then the vectors Ae1 , Ae2 are mutually perpendicular, so
the columns of A are perpendicular to each other. We want to show that they have the
same length.
Note that e1 +e2 ⊥ e1 −e2 , so by (3.4) and linearity Ae1 +Ae2 ⊥ Ae1 −Ae2 . Writing this in
the form (Ae1 +Ae2 )·(Ae1 −Ae2 ) = 0 and expanding, we are left with Ae1 ·Ae1 = Ae2 ·Ae2 ,
so ||Ae1 || = ||Ae2 ||. Therefore the columns of A are mutually perpendicular vectors with
the same length. Call this common length c. If c = 0 then A = O = 0 · I2 . If c 6= 0 then
the matrix (1/c)A has an orthonormal basis as its columns, so it is an orthogonal matrix.
Therefore A = c((1/c)A) is a scalar multiple of an orthogonal matrix. 
Since a composition of isometries is an isometry and isometries are invertible with the
inverse of an isometry being an isometry, isometries form a group under composition. We
will describe the elements of this group and show how the group law looks in that description.
Theorem 3.6. For A ∈ O2 (R) and w ∈ R2 , the function hA,w : R2 → R2 given by
hA,w (v) = Av + w = (tw A)(v)
is an isometry. Moreover, every isometry of R2 has this form for unique A and w.
Proof. The indicated formula always gives an isometry, since it is the composition of a
translation and orthogonal transformation, which are both isometries.
To show every isometry of R2 has the form hA,w for some A and w, let h : R2 → R2
be an isometry. By Theorem 2.1, h = k(v) + h(0) where k is an isometry of R2 fixing 0.
Theorem 2.2 tells us there is an A ∈ O2 (R) such that k(v) = Av for all v ∈ R2 , so
h(v) = Av + h(0) = hA,w (v)
where w = h(0).
8 KEITH CONRAD

If hA,w = hA0 ,w0 as functions on R2 , then evaluating both sides at 0 gives w = w0 .

Therefore Av + w = A0 v + w for all v, so Av = A0 v for all v, which implies A = A0 . 
Let Iso(R2 ) denote the group of isometries of R2 . Its elements have the form hA,w by
Theorem 3.6. Here is what composition of such mappings looks like:
hA,w (hA0 ,w0 (v)) = A(A0 v + w0 ) + w
= AA0 v + Aw0 + w
= hAA0 ,Aw0 +w (v).
This is similar to the multiplication law in the ax + b group:
 0 0  
aa0 ab0 + b
 
a b a b
= .
0 1 0 1 0 1
In fact, if we write an isometry hA,w ∈ Iso(R2 ) as a 3 × 3 matrix ( A0 w1 ), where the 0 in
the bottom is a row vector of 2 zeros, then the composition law in Iso(R2 ) is multiplication
of the corresponding 3 × 3 matrices, so 2 ) can be viewed as a subgroup of GL (R),
2 v

Iso(R
3
3
acting on R as the column vectors 1 in R (not a subspace!).

4. Geometric description of isometries of R2

We know from Theorem 3.6 what all the isometries of R2 look like by formulas. In this
section we describe what they are like geometrically.
Write an isometry h ∈ Iso(R2 ) as h(v) = Av + w with A ∈ O2 (R) and w ∈ R2 . By
Example 3.3, A is a rotation or reflection, depending on det A.
There turn out to be four possibilities for h: translations, rotations, reflections, and glide
reflections. A glide reflection is the composition of a reflection and a nonzero translation in
a direction parallel to the line of reflection. A picture of a glide reflection is in the figure
below, where the (horizontal) line of reflection is dashed and the translation is a movement
to the right.

The image above, which includes “before” and “after” states, suggests a physical inter-
pretation of a glide reflection: it is the result of turning the plane in space like a half-turn
of a screw. A more picturesque image, suggested to me by Michiel Vermeulen, is the effect
of successive steps with a left foot and then a right foot in the sand or snow (if your feet
are mirror reflections).
The possibilities for isometries of f are collected in Table 1 below. It describes how the
type of an isometry h is determined by det A and the geometry of the set of fixed points of h
(solutions to h(v) = v): empty, a point, a line, or the plane. (The only isometry belonging
to more than one of the four possibilities is the identity, which is both a translation and a
 ISOMETRIES OF THE PLANE AND LINEAR ALGEBRA 9

rotation, so we make the identity its own row in the table.) The table also shows that a
description of the fixed points can be obtained algebraically from A and w.

Isometry Condition Fixed pts

Identity A = I2 , w = 0 R2
Nonzero Translation A = I2 , w 6= 0 ∅
Nonzero Rotation det A = 1, A 6= I2 (I2 − A)−1 w
Reflection det A = −1, Aw = −w w/2 + ker(A − I2 )
Glide Reflection det A = −1, Aw 6= −w ∅
Table 1. Isometries of R2 : h(v) = Av + w, A ∈ O2 (R).

To justify the information in the table we move down the middle column. The first two
rows are obvious, so we start with the third row.
Row 3: Suppose det A = 1 and A 6= I2 , so A = ( cos θ − sin θ ) for some θ and cos θ 6= 1. We
sin θ cos θ
want to show h is a rotation. First of all, h has a unique fixed point: v = Av + w precisely
when w = (I2 − A)v. We have det(I2 − A) = 2(1 − cos θ) 6= 0, so I2 − A is invertible and
p = (I2 − A)−1 w is the fixed point of h. Then w = (I2 − A)p = p − Ap, so

(4.1) h(v) = Av + (p − Ap) = A(v − p) + p.

Since A is a rotation by θ around the origin, (4.1) shows h is a rotation by θ around P .

Rows 4, 5: Suppose det A = −1, so A = ( cos θ sin θ 2
sin θ − cos θ ) for some θ and A = I2 . We again
look at fixed points of h. As before, h(v) = v for some v if and only if w = (I2 − A)v.
But unlike the previous case, now det(I2 − A) = 0 (check!), so I2 − A is not invertible and
therefore w may or may not be in the image of I2 − A. When w is in the image of I2 − A,
we will see that h is a reflection. When w is not in the image of I2 − A, we will see that h
is a glide reflection.
Suppose the isometry h(v) = Av + w with det A = −1 has a fixed point. Then w/2 must
be a fixed point. Indeed, let p be a fixed point, so p = Ap + w. Since A2 = I2 ,

Aw = A(p − Ap) = Ap − p = −w,

so
w w 1 w
h =A + w = Aw + w = .
2 2 2 2
Conversely, if h(w/2) = w/2 then A(w/2) + w = w/2,, so Aw = −w.
Thus h has a fixed point if and only if Aw = −w, in which case
 w w
(4.2) h(v) = Av + w = A v − + .
2 2
Since A is a reflection across some line L through 0, (4.2) says h is a reflection across the
parallel line w/2 + L passing through w/2. See the figure below. (Algebraically, we can say
L = {v : Av = v} = ker(A − I2 ). Since A − I2 is not invertible and not identically 0, its
kernel really is 1-dimensional.)
10 KEITH CONRAD

v
Av

L w/2
w/2 + L h(v)
w
Now assume h has no fixed point, so Aw 6= −w. We will show h is a glide reflection. (The
formula h = Av +w shows h is the composition of a reflection and a nonzero translation, but
w need not be parallel to the line of reflection of A, which is ker(A − I2 ), so this formula for
h does not show directly that h is a glide reflection.) We will now take stronger advantage
of the fact that A2 = I2 .
Since O = A2 − I2 = (A − I2 )(A + I2 ) and A 6= ±I2 (after all, det A = −1), A + I2 and
A − I2 are not invertible. Therefore the subspaces
W1 = ker(A − I2 ), W2 = ker(A + I2 )
are both nonzero, and neither is the whole plane, so W1 and W2 are both one-dimensional.
We already noted that W1 is the line of reflection of A (fixed points of A form the kernel of
A − I2 ). It turns out that W2 is the line perpendicular to W1 . To see why, pick w1 ∈ W1
and w2 ∈ W2 , so
Aw1 = w1 , Aw2 = −w2 .
Then, since Aw1 · Aw2 = w1 · w2 by orthogonality of A, we have
w1 · (−w2 ) = w1 · w2 .
Thus w1 · w2 = 0, so w1 ⊥ w2 .
Now we are ready to show h is a glide reflection. Pick nonzero vectors wi ∈ Wi for i = 1, 2,
and use {w1 , w2 } as a basis of R2 . Write w = h(0) in terms of this basis: w = c1 w1 + c2 w2 .
To say there are no fixed points for h is the same as Aw 6= −w, so w 6∈ W2 . That is, c1 6= 0.
Then
(4.3) h(v) = Av + w = (Av + c2 w2 ) + c1 w1 .
Since A(c2 w2 ) = −c2 w2 , our previous discussion shows v 7→ Av + c2 w2 is a reflection
across the line c2 w2 /2 + W1 . Since c1 w1 is a nonzero vector in W1 , (4.3) exhibits h as the
composition of a reflection across the line c2 w2 /2 + W1 and a nonzero translation by c1 w1 ,
whose direction is parallel to the line of reflection, so h is a glide reflection.
We have now justified the information in Table 1. Each row describes a different kind of
isometry. Using fixed points it is easy to distinguish the first four rows from each other and
to distinguish glide reflections from isometries other than translations. A glide reflection
can’t be a translation since an isometry of R2 is uniquely of the form hA,w , and translations
have A = I2 while glide reflections have det A = −1.
Lemma 4.1. A composition of two reflections of R2 is a translation or a rotation.
Proof. The product of two matrices with determinant −1 has determinant 1, so the com-
position of two reflections has the form v 7→ Av + w where det A = 1. Such isometries
 ISOMETRIES OF THE PLANE AND LINEAR ALGEBRA 11

are translations or rotations by Table 1 (consider the identity to be a trivial translation or

rotation). 
Theorem 4.2. Each isometry of R2 is a composition of at most 2 reflections except for
glide reflections, which are a composition of 3 (and no fewer) reflections.
Proof. We check the theorem for each type of isometry in Table 1 besides reflections, for
which the theorem is obvious.
The identity is the square of every reflection.
For a translation t(v) = v + w, let A be the matrix representing the reflection across
the line w⊥ . Then Aw = −w. Set s1 (v) = Av + w and s2 (v) = Av. Both s1 and s2 are
reflections, and (s1 ◦ s2 )(v) = A(Av) + w = v + w since A2 = I2 .
Now consider a rotation, say h(v) = A(v − p) + p for some A ∈ O2 (R) with det A = 1
and p ∈ R2 . We have h = t ◦ r ◦ t−1 , where t is translation by p and r(v) = Av is a rotation
around the origin. Let A0 be a reflection matrix (e.g., A0 = ( 10 −1
0 )). Set s (v) = AA0 v and
1
0
s2 (v) = A v. Both s1 and s2 are reflections and r = s1 ◦ s2 (check). Therefore
(4.4) h = t ◦ r ◦ t−1 = (t ◦ s1 ◦ t−1 ) ◦ (t ◦ s2 ◦ t−1 ).
The conjugate of a reflection by a translation (or by any isometry, for that matter) is another
reflection, as an explicit calculation using Table 1 shows. Thus, (4.4) expresses the rotation
h as a composition of 2 reflections.
Finally we consider glide reflections. Since this is the composition of a translation and
a reflection, it is a composition of 3 reflections. We can’t use fewer reflections to get a
glide reflection, since a composition of two reflections is either a translation or a rotation
by Lemma 4.1 and we know that a glide reflection is not a translation or rotation (or
reflection). 
In Table 2 we record the minimal number of reflections whose composition can equal a
particular type of isometry of R2 .

Isometry Min. Num. Reflections dim(fixed set)

Identity 0 2
Nonzero Translation 2 0
Nonzero Rotation 2 0
Reflection 1 1
Glide Reflection 3 0
Table 2. Counting Reflections in an Isometry

That each isometry of R2 is a composition of at most 3 reflections can be proved geomet-

rically, without recourse to a prior classification of all isometries of the plane. We will give
a rough sketch of the argument. We will take for granted (!) that an isometry that fixes at
least two points is a reflection across the line through those points or is the identity. (This
is related to Corollary 2.3 when n = 2.) Pick an isometry h of R2 . We may suppose h is
not a reflection or the identity (the identity is the square of every reflection), so h has at
most one fixed point. If h has one fixed point, say P , choose Q 6= P . Then h(Q) 6= Q and
the points Q and h(Q) lie on a common circle centered at P (because h(P ) = P ). Let s be
the reflection across the line through P that is perpendicular to the line connecting Q and
h(Q). Then s ◦ h fixes P and Q, so s ◦ h is the identity or is a reflection. Thus h = s ◦ (s ◦ h)
12 KEITH CONRAD

is a reflection or a composition of two reflections. If h has no fixed points, pick a point P .

Let s be the reflection across the perpendicular bisector of the line connecting P and h(P ),
so s ◦ h fixes P . Thus s ◦ h has a fixed point, so our previous argument shows s ◦ h is either
the identity, a reflection, or the composition of two reflections, so h is the composition of
at most 3 reflections.
A byproduct of this argument, which did not use the classification of isometries, is another
proof that all isometries of R2 are invertible: every isometry is a composition of reflections
and reflections are invertible.

References
[1] A. Vogt, “On the linearity of form isometries,” SIAM Journal on Applied Mathematics 22 (1972), 553–
560.