Linear Maps from Rm to Rn

S Kumaresan
[email protected]

December 8, 2023

Abstract

We classify (find) all linear maps T : Rm → Rn . The plan is in the following list
of steps.
1. Find all linear maps from R to R. They are of the from f ( x ) = ax where
a = f (1).
2. Find all linear maps from Rn to R. They are of the from f ( x ) = ∑k ak xk where
a k = f ( e k ).
3. The projection maps πi : Rm → R, 1 ≤ i ≤ m, defined by π( x1 , . . . , xm ) = xi
are linear. Note that πi ( x ) = x · ei .
4. Composition of linear maps is a linear map.
5. Let T : Rm → Rn be linear. Let Ti := πi ◦ T. We ”know” what Ti looks like
from Item 2. Ti x = ai1 x1 + · · · + aim xm , 1 ≤ i ≤ n. Note that aij = Te j · ei .
6. Let Tx = (y1 , . . . , yn ). Then yi := πi (y) = πi ( Tx ) = (πi ◦ T )( x ).
7. Matrix representation of T : Rm → Rn . Let
 
    x1
a11 a12 . . . a1m a11 a12 . . . a1m  
 .. .. .. ..  .  .. .. .. ..   x2 
A :=  . . . .  Note that Tx =  . . . .  ..  .

.
an1 an2 . . . anm an1 an2 . . . anm
 
xm

8. What is Tei ? Geometric meaning of the matrix A of T: Aei is the i-th column of
A.
9. What is the geometric meaning of the rows? The i-th row constitutes the coor-
dinates of Ti x.
10. Matrix representation of a linear map T : Rm → Rn .

We start with the question: What are all the linear maps from R to R? Let f : R → R
be linear. Given x ∈ R, we write it as x = x · 1 where we treat x on the right side as
a scalar and 1 as a (basic) vector. That is, we consider {1} as a basis for R and express

1
the vector x as a scalar multiple of the basic vector 1. Since f is linear, f ( x ) = f ( x · 1) =
x f (1) = ax where a := f (1). Thus, if f : R → R is linear, then f is given by f ( x ) = ax
where a = f (1). Conversely, if a ∈ R and if we define f : R → R by f ( x ) = ax, then f
is linear. We also note that f (1) = a. Thus we have arrived at the following lemma.

Lemma 1. A map f : R → R is linear iff f ( x ) = ax for all x ∈ R, where a = f (1).

Let us now look at a linear map f : Rm → R. We consider Rm as the space of column

vectors. Let ei := (0, . . . , 0, 1, 0, . . . , 0)t be the standard basic vectors, 1 ≤ i ≤ m. If
x = ( x1 , . . . , xm )t ∈ Rm , we note that x = x1 e1 + · · · + xm em . Since f is linear we see that

f ( x ) = f ( x 1 e1 + · · · + x m e m )
= f ( x 1 e1 ) + · · · + f ( x m e m )
= x 1 f ( e1 ) + · · · + x m f ( e m ) .

If we let ai := f (ei ), we see that f ( x ) = a1 x1 + · · · + am xm . In conclusion, if f : Rm → R

is linear then f is of the from f ( x ) = a1 x1 + · · · + am xm where ai = f (ei ), 1 ≤ i ≤ m.
Conversely, if f : Rm → R is defined by f ( x ) = a1 x1 + · · · + am xm where ai ∈ R are
arbitrary, then f is linear and we have ai = f (ei ), 1 ≤ i ≤ m. We have thus arrived at
the following result.

Lemma 2. Any linear map T : Rm → R is of the form T ( x ) = a1 x1 + · · · + am xm and we

have ai = T (ei ), 1 ≤ i ≤ m. In particular, T is a first degree polynomial in the variables xi ,
1 ≤ i ≤ m, with zero constant term.

Recall the standard dot product on Rm : If x, y ∈ Rm , then x · y = x1 y1 + · · · + xm ym .

Using this dot product, we can reformulate the last lemma as follows.

Proposition 3. Let T : Rm → R be a linear map. Then there exists a unique vector a ∈ Rn

such that for all x ∈ Rm we have T ( x ) = x · a. The vector a = ( a1 , . . . , am ) is prescribed by
ai = T (ei ), 1 ≤ i ≤ m.

Let us look at the projection map πi : Rm → R defined by πi (( x1 , . . . , xm )t ) = xi ,

1 ≤ i ≤ m. It is easy to check that it is linear. By Lemma 2, we know that it is of the form
πi ( x ) = ∑k ak xk . What are ak ’s? Note that ak = 0 if k 6= i and ai = 1. Let us formulate
this as a lemma.

Lemma 4. Let the projection map πi : Rm → R be(defined by πi (( x1 , . . . , xm )t ). Then πi

1 k=i
is linear and we have πi ( x ) = ∑k ak xk where ak = In terms of the dot-product,
0 k 6= i.
πi ( x ) = x · ei , 1 ≤ i ≤ m.

2
 We claim that the composition of linear maps is linear.

Lemma 5. Let U. V and W be vector spaces. Let A : U → V and B : V → W be linear maps.

Then composition B ◦ A : U → W is a linear map.

Proof. This is an easy verification and is left to the reader.

Now, let a linear map T : Rm → Rn be given. Do you know how to generate n-linear
maps Ti : Rm → R? Lemmas 4-5 give us a way of generating such maps. How about
Ti := π ◦ T : Rm → R? We know how to express these maps Ti : Rm → R. They are
given by
Ti x := Ti ( x ) = ai1 x1 + ai2 x2 + · · · + aim xm , where aij ∈ R. (1)
But we wanted to know Tx! Observe that if we write Tx = y, then to know y is the
same as knowing its coordinates yi , 1 ≤ i ≤ n. That is easily achieved. Note that
yi = πi (y) = πi ( Tx ) = (πi ◦ T )( x ). Hence

y = Tx = ((π1 ◦ T )( x ), (π2 ◦ T )( x ), . . . , (πn ◦ T )( x ))t

= ( T1 ( x ), T2 ( x ), . . . , Tn ( x ))t
= ( a11 x1 + a12 x2 + · · · + a1m xm , . . . , an1 x1 + an2 x2 + · · · , anm xm )t (2)
 
  x1
a11 a12 . . . a1m  
 ..
= . .. . . ..   x2  = Ax, say. (3)
.
. .   ... 

an1 an2 . . . anm
xm

Observation 6. If T : Rm → Rn is a linear map, then we found the following:

(i) For x ∈ Rm , the vector Tx = ( p1 ( x ), . . . , pn ( x ))t where each pi is a first degree
polynomial with zero constant term.
(ii) Each pi is given by pi ( x ) = ai1 x1 + · · · + aim xm where aij = Ti (e j ) = πi ( Te j ) =
( Te j ) · ei .
(iii) There is an (n × m)-matrix A such that Tx = Ax where x is considered as a matrix
of size m × 1.

Conversely, if A is an (n × m)-matrix and if we define Tx := Ax for x ∈ Rm , then T

is a linear map from Rm to Rn . This follows from the properties of the matrix multipli-
cation. Thus we have proved the following theorem.

Theorem 7. A map T : Rm → Rn is linear iff there exists (n × m)-matrix A such that Tx =

Ax.

There are a few more observations that bring out the ‘geometric perspective’ to a
matrix. We already saw the geometric meaning of the rows. The entries of the i-th row

3
are the coefficients of the Ti x ∈ Rn expressed as a linear combination of the standard
basic vectors of Rn . The columns also admit a geometric meaning. Look at Tei = Aei .
This is nothing other than the i-th column of A.
The matrix A is unique: If Tx = Ax = Bx, then ( A − B)( x ) = 0 for all x ∈ Rm . In
particular, if we take x = ei , then the i-th column of the matrix A − B is the zero vector
in Rn , 1 ≤ i ≤ n. Thus each column of A − B is the zero vector in Rn and hence A − B
is the zero matrix. Hence, A = B.
The matrix A in (3) is known as the matrix of the linear map T. Why is the last
paragraph relevant here?
An observation: Note that the proofs of Lemmas 1-2 indicate the following result.

Theorem 8. Let V be a finite dimensional vector space and let W be a vector space. Let T : V →
W be linear. Let {v1 , . . . , vn } be a basis of V. If we “know” Tvi , say, Tvi = wi , 1 ≤ i ≤ n, then
we “know” Tv for any v ∈ V. If v = ∑i ai vi , then Tv = ∑i ai wi .

There is a ‘converse’ to the last result.

Theorem 9. Let V be a finite dimensional vector space and let W be a vector space. Fix a basis
{vk : 1 ≤ k ≤ n} of V. Let w1 , w2 , . . . , wn be a finite sequence of vectors in W. (Note that w j ’s
need not be distinct.) Define T : V → W by setting
n
Tv = a1 w1 + a2 w2 + · · · + an wn , where v = ∑ ak vk . (4)
k =1

Then T is a linear map from V to W.

Proof. If u, v ∈ V and if u = ∑k ak vk and v = ∑k bk vk , note that v + w = ∑k ( ak + bk )vk

and for λ ∈ R, λv = ∑k (λak )vk . Hence we have

T (u + v) = ∑(ak + bk )wk = ∑ ak wk + ∑ bk wk = Tu + Tv
k k k
T (λv) = ∑(λak )wk = λ(∑ ak wk ) = λTv.
k k

Hence T is linear.

What is significant in the last theorem is that we can “prescribe” arbitrary vectors
(from W) to Tvk and “extend T linearly” by (4) to any vector v. Theorems 8 and 9 sub-
stantiate the following important take-way: when dealing with a linear map T : V → W,
we should not be concerned with the explicit expressions for T in terms of “coordi-
nates”, but we should focus on its action on a (conveniently chosen) basis.

4
 To appreciate this, carry out the following exercise: Construct a linear map from R2
to R3 and another from R3 to R2 and express them in terms of coordinates! Use (u, v)
as coordinates for R2 and ( x, y, z) as coordinates for R3 .
Observation 6-(i) suggests we may define A : R3 → R2 and B : R2 → R3 as follows:

A( x, y, z) = ( p1 ( x, y, z), p2 ( x, y, z)) = (3x − 4y + 5z, − x + 2y + z), say. (5)

B(u, v) = (q1 (u, v), q2 (u, v), q3 (u, v)) = (3u + 2v, u − v, u + v), say. (6)

Theorems 8 and 9 suggest to define A : R3 → R2 , we choose arbitrary vectors w1 , w2 , w3 ∈

R2 and set Aei = wi , 1 ≤ i ≤ 3. For instance, take w1 = (3, −1), w2 = (−4, 2), w3 =
(5, 1). Then

A( x, y, z) = xAe1 + yAe2 + zAe3 = xw1 + yw2 + zw3 = (3x − 4y + 5z, − x + 2y + z).

Voilà! We got the expression for A in (5). Can you carry out a similar exercise for B?
That is, can you define Be1 , Be2 ∈ R3 suitably so that you get the expression for B in (6)?
Here e1 = (1, 0) and e2 = (0, 1) are the standard basis of R2 .

Remark 10. If I remember correctly, this article was written after recording the following
video. Though it is not an exact transcription, this article captures the spirit of the
lecture! Linear Maps RmtoRn: https://youtu.be/njIqG4-aQr4