Algebra Problem Sheet 04

Note: Vector spaces are over an arbitrary field F, unless otherwise stated.
Warmup questions
1. Find the column and the row ranks of the following matrix and confirm that they are equal. Write
A = BC, in two different ways, with B and C of maximal rank.
 
1 2 3
A = 1 2 −1
2 4 2

Since (2, 2, 4) = 2(1, 1, 2), while (3, −1, 2) is not a multiple of (1, 1, 2), we see that a basis for the
column space is (1, 1, 2), (3, −1, 2) so the column rank is 2.
Since (1, 2, −1) is not a multiple of (1, 2, 3), while (2, 4, 2) = (1, 2, 3) + (1, 2, −1), a basis for the row
space is (1, 2, 3), (1, 2, −1) so the row rank is 2.
We can find such a factorisation either by using the column basis as the columns of B, with colk (C)
being the coordinates of colk (A) in this basis, or the row basis as the rows of C, with rowi (B) being
the coordinates of rowi (A) in this basis. Thus
   
1 3   1 0  
1 2 0 1 2 3
A = 1 −1 = 0 1
0 0 1 1 2 −1
2 2 1 1

2. Let U , V , W be finite-dimensional and φ : U → V and ψ : V → W be linear maps.

(a) Show that rank(ψ ◦ φ) ≤ min{rank ψ, rank φ}.
(b) If φ is surjective, show that rank(ψ ◦ φ) = rank ψ.
(c) If ψ is injective, show that rank(ψ ◦ φ) = rank φ.

(a) First note that φ(U ) is a subspace of V and is itself a vector space as well. So ψ(φ(U )) is a subspace
of W and is contained in ψ(V ). Therefore ψ(φ(U )) is a subspace of ψ(V ). Hence

rank ψφ = dim ψ(φ(U )) ≤ dim ψ(V ) = rank ψ.

On the other hand, applying the Rank-Nullity-Theorem to the restriction

ψ|φ(U ) : φ(U ) → ψ(φ(U ))

gives dim ψ(φ(U )) ≤ dim φ(U ) and so

rank ψφ = dim ψ(φ(U )) ≤ dim φ(U ) = rank φ.

(b) If φ is surjective, then Im φ = V and hence Im(ψ ◦ φ) = ψ(V ) = Im ψ, so

rank ψφ = rank ψ.

(c) If ψ is injective, then, the restriction ψ|φ(U ) : φ(U ) → ψ(φ(U )) is an isomorphism. In particular,
ker ψ|φ(U ) = 0. So by Rank-Nullity-Theorem, dim ψ(φ(U )) = dim φ(U ). Therefore,

rank ψφ = rank φ.

3. P vn and β : w1 , . . . , wn be bases of V and W , respectively. Show that the map φ : V → W ,

Let α : v1 , . . . ,P
n n
λ
i=1 i iv →
7 i=1 λi wi is well-defined and is an isomorphism. (1) As α is a basis, for any v ∈ V ,
the linear combination
Xn
v= λi vi
i=1

is unique, that is, λ1 , . . . , λn are uniquely determined by v. So v is mapped to a unique vector

n
i=1 λi wi . Therefore, φ is well defined.
P

1
 Pn Pn
(2) We show that φ is linear. For any v = λi vi and u = i=1 µi vi , λ, µ ∈ F, we have
i=1
Pn
φ(λv + µu) = φ(P i=1 (λλi + µµi )vi )
n
= Pn i=1 (λλi +Pµµ i )wi
n
= i=1 λλi wi + i=1 µµi wi
= λφ(v) + µφ(u).
So φ is linear.
Pn
(3) We show that φ is an isomorphism. Suppose that φ(v) = 0 and v = i=1 λi vi . We have
n
X
0 = φ(v) = λi wi .
i=1

As β is a basis of W , λi = 0, ∀i. Therefore v = 0 and so ker φ = {0}, which implies that φ is injective.
Moreover, by the R-N-Theorem, dim Im φ = n = dim W . As Im φ ≤ W , by Corollary 1.4.6,
Im φ = W.
Therefore φ is also surjective and thus an isomorphism, as required.
Homework questions
4. Find the ranks of the following matrices, by finding a basis for either the row space or the column
space. Write each A = BC with B and C of maximal rank.
   
1 0 2 −1 1 2 1 0
(a) A = 2 0 4 −2 (b) A =  2 5 5 1
2 0 4 −2 −2 −3 0 3
(a) A basis for the row space is (1, 0, 2, −1) as every other row is a multiple of this. A basis for the
column space is (1, 2, 2) as every other column is a multiple of this. Thus the (row/column) rank is
1 and  
1

A = 2 1 0 2 −1
2
(b) Observe that neither of col1 (A), col2 (A) is a multiple of the other one. So the two column vectors
col1 (A), col2 (A) are L. I. Next solve the linear system A0 x = col3 (A), where A0 is the matrix
consisting of the first two columns of A.
     
1 2 1 R2−2R1 1 2 1 1 2 1
R3+2R1 R3−R2
 2 5 5  −→  0 1 3  −→  0 1 3 
−2 −3 0 0 1 2 0 0 −1
This shows that A0 x = col3 (A) is inconsistent and so col3 (A) is not linearly spanned by col1 (A),
col2 (A). Therefore col1 (A), col2 (A), col3 (A) is L. I. and thus a maximal L. I. list, i.e. a basis, since
the column space is a subspace of the 3-dimensional space F3 . So rank A = 3. (In this case the
column space is the same as F3 .)
Next compute the coordinate vector of col4 (A) with respect to the basis col1 (A), col2 (A), col3 (A)
by solving    
1 2 1 0
2 5 5 x = 1
−2 −3 0 3
as follows
     
1 2 1 0 1 2 1 0 1 0 0 −12
 2 5 5 1  −→  0 1 3 1  −→  0 1 0 7  (∗).
−2 −3 0 3 0 0 1 −2 0 0 1 −2
So the coordinate vector of col4 (A) is (−12, 7, −2) and we have the decomposition,
  
1 2 1 1 0 0 −12
A= 2 5 5 0 1 0 7 .
−2 −3 0 0 0 1 −2

2
 Since this means that A itself has maximal rank, the simplest factorisation of the required sort is
actually A = IA, which is what would be obtained by starting from all the rows as a basis of the
row space, that is, C = A.

Another way to see why col1 (A), col2 (A) and col3 (A) form a basis of the column space of
A. By Question 2.5, if φ : V → W is an isomorphism, then v1 , . . . vn are L. I. in V if and only if
φ(v1 ), . . . , φ(vn ) are L. I. in W . Note that

coli (P A) = φP (coli (A)) = P coli (A)

and matrices of EROs are invertible. So from the computation in (*), where the first 3 columns of
the right most matrix form a maximal L. I. list (i.e. a basis) in its column space. So the first 3
columns of A form a basis of the column space of A.
5. Suppose that V , W are vector spaces over F with dim V = n, dim W = m and that φ : V → W is
a linear map of rank r. Show that one can choose bases for V and W with respect to which φ is
represented by the matrix Dr = (dij ) ∈ Mm,n (F), given by
(
1 if i = j ≤ r,
dij =
0 otherwise.

Deduce that, if matrices A, B ∈ Mm,n (F) satisfy rank A = rank B, then they are equivalent, i.e.
B = P −1 AQ for invertible matrices P, Q. If the bases are v1 , . . . , vn for V and w1 , . . . , wm for W ,
then for φ to be represented by Dr it is necessary and sufficient that φ(vi ) = wi , for i ≤ r, and
φ(vi ) = 0, for i > r.
To find bases with this property we first choose vr+1 , . . . , vn (for some r) to be a basis of Ker φ and
extend it to a basis v1 , . . . , vn for V . Then set wi = φ(vi ), for i = 1, . . . , r. From the proof of the
Rank-Nullity-Theorem, we know that w1 , . . . , wr is a basis for Im φ, so that indeed r = rank φ. We
can then extend this to a basis w1 , . . . , wm for W , which has the required property.
If we apply this result to the linear maps φA , φB : Fn → Fm , then we will deduce that there are
invertible matrices P1 , Q1 , P2 , Q2 (whose columns are the respective bases) such that Dr = P1−1 AQ1
and also Dr = P2−1 BQ2 , where r = rank A = rank B. Then B = P −1 AQ, with P = P1 P2−1 and
Q = Q1 Q2−1 . That is, A and B are equivalent.
6. Give an example showing that the inequality in Q 2 (1) can be strict, using 2 × 2 matrices. That is,
find 2 × 2 matrices A, B such that rank(AB) < min{rank A, rank B}.
   
1 0 0 0
Let A = and B = . Then, rank A = rank B = 1, AB = 0 and so rank AB = 0, which
0 0 1 0
is strictly less than min{rank A, rank B}.
7. Let V be the vector space of all n × n matrices. What is the dimension of V ?
(a) For any A ∈ V define a map LA : V → V : B 7→ AB. Show that LA is linear.
(b) Suppose that AX = I, for some X ∈ V . Show that LA is surjective and deduce, from the Rank-
Nullity-Theorem, that it is therefore also injective.
(c) Compute LA (XA − I) and deduce that XA = I.
An n × n matrix is freely determined by its n2 entries in the coefficient field F, so the dimension of V
is n2 . In particular, this is finite.
(a) LA (λ1 B1 + λ2 B2 ) = A(λ1 B1 + λ2 B2 ) = λ1 AB1 + λ2 AB2 = λ1 LA (B1 ) + λ2 LA (B2 ), for all λ1 , λ2 ∈ F
and B1 , B2 ∈ V . Thus LA is linear.
(b) If AX = I, then for any Z ∈ V , we have LA (XZ) = AXZ = IZ = Z. Thus LA is surjective and
hence has rank n2 and nullity 0, by the Rank-Nullity-Theorem. Thus ker LA = {0} and so LA is
injective. (Note: this means that LA is an isomorphism.)
(c) LA (XA − I) = A(XA − I) = AXA − A = IA − A = 0. Hence, since LA is injective, XA − I = 0,
i.e. XA = I.
Additional questions

3
8. Let S1 , . . . , Sm ⊂ {1, . . . , n} be distinct subsets such that each contains an even number of elements,
while the intersection Si ∩ Sj of any distinct pair of sets contains an odd number of elements. Show
that m ≤ n. Show also that if m = n, then n is odd. Let F2 be the field with two elements, and define
A ∈ Mm,n (F2 ) by Aij = 1 if j ∈ Si and 0 otherwise. The hypotheses imply
• A(1) = 0 (where 1 ∈ Fn2 is the vector all of whose coefficients are 1).
• All diagonal entries of the m × m matrix AAT are 0, while all off-diagonal entries are equal to 1.
Part 1. implies that the kernel of A is non-trivial, so rank A ≤ n − 1. On the other hand, Part 2.
means that AAT x = 0 if and only if each entry of x equals the sum of all entries of x. Thus
• if m is even then AAT has trivial kernel, so rank AAT = m,
• if m is odd then ker AAT = {0, 1} (of dimension 1) so rank AAT = m − 1.
Because rank AAT ≤ rank A (by Q 2 (i)), we deduce that m ≤ n if m is odd, and m < n if m is even.