3.

RANK OF A MATRIX

3.1 LINEAR DEPENDENCE AND LINEAR INDEPENDENCE OF ROWS (COLUMNS) OF A MATRIX

In the following, we will call matrix-columns (matrix rows) simply columns (rows) and denote
them by lowercase letters. Columns are equal if they have the same size and all the corresponding
elements are equal.
Column A is called a linear combination of columns A1 , A2 ,…, Ak of the same size, if
A = a1 × A1 + a 2 × A2 + ... + a k × Ak , (3.1)
where a1 , a2 ,…, a k are arbitrary numbers. In that case we say that column A is decomposed into
columns A1 , A2 ,…, Ak , and numbers a1 , a2 ,…, a k are called the decomposition coefficients. A linear
combination A = 0 × A1 + + 0 × A2 + ... + 0 × Ak , where all the coefficients are equal to zero, is called trivial.
If the columns in (3.1) are given by
æ a1 ö æ a11 ö æ a1k ö
ç ÷ ç ÷ ç ÷
A = ç ! ÷ , A1 = ç ! ÷ , … , Ak = ç ! ÷ ,
ça ÷ ça ÷ ça ÷
è nø è n1 ø è nk ø

then the matrix equality (3.1) can be expressed in a form of element-to-element equalities
ai = a1 × ai1 + a 2 × ai 2 + ... + a k × aik , i = 1,..., n .

A linear combination of rows of the same size is defined in a similar way.

A set of columns A1 , A2 ,…, Ak of the same size is called a system of columns.
Any part of a system of columns system is called a subsystem. 1
 A system of k columns A1 , A2 ,…, Ak is called linearly dependent, if there exist such
numbers a1 , a 2 ,..., a k , not all equal to zero, that
a1 × A1 + a 2 × A2 + ... + a k × Ak = o . (3.2)

Hereinafter symbol o will denote a zero column of a corresponding size.

A system of k columns A1 , A2 ,…, Ak is called linearly independent, if equation (3.2) is
correct only if a1 = a 2 = ... = a k = 0 , i.e. when the linear combination on the left side of equation
(3.2) is trivial.
One column A1 composes a system as well: for A1 = o the system is linearly dependent, and
for A1 ¹ o - linearly independent.
For rows (row matrices) we get similar definitions.

2
 Example 3.1. By definition, determine linear dependence or linear independence of systems
of columns:
æ1ö æ 2ö æ1ö æ0ö
а) A1 = çç ÷÷ , A2 = çç ÷÷ ; b) A1 = çç ÷÷ , A2 = çç ÷÷ .
è 0ø è0ø è 0ø è 2ø
æ1ö æ 2ö
o а) Columns A1 = çç ÷÷ and A2 = çç ÷÷ are linearly dependent, because we can compose a non-
è 0ø è0ø
trivial linear combination, e.g., with coefficients a1 = 2 , a 2 = -1 , which is equal to a zero column:
æ 1ö æ 2ö æ 0ö
2 × çç ÷÷ - 1 × çç ÷÷ = çç ÷÷ ;
è 0ø è 0ø è 0ø

æ1ö æ0ö
b) columns A1 = çç ÷÷ and A2 = çç ÷÷ are linearly independent, because the equality
è 0ø è 2ø
æ 1ö æ 0ö æ 0ö ì1× a = 0
a1 × çç ÷÷ + a 2 × çç ÷÷ = çç ÷÷ , that matches the system í 1 , is correct only when a1 = a 2 = 0 .n
è 0ø è 2ø è 0ø î2 × a 2 = 0

3
 Properties of linearly dependent and linearly independent columns

The concepts of linear dependence and linear independence are defined for rows and
columns in a similar manner. Hence, the properties of linear dependence and independence, given
for columns, will be true for rows as well.
1. If there is a zero column in a system of columns, this system is linearly dependent.
2. If there are two equal columns in a system of columns, this system is linearly dependent.
3. If there are two proportional columns ( Ai = lA j ) in a system of columns, this system is

linearly dependent.
4. A system of k > 1 columns is linearly dependent when and only when at least one of the
columns is a linear combination of the others.
5. Any columns that are included in a linearly independent system, compose a linearly
independent subsystem.
6. A system of columns that contains a linearly dependent subsystem, is linearly dependent.
7. If a system of columns A1 , A2 ,…, Ak is linearly independent, but after the addition of
column A becomes linearly dependent, then column A can be uniquely decomposed into columns
A1 , A2 ,…, Ak , i.e. decomposition coefficients are single-valued.

8. Two nonzero columns A1 , A2 compose a linearly dependent system, if they are

4
proportional ( A1 = lA2 ), and a linearly independent system, if they are not proportional.
 3.2. BASIS MINOR AND MATRIX RANK

Basis minor of a matrix. Computing the rank of a matrix

Let A be a m´ n matrix, and k - a natural number not greater than m and n : k £ min{ m; n } .
A minor of order k of matrix A is the determinant of a matrix of the k -th order, composed
of the elements at the intersection of k arbitrarily chosen rows and k arbitrarily chosen columns
of A .
Denoting minors, we will write the numbers of the chosen rows as superscripts and the
numbers of the chosen columns as subscripts, in ascending order.

Example 3.3. Write down minors of different orders of the following matrices:
æ1 2 1 0ö
æ 1 2 3ö ç ÷
а) A = çç ÷÷ ; b) B = ç 0 2 2 3 ÷ .
è 4 5 6 ø ç1 4 3 3÷
è ø
o а) Matrix A of sizes 2´ 3 has six minors of the first order, for example, M 21 = det (a12 ) = 2 ,
12 2 3
and three minors of the second order, for example, M 23 = = -3 .
5 6
b) Matrix B of size 3´ 4 has 12 minors of the first order, e.g. M 23 = det (b32 ) = 4 , and 18 minors
12 2 1
of the second order, e.g. M 23 = =2, and four minors of the third order, e.g.
2 2
1 1 0
123
M 134 = 0 2 3 =0. n
1 3 3 5
 Let A be a m´ n matrix. A minor of A of order r is called basis, if it is nonzero and all
minors of order (r + 1) are equal to zero or do not exist.
The rank of a matrix is the order of its basis minor.
A zero matrix doesn’t have a basis minor. Thus, the rank of a zero matrix is, by definition,
equal to zero.

The rank of a matrix A is denoted by rg A . It can also be denoted by Rg A , rang A , rank A .

If all minors of order k of a matrix are equal to zero, all minors of higher order are also
equal to zero.

The rank of a matrix equals to the largest order of any nonzero minor of this matrix.

If a square matrix is nonsingular, its rank is equal to its order. If a square matrix is singular,
its rank is less than its order.

The rank of a block matrix is computed as the rank of an ordinary (numerical) matrix, i.e.
without paying attention to the block structure. In addition, the rank of a block matrix is not less
than the ranks of its blocks: rg ( A B) ³ rg A and rg (A B) ³ rg B .

6
54 3.2. БАЗИСНЫЙ МИНОР И РАНГ МАТРИЦЫ

Examle 3.4. Find all basis minors and ranks of the following matrices:
æ1 2ö
æ 0 0ö æ1 2ö æ 1 2 3ö ç ÷
а) O = çç ÷÷ ; b) A = (0 0 1) ; c) B = çç ÷÷ ; d) С = çç ÷÷ ; e) D = ç 0 0 ÷ ;
è 0 0ø è0 0ø è 2 4 6ø ç1 3÷
è ø
o а) Matrix O is zero, so all of its minors are equal to zero. A zero matrix doesn’t have any
basis minors, and it’s rank equals to zero by the definition: rg O = 0 .
b) One of the first-order minors of matrix A = (0 0 1) is nonzero: M 31 = 1 , and minors of the
second order don’t exist (since there’s only one row). Hence, the minor M 31 is basis and the rank of this
matrix is equal to 1.
æ1 2ö
c) For matrix B = çç ÷÷ , there are nonzero minors of the first order: M11 = 1 и M 21 = 2 . These
è0 0ø
12 1 2
minors are basis, because the only minor of the second order M 12 = is equal to zero. Hence,
0 0
rg B = 1 .
æ 1 2 3ö
d) All first-order minors minors of the first order of matrix С = çç ÷÷ , which are equal to its
è 2 4 6 ø
elements, are nonzero, and all the minors of the second order are equal to zero, because rows of the
matrix are proportional. Thus, the matrix has six basis minors and its rank is equal to 1.
13 1 2
e) Matrix D has a nonzero minor of the second order M 12 = =1, and minors of the third
1 3
13
order don’t exist (since there’re only two columns). Hence, M 12 – is the only basis minor and rg D = 2 .

7
 æ 1 2 3ö æ1 2 3ö æ1 2 2 0ö
ç ÷ ç ÷ ç ÷
f) F = ç 2 4 5÷ ; g) G = ç0 4 5÷ ; h) H = ç 0 2 2 3÷ .
ç 1 2 3÷ ç0 0 6÷ ç 0 0 0 0÷
è ø è ø è ø

12 12 12
f) Matrix F has six nonzero minors of the second order: M 12 , M13 , M 23 , M1223 , M1323 , M 23
23
,
and the only third-order minor, i.e. the determinant, is equal to zero, since matrix has two equal
rows (the first and the third ones). Hence, each of the mentioned minors of the second order is
basis and the rank of the matrix is equal to 2.

g) The determinant of matrix G (i.e. the minor of the third order) is nonzero:
123 123
det G = M123 = 1× 4 × 6 ¹ 0 . Hence, the minor M 123 is basis and rg G = 3 .

h) All third-order minors of this matrix are equal to zero, because the third row of these
minors is zero. So, only a minor of the second order, situated in the first two rows of the matrix,
can be basis. Searching through the six possible minors, we choose the nonzero ones:
12 12 1 2 12 12 2 0 12 1 0
M12 = M13 = , M 24 = M 34 = and M14 = .
0 2 2 3 0 3
Each of these five minors is basis. Hence, the rank of this matrix is equal to 2. n

8
 Properties of basis minor and matrix rank

1. In an arbitrary matrix A every row (column) is a linear combination of rows (columns),

in which the basis minor is situated.
2. The determinant is equal to zero if and only if one of its rows (columns) is a linear
combination of other rows (columns).
3. Applying elementary transformations to a matrix does not change its rank.
4. If a row (column) of a matrix is a linear combination of other rows (columns) of this
matrix, this row (column) can be deleted from the matrix without changing its rank.
5. If a matrix is reduced to the simplest form (1.3), then
rg A = rg L = r .
6. The rank of a matrix is equal to the maximum number of linearly independent rows of this
matrix.
7. The maximum number of linearly independent rows of a matrix is equal to the maximum
number of linearly independent columns:
rg A = rg AT .
8. Elementary row transformations preserve linear dependence (or linear independence) of
any system of columns of this matrix.
9. The rank of a matrix product is not larger than the ranks of factors:
rg( AB) £ min{ rg A, rg B } .

10. If A is a nonsingular square matrix, then rg ( AB) = rg B and rg (CA) = rg C , i.e. the rank of
a matrix does not change after multiplying it from the left or from the right by nonsingular square
matrix.
11. The rank of a sum of matrices is not larger than the sum of the ranks of summands:
rg( A + B) £ rg A + rg B . 9
 3.3. METHODS FOR COMPUTING MATRIX RANK

3.3.1. METHOD OF BOUNDING MINORS

Let A be an m´ n matrix. We will say that minor M ij11i2j2......ikjikk +j1k +1 of order (k + 1) bounds
(contains) minor M ij11i2j2......ikjk of order k .
Describing the method, we will write down indices of the chosen rows and columns, in
which the minor is situated, without putting them in ascending order. In so doing, the minor at issue
and the minor with indices put in order have equal absolute value and, maybe, are of different signs,
but it is of no importance for the method of bounding minors, because we only want to find out the
answer to the question: is the minor equal to zero or not.
1. Choose row i1 and column j1 , so that the minor of the first M ij11 = ai1 j1 is nonzero. If it is
possible, then rg A ³ 1 , else the process terminates and rg A = 0 .
2. Bound the minor M ij11 ¹ 0 by adding another row i2 ¹ i1 and another column j2 ¹ j1 to the
ai1 j1 ai1 j 2
chosen i1 -th row and the j1 -th column, so that the minor M ij11i2j2 = ¹ 0 . If it is possible,
ai2 j1 ai2 j 2
then rg A ³ 2 , else the process should terminates and rg A = 1 .
3. Bound the minor M ij11i2j2 ¹ 0 by adding to the previously chosen rows and columns another
row i3 and another column j3 in order to get the minor M ij11i2j2i3j3 ¹ 0 . If it is possible, rg A ³ 3 , else the
process terminats and rg A = 2 .
Continue the bounding process until it is terminated. Suppose we have found a nonzero
minor of order r : M ij11i2j2......irjr ¹ 0 , i.e. rg A ³ r . But all the minors of order (r + 1) , bounding it, are equal
to zero M ij11i2j2......irjirr +j1r +1 = 0 or do not exist (for r = m or r = n ). Then the process terminates and rg A = r . 10
 Example 3.5. Find ranks of matrices, using the method of bounding minors
æ1 0 2 1 3ö
æ 0 0ö æ3 9ö æ 0 2 3ö ç ÷
O = çç ÷÷ , A = çç ÷÷ , B = çç ÷÷ , С = ç 2 0 1 1 2÷ .
è 0 0ø è 2 4ø è 2 4 6ø ç 3 0 3 2 5÷
è ø

o Matrix O : 1. This matrix does not have any nonzero minors of the first order, because all of its elements are equal
to zero. Hence, rg O = 0 .
Matrix A : 1. Choose the first row ( i1 = 1 ) and the first column ( j1 = 1 ) of matrix A , at the intersection of which there is
a nonzero element a11 = 3 ¹ 0 . We have minor M11 = 3 ¹ 0 . Hence, rg A ³ 1 .
2. Add another row i2 = 2 and another column j2 = 2 to the previously chosen ones. We have a nonzero minor of the
3 9
second order M1122 = det A = =-6¹0. Hence, rg A ³ 2 .
2 4
3. Since we have used all the rows and columns of matrix A , there are no minors bounding M1122 ¹ 0 . Hence, rg A = 2 .
Matrix B : 1. Choose the first row and the second column of matrix at the intersection of which there is a nonzero
element b12 = 2 ¹ 0 . We have a minor M 21 = 2 ¹ 0 . Hence, rg B ³ 1 .
2. Add the second row and the third column to the previously chosen ones. We have a minor of the second order
2 3
M 21 23 = = 0 . The choice was unsuccessful, because we have got a zero minor. Let’s take the first column instead of the
4 6
2 0
third one. Then we have a nonzero minor of the second order M 21 12 = =4¹0. Hence, rg B ³ 2 .
4 2
3. We have used all the rows of matrix B . There are no minors of the third order, thus rg B = 2 .

11
 æ1 0 2 1 3ö
ç ÷
С = ç 2 0 1 1 2÷
ç 3 0 3 2 5÷
è ø

Matrix С : 1. Choose the first row ( i1 = 1 ) and the first column ( j1 = 1 ) of matrix С , at the
intersection of which there is a nonzero element a11 = 1 ¹ 0 . We have minor M11 = 1 ¹ 0 . Hence,
rg C ³ 1 .
2. Add another row i2 = 2 and another column j2 = 2 to the previously chosen ones. We
1 0
have a minor of the second order M 1122 = . Choosing the second column was unsuccessful,
2 0
because we have got a zero minor. Let’s choose the third column ( j2 = 3 ) instead. Then we have a
1 2
nonzero minor M1132 = = -3 ¹ 0 . Hence, rg C ³ 2 .
2 1
3. Bound minor M 1132 ¹ 0 . There are three bounding minors:
1 2 1 1 2 3 1 2 0
M 113243 = 2 1 1 = 0 , M 113253 = 2 1 2 = 0 , M 113223 = 2 1 0 = 0 .
3 3 2 3 3 5 3 3 0

All three determinants are equal to zero, since the third row is a sum of the first two. Thus, it’s
impossible to find a nonzero minor of the third order, i.e. the rank of matrix С is equal to 2. n

12
 3.3.2. GAUSSIAN ELIMINATION METHOD

Let A be an m´ n matrix. To calculate its rank we need to make the following steps:

1. Reduce the matrix to echelon form (see Gauss method in section 1.2.6).
2. Calculate the number r of nonzero rows of the obtained matrix. This number is equal to
the rank of matrix A .

This method is based on Property 8 (see section 3.2). The basis minor of a matrix in
echelon form (1.1) is a minor
1 * ! *
0 1 ! *
M = ,
" " # "
0 0 ! 1

composed of columns containing unity elements (at the beginning of each “step”). This determinant
of triangular form is nonzero (equals to 1), and each of its bounding minors (if it exists) is equal to
zero, because it contains a zero row.

13
 Example 3.6. Find ranks of matrices, using Gaussian elimination method
æ1 0 2 1 3ö
æ 0 0ö æ3 9ö æ 0 2 3ö ç ÷
O = çç ÷÷ , A = çç ÷÷ , B = çç ÷÷ . С = ç 2 0 1 1 2÷ .
è 0 0ø è 2 4ø è 2 4 6ø ç 3 0 3 2 5÷
è ø
o Matrix O : 1. A zero matrix is already in echelon form (see definition in section 1.2.6).
2. A number of nonzero rows is equal to zero. Hence, rg O = 0 .
Matrix A : 1. Reduce matrix A to echelon form (see example 1.18):

æ 1 3ö
A ~ çç ÷÷ .
è 0 1ø
2. There are two nonzero rows in this matrix. Hence, rg A = 2 .
Matrix B : 1. Reduce matrix B to echelon form (see example 1.18):
æ1 2 3 ö
B ~ çç ÷÷ .
è 0 1 1,5 ø
2. There are two nonzero rows in this matrix. Hence, rg B = 2 .
Matrix C : 1. Reduce matrix C to echelon form. Choose a11 = 1 as a pivot and make all the
other elements of the first column equal to zero: add the first row, multiplied by (- 2) , to the second
row, and the first row, multiplied by (- 3) , to the third one. We get matrix
æ1 0 2 1 3ö æ1 0 2 1 3 ö
ç ÷ ç ÷
С = ç 2 0 1 1 2÷ ~ ç 0 0 - 3 -1 - 4÷ ,
ç 3 0 3 2 5÷ ç 0 0 - 3 -1 - 4÷
è ø è ø
that has two equal rows. By Property 4 (see section 3.2), we delete one of the equal rows:
æ1 0 2 1 3 ö
çç ÷÷ .
è 0 0 - 3 -1 - 4ø
We got an echelon form of the matrix.
14
2. There are two nonzero rows in this matrix. Hence, rg С = 2 .
 æ1 2 3ö
ç ÷
ç0 0 0÷
D = ç2 1 3÷
ç ÷
ç1 1 2÷
ç3 2 5 ÷ø
è

Matrix D : 1. Reduce matrix D to echelon form. We delete the zero row and choose a11 = 1 as a
pivot element to make all other elements of the first column equal to zero:
æ1 2 3ö æ 1 2 3 ö
ç ÷ ç ÷
ç2 1 3÷ ç 0 - 3 - 3÷
D~ç ~ .
1 1 2÷ ç0 -1 -1÷
çç ÷ ç ÷
è3 2 5 ÷ø çè 0 - 4 - 4 ÷ø

The last three rows of the matrix are proportional. By Property 4 (see section 3.2) we can delete two
of them:
æ1 2 3 ö
çç ÷÷ .
è 0 - 1 - 1 ø
We have got an echelon form of the matrix.
2. There are two nonzero rows in this matrix. Hence, rg D = 2 .
Note, that rg C = rg D , because D = C T by Property 7 (see section 3.2). n

15
 INDIVIDUAL TASK

1. Calculate ranks of the matrices:

æ 1 2 1 2ö
æ1 m n m ö ç ÷
ç ÷ m n m n÷
а) ç 2 1 1 n ÷; b) çç
ç3 m +1 n +1 m + n÷ -m n - m n÷
è ø çç ÷
è 1 2 1 2 ÷ø
using the methods of bounding minors and Gaussian elimination, converting matrices into echelon
form.

16