Академический Документы
Профессиональный Документы
Культура Документы
Lectures
MODULE 17 MATRICES II
1. Inverse of a matrix
2. Systems of linear equations
3. Solution of sets of linear equations – elimination methods
4. Inverse of a matrix – using elimination method
1. Inverse of a matrix
It is very important to be able to obtain the inverse of a matrix.
Def. Given a square matrix A , if we can find a matrix B such that
AB = BA = I,
where I is the unit matrix, then B is called the inverse of A and written A−1 .
[It can be proved that if the inverse exists then it is unique, i.e. the inverse has only one form.]
Def. The matrix A is singular if |A| = 0 (i.e. detA = 0).
Def. The matrix A is non-singular if |A| 6= 0.
With this notation we can state:
if A is singular then A−1 does not exist;
adj A
if A is non-singular then A−1 does exist and A−1 = .
|A|
The above formula states the direct method (or cofactor method) of calculating the inverse.
Property of inverse Provided the inverses exist, (AB)−1 = B−1 A−1 .
[Proof If the expression on the RHS is the inverse then it must satisfy the rules stated earlier for an inverse.
For verification, note that
The result can be extended to a string of matrices: (ABCD)−1 = D−1 C−1 B−1 A−1 .
5 2 4
1 2
Ex 1. Find A−1 and B−1 when (a) A = , (b) B = 3 −1 2.
3 3
1 4 −3
(a) As discussed in module 16 the minor associated with each element is found by deleting the row and
column which passes through that element, and taking the determinant of what remains. The corresponding
cofactor is then calculated using Aij = (−1)i+j Mij . Hence,
3 3 (−1)1+1 3 (−1)1+2 3 3 −3
(Mij ) = , (Aij ) = = ,
2 1 (−1)2+1 2 (−1)2+2 1 −2 1
|A| = 1(3) − 2(3) = 3 − 6 = −3,
T 3 −2
adj A = (Aij ) = ,
−3 1
1
and therefore
3 −2
−1 −3 1 −1 2/3
A = = .
−3 1 −1/3
Thus finally
−5 22 8
adj B 1
B−1 = = 11 −19 2 .
|B| 49
13 −18 −11
2
In this module only systems which lead to a square matrix for A are considered. There are four cases which
arise depending on whether, or not, |A| = 0 and b = 0 .
For this system b 6= 0 and |A| = 3(2) − 3(2) = 6 − 6 = 0. It is clear from inspection of the pair of equations
that they are inconsistent (i.e. not compatible) so no solution exists.
The second system is
3x + 2y = 2 3 2 x 2
or = .
6x + 4y = 4 6 4 y 4
For this system also b 6= 0 and |A| = 3(2) − 3(2) = 6 − 6 = 0. Looking at the given pair of equations the
second one is twice the first, and can be ignored since it provides nothing new. For solutions of the remaining
3
equation 3x + 2y = 2 choose x = C, any constant, and then 2y = 2 − 3C which implies y = 1 − C .
2
Thus the solution is
x C
= , for any constant C.
y 1 − 32 C
3
Ex 2. Find the value of α for which the equations
αx + y − z = 0
y + 2z = 0
2y − z = 0
and hence non-trivial solutions exist only when α = 0. In this case the system of equations becomes
y− z=0
y + 2z = 0
2y − z = 0
x + 2y = 4 x + 2y = 4
→ (eqn. 2 − 2 × eqn. 1).
2x + y = 5 − 3y = −3
The second reduced equation implies y = 1 and hence substitution into the first reduced equation gives
x = 4 − 2(1) = 2 . The solution, therefore, is
x 2
= .
y 1
The basic elimination method for n linear equations in the n unknowns x1 , x2 , . . ., xn , which extends
the method outlined above, is
(i) retain the first equation for x1 in terms of x2 , . . ., xn and use this equation to eliminate x1 from the
remaining equations;
(ii) retain the second equation for x2 in terms of x3 , . . ., xn and use this equation to eliminate x2 from the
remaining lower equations in the reduced set;
(iii) repeat the process, until you arrive at the final equation in xn only, which you solve;
(iv) substitute back into the equations in the reduced set, using them in reverse order to find in turn
xn−1 , xn−2 , . . ., x1 .
Let us illustrate the method with the following example for 3 equations in 3 unknowns.
4
Ex 3. Use the elimination method to solve the equations
x1 + x2 = 3,
2x1 + x2 + x3 = 7,
x1 + 2x2 + 3x3 = 14.
The
elimination
method described above is equivalent to reducing A to upper triangular form, i.e.
a b c
0 d e with non-zero elements on the principal diagonal and all elements below it being zero. The
0 0 f
elimination procedures rely on manipulation of the ROWS of the matrix, equivalent to manipulation of the
original equations. The allowed row operations arise because of the operations which can be performed on
equations and systems of equations. These row operations are:
(i) any row can be multiplied by a constant;
(ii) a row can be added to (or subtracted from) any other row;
(iii) any two rows can be interchanged.
where we have used row 2 − 2 × row 1 and row 3 − row 1, in order to make the elements in the first column
(apart from the leading element) zero. Note that it is the rows in the matrix of coefficients and the same rows
in the column vector on the RHS that are changed, but the matrix of unknowns is unaltered. This follows
immediately from consideration of the equivalent set of equations. Completing the reduction we obtain
1 1 0 x1 3
0 −1 1 x2 = 1 , row 3 + row 2.
0 0 4 x3 12
x1 + x2 = 3,
− x2 + x3 = 1,
4x3 = 12.
Solving the final equation then implies x = 3 , and substituting into the second equation gives x2 = x3 − 1 =
3 − 1 = 2 . Finally from the first equation we deduce x1 = 3 − x2 = 3 − 2 = 1 . Thus the solution is
x1 1
x2 = 2 .
x3 3
[You should check that this solution satisfies the original set of three equations.]
5
The above elimination method can be modified to apply to special matrices, and the resulting algorithms
can be found in various texts. The basic idea is unchanged, however, and the various specialised algorithms
are not discussed further in this unit.
6
The first and second columns are now correct, so we must move on to the third row and third column. The
third element in this row is already 1, so we only need to use the third row to produce zeros above this
element in the third column
1 0 0 2/7 −5/7 1/7
1
→ 0 1 4/3 1/3 1/3 0 , row 1 − × row 3
3
0 0 1 1/7 1/7 −3/7
1 0 0 2/7 −5/7 1/7
4
→ 0 1 0 1/7 1/7 4/7 , row 2 − × row 3
3
0 0 1 1/7 1/7 −3/7
The matrix A has been changed to I, and the theory says that the right-hand side I will now be A−1 . Hence
2/7 −5/7 1/7
A−1 = 1/7 1/7 4/7 .
1/7 1/7 −3/7
The evaluation of A−1 by the above method appears lengthy but for large matrices it can be shown that it
is more efficient than determining the inverse using the cofactor method, and algorithms can be written for
computer implementation.
Before finishing this module another issue is briefly considered through the following example.
As you see the answers to (a) and (b) are hugely different despite the fact that the coefficients in the original
systems were almost identical. Results of this type arise when the determinant of the coefficients on the LHS is
7
approximately zero, and great care has to be taken in these situations to get accurate answers. Geometrically,
in each case above we are looking for the intersection of two straight lines. When the determinant is almost
zero the lines are almost parallel and any small change in the slope of one line can lead to major changes in
the point of intersection.
rec/00lm2