Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Applied and Computational Mathematics2015; 4(3): 207-213
Published online June 8, 2015 (http://www.sciencepublishinggroup.com/j/acm)
doi: 10.11648/j.acm.20150403.23
ISSN: 2328-5605 (Print); ISSN: 2328-5613 (Online)
Some Convalescent of Linear Equations
M. Rafique, Sidra Ayub
Department of Mathematics, Faculty of Science,
Email address: [email protected] (M. Rafique), [email protected]
To cite this article: M. Rafique, Sidra Ayub. Some Convalescent Methods
Mathematics. Vol. 4, No. 3, 2015, pp. 207-213. doi:
Abstract: In a variety of problems in the
linear equations, Ax = b, comprising n linear
and x = [x1 x2 . . .xn]T, b
= [b1 b2 . . .bn]
T are
to solve such systems of equations, including
Crout’s method and Cholesky’s method, which
and upper triangular matrices respectively.
Turing[2]-[11]
in 1948. Here, in this paper we
above mentioned system Ax = b, with the least[4]
needs about 2n3/3 operations, while Doolittle’s
are required to evaluate n2
number of unknown
2n2/3 operations. Accordingly this method
But, in contrast, the improved Doolittle’s,
(n–1)2 number of unknown elements of the
which requires evaluation of even less number
evaluate only (n–2)2 number of the said unknown
effort required for the purpose can substantially
Keywords: System of Equations, Matrix,
Cholesky’s Method
1. Introduction
There are numerous analytical and numerical
the solution of a linear system, Ax = b,
elimination method, and its modifications namely
In the case of Crout’s method this decomposition
Applied and Computational Mathematics
(http://www.sciencepublishinggroup.com/j/acm)
5613 (Online)
Methods for the Solution
Science, HITEC University, Taxila, Pakistan
[email protected] (S. Ayub)
Methods for the Solution of Systems of Linear Equations. Applied and Computational
doi: 10.11648/j.acm.20150403.23
the fields of physical sciences, engineering, economics,
linear equations in n unknowns x1, x2, …, xn, where A = [a
are the column vectors. There are many analytical as well
including Gauss elimination method, and its modification
which employ LU-decomposition method, where L = [
The LU-decomposition method was first introduced by
we have made an effort to modify the existing LU-decomposition
least possible endeavour. It may be seen that the Gauss
Doolittle’s and Crout’s methods require n2 operations. Accordingly,
unknown elements of the L and U matrices. Moreover,
requires evaluation of 2n2/3 number of unknown elements
Crout’s and Cholesky’s methods presented in this paper
the L and U matrices. Moreover, an innovative method
number of unknown elements of the L and U matrices.
unknown elements. Thus, by employing these methods,
substantially be reduced.
Matrix, Column Vector, Decomposition, Doolittle’s Method,
numerical methods for
b, including Gauss
namely Doolittle’s,
Crout's and Cholesky’s methods.
LU-decomposition of an nxn matrix
which is given as under,
=
decomposition is given as,
Solution of Systems
Applied and Computational
economics, etc., we are led to systems of
[aij] is an nxn coefficient matrix,
well as numerical methods[1}– [11]
modifications namely Doolittle’s method,
[iij] and u = [uij] are the lower
by the mathematician Alan M.
decomposition methods to solve the
Gauss elimination method[1], [2], [3],
Accordingly, in these methods we
Cholesky’s method[1]
requires
elements of the L and U matrices
paper require evaluation of only
is also presented in this paper
matrices. In this method we need to
methods, the computational time and
Method, Crout’s Method,
methods. In the Doolittle’s method
matrix A is of the form A = LU,
.
208 M. Rafique and Sidra Ayub: Some Convalescent Methods for the
It may be mentioned that both the above
For a symmetric and positive definite coefficient
under,
The above mentioned usual LU-decomposition
are well discussed in the books on numerical
In these methods the coefficient matrix A is
= LU, so that the linear system of equations
written as LUx = b. This equation may further
Ly = b, where, Ux = y. First we solve the former
y by forward substitution and then solve
backward substitution. Accordingly, we obtain
the given system of linear equations.
For numerical stability, pivoting is very
the coefficient matrix is diagonally dominant
and positive definite. Therefore, in all the methods
in this paper, pivoting has been implemented
out further steps.
2. Improved Doolittle’s Method
The usual Doolittle’s method, mentioned
devised by the American mathematician
Doolittle (1830-1913)[1], [3], [7]–[12]
, which
Here all aij are known while lij and uij are
are to be evaluated. For this purpose we
hand side matrices L and U, and equate
elements on both sides, i.e., set aij
correspondence enables us to evaluate
elements lij and uij. Here we see that for
coefficient matrix A, it is required to evaluate
number of unknown elements lij, and uij
M. Rafique and Sidra Ayub: Some Convalescent Methods for the Solution of Systems of Linear Equations
=
�������
��� 0 0 … 0�� � 0 … 0�� � � … 0… … … … … … … … … …… … … … … … … … … … .… … … … … … … … … … . .��� �� �� … ���
������
�������
1 u� u� 0 1 u 0 0 1 … … … … … … …… … … … … … …… … … … … … …
0 0 0
quoted decompositions lead to the same solution of any
coefficient matrix, the LU-decomposition by the Cholesky’s
=
decomposition methods
umerical linear algebra.
is decomposed as A
equations Ax = b can be
further be written as,
former equation for
the latter for x by
obtain the solution to
very essential unless
dominant or symmetric
methods discussed
implemented before carrying
Method
mentioned above, was
mathematician Myrick Hascall
appeared in U.S.
Coast and Geodetic Survey in
method, we consider a linear
unknowns, Ax = b. To start with
numerical stability. Thereafter,
is transcribed in such a way that
have the pivot element as unity,
if required, we may have to select
divide it throughout by the coefficient
and make it the pivot equation
main considerations for the decomposition
matrix A, as A = LU, are as under:
(i) The first column of the matrix
as that of matrix A, i.e., li1 = ai1,
(ii) The first row of the matrix
same as that of the matrix A, i.e.,
(iii) The diagonal elements of
unity, i.e., lii = 1 for all i.
Accordingly, we decompose
as,
=
are unknown which
multiply the right
the corresponding
= (lu)ij . This
all the unknown
for an nxn order
evaluate only (n–1)2
j of the L and U
matrices respectively. It may be
the above mentioned scheme
computational effort is substantially
above, after having carried
decomposition, we solve Ly =
of Ux = y for x, to get the required
system of equations. Here we
illustrate the above mentioned method.
Solution of Systems of Linear Equations
… u�� … u� … u�
… … … …… … … … .… … … … . .
0 … 1 ������
any linear system of equations.
Cholesky’s method is carried out as
1878. Here, in our improved
linear system of n equations in n
with pivoting is done to ensure
the system of linear equations
that the pivot equation is made to
unity, i.e., a11 = 1. For this purpose,
select the requisite equation and
coefficient of x1 for the needful,
equation. Following these steps, the
decomposition of the coefficient
under:
matrix L = [lij] is kept the same
, for all i.
matrix U = [uij] is also kept the
i.e., u1j = a1j, for all j.
of the matrix L= [lij] are kept as
the coefficient matrix A = LU
be appreciated that by following
scheme of LU-decomposition, the
substantially reduced. As described
out the above mentioned
b for y, followed by solution
required solution to the given
we discuss few examples to
method.
Applied and Computational Mathematics 2015;
2.1. Solution of a System of Three Equations
Unknowns
We consider the following system of equations,
5x1 + x2 + 2x3 = 19
6x1 + 24x2 – 12x3 = –12
2x1 + 3x2 + 8x3 = 39
We interchange 1st and 2
nd equations and
pivot equation by 6 so as to have a11 = 1,
method explained above. Then we decompose
matrix as A = LU, according to the scheme
Here the coefficient matrix A is of order 3x3,
find only (3–1)2 = 4 number of unknown
and U matrices, which turn out as, l32 = 0.26,
12, u33 = 8.84. Thus the coefficient matrix
as A = LU,
=
Having carried out the above decomposition,
=
After having carried out the above
decomposition, we solve Ly = b for y, where,
y5]T, and b = [– 4 –2 4 6 3]
T. Thus we obtain,
2.91 – 1.44]T. Thereafter, solving Ux = y
solution to the given system of equations as,
3. Improved Crout’s Method
The standard Crout’s method was formulated
American mathematician Prescott Durand
1984)[6]–[12]
. As in the case of improved Doolittle’s
here also, after having carried out pivoting,
equation is divided by the coefficient of
Here all aij are known while lij and uij are
are to be evaluated. For this purpose we
hand side matrices L and U, and equate
elements on both sides, i.e., set aij
Applied and Computational Mathematics 2015; 4(3): 207-213
Equations in Three
equations,
and then divide the
as required by the
decompose the coefficient
scheme described above.
3x3, so we need to
elements of the L
0.26, u22 = –19, u23 =
matrix gets decomposed
decomposition, we solve Ly
= b and obtain y = [–2 29 35.37]
get the solution to the given system
x = [2
2.2. Solution of a System of Five
Unknowns
We consider the following system
5x1 + 10x2 + 15x3 – 20x
2x1 + x2 + 5x3 + 3x4 +
3x1 + 4x2 + x3 + 2x4 +
4x1 + x2 + 3x3 + 5x4 +
2x1 + 4x2 – x3 + 2x4 +
We divide the pivot equation
and then decompose the coefficient
above and obtain (5–1)2 = 16 unknown
U matrices as, l32 = 0.67, l42 =
0.95, l54 = – 0.34, and, u22 =–3,
= – 7.33, u34 = 6.67, u35 = 1, u44
0.72. Accordingly, we obtain,
x
above mentioned
where, y = [y1 y2 y3 y4
obtain, y = [–4 6 12 –
y for x, we get the
as, x = [1 2 –1 1 2]T.
formulated by the
Durand Crout (1907 –
Doolittle’s method,
pivoting, the pivot
x1. Thereafter, the
main idea is to write out the system
the pivot equation has its pivot
the main considerations for
coefficient matrix A, as A = LU,
(i) As in the case of improved
li1 = ai1, for all i.
(ii) Same as in the case of improved
retain u1j = a1j, for all j.
(iii) The diagonal elements
unity, i.e., uii = 1 for all i.
Accordingly, we decompose
as,
=
are unknown which
multiply the right
the corresponding
= (lu)ij . This
correspondence enables us to
elements lij and uij.
We see that for an nxn order
required to evaluate only (n–1)
209
35.37]T. Next solving Ux = y we
system of equations as,
1 4]T
Five Equations in Five
system of linear equations,
20x4 + 5x5 = – 20
+ 2x5 = –2
+ 4x5 = 4
+ x5 = 6
+ 5x5 = 3
equation by 5 in order to make a11 = 1,
coefficient matrix as described
unknown elements of the L and
= 2.33, l32 = 0, l43 = 0.91, l53 =
u23 = –1, u24 = 11, u25 = 0, u33
44 = – 10.73, u45 = – 3.91, u55 =
system Ax = b in such a way that
pivot element a11 = 1. Furthermore,
the decomposition of the
LU, are as under:
improved Doolittle’s method, we keep
improved Doolittle’s method we
of the matrix U are kept as
the coefficient matrix A = LU
to evaluate all the unknown
order coefficient matrix A, it is
1)2 unknown elements lij, and uij
210 M. Rafique and Sidra Ayub: Some Convalescent Methods for the
of the L and U matrices respectively. It may
that by following the above mentioned
decomposition, the number of operations
reduced. As described above, after having
above mentioned decomposition, we solve
followed by Ux = y for x, to obtain the required
the given system of linear equations. Here
examples to illustrate the above mentioned
3.1. Solution of a System of Four Equations
Unknowns
We consider the following system of equations,
5x1 + 10x2 + 15x3 – 20x4 = –10
=
Next, solving Ly = b, where b is the column
12 8]T we obtain, y = [–2 –2 –1.91 1]
T, and
y, we get the solution to above system as x
3.2. Solution of a System of Five Equations
Unknowns
We consider the following system of equations,
16x1 – 8x2 + 8x3 + 16x4 + 24x5
3x1 + 2x2 + 2x3 + 2x4 + 4x5 = 9
3x1 + 3x2 – 3x3 + 2x4 – 4x5 = 7
Having carried out the above mentioned
we first solve Ly = b, where b = [2.5 9 7
[2.5 0.43 0.47 5.84 1]T. Next, we solve Ux
solution to the given system of equations
1]T.
4. Improved Cholesky’s Method
The usual Cholesky’s method for symmetric
M. Rafique and Sidra Ayub: Some Convalescent Methods for the Solution of Systems of Linear Equations
may be appreciated
mentioned scheme of LU-
operations are substantially
having carried out the
solve Ly = b for y,
required solution to
Here we discuss few
method.
Equations in Four
equations,
10
2x1 + x2 + 5x3 + 3x
3x1 + 4x2 + x3 + 2x
4x1 + x2 + 3x3 + 5x
We divide the pivot equation
then decompose the coefficient
Thus we obtain a total of (n
elements of the L and U matrices
l33 = –7.33, l43 = –0.33, l44 = –10.73,
u33 = 9, u34 = –0.91. Thus the
obtained as,
x
column vector, [–2 2
then solving Ux =
= [1 2 –1 1]T.
Equations in Five
equations,
5 = 40
9
7
5x1 – x2 – 1x3 + 3x
7x1 + x2 + 3x3 + 5x
We divide the pivot equation
= 1, and then decompose the coefficient
above. Thus we obtain (n – 1)
terms of the L and U matrices as:
l52 = 4.5, l33 = – 5.14, l43 = 3.71,
0.78 l55 = – 9.34, u23 = 0.14, u24
0.14, u35 = 1.53, and, u45 = 4.84.
desired LU decomposition of the
x
mentioned decomposition,
1 5]T. It gives y =
= y, and obtain the
as, x = [1 1 – 1 1
Method
symmetric and positive
definite coefficient matrix was
officer, geodesist and mathematician
(1876–1918)[1], [12]
. As discussed
algorithms pivoting is necessary
diagonally dominant, or symmetric
Here, in our improved method
in the case of improved Doolittle’s
decompose the coefficient matrix
=
�������
1 0 0 … 0�� 1 0 … 0�� c 1 … 0… … … … … … … … … …… … … … … … … … … … .… … … … … … … … … … . .��� c� c� … 1
������
�������1 a� a�0 c c0 0 c… … … … … …… … … … … …… … … … … … …0 0
Solution of Systems of Linear Equations
3x4 = 2
2x4 = 12
5x4 = 8
equation by 5 to make a11 = 1, and
nt matrix, as illustrated above.
(n–1)2 =(4–1)
2 = 9 unknown
matrices as, l22 = –3, l32 = –2, l42 = –7,
10.73, u23 = 0.33, u24 = – 3.67,
the decomposition A = LU, is
3x4 – 7x5 = 1
5x4 – 5x5 = 5
equation throughout by 16, to make a11
coefficient matrix as enunciated
1)2
= (5 – 1)2 = 16 unknown
as: l22 = 3.5, l32 = 4.5, l42 = 1.5,
3.71, l53 = 1.14, l44 = –1.78, l54 = –
24 = – 0.29, u34 = 0.056, u25 = –
4.84. Accordingly, we get the
the coefficient matrix as,
developed by French military
mathematician Andre-Louis Cholesky
discussed above, for all practical
necessary unless the matrix is
symmetric and positive definite.
we follow the same scheme as
Doolittle’s method. Accordingly, we
matrix A = LU as,
� … a��c … c�
… c�… … … … …
… … … … .… … … … . . 0 … c��
������
Applied and Computational Mathematics 2015;
Here all aij are known while cij are unknown
be evaluated. For this purpose we multiply
side matrices L and U, and equate the
elements on both sides. This correspondence
evaluate all the unknown elements cij.
We see that for an nxn order coefficient
to evaluate only (n–1)2 unknown elements
matrices respectively. It may be appreciated
the above mentioned scheme of LU-decomposition,
number of operations are substantially reduced.
above, after having carried out the
decomposition, we solve Ly = b for y, followed
x, to obtain the required solution to the given
equations. Here we discuss few example
above mentioned method.
=
Next, solving Ly = b, where b is the column
35 94 1]T we obtain, y = [15 –20 39 –16]
T
Ux = y, we get the solution to the given system
as x = [2 –3 4 –1]T.
4.2. Solution of a System of Five Equations
Unknowns
We consider the following system of equations,
12x1 + 18x2 + 24x3 + 30x4 + 12x
3x1 – x2 + x3 – 2x4 + x5 = 9
4x1 + x2 + 2x3 + x4 – x5 = 5
����� 1 1.5 2 2.5 13 � 1 1 � 2 14 1 2 1 � 15 � 2 1 2 12 1 � 1 1 1
���� =
�����
1 3 4 5 2
After carrying out the above mentioned
first we solve Ly = b, where b = [2 9 5 8 –1]
–5.73 –5.75 6.68]T. Thereafter, we solve Ux
the solution to the given system of equations
x = [1 –1 2 –1 1]T.
5. An Innovative LU-Decomposition
Method
We consider a linear system of n equations
Ax = b, where A is the coefficient matrix of
x and b are column vectors of order nx1, each.
carry out pivoting unless the coefficient matrix
Applied and Computational Mathematics 2015; 4(3): 207-213
unknown, which are to
multiply the right hand
the corresponding
correspondence enables us to
matrix A, we have
cij of the L and U
appreciated that by following
decomposition, the
reduced. As described
the above quoted
followed by Ux = y for
given system of linear
examples to illustrate the
4.1. Solution of a System of Four
Unknowns
We consider the following system
Decomposing the coefficient
1)2 = 9 unknown elements of the
0.5, l42 = 0, l43 = – 0.33, u22 =
u34 = – 3, u44 = 16. Thus the decomposition
as,
x
column vector, [15 –T, and then solving
system of equations
Equations in Five
equations,
12x5 = 24
5x1 – 2x2 + x3 + 2x
2x1 + x2 – x3 + x4 +
We divide the pivot equation
required by the method discussed
we decompose the coefficient
described above. In this case we
unknown terms of the L and U
1.73, l52 = 0.36, l43 = 0.25, l53 =
5.5, u23 = –5, u24 = – 9.5, u25 =
u35 = – 3.18, u44 = 6, u45
decomposition is given as,
� 0 0 0 1 0 0 0.91 1 0 1.73 0.25 1 0.36 2.19 0.042
���� x
����� 1 1.5 2
0 � 5.5 � 0 0 � 1.450 0 00 0 0
mentioned decomposition,
1]T, to get y = [2 3
Ux = y, and obtain
equations as,
Decomposition
equations in n unknowns,
of order nxn, while
each. First of all we
matrix is diagonally
dominant or symmetric and positive
necessitates to have the coefficients
as unity, i.e., ai1 = 1 for all i. Any
condition may be divided throughout
for the needful. Thereafter, we
matrix as, A = LU, with the following
(i) The first column of the
[lij] is kept the same as
1, for all i.
(ii) The first row of the upper
is also kept the same as
for j ≥ 2. The elements
matrix L are transformed
(iii) for i ≥ 2.
(iv) The diagonal elements
211
Four Equations in Four
system of equations,
coefficient matrix we obtain (n–1)2 =(4–
the L and U matrices as, l32 = –
4, u23 = – 2, u24 = 0, , u33 = 9,
decomposition A = LU, is given
2x4 + x5 = 8
+ x5 = – 1
equation by 12 to set a11 = 1, as
discussed in this paper. Thereafter,
coefficient matrix as, A = LU, as
we need to evaluate (5–1)2 = 16
U matrices as, l32 = 0.91, l42 =
= 2.19, l54 = 0.042, and u22 = –
= –2, u33 = – 1.45, u34 = –0.36,
45 = 0.25, u55 = 6.68. This
2 2.5 1� 5 � 9.5 � 2
45 � 0.36 � 3.180 6 0.250 0 6.68
����
positive definite. This method
coefficients of x1 in all the equations
Any equation not fulfilling this
throughout by the coefficient of x1
can decompose the coefficient
following provisions,
the lower triangular matrix L =
that of matrix A, i.e., li1= ai1 =
upper triangular matrix U = [uij]
as that of matrix A, i.e., u1j = a1j,
elements of the second column of
transformed as, li2 = ai2 – a12,
elements of the matrix U are kept as
212 M. Rafique and Sidra Ayub: Some Convalescent Methods for the
unity, i.e., uii =1, for all i.
(v) The remaining elements in the second
Thus, schematically, LU-decomposition of the coefficient matrix A = LU, is given as under,
=
After having accomplished the LU-decomposition
described above, we solve the equation Ly
solve the equation Ux = b for x, to obtain
given system of linear equations. Here
examples to elaborate the method discussed
5.1. Solution of a System of Three Equations
Unknowns
We consider the system of equations,
5x1 + x2 + 2x3 = 19
x1 + 4x2 – 2x3 = – 2
2x1 + 3x2 + 8x3 = 39.
We divide the first equation by 5 and the
2, throughout as required by the method elaborated
get,
=
Next, the coefficient matrix is decomposed
above. In this decomposition only (n – 2)2 =
(3–2)2 = 1 number of unknown, namely
be evaluated which comes to 4. Accordingly,
required decomposition A = LU, as,
After having carried out this decomposition,
system Ly = b for y to get y = [15 –5 4.33
solving the system Ux = y, for x we obtain
given system of equations as, x = [2 –3 4 –1]
By following the method discussed above,
linear system having any number of equations
number of unknowns.
6. Conclusion
The solution of a system of linear equations
M. Rafique and Sidra Ayub: Some Convalescent Methods for the Solution of Systems of Linear Equations
second row of matrix U are transcribed as: u2j
decomposition of the coefficient matrix A = LU, is given as under,
x
decomposition as
= b for y and next
the solution to the
we consider few
discussed above.
Equations in Three
the third equation by
elaborated above to
decomposed as described
=
namely l33 is required to
Accordingly, we obtain the
=
Further, solving Ly = b, we
Thereafter, solving Ux = y, we
required solution to the given systems
5.2. Solution of a System of Four
Unknowns
We consider the system of equations,
4x1 – 4x2 + 12x
–x1 + 5x2 – 5x
3x1 – 5x2 + 19x
2x1 – 2x2 + 3x
We divide the pivot equation
equation by –1, divide the third
divide the fourth equation by
transform the given system of equations
[ 15 35 31.33 0.5]. Here we need
(4–2)2 = 4 unknown number
matrices, which are obtained as:
and u34 = – 0.33. Thus we get A
= x
decomposition, we solve the
4.33 1]T. Thereafter,
the solution to the
1]T.
above, we can solve a
equations with the same
equations by means of
LU-decomposition of the coefficient
method that can be employed
numerically. It may be seen that
the usual Doolittle's, and Crout's
of a total of n2 number of unknown
matrices, and in the case of usual
required to evaluate 2n2/3 number
these matrices. However, the
and Cholesky’s methods need
number of elements of these matrices
LU-Decomposition method requires
Solution of Systems of Linear Equations
u2j = �� !�" ���#$"�
for j ≥ 3.
.
we obtain, y = [3.8 – 1.3 4]T.
get, x = [ 2 1 4]T, which is the
systems of equations.
Four Equations in Four
equations,
12x3 + 8x4 = 60
5x3 – 2x4 = – 35
19x3 + 3x4 = 94
3x3 + 21x4 = 1
equation by 4, multiply the second
third equation by 3, and also
by 2, as discussed above, and
equations as Ax = b, where b =
need to evaluate only (n –2)2 =
of elements of the L and U
as: l33 = 3, l43 = – 1.50, l44 = 8,
A = LU as under,
coefficient matrix is a plausible
employed analytically as well as
that for an nxn coefficient matrix,
Crout's methods require evaluation
unknown elements of the L and U
usual Cholesky’s method we are
number of unknown elements of
improved Doolittle’s, Crout’s
need evaluation of only (n –1)2
matrices, while the Innovative
requires evaluation of only (n –
Applied and Computational Mathematics 2015; 4(3): 207-213 213
2)2 number of unknown elements of the L and U matrices.
This difference becomes significant for systems of large
number of linear equations. As such a considerable amount of
computational time and energy can be saved by employing
the methods presented in this paper.
References
[1] E. Kreyszig: Advanced Engineering Mathematics, John Wiley, (2011)
[2] A. M. Turing: Rounding-Off Errors in Matrix Processes”, The Quarterly Journal of Mechanics and Applied Mathematics 1: 287-308.doi:10.1093/qjmam/1.1.287 (1948)
[3] Myrick H. Doolittle:. Method employed in the solution of normal equations and the adjustment of a triangulation, U.S. Coast and Geodetic Survey Report, Appendix 8, Paper No. 3, blz.. 115–120.( 1878)
[4] David Poole: Linear algebra: A Modern Introduction (2nd Edition), Thomson Brooks/Cole ISSN 0-534-99845-3 (2006)
[5] James R. Bunch, and John Hopkins: Triangular factorization and inversion by fast matrix multiplication, Mathematics of Computation, 28, 231-236, doi:10.2307/2005828, ISSN 0025-5718
[6] J H Wilkinson: The Algebraic Eigenvalue Problem, Oxford University Press (1988)
[7] B. Faires, Numerical Analysis, PWS Pub., Boston, (1993).
[8] G.H.Golub and C.F.Van Loan, Matrix Computations, John Hopkins, Baltimore, (1989).
[9] J.H. Mathews and K. D. Fink: Numerical Methods Using MATLAB (4th Edition)
[10] L. V. Fausett: Applied Numerical Analysis Using MATLAB, Pearson, (2009)
[11] S.C.Chapra and R.P.Canale, Numerical Methods for Engineers, Mc-Graw-Hill, New York,(1990).
[12] Petre Teodorescu, Nicolae-Doru Stanescu, Nicolae Pandrea: Numerical Analysis with Applications in Mechanics and Engineering, John Wiley (2013).