Applied Mathematics, 2012, 3, 342-345

http://dx.doi.org/10.4236/am.2012.34052 Published Online April 2012 (http://www.SciRP.org/journal/am)

A Generalized Symbolic Thomas Algorithm

Moawwad El-Mikkawy
Mathematics Department, Faculty of Science, Mansoura University, Mansoura, Egypt
Email: m_elmikkawy@yahoo.com

Received February 4, 2012; revised March 1, 2012; accepted March 8, 2012

ABSTRACT
The current paper is mainly devoted to construct a generalized symbolic Thomas algorithm that will never fail. Two
new efficient and reliable computational algorithms are given. The algorithms are suited for implementation using
computer algebra systems (CAS) such as Mathematica, Macsyma and Maple. Some illustrative examples are given.

Keywords: Tridiagonal Matrices; LU Factorization; Linear Systems; DETGTRI Algorithm; Thomas Algorithm

1. Introduction tridiagonal matrices of the form (1.1) and this can be

done using the efficient symbolic algorithm DETGTRI in
The n  n general tridiagonal matrix T takes the form:
[6]. The motivation of the present paper is to continue the
 d1 a1 0  0  study of the LU factorization [7] of the matrix Tn( k ) in
b d a   
 2 2 2  (1.2) using a new approach, different from the recent
T   0 b3 d3  0  (1.1) approach given in [8], in order to obtain a generalized
  symbolic Thomas algorithm that will never fail.
     an 1  The organization of the paper is as follows. In Section
 0  0 bn d n 
2, generalization of the DETGTRI algorithm in [6] is in-
This type of matrices frequently appear in many scien- vestigated. The generalization of the Thomas algorithm
tific and engineering applications. This area of research [7] is considered in Section 3. Some illustrative examples
is still very active and has been considered by many au- are given in Section 4 for the sake of illustration. A con-
thors. The interested reader may refer to, for instance to clusion is given in Section 5.
[1-3] and the references therein. The n  n general tri-
diagonal matrix T in (1.1) can be stored using only three 2. Generalization of the DETGTRI Algorithm
n-dimensional vectors since there is no need to store the We begin this section by giving the following result
zero elements. In this paper we shall focus on k-tridi- whose proof will be omitted.
agonal matrices [4] of order n  n of the form: Theorem 2.1. For k  1, 2, n  1 , the Doolittle LU
 d1 0  0 a1 0  0  factorization [7] of the k-tridiagonal matrix Tn( k ) in (1.2)
 0 d2 0  0 a2    is given by:

 0  0    0  Tn( k )  L(nk )U n( k ) , (2.1)
  where
0   d n k    an k 
Tn( k )   (1.2)  1 
bk 1 0      0   0 
   1 
 0 bk  2   0  0 
 
  0 
  0  0 d n1 0   
   0   1 
 0  0 bn 0  0 d n 
 bk 1 
where 1  k  n . For k  n , the matrix in (1.2) is a di-  0    
Ln   c1
(k )
 (2.2)
agonal matrix and the case k = 1 gives the ordinary tridi-
 bk  2 
agonal matrix in (1.1). As pointed out in [4,5], the matrix  0   0  
Tn( k ) plays an important role in describing generalized  c2 
k-Fibonacci numbers. Furthermore, the matrix Tn( k ) has     0  0 1 
very recently received attention by some authors. In [5],  
 

using direct sum of matrices, it is proved that det Tn( k )  0  0
bn
0  0 1
may be obtained as the product of exactly k ordinary  cn-k 

Copyright © 2012 SciRes. AM

 M. EL-MIKKAWY 343

and Algorithm 3.1. (Generalization of the Thomas nu-

c1 0  0 a1 0  0  meric algorithm [7])
 c2 0  0 a2    To solve the linear system of the form (3.1), we may
  proceed as follows:
  0    0 
  Step 1: Compute c1 , c2 , , cn using (2.4).
cn k    an k 
U n( k )  (2.3) Step 2: Compute z1 , z2 , , zn using
    0 
 
  0   yi i  1, 2, , k
 
cn 1 0  zi   bi zi -k
   yi  c i  k  1, k  2, , n
 cn   i -k

with Step 3: The solution of the system (3.1) is

di i  1, 2, , k
  zi
ci   bi ai -k (2.4) c i  n, n  1, , n  k  1
di  c i  k  1, k  2, , n  i
xi  
 i -k
 zi - ai xi  k i  n  k , n  k  1, ,1
Now we can formulate our first result.  ci
Algorithm 2.1. (Generalization of the DETGTRI al-
gorithm [6]) The Algorithm 3.1 will be referred to as k-Thomas al-
To compute the determinant of the k-tridiagonal matrix gorithm. The Thomas algorithm is a special case of the
Tn( k ) in (1.2) we may proceed as follows: k-Thomas algorithm when k = 1.
Step 1: Compute c1 , c2 , , cn to its simplest rational Note 3.1. Since the k-Thomas algorithm is based on
forms according to (2.4), setting cr   (  is just a the LU factorization of the matrix, Tn( k ) , then the k-
symbolic name) whenever cr  0, r  1, 2, , n .
n
Thomas numeric algorithm may fail if cr  0 ,
Step 2: Compute the polynomial p      cr to its r  1, 2, , n - k  as can be seen from (2.1)-(2.4). To
r 1

 
simplest rational form. Meanwhile, det Tn( k )  p  0  .
The Algorithm 2.1 will be referred to as k-DETGTRI
algorithm. The DETGTRI algorithm in [6] is a special
case of the k-DETGTRI algorithm when k = 1. Note that

the k-DETGTRI algorithm computes det Tn( k ) in linear
time.
3. Generalization of the Thomas Numeric
Algorithm
In this section, we are going to generalize the well-
known Thomas numeric algorithm [7]. The purpose of
the generalized algorithm is to solve the following linear
system for any k  1, 2, , n  1 .
The new algorithm can be stated as follows:
remove all cases in which the k-Thomas numeric algo-
rithm fails, it is convenient to give the following sym-
bolic version of the k-Thomas numeric algorithm de-
scribed above. The k-Thomas symbolic algorithm can be
stated as follows:
Algorithm 3.2. (Generalized symbolic Thomas al-
gorithm)
To solve the linear system of the form (3.1), we may
proceed as follows:
Step 1: Compute c1 , c2 , , cn to its simplest rational
forms using (2.4), setting cr   (  is just a symbolic
name) whenever cr  0, r  1, 2, , n .
Step 2: Compute z1 , z2 , , zn to its simplest rational
forms using:

 d1 0  0 a1 0  0   x1   y1 
 0 d2 0  0 a2     x2   y2 

 0  0    0   x3   y3 
    
 0   d nk    ank       
 . (3.1)
bk 1 0      0      
    
 0 bk  2   0  0    xn2   yn 2 
    0  0 d n1 0   xn1   yn1 
    
 0  0 bn 0  0 d n   xn   yn 

Copyright © 2012 SciRes. AM

344 M. EL-MIKKAWY

 yi i  1, 2, , k modified positions are written in bold):


zi   bi zi -k 2 0 0 0 1 0 0 0 0 0   x1   4 
 yi  c i  k  1, k  2, , n
 i -k
0 1 0
 0 0 1 0 0 0 0   x2   2 
Step 3: The solution of the system (3.1) is  0 0 1 0 0 0 2 0 0 0   x3   0 
    
0 0 0 3 0 0 0 4 0 0   x3  13
 zi
c i  n, n  1, , n  k  1 2 0 0 0 1 0 0 0 1 0   x5   6 
 i       .
xi    0 1 0 0 0 2 0 0 0 3  x6  10 
 zi - ai xi  k

i  n  k , n  k  1, ,1 0 0 3 0 0 0 5 0 0 0   x7   0 
ci     
0 0 0 2 0 0 0 3 0 0   x8   9 
evaluated at   0 .     
0 0 0 0 1 0 0 0 1 0   x9   0 
Before we consider the solution of the system (3.1) by
 0 0 0 0 0 5 0 0 0 3  x10   4 
using the k-Thomas algorithm, it is recommended to use
the k-DETGTRI algorithm to check the nonsingularity of (4.2)
the coefficient matrix Tn( k ) firstly. In this case the k-Thomas numeric algorithm fails
since c5 = 0 as can be easily checked. However, the
4. Some Illustrative Examples k-Thomas symbolic algorithm yields:
 c1  2, c2  1, c3  1, c4  3, c5   , c6  3, c7  11,
In this section we are going to give some illustrative
examples.
1
c8  , c9  
   1 , c  8 (Step 1);
Example 4.1. Consider the following k-tridiagonal sys- 3  10

tem with n = 10 and k = 4:  

Hence det Tn  528    1  0  528. Consequently
(k )

(k )
2 0 0 0 1 0 0 0 0 0   x1   4  the matrix T is nonsingular.
n
 z1  4, z2  2, z3  0, z4  13, z5  8, z6  12, z7  0,

0 1 0 0 0 1 0 0 0 0   x2   2 
1 2
 0 0 1 0 0 0 2 0 0 0   x3   0  z8  , z9   , z10  24 (Step 2);
     3 
0 0 0 3 0 0 0 4 0 0   x3  13
2  1
2 0 0 0 4 0 0 0 1 0   x5  12   x1   1, x2  1, x3  0, x4  3,
      .   1  0
 0 1 0 0 0 2 0 0 0 3  x6  10 
2 2
0 0 3 0 0 0 5 0 0 0   x7   0  x5 
  1  0 6 7  2, x  1, x  0, x  1, x
8 9
  1  0
 2,
    
0 0 0 2 0 0 0 3 0 0   x8   9  x10  3 (Step 3).
    
0 0 0 0 1 0 0 0 1 0   x9   0  Example 4.3. Consider the k-tridiagonal matrix
 0 0 0 0 0 5 0 0 0 3  x10   4 
2 0 0 0 1 0 0 0 0 0
(4.1) 0 1 0 0 0 1 0 0 0 0 

Applying the numeric or the symbolic version of the  0 0 1 0 0 0 2 0 0 0
k-Thomas algorithms gives:  
c1  2, c2  1, c3  1, c4  3, c5  3, c6  3, c7  11, 0 0 0 3 0 0 0 4 0 0
2 0 0 0 1 0 0 0 1 0
1 4 Tn( k )    . (4.3)
c8  , c9   , c10  8 (Step 1);
3 3  0 1 0 0 0 2 0 0 0 3
0 0 3 0 0
 
Hence det Tn( k )  2112 . Thus Tn( k ) is nonsingular. 
0 0 0 5 0

0 0 0 2 0 0 0 3 0 0
 z1  4, z2  2, z3  0, z4  13, z5  8, z6  12, z7  0,  
0 0 0 0 1 0 0 0 1 0 
1 8
z8  , z9   , z10  24 (Step 2);  0 0 0 0 0 5 0 0 0 5
3 3
 x1  1, x2  1, x3  0, x4  3, x5  2, x6  1, x7  0, x8  1, For this matrix, the k-DETGTRI algorithm gives:
x9  2, x10  3 (Step 3). 1
c1  2, c2  1, c3  1, c4  3, c5   , c6  3, c7  11, c8  ,
Example 4.2. Consider the following k-tridiagonal 3
system with n = 10 and k = 4 (this example is Example
c9  
   1
, c10   . (Step 1)
4.1 with slight modifications in some positions. The 

Copyright © 2012 SciRes. AM

 M. EL-MIKKAWY
 
Hence det Tn( k )  66    1  0  0. Therefore the
matrix Tn( k ) is singular.
5. Conclusion
A generalized symbolic Thomas algorithm, that will
never fail, is developed, for the first time, in this paper.
The algorithm is suited for implementation using the
Computer Algebra Systems (CAS) such as Macsyma,
Maple and Mathematica.
REFERENCES
[1] M. E. A. El-Mikkawy, “A Note on a Three-Term Recur-
rence for a Tridiagonal Matrix,” Applied Mathematics
and Computation, Vol. 139, No. 2-3, 2003, pp. 503-511.
doi:10.1016/S0096-3003(02)00212-6
[2] M. E. A. El-Mikkawy, “On the Inverse of a General Tri-
diagonal Matrix,” Applied Mathematics and Computation,
Vol. 150, No. 3, 2004, pp. 669-679.
doi:10.1016/S0096-3003(03)00298-4
[3] M. E. A. El-Mikkawy and A. Karawia, “Inversion of
Copyright © 2012 SciRes.
345
General Tridiagonal Matrices,” Applied Mathematics Let-
ters, Vol. 19, No. 8, 2006, pp. 712-720.
doi:10.1016/j.aml.2005.11.012
[4] M. El-Mikkawy and T. Sogabe, “A New Family of k-Fib-
onacci Numbers,” Applied Mathematics and Computation,
Vol. 215, No. 12, 2010, pp. 4456-4461.
doi:10.1016/j.amc.2009.12.069
[5] T. Sogabe and M. El-Mikkawy, “Fast Block Diagonaliza-
tion of k-Tridiagonal Matrices,” Applied Mathematics and
Computation, Vol. 218, No. 6, 2011, pp. 2740-2743.
doi:10.1016/j.amc.2011.08.014
[6] M. E. A. El-Mikkawy, “A Fast Algorithm for Evaluating
nth Order Tridiagonal Determinants,” Journal of Compu-
tational and Applied Mathematics, Vol. 166, No. 2, 2004,
pp. 581-584. doi:10.1016/j.cam.2003.08.044
[7] R. L. Burden and J. D. Faires, “Numerical Analysis,” 7th
Edition, Books & Cole Publishing, Pacific Grove, 2001.
[8] A. Yalciner, “The LU Factorization and Determinants of
the k-Tridiagonal Matrices,” Asian-European Journal of
Mathematics, Vol. 4, No. 1, 2011, pp. 187-197.
doi:10.1142/S1793557111000162
AM