SIAM J. MATRIX ANAL. APPL.


c 2005 Society for Industrial and Applied Mathematics
Vol. 27, No. 1, pp. 212217

BOUNDS FOR INVERSES OF TRIANGULAR TOEPLITZ MATRICES

Downloaded 08/04/12 to 152.17.52.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

KENNETH S. BERENHAUT , DANIEL C. MORTON , AND PRESTON T. FLETCHER

Abstract. This short note provides an improvement on a recent result of Vecchio on a norm
bound for the inverse of a lower triangular Toeplitz matrix with nonnegative entries. A sharper
asymptotic bound is obtained in addition to a version for matrices of nite order. The results are
shown to be nearly best possible under the given constraints.

Key words. Toeplitz matrix, inverse matrix, recurrence relation

AMS subject classications. 15A09, 39A10, 15A57, 15A60

DOI. 10.1137/04060826X

1. Introduction. This paper provides an improvement on a recent result of

Vecchio on inverses of lower triangular Toeplitz matrices. We refer the reader to
Vecchio [13] for a discussion of applications particularly those to the stability analysis
of linear methods for solving Volterra integral equations. Example 1 displays an
improvement in that realm. Other references on the topic, mentioned in [13], include
[1], [2], [7], [10], [11], and [12].
The matrices of interest here are (n + 1) (n + 1) truncations of innite lower
triangular matrices generated by sequences {ai }i0 , i.e.,

a0
a1 a0

(1.1)
An = a2 a1 a0 .


an a1 a0

We will prove the following theorem.

Theorem 1.1. Suppose that the sequence {ai }i0 satises

(1.2) a0 a1 a2 an a 0

for some constant a and all n. Then

 
2 1 1 a [ 2 ]+1
n
 1  if a > 0,
An  a a
2  n  + 1
0
(1.3)
a0 2 if a = 0,

and, in particular, if a > 0,

 1 
(1.4) An  2
a
independent of a0 .
Received by the editors April 13, 2004; accepted for publication (in revised form) by L. Reichel
April 14, 2005; published electronically August 31, 2005.
http://www.siam.org/journals/simax/27-1/60826.html
Department of Mathematics, Wake Forest University, Winston-Salem, NC 27109 (berenhks@

wfu.edu, mortdc0@wfu.edu, etpt1@wfu.edu). The rst author acknowledges nancial support from
a Sterge Faculty Fellowship.
212
 BOUNDS FOR INVERSES OF TOEPLITZ MATRICES 213

Note that Vecchio [13] obtained a comparable (to (1.4)) though less sharp result
 1 
(1.5) An  2 + 1 ,
Downloaded 08/04/12 to 152.17.52.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

a a0
and the methods used therein do not seem amenable to a bound for nite n as in
(1.3). Since we are primarily interested in reliable and applicable, explicit bounds
for nite n, standard Lyapunov stability methods (cf. [3], [4], [5], [6], [8], [9], and
references therein) are also not directly applicable.
Note (optimality of Theorem 1.1). Suppose a1 = a0 > 0 and ai a for 2 i n.
Also, for 0 i n, let
  i
1 a [2]
(1.6) bi = (1)i 1 .
a0 a0
It is easy to verify that in this case

b0
b1 b0

(1.7) A1
n =
b2 b1 b0 ,


bn b1 b0
and hence if a > 0,
 1  
n
(1.8) An  = |bi |

i=0

   i
a [2]
n
1
= 1
a
i=0 0
a0
   n +1  
2 a [2] 0 if n is odd,
= 1 1
a a0 |bn | if n is even.
1 i
If a = 0, bi = a0 (1) and hence
 1  
n
(1.9) An  = |bi |

i=0
n+1
=
a0
2  n   0 if n is odd,
= +1 1
a0 2 a0 if n is even.
Note that (1.8) and (1.9) compare quite favorably with (1.3), and that for a > 0,
(1.4) is in fact optimal under the monotonicity assumption in (1.2).
We close this section with an example.
Example 1. In [13], the dierence, En , between the analytical and numerical
solutions due to the application of a direct quadrature (DQ) method to the Volterra
integral equation (VIE)
 t
(1.10) y(t) = g(t) + k(t s)y(s) ds, t [0, T ], y, g, k 
0

was considered.
 214 K. S. BERENHAUT, D. C. MORTON, AND P. T. FLETCHER

In particular, it was shown that if k(t) 0 k  (t), for t 0, limt k(t) = k < 0
and h|k(0)| < c1 , then
 
Downloaded 08/04/12 to 152.17.52.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

2 1
(1.11) En  + n  ,
hk c2
where En = An n , and c1 , c2 , and An vary according to the DQ method used (see
[13]).
Applying (1.3) in place of (1.4) gives
 [ n2 ]+1
2 hk n 
En  1 1+
hk c2
2
(1.12) n  .
hk
The next section comprises a proof of Theorem 1.1.
2. Proof of Theorem 1.1. As in Vecchio [13], we have that the entries in An
and A1 1
n are related by a linear recurrence with b0 = a0 and


i1
(2.1) bi = i1j bj
j=0

for 1 i n, where
av+1
(2.2) v = .
a0
Note that if we set

(2.3) hi = a0 bi

for 0 i n, then h0 = 1 and


i1
(2.4) hi = i1j hj
j=0

for 1 i n, and hence we will restrict attention to the simpler sequence {hi }.
Now, set 1 = 1 and for 0 i n, let

(2.5) di = i1 i

be the successive dierence, which is nonnegative under the monotonicity assumption

in (1.2).
Also, dene the nonnegative sequence {Si } by Si = 0 for i 2, S1 = 1 and


i1
(2.6) Si = di1j Sj
j=1

for i 0.
 BOUNDS FOR INVERSES OF TOEPLITZ MATRICES 215

A key to obtaining our bound will be a splitting of hi into positive and negative
parts as given by Lemma 2.1. Note that (2.8) may be compared with the use of the
fundamental matrix {uj } in [13] given by
Downloaded 08/04/12 to 152.17.52.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php


n1
(2.7) un = 1 n1l ul
l=0

for n 0. In fact, it is not dicult to show that Si = ui+1 for all i 1.

The following lemma follows from (2.6) in [13]. For completeness, we give a proof
here.
Lemma 2.1. We have
(2.8) hi = Si1 Si2
for 0 i n.
Proof. Note that taking dierences in (2.6) gives

i
(2.9) Si Si1 = dij (Sj1 Sj2 )
j=0

for i 1. In addition, by (2.4)


i
(2.10) hi+1 hi = (i1j ij )hj 0 hi
j=0

for i 1, and hence


i
hi+1 = (i1j ij )hj + (1 0 )hi
j=0
i
(2.11) = dij hj .
j=0

Since S1 S2 = 1 = h0 and S0 S1 = d0 1 = h1 , {Sj Sj1 }j1 and

{hj }j>0 satisfy the
same recurrence and the lemma follows.
m
Now, let cm = j=0 dj = 1 m = 1 am+1 /a0 . We require the following lemma
concerning alternating sums over subsequences of {Si }.
Lemma 2.2. If 2 i1 < i2 < < ik = m is some increasing, nite, integer
sequence, then


k [ m+1
2 ]

(2.12) (1) Sij

j
cjm .
j=1 j=0

Proof. If m 0, (2.12) is immediate. Suppose (2.12) holds for m < N where

N > 0. For m = N , we have


k 
k 
N
(1)j Sij = (1)j dv Sij v1
j=1 j=1 v=0


N 
k
= dv (1)j Sij v1
v=0 j=1
 216 K. S. BERENHAUT, D. C. MORTON, AND P. T. FLETCHER
 i v 
k

N  2

dv j
cm
Downloaded 08/04/12 to 152.17.52.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

v=0 j=0


N [ N2 ]

dv j
cm
v=0 j=0

[ ]

N
2
j
= cm cm
j=0
N +1
[]
2

(2.13) j
cm ,
j=0

where the rst inequality in (2.13) follows by induction.

We are now in a position to prove Theorem 1.1.
Proof of Theorem 1.1. From Lemma 2.1, we have


n 
n
|hj | = |Sj1 Sj2 |
j=0 j=0


n1
=1+ (1)lj (Sj Sj1 )
j=0


n2
(2.14) = 2S1 + Zj Sj + (1)ln1 Sn1 ,
j=0

where for 0 j n 1,

0 if Sj Sj1 ,
(2.15) lj =
1 if Sj < Sj1 ,

and for 0 j n 2,

2 if lj = 0 and lj+1 = 1,
(2.16) Zj = (1)lj (1)lj+1 0 if lj = lj+1 ,

2 if lj = 1 and lj+1 = 0.

Lemma 2.2 and (2.14) give for some sequence 2 i1 < i2 < < ik = n 1,

 

n 
k1
|hj | = 2 S1 + (1) Sij j
+ (1)ln1 Sn1
j=0 j=1

[ n2 ]
 j
(2.17) 2 cn1 .
j=0
 BOUNDS FOR INVERSES OF TOEPLITZ MATRICES 217

Hence from (2.3) and (2.17),

 1  
n
An  = |bj |

Downloaded 08/04/12 to 152.17.52.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

j=0
n
1
= |hj |
a
j=0 0

[ n2 ]
2  j
c
a0 j=0 n1
[ n2 ]  j
2  an
(2.18) = 1
a0 j=0 a0
[ n2 ]  j
2  a
(2.19) 1 ,
a0 j=0 a0

and (1.3) and (1.4) follow.

Remark. Note that if knowledge of an is available, then (2.18) provides an im-
provement on (2.19).
Acknowledgment. We are very thankful to a referee for comments and insights
that improved this manuscript.

REFERENCES

[1] H. Brunner, M. R. Crisci, V. B. E. Russo, and A. Vecchio, A family of methods for Abel
integral equations of the second kind, J. Comput. Appl. Math., 34 (1991), pp. 211219.
[2] H. Brunner and P. J. van der Houwen, The Numerical Solution of Volterra Equations, CWI
Monogr. 3, North-Holland, Amsterdam, 1986.
[3] M. R. Crisci, V. B. Kolmanovskii, E. Russo, and A. Vecchio, Stability of continuous and
discrete Volterra integro-dierential equations by Liapunov approach, J. Integral Equations
Appl., 7 (1995), pp. 393411.
[4] M. R. Crisci, V. B. Kolmanovskii, E. Russo, and A. Vecchio, Boundedness of discrete
Volterra equations, J. Math. Anal. Appl., 211 (1997), pp. 106130.
[5] M. R. Crisci, V. B. Kolmanovskii, E. Russo, and A. Vecchio, Stability of dierence Volterra
equations: Direct Liapunov method and numerical procedure, Comput. Math. Appl., 36
(1998), pp. 7797.
[6] J. T. Edwards, N. J. Ford, J. A. Roberts, and L. Shaikhet, Stability of a discrete nonlinear
integro-dierential equation of convolution type, Stab. Control Theory Appl., 3 (2000),
pp. 2437.
[7] P. Henrici, Applied and Computational Complex Analysis, Vol. 1, John Wiley and Sons, New
York, 1974.
[8] V. B. Kolmanovskii and L. E. Shaikhet, General method of Lyapunov functionals construc-
tion for stability investigations of stochastic dierence equations, in Dynamical Systems
and Applications, World Sci. Ser. Appl. Anal. 4, World Scientic, River Edge, NJ, 1995,
pp. 397439.
[9] V. B. Kolmanovskii and L. E. Shaikhet, Some peculiarities of the general method of Lya-
punov functionals construction, Appl. Math. Lett., 15 (2002), pp. 355360.
[10] P. Linz, Analytical and Numerical Methods for Volterra Equations, SIAM, Philadelphia, 1985.
[11] M. Meneguette, A note on bordered isoclinal matrices, J. Comput. Appl. Math., 17 (1987),
pp. 375377.
[12] A. Vecchio, Stability results on some direct quadrature methods for Volterra integro-diferential
equations, Dynam. Systems Appl., 7 (1998), pp. 501518.
[13] A. Vecchio, A bound for the inverse of a lower triangular Toeplitz matrix, SIAM J. Matrix
Anal. Appl., 24 (2003), pp. 11671174.