Journal of Inequalities in Pure and

Applied Mathematics
http://jipam.vu.edu.au/
Volume 6, Issue 3, Article 63, 2005

ON INVERSES OF TRIANGULAR MATRICES WITH MONOTONE ENTRIES

KENNETH S. BERENHAUT AND PRESTON T. FLETCHER

D EPARTMENT OF M ATHEMATICS
WAKE F OREST U NIVERSITY
W INSTON -S ALEM , NC 27106
berenhks@wfu.edu
URL: http://www.math.wfu.edu/Faculty/berenhaut.html

fletpt1@wfu.edu

Received 26 August, 2004; accepted 24 May, 2005

Communicated by C.-K. Li

A BSTRACT. This note employs recurrence techniques to obtain entry-wise optimal inequalities
for inverses of triangular matrices whose entries satisfy some monotonicity constraints. The
derived bounds are easily computable.

Key words and phrases: Explicit bounds, Triangular matrix, Matrix inverse, Monotone entries, Off-diagonal decay, Recur-
rence relations.
2000 Mathematics Subject Classification. 15A09, 39A10, 26A48.

1. I NTRODUCTION
Much work has been done in the recent past to understand off-diagonal decay properties of
structured matrices and their inverses (cf. Benzi and Golub [1], Demko, Moss and Smith [4],
Eijkhout and Polman [5], Jaffard [6], Nabben [7] and [8], Peluso and Politi [9], Robinson and
Wathen [10], Strohmer [11], Vecchio [12] and the references therein).
This paper studies nonnegative triangular matrices with off-diagonal decay. In particular, let
l1,1

l2,1 l2,2
l l l

Ln = 3,1 3,2 3,3

.. . .
.. .
.. . ..


ln,1 ln,2 ln,3 ln,n

ISSN (electronic): 1443-5756

c 2005 Victoria University. All rights reserved.
We are very thankful to the referees for comments and insights that substantially improved this manuscript.
The first author acknowledges financial support from a Sterge Faculty Fellowship and an Archie fund grant.
166-04
2 K ENNETH S. B ERENHAUT AND P RESTON T. F LETCHER

be an invertible lower triangular matrix, and

x1,1

x2,1 x2,2
X n = L1 x3,1 x3,2 x3,3

n =
,
... .. .. ..
. . .
xn,1 xn,2 xn,3 xn,n
be its inverse.
We are interested in obtaining bounds on the entries in X n under the row-wise monotonicity
assumption
(1.1) 0 li,1 li,2 li,i1 li,i
for 2 i n.
As an added generalization, we will consider [li,j ] satisfying
li,1 li,2 li,i1
(1.2) 0 i1 ,
li,i li,i li,i
for some nondecreasing sequence = (1 , 2 , 3 , . . . ).
The paper proceeds as follows. Section 2 contains some recurrence-type lemmas, while the
main result, Theorem 3.1, and its proof are contained in Section 3. The paper closes with some
illustrative examples.

2. P RELIMINARY L EMMAS
In establishing our main results, we will employ recurrence techniques. In particular, suppose
{bi } and {i,j } satisfy the linear recurrence
i1
X
(2.1) bi = (i,k )bk , (1 i n),
k=0
with b0 = 1 and
(2.2) 0 i,0 i,1 i,2 i,i1 Ai ,
for i 1.
We will employ the following lemma, which reduces the scope of consideration in bounding
solutions to (2.1).
Lemma 2.1. Suppose that {bi } and {i,j } satisfy (2.1) and (2.2). Then, there exists a sequence
a1 , a2 , . . . , an , with 0 ai i for 1 i n, such that |bn | |dn |, where {di } satisfies d0 = 1,
and for 1 i n,
Pi1
j=ai (Ai )dj , if ai < i
(2.3) di = .
0, otherwise

In proving Lemma 2.1, we will refer to the following result on inner products.
Lemma 2.2. Suppose that p = (p1 , . . . , pn )0 and q = (q1 , . . . , qn )0 are n-vectors with
(2.4) 0 p1 p2 pn A.
Define
n
z }| { z }| {
(2.5) pn (, A) = (0, 0, . . . , 0, A, . . . , A, A)

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/
 I NVERSES OF T RIANGULAR M ATRICES WITH M ONOTONE E NTRIES 3

for 0 n. Then,
(2.6) min {pn (, A) q} p q max {pn (, A) q},
0n 0n
Pn
where p q denotes the standard dot product i=1 pi qi .
Proof. Suppose p is of the form
e e
z }|1 { z }|2 {
(2.7) (p1 , . . . , pj , k, . . . , k, A, . . . , A),
with 0 p1 p2 pj > k > A, e1 1 and e2 0. First, assume that p q > 0,
and consider S = ei=j+1
P 1 +j
qi . If S < 0 then, since k < A,
e e
z }|1 {z }|2 {
(2.8) (p1 , p2 , . . . , pj1 , pj , A, . . . , A A, . . . , A) q p q.
Otherwise, since k < pj ,
e
e
z }|1 { z }|2 {
(2.9) (p1 , p2 , . . . , pj1 , pj , pj , . . . , pj , A, . . . , A) q p q.
In either case, there is a vector of the form in (2.7) with strictly less distinct values, whose inner
product with q is at least as large as p q. Inductively, there exists a vector of the form in (2.7)
with e2 + e1 = n, with as large, or larger, inner product. Hence, we have reduced to the case
e e
z }|1 { z }|2 {
where p = (k, . . . , k, A, . . . , A), where e1 = 0P and en = 0 are permissible. If k = 0 or
e1 = 0, then p = pn (e1 , A). Otherwise, consider S = ei=1 1
qi . If S < 0, then
(2.10) pn (0, A) q p q.
If S 0,
(2.11) pn (e1 , A) q p q.
The result for the case p q > 0 now follows from (2.10) and (2.11).
The case when p q 0 is handled similarly, and the lemma follows. 
We now turn to a proof of Lemma 2.1.
Proof of Lemma 2.1. The proof, here, involves applying Lemma 2.2 to successively scale the
rows of the coefficient matrix
1,0

0 ... 0
2,0 2,1 . . .

0
,

. . .. .


.
. .
. . .
.


n,0 n,1 n,n1
while not decreasing the value of |bn | at any step.
First, define the sequences
i = (i,0 , . . . , i,i1 ) and

bk,j = (bk , . . . , bj ),
for 0 k j n 1 and 1 i n.
n and q = b0,n1 yields a vector
Now, note that applying Lemma 2.2 to the vectors p =
p (n , An ) (as in (2.5)) such that either
(2.12) p (n , An ) b0,n1
n b0,n1 = bn > 0

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/
4 K ENNETH S. B ERENHAUT AND P RESTON T. F LETCHER

or
(2.13) p (n , An ) b0,n1
n b0,n1 = bn 0
Hence, suppose that the entries of the k th through nth rows of the coefficient matrix are of the
form in (2.5), and express bn as a linear combination of b1 , b2 , . . . , bk i.e.
k
X
bn = Cik bi
i=1
k1
X
(2.14) = Ckk bk + Cik bi .
i=1

Now, suppose Ckk > 0. As before, applying k and q = b0,k1

Lemma 2.2 to the vectors p =
yields a vector pk (k , Ak ), such that
(2.15) pk (k , Ak ) b0,k1
k b0,k1 = bk .
Similarly, if Ckk 0, we obtain a vector pk (k , Ak ), such that
(2.16) pk (k , Ak ) b0,k1
k b0,k1 = bk .
Using the respective entries in pk (k , Ak ) in place of those in
k in (2.1) will not decrease the
value of bn . This completes the induction for the case bn > 0; the case bn 0 is similar, and
the lemma follows. 
Remark 2.3. A version of Lemma 2.4 for Ai 1 was recently applied in proving that all
symmetric Toeplitz matrices generated by monotone convex sequences have off-diagonal decay
preserved through triangular decompositions (see [2]).
Now, For a = (A1 , A2 , A3 , . . . ), with
(2.17) 0 A1 A2 A 3
define
( i
)
def
Y
(2.18) Zi (a) = max Av : 1 j i ,
v=j

for i 1.
We have the following result on bounds for linear recurrences.
Lemma 2.4. Suppose that a = (Aj ) satisfies the monotonicity constraint in (2.17). Then, for
i 1,
(2.19) sup{|bi | : {bj } and {i,j } satisfy (2.1) and (2.2)} = Zi (a).
Proof. Suppose that {bi } satisfies (2.1) and (2.2), and set i = Zi (a) and Mi = max{1, i }, for
i 1. From (2.18), we have
(2.20) Ai+1 Mi = i+1 ,
for i 1. By Lemma 2.1, we may find sequences {di } and {ai } satisfying (2.3) such that
(2.21) |dn | |bn |.
We will show that {di } satisfies the inequality
(2.22) |dl + dl+1 + + di | Mi ,
for 0 l i.

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/
 I NVERSES OF T RIANGULAR M ATRICES WITH M ONOTONE E NTRIES 5

Note that (2.22) (for i = n 1) and (2.3) imply that dn = 0 or an n 1 and

 
X n1 
|dn | =  (An )dj 
 
 
j=an
 
Xn1 
= An  dj 
 
 
j=an
An Mn1
(2.23) = n .

Since d0 = 1, d1 {0, A1 } and

max{|d1 |, |d0 + d1 |} = max{1, A1 , |1 A1 |}
= max{1, A1 }
(2.24) = M1 ,
i.e. the inequality in (2.22) holds for i = 1. Hence, suppose that (2.22) holds for i < N .
Rewriting dN , with v = aN , we have for 0 x N 1,
dx + dx+1 + + dN = (dx + dx+1 + + dN 1 ) An (dv + + dN 1 )
(
(1 AN )(dv + + dN 1 ) + (dx + + dv1 ), if v > x
(2.25) = .
(1 AN )(dx + + dN 1 ) AN (dv + + dx1 ), if v x
Let
(
dv + + dN 1 , if v > x
S1 = ,
dx + + dN 1 , if v x
and
(
dx + + dv1 , if v > x
S2 = .
dv + + dx1 , if v x
In showing that |dx + dx+1 + + dN | MN , we will consider several cases depending on
whether AN > 1 or AN 1, and the signs of S1 and S2 .
Case 1 (AN > 1 and S1 S2 > 0)
(1) v > x.
|dx + dx+1 + + dN | = |(1 AN )S1 + S2 |
max{AN |S1 |, AN |S2 |}
AN max{MN 1 , Mv1 }
AN MN 1
= N
(2.26) = MN ,

where the first inequality follows since (1 AN )S1 and S2 are of opposite signs and
An > 1. The second inequality follows from induction. The last equalities are direct
consequences of the definition of MN and the fact that AN > 1. The monotonicity of
{Mi } is employed in obtaining the third inequality.

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/
6 K ENNETH S. B ERENHAUT AND P RESTON T. F LETCHER

(2) v x.
|dx + dx+1 + + dN | = |(1 AN )S1 AN S2 |
|AN S1 + AN S2 |
= AN |S1 + S2 |
= AN |dv + dv+1 + + dN 1 |
AN MN 1
= N
(2.27) = MN .

In (2.27), the first inequality follows since (1 AN )S1 and AN S2 are of the same
sign.
Case 2 (AN > 1 and S1 S2 0)
(1) v > x.
|dx + dx+1 + + dN | = |(1 AN )S1 + S2 |
(2.28) = | AN S1 + (S1 + S2 )|.
If S1 and S1 + S2 are of the same sign, then
| AN S1 + (S1 + S2 )| max{AN |S1 |, |S1 + S2 |}
AN MN 1
(2.29) = MN .
Otherwise,
| AN S1 + (S1 + S2 )| | AN S1 + AN (S1 + S2 )|
= AN |S2 |
AN MN 1
(2.30) = MN .
(2) v x.
|dx + dx+1 + + dN | = |(1 AN )S1 AN S2 |
max{AN |S1 |, AN |S2 |}
AN MN 1
(2.31) = MN
Case 3 (AN 1 and S1 S2 > 0)
Note that for AN 1, Mi = 1 for all i.
(1) v > x.
|dx + dx+1 + + dN | = |(1 AN )S1 + S2 |
|S1 + S2 |
MN 1
(2.32) = MN .

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/
 I NVERSES OF T RIANGULAR M ATRICES WITH M ONOTONE E NTRIES 7

(2) v x.
|dx + dx+1 + + dN | = |(1 AN )S1 AN S2 |
max{|S1 |, |S2 |}
MN 1
(2.33) = MN .
Case 4 (AN 1 and S1 S2 0)
(1) v > x.
|dx + dx+1 + + dN | = |(1 AN )S1 + S2 |
max{|S1 |, |S2 |}
max{MN 1 , Mv1 }
(2.34) = MN .
(2) v x.
|dx + dx+1 + + dN | = |(1 AN )S1 AN S2 |
|S1 + S2 |
MN 1
(2.35) = MN .
Thus, in all cases |dx + dx+1 + + dN | MN and hence by (2.23), |dN | N . Equation
(2.19) now follows since, for 1 h n, |bn | = Ah Ah+1 An is attained for [i,j ] defined by


Ah , if i = h

(2.36) i,j = Ai , if i > h, j = i .


0, otherwise


We close this section with an elementary result (without proof) which will serve to connect
entries in L1
n with solutions to (2.1).

Lemma 2.5. Suppose M = [mi,j ]nn and y = [yi ]n1 , satisfy M y = (1, 0, . . . , 0)0 , with M
an invertible lower triangular matrix. Then, y1 = 1/m1,1 , and
i1  
X mi,j
(2.37) yi = yj ,
j=1
m i,i

for 2 i n.

3. T HE M AIN R ESULT
We are now in a position to prove our main result.
Theorem 3.1. Suppose = (i ) satisfies
(3.1) 0 1 2 3 ,
and set
def
(3.2) S = {i : i > 1}.

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/
8 K ENNETH S. B ERENHAUT AND P RESTON T. F LETCHER

As well, define {Wi,j } by

def
Y
(3.3) Wi,j = v .
T S
v (S {j,j+1,...,i2}) {i1}

Then, for 1 i n, |xi,i | 1/li,i and for 1 j < i n,

Wi,j
(3.4) |xi,j | .
lj,j

Proof. Suppose that n 1 and X n = L1 n . Solving for the sub-diagonal entries in the p
th

column of X n leads to the matrix equation

lp,p xp,p 1

lp+1,p lp+1,p+1 xp+1,p 0
. .. .. . = . .
.. . . .. ..
ln,p ln,p+1 ln,n xn,p 0

Applying Lemma 2.5 gives xp,p = 1/lp,p , and

i1  
X lp+i,p+j
(3.5) xp+i,p = xp+j,p ,
j=0
lp+i,p+i

for 1 i n p.
Now, note that (1.2) gives
lp+i,p lp+i,p+1 lp+i,p+i1
(3.6) 0 p+i1 .
lp+i,p+i lp+i,p+i lp+i,p+i
Hence by Lemma 2.4,

|xp+i,p | |xp,p |Zi ((p , p+1 , . . . , p+i1 ))

1
(3.7) = Wp+i,p ,
lp,p
for 1 i n p, and the theorem follows. 

4. E XAMPLES
In this section, we provide examples to illustrate some of the structural information contained
in Theorem 3.1.

Example 4.1 (Equally spaced Ai ). Suppose that Ai = Ci for i 1, where C > 0. Then, for
n 1,
1


nC, C 0, n1 ;

1 1

Zn (a) = (n)k C k , C nk+1 , nk , (2 k n 1);


n!C n , C (1, ),

where (n)k = n(n 1) (n k + 1).

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/
 I NVERSES OF T RIANGULAR M ATRICES WITH M ONOTONE E NTRIES 9

Consider the matrix

1 0 0 0 0 0 0

0.25 1 0 0 0 0 0
0.5 0.5 1 0 0 0 0


L7 = 0.75 0.75 0.75 1 0 0 0 ,


1 1 1 1 1 0 0

0 1.25 1.25 1.25 1.25 1 0
1.5 1.5 1.5 1.5 1.5 1.5 1
with (rounded to three decimal places)
1 0 0 0 0 0 0

0.25 1 0 0 0 0 0
0.375 0.5 1 0 0 0 0


(4.1) X7 = L1 = 0.281 0.375 0.75 1 0 0 0 .

7
0.094 0.125 0.25 1 1 0 0


1.25 0 0 0 1.25 1 0
1.875 0 0 0 0.375 1.5 1
Applying Theorem 3.1, with = (.25, .50, .75, 1.00, 1.25, 1.50, . . . ) gives the entry-wise bounds
1 0 0 0 0 0 0

0.25 1 0 0 0 0 0
0.5 0.5 1 0 0 0 0

(4.2) 0.75 0.75 0.75 1 0 0 0 .

1 1 1 1 1 0 0

1.25 1.25 1.25 1.25 1.25 1 0
1.875 1.875 1.875 1.875 1.875 1.5 1
Comparing (4.1) and (4.2), the absolute values of entry-wise ratios are
1

1 1
0.75 1 1


(4.3) 0.375 0.5 1 1 .

0.094 0.125 0.25 1 1

1 0 0 0 1 1
1 0 0 0 0.2 1 1
Note that here L7 was constructed so that |x7,1 | = W7,1 . In fact, as suggested by (2.19), for
each 4-tuple (, I, J, n) with 1 J I n, there exists a pair (Ln , X n ) satisfying (1.2) with
X n = (xi,j ) = L1
n , such that |xI,J | = WI,J .

Example 4.2 (Constant Ai ). Suppose that Ai = C for i 1, where C > 0. Then, for n 1,
(
C, if C 1
Zn (a) = .
C n , if C > 1
In [3], the following theorem was obtained when (2.2) is replaced with
(4.4) 0 i,j A,
for 0 j i 1 and i 1.

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/
10 K ENNETH S. B ERENHAUT AND P RESTON T. F LETCHER

Theorem 4.1. Suppose that A > 0 and m = [1/A], where square brackets indicate the greatest
integer function. If {j }
j=1 is defined by

(4.5) n = max{|bn | : {bi } and [i,j ] satisfy (2.1) and (4.4)},

for n 1, then


A, if n = 1

max(A, A2 ),




if n = 2

n2
  n1  3
(4.6) n = 2 2
A + A, if 3 n 2m + 1 .


(n 2)A2 , if n = 2m + 2





An1 + n2 , if n 2m + 3

Proof. See [3]. 

Thus, if the monotonicity assumption in (2.2) is dropped the scenario is much different. In
fact, in (4.6), {n } increases at an exponential rate for all A > 0. This leads to the following
question.
Open Question. Set
(4.7) n = max{|bn | : {bi } and [i,j ] satisfy (2.1) and i,j Ai for 0 j i 1}.
What is the value of n in terms of the sequence {Ai } and its assorted properties (eg. mono-
tonicity, convexity etc.)?

R EFERENCES
[1] M. BENZI, AND G. GOLUB, Bounds for the entries of matrix functions with applications to pre-
conditioning, BIT, 39(3) (1999), 417438.
[2] K.S. BERENHAUT AND D. BANDYOPADHYAY, Monotone convex sequences and Cholesky de-
composition of symmetric Toeplitz matrices, Linear Algebra and Its Applications, 403 (2005),
7585.
[3] K.S. BERENHAUT AND D.C. MORTON, Second order bounds for linear recurrences with nega-
tive coefficients, in press, J. of Comput. and App. Math., (2005).
[4] S. DEMKO, W. MOSS, AND P. SMITH, Decay rates for inverses of band matrices, Math. Comp.,
43 (1984), 491499.
[5] V. EIJKHOUT AND B. POLMAN, Decay rates of inverses of banded m-matrices that are near to
Toeplitz matrices, Linear Algebra Appl., 109 (1988), 247277.
[6] S. JAFFARD, Proprits des matrices bien localises" prs de leur diagonale et quelques applica-
tions, Ann. Inst. H. Poincar Anal. Non Linaire, 7(5) (1990), 461476.
[7] R. NABBEN, Decay rates of the inverse of nonsymmetric tridiagonal and band matrices, SIAM J.
Matrix Anal. Appl., 20(3) (1999), 820837.
[8] R. NABBEN, Two-sided bounds on the inverses of diagonally dominant tridiagonal matrices, Spe-
cial issue celebrating the 60th birthday of Ludwig Elsner, Linear Algebra Appl., 287(1-3) (1999),
289305.
[9] R. PELUSO, AND T. POLITI, Some improvements for two-sided bounds on the inverse of diago-
nally dominant tridiagonal matrices, Linear Algebra Appl., 330(1-3) (2001), 114.
[10] P.D. ROBINSON AND A.J. WATHEN, Variational bounds on the entries of the inverse of a matrix,
IMA J. Numer. Anal., 12(4) (1992), 463486.

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/
 I NVERSES OF T RIANGULAR M ATRICES WITH M ONOTONE E NTRIES 11

[11] T. STROHMER, Four short stories about Toeplitz matrix calculations, Linear Algebra Appl.,
343/344 (2002), 321344.
[12] A. VECCHIO, A bound for the inverse of a lower triangular Toeplitz matrix, SIAM J. Matrix Anal.
Appl., 24(4) (2003), 11671174.

J. Inequal. Pure and Appl. Math., 6(3) Art. 63, 2005 http://jipam.vu.edu.au/