Linear Algebra and its Applications 387 (2004) 277–286www.elsevier.com/locate/laaAn involutory Pascal matrixAshkan Ashrafi a, Peter M. Gibson b,∗aDepartment of Electrical and Computer Engineering, University of Alabama in Huntsville,Huntsville, AL 35899, USAbDepartment of Mathematical Sciences, University of Alabama in Huntsville, 204 Madison Hall,Huntsville, AL 35899, USAReceived 21 October 2003; accepted 17 February 2004Submitted by R.A. BrualdiAbstractAn involutory upper triangular Pascal matrix Un is investigated. Eigenvectors of Un andof UTn are considered. A characterization of Un is obtained, and it is shown how the resultscan be extended to matrices over a commutative ring with unity.© 2004 Elsevier Inc. All rights reserved.Keywords: Pascal matrices; Involutory matrices; Eigenvectors; Matrices over a ring1. IntroductionLet Un = (uij ) be the real upper triangular matrix of order n withuij = (−1)i−1j − 1i − 1for 1 i j n.For example,U5 =11 1 1 10 −1 −2 −3 −400 1 3 600 0 −1 −400 0 0 1.∗ Corresponding author.E-mail address: gibson@math.uah.edu (P.M. Gibson).0024-3795/$ - see front matter 2004 Elsevier Inc. All rights reserved.doi:10.1016/j.laa.2004.02.027278 A. Ashrafi, P.M. Gibson / Linear Algebra and its Applications 387 (2004) 277–286Such Pascal matrices are found in the book by Klein [2]. Moreover, the MATLABcommand pascal(n, 1) yields the lower triangular matrix UTn .Klein mentioned that U−1 n = Un (that is, Un is involutory). In fact a somewhatmore general result holds. Let p and q be integers with 1 p q n. Using theidentityδnk = nj=k(−1)j+knj jk,which can be found on page 44 of [3], it is not difficult to see that the principalsubmatrix of Un that lies in rows and column p, p + 1,...,q is involutory.The matrix Un is closely related to two other “Pascal matrices”. Let Pn = (pij )be the real lower triangular matrix of order n withpij =j − 1i − 1for 1 i j n,and let Sn = (sij ) be the real symmetric matrix of order n withsij =i + j − 2j − 1for i, j = 1, 2, . . . , n.Clearly Pn = UTn Dn for the n × n diagonal matrix Dn = ((−1)i−1δij ). Hence, it followsfrom the Cholesky factorization Sn = PnPTn obtained by Brawer and Pirovino[1] that Sn = (UTn Dn)(UTn Dn)T = UTn Un. Thus, the involutory matrices UTn and Uncan be used to obtain an LU factorization for Sn.Other properties of Un are investigated in this paper. Eigenvectors of Un and ofUTn are considered in Section 2. A characterization of Un is presented next, and thenit is shown how the results can be extended to matrices over a commutative ring withunity.2. EigenvectorsIt is easy to see that Un is similar to the diagonal matrix Dn = ((−1)i−1δij ). Wenow consider eigenvectors of Un. For each positive integer k, letxk =k0−k1...(−1)k−1 kk − 1.Lemma 2.1. For each positive integer k, xk is an eigenvector of Uk correspondingto the eigenvalue (−1)k−1.A. Ashrafi, P.M. Gibson / Linear Algebra and its Applications 387 (2004) 277–286 279Proof. SinceUk+1 =Uk xk0 (−1)k,we haveI = U2k+1 =I Ukxk + (−1)kxk0 1 and thus Ukxk = (−1)k−1xk.For integers 1 k n we define the vector ynk ∈ Rn by lettingynk =xk0.Let Yn1 = {ynk : k is odd} and Yn2 = {ynk : k is even}.Theorem 2.2. The set Yn1 is a basis for the eigenspace of Un corresponding tothe eigenvalue 1, and Yn2 is a basis for the eigenspace of Un corresponding to theeigenvalue −1 (when n 2).Proof. Lemma 2.1 implies that ynn = xn is an eigenvector of Un corresponding tothe eigenvalue (−1)n−1. Let 1 kUn =Uk A0 B.Using Lemma 2.1, we see thatUnynk =Uk A0 B xk0=(−1)k−1xk0= (−1)k−1ynk.Hence, for each 1 k n, ynk is an eigenvector of Un corresponding to the eigenvalue(−1)k−1. Moreover, it is easy to see that Yn1 and Yn2 are linearly independentsets.Let Hn = (hij ) be the upper triangular matrix withhij = (−1)i+jj − 1i − 12i−1 for 1 i j n,and let Mn = (mij ) be the lower triangular matrix withmij =i − 1j − 12n−j for 1 j i n.280 A. Ashrafi, P.M. Gibson / Linear Algebra and its Applications 387 (2004) 277–286For example,H6 =1 −1 1 −1 1 −10 2 −4 6 −8 1000 4 −12 24 −4000 0 8 −32 800 0 0 0 16 −800 0 0 0 0 32,M6 =32 0 0 0 0 016 16 0 0 0 08 16 8 0 0 04 12 12 4 0 02 8 12 8 2 01 5 10 10 5 1.It will be shown that the columns of Hn are eigenvectors of Un, and that the columnsof Mn are eigenvectors of UTn .Lemma 2.3. For each positive integer n, UnHn = HnDn.Proof. Clearly UnHn = (aij ) and HnDn = (bij ) are upper triangular. For 1 i j n, we see thataij = jk=i(−1)i−1k − 1i − 1(−1)k+jj − 1k − 12k−1= (−1)i+1 j−1k=i−1(−1)k+j−1 ki − 1 j − 1k2k= (−1)i+2j−1j − 1i − 12i−1= bij ,where we used the identity nk=m(−1)n+knk km2k−m =nm,which can be found on page 32 of [3].The columns of Hn yield different bases for the eigenspaces of Un than thosegiven in Theorem 2.2. Let Vn1 = {vnk : k is odd} and Vn2 = {vnk : k is even}, wherevnk is the kth column of Hn.A. Ashrafi, P.M. Gibson / Linear Algebra and its Applications 387 (2004) 277–286 281Theorem 2.4. The set Vn1 is a basis for the eigenspace of Un corresponding tothe eigenvalue 1, and Vn2 is a basis for the eigenspace of Un corresponding to theeigenvalue −1.Proof. Lemma 2.3 implies that vnk is an eigenvector of Un corresponding to theeigenvalue (−1)k−1. Moreover, Vn1 and Vn2 are linearly independent sets.
We now consider eigenvectors of UT
n . Let Wn1 = {wnk : k is odd} and Wn2 =
{wnk : k is even}, where wnk is the kth column of Mn. Define the diagonal matrices
Qn and Rn of order n by letting Qn = (2i−1δij ) and Rn = (2n−i
δij ).
Lemma 2.5. For each positive integer n, Mn = 2n−1(HT
n )−1.
Proof. We see that Hn = QnUnDn and Mn = RnUT
n Dn. Hence, using D2
n = I =
U2
n , it follows that
MnHT
n = (RnUT
n Dn)(DnUT
n Qn) = RnQn = 2n−1I,
and thus Mn = 2n−1(HT
n )−1.
Lemma 2.6. For each positive integer n, UT
n Mn = MnDn.
Proof. Using Lemma 2.3, we see that
UT
n (HT
n )
−1 = ((UnHn)
T)
−1 = ((HnDn)
T)
−1 = (HT
n )
−1Dn,
and it follows from Lemma 2.5 that UT
n Mn = MnDn.
Theorem 2.7. The set Wn1 is a basis for the eigenspace of UT
n corresponding to
the eigenvalue 1, and Wn2 is a basis for the eigenspace of UT
n corresponding to the
eigenvalue −1.
Proof. Lemma 2.6 implies that wnk is an eigenvector of UT
n corresponding to the
eigenvalue (−1)k−1. Moreover, Wn1 and Wn2 are linearly independent.
3. A characterization of Un
Let Kn = (kij ) be the (0,1)-matrix of order n with kij = 1 if and only if j = i + 1,
and let Gn = (gij ) = Un + (KT
n Un − Un)Kn. An easy computation shows that Gn
is a (0,1)-matrix with gij = 1 if and only if i = j = 1. Thus Gn is a symmetric
matrix. We will show that such symmetry and the property that each leading principal
submatrix is involutory characterizes ±Un for n 4. The following lemmas will be
used.
282 A. Ashrafi, P.M. Gibson / Linear Algebra and its Applications 387 (2004) 277–286
Lemma 3.1. Let X = (xij ) be an involutory matrix of order 2 such that x11 = 1,
X + (KT
2 X − X)K2 is symmetric and X /= U2. Then
X =
1 0
−1 −1
.
Proof. We see that
X + (KT
2 X − X)K2 =
1 x12 − 1
x21 1 − x21 + x22
.
Since this matrix is symmetric and X2 = I , it follows that
X =
1 x12
x12 − 1 −1
,
where x12 = 1 or x12 = 0.
Lemma 3.2. Let X be a matrix of order n 3 and let Y be the leading principal
submatrix of X of order n − 1. If X + (KT
n X − X)Kn is symmetric then Y +
(KT
n−1Y − Y )Kn−1 is symmetric.
Proof. Partition Kn and X as
Kn =
Kn−1 L
0 0
, X =
Y C
R d
.
We see that
X + (KT
n X − X)Kn =
Y + (KT
n−1Y − Y )Kn−1 C + (KT
n−1Y − Y )L
R + (LTY − R)Kn−1 d + (LTY − R)L
.
Lemma 3.3. Let X = (xij ) be a matrix of order 3 such that each leading principal
submatrix of X is involutory, x11 = 1, X + (KT
3 X − X)K3 is symmetric and
X /= U3. Then
X =
100
−1 −1 −2
001
.
Proof. It follows from Lemmas 3.1 and 3.2 that X = X1 or X = X2 where
X1 =
1 1 x13
0 −1 x23
x31 x32 x33
, X2 =
1 0 x13
−1 −1 x23
x31 x32 x33
.
In both cases, since X2 = I , we see that either x13 = x23 = 0 or x31 = x32 = 0. First
suppose that X = X1. It then follows that
A. Ashrafi, P.M. Gibson / Linear Algebra and its Applications 387 (2004) 277–286 283
G = X + (KT
3 X − X)K3 =
1 0 x13 − 1
0 0 x23 + 2
x31 x32 − x31 x33 − 1 − x32
.
Since G is symmetric, if x13 = x23 = 0, then x31 = −1 and x32 = 1. However, this
would imply that X2 =/ I . Hence, we must have x31 = x32 = 0. Therefore, since G
is symmetric, we see that x13 = 1 and x23 = −2. It now follows that X = U3. Thus
we assume that X = X2, and it follows that
G = X + (KT
3 X − X)K3 =
1 −1 x13
−1 1 x23 + 1
x31 x32 − 1 − x31 x33 − 1 − x32
.
Since G is symmetric, if x13 = x23 = 0, then x31 = 0 and x32 = 2. However, this
would imply that X2 =/ I . Hence, we must have x31 = x32 = 0. Therefore, since G
is symmetric, we see that x13 = 0 and x23 = −2. It now follows that
X =
100
−1 −1 −2
001
.
Lemma 3.4. Let X = (xij ) be a matrix of order 4 such that each leading principal
submatrix of X is involutory, x11 = 1, and X + (KT
4 X − X)K4 is symmetric. Then
X = U4.
Proof. It follows from Lemmas 3.2 and 3.3 that X = X1 or X = X2 where
X1 =
100 x14
−1 −1 −2 x24
001 x34
x41 x42 x43 x44
, X2 =
111 x14
0 −1 −2 x24
001 x34
x41 x42 x43 x44
.
In both cases, since X2 = I , we see that either x14 = x24 = x34 = 0 or x41 = x42 =
x43 = 0. First suppose that X = X1
การแปล กรุณารอสักครู่..
