18
This article was downloaded by: [University of Southern Queensland] On: 04 October 2014, At: 23:16 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK Linear and Multilinear Algebra Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/glma20 Convertible and m-convertible matrices Adam H. Berliner a a Department of Mathematics, Statistics, and Computer Science , St. Olaf College , 1520 St. Olaf Ave., Northfield , MN 55057 , USA Published online: 25 Aug 2011. To cite this article: Adam H. Berliner (2012) Convertible and m-convertible matrices, Linear and Multilinear Algebra, 60:3, 267-283, DOI: 10.1080/03081087.2011.591394 To link to this article: http://dx.doi.org/10.1080/03081087.2011.591394 PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms- and-conditions

Convertible and m -convertible matrices

  • Upload
    adam-h

  • View
    222

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Convertible and               m               -convertible matrices

This article was downloaded by: [University of Southern Queensland]On: 04 October 2014, At: 23:16Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Linear and Multilinear AlgebraPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/glma20

Convertible and m-convertible matricesAdam H. Berliner aa Department of Mathematics, Statistics, and Computer Science ,St. Olaf College , 1520 St. Olaf Ave., Northfield , MN 55057 , USAPublished online: 25 Aug 2011.

To cite this article: Adam H. Berliner (2012) Convertible and m-convertible matrices, Linear andMultilinear Algebra, 60:3, 267-283, DOI: 10.1080/03081087.2011.591394

To link to this article: http://dx.doi.org/10.1080/03081087.2011.591394

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Page 2: Convertible and               m               -convertible matrices

Linear and Multilinear AlgebraVol. 60, No. 3, March 2012, 267–283

Convertible and m-convertible matrices

Adam H. Berliner*

Department of Mathematics, Statistics, and Computer Science, St. Olaf College,1520 St. Olaf Ave., Northfield, MN 55057, USA

Communicated by B. Shader

(Received 25 August 2010; final version received 22 May 2011)

In this article, we consider the calculation of the permanent of a(0, 1)-matrix using determinants. We investigate when determinants ofmore than one signing of the original are used to calculate the permanentand we show that non-convertible matrices require at least four differentsignings. Then, we loosen the restriction that the matrices used to convertthe permanent are signings of the original and use this to reduce thenumber of determinants necessary to convert the permanent of the all 1’smatrix by considering a particular partition of the set Sn of permutationsof {1, . . . , n}. Finally, we construct a sequence of maximal convertiblematrices with a small number of nonzero entries, thus lowering the possibleupper bound for the number of nonzero entries of such a matrix, relative tothe order.

Keywords: permanent; convertible matrix; m-convertible matrix;sign-nonsingular matrix

AMS Subject Classifications: 15A15; 05C50; 15B35

1. Introduction

Suppose that A¼ [aij] is a (0, 1)-matrix of order n. Where appropriate, we will alsouse A[i, j] to denote the entry in the i-th row and j-th column of A. The permanent ofA, denoted per(A), is defined to be

perðAÞ ¼X�2Sn

Yni¼1

ai,�ðiÞ

!

where the sum extends over the set Sn of all permutations of {1, 2, . . . , n}. Much likethe determinant, the permanent of a matrix has one term for each of the nonzerodiagonals of the matrix. However, each term counts positively in the expansion of thepermanent. For example, if Jn is the n� n matrix for which every entry is equal to 1,we note that (Jn)¼ n!.

A matrix A is fully indecomposable if the rows and columns of A cannot bepermuted into a matrix that has a p� q block of 0’s where pþ q¼ n. For the entiretyof this article, we generally consider only fully indecomposable matrices. Let G be the

*Email: [email protected]

ISSN 0308–1087 print/ISSN 1563–5139 online

� 2012 Taylor & Francis

http://dx.doi.org/10.1080/03081087.2011.591394

http://www.tandfonline.com

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 3: Convertible and               m               -convertible matrices

bipartite graph whose vertices are {x1, . . . , xn, y1, . . . , yn} where xi and yj are joined byan edge if and only if aij¼ 1. Then, G is called the bipartite graph associated with Aand A is the bipartite adjacency matrix of G. Calculating per(A) is thus equivalent tocounting the number of perfect matchings of G.

Polya [16] asked if it was possible to characterize the matrices A whose permanentis equal to their determinant. This question carries great significance, as the

calculation of the permanent was shown by Valiant [19] to be a computationallydifficult problem, specifically a #P-complete problem. Calculating the determinant,on the other hand, is computationally simple and can be calculated in polynomialtime. Polya’s problem leads naturally to sign-nonsingular (SNS) matrices – matriceswhich are determined to be nonsingular based solely upon the positions of the zero,positive and negative entries (c.f. [3]). If A is SNS, then per(jAj)¼�det(A), and so wehave a way of using a determinant to calculate a permanent.

If A0 is a (0,�1)-matrix such that jA0j ¼A (entry-wise), then we say that A0 is asigning of A. We say that A is convertible if A has a nonzero permanent and there is asigning A0 of A such that

perðAÞ ¼ detðA0Þ

In this case, we say that A0 is a conversion of A.For example, the matrix

A ¼

1 1 0

1 1 1

1 1 1

264

375

is convertible, and one possible conversion is given by

A0 ¼

1 �1 0

1 1 �1

1 1 1

264

375

Suppose A is an n� n (0, 1)-matrix with the following properties:

(I) There is a row p containing exactly two nonzero entries apr and aps.(II) Column r contains exactly two nonzero entries apr and aqr.(III) aqs¼ 0.

Given A, we construct a matrix B of order n� 1 by making aqs¼ 1 and deletingrow p and column r. We call B the matrix obtained from A by elementary contractionof the columns r, s on the row p. We could similarly perform an elementarycontraction of two rows r, s on a column p of a matrix. The following theoremcharacterizes matrices which are not convertible [3,14].

THEOREM 1.1 Suppose A is a (0, 1)-matrix of order n whose determinant is not

identically zero. Then A is not convertible if and only if there exists a (0, 1)-matrix A0 oforder k and permutation matrices P and Q such that P(A0 � In�k)Q�A and A0 can becontracted to J3 by a series of elementary contractions.

Kasteleyn [8–10] worked with SNS matrices while studying the dimer problem instatistical mechanics. Building on this, Galluccio and Loebl [5] and Tesler [18] found

268 A.H. Berliner

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 4: Convertible and               m               -convertible matrices

that if A is the bipartite adjacency matrix of a genus g bipartite graph, then 2g � per(A)can be written as a sum of the determinants of 4g signings of A. Codenotti andResta [4] then used this result to calculate the permanents for several families ofsparse circulant matrices using four signings. In Section 2, we investigate thecalculation of the permanent of a matrix using determinants of several othermatrices.

In Section 3, we consider the structure of (0, 1)-matrices that are maximal withrespect to having a SNS signing. In particular, we construct a sequence of matriceswhich is maximal with respect to convertibility and yet have very few nonzeroentries. These results are based on those found in [1]. In Section 4, we conclude with abrief discussion of the notion of maximality as applied to the matrices studied inSection 2.

2. m-Convertible matrices

In this section, we consider the case in which A is non-convertible. A naturalquestion would be to ask how many matrices Ai are needed so that (a multiple of)per(A) can be written as the sum of determinants of the Ai. Thus, if A is a(0, 1)-matrix of order n, we want matrices A1, . . . ,Am such that

C perðAÞ ¼Xmi¼1

detðAiÞ ð1Þ

for some positive integer C. We investigate two ways of accomplishing this goal.

2.1. (Generic) m-convertibility

First, as in the case with convertibility, we require the Ai to be signings of A.Furthermore, we desire (1) to be generic. That is, replacing 1’s in A by algebraicallyindependent indeterminates yields an algebraic identity. In other words, for eachpermutation � 2Sn for which

Qni¼1 ai,�ðiÞ ¼ 1, the coefficient of

Qni¼1 ai,�ðiÞ on the

right-hand side of (1) equals C. If this happens for a particular permutation � 2Sn,we call � generic.

In general, suppose A1, . . . ,Am are signings of A and C is a positive integer suchthat (1) is generic. In this case, A is m-convertible and the collection A1, . . . ,Am iscalled an m-conversion of A. It is usually convenient, although not necessary, toconsider m to be as small as possible.

Ringel [17] proved that the complete bipartite graph Ki, j has genus

ði� 2Þð j� 2Þ þ 3

4

� �

Thus, using the results of Tesler [18] and Galluccio and Loebl [5], we know that Ji, j ism-convertible for

m ¼ 4 ðði�2Þð j�2Þþ3Þ=4b c ð2Þ

THEOREM 2.1 Suppose A is a non-convertible (0, 1)-matrix of order n� 3 whosedeterminant is not identically zero. If A is m-convertible, then m� 4.

Linear and Multilinear Algebra 269

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 5: Convertible and               m               -convertible matrices

Proof Since A is not convertible, we cannot have m¼ 1. If m¼ 2, then we must have

2 perðAÞ ¼ detðA1Þ þ detðA2Þ

since an m-conversion gives an algebraic identity. Thus, per(A)¼ det(A1)¼ det(A2),

which implies A is convertible. In the case m¼ 3, C¼ 1 or C¼ 3 in order for

C perðAÞ ¼ detðA1Þ þ detðA2Þ þ detðA3Þ

to be generic. If C¼ 3, then per(A)¼ det(A1)¼ det(A2)¼det(A3) and A would be

convertible. Therefore, we have C¼ 1 and

perðAÞ ¼ detðA1Þ þ detðA2Þ þ detðA3Þ

Suppose that A is such a matrix of smallest possible order n. If n¼ 3, A¼ J3 as there

is only one 3� 3 matrix which is not convertible. However, it is not possible to

obtain

perðJ3Þ ¼ detðA1Þ þ detðA2Þ þ detðA3Þ

where A1, A2, A3 are signings of J3, as the only possible determinants for signings of

J3 are 0,�4. Now suppose n43. We want each � 2Sn to be generic, so we assume

without loss of generality that A can be contracted to J3 by elementary contractions

and has the form

A ¼

0

B ... ..

.

0

� � � 0 1

0 � � � 0 1 1

26666664

37777775

where the first elementary contraction is of columns n� 1, n on row n. Now, suppose

we can write

perðAÞ ¼ detðA1Þ þ detðA2Þ þ detðA3Þ

where jA1j ¼ jA2j ¼ jA3j ¼A and each permutation is generic. Without loss of

generality, we can assume that each Ai has the form

Ai ¼

0

Bi... ..

.

0

� � � 0 1

0 � � � 0 1 1

26666664

37777775

If we let A0 be the (n� 1)� (n� 1) matrix obtained from A by the elementary

contraction of columns n� 1, n on row n, we have (with each permutation generic)

perðA0Þ ¼ detðA01Þ þ detðA02Þ þ detðA03Þ

270 A.H. Berliner

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 6: Convertible and               m               -convertible matrices

where

A0i ¼

Bi...

� � � �1

26664

37775

This results in a 3-conversion of the (n� 1)� (n� 1) matrix A0, a contradiction.Therefore, m� 4. g

If m¼ 4, then we must have

2 perðAÞ ¼ detðA1Þ þ detðA2Þ þ detðA3Þ þ detðA4Þ ð3Þ

where each permutation � 2Sn for whichQn

i¼1 ai,�ðiÞ 6¼ 0 contributes þ1 three timesand �1 one time on the right-hand side of (3).

2.2. General m-convertibility

Here, we still want to be able to write, in a generic way,

C perðAÞ ¼Xmi¼1

detðAiÞ ð4Þ

for some positive integer C. However, we no longer require the matrices Ai to besignings of the original matrix A. All we require is that Ai is a fully indecomposable(0, 1,�1)-matrix where jAij �A for 1� i�m. In this case, we call the collectionA1, . . . ,Am in (4) a general m-conversion of A. In particular, we consider A¼ Jn as acase where a general m-conversion uses fewer determinants than a genericm-conversion.

Setting i¼ j¼ n in (2), we know that Jn is m-convertible for

m ¼ 4 ððn�2Þ2þ3Þ=4b c ð5Þ

Our goal is to find a general m-conversion of Jn which uses fewer matrices than (5).For n41, let Pn be the (0, 1)-matrix of order n where Pn[i, j]¼ 1 if and only if

j¼ iþ 1 or i¼ n and j¼ 1. If the rows and columns of A can be permuted to result inthe matrix A0 ¼ InþPn, we will say that A is a full cycle matrix. If A is a full cyclematrix, then we calculate that per(A)¼ 2.

Suppose A is a (0, 1)-matrix of order n41. For 2� k� n, we will say that A isa simple k-permanent matrix if A has a full cycle matrix of order n as a submatrix,a row (column) with exactly k nonzero entries, and �(A)¼ 2nþ k� 2. As aconsequence, we note that if A is a simple k-permanent matrix, then per(A)¼ k. Inaddition, a full cycle matrix is a simple 2-permanent matrix. The main result here isthe following theorem.

THEOREM 2.2 For all n� 4 and 2� k� n, there is a collection fAi : 1 � i � n!kg of

simplek-permanent matrices of order n such that for each � 2Sn,Ynj¼1

Ai j, �ð j Þ½ ¼ 1 for exactly one value of i:

Linear and Multilinear Algebra 271

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 7: Convertible and               m               -convertible matrices

Thus, each of the n! terms in per(Jn) occurs as a nonzero term in the permanent of

exactly one Ai. Furthermore, simple k-matrices are convertible. Since InþPn is

convertible, we can change some of the extra 1’s into �1’s where appropriate to make

sure each nonzero permutation is positive in the determinant expansion. Therefore,

we get the following corollary.

COROLLARY 2.3 For n� 4, Jn has a general n!k-conversion for each k, 2� k� n.

For large enough values of n, this is a smaller number of matrices than (5). Before

we prove the theorem, we give some definitions and lemmas.Suppose that s and t are positive integers where s� t. We define P(t, s) to be the

set of permutations of length s on a set of size t. Given two elements �1, �2 of P(t, s),we say that they form a full cycle if, up to simultaneous reordering of the positions,

they have the form

�1 ¼ a1 a2 � � � as�1 as�2 ¼ a2 a3 � � � as a1

and we say that they form a full path if, up to simultaneous reordering of the

positions, they have the form

�1 ¼ a1 a2 � � � as�1 as

�2 ¼ a2 a3 � � � as asþ1

where a1, a2, . . . , asþ1 are distinct.We can interpret Theorem 2.2 to mean the following: for each k, 2� k� n, the

elements of Sn can be partitioned into sets of size k such that each set contains two

permutations �1, �2 which form a full cycle. Each of the other k� 2 permutations

agree with either �1 or �2 in every position except in one fixed position (dependent on

the particular set) in which none of the permutations agree. We will call such a set of

k elements of Sn a simple k-permanent set.

LEMMA 2.4 For t� 4, the set P(t, 2) can be partitioned into pairs such that each pair

forms a full path.

Proof Let t¼ 4. Then we take the following pairs: {12, 31}, {21, 13}, {14, 43},

{41, 34}, {23, 42} and {32, 24}.Now suppose t44. By induction, we assume we already have a partition for the

elements of P(t� 1, 2). Thus, we are left to partition the elements of P(t, 2)

containing t as an element. To do this we take the pairs {it, t(iþ 1)} along with the

pair {(t� 1)t, t1} for 1� i� t� 2. g

LEMMA 2.5 Theorem 2.2 holds for k¼ 2.

Proof We must partition the elements of Sn into pairs so that each pair forms a full

cycle. By Lemma 2.4, we can find a pairing of the elements of P(n, 2) such that each

pair has the form {ab, bc} where a, b, c are distinct. For a fixed pair {ab, bc}, let

d1, . . . , dn�3 denote the elements of {1, . . . , n} n {a, b, c}. We start by pairing the

following two elements of Sn:

�1 ¼ a b c d1 � � � dn�4 dn�3�2 ¼ b c d1 d2 � � � dn�3 a

ð6Þ

272 A.H. Berliner

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 8: Convertible and               m               -convertible matrices

We take all pairs resulting from the (n� 2)! simultaneous permutations ofpositions 3, . . . , n of (6). This results in a pairing for each element of Sn thatbegins with either ab or bc. We do this for each pairing of P(n, 2) given byLemma 2.4, and the result follows. g

Suppose that s, t are positive integers and that s5t. The Johnson graph J(t, s) isthe graph whose vertices are the s-subsets of [t]¼ {1, . . . , t} (we could use any set ofsize t), and two vertices v1 and v2 are connected by an edge if and only ifjv1\ v2j ¼ s� 1. For example, the Johnson graph J(n, 1) is the complete graph on nvertices and the graph J(5, 2) is the complement of the Petersen graph (Figure 1).

LEMMA 2.6 The vertices of J(t, s) can be partitioned into sets V1, . . . ,Vp such that

(i) jVij41 for 1� i� p,(ii) The graph induced on each Vi is complete.

Proof Without loss of generality, we may assume s � t2

� �, as J(t, t� s) is isomorphic

to J(t, s). For the case s¼ 1, J(t, 1) is a complete graph on t41 vertices.

Let s41. By induction, we assume that the result holds for J(t0 s0) where s0 � s andt05t (and 05s05t0). Thus, we can partition the s-subsets of [t� 1] to form completesubgraphs of J(t� 1, s) and we are left with partitioning the s-subsets of [t] whichcontain t as an element. In addition, we can partition the (s� 1)-subsets of [t� 1] toform complete subgraphs of J(t� 1, s� 1). Leaving that partition intact, we add theelement t to each of the (s� 1)-subsets of [t� 1], resulting in a partition of thes-subsets of [t] which contain t as an element into complete subgraphs of J(t, s). g

LEMMA 2.7 Suppose that s, t are positive integers and s5t. The elements of P(t, s)can be partitioned into L ordered sets ð�1,�2, . . . ,�pl Þ such that for 1� l�L, pl� 2 andthe pairs �1,�pl and �i, �iþ1 (1� i� pl� 1) form full paths. If t� 4 and if s¼ 2 ors¼ t� 1, then P(t, s) can be partitioned so that pl¼ 2 for all l.

{1, 2} {1,3}

{1,4}

{2,4}

{4,5}

{1,5}

{3,5}

{2,5}

{2,3}

{3,4}

Figure 1. A representation of J(5, 2).

Linear and Multilinear Algebra 273

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 9: Convertible and               m               -convertible matrices

Proof If s¼ 1, then the ordered set (1,2, . . . , t) works.Suppose s¼ 2. If t¼ 3, then we can partition P(3, 2) into the ordered sets

(12, 23, 31) and (13, 21, 32). If t� 4, the result follows by Lemma 2.4.Suppose s¼ t� 1 and t� 4. For each permutation � in St we remove the element

in the last position, resulting in a permutation �0 in P(t, t� 1). All elements ofP(t, t� 1) are uniquely obtained in this manner. By Lemma 2.5, we can partition theelements of St into pairs {�1,�2} so that each pair forms a full cycle. If we pair theelements of P(t, t� 1) in the same manner, then the resulting pairs f�01,�

02g will each

form a full path.Finally, suppose 3� s5t� 1. By Lemma 2.6, we can partition the vertices of the

Johnson graph J(t, s) into subsets of size at least 2 where the graph induced on eachof the subsets is complete. Consider one of the subsets V¼ {v1, . . . , vq} and make itinto an ordered set (v1, . . . , vq). For 1� i5q, let ri be the element of vi not in viþ1 andlet rq be the element of vq not in v1. Similarly, for 15i� q, let li be the element of vinot in vi�1 and let l1¼ lqþ1 be the element of v1 not in vq.

Since we are considering permutations in P(t, s), we let Wi be the set of allpermutations in P(t, s) whose elements come from vi. For 1� i� p and for any v2Wi,we define a mapping R by defining R(v) to be the element of Wiþ1 (whereWqþ1¼W1) obtained by the following procedure:

(i) Take the elements of v and shift them one position to the right (where thelast element becomes the first element).

(ii) Then, replace ri by liþ1 (if i¼ q, we replace rq with l1).

For example,

ri a1 a2 � � � as�1

becomes

as�1 liþ1 a1 � � � as�2

By construction, we see that v and R(v) form a full path. Thus, pick any �12W1 andcontinue to apply R. Since R is injective, the process of repeatedly applying R musteventually return to the original element �12W1. Thus, we obtain an ordered set(�1,�2, . . . ,�p) of size at least 2 such that �1, �p and �i, �iþ1 (1� i� p� 1) each formfull paths. If all of the elements of W1 have not been used, then we pick an unusedelement of W1 and repeat the process. Since W1, . . . ,Wq all have the same size, eachof the elements of W1, . . . ,Wq will eventually be in one of the ordered sets. Since(v1, . . . , vq) was arbitrary, the result follows. g

We are now ready to prove Theorem 2.2.

Proof of Theorem 2.2 By Lemma 2.5, we know the theorem holds for k¼ 2 and thuswe may assume k� 3. Suppose n� t (mod k) where 0� t� k� 1. We will split theproof up into cases based on the value of t. If t� 2, we will break the process into tsteps. At step j, we find k-permanent sets for some of elements of Sn which start withcertain elements of P(t, j). The unused elements of Sn will start with certain elementsof P(t, jþ 1) and are grouped together in such a way as to make it possible for severalof them to become part of a k-permanent set at the next step.

First, we partition the elements of {tþ 1, tþ 2, . . . , n} into sets Z1,Z2, . . . ,Zl ofsize k where l ¼ n�t

k and we write Zi ¼ fzi1, z

i2, . . . , zikg.

274 A.H. Berliner

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 10: Convertible and               m               -convertible matrices

Suppose t¼ 0. We denote the elements of [n]nZi by fdi1, . . . , difg. We then form the

k-permanent set given by the following k elements of Sn (where the first two

permutations form the required full cycle):

�1 ¼ zi1 zi2 � � � zik�1 zik di1 � � � dif�1 dif�2 ¼ zi2 zi3 � � � zik di1 di2 � � � dif zi1

�3 ¼ zi3 zi2 � � � zik di1 di2 � � � dif zi1

..

. ... ..

. ... ..

.

�k ¼ zik zi2 � � � zik�1 di1 di2 � � � dif zi1

ð7Þ

The (n� 1)! simultaneous permutations of columns 2, . . . , n of (7) yield a simple

k-permanent set for each element of Sn beginning with one of zi1, . . . , zik. Since Zi was

arbitrary, we obtain one simple k-permanent set for each element of Sn.Suppose t¼ 1. We now denote the elements of [n]n({1}[Zi) by fd

i1, . . . , difg. We

then form the k-permanent set given by the following k elements of Sn:

�1 ¼ zi1 zik zi2 � � � zik�2 zik�1 di1 � � � dif�1 dif 1

�2 ¼ zi2 1 zi3 � � � zik�1 di1 di2 � � � dif zik zi1

�3 ¼ zi3 1 zi2 � � � zik�1 di1 di2 � � � dif zik zi1

..

. ... ..

. ... ..

.

�k ¼ zik 1 zi2 � � � zik�2 zik�1 di1 � � � dif�1 dif zi1

ð8Þ

The (n� 1)! simultaneous permutations of columns 2, . . . , n of (8) yield a simple

k-permanent set for each element of Sn beginning with one of zi1, . . . , zik. Since Zi was

arbitrary, we obtain a simple k-permanent set for each element of Sn not beginning

with 1.Each element of Sn that begins with 1 must, more specifically, begin with 1 zij for

some i2 {1, . . . , l} and j2 {1, . . . , k}. Consider each of the k-permanent sets resulting

from one of the (n� 2)! possible simultaneous permutations of columns 3, . . . , n of

(8). The final permutation in the list begins with zik 1. By our construction, we can

interchange the first two columns of this last row and not change the fact that the k

permutations form a k-permanent set. If we do this, then instead of being left to find

a simple k-permanent set for each element of Sn beginning with 1, we are left to find

a simple k-permanent set for each element of Sn beginning with 1 zij for each

i2 {1, . . . , l} and j2 {1, . . . , k� 1} or beginning with zik 1 for each i2 {1, . . . , l}.For each i2 {1, . . . , l}, we form the following k-permanent set given by the

following k elements of Sn:

�01 ¼ 1 zi1 zi2 � � � zik�2 zik�1 di1 � � � dif�1 dif zik�02 ¼ zik 1 zi3 � � � zik�1 di1 di2 � � � dif zi1 zi2

�03 ¼ 1 zi2 zi3 � � � zik�1 di1 di2 � � � dif zi1 zik

..

. ... ..

. ... ..

. ... ..

. ...

�0k ¼ 1 zik�1 zi2 � � � zik�2 di1 di2 � � � dif zi1 zik

ð9Þ

The (n� 2)! simultaneous permutations of columns 3, . . . , n of (9) yield a simple

k-permanent set for each element of Sn beginning with one of 1 zi1, . . . , 1 zik�1, zik 1.

Linear and Multilinear Algebra 275

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 11: Convertible and               m               -convertible matrices

Since i was arbitrary, we obtain a simple k-permanent set for each element of Sn

beginning with 1 zij for each i2 {1, . . . , l} and j2 {1, . . . , k� 1} or beginning with zik 1

for each i2 {1, . . . , l}. As these simple k-permanent sets are pairwise disjoint, we are

finished.Now, suppose t41. As we did in the case t¼ 0, we can find simple k-permanent

sets for each of the permutations in Sn beginning with one of tþ 1, . . . , n. Thus, we

are left with the permutations beginning with one of 1, . . . , t (the elements of P(t, 1)).

Writing this out, we are left with the permutations in Sn beginning with i j where i 6¼ j,

i2 [t], and j2 [n]. We now break the process into t steps:

Step 1 (t42) If t¼ 2, see step (t� 1) below. For each value of m2 [t], we denote the

elements of [n]n({m, m1}[Zi) by fdi1, . . . , difg (where m1¼ 1 if m¼ t and m1¼mþ 1

otherwise). We form the k-permanent set given by the following k elements of Sn:

�1 ¼ m zi1 zik m1 di1 � � � dif�1 dif zi2 � � � zik�2 zik�1�2 ¼ m1 zik m di1 di2 � � � dif zi2 zi3 � � � zik�1 zi1

�3 ¼ m zi2 zik m1 di1 � � � dif�1 dif zi3 � � � zik�1 zi1

..

. ... ..

. ... ..

. ... ..

. ... ..

.

�k ¼ m zik�1 zik m1 di1 � � � dif�1 dif zi2 � � � zik�2 zi1

ð10Þ

The (n� 2)! simultaneous permutations of columns 3, . . . , n of (10) yield a simple

k-permanent set for elements of Sn beginning with one of mzi1, . . . ,mzik�1,m1 zik.

Since Zi was arbitrary and since we do this for each value of m, we obtain a simple

k-permanent set for each element of Sn beginning with c d where c2 [t] and

d2 {tþ 1, . . . , n}. Thus, we are left to deal with permutations of Sn beginning with

one of the elements of P(t, 2).

Step j (2� j� t� 2) At the end of the previous step, we are left to find simple

k-permanent sets for the elements of Sn that begin with an element of P(t, j). We

partition P(t, j) into ordered sets according to Lemma 2.7. Fix one of the ordered sets

(�1, �2, . . . ,�p) and then consider �m and �mþ1 (where �pþ1¼�1) for any fixed value

of m. Without loss of generality, we can assume �m and �mþ1 to have the form

�m ¼ x1 x2 � � � xj

�mþ1¼ x2 x3 � � � xjþ1

for some distinct x1, . . . ,xjþ12 [t].We now denote the elements of [n]n({x1, . . . , xjþ1}[Zi) by fd

i1, . . . , difg. We form

the k-permanent set given by the following k elements of Sn:

�1 ¼ x1 � � � xj zi1 zik xjþ1 zi2 � � � zik�2 zik�1 di1 � � � dif�1 dif�2 ¼ x2 � � � xjþ1 zik x1 zi2 zi3 � � � zik�1 di1 di2 � � � dif zi1

�3 ¼ x1 � � � xj zi2 zik xjþ1 zi3 � � � zik�1 di1 di2 � � � dif zi1

..

. ... ..

. ... ..

. ... ..

. ... ..

. ...

�k ¼ x1 � � � xj zik�1 zik xjþ1 zi2 � � � zik�2 di1 di2 � � � dif zi1

ð11Þ

276 A.H. Berliner

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 12: Convertible and               m               -convertible matrices

The (n� j� 1)! simultaneous permutations of columns jþ 2, . . . , n of (11) yield a

simple k-permanent set for each element of Sn beginning with one of

�m zi1, . . . ,�m zik�1,�mþ1 zik

Since Zi was arbitrary and since we do this for each value of m, we obtain a simple

k-permanent set for each element of Sn beginning with �d where �2 {�1, . . . ,�p} andd2 {tþ 1, . . . , n}. Also, since we do this for each ordered set in the partition of P(t, j),

this process results in a simple k-permanent set for each element of Sn beginning with

�d where �2P(t, j) and d2 {tþ 1, . . . , n}. Thus, we are left with permutations of Sn

beginning with one of the elements of P(t, jþ 1).

Step (t� 1) At the end of the previous step, we are left with grouping the elements

of Sn that begin with one of the elements of P(t, t� 1). By Lemma 2.7, we can

partition P(t, j) into pairs which form full paths. Fix one of the pairs {�1,�2}.Without loss of generality, we can assume �1 and �2 to have the form

�1 ¼ x1 x2 � � � xt�1

�2 ¼ x2 x3 � � � xt

where x1, . . . , xt are the elements of [t].We now denote the elements of [n]n({x1, . . . , xt}[Zi) by fd

i1, . . . , difg and form the

k-permanent set given by the following k elements of Sn:

�1 ¼ x1 � � � xt�1 zi1 zik xt zi2 � � � zik�2 zik�1 di1 � � � dif�1 dif�2 ¼ x2 � � � xt zik x1 zi2 zi3 � � � zik�1 di1 di2 � � � dif zi1

�3 ¼ x1 � � � xt�1 zi2 zik xt zi3 � � � zik�1 di1 di2 � � � dif zi1

..

. ... ..

. ... ..

. ... ..

. ... ..

. ...

�k ¼ x1 � � � xt�1 zik�1 zik xt zi2 � � � zik�2 di1 di2 � � � dif zi1

ð12Þ

The (n� t)! simultaneous permutations of columns tþ 1, . . . , n of (12) yield a simple

k-permanent set for elements of Sn beginning with one of �1zi1, . . . ,�1z

ik�1,�2z

ik.

Swapping �1 and �2, we get a simple k-permanent set for each element of Sn

beginning with one of �2zi1, . . . ,�2z

ik�1,�1z

ik. Since Zi was arbitrary, we do this for

Z1, . . . ,Zl�1, but not for Zl¼ {y1, . . . , yk}. Thus, we are left with elements of Sn

beginning with �d where �2P(t, t� 1) and d2Zl.As above, we get a simple k-permanent set for each element of Sn beginning with

�1y1, . . . ,�1yk�1, or �2 yk. We then form the k-permanent set given by the following k

elements of Sn:

�1 ¼ x2 � � � xt y1 x1 y2 � � � yk�2 yk�1 yk d1 � � � df�1 df�2 ¼ x1 � � � xt�1 xt y2 y3 � � � yk�1 yk d1 d2 � � � df y1

�3 ¼ x2 � � � xt y2 x1 y3 � � � yk�1 yk d1 d2 � � � df y1

..

. ... ..

. ... ..

. ... ..

. ... ..

. ...

�k ¼ x2 � � � xt yk�1 x1 y2 � � � yk�2 yk d1 d2 � � � df y1

ð13Þ

Linear and Multilinear Algebra 277

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 13: Convertible and               m               -convertible matrices

The (n� t)! simultaneous permutations of columns tþ 1, . . . , n of (13) yields a simple

k-permanent set for elements of Sn beginning with one of �2y1, . . . ,�2yk�1, �1xt.Thus, for each of the pairs {�1, �2} we are left with finding k-permanent sets for the

elements of Sn beginning with either �1yk or �2x1.

Step t At the end of step t� 1, we are left with grouping the elements of Sn that

begin with �1yk or �2x1 for each pair {�1, �2}.We denote the elements of [n]n({x1, . . . , xt, yk}[Zi) by fd

i1, . . . , difg. We then form

the k-permanent sets:

�1 ¼ x1 � � � xt�1 yk zi1 zi2 � � � zik�2 zik�1 zik xt di1 � � � dif�1 dif

�2 ¼ x2 � � � xt x1 zik zi3 � � � zik�1 di1 yk zi2 di2 � � � dif zi1

�3 ¼ x1 � � � xt�1 yk zi2 zi3 � � � zik�1 di1 zik xt di2 � � � dif zi1

..

. ... ..

. ... ..

. ... ..

. ... ..

. ... ..

.

�k ¼ x1 � � � xt�1 yk zik�1 zi2 � � � zik�2 di1 zik xt di2 � � � dif zi1

ð14Þ

and

�01 ¼ x2 � � � xt x1 zi1 zi2 � � � zik�2 zik�1 zik yk di1 � � � dif�1 dif

�02 ¼ x1 � � � xt�1 yk zik zi3 � � � zik�1 di1 xt zi2 di2 � � � dif zi1

�03 ¼ x2 � � � xt x1 zi2 zi3 � � � zik�1 di1 zik yk di2 � � � dif zi1

..

. ... ..

. ... ..

. ... ..

. ... ..

. ... ..

.

�0k ¼ x2 � � � xt x1 zik�1 zi2 � � � zik�2 di1 zik yk di2 � � � dif zi1

ð15Þ

The (n� t� 1)! simultaneous permutations of columns tþ 2, . . . , n of (14) and (15)

yield a simple k-permanent set for each element of Sn beginning with �1yk z or �2x1zfor all z2Zi. Since Zi is arbitrary, we do this for Z1, . . . ,Zl�1, but not for

Zl¼ {y1, . . . , yk}. Thus, for each pair {�1,�2} we are left with elements of Sn

beginning with �1yk j1 or �2x1 j2, where j12 {xt, y1, . . . , yk�1} and j22 {y1, . . . , yk}.We now denote the elements of [n]n({x1, . . . , xt}[Zl) by {d1, . . . , df} and form the

k-permanent sets:

�1 ¼ x1 x2 � � � xt�1 yk y1 y2 � � � yk�2 yk�1 xt d1 � � � df�1 df

�2 ¼ x2 x3 � � � xt x1 yk y3 � � � yk�1 d1 y2 d2 � � � df y1

�3 ¼ x1 x2 � � � xt�1 yk y2 y3 � � � yk�1 d1 xt d2 � � � df y1

..

. ... ..

. ... ..

. ... ..

. ... ..

. ... ..

.

�k ¼ x1 x2 � � � xt�1 yk yk�1 y2 � � � yk�2 d1 xt d2 � � � df y1

ð16Þ

278 A.H. Berliner

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 14: Convertible and               m               -convertible matrices

and

�01 ¼ x2 x3 � � � xt x1 y1 y2 � � � yk�2 yk�1 yk d1 � � � df�1 df�02 ¼ x1 x2 � � � xt�1 yk xt y3 � � � yk�1 d1 y2 d2 � � � df y1

�03 ¼ x2 x3 � � � xt x1 y2 y3 � � � yk�1 d1 yk d2 � � � df y1

..

. ... ..

. ... ..

. ... ..

. ... ..

. ... ..

.

�0k ¼ x2 x3 � � � xt x1 yk�1 y2 � � � yk�2 d1 yk d2 � � � df y1

ð17Þ

The (n� t� 1)! simultaneous permutations of columns tþ 2, . . . , n of (16) and (17)yield a simple k-permanent set for each element of Sn beginning with �1 yk j1 or�2x1j2, where j12 {xt, y1, . . . , yk�1} and j22 {y1, . . . , yk}. If we do this for each pair{�1, �2}, then we have successfully found a simple k-permanent set for each elementof Sn. g

3. Maximal convertible matrices

We now turn our attention back to a problem involving 1-convertible matrices.Several questions regarding the permanents, ranks and number of nonzero entries ofconvertible matrices have been addressed in [6,11–13]. We say that a matrix A ismaximal convertible if A is convertible and if it not possible to change a 0 to a 1 andremain convertible.

Given an n� n matrix A, we construct an (nþ 1)� (nþ 1) matrix B in thefollowing way:

B ¼

1 a11 � � � a1n

1

0

..

.

0

A

2666666664

3777777775

The matrix B is called the ( first) row copy of the matrix A. Other rows (or columns)may be copied similarly. The following theorem can be found in [2].

THEOREM 3.1 Suppose A and B are (0, 1)-matrices such that B is a row (column) copyof A. Then A is maximal convertible if and only if B is maximal convertible.

For a matrix A, denote the number of nonzero entries of A by �(A). We define

H3 ¼

1 1 1

1 1 1

0 1 1

264

375

and for n43 we define Hn by taking the first row copy of Hn�1. We call Hn theHessenberg matrix of order n, and �(Hn)¼ (n2þ 3n� 2)/2. It is shown in [3,6] that upto column and row permutations, the matrix Hn is the unique convertible matrix oforder n with the maximum number of nonzero entries.

Linear and Multilinear Algebra 279

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 15: Convertible and               m               -convertible matrices

The problem of finding maximal convertible matrices with the fewest number of

nonzero entries is more complicated. Define �n to be the least number of nonzero

entries a maximal convertible (0, 1)-matrix of order n can have. It is shown in [2] that,

for n43,

3n � �n � 4n� 4 ð18Þ

In fact, if n45, the lower bound for �n can only hold if the matrix has three nonzero

entries in each row and column. Hwang et al. [7] note that finding an explicit formula

for �n seems quite difficult. Although it is not known if the limit exists, they raise the

question of finding

limn!1

�n

n

If the limit is defined then (18) implies that

3 � limn!1

�n

n� 4 ð19Þ

Our goal is to find a sequence of matrices that lowers the upper bound given in

(19). We will use the following lemma, which is proved in [12,15].

THEOREM 3.2 Suppose A and B are maximal convertible (0, 1)-matrices of orders j

and k (respectively) and are of the form

A ¼A1 a2

a1 1

" #and B ¼

1 b2

b1 B1

" #

Then the matrix

A ? B ¼A1 a2 0

a1 1 b2b1a1 b1 B1

24

35 ð20Þ

is a maximal convertible matrix of order jþ k� 1.

THEOREM 3.3 Suppose A is a maximal convertible (0, 1)-matrix of order k containing

at least two rows (columns) with exactly three nonzero entries. If �(A)¼ 3kþ l for some

non-negative integer l, then

lim infn!1

�n

n� 3þ

lþ 6

k

Proof Without loss of generality, the first and last rows of A¼ [aij] have exactly

three nonzero entries and akk¼ 1. Let B be the matrix of order kþ 1 obtained by

taking a first row copy of A. We now recursively define a sequence of matrices Mt of

order kt in the following way: let M1¼A, and for t41 we define Mt¼Mt�1 ?B.By Theorems 3.1 and 3.2, Mt is a maximal convertible matrix for all t.

By the construction in (20), we see that for any matrices A1, A2 we have

�ðA1 ? A2Þ ¼ �ðA1Þ þ �ðA2Þ � 1þ ð�ðlast row of A1Þ � 1Þð�ðfirst column of A2Þ � 1Þ

280 A.H. Berliner

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 16: Convertible and               m               -convertible matrices

We know that �(A)¼ 3kþ l and �(B)¼ 3kþ lþ 5. Since it is the case that the last row

of B has exactly three nonzero entries with bkþ1,1¼ 0 and bkþ1,kþ1¼ 1, we see by

induction that the last row of Mt has exactly three nonzero entries, one of which isthe last entry of the row. Therefore,

�ðMtþ1Þ ¼ �ðMtÞ þ ð3kþ lþ 5Þ � 1þ ð3� 1Þð2� 1Þ ¼ �ðMtÞ þ 3kþ lþ 6:

Thus, by our construction,

�ðMtÞ ¼ 3kþ lþ ðt� 1Þð3kþ lþ 6Þ ð21Þ

and

�ðMtÞ

kt¼

tð3kþ lþ 6Þ � 6

kt¼ 3þ

lþ 6

k�

6

kt

Therefore,

limt!1

�ðMtÞ

kt¼ lim

t!13þ

lþ 6

k�

6

kt

� �¼ 3þ

lþ 6

k

Thus, for n¼ kt, we have constructed matrices of order n with t(3kþ lþ 6)� 6

nonzero entries. Therefore,

lim infn!1

�n

n� 3þ

lþ 6

k

g

COROLLARY 3.4 lim infn!1�n

n �277 .

Proof Consider the maximal convertible matrix

A ¼

1 1 0 1 0 0 00 1 1 0 1 0 00 0 1 1 0 1 00 0 0 1 1 0 11 0 0 0 1 1 00 1 0 0 0 1 11 0 1 0 0 0 1

2666666664

3777777775

A satisfies the conditions of Theorem 3.3, where l¼ 0 and k¼ 7. g

4. Concluding remarks

The following result for convertible matrices easily extends to m-convertiblematrices.

PROPOSITION 4.1 Suppose A is a (0, 1)-matrix and B is a row (column) copy of A.Then A is m-convertible if and only if B is m-convertible.

Proof Without loss of generality, we assume that B is the first row copy of A. If B ism-convertible, it follows that A is m-convertible since I1 � A�B. Now suppose A is

m-convertible and consider an m-conversion A1, . . . ,Am of A. For each Ai, construct

Linear and Multilinear Algebra 281

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 17: Convertible and               m               -convertible matrices

the (nþ 1)� (nþ 1) matrix Bi as follows:

Bi ¼

1 Ai 1, 1½ � � � Ai 1, n½

�1

0

..

.

0

Ai

26666664

37777775

Then B1, . . . ,Bm is an m-conversion of B. g

A natural question is to ask how the notion of maximality extends to

m-convertible matrices for any particular value of m. As noted in Theorem 3.1, if

A is a maximal convertible matrix, then any row or column copy of A is also a

maximal convertible matrix. This allowed us to study the structure of such matrices

more effectively. However, it appears much more difficult to check if this result

extends to m-convertible matrices.

CONJECTURE 4.2 If A is a maximal m-convertible matrix, then any row (or column)

copy of A is also a maximal m-convertible matrix.

References

[1] A.H. Berliner, Determinants, permanents, and the enumeration of forest-partitions, Ph.D.

Thesis, University of Wisconsin–Madison, August 2009.

[2] R.A. Brualdi and B.L. Shader, On sign-nonsingular matrices and the conversion of the

permanent into the determinant, in Applied Geometry and Discrete Mathematics, Volume 4

of DIMACS Series on Discrete Mathematics and Theoretical Computer Science,

P. Gritzmann and B. Sturmfels, eds., AMS, Providence, RI, 1991, pp. 117–134.[3] R.A. Brualdi and B.L. Shader, Matrices of Sign-solvable Linear Systems, Volume 16 of

Cambridge Tracts in Mathematics, Cambridge University Press, Cambridge, 1995.[4] B. Codenotti and G. Resta, Computation of sparse circulant permanents via determinants,

Linear Algebra Appl. 355 (2002), pp. 15–34.[5] A. Galluccio and M. Loebl, On the theory of Pfaffian orientations. I. Perfect matchings and

permanents, Electron. J. Combin. 6 (1999), 18pp. (electronic).[6] P.M. Gibson, Conversion of the permanent into the determinant, Proc. Amer. Math. Soc.

27 (1971), pp. 471–476.[7] S.G. Hwang, S.-J. Kim, and S.-Z. Song, On maximal convertible matrices,

Linear Multilinear Alg. 38(3) (1995), pp. 171–176.[8] P.W. Kasteleyn, The statistics of dimers on a lattice, Physica 27 (1961), pp. 1209–1225.[9] P.W. Kasteleyn, Dimer statistics and phase transitions, J. Math. Phys. 4 (1963),

pp. 287–293.[10] P.W. Kasteleyn, Graph theory and crystal physics, in Graph Theory and Theoretical

Physics, F. Harary, ed., Academic Press, London, 1967, pp. 43–110.[11] S.-J. Kim, On convertible (0, 1) matrices, Linear Multilinear Alg. 42 (1997), pp. 319–322.

[12] S.-J. Kim and T.-Y. Choi, A note on convertible {0, 1} matrices, Commun. Korean Math.

Soc. 12 (1997), pp. 841–849.

[13] S.-J. Kim and T.-Y. Choi, A note on convertible (0, 1) matrices. II, Commun. Korean

Math. Soc. 14 (1999), pp. 311–318.

282 A.H. Berliner

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014

Page 18: Convertible and               m               -convertible matrices

[14] C.H.C. Little, A characterization of convertible (0, 1)-matrices, J. Combinatorial TheorySer. B 18 (1975), pp. 187–208.

[15] T.J. Lundy, J. Maybee, and J. Van Buskirk, On maximal sign-nonsingular matrices,Linear Algebra Appl. 247 (1996), pp. 55–81.

[16] G. Polya, Aufgabe 424, Arch. Math. Phys. 20 (1913), p. 271.[17] G. Ringel, Das geschlecht des vollstandingen paaren graphen, Abh. Math. Sem. Univ.

Hamburg 28 (1965), pp. 138–150.

[18] G. Tesler, Matchings in graphs on non-orientable surfaces, J. Combin. Theory Ser. B 78(2000), pp. 198–231.

[19] L.G. Valiant, The complexity of computing the permanent, Theoret. Comput. Sci. 8 (1979),

pp. 189–201.

Linear and Multilinear Algebra 283

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uthe

rn Q

ueen

slan

d] a

t 23:

16 0

4 O

ctob

er 2

014