Electronic Journal of Linear Algebra

Volume 30 Volume 30 (2015)

Article 22

2015

Evaluation of a family of binomial determinants

Charles Helou
Pennsylvania State University, cxh22@psu.edu

James A. Sellers
Pennsylvania State University, sellersj@psu.edu

Follow this and additional works at: http://repository.uwyo.edu/ela

Part of the Algebra Commons
Recommended Citation
Helou, Charles and Sellers, James A.. (2015), "Evaluation of a family of binomial determinants", Electronic Journal of Linear Algebra,
Volume 30.
DOI: http://dx.doi.org/10.13001/1081-3810.2889

This Article is brought to you for free and open access by Wyoming Scholars Repository. It has been accepted for inclusion in Electronic Journal of
Linear Algebra by an authorized administrator of Wyoming Scholars Repository. For more information, please contact scholcom@uwyo.edu.

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

EVALUATION OF A FAMILY OF BINOMIAL DETERMINANTS

CHARLES HELOU AND JAMES A. SELLERS

Abstract. Motivated by a recent work about finite sequences where the n-th term is bounded
by n2 , some classes of determinants are evaluated such as the (n 2) (n 2) determinant






 xn xk + h 1
 , for n 3,
n = 

2kn1
nk1


0hn3

and more generally the n n determinant




 


 xi + j 
,

Dn = 

1in
i

1


1jn

for n 1,

where n, k, h, i, j are integers, (xk )1kn is a sequence of indeterminates over C and

binomial coefficient. It is proven that
Dn = 1

and

n = (1)

(n2)(n3)
2

A

B

is the usual

Key words. Determinants, Binomial coefficients, Row reduction.

AMS subject classifications. 11C20, 05A10, 11B65, 15B36.

1. Introduction. In a recent work [9], about finite sequences whose n-th term
does not exceed n2 , there appeared the determinant



 2

2


n k +h1
,
n = 
2kn1 
nk1

0hn3

with an integer n 3. One of the authors of [9], L. Haddad, conjectured after some
computations that n = 1. The authors of the present paper first proved that
n = (1)

(n2)(n3)
2

essentially by a process of row reduction. Then, following a suggestion by G.E. Andrews that this result should be true in a more general context, namely upon replacing
Received

by the editors on January 30, 2015. Accepted for publication on May 19, 2015.
Handling Editor: Joao Filipe Queiro.
Department of Mathematics, Penn State Brandywine, 25 Yearsley Mill Road, Media, PA 19063,
USA (cxh22@psu.edu).
Department of Mathematics, Penn State University Park, 104 McAllister Building, University
Park, PA 16802, USA (sellersj@psu.edu).
312

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

Evaluation of a Family of Binomial Determinants

313

n2 by xn and k 2 by xk , where (xk )1kn is an arbitrary sequence of indeterminates

over C, the proof was extended to this general case. We then realized that the problem
can be reduced to the evaluation of a simpler, more general family of determinants,
namely





 xi + j

,
Dn = 
1in 
i1

1jn

for all integers n 1. In what follows, we will establish that Dn = 1 and deduce that
(n2)(n3)
2
n = (1)
.
Results of a similar nature, involving determinants of matrices whose entries
involve binomial coefficients, can be found in [1, 2, 3, 4, 6, 10, 12]. In contrast to
these papers, we note that our determinant evaluations are strikingly simple and easy
to state. In fact, our result is a special case of the results contained in [10], but our
proof is more elementary, using only row reduction and induction.
We are thankful to L. Haddad for his conjecture and to G.E. Andrews for his
insightful suggestion.
2. The method of proof. First recall (e.g., [7] or [8]) that for an indeterminate


x over C and an integer n 0, the binomial coefficient nx is defined by
 
x (x 1) (x 2) (x n + 1)
x
,
=
n!
n

with the convention that

(2.1)

x
0

= 1. It satisfies the fundamental recurrence relation

  
 

x
x
x+1
+
=
,
n
n+1
n+1

for all n 0.

Let (xk )1kn be a sequence of indeterminates over C, with n 3, and consider the
(n 2) (n 2) determinant






 xn xk + h 1
.

n = n (x1 , . . . , xn ) = 
2kn1 
nk1

0hn3

First, setting i = k 1 and j = h + 1 allows one to rewrite n as







 xn xi+1 + j 2
.
n = 
1in2 
ni2

1jn2

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

314

C. Helou and J.A. Sellers

Second, the substitution i = n i 1 transforms n into








xn xni + j 2
,
n = n (x1 , . . . , xn ) = 

i 1
1i n2 

1jn2

which has the same rows as n but in reverse order. This order reversal consists in
respectively swapping each row of n with all the rows above it. The total number
of those row swaps is
(n 3) + (n 4) + + 2 + 1 =

(n 2)(n 3)
.
2

Therefore,
(2.2)

n = (1)

(n2)(n3)
2

n .

The problem is thus reduced to the determination of n .

Third, setting x = xn 2 gives







x

x
+
j
ni


.
n = 
1in2 
i1

1jn2

Fourth, setting yi = x xni yields








yi + j
.
n = 
1in2 
i1

1jn2

Finally, setting m = n 2 leads to the equality








yi + j

.

(2.3)
n = 
1im 
i1

1jm

The problem is thus reduced to the evaluation of the family of determinants







 xi + j
,

Dn = Dn (x1 , . . . , xn ) = 

1in
i1


1jn

for n 1, where x1 , x2 , . . . , xn are arbitrary indeterminates.

Our primary result is now the following:
Theorem 2.1. For any positive integer n, we have
Dn = 1.

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

Evaluation of a Family of Binomial Determinants

315

The proof proceeds by row reduction and is presented in the next section.
Corollary 2.2. For any integer n 3, we have
n = 1.

Proof. This follows from (2.3) and Theorem 2.1.

Corollary 2.3. For any integer n 3, we have
n = (1)

(n2)(n3)
2

Proof. This follows from (2.2) and Corollary 2.2.

Remark 2.4. An alternative method for deriving the last result is to set yk =
xn xnk 2, and i = k 1, j = h + 1. Then






 yni1 + j
.
n = 
1in2 
ni2

1jn2

Now, reversing the order of the rows, which consists in replacing i by n 1 i,

transforms n into







yi + j


.
n = 
1in2 
i1

1jn2

Moreover, the permutation that reverses the n 2 rows of n has n2

2 orbits,
namely {1, n 2}, {2, n 3}, . . . It follows that (see, e.g., [11]) the sign of is
() = (1)

n2
n2
2 2

= (1)

n2
2

Hence,
n = (1)

n2
2

n = (1)

n2
2

in view of our main theorem, which yields n = 1.

3. The proof of the main theorem. We start with two results about binomial
coefficients that will be used in the proof of Theorem 2.1.
Lemma 3.1. For any integers 0 a b and n 1, and any indeterminate x
over C, we have

 
 X

b1 
x+b
x+a
x+h

=
.
n
n
n1
h=a

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

316

C. Helou and J.A. Sellers

Proof. By the fundamental recurrence relation for binomial coefficients (2.1),


   

x+1
x
x

=
,
n
n
n1
for n 1. Hence, by iterating this recurrence numerous times, we have

 
 X
 
 X

b1 
b1 
x+b
x+a
x+h+1
x+h
x+h

=
.
n
n
n
n
n1
h=a

h=a

Lemma 3.2. For any integers 0 m n, we have



n  
X
k
n+1
=
.
m
m+1

k=m

Proof. For a fixed integer m 0, the proof can be completed by induction on

n m. Such an argument can be found in [5, p. 138]. This result also follows from
Lemma 3.1 by taking n = m + 1, a = 0, x = m, b = n m + 1.

i.e.,

We now proceed to prove Theorem 2.1 by row reduction. We start with







xi + j

Dn = (dij ) 1in  , where dij =
for 1 i, j n,
i1
1jn










Dn = 









x2 +1
1

x3 +1
2

x4 +1

x2 +2
1

x3 +2
2

x4 +2

x2 +3
1

x3 +3
2

x4 +3

xi +1
i1

xi +2
i1

xi +3
i1

..
.

..
.

xn +1
n1

..
.

..
.

xn +2
n1

..
.

..
.

xn +3
n1

1


x2 +j
1

x3 +j
2

x4 +j
3

..
.

xi +j
i1

..
.

xn +j
n1






x2 +n 
1

x3 +n 
2

x4 +n 

3
..  .
. 


xi +n 
i1 
.. 
. 


xn +n 
1

n1

Denoting the i-th row of Dn by Ri , the first row reduction step consists in replacing Ri by Ri di1 R1 for 2 i n. This gives







Dn =  dij 1in  ,
1jn

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

317

Evaluation of a Family of Binomial Determinants

where
(3.1)

dij

xi +j
i1

dij di1 =
d1j = 1,

xi +1
i1

Pj1

h=1

xi +h
i2

, if 2 i n, 1 j n,
if i = 1, 1 j n,

the last expression, for i 2, is obtained by using Lemma

vention that an empty sum is equal to 0. Thus,

1
1
1

1

0
1
2

j1


P2


Pj1 x3 +h


x3 +1
x3 +h
0

h=1
h=1
1
1
1


P2



Pj1
x4 +1
x4 +h
x4 +h
0

h=1
h=1
2
2
2

..
..
..
Dn =  ..
.
.
.
.




P2
Pj1 xi +h

0
xi +1
xi +h


h=1 i2
h=1 i2
i2
.
..
..
..
.
.
.
.
.





P
P
2
j1 xn +h

xn +1
xn +h
0

h=1 n2
h=1 n2
n2

Moreover, we obviously have

3.1, with the usual con-





Pn1 x3 +h


h=1
1

Pn1
x4 +h 
h=1
2

.
..

.
Pn1 xi +h

h=1 i2 

..

.

Pn1 xn +h


h=1 n2
1
n1

Dn = Dn .
Proposition 3.3. For 1 k n, let







(k)
Dn(k) =  dij 1in 


1jn

be the n n determinant obtained from Dn by applying k row reduction steps, each of

(k1)
(k1)
of the determinant Dn
, obtained
which consists in replacing the i-th row Ri
after k 1 such row reduction steps, by the row
(k)

Ri

(k1)

= Ri

(k1)

dik

(k1)

Rk

for k + 1 i n,

while the first k rows are unchanged, i.e.,

(k)

Ri

(k1)

= Ri

for 1 i k.

Then
(k)
dij

P
jk

h=1

d(k1) ,
ij

jh1
k1

xi +h
ik1

,

if k + 1 i n, 1 j n,
if 1 i k, 1 j n,

with the convention that an empty sum (here, when j k) is equal to 0.

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

318

C. Helou and J.A. Sellers

Proof. The proof is by induction on k. The first row reduction step was applied




to Dn right before this Proposition, and it gave Dn =  dij 1in , which satisfies
1jn
the stated equalities for dij as shown in (3.1) above. So the property holds for k =
(k1)

1.
 Assume that it holds for k 1, where 2 k n, i.e., assume that Dn




 (k1)
 satisfies
 dij
1in


1jn
P


j(k1) jh1

xi +h

h=1
k2
i(k1)1 , if k i n, 1 j n,
(k1)
=
dij

d(k2) ,
if 1 i k 1, 1 j n.
ij
(k)

Now, as for k + 1 i n, we have Ri

(k1)

(k)

dij = dij

(k1)

= Ri

(k1) (k1)
dkj ,

dik

(k1)

dik

(k1)

Rk

, i.e.,

for 1 j n,

and by the induction assumption, there hold

jk+1
X j h 1xi + h
(k1)
,
=
dij
ik
k2
h=1

(k1)
dik

kk+1
X 
h=1

(k1)
dkj

jk+1
X 
h=1

kh1
k2


 

xi + h
xi + 1
=
,
ik
ik


 jk+1
X j h 1
j h 1 xk + h
=
.
k2
kk
k2
h=1

Therefore, we get
(k)
dij

jk+1
X 


 jk+1
X j h 1xi + 1
j h 1 xi + h
=

k2
ik
k2
ik
h=1
h=1
jk+1
X j h 1 xi + h xi + 1
=

.
k2
ik
ik
h=1

Moreover, by Lemma 3.1,


 
 h1
X  xi + r 
xi + h
xi + 1
,

=
ik1
ik
ik
r=1
for i > k and h 1. Hence,
jk+1
X j h 1
X  xi + r  jk+1
X j h 1 h1
X  xi + r  jk
(k)
.
=
dij =
ik1
k2
ik1
k2
r=1
r=1
h=1

h=r+1

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

Evaluation of a Family of Binomial Determinants

319

Furthermore, by Lemma 3.2,

jk+1
X 

h=r+1

jh1
k2

jr2
X 

s=k2

 

s
jr1
=
.
k2
k1

Thus,
(k)

dij =



jk 
X
jr1
xi + r
,
k1
ik1
r=1

for k + 1 i n and 1 j n.
(k)

Also, for 1 i k, since Ri

(k)

(k1)

dij = dij

(k1)

= Ri

, we have

for 1 i k, 1 j n.

This shows that the property holds for k, and completes the induction.
Corollary 3.4. For 1 k n, the determinant







(k)
Dn(k) =  dij 1in 


1jn

obtained from Dn by applying k row reduction steps as described in Proposition 3.3 is

given by

j1
,
if 1 i k, 1 j n,
i1
(k)
dij =

Pjk jh1
xi +h
, if k + 1 i n, 1 j n,
h=1

k1

ik1

where if 0 m < n are integers then

m
n

(k)

= 0, and an empty sum is equal to 0.

Proof. Only the expression of dij for 1 i k and 1 j n needs to be

proved. The rest is contained in Proposition 3.3. This expression holds for k = 1
since by (3.1)


j1
d1j = d1j = 1 =
, for 1 j n.
11
Assume that the expression holds for k 1, where 2 k < n, i.e.,

j1
if 1 i k 1, 1 j n,
i1 ,
(k1)
=
dij

Pjk+1 jh1
xi +h
, if k i n, 1 j n.

h=1

k2

ik

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

320

C. Helou and J.A. Sellers

Then, by Proposition 3.3 and Lemma 3.2, we have



j1
(k1)
(k)
=
dij = dij
, for 1 i k 1, 1 j n,
i1
and
(k)

jk+1
X 


 jk+1
X j h 1
j h 1 xk + h
=
k2
0
k2
h=1
h=1
 

j2 
X
s
j1
=
=
, for 1 j n.
k2
k1
(k1)

dkj = dkj

s=k2

Hence,
(k)

dij =



j1
,
i1

for 1 i k, 1 j n,

and the expression holds for k.

Remark 3.5. We have
Dn = Dn(k) ,
since the determinant is invariant under the row reduction steps consisting of adding
to a row a multiple of another row. In particular,







j1

Dn(n) = 
1in 
i1

1jn

is the determinant of an upper triangular n n matrix whose diagonal entries are

(n)
dii = i1
i1 = 1 for 1 i n. Therefore,
Dn = Dn(n) = 1.

This concludes the proof of Theorem 2.1.

Remark 3.6. As noted in the introduction, our result is a special case of the


j +x
results contained in [10]. Indeed, in [10], Proposition 1, taking pj (x) = xj1
, which
1
, for 1 j n,
is a polynomial of degree j 1 in x, with leading coefficient aj = (j1)!
we get





n
Y
 Y
 xj + Xi
1
=


(Xj Xi ) ,


1in 
j1
(j 1)!

j=1
1jn

1i<jn

Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society
Volume 30, pp. 312-321, July 2015

ELA

http://math.technion.ac.il/iic/ela

Evaluation of a Family of Binomial Determinants

then specializing to Xi = i for 1 i n, we get




 Y

n
 xj + i

1

=



1in 
(j 1)!
j1

j=1
1jn

321

(j i) = 1.

1i<jn

Acknowledgment. We are thankful to the referee for a careful, thorough reading

of the paper, and for many helpful and interesting suggestions.
REFERENCES

[1] T. Amdeberhan and D. Zeilberger. Determinants through the looking glass. Adv. in Appl.
Math. (Special issue in honor of Dominique Foatas 65th birthday), 27(2-3):225230, 2001.
[2] G.E. Andrews. Pfaffs method. I. The Mills-Robbins-Rumsey determinant. Discrete Math.
(Selected papers in honor of Adriano Garsia), 193(1-3):4360, 1998.
[3] G.E. Andrews and W.H. Burge. Determinant identities. Pacific J. Math., 158(1):114, 1993.
[4] G.E. Andrews and D.W. Stanton. Determinants in plane partition enumeration. European J.
Combin., 19(3):273282, 1998.
[5] R. Brualdi. Introductory Combinatorics, fifth edition. Pearson Prentice Hall, 2010.
[6] W. Chu and L.V. di Claudio. Binomial determinant evaluations. Ann. Comb., 9(4):363377,
2005.
[7] L. Comtet. Advanced Combinatorics. D. Reidel Publishing Co., Dordrecht, 1974.
[8] R. Graham, D. Knuth, and O. Patashnik. Concrete Mathematics, second edition. AddisonWesley, 1994.
[9] L. Haddad and C. Helou. Finite sequences dominated by the squares. J. Integer Seq., 18:15.1.8,
2015.
[10] C. Krattenthaler. Advanced determinant calculus. The Andrews Festschrift (Maratea, 1998),
Sem. Lothar. Combin., 42:Article B42q, 1999.
[11] S. Nelson. Defining the sign of a permutation. Amer. Math. Monthly, 94(6):543545, 1987.
[12] A.M. Ostrowski. On some determinants with combinatorial numbers. J. Reine Angew. Math.,
216:2530, 1964.