On a practically solvable product-type system of difference equations of second order
Stevo Stevi´c
B1, 2and Dragana Rankovi´c
31Mathematical Institute of the Serbian Academy of Sciences Knez Mihailova 36/III, 11000 Beograd, Serbia
2King Abdulaziz University, Department of Mathematics, P.O. Box 80203, Jeddah 21589, Saudi Arabia
3Department of Physics and Mathematics, Faculty of Pharmacy, University of Belgrade Vojvode Stepe 450, Belgrade, Serbia
Received 20 May 2016, appeared 26 July 2016 Communicated by Leonid Berezansky
Abstract. The problem of solvability of the following second order system of difference equations
zn+1=αzanwbn, wn+1=βwcnzdn−1, n∈N0, wherea,b,c,d∈Z,α,β∈C\ {0},z−1,z0,w0∈C\ {0}, is studied in detail.
Keywords: system of difference equations, second order system, product-type system, practically solvable system.
2010 Mathematics Subject Classification: 39A10, 39A20.
1 Introduction
Investigation of various concrete nonlinear difference equations and systems has been of a great recent interest (see, e.g., [1–5], [10–35] and the references therein). Here we mention two subareas of interest. First, motivated by some classes of concrete nonlinear difference equations some experts started investigating the corresponding symmetric or cyclic systems of difference equations, as well as some modifications of the symmetric/cyclic systems (see, e.g., [4,10–12,14,15,19–21,23–27,29–33,35]). Second, one of the basic problems related to difference equations and systems is the problem of their solvability. Books [7–9] contain numerous classical results on the topic. Solvable difference equations and systems have appeared in the literature from time to time for a long time. It has turned out that some of recently studied concrete nonlinear difference equations and systems are solvable. For example, the solvability of systems in [4] and [21] was discovered during the investigation of the long-term behavior of their solutions. On the other hand, some papers which give solutions of quite concrete equations and systems, without any theoretical explanations, have appeared recently. One of the first papers of this type is [5]. These facts motivated some experts to work on the problem
BCorresponding author. Email: sstevic@ptt.rs
of solvability more seriously (see, e.g., [4,13,16,17], [19]-[30], [32–35]). Short note [16] can be regarded as a starting point for the serious investigation, and its importance is found in the method which is used for showing the solvability of an extension of the equation in [5].
The method was later used numerous times (see, e.g., [1,13,17,22,34]) and was essentially extended and developed in many of the other above mentioned papers on solvability.
Another fact of interest is to note that many difference equations and systems are ob- tained from product-type ones by some modifications of their right-hand sides (e.g., the equa- tions/systems in [18] and [31] are of this type). Product-type equations and systems are solvable for the case of positive coefficients and initial values. However, this is not always the case if some of them are real or complex numbers. Hence, the problem of their solvability in these cases is of some interest. An investigation in this direction was started by S. Stevi´c and some of his collaborators in [27], [29], [30] and [35], where some product-type systems are solved not only theoretically, but essentially also practically, that is, by presenting concrete formulas for solutions depending on the initial values and involving parameters.
This paper is devoted to the study of solvability of the following second order system of difference equations
zn+1 =αzanwbn, wn+1= βwcnzdn−1, n∈N0, (1.1) where a,b,c,d ∈ Z, α,β ∈ Cand z−1,z0,w0 ∈ C. Motivated by [27] system (1.1), unlike the ones in papers [29], [30] and [35], has some new parameters, namely, the coefficientsαandβ.
Since the casesα=0 orβ =0 are very simple, so of not much interest, from now on we will assume thatα,β∈C\ {0}.
Note that the domain of undefinable solutions [24] to system (1.1) is a subset of the fol- lowing set
U = {(z−1,z0,w0)∈ C3:z−1=0 or z0 =0 or w0=0}.
Thus we can regard that(z−1,z0,w0)belong toC3\ U, although in some cases C3\ U can be even equal toC3(for example, if a,b,canddare natural numbers).
For a system of difference equations of the form
zn = f(zn−1, . . . ,zn−k,wn−1, . . . ,wn−l)
wn =g(zn−1, . . . ,zn−s,wn−1, . . . ,wn−t), n∈ N0,
wherek,l,s,t ∈ N, is said that it issolvable in closed form if its general solution can be found in terms of initial valuesz−i,i=1, max{k,s},w−j,j=1, max{l,t}, delaysk,l,s,t, and index n only.
Let us also say that, as usual, the sums of the form∑mj=laj, for m< l, will be regarded to have value equal to zero.
2 Auxiliary results
Several auxiliary results, which will be used in the proofs of the main results are given in this section. The first lemma is well-known (see, e.g., [9]).
Lemma 2.1. Let i∈N0and
s(ni)(z) =1+2iz+3iz2+· · ·+nizn−1, n∈N, where z∈C.
Then
s(n0)(z) = 1−zn
1−z , (2.1)
s(n1)(z) = 1−(n+1)zn+nzn+1
(1−z)2 , (2.2)
for every z∈ C\ {1}and n ∈N.
Remark 2.2. Since by the above mentioned conventions(ni)(z) =∑nj=1jizj−1=0, forn=−1, 0, a simple calculation shows that (2.1) also holds for n = 0 and z 6= 0, while (2.2) holds for n=−1 andn=0 whenz 6=0. Note also thats(n1)(1) = n(n2+1).
Lemma 2.3. Letαi,βi, i=0,n, be complex numbers. Then
∑
n i=0αi
n−i j
∑
=0βj =
∑
n j=0βj
n−j i
∑
=0αi, (2.3)
for every n∈N0. Proof. Let
Sn={(i,j): 0≤i≤ n, 0≤ j≤n−i}.
From the definition of set Sn we see that variable jtakes all the values from 0 to n too, and that for a fixed j, from the inequalityj≤n−iwe have thati≤n−j, that is, the upper bound foriis equaln−j. Hence
Sn ={(i,j): 0≤ j≤n, 0≤i≤ n−j}.
Using this fact and by the changing order of summation, equality (2.3) easily follows.
Remark 2.4. Note that Lemma 2.3 is a simple sort of the Fubini theorem with the discrete measure.
Let
fn(u,v) =
∑
n i=0ui
n−i j
∑
=0vj, (2.4)
whereu,v∈Candn∈N0, and where we in this case, as usual, regard that 00=1.
Lemma 2.5. Let fn(u,v)be defined in(2.4). Then
fn(u,v) = fn(v,u), (2.5)
for every u,v∈Cand n ∈N0.
Moreover, the following formulas hold.
(a) If u6=16=v and u 6=v, then
fn(u,v) = v−u+un+2−vn+2+uvn+2−vun+2
(1−v)(1−u)(v−u) , n∈ N0. (2.6)
(b) If u=v6=1, then
fn(u,u) = 1−(n+2)un+1+ (n+1)un+2
(1−u)2 , n∈N0. (2.7)
(c) If u6=1and v=1, then
fn(u, 1) = n+1−(n+2)u+un+2
(1−u)2 , n∈N0. (2.8)
(d) If u=1and v 6=1, then
fn(1,v) = n+1−(n+2)v+vn+2
(1−v)2 , n∈N0. (2.9)
(e) If u =v=1, then
fn(1, 1) = (n+1)(n+2)
2 , n∈N0. (2.10)
Proof. By using Lemma2.3withαi =ui,βj =vj,i,j=0,n, equality (2.5) follows.
(a)By using the formula for the sum of a geometric progression three times (see (2.1)), the conditionsu6=16= vandu6=v, and some simple calculation, for the casev6=0, we obtain
fn(u,v) =
∑
n i=0ui
n−i
∑
j=0vj =
∑
n i=0ui
1−vn−i+1 1−v
= 1 1−v
n
i
∑
=0ui−vn+1
∑
n i=0u v
i
= 1 1−v
1−un+1
1−u −vvn+1−un+1 v−u
= v−u+un+2−vn+2+uvn+2−vun+2 (1−v)(1−u)(v−u) . Ifv=0 andu6=1, then
fn(u, 0) =
∑
n i=0ui = 1−un+1
1−u = −u+un+2
(1−u)(−u), (2.11)
which is nothing but formula (2.6) whenv=0.
(b) By using formula (2.1) twice, the conditions u = v 6= 1, and some simple calculation, we obtain
fn(u,u) =
∑
n i=0ui
n−i j
∑
=0uj =
∑
n i=0ui
1−un−i+1 1−u
= 1 1−u
n
i
∑
=0ui−un+1
∑
n i=01
= 1 1−u
1−un+1
1−u −(n+1)un+1
= 1−(n+2)un+1+ (n+1)un+2 (1−u)2 .
(c)By using formula (2.1), the conditionsu 6= 1 and v = 1, formula (2.2) withz = u, and some simple calculation, we obtain
fn(u, 1) =
∑
n i=0ui
n−i
∑
j=01=
∑
n i=0ui(n−i+1)
= (n+1)
∑
n i=0ui−u
∑
n i=1iui−1
= (n+1)1−un+1
1−u −u1−(n+1)un+nun+1 (1−u)2
= n+1−(n+2)u+un+2 (1−u)2 , as desired.
(d) By using equality (2.5) withu = 1, we get fn(1,v) = fn(v, 1). From this and by (2.8) with u→v, formula (2.9) follows.
(e)We have
fn(1, 1) =
∑
n i=0n−i
∑
j=01=
∑
n i=0(n−i+1) =
n+1 j
∑
=1j= (n+1)(n+2)
2 .
Remark 2.6. If we note that fn(u,u) =
∑
n i=0ui
n−i j
∑
=0uj =
∑
n i=0n−i
∑
j=0ui+j =
∑
n l=0(l+1)ul = s(n1+)1(u),
since the equation i+j=l hasl+1 nonnegative integer solutions, formula (2.7) also follows from (2.2). It is also easy to see that fn(u, 1) = uns(n1+)1(1u), u 6= 0, from which along with (2.2), formula (2.8) can be obtained. Note also that by using the above mentioned summing convention and some simple calculation is obtained that formulas (2.6), (2.8)–(2.10) holds for n=−1 (formula (2.7) holds forn=−1 ifu 6=0).
The following result is also known (for example, it is a consequence of the Lagrange interpolation formula).
Lemma 2.7. If all the zeros zj, j=1,k,of the polynomial
Pk(z) =γkzk+γk−1zk−1+· · ·+γ1z+γ0, are such that zi 6=zj, i6= j, then the following formulas hold:
∑
k j=1zlj Pk0(zj) =0 for every l∈ {0, 1, . . . ,k−2}, and
∑
k j=1zkj−1 Pk0(zj) = 1
γk.
3 Main results
The main results in this paper are proved in this section. The first one is devoted to the case b=0.
Theorem 3.1. Assume that a,c,d ∈ Z, b = 0, α,β ∈ C\ {0} and z−1,z0,w0 ∈ C\ {0}. Then system(1.1)is solvable in closed form.
Proof. In this case system (1.1) becomes
zn+1 =αzan, wn+1= βwcnzdn−1, n∈N0. (3.1) From the first equation in (3.1) it is not difficult to see that
zn =α∑
n−1
i=0 aiza0n, n∈N. (3.2)
From (3.2) we easily obtain that fora6=1 zn=α
1−an
1−a z0an, n∈N, (3.3)
while fora=1, we have
zn= αnz0, n∈N. (3.4)
By using (3.2) into the second equation in (3.1), we get wn= βαd∑
n−3
i=0 aizda0 n−2wcn−1, forn≥3. (3.5) Hence, by using (3.5) with n→n−1, we have that
wn= βαd∑
n−3
i=0 aizda0 n−2 βαd∑
n−4
i=0 aizda0 n−3wcn−2c
= β1+cαd∑
n−3
i=0 ai+dc∑ni=−04aizda0 n−2+dcan−3wcn2−2, (3.6) forn≥4.
Based on (3.5) and (3.6), assume that for some k≥2 we have proved wn= β∑
k−1 i=0ci
αd∑
k−1
j=0 cj∑ni=−0j−3ai zd∑k
−1 i=0cian−i−2
0 wcnk−k, (3.7)
forn≥ k+2.
Then by using (3.5) withn→n−kinto (3.7), we get wn =β∑
k−1 i=0ci
αd∑
k−1
j=0 cj∑ni=−0j−3ai zd∑
k−1 i=0cian−i−2
0 βαd∑
n−k−3
i=0 aizda0 n−k−2wcn−k−1ck
=β∑
k i=0ci
αd∑
kj=0 cj∑in=−0j−3ai
z0d∑ki=0cian−i−2wcnk−+1k−1 (3.8) forn≥ k+3.
From (3.5), (3.8) and the method of induction we see that (3.7) holds for all natural numbers kandnsuch that 1≤k ≤n−2.
By takingk=n−2 into (3.7) we get wn= β∑
n−3 i=0ci
αd∑
n−3
j=0 cj∑ni=−0j−3ai zd∑n
−3 i=0cian−i−2
0 wc2n−2, forn≥3. (3.9)
On the other hand, from the second equation in (3.1) with n=1, we have
w2= βwc1z0d= β(βwc0zd−1)czd0= β1+cwc02zcd−1zd0. (3.10) From (3.9) and (3.10), we get
wn =β∑
n−3 i=0 ci
αd∑
n−3
j=0 cj∑ni=−0j−3ai zd∑
n−3 i=0 cian−i−2
0 β1+cwc02zcd−1zd0cn−2
=β∑
n−1 i=0 ci
αd∑
n−3
j=0 cj∑ni=−0j−3ai
wc0nzd∑n
−2 i=0 cian−i−2
0 zdc−1n−1, (3.11)
forn≥3.
Now the subcasesa6=canda= cwill be considered separately.
Subcase a6= c.In this case from (3.11), we get wn= β∑
n−1 i=0 ci
αd∑
n−3
j=0 cj∑ni=−0j−3ai wc0nzd
an−1−cn−1 a−c
0 zdc−1n−1, n≥2. (3.12) Ifa6=1 andc6=1, then by formula (2.6) withn→n−3, u=candv= a, (3.12) becomes
wn= β
1−cn 1−cαd
a−c+cn−1−an−1+can−1−acn−1 (1−a)(1−c)(a−c) wc0nzd
an−1−cn−1 a−c
0 zdc−1n−1, n≥2. (3.13) Ifa6=canda=1, then by using formula (2.8) withn→n−3,u=candv=1, we get
wn =β
1−cn 1−c αd
n−2−(n−1)c+cn−1 (1−c)2 wc0nzd
1−cn−1 1−c
0 zdc−1n−1, n≥2, (3.14) while ifa6=candc=1, then by using formula (2.9) with n→n−3,u=1 andv=a, we get
wn =βnαd
n−2−(n−1)a+an−1 (1−a)2 w0zd
an−1−1 a−1
0 zd−1, n≥2. (3.15)
Subcase a=c. In this case from (3.11), we get wn= β∑
n−1 i=0 ai
αd∑
n−3
j=0 aj∑ni=−0j−3ai
wa0nzd0(n−1)an−2zda−1n−1, (3.16) forn≥3.
Ifa=c6=1, then by using formula (2.7) withn→n−3 andu=v =a, (3.16) becomes wn =β
1−an 1−a αd
1−(n−1)an−2+(n−2)an−1
(1−a)2 wa0nzd0(n−1)an−2zda−1n−1, n≥3, (3.17) while ifa=c=1, then by using formula (2.10) withn→n−3 andu=v =1, we get
wn= βnαd
(n−2)(n−1)
2 w0zd0(n−1)zd−1, n≥3, (3.18) finishing the proof of the theorem.
Corollary 3.2. Consider system (1.1) with a,c,d ∈ Z, b = 0 and α,β ∈ C\ {0}. Assume that z−1,z0,w0∈C\ {0}. Then the following statements are true.
(a) If a6=c, a 6=1and c6=1then the general solution to system(1.1)is given by(3.3)and(3.13).
(b) If a6=c and a =1then the general solution to system(1.1)is given by(3.4)and(3.14).
(c) If a6= c and c=1then the general solution to system(1.1)is given by(3.3)and(3.15).
(d) If a=c6=1then the general solution to system(1.1)is given by(3.3)and(3.17).
(e) If a=c=1then the general solution to system(1.1)is given by(3.4)and(3.18).
Theorem 3.3. Assume that a,b,c ∈ Z, d = 0, α,β ∈ C\ {0} and z−1,z0,w0 ∈ C\ {0}. Then system(1.1)is solvable in closed form.
Proof. A proof of the theorem was essentially given in [27], but we give a slightly modified for the completeness. In this case system (1.1) becomes
zn+1 =αznawbn, wn+1 = βwcn, n∈N0. (3.19) From the second equation in (3.19) we have that
wn=β∑
n−1
i=0 ciw0cn, n∈N. (3.20)
From (3.20) we easily obtain that forc6=1 wn= β
1−cn
1−cwc0n, n∈N, (3.21)
while forc=1, we have
wn= βnw0, n∈N. (3.22)
Employing (3.20) into the first equation in (3.19), we get zn =αβb∑
n−2
i=0 ciwbc0n−1zan−1, (3.23) forn≥2.
Hence, by using (3.23) with n→n−1 into itself, we have zn =αβb∑
n−2
i=0 ciwbc0n−1 αβb∑
n−3
i=0 ciw0bcn−2zan−2a
,
=α1+aβb∑
n−2
i=0 ci+ba∑ni=−03ciw0bcn−1+bacn−2zan2−2, (3.24) forn≥3.
Based on (3.23) and (3.24), assume that for somek≥2 we have proved zn= α∑
k−1 i=0ai
βb∑
k−1
j=0 aj∑ni=−0j−2ci wb∑
k−1 i=0aicn−i−1
0 zank−k, (3.25)
forn≥ k+1.
Then by using (3.23) with n→n−kinto (3.25), we get zn=α∑
k−1 i=0ai
βb∑
k−1
j=0 aj∑ni=−0j−2ci wb∑k
−1 i=0aicn−i−1
0 αβb∑
n−k−2
i=0 ciwbc0n−k−1zan−k−1ak
=α∑
ki=0ai
βb∑
k
j=0 aj∑ni=−0j−2ci
w0b∑ki=0aicn−i−1zank−+k1−1, (3.26) forn≥ k+2.
From (3.23), (3.26) and the method of induction we see that (3.25) holds for all natural numberskandnsuch that 1≤k ≤n−1.
By takingk=n−1 into (3.25) we get zn=α∑
n−2 i=0 ai
βb∑
n−2
j=0 aj∑ni=−0j−2ci wb∑
n−2 i=0 aicn−i−1
0 z1an−1, (3.27)
forn≥2.
By using the relationz1 =αz0awb0 into (3.27) it follows that zn =α∑
n−2 i=0 ai
βb∑
n−2
j=0 aj∑ni=−0j−2ci wb∑n
−2 i=0 aicn−i−1
0 (αz0awb0)an−1
=α∑
n−1 i=0 ai
βb∑
n−2
j=0 aj∑ni=−0j−2ci wb∑
n−1 i=0 aicn−i−1
0 za0n, (3.28)
forn≥2.
Now the subcasesa6=canda= cwill be considered separately.
Subcase a6= c.In this case from (3.28), we get zn =α∑
n−1 i=0 ai
βb∑
n−2
j=0 aj∑ni=−0j−2ci za0nwb
an−cn a−c
0 , n∈ N. (3.29)
Ifa6=1 andc6=1, then by formula (2.6) withn→n−2, u= aandv=cand (2.5), (3.29) becomes
zn =α
1−an 1−a βb
a−c+cn−an+can−acn (1−a)(1−c)(a−c) za0nwb
an−cn a−c
0 , n∈N. (3.30)
Ifa6=canda=1, then by using formula (2.9) withn→n−2,u=1 andv=c, we get zn= αnβb
n−1−nc+cn (1−c)2 z0wb
cn−1 c−1
0 , n ∈N, (3.31)
while ifa6=candc=1, then by using formula (2.8) with n→n−2,u= aandv=1, we get zn= α
1−an 1−a βb
n−1−na+an
(1−a)2 za0nwban
−1 a−1
0 , n ∈N. (3.32)
Subcase a=c. In this case from (3.28), we get zn= α∑
n−1 i=0 ai
βb∑
n−2
j=0 aj∑ni=−0j−2ai
za0nw0bnan−1, n≥2. (3.33) Ifa=c6=1, then by using formula (2.7) withn→n−2,u=v= a, (3.33) becomes
zn=α
1−an 1−a βb
1−nan−1+(n−1)an
(1−a)2 z0anwbna0 n−1, n≥2, (3.34) while ifa=c=1, then by using formula (2.10) withn→n−2, u=v=1, we get
zn= αnβb
(n−1)n
2 z0wbn0 , n∈N, (3.35)
completing the proof.
Corollary 3.4. Consider system (1.1) with a,b,c ∈ Z, d = 0 and α,β ∈ C\ {0}. Assume that z−1,z0,w0∈C\ {0}. Then the following statements are true.
(a) If a6=c, a 6=1and c6=1then the general solution to system(1.1)is given by(3.21)and(3.30).
(b) If a6=c and a =1then the general solution to system(1.1)is given by(3.21)and(3.31).
(c) If a6= c and c=1then the general solution to system(1.1)is given by(3.22)and(3.32).
(d) If a=c6=1, then the general solution to system(1.1)is given by(3.21)and(3.34).
(e) If a=c=1, then the general solution to system(1.1)is given by(3.22)and(3.35).
Theorem 3.5. Assume that a,b,c,d ∈ Z, bd 6= 0, α,β ∈ C\ {0}and z−1,z0,w0 ∈ C\ {0}. Then system(1.1)is solvable in closed form.
Proof. A simple inductive argument shows that the conditionsα,β∈C\ {0}andz−1,z0,w0∈ C\ {0}, along with the equations in (1.1) imply zn 6= 0 forn ≥ −1 and wn 6= 0 forn ∈ N0. Further, from the first equation in (1.1), for every well-defined solution, we have that
wbn= zn+1
αzna , n∈N0, (3.36)
while by taking the second equation in (1.1) to theb-th power, is obtained
wbn+1 = βbwbcnzbdn−1, n∈N0. (3.37) Using (3.36) into (3.37) we obtain
zn+2
αzan+1 = βbzcn+1
αczacn zbdn−1, n∈N0, that is,
zn+2 =α1−cβbzan++c1z−naczbdn−1, n∈N0. (3.38) Let
a1 =a+c, b1 =−ac, c1= bd, (3.39)
x1=1−c, y1= b. (3.40)
Then equation (3.38) can be written as
zn+2 =αx1βy1zan1+1zbn1zcn1−1, n∈N0. (3.41) By using recurrent relation (3.41) withn →n−1 into itself, we have
zn+2=αx1βy1(αx1βy1zan1zbn1−1zcn1−2)a1zbn1zcn1−1,
=αx1a1+x1βy1a1+y1zan1a1+b1zbn1−a11+c1zcn1−a12
=αx2βy2zan2zbn2−1zcn2−2, (3.42) forn∈ N, where
a2 := a1a1+b1, b2 :=b1a1+c1, c2:=c1a1, (3.43) x2:=x1a1+x1, y2:= y1a1+y1. (3.44) Assume that for ak ∈Nsuch that 2≤k ≤n+1, we have proved that
zn+2 =αxkβykzank+2−kzbnk+1−kzcnk−k, (3.45)