See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/276386581

Global solutions to fractional programming problem with ratio of nonconvex

functions

Article  in  Applied Mathematics and Computation · March 2015

DOI: 10.1016/j.amc.2014.08.060

CITATIONS READS

3 57

2 authors, including:

David Yang Gao

Federation University Australia
238 PUBLICATIONS   3,513 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

A Breakthrough Theory and Algorithm for Solving Chaotic Dynamics and NP-hard Problems in Mathematical Physics and Complex Systems, Sponsored by AFOSR
Tokyo, 16IOA082, US$120,000 (2016-2017) View project

New Research Grant from US Air Force View project

All content following this page was uploaded by David Yang Gao on 19 October 2017.

The user has requested enhancement of the downloaded file.

 Applied Mathematics and Computation 255 (2015) 66–72

Contents lists available at ScienceDirect

Applied Mathematics and Computation

journal homepage: www.elsevier.com/locate/amc

Global solutions to fractional programming problem with ratio

of nonconvex functions
N. Ruan a, D.Y. Gao a,b,⇑
a
School of Sciences, Information Technology and Engineering, Federation University Australia, Ballarat, VIC 3353, Australia
b
Research School of Engineering, Australian National University, Canberra, ACT 0200, Australia

a r t i c l e i n f o a b s t r a c t

Keywords: This paper presents a canonical dual approach for minimizing a sum of quadratic function
Nonconvex fractional programming and a ratio of nonconvex functions in Rn . By introducing a parameter, the problem is first
Sum-of-ratios equivalently reformed as a nonconvex polynomial minimization with elliptic constraint. It
Global optimization is proved that under certain conditions, the canonical dual is a concave maximization prob-
Canonical duality theory
lem in R2 that exhibits no duality gap. Therefore, the global optimal solution of the primal
problem can be obtained by solving the canonical dual problem.
Ó 2014 Elsevier Inc. All rights reserved.

1. Introduction

We intend to solve the following nonconvex fractional programming problem:

 
gðxÞ
ðPÞ : min P0 ðxÞ ¼ f ðxÞ þ : x2X ; ð1Þ
hðxÞ
where x ¼ ðx1 ; x2 ;    ; xn ÞT 2 Rn and
1 T T
f ðxÞ ¼ x Q x  f x;
2
 2
1 1
gðxÞ ¼ jBxj2  k  cT x;
2 2

1 T T
hðxÞ ¼ x Hx  b x;
2
with B 2 Rmn ; Q 2 Rnn being symmetric, H 2 Rnn negative definite, k 2 Rþ , and f; c; b 2 Rn are given vectors. In this paper,
1
the notation jv j denotes the Euclidean norm of v. Assume that l1
0 ¼ hðH bÞ > 0 and d 2 ð0; l1
0 , the feasible domain X is
defined by
X ¼ fx 2 Rn j hðxÞ P d > 0g;
which is a constraint of elliptic type.

⇑ Corresponding author at: School of Sciences, Information Technology and Engineering, Federation University Australia, Ballarat, VIC 3353, Australia.
E-mail addresses: [email protected] (N. Ruan), [email protected] (D.Y. Gao).

http://dx.doi.org/10.1016/j.amc.2014.08.060
0096-3003/Ó 2014 Elsevier Inc. All rights reserved.
 N. Ruan, D.Y. Gao / Applied Mathematics and Computation 255 (2015) 66–72 67

Problem ðPÞ belongs to a class of ‘‘sum-of-ratios’’ problems that have been actively studied for several decades. The ratios
often stand for efficiency measures representing performance-to-cost, profit-to-revenue or return-to-risk for numerous
applications in economics, transportation science, finance, and engineering (see [1–4,6,8–12,22–27,29,31–33]). The noncon-
vex function gðxÞ is the well-known double-well potential, which appears frequently in chaotic dynamics [15], phase transi-
tions of solids [16], and large deformation mechanics [21]. Depending on the nature of each application, the functions f ; g; h
can be affine, convex, concave, or neither. However, even for the simplest case in which f ; g; h are all affine functions, the
problem ðPÞ is still a global optimization problem that may have multiple local optima [5,28]. In particular, Freund and Jarre
[13] showed that the sum-of-ratios problem ðPÞ is NP-complete when f ; g are convex and h is concave.
Canonical duality theory is a powerful methodological theory which can be used for solving a large class of nonconvex/
nonsmooth/discrete problems in nonlinear analysis and global optimization [14,18,19]. The main goal of this paper is to
solve the problem ðPÞ by the canonical duality theory. In Section 2, the problem ðPÞ is first parameterized equivalently as
a nonconvex polynomial minimization ðP l Þ with an elliptic constraint. For each given parameter, the canonical dual problem
is derived by the standard canonical dual transformation. The global optimality condition is proposed in Section 3. An
example is illustrated in Section 4. Conclusion is provided in the last section.

2. Canonical dual problem

In order to solve the problem ðPÞ, we consider the following parameterized subproblem:
 
1 T
ðP l Þ : min Pl ðxÞ ¼ xT Qx  f x þ lgðxÞ : x 2 Xl ; ð2Þ
2
 
where l 2 l0 ; d1 and
X l ¼ fx 2 Rn j hðxÞ P l1 P d > 0g
is a convex set. We immediately have the following result:

Lemma 1. Problem ðPÞ is equivalent to ðP l Þ in the sense that

inf P0 ðxÞ ¼ inf inf Pl ðxÞ: ð3Þ
x2X l2½l0 ;d1  x2X l

Proof. It is easy to see that

   
1 T T gðxÞ 1 T T gðxÞ
inf P0 ðxÞ ¼ inf x Qx  f x þ ¼ inf inf x Qx  f x þ
x2X x2X 2 hðxÞ l2½l0 ;d1  hðxÞ¼l1 2 hðxÞ
   
1 T T 1 T
¼ inf inf x Q x  f x þ lgðxÞ P inf inf xT Q x  f x þ lgðxÞ ¼ inf inf Pl ðxÞ:
l2½l0 ;d1  hðxÞ¼l1 2 l2½l ;d1 
0
x2X l 2 l2½l0 ;d1  x2X l

Conversely,
   
1 T T 1 T T
inf inf x Q x  f x þ lgðxÞ ¼ inf inf x Q x  f x þ lgðxÞ
l2½l0 ;d1  x2X l 2 l2½l0 ;d1  hðxÞPl1 2
 
1 T T gðxÞ
P inf inf x Qx  f x þ ðsince gðxÞ > 0Þ ¼ inf P0 ðxÞ:
l2½l0 ;d1  hðxÞPl1 2 hðxÞ x2X

This completes the proof of the lemma. h

Now, for any l 2 ½l0 ; d1 , we define

Gl ð1; rÞ ¼ Q þ l1BT B  rH; ð4Þ

S þl ¼ fð1; rÞ 2 R2 j 1 P k; r P 0; Gl ð1; rÞ  0g; ð5Þ

where ‘’ means positive definiteness of a matrix. Let @S þ

l denote a singular hyper-surface defined by

@S þl ¼ f1 P k; r P 0j Gl ð1; rÞ  0; det Gl ð1; rÞ ¼ 0g: ð6Þ

Then, the parametrical canonical dual problem can be proposed as the following:
1 T l 2 r
Pdl ð1; rÞ ¼  ðf þ lc  rbÞ G1
l ð1; rÞðf þ lc  rbÞ  lk1  1 þ : ð7Þ
2 2 l
Therefore, the canonical dual problem ðP dl Þ can be proposed as the following:
68 N. Ruan, D.Y. Gao / Applied Mathematics and Computation 255 (2015) 66–72

n o
ðP dl Þ : sup Pdl ð1; rÞj ð1; rÞ 2 S þl : ð8Þ

Theorem 1 (Weak Duality). If there exists a global maximizer ð1l ; rl Þ of Pdl ð1; rÞ over S þ
l , then the vector

xl ¼ G1
l ð1l; ; rl Þðf þ lc  rl bÞ ð9Þ

is a global minimizer of ðP l Þ over X l and

Pl ðxÞ P Pdl ð1; rÞ; 8x 2 X l ; ð1; rÞ 2 S þl : ð10Þ

Proof. Let KðÞ : Rn ! R be the geometrical transformation [14,19,20] defined by

1
n ¼ KðxÞ ¼ jBxj2  k ð11Þ
2
and let
1 2
UðnÞ ¼ n ð12Þ
2
Then, Problem ðP l Þ in (2) can be written in the following form
 
1 T
min PðxÞ ¼ xT Q x  f x þ lðUðKðxÞÞ  rðhðxÞ  l1 Þjx 2 Rn : : ð13Þ
2

Let 1 be the dual variable of n, i.e., 1 ¼ rUðnÞ ¼ n, the Legendre conjugate U ð1Þ can be uniquely defined by

1 2
U  ð1Þ ¼ stafn1  UðnÞ j n P kg ¼ 1 ð14Þ
2
where 1 2 V a ¼ f1 2 Rj1 P kg.
By replacing UðKðxÞÞ with KðxÞT 1  U  ð1Þ in (13), the Gao–Strang total complementary function for this fractional problem
can be obtained as the following
 
1 1 T
Nðx; 1; rÞ ¼ xT Q x  f T x þ lðUðKðxÞÞ  r T
x Hx  b x  l1
2 2
 
1 T T  
1 T
¼ x Q x  f x þ l KðxÞ1  U ð1Þ  cT x  r xT Hx  b x  l1
2 2
1 T T l 2 r
¼ x Gl ð1; rÞx  ðf þ lc  rbÞ x  lk1  1 þ ;
2 2 l
where Gl ð1; rÞ is defined in (4). Note that Nðx; 1; rÞ is convex in x 2 Rn for any given ð1; rÞ 2 S þ
l and concave in ð1; rÞ for any
given x 2 Rn . By the criticality condition
@N
¼ Gl ð1; rÞx  ðf þ lc  rbÞ ¼ 0: ð15Þ
@x
We have xð1; rÞ ¼ G1
l ð1; rÞðf þ lc  rbÞ, which is the global minimizer of Nðx; 1; rÞ. Moreover,
1 T l r
minn Nðx; 1; rÞ ¼ Nðxð1; rÞ; 1; rÞ ¼ xð1; rÞT ðGl ð1; rÞÞxð1; rÞ  ðf þ lc  rbÞ xð1; rÞ  lk1  12 þ
x2R 2 2 l
1 T l r
¼ xð1; rÞT ðf þ lc  rbÞ  ðf  rbÞ xð1; rÞ  lk1  12 þ
2 2 l
1 T l 2 r
¼  ðf þ lc  rbÞ xð1; rÞ  lk1  1 þ
2 2 l
1 T 1 l r
¼  ðf þ lc  rbÞ Gl ð1; rÞðf þ lc  rbÞ  lk1  12 þ ¼ Pdl ð1; rÞ:
2 2 l
By the assumption, ð1l ; rl Þ is a global maximizer of P dl ð1; rÞ. If ð1l ; rl Þ is an interior of S þ
l , then
@ d
1l ; rl Þ ¼ 0, and
P ð
@1 l
@
Pd ð
@r l
1l ; rl Þ ¼ 0. Otherwise, we have @ d
1l ; rl Þ ¼ 0, rl ¼
P ð
@1 l
0; @@r P dl ð
1l ; rl Þ 6 0. In either case, if we denote
1
xl ¼ xð1l ; rl Þ ¼ Gl ð1l ; rl Þðf þ lc  rl bÞ, we have
@ d 1
P ð1 ; rl Þ ¼ lð xTl BT Bxl  k  1Þ ¼ 0;
@1 l l 2
@ d 1 1
P ð1 ; rl Þ ¼  xTl ð Hxl  bÞ 6 0:
@r l l l 2
 N. Ruan, D.Y. Gao / Applied Mathematics and Computation 255 (2015) 66–72 69

That is,
1
1 ¼ jBxl j2  k;
2
1 T T
x Hxl  b xl  l1 P 0:
2 l
Therefore, xl 2 X l , and for any ð1; rÞ 2 S þ
l , we have

Pdl ð1; rÞ 6 P dl ð1l ; rl Þ ¼ minn Nðx; 1l ; rl Þ ¼ Nðxl ; 1l ; rl Þ ¼ min Nðx; 1l ; rl Þ

x2R x2X l

1 T T 1 T
¼ x Q x  f x þ lðKðxÞT 1  U  ð1Þ  cT xÞ  rð xT Hx  b x  l1 Þ
2 2
2
1 T 1 T 1 1
6 xT Q x  f x þ lðKðxÞT 1  U  ð1Þ  cT xÞ ¼ xT Qx  f x þ lð ð jBxj2  kÞ  cT xÞ ¼ Pl ðxÞ:
2 2 2 2
This completes the proof. h

3. Sufficient condition for global optimality

In this section we prove that canonical dual problem is a concave maximization problem. And we derive global optimality
condition based on canonical duality theory.

Lemma 2. For any l 2 ½l0 ; d1 , the canonical dual function Pdl ðr; 1Þ is a two-dimensional concave function over S þl .

Proof. Notice that the Hessian Matrix of the dual objective function is
!
2 d H r2 H r1
r P l ðr; 1Þ ¼ S ¼ ;
H1r H12
where
T
Hr2 ¼ ðHxl  bÞ G1
l ð1; rÞðHxl  bÞ;
H12 ¼ l2 xTl ðBT BÞG1 T
l ð1; rÞðB BÞxl  l;
T T
Hr1 ¼ lðHxl  bÞ G1
l ð1; rÞðB BÞxl ;
H1r ¼ lxTl ðBT BÞG1
l ð1; rÞðHxl  bÞ:

In order to show the dual function is a concave function, it is equivalent to show that
!
Hr2 H r1
S0 ¼ ;
H1r H12 þ l

is semi-negative definite. By Sylvester’s Criterion, it suffices to show that all the leading principal minors have a non-positive
determinant. Obviously, the first 1  1 leading principal minors have non-positive determinants, since

ðHxl  bÞG1 ð1; rÞðHxl  bÞ 6 0: ð16Þ

Now we need to show detðS0 Þ 6 0. Note that
T ! !
ðHxl  bÞ G1
l ð1; rÞ 0 Hxl  b lðBT BÞxl
S0 ¼ CD ¼
0 lxTl BT BG1
l ð1; rÞ Hxl  b lðBT BÞxl

Apparently, RankðCDÞ 6 RankðDÞ 6 n. We can make a conclusion that det S0 ¼ 0. Thus, S is semi-negative definite, which
implies that dual function is concave function. h

Theorem 2 (Strong duality). If ð1l ; rl Þ is a critical point of P dl ð1; rÞ over S þ d

l , then ðP l Þ is perfectly dual to ðP l Þ in the sense that
the vector

xl ¼ G1
l ð1l ; rl Þðf þ lc  rl bÞ ð17Þ

is a global minimizer of ðP l Þ and ð1l ; rl Þ is a global maximizer of ðP dl Þ, and

minPl ðxÞ ¼ Pl ðxl Þ ¼ Pdl ð1l ; rl Þ ¼ maxþ Pdl ð1; rÞ: ð18Þ
x2X l ð1;rÞ2S l
70 N. Ruan, D.Y. Gao / Applied Mathematics and Computation 255 (2015) 66–72

Proof. The proof basically follows that of the former weak duality Theorem, The only difference lies in the assumption that
ð1l ; rl Þ is a critical point of Pdl ð1; rÞ over Sþ
l. In this case, @ d
1l ; rl Þ ¼ 0, and @@r Pdl ð1l ; rl Þ ¼ 0. So
P ð
@1 l
1 T
xl ¼ xð1l ; rl Þ ¼ Gl ð1l ; rl Þðf þ lc  rl bÞ is on the boundary of X l . That is, 12 xT Hx  b x  l1 ¼ 0. This further implies that

Pdl ð1l ; rl Þ ¼ Nðxl ; 1l ; rl Þ ¼ Pl ðxl Þ ð19Þ

and the Eq. (18) follows naturally. h

The above results immediately lead to the following sufficient condition for finding the global optimizer of problem ðPÞ:

Corollary 1. If there exists a critical point ð1l ; rl Þ 2 S þ

l for every l 2 ½l0 ; d1 , then

minP0 ðxÞ ¼ min Pdl ð1l ; rl Þ: ð20Þ

x2X l2½l0 ;d1 

4. Numerical example

In order to demonstrate the application of the theoretical results presented in this paper, let us consider the following
one-dimensional problem
2
0:5ð0:5x2  2Þ  0:5x
min P0 ðxÞ ¼ 5x2  x þ
0:5x2  x
over the feasible domain

X ¼ fx 2 R j 0:5x2  x 0:01g ¼ f1:9899 6 x 6 0:01005g:

The target function P 0 ðxÞ has singularities at x ¼ 2 and x ¼ 0.
By Lemma 1, we have minx2X P0 ðxÞ ¼ minl2½2;100 minx2X l Pl ðxÞ with
2
Pl ðxÞ ¼ 5x2  x þ lð0:5ð0:5x2  2Þ  0:5xÞ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
and X l ¼ f1  1  2l1 6 x 6 1 þ 1  2l1 g. For each l 2 ½2; 100, the canonical dual function can be obtained as
1 T l 2 r ð1 þ 0:5l  rÞ2 l r
Pdl ð1; rÞ ¼  ðf þ lc  rbÞ G1
l ð1; rÞðf þ lc  rbÞ  lk1  1 þ ¼   2l1  12 þ
2 2 l 2ð10 þ l1 þ rÞ 2 l
d
over S þ
l ¼ f1 2 R; r 2 Rj 1 P 2; r P 0; l1 þ r > 10g. The partial derivatives of P l ð1; rÞ become

@ d lð1 þ 0:5l  rÞ2

Pl ð1; rÞ ¼  2l  l1;
@1 2ð10 þ l1 þ rÞ2
and
 2
@ d 1 1 þ 0:5l  r 1 þ 0:5l  r 1
Pl ð1; rÞ ¼ þ þ :
@r 2 10 þ l1 þ r 10 þ l1 þ r l
It is obvious that
@ d
lim P ð1; rÞ < 0 for 8l 2 ½2; 100;
1!1 @ 1 l

@ d 1 1 1 1
lim P ð1; rÞ ¼  1 þ ¼  þ < 0 for l > 2:
r!1 @ r l 2 l 2 l
Therefore, the maximizer ð1l ; rl Þ of Pdl ð1; rÞ does not exist on l ¼ 2.
Suppose the maximizer of P dl ð1; rÞ is ð1l ; rl Þ for given l, we define Pd ðlÞ ¼ P dl ð1; rÞ. To minimize P d ðlÞ over l 2 ½l0 ; d1 ,
we use the line search with the Armijo’s rule. Suppose the current iterate is at lk 2 ½l0 ; d1 , we may approximate the deriv-
ative of P d ðlÞ at l ¼ lk by
d d
d d : P ðlk þ Þ  P ðlk Þ
dk ¼ P ðlÞjl¼lk ¼
dl 
where  is a selected parameter and the two terms in the numerator can be evaluated by the interior point method in the
optimization Toolbox within the Matlab environment. If dk > 0, the full step size s can be taken as the distance fro the left
boundary to lk , i.e., s ¼ l0  lk . Otherwise, we take s ¼ d1  lk from the other end. Then we select two parameters such
 N. Ruan, D.Y. Gao / Applied Mathematics and Computation 255 (2015) 66–72 71

Fig. 1. Graph of P d4:8863 ð1; rÞ for Example 1.

that parameter a is close to 0 for scaling the slope dk , and parameter b 2 ð0; 1Þ for scaling the full step size s. Choose m to be
the first non-negative integer such that

m ¼ minfn P 0jPd ðlk þ ðbn sÞðadk ÞÞ < P d ðlk Þ þ ðbn sÞðadk Þg:
Then, we can update

lkþ1 ¼ lk þ ðbm sÞðadk Þ

and repeat until dk is nearly 0.
In our example, we use b ¼ 2=3; a ¼ 0:001 and  ¼ 0:01. It reaches the global minimum of P dl ð1; rÞ at l ¼ 4:8863 with a
value of 9:0851. The corresponding primal solution is x ¼ 1:7686, and Pl ðxÞ ¼ 9:0851. See Fig. 1 for the graph of
P d4:8863 ð1; rÞ.

5. Conclusions

The paper presents a new method for solving a special fractional programming problem with sum of a quadratic function
and the ratio of nonconvex function and quadratic function as its target function. By using the canonical duality theory, this
challenging problem is reformulated as a concave maximization dual problem over a convex domain, which can be solved
easily if the canonical dual function P dl ðr; 1Þ has a critical point in S þ
l . Otherwise, the primal problem could be really NP-hard,
which is a conjuncture proposed in [17]. Recently, we noted a new methodology has been proposed by Sergeryev [7,30] for
performing calculations with infinite and infinitesimal quantities, by introducing an infinite unit of measure expressed by
the numeral grossone. We believe this new concept could be important for the future study of the fractional optimization.

Acknowledgement

This paper was partially supported by a grant (AFOSR FA9550–10-1–0487) from the US Air Force Office of Scientific
Research. Dr. Ning Ruan was supported by a funding from the Australian Government under the Collaborative Research Net-
works (CRN) program.

References

[1] Y. Almogy, O. Levin, A class of fractional programming problems, Oper. Res. 19 (1971) 57–67.
[2] H.P. Benson, Global optimization algorithm for the nonlinear sum of ratios problem, J. Optim. Theory Appl. 112 (2002) 1–29.
[3] H.P. Benson, Using concave envelopes to globally solve the nonlinear sum of ratios problems, J. Global Optim. 22 (2002) 343–364.
[4] H.P. Benson, On the global optimization of sum of linear fractional functions over a convex set, J. Optim. Theory Appl. 121 (2004) 19–39.
[5] A. Cambini, J.P. Crouzeix, L. Martein, On the pseudoconvexity of a quadratic fractional function, Optimization 51 (2002) 677–687.
[6] C.S. Colantoni, R.P. Manes, A. Whinston, Programming, profit rates, and pricing decisions, Acc. Rev. 44 (1969) 467–481.
[7] S.D. Cosmis, R.D. Leone, The use of grossone in mathematical programming and operations research, Appl. Math. Comput. 218 (2012) 8029–8038.
[8] B.D. Craven, Fractional programming, Sigma Series in Applied Mathematics, 4, Heldermann Verlag, Berlin, Germany, 1988.
[9] G. Dahl, A note on diagonally dominant matrices, Linear Algebr. Appl. 317 (2000) 217–224.
72 N. Ruan, D.Y. Gao / Applied Mathematics and Computation 255 (2015) 66–72

[10] R.J. Duffin, E.L. Peterson, Geometric programming with signomials, J. Optim. Theory Appl. 11 (1973) 3–35.
[11] J.E. Falk, S.W. Palocsay, Optimizing the sum of linear fractional functions, in: C.A. Floudas, P.M. Pardalos (Eds.), Recent Advances in Global
Optimization, Princeton University Press, Princeton, NJ, 1992, pp. 221–258.
[12] S.C. Fang, D.Y. Gao, R.L. Sheu, S.Y. Wu, Canonical dual approach to solving 0–1 quadratic programming problems, J. Ind. Manag. Optim. 4 (2008) 125–
142.
[13] R.W. Freund, F. Jarre, Solving the sum-of-ratios problem by an interior-point method, J. Global Optim. 19 (2001) 83–102.
[14] D.Y. Gao, Duality Principles in Nonconvex Systems: Theory, Methods and Applications, Kluwer Academic Publishers, Dordrecht/Boston/London, 2000.
454 pp.
[15] D.Y. Gao, Analytic solution and triality theory for nonconvex and nonsmooth variational problems with applications, Nonlinear Anal. 42 (2000) 1161–
1193.
[16] D.Y. Gao, Complementary variational principle, algorithm, and complete solutions to phase transitions in solids governed by Landau–Ginzburg
equation, Math. Mech. Solid 9 (2004) 285–305.
[17] D.Y. Gao, Solutions and optimality to box constrained nonconvex minimization problems, J. Ind. Manage. Optim. 3 (2007) 293–304.
[18] D.Y. Gao, Canonical duality theory: unified understanding and generalized solutions for global optimization, Comput. Chem. Eng. 33 (2009) 1964–
1972.
[19] D.Y. Gao, H.D. Sherali, Canonical Duality Theory: Connections between nonconvex mechanics and global optimization, Adv. Mech. Math. Springer 17
(2009) 257–326.
[20] D.Y. Gao, G. Strang, Geometric nonlinearity: Potential energy, complementary energy, and the gap function, Q. Appl. Math. 47 (1989) 487–504.
[21] H.A.F.A. Santos, D.Y. Gao, Canonical dual finite element method for solving post-buckling problems of a large deformation elastic beam, Int. J. Nonlinear
Mech. 47 (2012) 240247.
[22] P.K. Kanchan, A.S.B. Holland, B.N. Sahney, Transportation techniques in linear-plus-fractional programming, Cah. du CERO 23 (1981) 153–157.
[23] H. Konno, K. Fukaishi, A branch and bound algorithm for solving low rank linear multiplicative and fractional programming problems, J. Global Optim.
18 (2000) 283–299.
[24] H. Konno, M. Inori, Bond portfolio optimization by bilinear fractional programming, J. Oper. Res. Soc. Jpn. 32 (1989) 143–158.
[25] H. Konno, H. Watanabe, Bond portfolio optimization problems and their application to index tracking: a partial optimization approach, J. Oper. Res.
Soc. Jpn. 39 (1996) 295–306.
[26] T. Kuno, A branch-and-bound algorithm for maximizing the sum of several linear ratios, J. Global Optim. 22 (2002) 155–174.
[27] N.T.H. Phuong, H. Tuy, A unified monotonic approach to generalized linear fractional programming, J. Global Optim. 26 (2003) 229–259.
[28] S. Schaible, A note on the sum of a linear and linear-fractional function, Nav. Res. Logist. Q. 24 (1977) 691–693.
[29] S. Schaible, Fractional programming, in: R. Horst, P.M. Pardalos (Eds.), Handbook of Global Optimization, Kluwer Academic Publishers, 1995, pp. 495–
608.
[30] Y.D. Sergeyev, Arithmetic of Infinity, Edizioni Orizzonti Meridionali, CS, 2003.
[31] C.H. Scott, T.R. Jefferson, Duality of a nonconvex sum of ratios, J. Optim. Theory Appl. 98 (1998) 151–159.
[32] I.M. Stancu-Minasian, Applications of the fractional programming, Econ. Comput. Econ. Cybern. Stud. Res. 1 (1980) 69–86.
[33] W.Y. Wu, R.L. Sheu, S.I. Birbil, Solving the sum-of-ratios problem by a stochastic search algorithm, J. Global Optim. 42 (2008) 91–109.

View publication stats