Hindawi Publishing Corporation

Mathematical Problems in Engineering

Volume 2013, Article ID 568098, 8 pages
http://dx.doi.org/10.1155/2013/568098

Research Article
Engineering Design by Geometric Programming

Chia-Hui Huang
Department of Business Administration, National Taipei College of Business, No. 321, Sec. 1, Jinan Road,
Zhongzheng District, Taipei 100, Taiwan

Correspondence should be addressed to Chia-Hui Huang; leohuang@webmail.ntcb.edu.tw

Received 7 June 2013; Accepted 19 July 2013

Academic Editor: Jung-Fa Tsai

Copyright © 2013 Chia-Hui Huang. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A geometric program (GP) is a type of mathematical optimization problem characterized by objective and constraint functions,
where all functions are of signomial form. The importance of GP comes from two relatively recent developments: (i) new methods
can solve even large-scale GP extremely efficiently and reliably; (ii) a number of practical problems have recently been found to be
equivalent to or approximated by GP. This study proposes an optimization approach for solving GP. Our approach is first to convert
all signomial terms in GP into convex and concave terms. Then the concave terms are further treated with the proposed piecewise
linearization method where only ⌈log2 (𝑚 − 1)⌉ binary variables are used. It has the following features: (i) it offers more convenient
and efficient means of expressing a piecewise linear function; (ii) fewer 0-1 variables are used; (iii) the computational results show
that the proposed method is much more efficient and faster than the conventional one, especially when the number of break points
becomes large. In addition, the engineering design problems are illustrated to evaluate the usefulness of the proposed methods.

1. Introduction GP has been applied in many fields of applications includ-

ing analog/digital circuit design [2–8], chemical engineering
Consider the following geometric program (GP) [1]: [9–11], mechanical engineering [12–19], power control [20],
min 𝑓0 (𝑥) (1) and communication network systems [21–23].
Obtaining the optimal solutions for GP is not straightfor-
s.t. 𝑓𝑘 (𝑥) ≤ 1, 𝑘 = 1, 2, . . . , 𝑝, (2) ward because the signomial terms in the objective function
and constraints cannot be solved directly. As a result, many
𝑔𝑙 (𝑥) = 1, 𝑙 = 1, 2, . . . , 𝑞, (3) approaches have been developed. Coello and Cortés [24] pro-
posed a genetic algorithm with an artificial immune system
𝑥 = (𝑥1 , 𝑥2 , . . . , 𝑥𝑛 ) , (4)
to solve a GP in engineering optimization. Nevertheless, this
where 𝑔𝑙 (𝑥), 𝑙 = 1, 2, . . . , 𝑞, are monomials as in (5), 𝑐𝑙 ≥ 0, method can only obtain the local optima. Horst and Tuy [25]
and the exponential constants 𝛼𝑖𝑙 ∈ R, 𝑖 = 1, 2, . . . , 𝑛, 𝑙 = introduced an analytical approach for solving a problem with
1, 2, . . . , 𝑞: Lipschitzian objective and constraints. The restriction of this
approach is to find the global optimum only if the range of
𝑛
𝛼 variables can be reduced by analytical techniques. Sherali and
𝑔𝑙 (𝑥) = 𝑐𝑙 ∏𝑥𝑖 𝑖𝑙 (5) Tuncbilek [26] developed a reformulation-linearization tech-
𝑖=1
nique (RLT) which generates polynomial implied constraints
and 𝑓𝑘 (𝑥), 𝑘 = 0, 1, . . . , 𝑝, are posynomials as in (6) which are and then linearizes the resulting problem by introducing new
sum of one or more monomials, 𝑐ℎ𝑘 ≥ 0, and 𝛼𝑖ℎ𝑘 ∈ R, 𝑖 = variables. Lin and Tsai [14] introduced a generalized method
1, 2, . . . , 𝑛, ℎ = 1, 2, . . . , 𝐻𝑘 , 𝑘 = 0, 1, . . . , 𝑝: to find multiple optimal solutions of signomial discrete
𝐻𝑘
programming problems with free variables. By means of
𝑛
𝑓𝑘 (𝑥) = ∑ 𝑐ℎ𝑘 ∏𝑥𝑖 𝑖ℎ𝑘 .
𝛼
(6) variable substitution and convexification strategies, a signo-
ℎ=1 𝑖=1
mial discrete programming problem with free variables is
2 Mathematical Problems in Engineering

first converted into another convex mixed-integer nonlinear
programming problem solvable to obtain an exactly global
optimum. Tsai [17] proposed a novel method to solve signo-
mial discrete programming problems. The signomial terms
are first convexified following the convexification strategies.
The original program is then converted into a convex integer
program solvable by commercialized packages to obtain
globally optimal solutions. Tsai and Lin [19] developed an effi-
cient method to solve a posynomial geometric program with
separable functions. Power transformations and exponential
transformations are utilized to convexify and underestimate
posynomial terms. The original program therefore can be
converted into a convex mixed-integer nonlinear program
solvable to obtain a global optimum.
This paper develops an optimization approach for solving
GP. Our approach is listed as follows.
Lemma 1. Let 𝐴 be a nonzero 𝑛 × 𝑛 symmetric matrix.
(a) If 𝐴 is positive semidefinite, then Δ 𝑘 ≥ 0, 𝑘 =
1, 2, . . . , 𝑛, and not all Δ 𝑘 = 0.
(b) 𝐴 is positive definite if and only if Δ 𝑘 > 0, for all 𝑘 =
1, 2, . . . , 𝑛.
The converse of Lemma 1(a) is false. For example, let 𝐴 =
10 0
we have ⟨𝑥, 𝐴𝑥⟩ = 𝑥12 − 𝑥32 which is not always
[ 0 0 0 ];
0 0 −1
nonnegative for all 𝑥 ∈ R3 . But Δ 1 = 1 ≥ 0, Δ 2 = 0 ≥ 0,
and Δ 3 = 0 ≥ 0. In fact, the converse of Lemma 1(a) is true
only for 𝑛 = 2; see [27, page 112].
Lemma 2. Let 𝐴 be a nonzero symmetric matrix. Then, the
following will hold.

(i) Convert all signomial terms into convex and concave (a) 𝐴 is 𝑝.𝑠.𝑑. if and only if all of its eigenvalues are
ones. nonnegative.

(ii) The concave terms are further treated with the pro-
posed piecewise linearization method where only
⌈log2 (𝑚 − 1)⌉ binary variables are used.
The rest of this paper is organized as follows. Section 2
introduces the proposed methods. Section 3 provides some
numerical examples to illustrate the modeling idea and the
usefulness of the proposed method. Section 4 gives our
conclusions.
(b) 𝐴 is 𝑝.𝑑. if and only if all of its eigenvalues are positive.
Proposition 3. Let 𝑓1 : R𝑛++ → R be defined as 𝑓1 (𝑥) =
𝛼
𝑐1 ∏𝑛𝑖=1 𝑥𝑖 𝑖 , where 𝑐1 > 0 and 𝛼𝑖 ≤ 0 for all 𝑖 = 1, 2, . . . , 𝑛. Then
𝑓1 is a convex function.
Proof. Since 𝑐1 > 0, it is enough to show that 𝑓̃1 (𝑥) = ∏𝑛𝑖=1 𝑥𝑖 𝑖
𝛼
is convex. Let 𝑔(𝑥) = ln 𝑓̃1 (𝑥) = ∑𝑛𝑖=1 ln 𝑥𝑖 𝑖 = ∑𝑛𝑖=1 𝛼𝑖 ln 𝑥𝑖 .
𝛼
Then, we have

𝛼1 𝛼2 𝛼 𝑇
2. Proposed Methods ∇𝑔 (𝑥) = [ ⋅⋅⋅ 𝑛] ,
𝑥1 𝑥2 𝑥𝑛
In this section, the proposed methods will be presented. First, −𝛼1
we recall what a signomial function is. A function 𝑓 : R𝑛++ → 0 ⋅⋅⋅ 0
R defined as [ 𝑥12 ]
[ −𝛼2 ] (9)
[ 0 ⋅⋅⋅ 0 ]
[ 𝑥22 ]
𝑛
∇ 𝑔 (𝑥) = [
2
[
].
𝑓 (𝑥) = 𝑐∏𝑥𝑖 𝑖 ,
𝛼
(7) [ .. .. .. ]
]
𝑖=1 [ . . d . ]
[ −𝛼𝑛 ]
0 0 ⋅⋅⋅
where 𝑐 > 0 and 𝛼𝑖 ∈ R for all 𝑖, is called a monomial function [ 𝑥𝑛2 ]
or simply a monomial [1]. Note that the exponents 𝛼𝑖 of a
monomial can be any real numbers, but the coefficient 𝑐 must Because 𝛼𝑖 ≤ 0 for all 𝑖 = 1, 2, . . . , 𝑛, we know that
be nonnegative. A sum of monomials, namely, a function of all eigenvalues of ∇2 𝑔(𝑥) are nonnegative, which implies
the form (by Lemma 1(a)) that ∇2 𝑔(𝑥) is positive semidefinite. Thus,
𝑔(𝑥) = ln 𝑓(𝑥) is a convex function which yields that 𝑓̃1 (𝑥)
𝑁 𝑛
𝛼 is a convex function.
𝑓 (𝑥) = ∑ 𝑐𝑘 ∏𝑥𝑖 𝑖𝑘 , (8)
𝑘=1 𝑖=1
Proposition 4. Let 𝑓2 : R𝑛++ → R be defined as 𝑓2 (𝑥) =
𝛼
where 𝑐𝑘 > 0 and 𝛼𝑖𝑘 ∈ R, is called a posynomial function 𝑐2 ∏𝑛𝑖=1 𝑥𝑖 𝑖 , where 𝑐2 < 0 and 𝛼𝑖 > 0 for all 𝑖 = 1, 2, . . . , 𝑛 with
𝑛
with 𝑁 terms or simply a posynomial. A signomial is a 1 − ∑𝑖=1 𝛼𝑖 ≥ 0. Then 𝑓2 is a convex function.
linear combination of monomials of some positive variables
Proof. It is not hard to compute that [∇𝑓2 (𝑥)]𝑖 =
𝑥1 , . . . , 𝑥𝑛 . 𝛼 −1 𝛼
Let 𝑓 be defined on an open convex set 𝐷 ⊆ R𝑛 𝑐2 𝛼𝑖 𝑥𝑖 𝑖 ∏𝑛𝑗=1,𝑗 ≠ 𝑖 𝑥𝑗 𝑗 . In other words,
and be twice differentiable; it is known that (i) 𝑓 is convex
𝛼 −1 𝛼
on 𝐷 if and only if the Hessian matrix ∇2 𝑓(𝑥) is positive 𝑐2 𝛼1 𝑥1 1 𝑥2 2 ⋅ ⋅ ⋅ 𝑥𝑛𝛼𝑛
semidefinite (p.s.d. for short) at each 𝑥 ∈ 𝐷; (ii) if ∇2 𝑓(𝑥) [𝑐 𝛼 𝑥𝛼1 𝑥𝛼2 −1 ⋅ ⋅ ⋅ 𝑥𝛼𝑛 ]
[2 2 1 2 𝑛 ]
is positive definite (p.d. for short) at each 𝑥 ∈ 𝐷, then 𝑓 is ∇𝑓2 (𝑥) = [ .. ]. (10)
[ . ]
strictly convex. For convenience, we denote by Δ 𝑘 the leading 𝛼 𝛼 𝛼 −1
principal minors of 𝐴. [𝑐2 𝛼𝑛 𝑥1 1 𝑥2 2 ⋅ ⋅ ⋅ 𝑥𝑛 𝑛 ]
Mathematical Problems in Engineering 3

In addition, it can be verified that fact. To see this, we arrange all powers 𝛼𝑖 in a decreasing order.
𝛼𝑖 𝛼𝑗 In other words, without loss of generality, we assume that
{
{ 𝑓 (𝑥) , 𝑖 ≠ 𝑗,
𝜕 𝑓2 (𝑥) 2
{ 𝜕𝑥𝑖 𝜕𝑥𝑗 2 𝛼1 > 𝛼2 ≥ ⋅ ⋅ ⋅ ≥ 𝛼𝑛 . (16)
[∇2 𝑓2 (𝑥)]𝑖𝑗 = = { 𝛼 (𝛼 − 1) (11)
𝜕𝑥𝑖 𝜕𝑥𝑗 {
{ 𝑖 𝑖 𝑓2 (𝑥) , 𝑖 = 𝑗, Notice that condition (15) implies that 𝛼1 is positive and all
{ 𝜕𝑥𝑖2 the other 𝛼2 , . . . , 𝛼𝑛 are nonpositive with 𝛼1 ≥ 1 − ∑𝑛𝑖=2 𝛼𝑖 . As
Namely, mentioned in Proposition 3, we only need to show that the
function 𝑓̃1 (𝑥) = ∏𝑛𝑖=1 𝑥𝑖 𝑖 is convex. By similar arguments as
𝛼
2
∇ 𝑓2 (𝑥) in the proof of Proposition 4, we know that
𝑛 𝑛 𝑛
−2 𝛼𝑖 𝛼 𝛼 𝑖 𝑖
[𝑐2 𝛼1 (𝛼1 − 1) 𝑥1 ∏𝑥𝑖 𝑐2 𝛼1 𝛼2 𝑥1−1 𝑥2−1 ∏𝑥𝑖 𝑖 ⋅⋅⋅ 𝑐2 𝛼1 𝛼𝑛 𝑥1−1 𝑥𝑛−1 ∏𝑥𝑖 𝑖 ]
[
[ 𝑛
𝑖=1 𝑖=1
𝑛
𝑖=1
𝑛
]
]
̂ 𝑖 = (−1)𝑖 (∏𝛼𝑗 𝑥𝑖𝛼𝑗 −2 ) (1 − ∑𝛼𝑗 ) ,
Δ (17)
[ 𝑐 𝛼 𝛼 𝑥−1 𝑥−1 ∏𝑥𝛼𝑖 𝑗
𝑐2 𝛼2 𝛼𝑛 𝑥2−1 𝑥𝑛−1 ∏𝑥𝑖 𝑖 ]
𝛼 𝛼
[ 𝑐2 𝛼2 (𝛼2 − 1) 𝑥2−2 ∏𝑥𝑖 𝑖 ⋅⋅⋅ ] 𝑗=1 𝑗=1
[ 2 2 1 2 1 𝑖=1 𝑖 ]
[ 𝑖=1 𝑖=1 ]
=[ . . . ].
[ ]
[
[
.
.
.
. d
.
. ]
] where Δ ̂ 𝑖 denotes the 𝑖th principal minor of the Hessian
[ 𝑛 𝑛 𝑛 ]
[ 𝑐 𝛼 𝛼 𝑥−1 𝑥−1 ∏𝑥𝛼𝑖
[ 2 𝑛 1 𝑛 1 𝑖 𝑐2 𝛼𝑛 𝛼2 𝑥𝑛−1 𝑥2−1 ∏𝑥𝑖 𝑖
𝛼 −2
⋅ ⋅ ⋅ 𝑐2 𝛼𝑛 (𝛼𝑛 − 1) 𝑥𝑛 ∏𝑥𝑖 ]
𝛼𝑖 ]
matrix of 𝑓̃1 (𝑥). From conditions (15) and (16), it is easily
𝑖=1 𝑖=1 𝑖=1
[ ] verified that (1 − ∑𝑖𝑗=1 𝛼𝑗 ) < 0 for each 𝑖. It is also not hard
(12)
to observe that ∏𝑖𝑗=1 𝛼𝑗 is positive if 𝑖 is odd and is negative if
Moreover, the determinant of ∇2 𝑓2 (𝑥) can be computed and 𝑖 is even. In summary, there holds
be shown by induction as
𝑖
𝑖𝛼 −2
𝑛 𝑛 (−1)𝑖 (∏𝛼𝑗 𝑥𝑗 𝑗 ) < 0. (18)
2 𝑛 𝑛𝛼 −2
det [∇ 𝑓2 (𝑥)] = (−𝑐2 ) (∏𝛼𝑖 𝑥𝑖 𝑖 ) (1 − ∑𝛼𝑖 ) . (13) 𝑗=1
𝑖=1 𝑖=1
̂ 𝑖 > 0 for each 𝑖. Thus,
The above two inequalities yield that Δ
We will complete the proof by discussing the following two following the same arguments as in Proposition 4, we can
cases. conclude that 𝑓1 is also a convex function under condition
(15).
Case i. If 1 − ∑𝑛𝑖=1 𝛼𝑖 = 0, then it is not hard to verify
that ∇2 𝑓2 (𝑥) 𝑥 is a zero vector for any 𝑥 ∈ R𝑛 . Hence, Remark 5. If 𝑐 > 0, 𝛼𝑖 < 0, for 𝑖 = 1, 2, . . . , 𝑚, and 𝛼𝑖 > 0,
𝑥𝑇 ∇2 𝑓2 (𝑥) 𝑥 = 0 which says that ∇2 𝑓2 (𝑥) is a positive 𝑖 = 𝑚 + 1, 𝑚 + 2, . . . , 𝑛, then 𝑓(𝑥) can be converted into the
semidefinite matrix by definition. Therefore, 𝑓2 (𝑥) is convex following convex function:
under this case.
𝑚 𝑛
𝛼 −𝛼
Case ii. If 1 − ∑𝑛𝑖=1 𝛼𝑖 > 0, then we know from (13) that 𝑓󸀠 (𝑥) = 𝑐 (∏𝑥𝑖 𝑖 ) ( ∏ 𝑧𝑖 𝑖 ) , (19)
𝑖=1 𝑖=𝑚+1
𝑖 𝑖
𝑖 𝑖𝛼 −2
Δ 𝑖 = (−𝑐2 ) (∏𝛼𝑗 𝑥𝑗 𝑗 ) (1 − ∑𝛼𝑗 ) , (14) where 𝑧𝑖 = 𝑥𝑖−1 , 𝑖 = 𝑚 + 1, 𝑚 + 2, . . . , 𝑛.
𝑗=1 𝑗=1
Remark 6. If 𝑐 < 0, 𝛼𝑖 > 0, for 𝑖 = 1, 2, . . . , 𝑚, and 𝛼𝑖 < 0,
where Δ 𝑖 denotes the 𝑖th principal minor of the Hessian 𝑖 = 𝑚 + 1, 𝑚 + 2, . . . , 𝑛, then 𝑓(𝑥) can be converted into the
matrix of 𝑓2 (𝑥). Note that 𝑐2 < 0, 𝛼𝑖 > 0 for all 𝑖 = 1, 2, . . . , 𝑛, following convex function:
and 1 − ∑𝑛𝑖=1 𝛼𝑖 > 0. Therefore, it can be seen that Δ 𝑖 > 0
𝑚 𝑛
for all 𝑖 = 1, 2, . . . , 𝑛, which implies (by Lemma 1(b)) that 𝛼 /𝑡 −𝛼𝑖 /𝑡
∇2 𝑓2 (𝑥) is a positive definite matrix. This says that 𝑓2 (𝑥) is 𝑓󸀠 (𝑥) = 𝑐 (∏𝑧𝑖 𝑖 ) ( ∏ 𝑧𝑖 ), (20)
𝑖=1 𝑖=𝑚+1
strictly convex under this case.
From all the previous, the desired result follows. where
We want to point out that our results also provide an (i) 𝑡 is the smallest positive integer satisfying 1 −
alternative proof for the main result (Theorem 7) of [28]. (1/𝑡)(∑𝑚 𝑛
𝑖=1 𝛼𝑖 − ∑𝑖=𝑚+1 𝛼𝑖 ) ≥ 0.
Indeed, Maranas and Floudas [28] further discuss another
(ii) 𝑧𝑖 = 𝑥𝑖𝑡 , for 𝛼𝑖 > 0, 𝑖 = 1, 2, . . . , 𝑚.
condition as follows:
𝑛
(iii) 𝑧𝑖 = 𝑥𝑖−𝑡 , for 𝛼𝑖 < 0, 𝑖 = 𝑚 + 1, 𝑚 + 2, . . . , 𝑛.
∃𝑗 such that 𝛼𝑗 ≥ 1 − ∑ 𝛼𝑖 ,
𝑖 ≠ 𝑗
Obviously, all signomial terms are converted into convex
(15)
and concave terms by the proposed methods. All concave
𝛼𝑖 ≤ 0, ∀𝑖 ≠ 𝑗, 𝑖 = 1, 2, . . . 𝑛, terms in the GP problem can be further represented by an
effective piecewise linearization method as follows.
to guarantee that 𝑓1 defined as in Proposition 3 is a convex Conventional methods for linearizing a concave function
function. Our approach can be also employed to verify this with 𝑚 break points require 𝑚 − 1 binary variables. However,
4 Mathematical Problems in Engineering

̂
f(x) Table 1: Information of interval of 𝑥.
f(x) Interval of 𝑥 𝜃 𝑢1 𝑢2 𝑢3 𝑢4 𝐺(𝜃) ‖𝐺(𝜃)‖
sm−1
1≤𝑥≤2 1 0 0 0 0 ⌀ 0
sk
2≤𝑥≤3 2 1 0 0 0 1 1
3≤𝑥≤4 3 0 1 0 0 2 1
s2 4≤𝑥≤5 4 1 1 0 0 1, 2 2
5≤𝑥≤6 5 0 0 1 0 3 1
s1 6≤𝑥≤7 6 1 0 1 0 1, 3 2
7≤𝑥≤8 7 0 1 1 0 2, 3 2
··· ··· 8≤𝑥≤9 8 1 1 1 0 1, 2, 3 3
9 ≤ 𝑥 ≤ 10 9 0 0 0 1 4 1
10 ≤ 𝑥 ≤ 11 10 1 0 0 1 1, 4 2

a1 a2 a3 ··· ak ak+1 ··· am−1
Figure 1: A piecewise linear function of 𝑓(𝑥).
when 𝑚 becomes large, the computation will be very time
consuming and may cause a heavy computational burden. An
effective piecewise linearization method proposed by Huang
[29] is presented in which only ⌈log2 (𝑚 − 1)⌉ binary variables
are used.
Consider a grid of 𝑚 points throughout the interval of
[𝑎1 , 𝑎𝑚 ] and denote the break points 𝑎1 , 𝑎2 , . . . , 𝑎𝑚 , where
𝑎1 < 𝑎2 < ⋅ ⋅ ⋅ < 𝑎𝑚 . 𝑓(𝑎𝑘 ) is evaluated at each 𝑎𝑘 , and
two adjacent break points (𝑎𝑘 , 𝑓(𝑎𝑘 )) and (𝑎𝑘+1 , 𝑓(𝑎𝑘+1 )) are
connected with a straight line as shown in Figure 1.
Let 𝜃 be an integer, 1 ≤ 𝜃 ≤ 𝑚 − 1, expressed as
x
am The following proposition is deduced.
Proposition 7. Given a set of binary variables 𝐷 =
{𝑢1 , 𝑢2 , . . . , 𝑢ℎ }, where 𝑢𝑗 ∈ {0, 1}, 𝑗 = 1, 2, . . . , ℎ, and ℎ =
⌈log2 (𝑚 − 1)⌉, there is one and only one 𝑘, 1 ≤ 𝑘 ≤ 𝑚 − 1, such
that
(i) 𝑞𝑘 = 0 and ‖𝐺(𝑘)‖ = ∑𝑢𝑗 ∈𝐷,𝑗∈𝐺(𝑘) 𝑢𝑗 − ∑𝑢𝑗 ∈𝐷,𝑗∉𝐺(𝑘) 𝑢𝑗 ,
(ii) 𝑞𝜃 ≥ 1 for 1 ≤ 𝜃 ≤ 𝑚 − 1, 𝜃 ≠ 𝑘.
Taking Table 1 for instance, given 1 ≤ 𝑥 ≤ 11, 𝑚 = 11. Let
𝐷 = {𝑢1 , 𝑢2 , 𝑢3 , 𝑢4 } = {1, 0, 0, 1}; then we have
𝑞1 = ‖𝐺 (1)‖ + 𝑢1 + 𝑢2 + 𝑢3 + 𝑢4 = 0 + 1 + 0 + 0 + 1 = 2
𝑞2 = ‖𝐺 (2)‖ − 𝑢1 + 𝑢2 + 𝑢3 + 𝑢4 = 1 − 1 + 0 + 0 + 1 = 1

ℎ ..
.
𝜃 = 1 + ∑2𝑗−1 𝑢𝑗 , (21)
𝑗=1 𝑞8 = ‖𝐺 (8)‖ − 𝑢1 − 𝑢2 − 𝑢3 + 𝑢4 = 3 − 1 − 0 − 0 + 1 = 3
where ℎ = ⌈log2 (𝑚 − 1)⌉ and 𝑢𝑗 ∈ {0, 1}. Given the values of 𝑞9 = ‖𝐺 (9)‖ + 𝑢1 + 𝑢2 + 𝑢3 − 𝑢4 = 1 + 1 + 0 + 0 − 1 = 1
𝑢𝑗 , 1 ≤ 𝑗 ≤ ℎ, there is a unique 𝜃.
Let 𝐺(𝜃) ⊆ {1, 2, . . . , ℎ} be a set composed of all indices 𝑞10 = ‖𝐺 (10)‖ − 𝑢1 + 𝑢2 + 𝑢3 − 𝑢4 = 2 − 1 + 0 + 0 − 1 = 0.
corresponding to a given value 𝜃, such that (24)

∑ 2𝑗−1 = 𝜃 − 1. (22)
Since 𝑞10 = 0 and 𝑞𝜃 ≥ 1 for 1 ≤ 𝜃 ≤ 10, 𝜃 ≠ 10, then
𝑗∈𝐺(𝜃) ̂ = 𝑓(𝑎 ) +
only 𝑎10 ≤ 𝑥 ≤ 𝑎11 is activated and results in 𝑓(𝑥) 10
𝑠10 (𝑥 − 𝑎10 ).
For instance, 𝐺(1) = ⌀, 𝐺(2) = {1}, 𝐺(3) = {2}, 𝐺(4) = We then have the following theorem.
{1, 2}, . . ., 𝐺(10) = {1, 4}, and so forth.
Denote by ‖𝐺(𝜃)‖ the number of elements in 𝐺(𝜃). For ̂
Theorem 8. Consider a piecewise linear function 𝑓(𝑥) of a
instance, given 1 ≤ 𝑥 ≤ 11, then ℎ = ⌈log2 (𝑚 − 1)⌉ = 4.
concave function 𝑓(𝑥), where 𝑥 is divided into 𝑚 − 1 intervals
‖𝐺(1)‖ = 0, ‖𝐺(2)‖ = 1, ‖𝐺(4)‖ = 2, and so forth. A list of all
by 𝑚 break points, 𝑎1 < 𝑎2 < ⋅ ⋅ ⋅ < 𝑎𝑚 . Given a set of binary
𝜃, 𝑢𝑗 , 𝐺(𝜃), and ‖𝐺(𝜃)‖ is shown in Table 1.
variables 𝐷 = {𝑢1 , 𝑢2 , . . . , 𝑢ℎ }, where ℎ = ⌈log2 (𝑚 − 1)⌉, 𝑥 and
Now we intend to use 𝑘 as an index to indicate that the ̂ can be expressed as the following linear inequalities:
𝑘th interval is activated, which results in 𝑎𝑘 ≤ 𝑥 ≤ 𝑎𝑘+1 and 𝑓(𝑥)
̂ = 𝑓(𝑎 ) + 𝑠 (𝑥 − 𝑎 ).
𝑓(𝑥) 𝑎𝜃 − 𝑀𝑞𝜃 ≤ 𝑥 ≤ 𝑎𝜃+1 + 𝑀𝑞𝜃 ,
𝑘 𝑘 𝑘 (25)
Define 𝑚 − 1 functions 𝑞𝜃 based on the combination of 𝑢𝑗
as follows: 𝑓 (𝑎𝜃 ) + 𝑠𝜃 (𝑥 − 𝑎𝜃 ) − 𝑀󸀠 𝑞𝜃
(26)
𝑞𝜃 = ‖𝐺 (𝜃)‖ − ∑ 𝑢𝑗 + ∑ 𝑢𝑗 ,
(23) ≤ 𝑓̂ (𝑥) ≤ 𝑓 (𝑎𝜃 ) + 𝑠𝜃 (𝑥 − 𝑎𝜃 ) + 𝑀󸀠 𝑞𝜃 ,
𝑢𝑗 ∈𝐷,𝑗∈𝐺(𝜃) 𝑢𝑗 ∈𝐷,𝑗∉𝐺(𝜃)

where 𝜃 = 1, 2, . . . , 𝑚 − 1, 𝑗 = 1, 2, . . . , ℎ, and ℎ = ⌈log2 (𝑚 − 𝑞𝜃 = ‖𝐺 (𝜃)‖ − ∑ 𝑢𝑗 + ∑ 𝑢𝑗 ,

(27)
𝑢𝑗 ∈𝐷,𝑗∈𝐺(𝜃) 𝑢𝑗 ∈𝐷,𝑗∉𝐺(𝜃)
1)⌉.
Mathematical Problems in Engineering 5

where 𝑠𝜃 = (𝑓(𝑎𝜃+1 )−𝑓(𝑎𝜃 ))/(𝑎𝜃+1 −𝑎𝜃 ), 𝜃 = 1, 2, . . . , 𝑚−1. 𝑀, x7

and 𝑀󸀠 are large positive values, 𝑀 = max{𝑎𝑚 , 1} and 𝑀󸀠 =
max{𝑓(𝑎1 ), 𝑓(𝑎2 ), . . . , 𝑓(𝑎𝑚 ), 1}.

Proof. Given the value of a specific 𝐷 = 𝐷0 , 𝐷0 =

{𝑢10 , 𝑢20 , . . . , 𝑢ℎ0 }, consider the following two cases.
x4
Case i. There is one and only one integer 𝜃0 , 𝜃0 = 1 +
x2 x3
∑ℎ𝑗=1 2𝑗−1 𝑢𝑗0 such that

(a) 𝐺(𝜃0 ) = {𝑗 | 𝑢𝑗0 = 1, x5 x1

(b) ‖𝐺(𝜃0 )‖ = ∑𝑗∈𝐺(𝜃0 ) 𝑢𝑗0 ,

(c) ∑𝑢𝑗0 ∈𝐷,𝑗∉𝐺(𝜃0 ) 𝑢𝑗0 = 0.

Expressions (a), (b), and (c) force 𝑞𝜃0 = 0 (from (27)). Then,
̂ =
𝑎𝜃0 ≤ 𝑥 ≤ 𝑎𝜃0 +1 is activated (from (25)) and results in 𝑓(𝑥) x6
𝑓(𝑎𝜃0 ) + 𝑠𝜃0 (𝑥 − 𝑎𝜃0 ) (from (26)).
Figure 2: Schematic of the welded beam problem, Example 2.
Case ii. For other 𝑚 − 1 integers 1 ≤ 𝜃󸀠 ≤ 𝑚, 𝜃󸀠 ≠ 𝜃0 , since
‖𝐺(𝜃󸀠 )‖ > ∑𝑗∈𝐺(𝜃󸀠 ) 𝑢𝑗0 + ∑𝑢𝑗0 ∈𝐷,𝑗∉𝐺(𝜃󸀠 ) 𝑢𝑗0 , we then have 𝑞𝜃󸀠 ≥ 1.
In this case, inequalities (25) and (26) are still correct but not contact stress of the gear tooth), ℎ3 , ℎ4 (upper bounds on the
being activated. transverse deflection of shafts 1, 2), ℎ5 , ℎ6 (upper bounds on
the stresses in shafts 1, 2), ℎ7 , ℎ8 , ℎ9 (dimensional restrictions
Compared with the conventional piecewise linearization
based on space and experience), and ℎ10 , ℎ11 (dimensional
methods [30–36], the number of newly added binary vari-
requirements for shafts based on experience).
ables in the proposed method for a piecewise linear function
The problem can be formulated as
with 𝑚 break points is significantly reduced from 𝑚 − 1 to
⌈log2 (𝑚 − 1)⌉. The algorithm is listed in Algorithm 1 based on min 2.6179𝑥1 𝑥22 𝑥32 + 11.7287𝑥1 𝑥22 𝑥3
Theorem 8.
− 33.84556𝑥1 𝑥22 − 1.508𝑥1 𝑥62
Algorithm 1 (the proposed piecewise linearization method). (28)
Step 1. Identify the number of subintervals and choose the − 1.508𝑥1 𝑥72 + 7.4777𝑥63 + 7.4777𝑥73
break points, 𝑎𝑖 , 𝑖 = 1, 2, . . . , 𝑚.
Step 2. Calculate the values of ℎ = ⌈log2 𝑚 − 1⌉ based on + 0.7854𝑥4 𝑥62 + 0.7854𝑥5 𝑥72
the number of break points selected and obtain 𝑞𝜃 , 𝜃 =
1, 2, . . . , 𝑚 − 1, based on the combination of 𝑢ℎ as (27). s.t. ℎ1 = −𝑥1 𝑥22 𝑥3 + 27.0 ≤ 0, (29)

̂ as (25)
Step 3. Generate the set of constraints for 𝑥 and 𝑓(𝑥) ℎ2 = −𝑥1 𝑥22 𝑥32 + 397.5 ≤ 0, (30)
and (26), respectively, and solve the resulting problem.
ℎ3 = −𝑥2 𝑥3 𝑥64 + 1.93𝑥43 ≤ 0, (31)

3. Numerical Examples ℎ4 = −𝑥2 𝑥3 𝑥74 + 1.93𝑥53 ≤ 0, (32)

In this section, we have conducted some engineering design ℎ5 = 555025.0𝑥42 + 1.690 × 107 𝑥22 𝑥32
problems to evaluate the usefulness of the proposed methods. (33)
− 12100.0𝑥22 𝑥32 𝑥66 ≤ 0,
Example 1. Consider the following engineering design prob-
lem of a speed reducer which was proposed by Golinski [37] ℎ6 = 555025.0𝑥52 + 1.575 × 108 𝑥22 𝑥32
as in Figure 2. (34)
− 7225.0𝑥22 𝑥32 𝑥76 ≤ 0,
The objective of this problem is to minimize the weight
of the speed reducer while satisfying a number of constraints ℎ7 = 𝑥2 𝑥3 − 40.0 ≤ 0, (35)
imposed by gear and shaft design practices. There are seven
ℎ8 = −𝑥1 + 5𝑥2 ≤ 0, (36)
design variables, 𝑥1 (width of the gear face, cm), 𝑥2 (teeth
module, cm), 𝑥3 (number of pinion teeth), 𝑥4 (shaft 1 length ℎ9 = 𝑥1 − 12𝑥2 ≤ 0, (37)
between bearings, cm), 𝑥5 (shaft 2 length between bearings,
cm), 𝑥6 (diameter of shaft 1, cm), and 𝑥7 (diameter of shaft 2, ℎ10 = −𝑥4 + 1.5𝑥6 + 1.9 ≤ 0, (38)
cm). The constraints are characterized by ℎ1 (upper bound on
the bending stress of the gear tooth), ℎ2 (upper bound on the ℎ11 = −𝑥5 + 1.1𝑥7 + 1.9 ≤ 0, (39)
6 Mathematical Problems in Engineering

2.6 ≤ 𝑥1 ≤ 3.6, 0.7 ≤ 𝑥2 ≤ 0.8, Table 2: Comparison of Example 1.

(40)
17.0 ≤ 𝑥3 ≤ 28.0, 7.3 ≤ 𝑥4 ≤ 8.3, Item Coello and Cortés [24] Proposed method
𝑥1 3.50 3.4956520
7.3 ≤ 𝑥5 ≤ 8.3, 2.9 ≤ 𝑥6 ≤ 3.9, 5.0 ≤ 𝑥7 ≤ 5.5.
𝑥2 0.70 0.7000002
(41)
𝑥3 17.0 17.0
𝑥4 7.300008 7.30000007
This problem has been solved by Coello and 𝑥5 7.715322 7.7120386
Cortés [24] using genetic algorithm. The solution is 𝑥6 3.350215 3.3433720
(𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 , 𝑥5 , 𝑥6 , 𝑥7 ) = (3.5, 0.7, 17.0, 7.300008, 7.715322, 𝑥7 5.286655 5.2853520
3.350215, 5.286655) with the objective value 2994.3419. Objective 2994.3419 2989.5140
For instance, the nonlinear terms 𝑥1 𝑥22 𝑥32 , 𝑥1 𝑥22 𝑥3 , and
𝑥4 𝑥62 can be convexified as 𝑧1,1 −1 −2 −2 −1 −2 −1 −1 −2
𝑧2,1 𝑧3,1 , 𝑧1,1 𝑧2,1 𝑧3,1 , and 𝑧4,1 𝑧6,1 x3 F F x
1
−1
by Remark 5, where 𝑧𝑘,1 = 𝑥𝑘 , 𝑘 = 1, 2, . . . , 7. The
nonlinear terms −𝑥1 𝑥22 , −𝑥1 𝑥62 , and −𝑥1 𝑥72 can be convexified x6
1/3 2/3 1/3 2/3 1/3 2/3
as −𝑧1,2 𝑧2,2 , −𝑧1,2 𝑧6,2 , and −𝑧1,2 𝑧7,2 by Remark 6, where
x2
𝑧1,2 = 𝑥1 , 𝑧2,2 = 𝑥2 , 𝑧6,2 = 𝑥63 , and 𝑧7,2 = 𝑥73 . The
3 3

concave terms can be further represented by the proposed

piecewise linearization method. Take 𝑧1,1 = 𝑥1−1 for instance;
suppose 41 equal-distance break points are used to linearize
𝑧1,1 = 𝑥1−1 , as given by 𝑎1,1 = 2.600, 𝑎1,2 = 2.625, 𝑎1,3 = L x4
2.650, . . . , and 𝑎1,41 = 3.600. Then 𝑧1,1 = 𝑥1−1 can be
Figure 3: Golinski’s speed reducer problem, Example 1.
represented by Algorithm 2 as follows.

Algorithm 2. Step 1. The proposed method requires the use 2.600−1 + 𝑠1,1 (𝑥1 − 2.600) − 𝑀󸀠 𝑞1,1
of ⌈log2 (41 − 1)⌉ = 6 binary variables to linearize 𝑧1,1 =
𝑥1−1 in a piecewise manner. Compared with the conventional ≤ 𝑧1,1 ≤ 2.600−1 + 𝑠1,1 (𝑥1 − 2.600) + 𝑀󸀠 𝑞1,1
piecewise linearization methods, the number of newly added
2.625−1 + 𝑠1,2 (𝑥1 − 2.625) − 𝑀󸀠 𝑞1,2
binary variables for a piecewise linear function with 41 break
points is significantly reduced from 40 to 6.
≤ 𝑧1,1 ≤ 2.625−1 + 𝑠1,2 (𝑥1 − 2.625) + 𝑀󸀠 𝑞1,2 (44)
Step 2. ℎ1,1 = ⌈log2 (41 − 1)⌉ = 6. A set of 𝑞1,𝜃 , 𝜃 = 1, 2, . . . , 40, ..
.
based on the combination of 𝑢1,ℎ1,1 is expressed as (42)
3.575−1 + 𝑠1,40 (𝑥1 − 3.575) − 𝑀󸀠 𝑞1,40

𝑞1,1 = 0 + 𝑢1,1 + 𝑢1,2 + 𝑢1,3 + 𝑢1,4 + 𝑢1,5 + 𝑢1,6 ≤ 𝑧1,1 ≤ 3.575−1 + 𝑠1,40 (𝑥1 − 3.575) + 𝑀󸀠 𝑞1,40 ,
−1 −1
𝑞1,2 = 1 − 𝑢1,1 + 𝑢1,2 + 𝑢1,3 + 𝑢1,4 + 𝑢1,5 + 𝑢1,6 where 𝑠1,𝜃 = (𝑎1,𝜃+1 − 𝑎1,𝜃 )/(𝑎1,𝜃+1 − 𝑎1,𝜃 ), 𝜃 = 1, 2, . . . , 40. 𝑀
󸀠
(42) and 𝑀 are large positive values.
..
.
Solving the transformed program with MATLAB R13,
𝑞1,40 = 4 − 𝑢1,1 − 𝑢1,2 − 𝑢1,3 + 𝑢1,4 + 𝑢1,5 − 𝑢1,6 . Table 2 shows the results computed by Coello and Cortés [24]
and the proposed method.

Example 2 (see [38]). Consider the design of a welded beam

Step 3. Based on Theorem 8, 𝑥1 and 𝑧1,1 can be expressed as
which is to be rigidly attached to a fixed support and is
the following linear inequalities:
subjected to a load as illustrated in Figure 3. The welded beam
is to consist of 1010 steel and is to support a force 𝐹 of 6000 lb.
The length 𝐿 is assumed to be specified at 14 in.
2.600 − 𝑀𝑞1,1 ≤ 𝑥1 ≤ 2.625 + 𝑀𝑞1,1
2.625 − 𝑀𝑞1,2 ≤ 𝑥1 ≤ 2.650 + 𝑀𝑞1,2 Three methods (brazed, bolted, and welded connections)
of attachment and three materials (steel, brass, and alu-
(43) minum) are considered. The choices of attachment and mate-
..
. rials are to be made, which minimizes cost while satisfying
a limit on the allowable bending stress, the buckling load,
3.575 − 𝑀𝑞1,40 ≤ 𝑥1 ≤ 3.600 + 𝑀𝑞1,40 , and the allowable deflection at the load application point.
Mathematical Problems in Engineering
The various materials have different elastic, shear moduli, and
different costs.
The design variables are defined as 𝑥1 (a continuous
variable for thickness of weld), 𝑥2 (a continuous variable
for width of beam), 𝑥3 (a continuous variable for length of
connection), 𝑥4 (a continuous variable for thickness of beam),
𝑥5 (a continuous variable for diameter of bolts), 𝑥6 (a 0-1
variable for brazed connections), 𝑥7 (a 0-1 variable for bolted
connections), 𝑥8 (a 0-1 variable for welded connections), 𝑥9
(a 0-1 variable for steel material), 𝑥10 (a 0-1 variable for brass
material), and 𝑥11 (a 0-1 variable for aluminum material).
The problem can be formulated as (45)–(50)
min 1.2𝑥2 𝑥3 𝑥6 + 𝑥5 𝑥7 + 𝑥12 𝑥3 𝑥8
+ 0.04811𝐿𝑥2 𝑥4 𝑥9 + 0.04811𝑥2 𝑥3 𝑥4 𝑥9
+ 0.06𝐿𝑥2 𝑥4 𝑥10 + 0.06𝑥2 𝑥3 𝑥4 𝑥10
+ 0.015𝐿𝑥2 𝑥4 𝑥11 + 0.015𝑥2 𝑥3 𝑥4 𝑥11
6𝐹𝐿
s.t ℎ1 = −𝜎allowable + ≤ 0,
𝑥22 𝑥4
ℎ2 = −𝑃𝑐 + 6000 ≤ 0,
3 University Press, Cambridge, UK, 2004.
4𝐹𝐿
ℎ3 = −𝛿max + ≤ 0,
𝐸𝑥23 𝑥4
𝑥6 + 𝑥7 + 𝑥8 = 1,
𝑥9 + 𝑥10 + 𝑥11 = 1,
where 𝜎allowable is the maximum bar bending stress. 𝑃𝑐 is the
bar buckling load which is a function of the beam dimensions
and material properties. 𝛿max is the maximum bar deflection.
𝐸 is the Young’s modulus, 30 × 106 psi. More detailed
description including the derivation of 𝑃𝑐 can be found in
Reklaitis et al. [38].
Following the same procedure as in Example 1,
the nonlinear terms 𝑥2 𝑥3 𝑥6 , 𝑥5 𝑥7 , 𝑥12 𝑥3 𝑥8 , 𝑥2 𝑥4 𝑥9 ,
𝑥2 𝑥3 𝑥4 𝑥9 , 𝑥2 𝑥4 𝑥10 , 𝑥2 𝑥3 𝑥4 𝑥10 , 𝑥2 𝑥4 𝑥11 , and 𝑥2 𝑥3 𝑥4 𝑥11
can be convexified by Remark 5. The concave terms are
further represented by the proposed piecewise lineariza-
tion method. Solving the transformed program with
MATLAB R13, we can obtain the global optimum with
(𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 , 𝑥5 , 𝑥6 , 𝑥7 , 𝑥8 , 𝑥9 , 𝑥10 , 𝑥11 ) = (0.320, 14.109,
5.720, 0.321, 4.992, 0, 0, 1, 0, 0, 1) and the objective value is
1.969 within the tolerance error 0.0001.
4. Conclusions
This paper proposes an optimization approach for solving
geometric programming problems. Our approach is first to
convert all signomial terms in GP into convex and concave
terms by the proposed methods. Then the concave terms
are further treated with the proposed piecewise linearization
method where only ⌈log2 (𝑚 − 1)⌉ binary variables are used.
It has the following features.
(i) It offers more convenient and efficient means of
expressing a piecewise linear function.
7
(ii) Fewer 0-1 variables are used.
(iii) The computational results show that the proposed
method is much more efficient and faster than the
conventional one, especially when the number of
break points becomes large.
Numerical examples in real applications are illustrated to
demonstrate the usefulness of the proposed methods.
Acknowledgments
The author would like to thank Distinguished Chair Professor
Gwo-Hshiung Tzeng for reading the first draft of this paper
and giving him many useful comments to improve this paper.
The author also thanks the anonymous referees for their
(45) careful reading and fruitful comments on the paper. The
author especially acknowledges the National Science Council,
Taiwan, for financially supporting this research under Grant
no. NSC 102–2410-H-141-012-MY2.
(46)
References
(47) [1] S. Boyd and L. Vandenberghe, Convex Optimization, Cambridge
(48) [2] S. P. Boyd, S.-J. Kim, D. D. Patil, and M. A. Horowitz, “Digital
circuit optimization via geometric programming,” Operations
Research, vol. 53, no. 6, pp. 899–932, 2005.
(49)
[3] C. Chu and D. F. Wong, “VLSI circuit performance optimization
(50) by geometric programming,” Annals of Operations Research, vol.
105, pp. 37–60, 2001.
[4] M. Del Mar Hershenson, S. P. Boyd, and T. H. Lee, “Optimal
design of a CMOS op-amp via geometric programming,” IEEE
Transactions on Computer-Aided Design of Integrated Circuits
and Systems, vol. 20, no. 1, pp. 1–21, 2001.
[5] T.-S. Moh, T.-S. Chang, and S. L. Hakimi, “Globally optimal
floorplanning for a layout problem,” IEEE Transactions on
Circuits and Systems I, vol. 43, no. 9, pp. 713–720, 1996.
[6] S. S. Mohan, M. D. M. Hershenson, S. P. Boyd, and T. H.
Lee, “Simple accurate expressions for planar spiral inductances,”
IEEE Journal of Solid-State Circuits, vol. 34, no. 10, pp. 1419–1420,
1999.
[7] S. S. Mohan, M. Del Mar Hershenson, S. P. Boyd, and T. H.
Lee, “Bandwidth extension in CMOS with optimized on-chip
inductors,” IEEE Journal of Solid-State Circuits, vol. 35, no. 3, pp.
346–355, 2000.
[8] E. Rosenberg, “Optimal module sizing in VLSI floorplanning
by nonlinear programming,” Zeitschrift für Operations Research.
Mathematical Methods of Operations Research, vol. 33, no. 2, pp.
131–143, 1989.
[9] R. J. Clasen, “The solution of the chemical equilibrium pro-
gramming problem with generalized benders decomposition,”
Operations Research, vol. 32, no. 1, pp. 70–79, 1984.
[10] C. D. Maranas and C. A. Floudas, “Global optimization in
generalized geometric programming,” Computers and Chemical
Engineering, vol. 21, no. 4, pp. 351–369, 1997.
[11] T. W. Wall, D. Greening, and R. E. D. Woolsey, “Solving
complex chemical equilibria using a geometric-programming
based technique,” Operations Research, vol. 34, no. 3, pp. 345–
355, 1986.
8 Mathematical Problems in Engineering

[12] N. Jha, “A discrete data base multiple objective optimization of [31] K. L. Croxton, B. Gendron, and T. L. Magnanti, “A comparison
milling operation through geometric programming,” Journal of of mixed-integer programming models for nonconvex piece-
Engineering For Industry, vol. 112, pp. 368–374, 1990. wise linear cost minimization problems,” Management Science,
[13] H.-L. Li, J.-F. Tsai, and C. A. Floudas, “Convex underestimation vol. 49, no. 9, pp. 1268–1273, 2003.
for posynomial functions of positive variables,” Optimization [32] C. A. Floudas, Nonlinear and Mixed-Integer Optimization:
Letters, vol. 2, no. 3, pp. 333–340, 2008. Fundamentals and Applications, Oxford University Press, 1995.
[14] M.-H. Lin and J.-F. Tsai, “Finding multiple optimal solutions of [33] C. A. Floudas, Deterministic Global Optimization: Theory, Meth-
signomial discrete programming problems with free variables,” ods and Applications, Kluwer Academic Publishers, Boston,
Optimization and Engineering, vol. 12, no. 3, pp. 425–443, 2011. Mass, USA, 2000.
[15] A. I. Sönmez, A. Baykasoǧlu, T. Dereli, and I. H. Filiz, “Dynamic [34] S. Kontogiorgis, “Practical piecewise-linear approximation for
optimization of multipass milling operations via geometric monotropic optimization,” INFORMS Journal on Computing,
programming,” International Journal of Machine Tools and vol. 12, no. 4, pp. 324–340, 2000.
Manufacture, vol. 39, no. 2, pp. 297–320, 1999. [35] M. Padberg, “Approximating separable nonlinear functions via
[16] J.-F. Tsai, “Treating free variables in generalized geometric mixed zero-one programs,” Operations Research Letters, vol. 27,
programming problems,” Computers and Chemical Engineering, no. 1, pp. 1–5, 2000.
vol. 33, no. 1, pp. 239–243, 2009. [36] H. Topaloglu and W. B. Powell, “An algorithm for approximat-
ing piecewise linear concave functions from sample gradients,”
[17] J.-F. Tsai, “Global optimization for signomial discrete program-
Operations Research Letters, vol. 31, no. 1, pp. 66–76, 2003.
ming problems in engineering design,” Engineering Optimiza-
tion, vol. 42, no. 9, pp. 833–843, 2010. [37] J. Golinski, “An adaptive optimization system applied to
machine synthesis,” Mechanism and Machine Theory, vol. 8, no.
[18] J.-F. Tsai and M.-H. Lin, “Global optimization of signomial
4, pp. 419–436, 1973.
mixed-integer nonlinear programming problems with free
variables,” Journal of Global Optimization, vol. 42, no. 1, pp. 39– [38] G. V. Reklaitis, A. Ravindran, and K. M. Ragsdell, Engineering
49, 2008. Optimization, Methods and Applications, John Wiley & Sons,
New York, NY, USA, 2nd edition, 2006.
[19] J.-F. Tsai and M.-H. Lin, “An efficient global approach for posyn-
omial geometric programming problems,” INFORMS Journal
on Computing, vol. 23, no. 3, pp. 483–492, 2011.
[20] M. Chiang, C. W. Tan, D. P. Palomar, D. O’Neill, and D. Julian,
“Power control by geometric programming,” IEEE Transactions
on Wireless Communications, vol. 6, no. 7, pp. 2640–2651, 2007.
[21] M. Chiang and S. Boyd, “Geometric programming duals of
channel capacity and rate distortion,” IEEE Transactions on
Information Theory, vol. 50, no. 2, pp. 245–258, 2004.
[22] A. Dutta and D. V. Rama, “An optimization model of commu-
nications satellite planning,” IEEE Transactions on Communica-
tions, vol. 40, no. 9, pp. 1463–1473, 1992.
[23] S. Kandukuri and S. Boyd, “Optimal power control in
interference-limited fading wireless channels with outage-
probability specifications,” IEEE Transactions on Wireless Com-
munications, vol. 1, no. 1, pp. 46–55, 2002.
[24] C. A. C. Coello and N. C. Cortés, “Hybridizing a genetic algo-
rithm with an artificial immune system for global optimization,”
Engineering Optimization, vol. 36, no. 5, pp. 607–634, 2004.
[25] R. Horst and H. Tuy, Global Optimization: Deterministic
Approaches, Springer, New York, NY, USA, 2003.
[26] H. D. Sherali and C. H. Tuncbilek, “A global optimization
algorithm for polynomial programming problems using a
reformulation-linearization technique,” Journal of Global Opti-
mization, vol. 2, no. 1, pp. 101–112, 1992.
[27] L. D. Berkovitz, Convexity and Optimization in R𝑛 , John Wiley
& Sons, New York, NY, USA, 2002.
[28] C. D. Maranas and C. A. Floudas, “Finding all solutions of
nonlinearly constrained systems of equations,” Journal of Global
Optimization, vol. 7, no. 2, pp. 143–182, 1995.
[29] C.-H. Huang, “An effective linear approximation method for
separable programming problems,” Applied Mathematics and
Computation, vol. 215, no. 4, pp. 1496–1506, 2009.
[30] M. S. Bazaraa, H. D. Sherali, and C. M. Shetty, Nonlinear
Programming: Theory and Algorithms, John Wiley & Sons, New
York, NY, USA, 2nd edition, 1993.
Advances in Advances in Journal of Journal of
Operations Research
Hindawi Publishing Corporation
Decision Sciences
Hindawi Publishing Corporation
Applied Mathematics
Hindawi Publishing Corporation
Algebra
Hindawi Publishing Corporation
Probability and Statistics
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

The Scientific International Journal of

World Journal
Hindawi Publishing Corporation
Differential Equations
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Submit your manuscripts at

http://www.hindawi.com

International Journal of Advances in

Combinatorics
Hindawi Publishing Corporation
Mathematical Physics
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of Journal of Mathematical Problems Abstract and Discrete Dynamics in

Complex Analysis
Hindawi Publishing Corporation
Mathematics
Hindawi Publishing Corporation
in Engineering
Hindawi Publishing Corporation
Applied Analysis
Hindawi Publishing Corporation
Nature and Society
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

International
Journal of Journal of
Mathematics and
Mathematical
Discrete Mathematics
Sciences

Journal of International Journal of Journal of

Hindawi Publishing Corporation Hindawi Publishing Corporation Volume 2014

http://www.hindawi.com
Function Spaces
Hindawi Publishing Corporation
Volume 2014 http://www.hindawi.com http://www.hindawi.com
Stochastic Analysis
Hindawi Publishing Corporation
Optimization
Hindawi Publishing Corporation
Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014