2 Journal of Applied Mathematics
by using the change of variable =
𝑇
,where is a
lower triangular matrix. e rank-two relaxation heuristics
algorithm in [11] relaxed the max-cut problem to form
an unconstrained optimization problem by replacing each
binary variable with one unit vector in space
2
and then
using polar coordinates. In [12], the rank-two SDP relaxation
model is proposed for maximal independent set problem.
Based on the low-rank decomposition of the semidenite
matrix, Liu et al. [13] proposed a feasible direction method
to solve a nonlinear programming model of binary quadratic
programming.
Inthepaper,weproposearank-twofeasibledirection
method for the binary quadratic programming. we restrict
therankofmatrixvariabletobetwointhesemidenite
programming relaxation and obtain a quadratic objective
function with simple quadratic constraints. A feasible direc-
tion method is used to solve the nonlinear programming. We
alsogivetheanalysisoftheconvergenceandthecomplexityof
the method. e randomized algorithm is used to obtain the
suboptimal solution of the binary quadratic programming.
At last, we compare our method with the randomized
algorithm based on the interior point method and the feasible
direction method on max-cut problem. Simulation results
show that our method costs less CPU time than the two
methods.
2. The SDP Relaxation Method for
Binary Quadratic Programming
In this section, we introduce the SDP relaxation of binary
quadratic programming problem [1].
In problem (1), let =+1, =[
𝑇
,
𝑛
]
𝑇
(
𝑛
=1),and
1
=
𝑄𝑟
𝑟
𝑇
0
;thenproblem(1)canbeformulatedas
min
𝑇
1
s.t.∈
{
−1,1
}
𝑛
.
(2)
It is well known that problem (2)isalso-hard [1].
Let
2
=
1
−(
max
(
1
)+1)
𝑛
,where
max
(
1
)denotes
the largest eigenvalue of the matrix
1
and
𝑛
denotes the unit
matrix; then
2
is a negative denite matrix. Problem (2)is
equivalent to the following problem below:
min
𝑇
2
+
max
1
+1
s.t.∈
{
−1,1
}
𝑛
.
(3)
Letting =
𝑇
and ignoring the constant term, then
problem (3) is equivalent to the following problem:
min
2
s.t.
𝑖𝑖
=1, =1,2,...,
rank
(
)
=1
0,
(4)
where
2
=tr(
𝑇
2
) and
𝑖𝑖
is the diagonal elements
of matrix . In addition, 0denotes that matrix is
semidenite. Ignoring the nonconvex “rank one” constraint,
the SDP relaxation is given as follows [1]:
min
2
s.t.
𝑖𝑖
=1, =1,2,...,
0.
(5)
Interior point methods have been proved to be quite
ecient for small and moderate scale SDP. In the 0.878
randomized method by Goemans and Williamson [8], the
author solved the SDP relaxation problem (5)byinterior
point methods. However, interior point methods are second-
order method, so they are quite time and memory intensive
andnotadaptedtolargescalebinaryquadraticproblems.e
complexity of the primal-dual interior point method based on
AHO search direction for the SDP relaxation (5)ofthemax-
cut is (
4.5
ln(1/))[14, 15].
3. The Rank-Two SDP Relaxation for
Binary Quadratic Programming
In [11], the rank-two SDP relaxation model is proposed for
max-cut problem based on the polar direction. In [12], the
rank-two SDP relaxation model is proposed for maximal
independent set problem. In this section, we present a rank-
two SDP relaxation based on the rank-two approximate
matrix for binary quadratic programming.
In SDP relaxation problem (5), let =−
2
;wehave
max
s.t.
𝑖𝑖
=1, =1,2,...,
0.
(6)
Obviously, matrix is positive denite.
Let =
𝑇
+
𝑇
, , ∈
𝑛
[12]; then =
𝑇
+
𝑇
and
𝑖𝑖
=
2
𝑖
+
2
𝑖
=1.Weobtainthe
rank-two SDP relaxation of binary quadratic programming
as follows:
max
𝑇
+
𝑇
s.t.
2
𝑖
+
2
𝑖
=1, =1,2,...,;
(7)
where
𝑖
and
𝑖
are the elements of vector and .Obviously,
matrix satises that rank() = 2, 0,soproblem(7)is
arank-twoSDPrelaxationproblem.
Problem (7) is also a nonlinear programming with
quadratic objective function and constraints. Compared to
the
2
variables of SDP relaxation, the rank-two relaxation has
only 2 variables, so this approach possesses scalability for
solving large-scale binary quadratic programming problems.
Let
,=
𝑇
+
𝑇
,
𝑖
, =
2
𝑖
+
2
𝑖
−1,
=1,2,...,,
(8)