 Research
 Open access
 Published:
Kernel function based interiorpoint methods for horizontal linear complementarity problems
Journal of Inequalities and Applications volume 2013, Article number: 215 (2013)
Abstract
It is well known that each kernel function defines an interiorpoint algorithm. In this paper we propose new classes of kernel functions whose form is different from known kernel functions and define interiorpoint methods (IPMs) based on these functions whose barrier term is exponential power of exponential functions for {P}_{\ast}(\kappa )horizontal linear complementarity problems (HLCPs). New search directions and proximity measures are defined by these kernel functions. We obtain so far the best known complexity results for large and smallupdate methods.
1 Introduction
In this paper we consider {P}_{\ast}(\kappa )horizontal linear complementarity problem (HLCP) as follows.
Given \{M,N\}, a {P}_{\ast}(\kappa )pair, M,N\in {\mathbf{R}}^{n\times n}, q\in {\mathbf{R}}^{n}, and \kappa \ge 0, find a pair (x;s)\in {\mathbf{R}}^{2n} such that
Note that \{M,N\} is called a {P}_{\ast}(\kappa )pair if Mx+Ns=0 implies that
where {I}_{+}(x):=\{i\in I:{x}_{i}{s}_{i}\ge 0\}, {I}_{}(x):=\{i\in I:{x}_{i}{s}_{i}<0\}, and I:=\{1,2,\dots ,n\}.
{P}_{\ast}(\kappa )HLCPs have many applications in economic equilibrium problems, noncooperative games, traffic assignment problems, and optimization problems [1, 2]. {P}_{\ast}(\kappa )HLCP (1) includes the standard linear complementarity problem (LCP), linear, and quadratic optimization problems. Indeed, when N is nonsingular, then {P}_{\ast}(\kappa )HLCP reduces to {P}_{\ast}(\kappa )LCP. Furthermore, when \kappa =0, {P}_{\ast}(0)HLCP is monotone LCP.
Recently, Bai et al. [3] defined the concept of eligible kernel functions which require four conditions and proposed primaldual IPMs for linear optimization (LO) problems based on these functions, and some of these methods achieved the best known complexity results for both large and smallupdate methods. Cho [4] and Cho et al. [5] extended these algorithms for LO to {P}_{\ast}(\kappa )linear complementarity problems (LCPs) and obtained the similar complexity results as LO problems for largeupdate methods. Amini et al. [6, 7] introduced new IPMs based on parametric versions of kernel functions in [3] and obtained the better iteration bounds than the bound of the algorithm in [3] with numerical tests. Wang et al. [2] generalized polynomial IPMs for LO problem to {P}_{\ast}(\kappa )HLCP based on a finite kernel function, which was first defined in [8], and obtained the same iteration bounds for large and smallupdate methods as an LO problem. Ghami et al. [9] extended IPMs for LO problems to the {P}_{\ast}(\kappa )LCPs based on eligible kernel functions, which were defined in [3], and proposed large as well as smallupdate methods. Lesaja et al. [10] also proposed IPMs for {P}_{\ast}(\kappa )LCPs based on ten kernel functions which were defined for LO problems. Ghami et al. [11] proposed IPM for an LO problem based on a kernel function whose barrier term is a trigonometric function. However, this method does not have the best known iteration bound for a largeupdate method. Cho et al. [12] defined a new kernel function, whose barrier term is the exponential power of the exponential function for LO problems, and obtained the best known iteration bounds for large and smallupdate methods.
Motivated by these works, we introduce new classes of eligible kernel functions, which are different from known kernel functions in [3, 6, 7] and have the exponential power of exponential barrier term, and propose a complexity analysis of the IPMs for {P}_{\ast}(\kappa )HLCP based on these kernel functions. We show that these algorithms have \mathcal{O}((1+2\kappa )\sqrt{n}lognlog\frac{n{\mu}^{0}}{\u03f5}) and \mathcal{O}((1+2\kappa )\sqrt{n}log\frac{n{\mu}^{0}}{\u03f5}) iteration bounds for large and smallupdate methods, respectively, which are currently the best known iteration bounds for such methods.
The paper is organized as follows. In Section 2 we propose some basic concepts and a generic interior point algorithm for {P}_{\ast}(\kappa )HLCP. In Section 3 we introduce new classes of eligible kernel functions and their technical properties. Finally, we derive the framework for analyzing the iteration bounds and the complexity results of the algorithms based on these kernel functions in Section 4.
Notational conventions: {\mathbf{R}}_{+}^{n} and {\mathbf{R}}_{++}^{n} denote the sets of ndimensional nonnegative vectors and positive vectors, respectively. For x,s\in {\mathbf{R}}^{n}, {x}_{min}, xs, and (x;s) denote the smallest component of the vector x, the componentwise product of the vectors x and s, and the column vector {({x}^{T},{s}^{T})}^{T}, respectively. We denote by D the diagonal matrix from a vector d, i.e., D=diag(d). e denotes the ndimensional vector of ones. For f(x), g(x):{\mathbf{R}}_{++}\to {\mathbf{R}}_{++}, f(x)=\mathcal{O}(g(x)) if f(x)\le {c}_{1}g(x) for some positive constant {c}_{1} and f(x)=\mathrm{\Theta}(g(x)) if {c}_{2}g(x)\le f(x)\le {c}_{3}g(x) for some positive constants {c}_{2} and {c}_{3}.
2 Preliminaries
In this section we recall some basic definitions and introduce a generic interior point algorithm for {P}_{\ast}(\kappa )HLCP.
Definition 2.1 [13]
Let M\in {\mathbf{R}}^{n\times n}, x\in {\mathbf{R}}^{n}, and \kappa \ge 0.

(i)
M is called a positive semidefinite matrix if {x}^{T}(Mx)\ge 0.

(ii)
M is called a {P}_{0}matrix if there exists an index i\in I such that {x}_{i}\ne 0 and {x}_{i}{[Mx]}_{i}\ge 0.

(iii)
M is called a {P}_{\ast}(\kappa )matrix if
(1+4\kappa )\sum _{i\in {I}_{+}(x)}{x}_{i}{[Mx]}_{i}+\sum _{i\in {I}_{}(x)}{x}_{i}{[Mx]}_{i}\ge 0,
where {[Mx]}_{i} denotes the i th component of the vector Mx, {I}_{+}(x)=\{i\in I:{x}_{i}{[Mx]}_{i}\ge 0\}, and {I}_{}(x)=\{i\in I:{x}_{i}{[Mx]}_{i}<0\}.
Definition 2.2 [14]
Let M,N\in {\mathbf{R}}^{n\times n}, x,s\in {\mathbf{R}}^{n}, and \kappa \ge 0.

(i)
\{M,N\} is called a monotone pair if Mx+Ns=0 implies {x}^{T}s\ge 0.

(ii)
\{M,N\} is called a {P}_{0}pair if Mx+Ns=0 and (x;s)\ne 0 implies that there exists an index i\in I such that {x}_{i}\ne 0 or {s}_{i}\ne 0, and {x}_{i}{s}_{i}\ge 0.

(iii)
\{M,N\} is called a {P}_{\ast}(\kappa )pair if Mx+Ns=0 implies that {x}^{T}s\ge 4\kappa {\sum}_{i\in {I}_{+}}{x}_{i}{s}_{i}, where {I}_{+}(x)=\{i\in I:{x}_{i}{s}_{i}\ge 0\}.
Lemma 2.3 If \{M,N\} is a {P}_{0}pair, then
is a nonsingular matrix for any positive diagonal matrices X,S\in {\mathbf{R}}^{n\times n}.
Proof Assume that the matrix {M}^{\prime} is singular. Then {M}^{\prime}\zeta =0 for some nonzero \zeta =(\xi ;\eta )\in {R}^{2n}, i.e., M\xi +N\eta =0 and {s}_{i}{\xi}_{i}+{x}_{i}{\eta}_{i}=0, i\in I. Hence (\xi ;\eta )\ne 0, and we have an index i\in I such that {\xi}_{i}\ne 0 or {\eta}_{i}\ne 0, and {\xi}_{i}{\eta}_{i}\ge 0, since \{M,N\} is a {P}_{0}pair. On the other hand, {\xi}_{i}{\eta}_{i}={x}_{i}{({\eta}_{i})}^{2}/{s}_{i}<0. This is a contradiction. This completes the proof. □
Since the class of {P}_{0}pairs includes the class of {P}_{\ast}(\kappa )pairs, we obtain the following corollary.
Corollary 2.4 Let \{M,N\} be a {P}_{\ast}(\kappa )pair and x,s\in {\mathbf{R}}_{++}^{n}. Then all c\in {\mathbf{R}}^{n} the system
has a unique solution (\mathrm{\Delta}x;\mathrm{\Delta}s).
The basic idea of generic IPMs is to replace the second equation of (1) by the parameterized equation xs=\mu \mathbf{e} with \mu >0, i.e., we consider the following system:
Without loss of generality, we assume that (1) satisfies the interiorpoint condition (IPC), i.e., there exists ({x}^{0};{s}^{0})>0 such that M{x}^{0}+N{s}^{0}=q [15]. Since \{M,N\} is a {P}_{\ast}(\kappa )pair and (1) satisfies IPC, the system (2) has a unique solution (x(\mu );s(\mu )) for each \mu >0, which is called the μcenter. The set of μcenters is called the central path of (1). The limit of the central path exists, and since the limit point satisfies (1), it naturally yields the solution for (1) [16]. IPMs follow this central path approximately and approach the solution of (1) as \mu \to 0.
For given (x;s):=({x}^{0};{s}^{0}), by applying Newton’s method to the system (2), we have the Newtonsystem as follows:
By taking a step along the search direction (\mathrm{\Delta}x;\mathrm{\Delta}s), we define a new iteration ({x}_{+};{s}_{+}), where for some \alpha \ge 0,
To have the motivation of a new algorithm, we define the following scaled vectors:
Using (5), we can rewrite the Newtonsystem (3) as follows:
where \overline{M}:=DMD, \overline{N}:=DND, and D:=diag(d). Note that the righthand side of the second equation of (6) equals the negative gradient of the logarithmic barrier function {\mathrm{\Psi}}_{l}(v):={\sum}_{i=1}^{n}{\psi}_{l}({v}_{i}) and {\psi}_{l}(t)=\frac{{t}^{2}1}{2}logt, i.e.,
The interiorpoint algorithm works as follows. Assume that we are given a strictly feasible point (x;s) which is in a τneighborhood of the given μcenter. Then we update μ to {\mu}_{+}=(1\theta )\mu for some fixed \theta \in (0,1) and solve the system (3) to obtain the search direction. The positivity condition of a new iteration is ensured with the right choice of the step size α. This procedure is repeated until we find a new iteration ({x}_{+};{s}_{+}) that is in a τneighborhood of the {\mu}_{+}center and then we let \mu :={\mu}_{+} and (x;s):=({x}_{+};{s}_{+}). We repeat the process until n\mu <\epsilon (see Algorithm 1).
If \tau =\mathcal{O}(n) and \theta =\mathrm{\Theta}(1), then the algorithm is called a largeupdate method. When \tau =\mathcal{O}(1) and \theta =\mathrm{\Theta}(\frac{1}{\sqrt{n}}), we call the algorithm a smallupdate method.
3 New kernel function
In this section we define new classes of kernel functions and give their essential properties.
\psi :{\mathbf{R}}_{++}\to {\mathbf{R}}_{+} is called a kernel function if ψ is twice differentiable and satisfies the following conditions:
We define new classes of kernel functions {\psi}_{j}(t), j\in \{1,2\}, in Table 1 and give the first three derivatives of {\psi}_{j}(t), j\in \{1,2\}, in Table 2 and Table 3.
In the following lemma, we show that \psi (t):={\psi}_{j}(t), j\in \{1,2\}, are eligible [3].
Lemma 3.1 Let \psi (t):={\psi}_{j}(t), j\in \{1,2\}, be defined as in Table 1. Then {\psi}_{j}, j\in \{1,2\}, satisfy the following eligible conditions:

(a)
t{\psi}^{\u2033}(t)+{\psi}^{\prime}(t)>0, t>0, i.e., ψ is exponential convex,

(b)
t{\psi}^{\u2033}(t){\psi}^{\prime}(t)>0, t>0,

(c)
{\psi}_{j}^{(3)}(t)<0, t>0,

(d)
2{({\psi}^{\u2033}(t))}^{2}{\psi}^{\prime}(t){\psi}_{j}^{(3)}(t)>0, t>0.
Proof From Table 4, Table 3, and Table 5, we show that {\psi}_{j}(t), j\in \{1,2\}, satisfy eligible conditions (a)(d). □
Remark 3.2 For {\psi}_{j}(t), j\in \{1,2\}, let {\psi}_{b1}(t)={\psi}_{1}(t)\frac{e({t}^{2}1)}{2}, {\psi}_{b2}(t)={\psi}_{2}(t)\frac{{t}^{2}1}{2}.
From Table 2,
Since {\psi}_{bj}^{\prime}(t)<0, j\in \{1,2\}, from Table 2, {\psi}_{bj}(t), j\in \{1,2\}, are monotonically decreasing with respect to t>0.
Let {\rho}_{j}:[0,\mathrm{\infty})\to (0,1] and {\varrho}_{j}:[0,\mathrm{\infty})\to [1,\mathrm{\infty}) denote the inverse functions of the restriction of \frac{1}{2}{\psi}_{j}^{\prime}(t) for 0<t\le 1 and {\psi}_{j}(t) for t\ge 1, respectively, j\in \{1,2\}. Then
and
Lemma 3.3 Let {\rho}_{j}(z), j\in \{1,2\}, be defined as in (10). Then we have, for p\ge 1, r\ge 1,

(i)
{\rho}_{1}(z)\ge {(log(e+{p}^{1}log(e+2z)))}^{\frac{1}{r}}, z\ge 0,

(ii)
{\rho}_{2}(z)\ge {(1+log(1+{p}^{1}log(1+2z)))}^{\frac{1}{r}}, z\ge 0.
Proof For (i), using (10) and Table 2, we have the equation
Since 0<t\le 1,
By taking the natural logarithm on both sides of (12), we have {e}^{{t}^{r}}\le e+{p}^{1}log(e+2z). Hence we have
By the same way as (i), we obtain the result (ii). This completes the proof. □
Lemma 3.4 Let {\psi}_{j}(t), j\in \{1,2\}, be defined as in Table 1. Then we have

(i)
\frac{e}{2}{(t1)}^{2}\le {\psi}_{1}(t)\le \frac{1}{2e}{({\psi}_{1}^{\prime}(t))}^{2}, t>0,

(ii)
\frac{1}{2}{(t1)}^{2}\le {\psi}_{2}(t)\le \frac{1}{2}{({\psi}_{2}^{\prime}(t))}^{2}, t>0.
Proof For (i), using the first condition of (8) and (9), we have
which proves the first inequality. The second inequality is obtained as follows:
For (ii), by the same way as above, we obtain the result. This completes the proof. □
Lemma 3.5 Let {\varrho}_{j}(u), j\in \{1,2\}, be defined as in (11). Then we have

(i)
{\varrho}_{1}(u)\le 1+\sqrt{\frac{2u}{e}}, u\ge 0,

(ii)
{\varrho}_{2}(u)\le 1+\sqrt{2u}, u\ge 0.
Proof For (i), using the first inequality in Lemma 3.4, we have u={\psi}_{1}(t)\ge \frac{e}{2}{(t1)}^{2}. Then we have
Similarly, we obtain the result (ii). This completes the proof. □
In this paper we replace the logarithmic barrier function {\mathrm{\Psi}}_{l}(v) in (7) by a strictly convex function \mathrm{\Psi}(v) as follows:
where
and {\psi}_{j}(t), j\in \{1,2\}, are defined in Table 1. Since \mathrm{\Psi}(v) is strictly convex and minimal at v=\mathbf{e}, we have
Using (5) and (13), we modify the Newtonsystem (3) as follows:
By Corollary 2.4, the system (15) has a unique solution (\mathrm{\Delta}x;\mathrm{\Delta}s) which is the modified Newton search direction. Consequently, we use \mathrm{\Psi}(v) as the proximity function to find a search direction and to measure the proximity between the current iteration and the μcenter. We also define the normbased proximity measure {\delta}_{j}(v), j\in \{1,2\}, as follows:
The following lemma gives a relation between two proximity measures.
Lemma 3.6 Let {\delta}_{j}(v) and {\mathrm{\Psi}}_{j}(v), j\in \{1,2\}, be defined as in (16) and (14), respectively. Then we have

(i)
{\delta}_{1}(v)\ge \sqrt{\frac{e{\mathrm{\Psi}}_{1}(v)}{2}},

(ii)
{\delta}_{2}(v)\ge \sqrt{\frac{{\mathrm{\Psi}}_{2}(v)}{2}}.
Proof For (i), using (16) and the second inequality in Lemma 3.4, we have
Hence we have {\delta}_{1}(v)\ge \sqrt{\frac{e{\mathrm{\Psi}}_{1}(v)}{2}}.
For (ii), by the same way as above, we obtain the result. This completes the proof. □
Using the eligible conditions (b) and (c) in Lemma 3.1, we obtain the following lemma.
Lemma 3.7 (Theorem 3.2 in [3])
Let {\varrho}_{j}, j\in \{1,2\}, be defined as in (11). Then we have
In the following lemma, we give upper bounds of {\mathrm{\Psi}}_{j}(v), j\in \{1,2\}, after a μupdate.
Lemma 3.8 Let {\mathrm{\Psi}}_{j}(v), j\in \{1,2\}, be defined as in (14), 0<\theta <1, and {v}_{+}=\frac{v}{\sqrt{1\theta}}. If {\mathrm{\Psi}}_{j}(v)\le \tau, j\in \{1,2\}, then we have

(i)
{\mathrm{\Psi}}_{1}({v}_{+})\le \frac{en\theta +2\tau +2\sqrt{2en\tau}}{2(1\theta )} or {\mathrm{\Psi}}_{1}({v}_{+})\le \frac{{\psi}_{1}^{\u2033}(1){(\sqrt{\frac{2\tau}{e}}+\theta \sqrt{n})}^{2}}{2(1\theta )},

(ii)
{\mathrm{\Psi}}_{2}({v}_{+})\le \frac{n\theta +2\tau +2\sqrt{2n\tau}}{2(1\theta )} or {\mathrm{\Psi}}_{2}({v}_{+})\le \frac{{\psi}_{2}^{\u2033}(1){(\sqrt{2\tau}+\theta \sqrt{n})}^{2}}{2(1\theta )}.
Proof For the first inequality of (i), using Remark 3.2 with {\psi}_{b1}(1)=0 and {\psi}_{b1}^{\prime}(t)<0, we get
Using Lemma 3.7, (17), and Lemma 3.5(i), we have
For the second inequality of (i), using Taylor’s theorem, {\psi}_{1}(1)={\psi}_{1}^{\mathrm{\prime}}(1)=0 and {\psi}_{1}^{(3)}(t)<0, we have
for some ξ, 1\le \xi \le t. Since \frac{1}{\sqrt{1\theta}}\ge 1 and {\varrho}_{1}(\frac{\tau}{n})\ge 1, we have \frac{{\varrho}_{1}(\frac{\tau}{n})}{\sqrt{1\theta}}\ge 1. Using Lemma 3.7, (18), and Lemma 3.5(i), we have
where the last inequality holds from 1\sqrt{1\theta}=\frac{\theta}{1+\sqrt{1\theta}}\le \theta, 0<\theta <1.
By the same way as the proof of (i), we obtain the result (ii). This completes the proof. □
Define
and
We will use {\overline{\mathrm{\Psi}}}_{j,0} and {\tilde{\mathrm{\Psi}}}_{j,0} for the upper bounds of {\mathrm{\Psi}}_{j}(v) from (14) for large and smallupdate methods, respectively, j\in \{1,2\}.
Remark 3.9 For the largeupdate method with \tau =\mathcal{O}(n) and \theta =\mathrm{\Theta}(1), {\overline{\mathrm{\Psi}}}_{j,0}=\mathcal{O}(n), j\in \{1,2\}, and for the smallupdate method with \tau =\mathcal{O}(1) and \theta =\mathrm{\Theta}(\frac{1}{\sqrt{n}}), {\tilde{\mathrm{\Psi}}}_{j,0}=\mathcal{O}({\psi}_{j}^{\u2033}(1)), j\in \{1,2\}.
For fixed μ, if we take a step size α, using (4) and (5), we have new iterations
and
For fixed \mu >0,
For notational convenience, let \mathrm{\Psi}(v):={\mathrm{\Psi}}_{j}(v) and \psi (t):={\psi}_{j}(t), j\in \{1,2\}.
For \alpha >0, we define
where f(\alpha ) is the difference of proximities between a new iteration and a current iteration for fixed μ. By the condition (a) in Lemma 3.1, we have
Hence we have f(\alpha )\le {f}_{1}(\alpha ), where
Then, we have f(0)={f}_{1}(0)=0. Differentiating {f}_{1}(\alpha ) with respect to α, we have
where {[{d}_{x}]}_{i} and {[{d}_{s}]}_{i} denote the i th components of the vectors {d}_{x} and {d}_{s}, respectively. Using (13) and (16), we have
By taking the derivative of {f}_{1}^{\prime}(\alpha ) with respect to α, we have
Since {f}_{1}^{\u2033}(\alpha )>0, {f}_{1}(\alpha ) is strictly convex in α unless {d}_{x}={d}_{s}=0. Since \{M,N\} is a {P}_{\ast}(\kappa )pair and M\mathrm{\Delta}x+N\mathrm{\Delta}s=0 from (15), for (\mathrm{\Delta}x;\mathrm{\Delta}s)\in {\mathbf{R}}^{2n},
where {I}_{+}=\{i\in I:{[\mathrm{\Delta}x]}_{i}{[\mathrm{\Delta}s]}_{i}\ge 0\}, {I}_{}=I{I}_{+}. Since {d}_{x}{d}_{s}=\frac{{v}^{2}\mathrm{\Delta}x\mathrm{\Delta}s}{xs}=\frac{\mathrm{\Delta}x\mathrm{\Delta}s}{\mu} and \mu >0, we have
For notational convenience, we denote \mathrm{\Psi}:={\mathrm{\Psi}}_{j}(v) and \delta :={\delta}_{j}(v), j\in \{1,2\}.
In the following lemmas, we state same technical properties in [5].
Lemma 3.10 (Lemma 4.4 in [5])
{f}_{1}^{\mathrm{\prime}}(\alpha )\le 0 if α satisfies
Lemma 3.11 (Lemma 4.5 in [5])
Let \rho :={\rho}_{j}(\delta ), j\in \{1,2\}, be defined as in (10). Then, in the worst case, the largest step size α satisfying (21) is given by
Lemma 3.12 (Lemma 4.6 in [5])
Let ρ and \overline{\alpha} be defined as in Lemma 3.11. Then
Define
Then we have \tilde{\alpha}\le \overline{\alpha}.
Lemma 3.13 Let \tilde{\alpha} be defined as in (22). Then for \kappa \ge 0, we have
Lemma 3.14 (Lemma 4.10 in [5])
The righthand sides in Lemma 3.13 are monotonically decreasing with respect to δ.
Lemma 3.15 (Proposition 1.3.2 in [17])
Let {t}_{0},{t}_{1},\dots ,{t}_{K} be a sequence of positive numbers such that
where \lambda >0 and 0<\gamma \le 1. Then K\le \lceil \frac{{t}_{0}^{\gamma}}{\lambda \gamma}\rceil.
We define the value of \mathrm{\Psi}(v) after the μupdate as {\mathrm{\Psi}}_{0}, and the subsequent values in the same outer iteration are denoted as {\mathrm{\Psi}}_{k}, k=1,2,\dots . Then we have
Theorem 3.16 Let a {P}_{\ast}(\kappa )HLCP be given. If \tau \ge 1, then the upper bound of a total number of iterations is given by
Proof From Lemma 3.15 and Lemma II. 17 in [18], the number of inner and outer iterations is given by \lceil \frac{{\mathrm{\Psi}}_{0}^{\gamma}}{\lambda \gamma}\rceil and \lceil \frac{1}{\theta}log\frac{n{\mu}^{0}}{\u03f5}\rceil, respectively. For the total number of iterations, we multiply the number of inner iterations by that of outer iterations. Hence we have the desired results. This completes the proof. □
4 Application to new kernel functions
For the complexity analysis, we follow a similar framework in [3] for LO problems.
We apply the framework in Table 6 to the specific kernel function
Step 1: Using Lemma 3.3, {\rho}_{1}(z)\ge {(log(e+{p}^{1}log(e+2z)))}^{\frac{1}{r}}, z\ge 0.
Step 2: By Lemma 3.5, the inverse function of {\psi}_{1}(t) for t\ge 1 satisfies
Step 3: Using Lemma 3.6, we obtain
Step 4: Using (19) and {\psi}_{1}^{\u2033}(1)=e(pre+2r+2) from Table 2, we have the following:

(i)
For the largeupdate method, {\mathrm{\Psi}}_{0}\le \frac{en\theta +2\tau +2\sqrt{2en\tau}}{2(1\theta )}:={\overline{\mathrm{\Psi}}}_{1,0}.

(ii)
For the smallupdate method, {\mathrm{\Psi}}_{0}\le \frac{e(pre+2r+2){(\sqrt{\frac{2\tau}{e}}+\theta \sqrt{n})}^{2}}{2(1\theta )}:={\tilde{\mathrm{\Psi}}}_{1,0}.
Step 5: Define {L}_{1}({\mathrm{\Psi}}_{1},p):=e+{p}^{1}log(e+2\sqrt{2e{\mathrm{\Psi}}_{1}}). Using {\psi}_{1}^{(3)}(t)<0, Step 1, 1+\frac{1}{\sqrt{1+2\kappa}}\le 2, and Table 2, we have
where the last inequality follows from the assumption {\mathrm{\Psi}}_{1}\ge \tau \ge 1. Using Lemma 3.13, Lemma 3.14, Lemma 3.6, and (23), we have
where the last inequality follows from {L}_{1}({\mathrm{\Psi}}_{1,0},p):=e+{p}^{1}log(e+2\sqrt{2e{\mathrm{\Psi}}_{1,0}}) and the assumption {\mathrm{\Psi}}_{1,0}\ge {\mathrm{\Psi}}_{1}.
Step 6: Using Theorem 3.16, Step 4 with {\mathrm{\Psi}}_{1,0}\le {\overline{\mathrm{\Psi}}}_{1,0}, and {\mathrm{\Psi}}_{1,0}\le {\tilde{\mathrm{\Psi}}}_{1,0}, and Step 5 with \gamma =\frac{1}{2} and \frac{1}{\lambda}=8(1+2\kappa ){L}_{1}({\mathrm{\Psi}}_{1,0},p){(log{L}_{1}({\mathrm{\Psi}}_{1,0},p))}^{\frac{2(r+1)}{r}}(pr{L}_{1}({\mathrm{\Psi}}_{1,0},p)+2r+1), we have the upper bounds of the total number of iterations for large and smallupdate methods as follows.

(i)
For largeupdate methods,
\lceil 8(1+2\kappa ){L}_{1}({\overline{\mathrm{\Psi}}}_{1,0},p){(log{L}_{1}({\overline{\mathrm{\Psi}}}_{1,0},p))}^{\frac{2(r+1)}{r}}(pr{L}_{1}({\overline{\mathrm{\Psi}}}_{1,0},p)+2r+1){\overline{\mathrm{\Psi}}}_{1,0}^{\frac{1}{2}}\frac{1}{\theta}log\frac{n{\mu}^{0}}{\u03f5}\rceil ,where {L}_{1}({\overline{\mathrm{\Psi}}}_{1,0},p):=e+{p}^{1}log(e+2\sqrt{2e{\overline{\mathrm{\Psi}}}_{1,0}}).

(ii)
For smallupdate methods,
\lceil 8(1+2\kappa ){L}_{1}({\tilde{\mathrm{\Psi}}}_{1,0},p){(log{L}_{1}({\tilde{\mathrm{\Psi}}}_{1,0},p))}^{\frac{2(r+1)}{r}}(pr{L}_{1}({\tilde{\mathrm{\Psi}}}_{1,0},p)+2r+1){\tilde{\mathrm{\Psi}}}_{1,0}^{\frac{1}{2}}\frac{1}{\theta}log\frac{n{\mu}^{0}}{\u03f5}\rceil ,
where {L}_{1}({\tilde{\mathrm{\Psi}}}_{1,0},p):=e+{p}^{1}log(e+2\sqrt{2e{\tilde{\mathrm{\Psi}}}_{1,0}}).
Step 7: Using Step 6 and Remark 3.9, for the largeupdate method with p=log(e+2\sqrt{2e{\overline{\mathrm{\Psi}}}_{1,0}})=\mathcal{O}(logn) and r=1, the algorithm has \mathcal{O}((1+2\kappa )\sqrt{n}lognlog\frac{n{\mu}^{0}}{\u03f5}) complexity. For the smallupdate method with p=1 and r=1, the algorithm has \mathcal{O}((1+2\kappa )\sqrt{n}log\frac{n{\mu}^{0}}{\u03f5}) complexity. These are currently the best known complexity results.
Remark 4.1 For the kernel function {\psi}_{2}(t) in Table 1, by applying the framework, the algorithms have \lceil 8(1+2\kappa ){L}_{2}({\overline{\mathrm{\Psi}}}_{2,0},p){(log{L}_{2}({\overline{\mathrm{\Psi}}}_{2,0},p))}^{\frac{2(r+1)}{r}}(pr{L}_{2}({\overline{\mathrm{\Psi}}}_{2,0},p)+2r+1){\overline{\mathrm{\Psi}}}_{2,0}^{\frac{1}{2}}\frac{1}{\theta}log\frac{n{\mu}^{0}}{\u03f5}\rceil and \lceil 8(1+2\kappa ){L}_{2}({\tilde{\mathrm{\Psi}}}_{2,0},p){(log{L}_{2}({\tilde{\mathrm{\Psi}}}_{2,0},p))}^{\frac{2(r+1)}{r}}(pr{L}_{2}({\tilde{\mathrm{\Psi}}}_{2,0},p)+2r+1){\tilde{\mathrm{\Psi}}}_{2,0}^{\frac{1}{2}}\frac{1}{\theta}log\frac{n{\mu}^{0}}{\u03f5}\rceil iteration bounds for large and smallupdate methods, respectively, where {L}_{2}({\overline{\mathrm{\Psi}}}_{2,0},p):=1+{p}^{1}log(1+2\sqrt{2{\overline{\mathrm{\Psi}}}_{2,0}}) and {L}_{2}({\tilde{\mathrm{\Psi}}}_{2,0},p):=1+{p}^{1}log(1+2\sqrt{2{\tilde{\mathrm{\Psi}}}_{1,0}}). By taking p=log(1+2\sqrt{2{\overline{\mathrm{\Psi}}}_{2,0}})=\mathcal{O}(logn) and r=1, the algorithm has \mathcal{O}((1+2\kappa )\sqrt{n}lognlog\frac{n{\mu}^{0}}{\u03f5}) complexity for largeupdate methods. Choosing p=1 and r=1, the algorithm has \mathcal{O}((1+2\kappa )\sqrt{n}log\frac{n{\mu}^{0}}{\u03f5}) for smallupdate methods. In conclusion, we obtain so far the best known iteration bounds of large and smallupdate methods for kernel functions {\psi}_{j}, j\in \{1,2\}, in Table 1.
References
Cottle RW, Pang JS, Stone RE: The Linear Complementarity Problem. Academic Press, San Diego; 1992.
Wang GQ, Bai YQ:Polynomial interiorpoint algorithm for {P}_{\ast}(\kappa ) horizontal linear complementarity problem. J. Comput. Appl. Math. 2009, 233: 248–263. 10.1016/j.cam.2009.07.014
Bai YQ, Ghami ME, Roos C: A comparative study of kernel functions for primaldual interiorpoint algorithms in linear optimization. SIAM J. Optim. 2004, 15: 101–128. 10.1137/S1052623403423114
Cho GM:A new largeupdate interior point algorithm for {P}_{\ast}(\kappa ) linear complementarity problems. J. Comput. Appl. Math. 2008, 216: 256–278.
Cho GM, Kim MK:A new largeupdate interior point algorithm for {P}_{\ast}(\kappa ) LCPs based on kernel functions. Appl. Math. Comput. 2006, 182: 1169–1183. 10.1016/j.amc.2006.04.060
Amini K, Haseli A: A new proximity function generating the best known iteration bounds for both largeupdate and smallupdate interiorpoint methods. ANZIAM J. 2007, 49: 259–270. 10.1017/S1446181100012827
Amini K, Peyghami MR:Exploring complexity of large update interiorpoint methods for {P}_{\ast}(\kappa ) linear complementarity problem based on kernel function. Appl. Math. Comput. 2009, 207: 501–513. 10.1016/j.amc.2008.11.002
Bai YQ, Ghami ME, Roos C: A new efficient largeupdate primaldual interiorpoint method based on a finite barrier. SIAM J. Optim. 2003, 13: 766–782.
Ghami ME, Steihaug T:Kernelfunction based primaldual algorithms for {P}_{\ast}(\kappa ) linear complementarity problems. RAIRO. Rech. Opér. 2010, 44: 185–205.
Lesaja G, Roos C:Unified analysis of kernelbased interiorpoint methods for {P}_{\ast}(\kappa )linear complementarity problems. SIAM J. Optim. 2010, 20: 3014–3039. 10.1137/090766735
Ghami ME, Guennoun ZA, Bouali S, Steihaug T: Interiorpoint methods for linear optimization based on a kernel function with a trigonometric barrier term. J. Comput. Appl. Math. 2012, 236: 3613–3623. 10.1016/j.cam.2011.05.036
Cho GM, Cho YY, Lee YH: A primaldual interiorpoint algorithm based on a new kernel function. ANZIAM J. 2010, 51: 476–491. 10.1017/S1446181110000908
Kojima M, Megiddo N, Noma T, Yoshise A Lecture Notes in Computer Science 538. In A Unified Approach to Interior Point Algorithms for Linear Complementarity Problems. Springer, Berlin; 1991.
Tütüncü RH, Todd MJ: Reducing horizontal linear complementarity problems. Linear Algebra Appl. 1995, 223/224: 717–729.
Jansen B, Roos K, Terlaky T, Yoshise A: Polynomiality of primaldual affine scaling algorithms for nonlinear complementarity problems. Math. Program. 1997, 78: 315–345.
Xiu N, Zhang J: A smoothing GaussNewton method for the generalized HLCP. J. Comput. Appl. Math. 2001, 129: 195–208. 10.1016/S03770427(00)005501
Peng J, Roos C, Terlaky T: SelfRegularity: A New Paradigm for PrimalDual InteriorPoint Algorithms. Princeton University Press, Princeton; 2002.
Roos C, Terlaky T, Vial JP: Theory and Algorithms for Linear Optimization, an Interior Approach. Wiley, Chichester; 1997.
Acknowledgements
This research of the first author was supported by the Basic Science Research Program through NRF funded by the Ministry of Education, Science, and Technology (No. 2012005767) and by the Research Fund Program of Research Institute for Basic Science, Pusan National University, Korea, 2012, Project No. RIBSPNU2012102.
Author information
Authors and Affiliations
Corresponding author
Additional information
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
All authors have equally contributed in designing a new algorithm and obtaining complexity results. All authors read and approved the final manuscript.
Authors’ original submitted files for images
Below are the links to the authors’ original submitted files for images.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 2.0 International License ( https://creativecommons.org/licenses/by/2.0 ), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
About this article
Cite this article
Lee, YH., Cho, YY. & Cho, GM. Kernel function based interiorpoint methods for horizontal linear complementarity problems. J Inequal Appl 2013, 215 (2013). https://doi.org/10.1186/1029242X2013215
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/1029242X2013215