Abstract
The weighted complementarity problem (denoted by WCP) significantly extends the general complementarity problem and can be used for modeling a larger class of problems from science and engineering. In this paper, by introducing a one-parametric class of smoothing functions which includes the weight vector, we propose a smoothing Newton algorithm with nonmonotone line search to solve WCP. We show that any accumulation point of the iterates generated by this algorithm, if exists, is a solution of the considered WCP. Moreover, when the solution set of WCP is nonempty, under assumptions weaker than the Jacobian nonsingularity assumption, we prove that the iteration sequence generated by our algorithm is bounded and converges to one solution of WCP with local superlinear or quadratic convergence rate. Promising numerical results are also reported.
Similar content being viewed by others
References
Asadi, S., Darvay, Z., Lesaja, G., Mahdavi-Amiri, N., Potra, F.A.: A full-Newton step interior-point method for monotone weighted linear complementarity problems. J. Optim. Theory Appl. (2020). https://doi.org/10.1007/s10957-020-01728-4
Anstreicher, K.M.: Interior-point algorithms for a generalization of linear programming and weighted centering. Optim. Methods Softw. 27, 605–612 (2012)
Burke, J., Xu, S.: A non-interior predictor–corrector path following algorithm for the monotone linear complementarity problem. Math. Program. 87, 113–130 (2000)
Chen, J.-S., Pan, S.H.: A survey on SOC complementarity functions and solution methods for SOCPs and SOCCPs. Pac. J. Optim. 8, 33–74 (2012)
Clarke, F.H.: Optimization and Nonsmooth Analysis. Wiley, New York (1983)
Faraut, J., Korányi, A.: Analysis on Symmetric Cones. Oxford Mathematical Monographs, Oxford University Press, New York (1994)
Fan, J.Y.: The modified Levenberg–Marquardt method for nonlinear equations with cubic convergence. Math. Comput. 81, 447–466 (2012)
Fan, J.Y.: Accelerating the modified Levenberg–Marquardt method for nonlinear equations. Math. Comput. 83, 1173–1187 (2014)
Fukushima, M., Luo, Z.Q., Tseng, P.: Smoothing functions for second-order cone complementarity problems. SIAM J. Optim. 12, 436–460 (2002)
Gowda, M.S., Sznajder, R., Tao, J.: Some P-properties for linear transformations on Euclidean Jordan algebras. Linear Algebra Appl. 393, 203–232 (2004)
Grippo, L., Lampariello, F., Lucidi, S.: A truncated Newton method with nonmonotone line search for unconstrained optimization. J. Optim. Theory Appl. 60, 401–419 (1989)
Hager, W.W., Zhang, H.: A nonmonotone line search technique and its application to unconstrained optimization. SIAM J. Optim. 14, 1043–1056 (2004)
Huang, Z.H., Hu, S.L., Han, J.Y.: Convergence of a smoothing algorithm for symmetric cone complementarity problems with a nonmonotone line search. Sci. China Ser. A: Math. 52, 833–848 (2009)
Huang, Z.H., Ni, T.: Smoothing algorithms for complementarity problems over symmetric cones. Comput. Optim. Appl. 45, 557–579 (2010)
Huang, Z.H., Liu, X.H.: Extension of smoothing Newton algorithms for solve linear programming over symmetric cones. J. Syst. Sci. Complex. 24, 195–206 (2011)
Jian, Z.: A smoothing Newton algorithm for weighted linear complementarity problem. Optim. Lett. 10, 499–509 (2016)
Kong, L., Sun, J., Xiu, N.: A regularized smoothing Newton method for symmetric cone complementarity problems. SIAM J. Optim. 19, 1028–1047 (2008)
Kong, L.: Quadratic convergence of a smoothing Newton method for symmetric cone programming without strict complementarity. Positivity 16, 297–319 (2012)
Li, Y.M., Wei, D.Y.: A generalized smoothing Newton method for the symmetric cone complementarity problem. Appl. Math. Comput. 264, 335–345 (2015)
Liu, X.H., Huang, Z.H.: A smoothing Newton algorithm based on a one-parametric class of smoothing functions for linear programming over symmetric cones. Math. Math. Oper. Res. 70, 385–404 (2009)
Liu, L.X., Liu, S.Y., Wu, Y.: A smoothing Newton method for symmetric cone complementarity problem. J. Appl. Math. Comput. 47, 175–191 (2015)
Lu, N., Huang, Z.H.: A smoothing Newton algorithm for a class of non-monotonic symmetric cone linear complementarity problems. J. Optim. Theory Appl. 161, 446–464 (2014)
Luo, Z.Q.: New error bounds and their applications to convergence analysis of iterative algorithms. Math. Program. Ser. B 88, 341–355 (2000)
Ma, C.F.: A regularized smoothing Newton method for solving the symmetric cone complementarity problem. Math. Comput. Model. 54, 2515–2527 (2011)
Mousavi, S., Gao, Z., Han, L., Lim, A.: Quadratic Surface Support Vector Machine with L1 Norm Regularization. arXiv preprint arXiv:1908.08616 (2019)
Narushima, Y., Sagara, N., Ogasawara, H.: A smoothing Newton method with Fischer–Burmeister function for second-order cone complementarity problems. J. Optim. Theory Appl. 149, 79–101 (2011)
Ni, T., Gu, W.Z.: Smoothing Newton algorithm for symmetric cone complementarity problems based on a one-parametric class of smoothing functions. J. Appl. Math. Comput. 35, 73–92 (2011)
Pang, J.S.: Error bounds in mathematical programming. Math. Program. 79, 299–332 (1997)
Pan, S., Chen, J.-S.: A one-parametric class of merit functions for the symmetric cone complementarity problem. J. Math. Anal. Appl. 355, 195–215 (2009)
Potra, F.A.: Weighted complementarity problems—a new paradigm for computing equilibria. SIAM J. Optim. 22, 1634–1654 (2012)
Potra, F.A.: Sufficient weighted complementarity problems. Comput. Optim. Appl. 64(2), 467–488 (2016)
Qi, L., Sun, J.: A nonsmooth version of Newton’s method. Math. Program. 58, 353–367 (1993)
Sun, D., Sun, J.: Löwner’s operator and spectral functions in Euclidean Jordan algebras. Math. Oper. Res. 33(2), 421–445 (2008)
Tang, J.Y.: A variant nonmonotone smoothing algorithm with improved numerical results for large-scale LWCPs. Comput. Appl. Math. 37, 3927–3936 (2018)
Tang, J.Y., Dong, L., Fang, L., Zhou, J.C.: The convergence of a modified smoothing-type algorithm for the symmetric cone complementarity problem. J. Appl. Math. Comput. 43, 307–328 (2013)
Tang, J.Y., Zhou, J.C., Fang, L.: Strong convergence properties of a modified nonmonotone smoothing algorithm for the SCCP. Optim. Lett. 12, 411–424 (2018)
Yoshise, A.: Interior point trajectories and a homogeneous model for nonlinear complementarity problems over symmetric cones. SIAM J. Optim. 17, 1129–1153 (2006)
Zhang, J.L., Chen, J.: A smoothing Levenberg–Marquardt type method for LCP. J. Comput. Math. 22, 735–752 (2004)
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Florian Potra.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This research was partially supported by the National Natural Science Foundation of China (11601466), China Scholarship Council and Nanhu Scholars Program for Young Scholars of XYNU, and by the USA National Science Foundation (1522654, 1819161).
Appendix: The proof of the example given in remark (ii) for Assumption 5.1
Appendix: The proof of the example given in remark (ii) for Assumption 5.1
Since M is positive definite, all of its diagonal elements are greater than zero. Without loss of generality, here we assume that \(M-I_n\) is positive definite where \(I_n\) represents the \(n\times n\) identity matrix. Consider the smoothing function
where \(t\in [1,2]\) and \(\tau \in [0,4)\). By (3.6), we have
Then we can reformulate the problem (5.7) as the following nonlinear smooth equations:
and apply Algorithm 4.1 to solve it. Let
Then, \(\mathcal {H}(z)\) is continuously differentiable at any \(z\in \mathcal {R}_{++}\times \mathcal {R}^{2n}\) with its Jacobian
where
Since M is positive definite, \(\mathcal {H}'(z)\) is nonsingular at any \(z\in \mathcal {R}_{++}\times \mathcal {R}^{2n}\) and we can find its inverse
where
In what follows, we divide the analysis into three parts.
Part 1. We show that \(\mathbf{diag }(d_x)\) and \(\mathbf{diag }(d_s)\) are positive semidefinite and bounded when \(\mu \rightarrow 0^+\). This result holds by noticing that \(g(\mu ,a,b)\) can be written as
and hence for any \((\mu ,a,b)\in \mathcal {R}_+\times \mathcal {R}^2\),
Part 2. We show that \(\Vert [\mathbf{diag }(d_x)+\mathbf{diag }(d_s)M]^{-1}\Vert \le \frac{\sqrt{n}\mathrm {cond}(M-I_n)}{2-\sqrt{\tau }}\) when \(\mu \rightarrow 0^+\), where \(\mathrm {cond}(M-I_n)\) is the conditional number of \(M-I_n\). First, we have
where
Since \(\big |\frac{\tau /2(a+b)}{g(\mu ,a,b)}\big |\le \big |\frac{\tau /2(a+b)}{\sqrt{a^2+b^2+(\tau -2)ab}}\big |\le \sqrt{\tau }\) holds for any \(\tau \in [0,4)\) and \((\mu ,a,b)\in \mathcal {R}_+\times \mathcal {R}^2\), we have
Hence, \(\mathbf {diag}(d_{xs})\) is positive definite and so is \(\mathbf {diag}(d_{xs})^{-1}\). Since \(M-I_n\) is positive definite, it is invertible and \((M-I_n)^{-1}\) is also positive definite. So, we have
When \(\mu \rightarrow 0^+\), since \(\mathbf{diag }(d_s)\) is positive semidefinite by Part 1, \(\mathbf {diag}(d_{xs})^{-1}\mathbf{diag }(d_s)\) is positive semidefinite. So, \((M-I_n)^{-1}+\mathbf {diag}(d_{xs})^{-1}\mathbf{diag }(d_s)\) is positive definite. Hence, when \(\mu \rightarrow 0^+,\) \(\mathbf{diag }(d_x)+\mathbf{diag }(d_s)M\) is nonsingular and
which implies that
where \(\mathrm {cond}(M-I_n):=\Vert M-I_n\Vert \Vert (M-I_n)^{-1}\Vert \).
Part 3. We show that when \(\mu \rightarrow 0^+\), if \(w>0\), then \(\Vert d_\mu \Vert \le \varrho \) where \(\varrho >0\) is a constant, and if \(w\ge 0\), then \(\Vert d_\mu \Vert \le \frac{\sqrt{n}t}{\mu ^{1-\frac{t}{2}}}\). This result holds since for any \((\mu ,a,b)\in \mathcal {R}_+\times \mathcal {R}^2\),
Therefore, when \(\mu \rightarrow 0^+\), from Parts 1,2 and 3, we have that \(z_{22},z_{23},z_{32}\) and \(z_{33}\) are bounded, and \(z_{21}, z_{31}\) are bounded or
Hence, for any \(t\in [1,2]\), we can conclude that for any \(z=(\mu ,x,s)\in \mathcal {R}_{++}\times \mathcal {R}^{2n}\), when \(\mu \rightarrow 0^+\), \(\Vert \mathcal {H}'(z)^{-1}\Vert \) is bounded or there exists a constant \({C}_t>0\) such that
Since \(1-\frac{t}{2}\in [0,1/2],\) there exist constants \(C>0\) and \(d\in [0,1/2)\) such that Assumption 5.1 holds. This completes the proof.\(\square \)
The proof of Lemma 4
For any \(\xi =(\mu ,a,b)^T, \xi '=(\mu ',a',b')^T\in \mathcal {R}^3\), we have
This proves (i). Now we prove (ii) by considering the following three cases.
Case 1. If \(c>0\), then \(\psi _c\) is differentiable at any \(\xi \in \mathcal {R}^3\) with
and
where
which yields
By (7.1), for any \(u,v\in \mathcal {R}^3\) we have
For any \(\xi \in \mathcal {R}^3\) and \(h\in \mathcal {R}^3\), \(\psi _c\) is differentiable at \(\xi +h\) and hence \( \partial {\psi }_c(\xi +h)=\{{\psi }_c'(\xi +h)^T\}\). So, from (7.2) we have for any \(V\in \partial {\psi }_c(\xi +h)\) and \(h\rightarrow 0\),
Case 2. If \(c=0\) and \(\xi =0\), then for any nonzero direction vector \(h=(\tilde{\mu },\tilde{a},\tilde{b})^T\in \mathcal {R}^3\), \({\psi }_c\) is smooth at the point \(0+h=h\). Thus, \(V\in \partial {\psi }_c(0+h)=\{{\psi }_c'(h)^T\}\) is uniquely given by
Then, for any \(V\in \partial {\psi }_c(0+h)\) and \(h\rightarrow 0\),
Case 3. If \(c=0\) and \(\xi \ne 0\), then \(\psi _c\) is differentiable at \(\xi \) and from (7.1) we have
Next we show that \({\psi }_c'(\xi )\) is locally Lipschitz continuous at \(\xi \) with the constant \(\frac{6\sqrt{2}}{\Vert \xi \Vert }\). In fact, let \(u,v\in N(\xi ,\frac{\Vert \xi \Vert }{2})\), we have
By (7.3), we have
Since \(u,v\in N(\xi ,\frac{\Vert \xi \Vert }{2})\), we have
It follows that
This together with (7.5) yields
Thus, from (7.4) and (7.6), we have for any \(u,v\in N(\xi ,\frac{\Vert \xi \Vert }{2})\),
Now we show that \({\psi }_c(\xi )\) is strongly semismooth at \(\xi \ne 0\). Since \(\xi \ne 0\), when \(h\rightarrow 0\), \(\xi +h\ne 0\) and hence \( \partial {\psi }_c(\xi +h)=\{{\psi }_c'(\xi +h)^T\}\). So, from (7.7), similarly as the proof of Case 1, we have for any \(V\in \partial {\psi }_c(\xi +h)\) and \(h\rightarrow 0\),
Thus, the proof is completed.\(\square \)
The proof of Lemma 5
Let \(\mathcal {{H}}(z)\) be defined by (5.37). Then, \(\mathcal {{H}}(z)\) is continuously differentiable at any \(z=(\mu ,x,s,y)\in \mathcal {R}_{++}\times \mathcal {R}^{2n+m}\) and its Jacobian is
where
We now divide the proof by the following three parts.
Part (i) We show that \(\mathcal {{{H}}}(z)\) is Lipschitz continuous on \(\mathcal {R}^{1+2n+m}\). In fact, since \((Px+Qs+Ry-a)'=[P, Q, R]\), we have that \(Px+Qs+Ry-a\) is Lipschitz continuous on \(\mathcal {R}^{2n+m}\). Moreover, from (i) of Lemma 4, \({\psi }_c\) is Lipschitz continuous on \(\mathcal {R}^{3}\). Hence, \(\mathcal {{{H}}}(z)\) is Lipschitz continuous on \(\mathcal {R}^{1+2n+m}\).
Part (ii) For any \(\theta >0\), we show that \(\mathcal {{{H}}}'(z)\) is bounded and Lipschitz continuous on the set
In fact, for any \(i=1,...,n,\) since
\(D_\mu \), \(D_x\) and \(D_s\) are bounded and hence \(\mathcal {{H}}'(z)\) is bounded. Let
For any \(\upsilon \in \varGamma \), define
Since \(\mu \ge \theta >0,\) \(f_c(\upsilon )\) is continuously differentiable and
which yields
By noticing that
we have
Thus, for any \(\tilde{\upsilon }, \upsilon \in \varGamma \), we have
This implies that \(f_c\) is Lipschitz continuous on \(\varGamma \) and hence \(D_x\) is Lipschitz continuous on \(\varOmega \). By a similar way, we can show that \(D_s\) and \(D_\mu \) are also Lipschitz continuous on \(\varOmega \) and so is \(\mathcal {{{H}}}'(z)\).
Part (iii) We show that \(\mathcal {{{M}}}'(z)\) is Lipschitz continuous on the set \(\varTheta \) defined by (5.9). In fact, \(\mathcal {{{M}}}(z)\) is continuous differentiable at any \(z\in \mathcal {R}_{++}\times \mathcal {R}^{2n+m}\) and
So, for any \(\tilde{z},z\in \varTheta \), by Parts (i) and (ii) and \(\Vert \mathcal {H}(\tilde{z})\Vert \le 2 \Vert \mathcal {H}(z^0)\Vert \), there exists a constant \(M>0\) such that
This completes the proof.\(\square \)
Rights and permissions
About this article
Cite this article
Tang, J., Zhang, H. A Nonmonotone Smoothing Newton Algorithm for Weighted Complementarity Problem. J Optim Theory Appl 189, 679–715 (2021). https://doi.org/10.1007/s10957-021-01839-6
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10957-021-01839-6
Keywords
- Smoothing Newton algorithm
- Jacobian nonsingularity
- Superlinear/quadratic convergence
- Weighted complementarity problem
- Symmetric cone