Movatterモバイル変換


[0]ホーム

URL:


Skip to main content

Advertisement

Springer Nature Link
Log in

An infeasible interior-point arc-search algorithm for nonlinear constrained optimization

  • Original Paper
  • Published:
Numerical Algorithms Aims and scope Submit manuscript

Abstract

In this paper, we propose an infeasible arc-search interior-point algorithm for solving nonlinear programming problems. Most algorithms based on interior-point methods are categorized as line search since they compute a next iterate on a straight line determined by a search direction which approximates the central path. The proposed arc-search interior-point algorithm uses an arc for the approximation. We discuss convergence properties of the proposed algorithm. We also conduct numerical experiments on the CUTEst benchmark problems and compare the performance of the proposed arc-search algorithm with that of a line-search algorithm. Numerical results indicate that the proposed arc-search algorithm reaches the optimal solution using fewer iterations but longer times than a line-search algorithm. A modification that leads to a faster arc-search algorithm is also discussed.

This is a preview of subscription content,log in via an institution to check access.

Access this article

Log in via an institution

Subscribe and save

Springer+ Basic
¥17,985 /Month
  • Get 10 units per month
  • Download Article/Chapter or eBook
  • 1 Unit = 1 Article or 1 Chapter
  • Cancel anytime
Subscribe now

Buy Now

Price includes VAT (Japan)

Instant access to the full article PDF.

Fig. 1
Fig. 2
Fig. 3

Similar content being viewed by others

References

  1. Byrd, R. H., Hribar, M. E., Nocedal, J.: An interior point algorithm for large-scale nonlinear programming. SIAM J. Optim.9, 877–900 (1999)

    Article MathSciNet  Google Scholar 

  2. Byrd, R. H., Gilbert, J. C., Nocedal, J.: A trust region method based on interior point techniques for nonlinear programming. Math. Progr.89, 149–185 (2000)

    Article MathSciNet  Google Scholar 

  3. El-Bakry, A. S., Tapia, R. A., Tsuchiya, T., Zhang, Y.: On the formulation and theory of the Newton interior-point method for nonlinear programming. J. Optim. Theory Appl.89, 507–541 (1996)

    Article MathSciNet  Google Scholar 

  4. Forsgren, A., Gill, P. E.: Primal-dual interior methods for nonconvex nonlinear programming. SIAM J. Optim.8, 1132–1152 (1998)

    Article MathSciNet  Google Scholar 

  5. Gould, N., Scott, J.: A note on performance profiles for benchmarking software. ACM Trans. Math. Softw.43(2), 15 (2016)

    Article MathSciNet  Google Scholar 

  6. Gould, N. I., Orban, D., Toint, P.L.: CUTEst: a constrained and unconstrained testing environment with safe threads for mathematical optimization. Comput. Optim. Appl.60(3), 545–557 (2015)

    Article MathSciNet  Google Scholar 

  7. Kheirfam, B.: An arc-search infeasible interior-point algorithm for horizontal linear complementarity problem in the\(N^{-\infty }\) neighbourhood of the central path. Int. J. Comput. Math.94(1), 2271–2282 (2017)

    Article MathSciNet  Google Scholar 

  8. Lu, T., Shiou, S.: Inverses of 2 × 2 block matrices. Comput. Math. Appl.43, 119–129 (2002)

    Article MathSciNet  Google Scholar 

  9. Lustig, I., Marsten, R., Shannon, D.: Computational experience with a primal-dual interior-point method for linear programming. Linear Algebra Appl.152, 192–222 (1991)

    Article MathSciNet  Google Scholar 

  10. Lustig, I., Marsten, R., Shannon, D.: On implementing Mehrotra’s predictor-corrector interior-point method for linear programming. SIAM J. Optim.2, 432–449 (1992)

    Article MathSciNet  Google Scholar 

  11. Mehrotra, S.: On the implementation of a primal-dual interior point method. SIAM J. Optim.2, 575–601 (1992)

    Article MathSciNet  Google Scholar 

  12. Nocedal, J., Wachter, A., Waltz, R. A.: Adaptive barrier update strategies for nonlinear interior methods. SIAM J. Optim.19, 1674–1693 (2009)

    Article MathSciNet  Google Scholar 

  13. Plantenga, T.: A trust region method for nonlinear programming based on primal interior-point techniques. SIAM J. Optim.20, 282–305 (1998)

    MathSciNet MATH  Google Scholar 

  14. Tits, A. L., Wachter, A., Bakhtiarl, S., Urban, T. J., Lawrence, C. T.: A primal-dual method for nonlinear programming with strong global and local convergence properties. Math. Progr.8, 1132–1152 (1998)

    Google Scholar 

  15. Ulbrich, M., Ulbrich, S., Vicente, L. N.: A globally convergent primal-dual interior-point filter method for nonlinear programming. Math. Progr.100, 379–410 (2004)

    Article MathSciNet  Google Scholar 

  16. Vanderbei, R., Shanno, D.: An interior-point algorithm for nonconvex nonlinear programming. Comput. Optim. Appl.13, 231–252 (1999)

    Article MathSciNet  Google Scholar 

  17. Wright, S.: Primal-Dual Interior-Point Methods. SIAM, Philadelphia (1997)

    Book  Google Scholar 

  18. Yang, Y.: A polynomial arc-search interior-point algorithm for convex quadratic programming. Eur. J. Oper. Res.215(1), 25–38 (2011)

    Article MathSciNet  Google Scholar 

  19. Yang, Y.: A polynomial arc-search interior-point algorithm for linear programming. J. Optim. Theory Appl.158, 859–873 (2013)

    Article MathSciNet  Google Scholar 

  20. Yang, Y.: Curvelp-a matlab implementation of an infeasible interior-point algorithm for linear programming. Numer. Algorithms74, 967–996 (2017)

    Article MathSciNet  Google Scholar 

  21. Yang, Y.: Two computationally efficient polynomial-iteration infeasible interior-point algorithms for linear programming. Numer. Algorithms79, 957–992 (2018)

    Article MathSciNet  Google Scholar 

  22. Yang, Y., Yamashita, M.: An arc-searchO(nL) infeasible-interior-point algorithm for linear programming. Optim. Lett.12(1), 781–798 (2018)

    Article MathSciNet  Google Scholar 

  23. Yang, X., Liu, H., Zhang, Y.: An arc-search infeasible-interior-point method for symmetric optimization in a wide neighborhood of the central path. Optim. Lett.11(1), 135–152 (2017)

    Article MathSciNet  Google Scholar 

  24. Ye, Y.: Interior Point Algorithms: Theory and Analysis. Wiley, New York (1997)

    Book  Google Scholar 

  25. Zhang, M., Yuan, B., Zhou, Y., Luo, X., Huang, Z.: A primal-dual interior-point algorithm with arc-search for semidefinite programming. Optim. Lett.13(5), 1157–1175 (2019)

    Article MathSciNet  Google Scholar 

  26. Zhao, G., Zhu, J.: Analytical properties of the central trajectory in interior point methods. In: Du, D. Z., Sun, J. (eds.) Advances in Optimization and Approximation, pp 362–375. Springer, Boston (1994)

Download references

Funding

This research was partially supported by JSPS KAKENHI (Grant Number: 18K11176).

Author information

Authors and Affiliations

  1. Department of Mathematical and Computing Science, Tokyo Institute of Technology, Tokyo, Japan

    Makoto Yamashita & Einosuke Iida

  2. US NRC, Office of Research, 21 Church Street, Rockville, MD, USA

    Yaguang Yang

Authors
  1. Makoto Yamashita

    You can also search for this author inPubMed Google Scholar

  2. Einosuke Iida

    You can also search for this author inPubMed Google Scholar

  3. Yaguang Yang

    You can also search for this author inPubMed Google Scholar

Corresponding author

Correspondence toMakoto Yamashita.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendices

Appendix: 1: Derivatives

In this section, we give notation related to derivatives. The Hessian matrix of\(f : {\mathbb {R}}^{n} \to {\mathbb {R}}\) is

$$ \begin{array}{@{}rcl@{}} \nabla^{2} f(x) = \left[ \begin{array}{cccc} \frac{\partial^{2} f}{\partial x_{1} \partial x_{1}} & \frac{\partial^{2} f}{\partial x_{1} \partial x_{2}} & {\cdots} & \frac{\partial^{2} f}{\partial x_{1} \partial x_{n}} \\ {\vdots} & {\vdots} & {\ddots} & {\vdots} \\ \frac{\partial^{2} f}{\partial x_{n} \partial x_{1}} & \frac{\partial^{2} f}{\partial x_{n} \partial x_{2}} & {\cdots} & \frac{\partial^{2} f}{\partial x_{n} \partial x_{n}} \end{array} \right] \in {\mathbb{R}}^{n \times n}. \end{array} $$

The Jacobian for\(h : {\mathbb {R}}^{n} \to {\mathbb {R}}^{m}\) is

$$ \begin{array}{@{}rcl@{}} \nabla h(x) = \left[ \begin{array}{cccc} \frac{\partial h_{1}}{\partial x_{1} } & \frac{\partial h_{2}}{\partial x_{1} } & {\cdots} & \frac{\partial h_{m}}{\partial x_{1} } \\ {\vdots} & {\vdots} & {\ddots} & {\vdots} \\ \frac{\partial h_{1}}{\partial x_{n} } & \frac{\partial h_{2}}{\partial x_{n} } & \cdots & \frac{\partial h_{m}}{\partial x_{n} } \end{array} \right] =[\nabla h_{1}(x), \cdots, \nabla h_{m}(x) ] \in {\mathbb{R}}^{n \times m}. \end{array} $$

The Jacobian for\(g : {\mathbb {R}}^{n} \to {\mathbb {R}}^{p}\) is

$$ \begin{array}{@{}rcl@{}} \nabla g(x) = \left[ \begin{array}{cccc} \frac{\partial g_{1}}{\partial x_{1} } & \frac{\partial g_{2}}{\partial x_{1} } & {\cdots} & \frac{\partial g_{p}}{\partial x_{1} } \\ {\vdots} & {\vdots} & {\ddots} & {\vdots} \\ \frac{\partial g_{1}}{\partial x_{n} } & \frac{\partial g_{2}}{\partial x_{n} } & \cdots & \frac{\partial g_{p}}{\partial x_{n} } \end{array} \right] =[\nabla g_{1}(x), \cdots, \nabla g_{p}(x) ] \in {\mathbb{R}}^{n \times p}. \end{array} $$

For the right-hand side of (8), we use

$$ \begin{array}{@{}rcl@{}} {\nabla_{x}^{3}} L(x,y,z)\dot{x} \dot{x} &=& \frac{\partial \left( \frac{\partial^{2} L(x,y,z)}{\partial x^{2}} \dot{x} \right) } {\partial x} \dot{x} = \sum\limits_{i=1}^{n} \dot{x}_{i} \frac{\partial}{\partial x} \left[ \begin{array}{c} \frac{\partial^{2} L(x,y,z)}{\partial x_{1} \partial x_{i}} \\ {\vdots} \\ \frac{\partial^{2} L(x,y,z)}{\partial x_{n} \partial x_{i}} \end{array} \right] \dot{x} \\ {\nabla_{x}^{2}} h(x)\dot{y} \dot{x} &=& \frac{\partial \left( \frac{\partial h(x)}{\partial x } \dot{y} \right) } {\partial x} \dot{x} = \sum\limits_{i=1}^{m} \dot{y}_{i} \frac{\partial}{\partial x} \left[ \begin{array}{c} \frac{\partial h_{i}(x)}{\partial x_{1} } \\ {\vdots} \\ \frac{\partial h_{i}(x)}{\partial x_{n}} \end{array} \right] \dot{x} = \sum\limits_{i=1}^{m} \dot{y}_{i} \left( {\nabla_{x}^{2}} h_{i}(x) \right) \dot{x} \\ {\nabla_{x}^{2}}g(x)\dot{z} \dot{x} &=& \frac{\partial \left( \frac{\partial g(x)}{\partial x } \dot{z} \right) } {\partial x} \dot{x} = \sum\limits_{i=1}^{n} \dot{z}_{i} \frac{\partial}{\partial x} \left[ \begin{array}{c} \frac{\partial g_{i}(x)}{\partial x_{1} } \\ {\vdots} \\ \frac{\partial g_{i}(x)}{\partial x_{n}} \end{array} \right] \dot{x} = \sum\limits_{i=1}^{n} \dot{z}_{i} \left( {\nabla_{x}^{2}} g_{i}(x) \right) \dot{x} \\ {\nabla_{x}^{2}} h(x)^{\mathrm{T}} \dot{x} \dot{x} &=& \left( \frac{\partial \left( \left( \frac{\partial h(x)} {\partial x } \right)^{\mathrm{T}} \dot{x} \right) }{\partial x} \right)^{\mathrm{T}} \dot{x} = \left[ \begin{array}{c} \dot{x}^{\mathrm{T}} \left( {\nabla_{x}^{2}} h_{1}(x) \right) \dot{x} \\ {\vdots} \\ \dot{x}^{\mathrm{T}} \left( {\nabla_{x}^{2}} h_{m}(x) \right) \dot{x} \end{array} \right] \\ {\nabla_{x}^{2}} g(x)^{\mathrm{T}} \dot{x} \dot{x} &=& \left( \frac{\partial \left( \left( \frac{\partial g(x)} {\partial x } \right)^{\mathrm{T}} \dot{x} \right) }{\partial x} \right)^{\mathrm{T}} \dot{x} = \left[ \begin{array}{c} \dot{x}^{\mathrm{T}} \left( {\nabla_{x}^{2}} g_{1}(x) \right) \dot{x} \\ {\vdots} \\ \dot{x}^{\mathrm{T}} \left( {\nabla_{x}^{2}} g_{p}(x) \right) \dot{x} \end{array} \right]. \end{array} $$

??

Appendix: 2: The largest step angle

In this section, we give analytical forms to compute the largest\(\alpha _{w_{i}}\) and\(\alpha _{s_{i}}\) for eachi in (12). For simplicity, here, we drop the indexi and the iteration numberk; for example,\({w_{i}^{k}}\) is simply written asw. For (11a), we should have

$$ \begin{array}{@{}rcl@{}} w(\alpha)=w - \dot{w}\sin(\alpha)+\ddot{w}(1-\cos(\alpha)) \ge \delta w, \end{array} $$

or equivalently,

$$ w - \delta w +\ddot{w} \ge \dot{w}\sin(\alpha)+\ddot{w}\cos(\alpha). $$
(28)

We split this computation into seven cases by the signs of\(\dot {w}\) and\(\ddot {w}\).

Case 1

(\(\dot {w}=0\)and\(\ddot {w}\ne 0\))

If\(\ddot {w} \ge -(1-\delta ) w\), thenw(α) ≥δw holds for\(\alpha \in [0, \frac {\pi }{2}]\). If\(\ddot {w} \le -(1-\delta )w < 0\), to meet (28), we must have\(\cos \limits (\alpha ) \ge \frac {w-\delta w +\ddot {w}}{\ddot {w}} \ge 0\), or,\(\alpha \le \cos \limits ^{-1}\left (\frac {w-\delta w +\ddot {w}} {\ddot {w}} \right )\). Therefore,

$$ \alpha_{w} = \left\{ \begin{array}{ll} \frac{\pi}{2} & \quad \text{if} w-\delta w +\ddot{w} \ge 0 \\ \cos^{-1}\left( \frac{w-\delta w +\ddot{w}} {\ddot{w}} \right) & \quad \text{if} w-\delta w +\ddot{w} \le 0. \end{array} \right. $$

Case 2

(\(\ddot {w}=0\)and\(\dot {w}\ne 0\))

If\(\dot {w} \le (1 - \delta ) w\), thenw(α) ≥δw holds for any\(\alpha \in [0, \frac {\pi }{2}]\). If\(\dot {w} \ge (1 - \delta ) w > 0\), to meet (28), we must have\(\sin \limits (\alpha ) \le \frac {w-\delta w }{\dot {w}}\), or\(\alpha \le \sin \limits ^{-1}\left (\frac {w-\delta w } {\dot {w}} \right )\). Therefore,

$$ \alpha_{w} = \left\{ \begin{array}{ll} \frac{\pi}{2} & \quad \text{if} \dot{w} \le w-\delta w \\ \sin^{-1}\left( \frac{w-\delta w } {\dot{w}} \right) & \quad \text{if} \dot{w} \ge w-\delta w. \end{array} \right. $$

Case 3

(\(\dot {w}>0\)and\(\ddot {w}>0\))

Let\(\beta = \sin \limits ^{-1} \left (\frac {\ddot {w} } {\sqrt {\dot {w}^{2}+\ddot {w}^{2}}} \right )\). We can express\(\dot {w}=\sqrt {\dot {w}^{2}+\ddot {w}^{2}}\cos \limits (\beta )\) and\(\ddot {w}=\sqrt {\dot {w}^{2}+\ddot {w}^{2}}\sin \limits (\beta )\). Then, (28) can be rewritten as

$$ w-\delta w + \ddot{w} \ge \sqrt{\dot{w}^{2}+\ddot{w}^{2}} \sin(\alpha + \beta). $$
(29)

If\(\ddot {w} + w- \delta w \ge \sqrt {\dot {w}^{2}+\ddot {w}^{2}}\), thenw(α) ≥δw holds for any\(\alpha \in [0, \frac {\pi }{2}]\). If\(\ddot {w} + w- \delta w \le \sqrt {\dot {w}^{2}+\ddot {w}^{2}}\), to meet (29), we must have\(\sin \limits (\alpha + \beta ) \le \frac {w-\delta w + \ddot {w}} {\sqrt {\dot {w}^{2}+\ddot {w}^{2}}}\), or\(\alpha + \beta \le \sin \limits ^{-1}\left (\frac {w-\delta w + \ddot {w} } {\sqrt {\dot {w}^{2}+\ddot {w}^{2}}} \right )\). Therefore,

$$ \alpha_{w} = \left\{ \begin{array}{ll} \frac{\pi}{2} & \quad \text{if} w-\delta w + \ddot{w} \ge \sqrt{\dot{w}^{2}+\ddot{w}^{2}} \\ \sin^{-1}\left( \frac{w-\delta w + \ddot{w} } {\sqrt{\dot{w}^{2}+\ddot{w}^{2}}} \right) - \sin^{-1}\left( \frac{\ddot{w} } {\sqrt{\dot{w}^{2}+\ddot{w}^{2}}} \right) & \quad \text{if} w-\delta w + \ddot{w} \le \sqrt{\dot{w}^{2}+\ddot{w}^{2}}. \end{array} \right. $$

Case 4

(\(\dot {w}>0\)and\(\ddot {w}<0\))

Let\(\beta = \sin \limits ^{-1} \left (\frac {-\ddot {w} } {\sqrt {\dot {w}^{2}+\ddot {w}^{2}}} \right )\). We can express\(\dot {w}=\sqrt {\dot {w}^{2}+\ddot {w}^{2}}\cos \limits (\beta )\) and\(\ddot {w}=-\sqrt {\dot {w}^{2}+\ddot {w}^{2}}\sin \limits (\beta )\). Then, (28) can be rewritten as

$$ w-\delta w + \ddot{w} \ge \sqrt{\dot{w}^{2}+\ddot{w}^{2}} \sin(\alpha - \beta). $$
(30)

If\(\ddot {w} + w- \delta w \ge \sqrt {\dot {w}^{2}+\ddot {w}^{2}}\), thenw(α) ≥δw holds for any\(\alpha \in [0, \frac {\pi }{2}]\). If\(\ddot {w} +w- \delta w \le \sqrt {\dot {w}^{2}+\ddot {w}^{2}}\), to meet (30), we must have\(\sin \limits (\alpha - \beta ) \le \frac {w-\delta w + \ddot {w}} {\sqrt {\dot {w}^{2}+\ddot {w}^{2}}}\), or\(\alpha - \beta \le \sin \limits ^{-1}\left (\frac {w-\delta w + \ddot {w} } {\sqrt {\dot {w}^{2}+\ddot {w}^{2}}} \right )\). Therefore,

$$ \alpha_{w} = \left\{ \begin{array}{ll} \frac{\pi}{2} & \quad \text{if} w-\delta w + \ddot{w} \ge \sqrt{\dot{w}^{2}+\ddot{w}^{2}} \\ \sin^{-1}\left( \frac{w-\delta w + \ddot{w} } {\sqrt{\dot{w}^{2}+\ddot{w}^{2}}} \right) + \sin^{-1}\left( \frac{-\ddot{w} } {\sqrt{\dot{w}^{2}+\ddot{w}^{2}}} \right) & \quad \text{if} w-\delta w + \ddot{w} \le \sqrt{\dot{w}^{2}+\ddot{w}^{2}}. \end{array} \right. $$

Case 5

(\(\dot {w}<0\)and\(\ddot {w}<0\))

Let\( \beta = \sin \limits ^{-1} \left (\frac {-\ddot {w} } {\sqrt {\dot {w}^{2}+\ddot {w}^{2}}} \right )\). We can express\(\dot {w}=-\sqrt {\dot {w}^{2}+\ddot {w}^{2}}\cos \limits (\beta )\) and\(\ddot {w}=-\sqrt {\dot {w}^{2}+\ddot {w}^{2}}\sin \limits (\beta )\). Then, (28) can be rewritten as

$$ w-\delta w + \ddot{w} \ge -\sqrt{\dot{w}^{2}+\ddot{w}^{2}} \sin(\alpha +\beta), $$
(31)

If\(\ddot {w} +(w- \delta w) \ge 0\), thenw(α) ≥δw holds for any\(\alpha \in [0, \frac {\pi }{2}]\). If\(\ddot {w} +(w- \delta w) \le 0\), to meet (31), we must have\(\sin \limits (\alpha + \beta ) \ge \frac {-(w-\delta w + \ddot {w})} {\sqrt {\dot {w}^{2}+\ddot {w}^{2}}}\), or\(\alpha + \beta \le \pi - \sin \limits ^{-1} \left (\frac {-(w-\delta w + \ddot {w})}{\sqrt {\dot {w}^{2}+\ddot {w}^{2}}} \right )\). Therefore,

$$ \alpha_{w} = \left\{ \begin{array}{ll} \frac{\pi}{2} & \quad \text{if} w-\delta w + \ddot{w} \ge 0 \\ \pi - \sin^{-1} \left( \frac{-(w-\delta w + \ddot{w}) } {\sqrt{\dot{w}^{2}+\ddot{w}^{2}}} \right) - \sin^{-1}\left( \frac{-\ddot{w} } {\sqrt{\dot{w}^{2}+\ddot{w}^{2}}} \right) & \quad \text{if} w-\delta w + \ddot{w} \le 0. \end{array} \right. $$

Case 6

(\(\dot {w}<0\)and\(\ddot {w}>0\))

Clearly, (28) always holds for any\(\alpha \in [0, \frac {\pi }{2}]\). Therefore, we can take

$$ \alpha_{w} = \frac{\pi}{2}. $$
(32)

Case 7

(\(\dot {w}=0\)and\(\ddot {w}=0\))

Clearly, (28) always holds for any\(\alpha \in [0, \frac {\pi }{2}]\). Therefore, we can take

$$ \alpha_{w} = \frac{\pi}{2}. $$
(33)

Similar analysis can be performed for (11b), then similar analytical forms are derived forαs.

Appendix: 3: Details on numerical results

Tables 1 and 2 report the objective value, the numbers of iterations, and the computation time (in seconds) of the proposed arc-search algorithm (Algorithm 1) and the line-search algorithm (Algorithm 2) for QCQP and Other type problems. The symbol “Unattained” indicates that the algorithms stopped prematurely, mainly because of the numerical errors. We excluded the problems that all the three algorithms (Algorithm 1, Algorithm 1 with\(\ddot {\ddot {v}}\), and Algorithm 2) stopped with “Unattained.”

Table 1 Results on QCQP problems
Table 2 Results on Others

Rights and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Yamashita, M., Iida, E. & Yang, Y. An infeasible interior-point arc-search algorithm for nonlinear constrained optimization.Numer Algor89, 249–275 (2022). https://doi.org/10.1007/s11075-021-01113-w

Download citation

Keywords

Mathematics Subject Classification (2010)

Access this article

Subscribe and save

Springer+ Basic
¥17,985 /Month
  • Get 10 units per month
  • Download Article/Chapter or eBook
  • 1 Unit = 1 Article or 1 Chapter
  • Cancel anytime
Subscribe now

Buy Now

Price includes VAT (Japan)

Instant access to the full article PDF.

Advertisement


[8]ページ先頭

©2009-2025 Movatter.jp