Abstract
In the last two decades, many descent methods for multiobjective optimization problems were proposed. In particular, the steepest descent and the Newton methods were studied for the unconstrained case. In both methods, the search directions are computed by solving convex subproblems, and the stepsizes are obtained by an Armijo-type line search. As a consequence, the objective function values decrease at each iteration of the algorithms. In this work, we consider nonmonotone line searches, i.e., we allow the increase of objective function values in some iterations. Two well-known types of nonmonotone line searches are considered here: the one that takes the maximum of recent function values, and the one that takes their average. We also propose a new nonmonotone technique specifically for multiobjective problems. Under reasonable assumptions, we prove that every accumulation point of the sequence produced by the nonmonotone version of the steepest descent and Newton methods is Pareto critical. Moreover, we present some numerical experiments, showing that the nonmonotone technique is also efficient in the multiobjective case.
Similar content being viewed by others
Notes
In the original version, either the bounds L and U, or the variable n can be modified.
It is an adaptation of a single-objective optimization problem to the multiobjective setting. Since the problem is originally unconstrained, we also added some bound constraints.
References
Das, I., Dennis, J.E.: Normal-boundary intersection: a new method for generating the Pareto surface in nonlinear multicriteria optimization problems. SIAM J. Optim. 8(3), 631–657 (1998)
Dolan, E.D., Moré, J.J.: Benchmarking optimization software with performance profiles. Math. Program. 91(2), 201–213 (2002)
Fazzio, N., Schuverdt, M.L.: Convergence analysis of a nonmonotone projected gradient method for multiobjective optimization problems (2018) (submitted)
Fliege, J., Graña Drummond, L.M., Svaiter, B.F.: Newton’s method for multiobjective optimization. SIAM J. Optim. 20(2), 602–626 (2009)
Fliege, J., Svaiter, B.F.: Steepest descent methods for multicriteria optimization. Math. Methods Oper. Res. 51(3), 479–494 (2000)
Fukuda, E.H., Graña Drummond, L.M.: A survey on multiobjective descent methods. Pesquisa Oper. 34(3), 585–620 (2014)
Grippo, L., Lampariello, F., Lucidi, S.: A nonmonotone line search technique for Newton’s method. SIAM J. Numer. Anal. 23(4), 707–716 (1986)
Jin, Y., Olhofer, M., Sendhoff, B.: Dynamic weighted aggregation for evolutionary multi-objective optimization: Why does it work and how? In: GECCO’01 Proceedings of the 3rd Annual Conference on Genetic and Evolutionary Computation, pp. 1042–1049 (2001)
Kim, I.Y., de Weck, O.L.: Adaptive weighted-sum method for bi-objective optimization: Pareto front generation. Struct. Multidiscip. Optim. 29(2), 149–158 (2005)
Laumanns, M., Thiele, L., Deb, K., Zitzler, E.: Combining convergence and diversity in evolutionary multiobjective optimization. Evolut. Comput. 10(3), 263–282 (2002)
Luc, D.T.: Scalarization of vector optimization problems. J. Optim. Theory Appl. 55(1), 85–102 (1987)
Mita, K.: Nonmonotone line search in multiobjective settings (in Japanese). Undergraduate research, Kyoto University(2017)
Mita, K., Fukuda, E.H., Yamashita, N.: On using nonmonotone line search techniques in steepest descent methods for multiobjective optimization (in Japanese). In: Proceedings of the 61st Annual Conference of the Institute of Systems, Control and Information Engineers (2017)
Moré, J.J., Garbow, B.S., Hillstrom, K.E.: Testing unconstrained optimization software. ACM T. Math. Softw. 7(1), 17–41 (1981)
Nocedal, J.: Updating quasi-newton matrices with limited storage. Math. Comput. 35(151), 773–782 (1980)
Ogata, Y., Saito, Y., Tanaka, T., Yamada, S.: Sublinear scalarization methods for sets with respect to set-relations. Linear Nonlinear Anal. 3(1), 121–132 (2017)
Qu, S., Ji, Y., Jiang, J., Zhang, Q.: Nonmonotone gradient methods for vector optimization with a portfolio optimization application. Eur. J. Oper. Res. 263, 356–366 (2017)
Stadler, W., Dauer, J.: Multicriteria optimization in engineering: a tutorial and survey. In: Kamat, M.P. (ed.) Progress in Aeronautics and Astronautics: Structural Optimization: Status and Promise, vol. 150, pp. 209–249. American Institute of Aeronautics and Astronautics, Reston (1992)
Toint, Ph.L.: Test problems for partially separable optimization and results for the routine PSPMIN. Tech. Rep. 83/4, Department of Mathematics, University of Namur, Brussels (1983)
Zhang, H., Hager, W.W.: A nonmonotone line search technique and its application to unconstrained optimization. SIAM J. Optim. 14(4), 1043–1056 (2004)
Zitzler, E., Deb, K., Thiele, L.: Comparison of multiobjective evolutionary algorithms: empirical results. Evolut. Comput. 8(2), 173–195 (2000)
Acknowledgements
We would like to thank the anonymous referees for their suggestions, which improved the original version of the paper.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was supported by the Kyoto University Foundation, and the Grant-in-Aid for Scientific Research (C) (17K00032 and 19K11840) from Japan Society for the Promotion of Science.
Appendix A
Appendix A
Here, we list the test problems used in Sect. 7. For each problem, we state the original reference, the number of variables n, the number of objective functions m, the convexity property, the objective functions, and the bounds L and U of the box constraints.
-
1.
Das and Dennis (DD1) [1]: \(n=5\), \(m=2\), nonconvex,Footnote 1\(^{,}\)Footnote 2
$$\begin{aligned} F_{1}(x)&=x_{1}^{2}+x_{2}^{2}+x_{3}^{2}+x_{4}^{2}+x_{5}^{2},\\ F_{2}(x)&=3x_{1}+2x_{2}-\frac{x_{3}}{3}+0.01(x_{4}-x_{5})^{3}, \end{aligned}$$\(L=(-20, \ldots , -20)^{\top }\), and \(U=(20, \ldots , 20)^{\top }\).
-
2.
Fliege, Graña Drummond and Svaiter (FDS) [4]: \(n=10\), \(m=3\), convex,\(^2\)
$$\begin{aligned} F_{1}(x)&=\frac{1}{n^{2}}\sum _{i=1}^{n}i(x_{i}-i)^{4},\\ F_{2}(x)&=\exp \left( \sum _{i=1}^{n}\frac{x_{i}}{n}\right) +\Vert x\Vert _{2}^{2},\\ F_{3}(x)&=\frac{1}{n(n+1)}\sum _{i=1}^{n}i(n-i+1)e^{-x_{i}}, \end{aligned}$$\(L=(-2, \ldots , -2)^{\top }\), and \(U=(2,\ldots , 2)^{\top }\).
-
3.
Jin, Olhofer and Sendhoff (JOS1) [8]: \(n=5\), \(m=2\), quadratic convex,\(^2\)
$$\begin{aligned} F_{1}(x)&=\frac{1}{n}\sum _{i=1}^{n}x_{i}^{2},\\ F_{2}(x)&=\frac{1}{n}\sum _{i=1}^{n}(x_{i}-2)^{2}, \end{aligned}$$\(L=(-2,\ldots , -2)^{\top }\), and \(U=(2,\ldots , 2)^{\top }\).
-
4.
Kim and Weck (KW2) [9]: \(n=2\), \(m=2\), nonconvex,
$$\begin{aligned} F_{1}(x) =&-3(1-x_{1})^{2} \exp (-x_{1}^{2}-(x_{2}+1)^{2})\\&+ 10\left( \frac{x_{1}}{5}-x_{1}^{3}-x_{2}^{5}\right) \exp (-x_{1}^{2}-x_{2}^{2})\\&+3 \exp (-(x_{1}+2)^{2}-x_{2}^{2})-0.5(2x_{1}+x_{2}),\\ F_{2}(x) =&-3(1+x_{2})^{2} \exp (-x_{2}^{2}-(1-x_{1})^{2}) \\&+ 10\left( -\frac{x_{2}}{5}+x_{2}^{3}+x_{1}^{5}\right) \exp (-x_{1}^{2}-x_{2}^{2})\\&+ 3 \exp (-(2-x_{2})^{2}-x_{1}^{2}), \end{aligned}$$\(L=(-3,-3)^{\top }\), and \(U=(3,3)^{\top }\).
-
5.
Stadler and J. Dauer (SD) [18]: \(n=4\), \(m=2\), convex,
$$\begin{aligned} F_{1}(x)&=2x_{1}+\sqrt{2}x_{2}+\sqrt{2}x_{3}+x_{4},\\ F_{2}(x)&=\frac{2}{x_{1}}+\frac{2\sqrt{2}}{x_{2}}+\frac{2\sqrt{2}}{x_{3}}+\frac{2}{x_{4}}, \end{aligned}$$\(L=\left( 1, \sqrt{2}, \sqrt{2}, 1\right) ^{\top }\), and \(U=(3, 3, 3, 3)^{\top }\).
-
6.
Zitzler, Deb and Thiele (ZDT1) [21]: \(n=30\), \(m=2\), convex,\(^2\)
$$\begin{aligned} F_{1}(x)&=x_{1},\\ F_{2}(x)&=g(x)\left( 1-\sqrt{\frac{x_{1}}{g(x)}}\right) , \end{aligned}$$with \(g(x)=1+9\sum _{i=2}^{n} x_{i}/(n-1)\), \(L=(0,\ldots , 0)^{\top }\), and \(U=\left( \frac{1}{100},\ldots , \frac{1}{100}\right) ^{\top }\).
-
7.
Zitzler, Deb and Thiele (ZDT4) [21]: \(n=10\), \(m=2\), nonconvex,\(^2\)
$$\begin{aligned} F_{1}(x)&=x_{1},\\ F_{2}(x)&=g(x)\left( 1-\sqrt{\frac{x_{1}}{g(x)}}\right) , \end{aligned}$$with \(g(x)=1+10(n-1)+\sum _{i=2}^{n}\left( x_{i}^{2}-10\cos (4\pi x_{i})\right) \), \(L=\left( \frac{1}{100}, -5,\ldots , -5\right) ^{\top }\), and \(U=(1, 5,\ldots , 5)^{\top }\).
-
8.
Toint (TOI4) [19, Problem 4]: \(n=4\), \(m=2\), convex,Footnote 3
$$\begin{aligned} F_1(x)&=x_1^2+x_2^2+1,\\ F_2(x)&=0.5\left( (x_1-x_2)^2+(x_3-x_4)^2\right) +1, \end{aligned}$$\(L=(-2,\ldots , -2)^{\top }\), and \(U=(5,\ldots , 5)^{\top }\).
-
9.
TRIDIA [19, Problem 8]: \(n=3\), \(m=3\), convex,\(^{22}\)
$$\begin{aligned} F_1(x)&=(2x_{1}-1)^{2}, \\ F_2(x)&=2(2x_{1}-x_{2})^{2}, \\ F_3(x)&=3(2x_{2}-x_{3})^{2}, \end{aligned}$$\(L=(-1, -1, -1)^{\top }\), and \(U=(1, 1, 1)^{\top }\).
-
10.
Shifted TRIDIA [19, Problem 9]: \(n=4\), \(m=4\), nonconvex,\(^{23}\)
$$\begin{aligned} F_{1}(x)&=(2x_{1}-1)^{2}+x_{2}^{2},\\ F_{i}(x)&=i(2x_{i-1}-x_{i})^{2}-(i-1)x_{i-1}^{2}+ix_{i}^{2} \quad i=2,3,\\ F_{4}(x)&=4(2x_{3}-x_{4})^{2}-3x_{3}^{2}, \end{aligned}$$\(L=(-1,\ldots , -1)^{\top }\), and \(U=(1,\ldots , 1)^{\top }\).
-
11.
Rosenbrock [19, Problem 10]: \(n=4\), \(m=3\), nonconvex,\(^{23}\)
$$\begin{aligned} F_{i}(x)&=100(x_{i+1}-x_{i}^{2})^{2}+(x_{i+1}-1)^{2}, \quad i=1,2,3, \end{aligned}$$\(L=(-2,\ldots , -2)^{\top }\), and \(U=(2,\ldots , 2)^{\top }\).
-
12.
Helical valley [14, Problem (7)]: \(n=3\), \(m=3\), nonconvex,\(^3\)
$$\begin{aligned} F_{1}(x)&=\left\{ \begin{array}{ll} \displaystyle { \left[ 10\left( x_3-\frac{5}{\pi }\arctan \left( \frac{x_2}{x_1}\right) \right) \right] ^2,} &{} \text{ if }\;x_1>0,\\ \displaystyle { \left[ 10\left( x_3-\frac{5}{\pi }\arctan \left( \frac{x_2}{x_1}\right) -5\right) \right] ^2,} &{} \text{ if }\;x_1<0 \end{array}\right. \\ F_{2}(x)&=\left( 10\left( (x_1^2+x_2^2)^{1/2}-1\right) \right) ^2,\\ F_{3}(x)&=x_3^2, \end{aligned}$$\(L=(-2, -2, -2)^{\top }\), and \(U=(2, 2, 2)^{\top }\).
-
13.
Gaussian [14, Problem (9)]: \(n=3\), \(m=15\), nonconvex,\(^3\)
$$\begin{aligned} F_{i}(x)&=x_1\exp \left( \frac{-x_2(t_i-x_3)^2}{2}\right) -y_i, \end{aligned}$$where \(t_i=(8-i)/2\), \(i=1,\ldots ,m\) and \(y_i\) is given as
i
1,15
2,14
3,13
4,12
5,11
6,10
7,9
8
\(y_i\)
0.0009
0.0044
0.0175
0.0540
0.1295
0.2420
0.3521
0.3989
\(L=(-2, -2, -2)^{\top }\), and \(U=(2, -2, 2)^{\top }\).
-
14.
Brown and Dennis [14, Problem (16)]: \(n=4\), \(m=5\), nonconvex,\(^3\)
$$\begin{aligned} F_i(x)&=\left( x_1+t_i x_2-e^{t_i}\right) ^2+\left( x_3+x_4\sin (t_i)-\cos (t_i)\right) ^2, \end{aligned}$$where \(t_i=i/5\), \(L=(-25, -5, -5, -1)^{\top }\), and \(U=(25, 5, 5, 1)^{\top }\).
-
15.
Trigonometric [14, Problem (26)]: \(n=4\), \(m=4\), nonconvex,\(^{23}\)
$$\begin{aligned} F_i(x)&=\left( n-\sum _{j=1}^{n}\cos x_j+i\left( 1-\cos x_i\right) -\sin x_i\right) ^2, \quad i=1,\ldots ,4, \end{aligned}$$\(L=(-1,\ldots , -1)^{\top }\), and \(U=(1,\ldots , 1)^{\top }\).
-
16.
Linear function – rank 1 [14, Problem (33)]: \(n=10\), \(m=4\), convex,\(^{23}\)
$$\begin{aligned} F_i(x)&=\left( i\left( \sum _{j=1}^{n}jx_j\right) -1\right) ^2, \quad i=1,\ldots ,4, \end{aligned}$$\(L=(-1,\ldots , -1)^{\top }\), and \(U=(1,\ldots , 1)^{\top }\).
Rights and permissions
About this article
Cite this article
Mita, K., Fukuda, E.H. & Yamashita, N. Nonmonotone line searches for unconstrained multiobjective optimization problems. J Glob Optim 75, 63–90 (2019). https://doi.org/10.1007/s10898-019-00802-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10898-019-00802-0