沈 潔,李 軒,李 娜
(遼寧師范大學(xué) 數(shù)學(xué)學(xué)院,遼寧 大連 116029)
關(guān)于雙穩(wěn)定束方法對(duì)偶問(wèn)題的研究
沈 潔,李 軒,李 娜
(遼寧師范大學(xué) 數(shù)學(xué)學(xué)院,遼寧 大連 116029)
對(duì)于帶有非線性約束優(yōu)化問(wèn)題,本文在迫近束方法的思想基礎(chǔ)上將水平束方法與其結(jié)合,應(yīng)用雙穩(wěn)定束方法解決此優(yōu)化問(wèn)題.本文不僅從其對(duì)偶問(wèn)題的角度研究了解的形式及相關(guān)性質(zhì),發(fā)現(xiàn)解的表現(xiàn)形式不盡相同,而且得出該解與之前迭代點(diǎn)的次梯度的凸組合有關(guān)的結(jié)論.進(jìn)一步我們發(fā)現(xiàn)次梯度值和額定下降具有與單純用迫近束方法從對(duì)偶問(wèn)題角度解無(wú)約束優(yōu)化問(wèn)題相類似性質(zhì).
雙穩(wěn)定束方法;約束問(wèn)題;對(duì)偶問(wèn)題;次梯度
非光滑優(yōu)化問(wèn)題在眾多領(lǐng)域有廣泛應(yīng)用,前人在理論與算法方面都進(jìn)行了廣泛的研究,見(jiàn)[1-8]。文獻(xiàn)[9]從對(duì)偶空間出發(fā)研究了無(wú)約束懲罰子問(wèn)題
的解的表達(dá)形式及其相關(guān)性質(zhì).文獻(xiàn)[10]中利用束方法研究了非光滑約束優(yōu)化問(wèn)題
其中f:Rn→R是非光滑凸函數(shù),χ∈Rn是一個(gè)非空的凸閉集(典型的多面體).文獻(xiàn)[10]中,作者將分別利用迫近束方法和水平束方法構(gòu)造的目標(biāo)函數(shù)的近似模型進(jìn)行比較,結(jié)合對(duì)校正不同參數(shù)方法的難易的討論以及最終在一定條件下解的一致性,作者將兩種束方法結(jié)合產(chǎn)生新的束方法,即雙穩(wěn)定束方法.在求解過(guò)程中,[10]將對(duì)原問(wèn)題的求解轉(zhuǎn)換成對(duì)一系列帶有水平約束的二次規(guī)劃問(wèn)題的求解:
(1.1)
等價(jià)于
(1.2)
(2.1)
(2.2)
(2.3)
且有以下關(guān)系式成立:
證明:?jiǎn)栴}(1.2)可寫(xiě)成關(guān)于參變量r,lk的二次規(guī)劃問(wèn)題,
(2.4)
其中
考慮后面的對(duì)偶問(wèn)題,對(duì)每一個(gè)固定的μ,定義x(μ,λ)=argminxL(x,μ,λ),它的最優(yōu)性條件為
(2.5)
(2.6)
接著證明(ii),首先注意到,由于不存在對(duì)偶間隙,原始問(wèn)題(1.2)的最優(yōu)值等于對(duì)偶問(wèn)題(2.6)的最優(yōu)值:
(2.7)
為了證明(2.3)式,注意到對(duì)于所有的x∈χ有下式成立:
[1]朱 宏.森林救火的優(yōu)化模型[J].吉林師范大學(xué)學(xué)報(bào)(自然科學(xué)版), 2004, 1: 96~97.
[2]方石銀.步進(jìn)電動(dòng)機(jī)起動(dòng)過(guò)程中脈沖頻率的優(yōu)化控制[J].吉林師范大學(xué)學(xué)報(bào)(自然科學(xué)版),2008,29(4):35~37.
[3]J.D.Buys.Dual algorithms for constrained optimization.PhD thesis[M].Rijksuniversiteit te Leiden, Leiden, The Netherlands,1972.
[4]景書(shū)杰, 曹香蓮.等式約束廣義幾何規(guī)劃的一種優(yōu)化算法[J].吉林師范大學(xué)學(xué)報(bào)(自然科學(xué)版), 2008,29(3):51~53.
[5]J.E.Dennis and R.B.Schnabel.Numerical Methods for Unconstrained Optimization and Nonlinear Equations[M].Prentice-Hall,Inc.,Englewood Cliffs,NJ,1983.
[6]費(fèi)紹金.眼科病床合理安排的優(yōu)化模型[J].吉林師范大學(xué)學(xué)報(bào)(自然科學(xué)版),2011,32(4):85~91.
[7]武慧虹, 錢(qián)淑渠,李 俊.動(dòng)態(tài)環(huán)境優(yōu)化問(wèn)題及算法綜述[J].吉林師范大學(xué)學(xué)報(bào)(自然科學(xué)版), 2013,34(2):121~124.
[8]C.Sagastizabal.Composite proximal bundle method[J].Mathematical Programming.(2012).10.1007/s10107-012-0600-5.
[9]J.Frédéric Bonnans,J.Charles Gilbert,L.Claude,et al.Bundle Methods.The Quest for Descent Numerical Optimization,2003,Part II:150~152.
[10]W.Oliveira,M.Solodov.A Doubly Stabilized Bundle Method For Nonsmooth Convex Optimization[OL].www.optimization-online.April 2013.
ResearchontheDualProblemofDoublyStabilizedBundleMethodForConvexOptimization
SHENJie,LIXuan,LINa
(School of Mathematics,Liaoning Normal University,Dalian 116029,China)
For nonlinear constraint optimization problem, we try to solve it using doubly stabilized bundle method by combining the proximal bundle method with the level bundle method.This paper studies the form and the corresponding properties of the solution of subproblem from the viewpoint of the dual problem, finds that the forms of the solution are not similar, and comes to the conclusion that the solution is related to the convex combinations of the subgradient of the previous iteration.Furthermore, we find that the subgradients and the prediction descent have similar properties with the case in which we simply use the proximal bundle method to solve the unconstrained optimization problems from the point of view of dual problem.
doubly stabilized bundle method;constraint optimization;duality problem;subgradient
梁懷學(xué))
2014-04-26
國(guó)家自然科學(xué)基金項(xiàng)目(11301246,11171138)
沈 潔(1973-),女,遼寧省沈陽(yáng)市人,現(xiàn)為遼寧師范大學(xué)副教授,博士,碩士生導(dǎo)師.研究方向:運(yùn)籌學(xué)與控制論.
O221.2
A
1674-3873-(2014)03-0064-04