Interior Point Methods Optimality conditions.pdf

Interior Point Methods Optimality conditions.pdf

  1. 1、本文档共29页,可阅读全部内容。
  2. 2、原创力文档(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
  3. 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载
  4. 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
查看更多
Interior Point Methods Optimality conditions

A. Grothey, J. Gondzio Direct Parallel Solution of Systems of Dimension 109 School of Mathematics T H E U N I V E R S I T Y O F E D I N B U R G H Direct Parallel Solution of Linear Systems of Dimension 109 Andreas Grothey, Jacek Gondzio SIAM PP 2006, San Francisco 1 A. Grothey, J. Gondzio Direct Parallel Solution of Systems of Dimension 109 OOPS (Object Oriented Parallel Solver) ? Mantra: “Truly large scale problems are not only sparse but structured” (due to e.g. dynamics, uncertainty, spatial distribution etc.) ? Exploiting structure is key to building efficient IPMs for large problems: – Faster linear algebra – Reduced memory use (by use of implicit factorization) – Possibility to exploit (massive) parallelism – We assume that structure is known! ? no automatic detection. ? OOPS is a general purpose (parallel) IPM solver – Not tuned to any particular hardware – Not tuned to any particular problem (structure). ? OOPS currently solves LP/QP problems. ? NLP extension solves nonlinear financial planning problems SIAM PP 2006, San Francisco 2 A. Grothey, J. Gondzio Direct Parallel Solution of Systems of Dimension 109 Interior Point Methods min cx + 1 2 xQx s.t. Ax = b x ≥ 0 (QP) Optimality conditions: c + Qx ? Ay ? z = 0 Ax = b XZe = 0 (μe) x, z ≥ 0 ? Newton Step:[ ?Q ? Θ A A 0 ] [ ?x ?y ] = [ r h ] := [ Qx + c ? Ay ? μX?1e b ? Ax ] (NS-QP) where Θ = X?1Z, X = diag(x), Z = diag(z) SIAM PP 2006, San Francisco 3 A. Grothey, J. Gondzio Direct Parallel Solution of Systems of Dimension 109 Linear Algebra of IPMs Main work: solve [ ?Q ? Θ A A 0 ] ︸ ︷︷ ︸ Φ [ ?x ?y ] = [ r h ] for several right-hand-sides at each iteration ? Two stage solution procedure ? factorize Φ = LDL ? backsolve(s) to compute direction (?x, ?y) + corrections ? Φ changes numerically but not structurally at each iteration Key to efficient implementation is exploiting structure of Φ in these two steps SIAM PP 2006, San Francisco 4 A. Grothey, J. Gondzio Direct Parallel Solution of Systems of Dimension 109

文档评论(0)

l215322 + 关注
实名认证
内容提供者

该用户很懒,什么也没介绍

1亿VIP精品文档

相关文档