An SDP approach for non-convex quadraticfractional programming problems
博士 === 國立成功大學 === 數學系應用數學碩博士班 === 103 === In this dissertation, we are concerned with two types of quadratic fractional programming problems: one is a single ratio quadratic fractional problem with a twosided quadratic constraint (P) and the other is a sum-of-ratios problem (Q). Although we pose the...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | en_US |
Published: |
2015
|
Online Access: | http://ndltd.ncl.edu.tw/handle/13113879053888063873 |
id |
ndltd-TW-103NCKU5507003 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-103NCKU55070032016-02-29T04:30:49Z http://ndltd.ncl.edu.tw/handle/13113879053888063873 An SDP approach for non-convex quadraticfractional programming problems An SDP approach for non-convex quadraticfractional programming problems Van-BongNguyen 阮文蓬 博士 國立成功大學 數學系應用數學碩博士班 103 In this dissertation, we are concerned with two types of quadratic fractional programming problems: one is a single ratio quadratic fractional problem with a twosided quadratic constraint (P) and the other is a sum-of-ratios problem (Q). Although we pose the problems (P) and (Q) in the form of ratios of quadratic functions, after some parameterization scheme, the fractional structure can be exchanged with a family of special types of quadratically constrained quadratic programming (QCQP) problems. The QCQP problem is a long-standing difficult non-convex optimization problem which has attracted much attention in literature. Therefore, in dealing with (P) and (Q), we not only face the traditional QCQP problem, but also an additional parameter coming from the ratio structure. Since we feel that we can address the two issues all together, we directly attack the quadratic fractional programming but apparently many results are new and useful to QCQP alone. Our main idea is to probe the hidden convexity of (P) and (Q), in spite that both (P) and (Q) are themselves non-convex. Our main tool is a powerful mechanism, known as the S-lemma, together with the semi-definite relaxation and the rank-one decomposition technique. In order to apply the aforementioned tools for solving (P) and (Q), we transform or divide the problems into a few subcases; study new versions of S-lemma; and modify the rank-one decomposition procedure. As a result, we prove that (P) is indeed in the class of P, and the NP-hard problem (Q) can be reduced to a much simpler one-dimensional search problem. Our solution method is novel, and can be easily implemented on the computer. Our computational results show that the algorithms are very efficient in finding the global optimal solution compared with other existing methods. Ruey-Lin Sheu 許瑞麟 2015 學位論文 ; thesis 98 en_US |
collection |
NDLTD |
language |
en_US |
format |
Others
|
sources |
NDLTD |
description |
博士 === 國立成功大學 === 數學系應用數學碩博士班 === 103 === In this dissertation, we are concerned with two types of quadratic fractional programming problems: one is a single ratio quadratic fractional problem with a twosided quadratic constraint (P) and the other is a sum-of-ratios problem (Q). Although we pose the problems (P) and (Q) in the form of ratios of quadratic functions, after some parameterization scheme, the fractional structure can be exchanged with a family of special types of quadratically constrained quadratic programming (QCQP) problems. The QCQP problem is a long-standing difficult non-convex optimization problem which has attracted much attention in literature. Therefore, in dealing with (P) and (Q), we not only face the traditional QCQP problem, but also an additional parameter coming from the ratio structure. Since we feel that we can address the two issues all together, we directly attack the quadratic fractional programming but apparently many results are new and useful to QCQP alone. Our main idea is to probe the hidden convexity of (P) and (Q), in spite that both (P) and (Q) are themselves non-convex. Our main tool is a powerful mechanism, known as the S-lemma, together with the semi-definite relaxation and the rank-one decomposition technique.
In order to apply the aforementioned tools for solving (P) and (Q), we transform or divide the problems into a few subcases; study new versions of S-lemma; and modify the rank-one decomposition procedure. As a result, we prove that (P) is indeed in the class of P, and the NP-hard problem (Q) can be reduced to a much simpler one-dimensional search problem. Our solution method is novel, and can be easily implemented on the computer. Our computational results show that the algorithms are very efficient in finding the global optimal solution compared with other existing methods.
|
author2 |
Ruey-Lin Sheu |
author_facet |
Ruey-Lin Sheu Van-BongNguyen 阮文蓬 |
author |
Van-BongNguyen 阮文蓬 |
spellingShingle |
Van-BongNguyen 阮文蓬 An SDP approach for non-convex quadraticfractional programming problems |
author_sort |
Van-BongNguyen |
title |
An SDP approach for non-convex quadraticfractional programming problems |
title_short |
An SDP approach for non-convex quadraticfractional programming problems |
title_full |
An SDP approach for non-convex quadraticfractional programming problems |
title_fullStr |
An SDP approach for non-convex quadraticfractional programming problems |
title_full_unstemmed |
An SDP approach for non-convex quadraticfractional programming problems |
title_sort |
sdp approach for non-convex quadraticfractional programming problems |
publishDate |
2015 |
url |
http://ndltd.ncl.edu.tw/handle/13113879053888063873 |
work_keys_str_mv |
AT vanbongnguyen ansdpapproachfornonconvexquadraticfractionalprogrammingproblems AT ruǎnwénpéng ansdpapproachfornonconvexquadraticfractionalprogrammingproblems AT vanbongnguyen sdpapproachfornonconvexquadraticfractionalprogrammingproblems AT ruǎnwénpéng sdpapproachfornonconvexquadraticfractionalprogrammingproblems |
_version_ |
1718196606695636992 |