Semidefinite Relaxation of Quadratic Optimization Problems
Summary (2 min read)
Introduction
- The promising empirical approximation performance of SDR has motivated new endeavors, leading to the creation of new research trends in some cases.
- The authors will also endeavor to touch on some advanced, key theoretical results by highlighting their practical impacts and implications.
III. APPLICATION: MIMO DETECTION
- Let us show an example of SDR application before proceeding to further advanced concepts and applications.
- The problem the authors consider is MIMO detection, a frequently encountered problem in digital communications.
- Sphere decoding has been found to be computationally fast for small to moderate problem sizes; e.g., N ≤ 20.
- The authors now show some simulation results to illustrate how well SDR performs in practice.
IV. RANDOMIZATION AND PROVABLE APPROXIMATION ACCURACIES
- The intuitive ideas behind randomization are not difficult to see, yet the theoretical implications that follow are far from trivial— many theoretical approximation accuracy results for SDRs are proven using randomization.
- Remarkably, one can see that there is a significant amount of x(ξ) that lie close to the optimal QCQP solutions.
- To give some flavor of these approximation accuracy results, let us first consider Problem (16).
- In addition, if the authors adopt the randomization procedure in Box 2, then the expected objective value of the randomized solution x̂ will satisfy γvQP ≤ E{v(x̂)} ≤ vQP (21) with the same constant γ.
V. EXTENSION TO MORE GENERAL CASES
- For ease of exposition of the SDR idea, the authors have only concentrated on the real-valued homogeneous QCQPs in previous sections.
- Readers are referred to [48], [64] for another interpretation of SDR in the inhomogeneous case.
VI. APPLICATION: TRANSMIT B1 SHIM IN MRI
- At this point readers may have the following concern: since SDR is an approximation method, as an alternative the authors may also choose to approximate a nonconvex QCQP by an available nonlinear programming method (NPM) (e.g., sequential quadratic programming, available in the MATLAB Optimization Toolbox).
- Thus, one can consider a two-stage approach, in which SDR is used to provide a starting point for an NLM.
- The application involved is transmit B1 shimming in magnetic resonance imaging (MRI) [37].
- Figs. 9(d) and (e) show the results for SDR randomized solutions, where the number of randomizations is L = 200.
VII. RANK REDUCTION IN SDP
- As the readers may have noticed by now, one of the recurring themes in the SDR methodology is the following.
- For a real-valued (resp. complex-valued) homogeneous QCQP with 2 (resp. 3) constraints or less, SDR is not just a relaxation, also known as To summarize.
- +nonlinear prog., realization 1 two-stage opt.: SDR w/ rand.
- If one allows the linear constraints in a given SDP to be satisfied only approximately, then it is possible to find a solution matrix whose rank is much smaller than O(√m).
VIII. APPLICATION: SENSOR NETWORK LOCALIZATION
- Let us now consider another practical problem to which the SDR technique can be applied, namely, the sensor network localization (SNL) problem.
- Under the above setting, their goal is to determine the coordinates of the sensors in, say R2, so that the distances induced 3This can be achieved using, e.g., the arrival time or difference in arrival time of the signal, the received signal strength, or angle of arrival measurements (see, e.g., [77], [78] and references therein).
- One can derive a computationally efficient SDR of Problem (33) as follows.
IX. CONCLUSION AND DISCUSSION
- Comprehensive coverage of the SDR technique, from its practical deployments and scope of applicability to key theoretical results.the authors.
- The authors have also showcased several representative applications, namely MIMO detection, B1 shimming in MRI and sensor network localization.
- Another important application, namely downlink transmit beamforming, is described in the article [1] in this special issue.
- Due to space limit, the authors are unable to cover many other beautiful applications of the SDR technique, although they have done their best to illustrate the key intuitive ideas that resulted in those applications.
- The authors hope that this introductory paper will serve as a good starting point for readers who would like to apply the SDR technique to their applications, and to locate specific references either in applications or theory.
Did you find this useful? Give us your feedback
Citations
1,079 citations
Cites background from "Semidefinite Relaxation of Quadrati..."
...It has been established that the SDR is tight when the number of constraints is less than three for a complex-valued homogeneous QCQP problem [51]....
[...]
796 citations
Additional excerpts
...5...
[...]
735 citations
Cites background or methods from "Semidefinite Relaxation of Quadrati..."
...If we use the element-wise sign function as mentioned in [35] and [37], the recovered vectors may not satisfy the constraints (3a)....
[...]
...As in [35], near optimality could be achieved at a small value of L in comparison with the size of the decision space....
[...]
...Following [35], F2 is transformed to an equivalent homogeneous quadratic constrained quadratic programming (QCQP)....
[...]
626 citations
607 citations
Additional excerpts
...[8], semidefinite relaxation based schemes [9]), admission...
[...]
References
7,655 citations
3,949 citations
"Semidefinite Relaxation of Quadrati..." refers background in this paper
...Readers are referred to [48], [64] for another interpretati on of SDR in the inhomogeneous case....
[...]
3,932 citations
"Semidefinite Relaxation of Quadrati..." refers background in this paper
...Since then, we have seen a number of dedicated theoretical analyse s that establish the SDR approximation accuracy under differ ent problem settings [3]–[11], and that have greatly improved o ur understanding of the capabilities of SDR....
[...]
...), Cij ≥ 0, ∀i 6= j Goemans-Williamson [3],...
[...]
...The idea of SDR can already be found in an early paper of Lovász in 1979 [2], but it was arguably the seminal work of Goemans and Williamson in 1995 [3] that sparked the significant interest in and rapid development of SDR techniques....
[...]
...In the seminal work of Goemans and Williamson [3], it is shown that whenCij ≤ 0 for all i 6= j, one has...
[...]
...[3] M. X. Goemans and D. P. Williamson, “Improved approximation algorithms for maximum cut and satisfiability problem using semi-definite programming,”J. ACM, vol. 42, pp. 1115–1145, 1995....
[...]
2,202 citations
1,733 citations
"Semidefinite Relaxation of Quadrati..." refers background in this paper
...The idea of SDR can already be found in an early paper of Lovász in 1979 [2], but it was arguably the seminal work of Goemans and Williamson in 1995 [3] that sparked the significant interest in and rapid development of SDR techniques....
[...]
Related Papers (5)
Frequently Asked Questions (12)
Q2. What is the goal of transmit b1 shimming?
The goal of transmit B1 shimming is to design the transmit amplitudes and phases of the RF coils such that the resultant B1 map (or the MR image) is as uniform as possible.
Q3. What are the current applications of complex-valued SDR?
The current applications of complex-valued SDR lie in various kinds of beamforming problems [1], [15], [16], [27], [35], [37], [61].
Q4. What is the relevant application of a separable QCQP?
A relevant application for separable QCQPs is the unicast downlink transmit beamforming problem [65]; see [1] in this special issue for the problem description.
Q5. How can one find a solution matrix with a rank of m?
if one allows the linear constraints ina given SDP to be satisfied only approximately, then it is possible to find a solution matrix whose rank is much smaller than O(√m).
Q6. What is the transmit vector of the RF coil array?
Cn be the transmit vector of the RF coil array, where n is the number of RF coils and xi is a complex variable characterizing the transmit amplitude and phase of the ith RF coil.
Q7. What is the first step in deriving an SDR of Problem (4)?
Sn denotes the set of all real symmetric n×n matrices; and b1, . . . , bm ∈ R. A crucial first step in deriving an SDR of Problem (4) is to observe thatxTCx = Tr(xTCx) = Tr(CxxT ),xTAix = Tr(x TAix) = Tr(Aixx T ).
Q8. how many constraints are there in a QCQP?
Most of these approaches rely on so-called rank-one decomposition theorems, which allow one to extract an optimal QCQP solution from the SDR solution, provided that the number of constraints in the QCQP is not too large—say, at most 3 for the complex-valued homogeneous QCQP.
Q9. How many iterations of the gradient search procedure?
The initial positions of the sensors as computed by the SDP are denoted by stars ‘∗’, and the tail end of a trajectory gives the computed position of a sensor after 50 iterations of the gradient search procedure.
Q10. What is the definition of a separable QCQP?
C. Separable QCQPs: Consider a QCQP of the formmin x1,...,xk∈Cn∑k i=1 x H i Cixis.t. ∑k l=1 x H l Ai,lxl Di bi, i = 1, . . . ,m.(27) Problem (27) is called a separable QCQP.
Q11. What are the applications of the aforementioned tightness results?
The authors note that the aforementioned tightness results have already found many applications in signal processing and communications; see, e.g., [32], [33], [40], [41], [71]–[75].
Q12. How many iterations of the gradient search procedure are still nowhere near the true positions?
As can be seen from the figure, even after 50 iterations, the computed positions of the sensors are still nowhere close to the true positions.