Distributed constrained convex optimization and consensus via dual decomposition and proximal minimization
Summary (2 min read)
Introduction
- Optimization in multi-agent networks has attracted significant interest from both the control and the operations research community, and has already found numerous applications in different domains, like power systems [1], [2], wireless networks [3], [4], robotics [5], etc.
- They then exchange the outcome of this computation (but not their private information) with neighboring agents, and the process is then repeated on the basis of the information received.
- Applying the methodologies of the aforementioned references to this problem, though possible, would unnecessarily increase the computational and communication effort, since it would require each agent to maintain an estimate of the decision vectors of all other agents when solving its local optimization program, and to communicate it to its neighboring agents.
- In particular, the contributions of their paper can be summarized as follows: 1) We extend dual decomposition based algorithms to a distributed setting, accounting for possibly time-varying network connectivity.the authors.the authors.
A. Problem statement and proposed solution
- Moreover, agent i may not be willing to share information about fi(·), Xi, and gi(·), with other agents, due to privacy issues.
- To account for information privacy and facilitate the development of a computationally tractable solution, the authors seek for a distributed strategy.
- In principle, (6) fits the framework of algorithms like [7], [10], the concave function ϕi(·) is implicitly defined through an optimization problem parametric in λ.
- In particular, the update of the local primal vector xi(k + 1) (step 7) is the same as in dual decomposition, whereas, in contrast to dual decomposition, the update of the dual vector (step 8) involves also a proximal term, which facilitates consensus among the agents.
B. Structural and communication assumptions
- The authors then impose the following connectivity and communication assumption.
- The graph (V,E∞) is strongly connected, i.e., for any two nodes there exists a path of directed edges that connects them.
C. Statement of the main results
- Under Assumptions 1-6, Algorithm 1 converges and agents reach consensus to a common vector of Lagrange multipliers.
- In particular, their local estimates λi(k) converge to some optimal dual solution, while the vector x̂(k) = [x̂.
- This is formally stated in the following theorems.
- Theorem 1. [Dual Optimality] Consider Assumptions 1-6.
D. Sketch of the proof
- The proofs of Theorems 1 and 2 are quite technical and require the derivation of several intermediate results, therefore they are omitted in the interest of space.
- In the following the authors provide a sketch of the main idea behind their proofs, while for more details the interested reader is referred to [23].
- This is established by showing that the sequence {λi(k)}k≥0 achieves the optimal value of the dual function across a subsequence, and relies on Proposition 4 of [7].
- Finally, the proof of Theorem 2 follows from [24], extending its derivations to deal with the considered distributed context.
III. NUMERICAL EXAMPLE
- For the sake of simplicity the authors assumed that the network does not change across iterations.
- Since problem (13) has a unique coupling constraint, there is just one Lagrange multiplier λ ∈ R+.
- The authors ran the Algorithm 1 for 1000 iterations.
- By inspection of Figure 2, the average converge quite fast to the optimal Lagrange multipliers of (13) (red triangles), whereas all agents gradually reach consensus on those values.
- Figure 3 shows the evolution of the primal objective value∑m i=1 fi(xi) (upper plot), and constraint violation in terms of ‖ ∑m i=1 gi(xi)‖∞ (lower plot), where xi is replaced by two different sequences: xi(k) (blue solid lines), and x̂i(k) (orange dashed lines), where the latter is given by (7).
IV. CONCLUDING REMARKS
- A novel distributed algorithm to deal with a class of convex optimization programs that exhibit a separable structure was developed.
- The authors considered an iterative scheme based on a combination of dual decomposition and proximal minimization, and they showed that this scheme converges to some optimal dual solution of the centralized problem counterpart, while the primal iterates generated by the algorithm converge to the set of optimal primal solutions.
- Current work concentrates on a convergence rate analysis and further comparison with gradient/subgradient methods.
- Moreover, the authors aim at relaxing the convexity assumption by extending the results of [29] to a distributed set-up, quantifying the duality gap incurred in case of mixed-integer programs.
- From an application point of view, the main focus is on applying the proposed algorithm to the problem of energy efficient control of a building network [30].
Did you find this useful? Give us your feedback
Citations
126 citations
Cites background from "Distributed constrained convex opti..."
...A preliminary version of this work is given in [10]....
[...]
40 citations
33 citations
5 citations
5 citations
References
17,433 citations
"Distributed constrained convex opti..." refers methods in this paper
...To exploit the particular problem structure and alleviate these difficulties, dual decomposition techniques (see [11], and references therein), or approaches based on the alternating direction method of multipliers [12], are often employed, relying on the separable structure of the problem after dualizing the coupling constraint....
[...]
3,238 citations
"Distributed constrained convex opti..." refers background in this paper
...In particular, in [6], [7], [8], [9] a gradient/subgradient based consensus approach is followed to address problems where agents with their own objective functions and constraints are coupled via a common decision vector....
[...]
...For details about the interpretation of Assumptions 5 and 6, the reader is referred to [6], [10], [7]....
[...]
1,773 citations
1,761 citations
Related Papers (5)
Frequently Asked Questions (7)
Q2. what is the weight of agent i to the solution of agent j?
Coefficient aij(k) is the weight that agent i attributes to the solution of agent j at iteration k; aij(k) = 0 means that agent j does not send any information to agent i at iteration k.
Q3. What is the vector of ni decision variables of agent i?
Consider a time-varying network of m agents that communicate to solve the following optimization programP : min {xi∈Xi}mi=1 m∑ i=1 fi(xi)subject to: m∑ i=1 gi(xi) ≤ 0, (1)where for each i = 1, . . . ,m, xi ∈ Rni is the vector of ni decision variables of agent i, fi(·) : Rni →
Q4. Why is the objective and constraint functions in P separable?
Notice that, due to the separable structure of the objective and the constraint functions in P ,ϕ(λ) = m∑ i=1 ϕi(λ) = m∑ i=1 min xi∈Xi Li(xi, λ).
Q5. What is the way to define gi?
It is perhaps worth noticing that this setup comprises also equality coupling constraints like∑mi=1 g̃i(xi) = 0. To this purpose it is enough to define gi = [g̃ > i −g̃>i ]>.
Q6. What is the difference between a primal and a dual update step?
Such an auxiliary sequence is referred to as primal recovery procedure and it is often used in dual decomposition methods, since it has better convergence properties compared to {xi(k)}k≥0 [24], [22], [21].
Q7. What is the purpose of this article?
To account for information privacy and facilitate the development of a computationally tractable solution, the authors seek for a distributed strategy.