Minimizing Effective Resistance of a Graph
read more
Citations
Using circuit theory to model connectivity in ecology, evolution, and conservation.
DropEdge: Towards Deep Graph Convolutional Networks on Node Classification.
Kron Reduction of Graphs with Applications to Electrical Networks
Design of Optimal Sparse Feedback Gains via the Alternating Direction Method of Multipliers
Chemical Graph Theory
References
Convex Optimization
Using SeDuMi 1.02, a MATLAB toolbox for optimization over symmetric cones
Topics in Matrix Analysis
Algebraic connectivity of graphs
Fast linear iterations for distributed averaging
Related Papers (5)
Frequently Asked Questions (14)
Q2. What is the optimal duality gap for the ERMP?
Since the ERMP is convex, has only linear equality and inequality constraints, and Slater’s condition is satisfied (for example, by g = (1/m)1), the authors know that the optimal duality gap for the ERMP (31) and the dual problem (32) is zero.
Q3. What is the effective resistance between two nodes of a circuit?
(The effective resistance between two nodes of a circuit is defined as the ratio of voltage across the nodes to the current flow injected into them.)
Q4. What is the condition for the optimality of a convex problem?
Since the ERMP is a convex problem with differentiable objective, a necessary and sufficient condition for optimality of a feasible g is∇RTtot(ĝ − g) ≥ 0 for all ĝ with 1T ĝ = 1, ĝ ≥ 0Copyright © by SIAM.
Q5. What is the way to determine the optimal value of R!tot?
Since R!tot cannot decrease as edges are removed from the graph, the authors conclude that the largest value of R!tot among graphs with n nodes is obtained for a tree.
Q6. How do the authors form the reduced conductance matrix G?
The authors form the reduced conductance matrix G̃ by removing, say, the kth row and column of G. Let ṽ, ẽi, and ẽj be, respectively, the vectors v, ei, and ej , each with the kth component removed.
Q7. How many steps did the authors need to take to converge?
Using β = 1 and relative tolerance * = 0.001, the authors have found the algorithm to be very effective, never requiring more than 20 or so steps to converge for the many graphs the authors tried.
Q8. What is the main computational effort in computing the Newton step?
The main computational effort is in computing the Newton step (i.e., step 2), which requires O(m3) arithmetic operations if no structure in the equations is exploited.
Q9. What is the optimality condition for a circuit driven by a random current?
The authors suppose the circuit is driven by a random current excitation J with zero mean and covariance EJJT = The author−11T /n. By (25), the authors have ∂Rtot/∂gl = −nE v2l , where vl is the (random) voltage appearing across edge l.
Q10. What is the optimality condition for this problem?
The optimality conditions for this problem are simply that 1T g = 1 (feasibility) and that all components of the gradient, ∇Rtot, are equal; specifically, ∇Rtot(g) = −Rtot1.Acknowledgment.
Q11. What is the definition of a graph that is edge-transitive?
(For more on symmetries in convex optimization in general, see, e.g., [7, Ex. 4.4]; for more in the context of graph optimization problems, see [5, 18].)A graph is edge-transitive if all pairs of edges are symmetric, i.e., for any two edges, there is an automorphism of the graph that maps one edge to the other.
Q12. What is the formula for the total effective resistance?
The total effective resistance can also be expressed in terms of the reduced conductance matrix G̃. Multiplying (12) on the left and right by 1T and 1 and dividing by 2, the authors haveRtot = nTr G̃−1 − 1T G̃−11 = nTr(I − 11T /n)G̃−1.(16)(Note that G̃ ∈ R(n−1)×(n−1), so the vectors denoted 1 in this formula have dimension n − 1.)The total effective resistance can also be written in terms of an integral:Rtot = nTr ∫ ∞0 (e−tG − 11T /n) dt.(17)This can be seen as follows.
Q13. What is the effective resistance between nodes i and j?
In this case the effective resistance between nodes i and j can be expressed asRij = ∑ 1gl ,where the sum is over edges that lie on the (unique) path between i and j.
Q14. How does the algorithm differ from the standard barrier method?
This algorithm differs from the standard barrier method in [7, section 11.3] in two ways: the exit condition uses the duality gap η̂ from (35), and the parameter t is updated in every step of the interior-point method using η̂.