We consider phase transitions in random graphs with constant average degree, like the sudden appearance of the giant component or the giant k-core [2] at some critical average degree. Small neighbourhoods in such random graphs closely resemble branching trees. Frequently, the exact numerical values of the critical average degrees and the expected sizes of the aforementioned giant subgraphs can be `predicted' in a semi-heuristic manner from studying `corresponding' branching trees instead, which is far simpler than the rigorous analysis of the phase transition in the random graphs. It is a major goal to turn this observation, which - following [2] - we shall call the `Branching Tree Connection', into a rigorous proof technique. Goerdt and Molloy have achieved this for the k-core in the model of random faulty configurations in [1]. We prove the sudden appearance of a giant subgraph that is `almost the k-core', its size being sharply concentrated around what is predicted by the Branching Tree Connection. Our proof, based on the Principle of Deferred Decisions and a new application of the Simple Concentration Bound, essentially employs the same recursive equations as used for analysing a `corresponding' phase transition in branching trees, thus providing structural insight into why the Branching Tree Connection predicts the correct values. Adapting ideas from [1], we show that the aforementioned subgraph which is `almost the k-core' can be `purged', yielding a giant k-core upon deletion of only o(n) nodes. Motivated by a new phase transition phenomenon in branching trees we have found a novel subgraph of k-partite graphs, the magic subgraph. We can empirically demonstrate its sudden appearance. Both critical average degree and size are in good accordance with the Branching Tree Connection. The fact that it appears at a critical average degree of 4.91... suggests a relation with the Antivoter Phenomenon~([3]). Moreover, when it appears, the magic subgraph seems to be `almost uniquely k-colourable', which may turn out to be relevant for understanding the threshold for k-colourability in non-k-partite random graphs with constant average degree. [1] Goerdt/Molloy, Analysis of edge deletion processes on faulty random regular graphs, Proceedings Latin 2000, pp. 38-47, 2000. [2] Pittel/Spencer/Wormald, Sudden emergence of a giant k-core in a random graph, Journal of Combinatorial Theory Series B (67), pp. 111-151, 1996. [3] Petford/Welsh, A randomised 3-colouring algorithm, Discrete Mathematics (74), pp. 253-261, 1989.
«
We consider phase transitions in random graphs with constant average degree, like the sudden appearance of the giant component or the giant k-core [2] at some critical average degree. Small neighbourhoods in such random graphs closely resemble branching trees. Frequently, the exact numerical values of the critical average degrees and the expected sizes of the aforementioned giant subgraphs can be `predicted' in a semi-heuristic manner from studying `corresponding' branching trees instead, which...
»