We consider two versions of the problem of testing graph isomorphism in the bounded-degree graph model: A version in which one graph is fixed, and a version in which the input consists of two graphs. We essentially determine the query complexity of these testing problems in the special case of n-vertex graphs with connected components of \(\textrm{poly}(\log n)\) size. This is done by showing that these problems are computationally equivalent (up to polylogarithmic factors) to corresponding problems regarding isomorphism between sequences (over a large alphabet). Ignoring the dependence on the proximity parameter, our main results are: Testing isomorphism between two sequences is shown to be related to testing that two distributions are identical, and this relation yields reductions in three of the four relevant cases. Failing to reduce the problem of testing the equality of two input distributions to the problem of testing isomorphism between two input sequences, we adapt the proof of the lower bound on the complexity of the first problem to the second problem. This adaptation constitutes the main technical contribution of the current work. We stress that determining the complexity of testing graph isomorphism (in the bounded-degree graph model), in the general case (e.g., for expander graphs), is left open.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Testing Isomorphism in the Bounded-Degree Graph Model

  • Oded Goldreich

摘要

We consider two versions of the problem of testing graph isomorphism in the bounded-degree graph model: A version in which one graph is fixed, and a version in which the input consists of two graphs. We essentially determine the query complexity of these testing problems in the special case of n-vertex graphs with connected components of \(\textrm{poly}(\log n)\) size. This is done by showing that these problems are computationally equivalent (up to polylogarithmic factors) to corresponding problems regarding isomorphism between sequences (over a large alphabet). Ignoring the dependence on the proximity parameter, our main results are: Testing isomorphism between two sequences is shown to be related to testing that two distributions are identical, and this relation yields reductions in three of the four relevant cases. Failing to reduce the problem of testing the equality of two input distributions to the problem of testing isomorphism between two input sequences, we adapt the proof of the lower bound on the complexity of the first problem to the second problem. This adaptation constitutes the main technical contribution of the current work. We stress that determining the complexity of testing graph isomorphism (in the bounded-degree graph model), in the general case (e.g., for expander graphs), is left open.