The Ginibre inequality

Abstract

In the Ising-type models of statistical mechanics and the related quantum field theories, an inequality of Ginibre implies useful positivity and monotonicity properties: the Griffiths correlation inequalities. Essentially, the Ginibre inequality states that certain functions on the cycle group of a graph are positive definite. This has been proved for arbitrary graphs when the spin dimension is 1 or 2 (classical Ising or plane rotator models). We give a counterexample to show that these spin dimensions are the only ones for which the Ginibre inequality is generally true: there are graphs for which it never holds when the spin dimension is at least 3. On the other hand, we show that for any graph the inequality holds for the apparent leading term in the largespin-dimension limit. (The leading term vanishes in the graph of the counterexample.) Based on these results, one expects the Ginibre inequality to be true in most instances, with infrequent exceptions. A numerical survey supports this. The surprising failure of the Ginibre inequality in higher dimensions need not necessarily mean the Griffiths inequalities fail as well, but a different approach to them is required.