Term graphs are a natural generalization of terms in which structure sharing is allowed. Structure sharing makes term graph rewriting a time- and space-efficient method for implementing term rewrite systems. Certain structure sharing schemes can lead to a situation in which a term graph component is rewritten to another component that contains the original. This phenomenon, called redex capturing, introduces cycles into the term graph which is being rewritten—even when the graph and the rule themselves do not contain cycles. In some applications, redex capturing is undesirable, such as in contexts where garbage collectors require that graphs be acyclic. In other applications, for example in the use of the fixed-point combinator Y, redex capturing acts as a rewriting optimization. We show, using results about infinite rewritings of trees, that term graph rewriting with arbitrary structure sharing (including redex capturing) is sound for left-linear term rewrite systems.