Note: this section is a work in progress!

[1] Logic programming has grown out of a long history of research in automatic theorem proving. Early theorem-proving programs could accomplish very little, because they exhaustively searched the space of possible proofs. The major breakthrough that made such a search plausible was the discovery in the early 1960s of the unification algorithm and the resolution principle (Robinson 1965). Resolution was used, for example, by Green and Raphael (1968) (see also Green 1969) as the basis for a deductive question-answering system. During most of this period, researchers concentrated on algorithms that are guaranteed to find a proof if one exists. Such algorithms were difficult to control and to direct toward a proof. Hewitt (1969) recognized the possibility of merging the control structure of a programming language with the operations of a logic-manipulation system, leading to the work in automatic search mentioned in section 4.3.1 (footnote 1). At the same time that this was being done, Colmerauer, in Marseille, was developing rule-based systems for manipulating natural language (see Colmerauer et al. 1973). He invented a programming language called Prolog for representing those rules. Kowalski (1973; 1979) in Edinburgh, recognized that execution of a Prolog program could be interpreted as proving theorems (using a proof technique called linear Horn-clause resolution). The merging of the last two strands led to the logic-programming movement. Thus, in assigning credit for the development of logic programming, the French can point to Prolog's genesis at the University of Marseille, while the British can highlight the work at the University of Edinburgh. According to people at MIT, logic programming was developed by these groups in an attempt to figure out what Hewitt was talking about in his brilliant but impenetrable Ph.D. thesis. For a history of logic programming, see Robinson 1983.
[2] To see the correspondence between the rules and the function, let x in the function (where x is nonempty) correspond to pair(u, v) in the rule. Then z in the rule corresponds to the append of tail(x) and y.
[3] This certainly does not relieve the user of the entire problem of how to compute the answer. There are many different mathematically equivalent sets of rules for formulating the append relation, only some of which can be turned into effective devices for computing in any direction. In addition, sometimes what is information gives no clue how to compute an answer. For example, consider the problem of computing the $y$ such that $y^2 = x$.
[4] Interest in logic programming peaked during the early 80s when the Japanese government began an ambitious project aimed at building superfast computers optimized to run logic programming languages. The speed of such computers was to be measured in LIPS (Logical Inferences Per Second) rather than the usual FLOPS (FLoating-point Operations Per Second). Although the project succeeded in developing hardware and software as originally planned, the international computer industry moved in a different direction. See Feigenbaum and Shrobe 1993 for an overview evaluation of the Japanese project. The logic programming community has also moved on to consider relational programming based on techniques other than simple pattern matching, such as the ability to deal with numerical constraints such as the ones illustrated in the constraint-propagation system of section 3.3.5.
4.4 Logic Programming