A.: Ideas for a high-level proof strategy language

ABSTRACT Finding ways to prove theorems mechanically was one of the earliest challenges tackled by the AI community. Notable progress has been made but there is still always a limit to any set of heuristic search techniques. From a proof done by human users, we wish to find out whether AI techniques can also be used to learn from a human user. AI4FM (Artificial Intelligence for Formal Methods) is a four-year project that starts officially in April 2010 (see www.AI4FM.org). It focuses on helping users of "formal methods" many of which give rise to proof obligations that have to be (mechanically) verified (by a theorem prover). In industrial-sized developments, there are often a large number of proof obligations and, whilst many of them succumb to similar proof strategies, those that remain can hold up engineers trying to use formal methods. The goal of AI4FM is to learn enough from one manual proof, to discharge proof obligations automatically that yield to similar proof strategies. To achieve this, a high-level (proof) strategy language is required, and in this paper we outline some ideas of such language, and towards extracting them. * During this work Gudmund Grov has been employed jointly by University of Edinburgh and Newcastle University. and constrained use of Z [FW08] -is the so-called "posit and prove" approach: a designer posits development steps and then justifies that they satisfy earlier specifications by discharging (often automatically generated) proof obligations (POs). A large proportion of these POs can be discharged by automatic theorem provers but "some" proofs require user interaction. Quantifying "some" is hard since it depends on many factors such as the domain, technology and methodology used -it could be as little as 3% or as much as 40%. For example, the Paris Metro line 14, developed in the Bmethod, generated 27, 800 POs (of which around 2, 250 required user-interaction) [Abr07] -the need for interactive proofs is clearly still a bottleneck in industrial application of FM, notwithstanding high degree of automation. THE FORMAL METHODS PROBLE