At present, we would say that CMSC is at a consolidation phase. Models of cement paste and concrete have been made, and many individual properties can be computed and compared to experiment. Much of the current state of affairs can be seen in the Electronic Monograph developed by us and others, which contains, in HTML book form, a record of most of the developments in CMSC over the last 11 years [24]. As part of this monograph, a "digital tool-kit" has been described, detailing what computational tools are available to operate on computer-based models of random materials [25]. The monograph also describes experimental materials science results obtained at NIST, many of which are strongly coupled to computational results.
There is increasing synergy between the experimental and computational aspects of the materials science of concrete in our group at NIST. This has been true for centuries in physics, but is only now being made use of in understanding the microstructure-property relationships in concrete. For example, the NIST 3-D cement hydration model (now called CEMHYD3D) is only able to accurately model microstructure development and properties because of the extreme care taken in generating representative, in terms of particle size and phase distribution, starting 3-D cement particle images. The basis for these starting images is an image segmentation algorithm applied to real experimental 2-D cement powder images and a correlated image generation algorithm originally developed in the 1970´s [26] .
Currently, the cement paste microstructure model is being improved and revised almost daily [27, 28]. This model, the concrete microstructure model, and many of the application programs have been documented and are freely available to any user [29]. More and more people, including industrial companies, are starting to use the software, at least partly due to the ACBM/NIST Computer Modelling Workshop, which is now in its 11 th year. Over 200 people from academia, industry, and government, including almost 100 students, have been introduced to computational materials science via this workshop, and we are now approaching a "critical mass" necessary for much more activity in this area to take place.
New uses have been proposed for the computational materials science tools that have been developed, with most involving combinations of tools to study phenomena that are more complicated than those studied to date. For example, studying the early-age cracking of high performance concrete involves first hydrating the cement paste and computing the self-desiccation in terms of which pores dry out first. Then elastic and viscoelastic computations must be made for stresses, allowing for capillary tension in the pore water and possible changes in the surface energy of the pore water as ions are dissolved in it. These stresses can in turn cause cracking, which can then alter the pore structure and thus alter the self-desiccation. A different combination of tools have been used to simulate and study the drying shrinkage of porous Vycor glass [30]. Other groups are also carrying on CMSC activities, mention of which are beyond the scope of this short paper.