This paper models knowledge di usion as a barter process in which agents exchange di erent types of knowledge. This is intended to capture the observed practice of informal knowledge trading. Agents are located on a network and are directly connected with a small number of other agents. Agents repeatedly meet those with whom direct connections exist and trade if mutually proÿtable trades exist. In this way knowledge di uses throughout the economy. We examine the relationship between network architecture and di usion performance. We consider the space of structures that fall between, at one extreme, a network in which every agent is connected to n nearest neighbours, and at the other extreme a network with each agent being connected to, on average, n randomly chosen agents. We ÿnd that the performance of the system exhibits clear 'small world' properties, in that the steady-state level of average knowledge is maximal when the structure is a small world (that is, when most connections are local, but roughly 10 percent of them are long distance). The variance of knowledge levels among agents is maximal in the small world region, whereas the coe cient of variation is minimal. We explain these results as re ecting the dynamics of knowledge transmission as a ected by the architecture of connections among agents. ?

In this paper, we model the formation of innovation networks as they emerge from bilateral decisions. In contrast to much of the literature, here firms only consider knowledge production, and not network issues, when deciding on partners. Thus, we focus attention on the effects of the knowledge and information regime on network formation. The effectiveness of a bilateral collaboration is determined by cognitive, relational, and structural embeddedness. Innovation results from the recombination of knowledge held by the partners to the collaboration, and its success is determined in part by the extent to which firms' knowledge complement each other. Previous collaborations (relational embeddedness) increase the probability of a successful collaboration, as does information gained from common third parties (structural embeddedness). Repeated alliance formation creates a network. Two features are central to the innovation process: how firms pool their knowledge resources, and how firms derive information about potential partners. When innovation is decomposable into separate subtasks, networks tend to be dense; when structural embeddedness is important, networks become cliquish. For some regions in this parameter space, small worlds emerge.networks, innovation, knowledge, collaborative R& D, embeddedness