1: \begin{abstract}

2: We define the distance between two information structures as the largest

3: possible difference in value across all zero-sum games. We provide

4: a tractable characterization of distance and use it to discuss the

5: relation between the value of information in games versus single-agent

6: problems, the value of additional information, informational substitutes,

7: complements, or joint information. The convergence to a countable

8: information structure under value-based distance is equivalent to

9: the weak convergence of belief hierarchies, implying, among other

10: things, that for zero-sum games, approximate knowledge is equivalent

11: to common knowledge. At the same time, the space of information structures

12: under the value-based distance is large: there exists a sequence of

13: information structures where players acquire increasingly more information,

14: and $\varepsilon>0$ such that any two elements of the sequence have

15: distance of at least $\varepsilon$. This result answers by the negative

16: the second (and last unsolved) of the three problems posed by J.F.

17: Mertens in his paper ``Repeated Games'', ICM 1986. 

18: \end{abstract}

19: