The category of 3-computads is not cartesian closed (Q942212)

The computads defined by the reviewer for strict \(n\)-categories were vastly generalized by \textit{M.~A. Batanin} [Contemp. Math. 230, 37--57 (1998; Zbl 0932.18004)] for a wide class of structures monadic over globular sets (= \(\omega\)-graphs). A 1-graph is just a (directed) graph; it is discrete when the source and target functions are identity functions. A 2-graph (or 2-globular set) is a graph in the category of 1-graphs whose vertex 1-graph is discrete; a 2-graph is discrete when all source and target functions are identities. Continuing, we obtain an inductive definition of \(n\)-graph. The category of \(n\)-graphs is a presheaf category and so a topos and so cartesian closed. A 1-computad is a 1-graph. Every category has an underlying 1-graph and that underlying functor has a left adjoint which assigns free categories to each 1-computad. A 2-computad is a 2-graph in which the underlying 1-graph underlies a free category on a 1-computad. Each (strict) 2-category has an underlying 2-computad and we have free 2-categories on 2-computads. A 3-computad whose underlying 2-computad underlies a free 2-category on a 2-computad. Batanin [loc. cit.], following \textit{A. Carboni} and \textit{P. Johnstone} [Math. Struct. Comput. Sci. 5, No. 4, 441--459 (1995; Zbl 0849.18002)], claimed the category of 3-computads was a topos. The present paper proves this false by showing the category of 3-computads is not even cartesian closed. The impediment has to do with commutativity under composition for 2-cells between identities; a familiar difficulty when looking at computads as higher rewriting systems. Batanin has shown however that if we look at 3-computads for weak 3-categories (rather than for strict 3-categories), a topos is obtained.