Dear Janko,
Thanks for the comments. Regarding your interpretation of special relativity, I agree with your conclusion that "spacetime" does not exist in the absence of "matter-energy," and my approach says the same thing, but in my approach this is a direct consequence of a hypothesis about what "spacetime" and "matter-energy" really are at the fundamental scale (the causal-metric hypothesis), not deduced from an argument on the basis of an existing theory (such as your argument involving the trolley and the rocket in special relativity). Arguments like yours are encouraging for such a hypothesis because they reach downward from existing theory toward a more fundamental theory, while my ideas are trying to reach upward to recover existing theory in some approximation.
The strong form of the causal metric hypothesis treats what we call "spacetime" and "matter-energy" as manifestations of causal structure. Neither has an independent meaning at the fundamental scale; they emerge together. Other "causal theories" (such as causal set theory and causal dynamical triangulations) sometimes impose "matter fields" on causal structure, essentially by adding "weights" to the edges or vertices of a causal graph (the weights might be numbers or various types of mathematical objects such as spinors or elements of SU(2)). There might be some use in trying this, but I would prefer to try to use only the graph-theoretic structure.
This might seem to contain too little information to describe things like particles in the standard model, but I think there might be enough information present after all. Metric recovery theorems show that causal structure can give a metric up to a conformal factor, and this can be supplied by appropriate volume information, which can also be associated with the causal graph by an appropriate rule (Sorkin uses the simplest possible rule: a constant discrete metric in his "order plus number equals geometry" motto in causal set theory).
Many of the properties of the particles of standard model come from the representation theory of the Poincare group of symmetries of Minkowski spacetime, so already there is enough information in causal graphs to recover these properties. But there is more information: not only are other metrics recoverable as well, but the microstructure contains local information which has not yet been used. Also, the causal graphs I use are more general than causal sets; for instance, a given causal set corresponds to an entire equivalence class of "degenerate" non-transitive causal graphs, each of which specify information that is lost when taking the transitive closure.
At the most naive intuitive level, the reason why no spacetime exists in the absence of energy is because energy involves causal relations; i.e., "things are happening" in an energetic system. As the energy goes to zero, all interactions (relations) cease; there is no causal structure because "nothing is being caused." Hence, there is no emergent geometry and no spacetime.
Bear in mind, though, that all this is based on the hypothesis that spacetime is only a way of talking about causal structure. If this hypothesis is wrong (e.g. if there is a "background manifold") then this whole idea falls through. Take care,
Ben