Computational Processes and Incompleteness

We introduce a formal definition of Wolfram’s notion of computational process based on cellular automata, a physics-like model of computation. There is a natural classification of these processes into decidable, intermediate and complete. It is shown that in the context of standard finite injury priority arguments one cannot establish the existence of an intermediate computational process.

💡 Research Summary

The paper sets out to give a rigorous definition of Stephen Wolfram’s informal notion of a “computational process” by anchoring it in the well‑studied framework of cellular automata (CA). A computational process is taken to be the infinite space‑time evolution that results from a fixed local transition rule applied to an initial configuration. This model is deliberately “physics‑like”: it respects locality, simultaneity, and deterministic update rules, thereby capturing the intuitive idea that natural computation proceeds through uniform, locally interacting components.

With this formalism in place, the authors introduce a natural trichotomy of processes based on the complexity of the language (or output set) they generate:

1. Decidable processes – the associated language is recursive; for any input the automaton’s behaviour can be predicted in finite time. These correspond to very simple CA rules whose dynamics are ultimately regular or eventually periodic.

2. Complete processes – the language is recursively enumerable but not recursive; the process can simulate any Turing machine. In CA terms this means that the rule is capable of universal computation, a fact already known for a number of elementary rules (e.g., Rule 110).

3. Intermediate processes – the language is recursively enumerable yet not complete. In classical recursion theory such sets exist (Friedberg–Muchnik constructions), and they occupy the “middle ground” between decidable and complete. The paper asks whether an analogous middle class can be realized by a genuine physical computation model.

To address this, the authors adapt the standard finite‑injury priority argument, a staple of recursion theory used to build intermediate sets while satisfying an infinite list of requirements of decreasing priority. The adaptation proceeds by treating each requirement as a constraint on the CA’s evolution (e.g., forcing a particular pattern to appear or to be avoided). The key observation is that the locality of CA updates makes it impossible to enforce the global, often non‑local, adjustments that a priority construction demands.

The proof sketch shows that any attempt to embed a finite‑injury construction into a CA inevitably leads to a clash: satisfying a higher‑priority requirement may require altering the state of infinitely many cells, which cannot be achieved by a finite number of local updates. Moreover, the simultaneous nature of CA updates prevents the “injury‑repair” cycle that underlies the priority method; once a lower‑priority pattern is destroyed, there is no mechanism to resurrect it without violating the deterministic rule. Consequently, the authors conclude that standard finite‑injury priority arguments cannot be used to exhibit an intermediate computational process in the cellular‑automaton model. In other words, within this physically motivated framework, one cannot prove the existence of a process that is neither decidable nor complete using the usual recursion‑theoretic tools.

The paper’s broader significance lies in the contrast it draws between abstract recursion theory and concrete physical computation. While intermediate r.e. sets are known to exist mathematically, the additional constraints of locality, uniformity, and determinism appear to collapse the complexity spectrum into a dichotomy: processes are either trivially decidable or fully universal. This supports Wolfram’s conjecture that natural computation tends to fall into one of two regimes, and it suggests that any “middle‑ground” phenomena would have to arise from non‑standard models—perhaps non‑deterministic, quantum, or otherwise enriched cellular automata.

Finally, the authors outline several avenues for future work: exploring priority constructions that allow “infinite injury,” investigating non‑deterministic or probabilistic CA variants, and examining whether quantum cellular automata can host intermediate processes. Such extensions could reveal whether the absence of an intermediate class is a peculiarity of the classical CA model or a deeper feature of physically realizable computation.