Forbidden-Context & Ordered Grammar Systems

Abstract: In this paper, we consider combining the ideas of forbidden random context grammars as well as of ordered grammars with cooperating distributed grammar systems (CDGS). We focus on investigating their generative capacities. Both ideas can be added to CDGS in two ways: either having (e.g.) a strict order of the rules in each component, or having a strict order on the components. This leads to four different scenarios, only some of them have been addressed in the literature before. While in the area of CDGS, many inclusions among language classes have been %are still open questions for decades, the proposed addition of forbidden random context and ordered regulation variants leads to a clear picture which allows us to get down to only five different classes of languages well known from classical regulated rewriting. This way, we also solve some open problems from the literature.

• The paper shows that ordered and forbidden-context CDGS reach RE for erasing systems in most =k and ≥k modes with k≥2, while t-mode remains limited to ORD or E(P)T0L depending on regulation placement.
• Forbidden-context regulation collapses the =k hierarchy from level 2, establishes equivalences with ordered grammars and graph-controlled grammars with appearance checking, and shows forbidding conditions can replace permitting conditions in key modes.
• The results settle longstanding inclusion questions for priority-controlled systems, but leave the exact power of ≥k-mode entry conditions, the non-erasing ORD versus fRC problem, and descriptional complexity open.

Motivation and setting

Cooperating distributed grammar systems (CDGS) with context-free rules were introduced as formal models of blackboard-style problem solving, with derivation modes ($=k$, $\leq k$, $\geq k$, $*$, $t$) governing how many steps a component performs before handing the sentential form back. Despite decades of study, the inclusion structure among CDGS language classes remains largely unresolved — the classical monograph of Csuhaj-Varjú et al. leaves many relations open, and no classical CDGS variant is known to characterize the class RE of recursively enumerable languages. The paper under review (Fernau, Kuppusamy, Schulz (2604.19963)) takes a different route: rather than resolving these open questions directly, it enriches CDGS with two well-understood regulation mechanisms from regulated rewriting — forbidden random context conditions and strict orderings on rules — applied either to the rules within each component or to the components themselves. This yields four scenarios, only some previously studied, and produces a remarkably clean classification: nearly all resulting language classes collapse onto five known classes from classical regulated rewriting (RE, $\textsf{GC}_{ac}$, $\textsf{fRC}=\textsf{ORD}$, E(P)T0L), thereby settling several long-standing open problems.

Ordered cooperating distributed grammar systems

An ordered grammar (Friš, 1968) attaches a strict partial order $>$ to its context-free rules; a rule may be applied only if no strictly larger rule is applicable at the current sentential form. An OCDGS equips each component of a CDGS with such an ordering. A small system over $\{A,B,C\}$ illustrates the mechanism: blocking rules of the form $X \to X$ placed above productive rules prevent components from terminating prematurely, and a three-component system generates $\leq k$0 in $\leq k$1-mode.

The central result for this model is that ordered CDGS simulate graph-controlled grammars with appearance checking:

$\leq k$2

The simulation uses $\leq k$3 components per graph-controlled grammar, with orderings of dimension two. Each labeled rule $\leq k$4 is simulated by three components handling the failure case (via an ordering that makes the step productive only when $\leq k$5 is absent) and the success case (via a two-component protocol that marks one occurrence of $\leq k$6 as $\leq k$7 before rewriting). Since $\leq k$8, this yields computational completeness for the erasing case across all modes $\leq k$9 — a notable contrast with classical CDGS, where no mode is known to reach RE. For the non-erasing case, the authors exploit closure of $\geq k$0 under left derivatives, decomposing $\geq k$1 into left derivatives per terminal symbol and simulating each separately.

In contrast, the $\geq k$2-mode behaves differently. Because a component can only be left when no rule is applicable, blocking tricks fail, and indeed:

$\geq k$3

The proof marks every nonterminal with the index of the currently active component plus transition markers, then flattens the whole system into a single ordered grammar whose ordering enforces component exclusivity. Consequently, adding multiple ordered components gives no generative power over a single ordered grammar in $\geq k$4-mode — a structural limitation specific to that mode.

Forbidden random context as grammar-system components

A random context rule $\geq k$5 applies only if all symbols of $\geq k$6 occur and none of $\geq k$7 occurs; forbidding-only variants define fRC grammars, known since Mayer (1972) to equal ordered grammars. Masopust (2009) had shown $\geq k$8, hence $\geq k$9 with erasing. The paper extends this equivalence to all other modes via the classical Cremers–Mayer translation between orderings and forbidding contexts:

$*$0

and shows that weak modes collapse to single-component systems: $*$1 for $*$2.

The most striking result here concerns mode hierarchies. For classical CDGS, whether the $*$3-mode hierarchy collapses has been open for decades. With forbidden context attached to rules, the hierarchy collapses entirely from level 2:

$*$4

The construction introduces counter nonterminals $*$5 that store pending right-hand sides while counting up to $*$6 forced steps within a simulated component, together with selector symbols $*$7 enforcing that only one original component is active. Combined with the earlier equalities, this places $*$8 exactly at $*$9 (erasing case), while level 1 remains strictly below via the known hierarchy $t$0 inherited from ordered grammars.

Entry-condition control versus rule-level control

The paper also revisits the "internal control" variant of Csuhaj-Varjú et al., where random context conditions act as entry filters on whole components rather than on individual rules. Two results calibrate this mechanism's strength. First, forbidding entry conditions add nothing in $t$1-mode:

$t$2

proved by a priming construction that defers forbidden-context checks to the first step of each component. Since E(P)T0L is strictly contained in ORD, entry-level forbidding is strictly weaker than rule-level forbidding in $t$3-mode. Second, in $t$4-mode, entry filters with only forbidding conditions already achieve full power:

$t$5

which considerably strengthens the book result $t$6 by showing that permitting conditions are unnecessary there. The harder direction encodes pairs of consecutive rule applications of an fRCCDGS component into dedicated frc-components using marker nonterminals $t$7 and a separator $t$8, so that each $t$9 block of the simulating system faithfully reproduces two-step blocks of the simulated one.

Priorities on components

Mitrana, Păun, and Rozenberg (1994) studied CDGS with priorities — a strict ordering on components, where a component may act only if no higher-priority component can. The paper proves that this component-level ordering is equivalent to component-level forbidding entry conditions for most modes:

$\textsf{GC}_{ac}$0

The forward direction attaches failure components $\textsf{GC}_{ac}$1 ranked above $\textsf{GC}_{ac}$2; the reverse direction follows by chaining through the results above, or directly for $\textsf{GC}_{ac}$3/$\textsf{GC}_{ac}$4 modes by placing all left-hand sides of higher-priority productions into the forbidden sets of lower-priority ones. As a corollary, the paper resolves an open question from Mitrana et al.: $\textsf{GC}_{ac}$5 is strictly included in $\textsf{GC}_{ac}$6 for the $\textsf{GC}_{ac}$7- and $\textsf{GC}_{ac}$8-modes, and moreover $\textsf{GC}_{ac}$9.

For the $\textsf{fRC}=\textsf{ORD}$0-mode, the picture is less complete. The paper establishes $\textsf{fRC}=\textsf{ORD}$1 (witnessed by $\textsf{fRC}=\textsf{ORD}$2, which lies outside $\textsf{fRC}=\textsf{ORD}$3 by Hauschildt–Jantzen but inside $\textsf{fRC}=\textsf{ORD}$4), and shows the collapse $\textsf{fRC}=\textsf{ORD}$5 for all $\textsf{fRC}=\textsf{ORD}$6. However, whether the chain

$\textsf{fRC}=\textsf{ORD}$7

contains strict inclusions remains open — the authors explicitly note they are "not sure at all" about the power of $\textsf{fRC}=\textsf{ORD}$8 relative to $\textsf{fRC}=\textsf{ORD}$9.

Limitations and open questions

Several caveats qualify the results. The equivalence $>$0 inherits a more than fifty-year-old open problem: whether disallowing erasing rules strictly weakens ordered grammars. The $>$1-mode inclusions mentioned above are unresolved. Permitting random context conditions are deliberately excluded from the study, despite evidence (e.g., left-permitting variants in the literature) that they may behave differently. Finally, descriptional complexity — number of components, nonterminals, or rules per component — is touched only incidentally (e.g., two components suffice for $>$2, one for $>$3); a systematic treatment is absent.

Conclusion

This paper demonstrates that augmenting CDGS with forbidden-context or ordering regulations, whether at rule level or component level, dissolves much of the murky inclusion structure of classical grammar systems into a crisp landscape anchored at five classical language classes. Rule-level regulation in modes $>$4 ($>$5) yields computational completeness, weak modes collapse to single-grammar power, and the $>$6-mode sits at E(P)T0L or ORD depending on where the condition is attached. Along the way, the mode-hierarchy collapse for fRCCDGS and the resolution of the PCDGS-versus-matrix-grammars question settle problems open for decades, while leaving the $>$7-mode with entry filters as the principal remaining gap.