n-Nested-Simulation Preorder: Hierarchy & Complexity

• n-Nested-simulation preorder is a hierarchy of behavioral relations on finite labelled transition systems that refines simulation by adding a reverse condition at a lower nesting level.
• It is characterized through fragments of Hennessy–Milner logic incorporating nested negation and modal operators to capture finer behavioral distinctions.
• The approach employs nested fixed-point semantics and game-based decision procedures, highlighting a complexity jump from DP/NP at n=2 to PSPACE-completeness for n≥3.

Searching arXiv for recent and foundational papers on nested simulation preorder, characteristic formulae, and simulation complexity. The $n$-nested-simulation preorder is a hierarchy of behavioural preorders on finite labelled transition systems that extends plain simulation by combining forward simulation with a reverse requirement at one lower nesting level. Introduced by Groote and Vaandrager and situated by van Glabbeek within the linear-time/branching-time spectrum, it yields increasingly fine behavioural distinctions as $n$ grows, while remaining strictly coarser than bisimilarity. Two complementary lines of work are central to its study: a modal-logic account via fragments of Hennessy–Milner logic that characterize the preorder and its kernel, and a nested fixed-point account that treats $n$-nested simulation as a recursively defined greatest-fixed-point semantics (Aceto et al., 17 Sep 2025, Aceto et al., 2012).

1. Formal definition and inductive structure

The standard setting is a finite labelled transition system $\mathcal{L}=(P,A,\rightarrow)$, where $P$ is a finite set of states, $A$ is a finite, non-empty set of actions, and $\rightarrow \subseteq P \times A \times P$. One writes $p \xrightarrow{a} q$ for $(p,a,q)\in\rightarrow$. Processes are assumed finite and loop-free, and are often described by the CCS-style syntax

$p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$

The depth $n$0 of a process is the length of its longest trace (Aceto et al., 17 Sep 2025).

The simulation preorder $n$1 is the largest relation such that every transition of the left-hand process can be matched by a transition of the right-hand process with the same label, and the successors remain related: $n$2

The $n$3-nested-simulation preorders $n$4 are defined inductively. For $n$5, $n$6. For $n$7, $n$8 is the largest relation such that

$n$9

Thus the relation at level $n$0 is simulation in one direction together with a weaker reverse condition at level $n$1. For $n$2, this specializes to forward simulation plus ordinary simulation in the reverse direction. For general $n$3, the definition continues inductively in the same way (Aceto et al., 17 Sep 2025).

Each preorder has a kernel, the induced equivalence

$n$4

The hierarchy is strict: $n$5 Hence increasing $n$6 yields finer behavioural distinctions, but bisimilarity remains strictly finer than every $n$7-nested-simulation preorder (Aceto et al., 17 Sep 2025).

2. Modal characterization by fragments of Hennessy–Milner logic

The preorder is characterized by fragments of Hennessy–Milner logic. Full HML has syntax

$n$8

with the standard semantics for $n$9, $\mathcal{L}=(P,A,\rightarrow)$0, Boolean connectives, and negation (Aceto et al., 17 Sep 2025).

For simulation, the characterizing fragment contains only positive Boolean structure and diamonds: $\mathcal{L}=(P,A,\rightarrow)$1 For $\mathcal{L}=(P,A,\rightarrow)$2-nested simulation with $\mathcal{L}=(P,A,\rightarrow)$3, the fragment is defined by

$\mathcal{L}=(P,A,\rightarrow)$4

Syntactically, $\mathcal{L}=(P,A,\rightarrow)$5 contains positive combinations of $\mathcal{L}=(P,A,\rightarrow)$6-formulas together with negations of formulas one level lower. The backward clause in the semantic definition is therefore mirrored by restricted negation in the logic (Aceto et al., 17 Sep 2025).

If $\mathcal{L}=(P,A,\rightarrow)$7, then the logical characterization is

$\mathcal{L}=(P,A,\rightarrow)$8

Thus the preorder is exactly inclusion of theories in the fragment $\mathcal{L}=(P,A,\rightarrow)$9, and the kernel $P$0 is equality of theories in that fragment. In this sense the fragments $P$1 are behaviourally complete for $P$2-nested simulation (Aceto et al., 17 Sep 2025).

A recurrent misunderstanding is to treat the hierarchy as merely syntactic. The theory-inclusion theorem rules that out: the hierarchy is semantic, and the increasing use of lower-level negation corresponds directly to increasingly discriminating behavioural preorders.

3. Characteristic formulae, kernels, and primality

Given a logic $P$3, a formula $P$4 is characteristic for a process $P$5 within $P$6 when

$P$7

For $P$8, this is equivalent to

$P$9

The truth set of $A$0 is therefore exactly the upward cone of $A$1 under the preorder (Aceto et al., 17 Sep 2025).

Characteristic formulae modulo the kernel are more selective. A formula $A$2 is characteristic for $A$3 modulo $A$4 when

$A$5

equivalently, iff $A$6. A convenient reformulation is that $A$7 is characteristic modulo $A$8 iff $A$9 is satisfiable and all its models are pairwise $\rightarrow \subseteq P \times A \times P$0-equivalent (Aceto et al., 17 Sep 2025).

The key logical notion used to connect characteristic formulae to decision problems is primality. A formula $\rightarrow \subseteq P \times A \times P$1 is prime in $\rightarrow \subseteq P \times A \times P$2 if

$\rightarrow \subseteq P \times A \times P$3

Unsatisfiable formulas are trivially prime, since they entail everything (Aceto et al., 17 Sep 2025).

For every $\rightarrow \subseteq P \times A \times P$4, the decisive equivalence is: $\rightarrow \subseteq P \times A \times P$5 This turns the existence of a characteristic formula from a semantic question about processes into the conjunction of two logical decision problems. For $\rightarrow \subseteq P \times A \times P$6, the modulo-kernel case is essentially degenerate: no formulas in $\rightarrow \subseteq P \times A \times P$7 are characteristic modulo $\rightarrow \subseteq P \times A \times P$8 in a nontrivial way (Aceto et al., 17 Sep 2025).

4. Nested fixed points and characteristic declarations

A separate but complementary account treats $\rightarrow \subseteq P \times A \times P$9-nested simulation as a semantics defined by nested greatest fixed points. This perspective was developed to address a limitation of earlier single-fixed-point frameworks, which do not suitably cover semantics defined by nested fixed points, such as the $p \xrightarrow{a} q$0-nested simulation semantics for $p \xrightarrow{a} q$1 greater than $p \xrightarrow{a} q$2 (Aceto et al., 2012).

Let $p \xrightarrow{a} q$3 be the monotone function whose greatest fixed point is standard simulation, and let $p \xrightarrow{a} q$4 be its inverse-simulation counterpart. Then

$p \xrightarrow{a} q$5

For $p \xrightarrow{a} q$6, the recursive functions are

$p \xrightarrow{a} q$7

and the corresponding preorders are their greatest fixed points (Aceto et al., 2012).

Characteristic formulae are then obtained through declarations over variables indexed by processes. For standard simulation and inverse simulation one uses

$p \xrightarrow{a} q$8

$p \xrightarrow{a} q$9

The nested declarations are defined inductively by

$(p,a,q)\in\rightarrow$0

$(p,a,q)\in\rightarrow$1

Here $(p,a,q)\in\rightarrow$2 and $(p,a,q)\in\rightarrow$3 are treated as constants at the next level of the logic (Aceto et al., 2012).

The resulting theorem is that, for every $(p,a,q)\in\rightarrow$4,

$(p,a,q)\in\rightarrow$5

This gives characteristic formulae for $(p,a,q)\in\rightarrow$6-nested simulation in a structured hierarchy of logics with nested greatest fixed points, and makes explicit how the semantic nesting is reflected in the logic (Aceto et al., 2012).

5. Complexity of satisfiability, primality, and characteristic formulae

The recent complexity classification focuses not on deciding $(p,a,q)\in\rightarrow$7 directly, but on deciding whether a formula in $(p,a,q)\in\rightarrow$8 is satisfiable, prime, or characteristic for some process. For action sets $(p,a,q)\in\rightarrow$9 with $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$0, satisfiability for $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$1 is in $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$2, satisfiability for $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$3 is $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$4-complete, satisfiability for $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$5 is $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$6-complete, and for all $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$7, satisfiability for $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$8 is $p ::= \mathtt{0} ~\mid~ a.p ~\mid~ p + p.$9-complete. The Formula Primality Problem is coNP-complete for $n$00 and $n$01-complete for $n$02 when $n$03 (Aceto et al., 17 Sep 2025).

These results imply a sharp complexity jump between $n$04 and $n$05. For all $n$06 and $n$07, deciding whether $n$08 is characteristic for some process within $n$09 is $n$10-complete, and deciding whether it is characteristic modulo $n$11 is also $n$12-complete. The reason is exactly the equivalence

$n$13

combined with $n$14-completeness of both components for $n$15 (Aceto et al., 28 May 2025).

The case $n$16 is more delicate. Primality in $n$17 is coNP-complete, while satisfiability is $n$18-complete, so deciding whether a formula is characteristic for a process within $n$19, or modulo $n$20, lies in $n$21, the class of languages expressible as the intersection of one language in $n$22 and one in coNP. Earlier lower bounds include US-hardness and coNP-hardness (Aceto et al., 17 Sep 2025).

This classification shows that the logical problems associated with $n$23-nested simulation remain comparatively moderate at level $n$24, but become $n$25-complete from level $n$26 onward.

6. Decision procedures, comparisons, and broader significance

For $n$27, the upper bounds are obtained through two-player perfect-information games. In the $n$28 games, player $n$29 builds two labelled trees corresponding to processes satisfying a formula, and player $n$30 must demonstrate that these processes are $n$31-nested-simulation equivalent. In the $n$32 games, initiated on a satisfiable $n$33, player $n$34 builds two trees and player $n$35 builds a third; the rules are arranged so that $n$36 has a winning strategy iff $n$37 is prime in $n$38. These games have polynomially bounded depth, are zero-sum and perfect-information, and use $n$39-oracle calls per round, which yields the $n$40 upper bounds (Aceto et al., 17 Sep 2025).

For $n$41, the proof method is constructive rather than game-based. The algorithm $n$42 nondeterministically constructs small processes satisfying $n$43, and $n$44 runs $n$45 twice, computes the maximal lower bound $n$46, and uses its existence and satisfaction of $n$47 to decide primality. The maximal lower bound, when defined, can be computed in polynomial time and is of polynomial size (Aceto et al., 17 Sep 2025).

In the broader landscape, nested simulation sits between basic simulation and bisimulation. For standard simulation and some related preorders over constant-size action sets, satisfiability and primality can be decided in polynomial time; with larger action sets, ready simulation logic has $n$48-complete satisfiability and coNP-complete primality; for bisimilarity characterized by HML, primality and characteristic-formula checking are $n$49-complete. For $n$50, nested simulation therefore aligns more closely with bisimulation in logical complexity than with plain simulation (Aceto et al., 28 May 2025).

A plausible implication emerges from fine-grained complexity results for ordinary simulation. Deciding simulation preorder on deterministic labelled transition systems and simulation equivalence on general labelled transition systems has conditional quadratic lower bounds under the Strong Exponential Time Hypothesis; this means that deciding simulation is inherently quadratic under SETH. Since $n$51-nested simulation refines simulation, this suggests that general, graph-size-parameterized algorithms for $n$52-nested simulation preorder are unlikely to admit subquadratic runtimes without strong structural restrictions such as bounded branching, bounded depth, or restricted alphabets (Groote et al., 2024).

Taken together, these results identify $n$53-nested simulation as a structurally rich hierarchy whose semantic definition, modal characterization, fixed-point treatment, and logical decision complexity are tightly aligned. The hierarchy refines simulation strictly, remains strictly coarser than bisimilarity, admits characteristic formulae through satisfiable prime formulas, and exhibits a clear complexity threshold: the passage from $n$54-nested to $n$55-nested simulation marks a transition from $n$56/coNP phenomena to full $n$57-completeness.