---
title: 2-Nested-Simulation Preorder
url: https://www.emergentmind.com/topics/2-nested-simulation-preorder
type: topic
---

# 2-Nested-Simulation Preorder

The 2-nested-simulation preorder is the second level of the nested-simulation hierarchy introduced by Groote and Vaandrager. It refines ordinary simulation by combining a forward simulation clause with a reverse constraint one nesting level lower, and it is characterized modally by the fragment \(\mathcal{L}_{2S}\) of Hennessy–Milner logic. In the cited 2025 analyses, the preorder is studied chiefly through the decision problem of whether a formula is characteristic for some process, a question that is equivalent to satisfiability together with primeness in \(\mathcal{L}_{2S}\). The resulting complexity profile makes the \(2S\) case a boundary point: harder than plain simulation, but strictly easier than the \(n\ge 3\) nested-simulation levels [2509.14089].

## 1. Definition and position in the hierarchy

For finite, loop-free labelled transition systems, ordinary simulation is defined by
\[
p \preceq_S q \iff \forall\, p \xrightarrow{a} p'\;\exists\, q \xrightarrow{a} q' \text{ such that } p' \preceq_S q'.
\]
The \(n\)-nested simulation preorder is defined inductively by setting \(\preceq_{1S}=\preceq_S\), and for \(n>1\),
\[
p \preceq_{nS} q \iff \Big(\forall\, p \xrightarrow{a} p'\;\exists\, q \xrightarrow{a} q' \text{ with } p' \preceq_{nS} q'\Big)\;\wedge\; q \preceq_{(n-1)S} p.
\]
Specializing to \(n=2\),
\[
p \preceq_{2S} q \iff \Big(\forall\, p \xrightarrow{a} p' \;\exists\, q \xrightarrow{a} q' \text{ such that } p' \preceq_{2S} q'\Big)\;\wedge\; q \preceq_S p.
\]
Thus \(2\)-nested simulation strengthens ordinary simulation by requiring not only the usual forward matching condition, but also a reverse simulation condition at one lower nesting level [2505.22277].

Each \(\preceq_{nS}\) is a preorder, and its kernel is the induced equivalence
\[
p \equiv_{nS} q \iff p \preceq_{nS} q \text{ and } q \preceq_{nS} p.
\]
For \(n=2\), this yields \(\equiv_{2S}\). The hierarchy is strict:
\[
{\sim}\subsetneq {}_{nS}, \qquad {}_{(n+1)S}\subsetneq {}_{nS} \text{ for every } n\ge 1.
\]
Accordingly, \(2\)-nested simulation is strictly finer than simulation, but coarser than \(3\)-nested simulation and bisimilarity. This places it at the first level where nesting changes the semantics in a substantial way.

## 2. Modal characterization by \(\mathcal{L}_{2S}\)

The modal language characterizing \(2\)-nested simulation extends the positive simulation fragment by a restricted form of negation. For simulation,
\[
\varphi_S ::= \top \mid \bot \mid \varphi_S\wedge \varphi_S \mid \varphi_S\vee \varphi_S \mid \langle a\rangle \varphi_S,
\]
whereas for \(n\ge 2\),
\[
\varphi_{nS} ::= \top \mid \bot \mid \varphi_{nS}\wedge \varphi_{nS} \mid \varphi_{nS}\vee \varphi_{nS} \mid \langle a\rangle \varphi_{nS} \mid \neg \varphi_{(n-1)S}.
\]
Hence
\[
\varphi_{2S} ::= \top \mid \bot \mid \varphi_{2S}\wedge \varphi_{2S} \mid \varphi_{2S}\vee \varphi_{2S} \mid \langle a\rangle \varphi_{2S} \mid \neg \varphi_S.
\]
This is the first level at which negation appears, but it is confined to \(\mathcal{L}_S\)-formulae [2505.22277].

Truth is given by the usual Hennessy–Milner semantics; in particular,
\[
p \models \langle a\rangle \varphi \text{ iff there is some } p \xrightarrow{a} q \text{ such that } q\models \varphi,
\]
and negation is interpreted classically:
\[
p \models \neg \varphi \iff p \not\models \varphi.
\]

The fundamental modal characterization is
\[
p \preceq_{nS} q \iff \mathcal{L}_{nS}(p)\subseteq \mathcal{L}_{nS}(q),
\]
and therefore, in particular,
\[
p \preceq_{2S} q \iff \mathcal{L}_{2S}(p)\subseteq \mathcal{L}_{2S}(q).
\]
This identifies \(2\)-nested simulation with logical inclusion over \(\mathcal{L}_{2S}\). A formula \(\varphi\in\mathcal{L}_{2S}\) is characteristic for a process \(p\) within \(\mathcal{L}_{2S}\) precisely when
\[
q\models \varphi \iff \mathcal{L}_{2S}(p)\subseteq \mathcal{L}_{2S}(q),
\]
equivalently,
\[
q\models \varphi \iff p \preceq_{2S} q.
\]
Such a formula defines exactly the upward cone of \(p\) under \(\preceq_{2S}\).

## 3. Characteristic formulae, primeness, and kernel semantics

A formula \(\varphi\) is satisfiable if there is a process that satisfies it. Primeness is defined relative to a logic \(\mathcal{L}\) by
\[
\varphi \models \varphi_1\vee \varphi_2 \;\Longrightarrow\; \varphi \models \varphi_1 \text{ or } \varphi \models \varphi_2
\]
for all \(\varphi_1,\varphi_2\in\mathcal{L}\). Every unsatisfiable formula is trivially prime. In the \(2S\) setting, \(\varphi\in\mathcal{L}_{2S}\) is prime exactly when this implication holds for all \(\varphi_1,\varphi_2\in\mathcal{L}_{2S}\) [2509.14089].

The central equivalence is:
\[
\varphi\in \mathcal{L}_{nS} \text{ is characteristic for some process within } \mathcal{L}_{nS}
\iff
\varphi \text{ is satisfiable and prime in } \mathcal{L}_{nS}.
\]
Thus, for \(n=2\),
\[
\varphi\in \mathcal{L}_{2S} \text{ is characteristic within } \mathcal{L}_{2S}
\iff
\varphi \text{ is satisfiable and prime in } \mathcal{L}_{2S}.
\]
This equivalence explains why satisfiability and primeness are the decisive computational subproblems.

The preorder and its kernel lead to two different notions of characteristic formula. Within \(\mathcal{L}_{2S}\), a characteristic formula identifies an upward closure under \(\preceq_{2S}\). Modulo the kernel \(\equiv_{2S}\), a formula is characteristic for \(p\) when
\[
q \models \varphi \iff \mathcal{L}_{2S}(p)=\mathcal{L}_{2S}(q),
\]
equivalently,
\[
q\models \varphi \iff p\equiv_{2S} q.
\]
So kernel-level characterization isolates exactly one \(\equiv_{2S}\)-equivalence class rather than an upward cone.

The cited examples of primeness are given in \(\mathcal{L}_S\): \(\langle a\rangle \top\) is prime, whereas \(\langle a\rangle \top \vee \langle b\rangle \top\) is not. These examples illustrate the same logical notion used at the \(2S\) level.

## 4. Complexity profile of the \(2S\) case

The 2025 results establish a sharp split between the \(2\)-nested case and the levels \(n\ge 3\). For \(\mathcal{L}_{2S}\), satisfiability is NP-complete and the Formula Primality problem is coNP-complete. For \(n\ge 3\), satisfiability, primeness, and deciding characteristic formulae are PSPACE-complete [2505.22277].

| Problem | \(2S\) result | Higher-level comparison |
|---|---|---|
| Satisfiability in \(\mathcal{L}_{2S}\) | NP-complete | PSPACE-complete for \(n\ge 3\) |
| Formula Primality in \(\mathcal{L}_{2S}\) | coNP-complete | PSPACE-complete for \(n\ge 3\) |
| Characteristic formula decision | In DP when \(|Act|>1\) | PSPACE-complete for \(n\ge 3\) |

The class
\[
DP=\{L_1\cap L_2 \mid L_1\in NP \text{ and } L_2\in coNP\}
\]
is used exactly as follows. Since characteristic-within-\(\mathcal{L}_{2S}\) is equivalent to satisfiable \(+\) prime, and satisfiability is in NP while primeness is in coNP, deciding whether a formula is characteristic for a process within \(\mathcal{L}_{2S}\) is in DP. For characteristic modulo \(\equiv_{2S}\), the decomposition is satisfiability together with the condition that every two processes satisfying the formula are \(\equiv_{2S}\)-equivalent; the first component is in NP and the second in coNP, so this problem is also in DP. The exact stated result is: let \(|Act|>1\). Deciding whether a formula in \(\mathcal{L}_{2S}\) is characteristic for a process within \(\mathcal{L}_{2S}\), or modulo \(\equiv_{2S}\), is in DP.

The lower bounds are weaker than DP-completeness. When \(|Act|>1\), deciding whether \(\varphi\in\mathcal{L}_{2S}\) is characteristic within \(\mathcal{L}_{2S}\) is US-hard, and deciding whether \(\varphi\) is characteristic modulo \(\equiv_{2S}\) is coNP-hard. A recurrent misconception is therefore avoided in the cited work: the result for characteristic-formula decision at \(n=2\) is DP membership, not DP-completeness.

The reason the \(2S\) case is easier than \(n\ge 3\) is explicit: for a satisfiable formula \(\varphi\in\mathcal{L}_{2S}\), there is always a tableau for \(\varphi\)—and so a corresponding process satisfying \(\varphi\)—of polynomial size. This small-model property supports NP for satisfiability and coNP for primeness, whereas the higher nested levels require PSPACE game machinery.

## 5. Polynomial witness machinery: tableau construction, maximal lower bounds, and primality

The distinctive algorithmic contribution for \(2S\) is a direct coNP procedure built from three components: a nondeterministic process constructor \(\mathtt{ConPro}\), a polynomial-time maximal-lower-bound procedure \(\mathtt{MLB}\), and the primality checker \(\mathtt{Prime}_{2S}\) [2509.14089].

\(\mathtt{ConPro}(\varphi)\) is a nondeterministic extension of a tableau construction for \(\varphi\in\mathcal{L}_{2S}\). If \(\varphi\) is satisfiable, some execution outputs a process satisfying \(\varphi\); if \(\varphi\) is unsatisfiable, executions stop without output. The cited presentation emphasizes that lines \(1\)–\(19\) implement the tableau construction, while lines \(20\)–\(30\) add bounded “box-generated” witness structure exposing failures of simulation when needed. Two size bounds are crucial: if \(\mathtt{ConPro}(\varphi)\) outputs a process \(r\), then \(r\models\varphi\) and its depth is at most \(md(\varphi)+1\); moreover, the output size is at most
\[
4|\varphi|.
\]
This polynomial bound is the small-model property exploited throughout.

The maximal lower bound with respect to \(\preceq_{2S}\), written \({}_{2S}(p,q)\) or \(\mathtt{MLB}(p,q)\), is the second decisive ingredient. Its special structural property is:
\[
{}_{2S}(p,q) \text{ exists iff } p\equiv_S q.
\]
Moreover, one can decide in polynomial time whether \({}_{2S}(p,q)\) exists; if it exists, it can be computed in polynomial time and has size at most
\[
2|p||q|.
\]
The constructive definition used in the proof is
\[
(p,q)=\sum \{\, a.(p',q') \mid a\in Act,\ p\xrightarrow{a} p',\ q\xrightarrow{a} q',\ p'\equiv_S q' \,\}.
\]
This recursively builds the common lower bound from simulation-equivalent successors.

The key semantic equivalence is that for \(\varphi\in\mathcal{L}_{2S}\) and \(p,q\models\varphi\), the following are equivalent:
\[
{}_{2S}(p,q) \text{ exists and } {}_{2S}(p,q)\models\varphi,
\]
and
\[
\exists r\; \bigl(r\models\varphi \wedge r\preceq_{2S} p \wedge r\preceq_{2S} q\bigr).
\]
This allows primality to be checked by examining only the maximal lower bound.

The algorithm \(\mathtt{Prime}_{2S}\) operates by running \(\mathtt{ConPro}(\varphi)\) twice to obtain \(p_1,p_2\). If either construction fails to output, the branch accepts, corresponding to unsatisfiability and hence trivial primeness. Otherwise it computes
\[
g \gets \mathtt{MLB}(p_1,p_2).
\]
If \(g\) does not exist, the branch rejects; if \(g\models\varphi\), it accepts; otherwise it rejects. The correctness criterion is universal:
\[
\varphi \text{ is prime iff } \mathtt{Prime}_{2S}(\varphi) \text{ always accepts.}
\]
Because each execution is polynomial-time, universal acceptance yields a coNP algorithm, giving coNP-completeness for Formula Primality in \(\mathcal{L}_{2S}\).

For the kernel-style problem, the cited work also describes \(\mathtt{EquivProc}_{2S}\), which checks whether every two processes satisfying \(\varphi\) are \(\equiv_{2S}\)-equivalent. This condition is in coNP, and checking \(p\equiv_{2S} q\) for given processes can be done in polynomial time.

## 6. Relation to ordinary simulation algorithms and scope of extrapolation

A separate line of work studies efficient algorithms for ordinary simulation preorder rather than nested simulation. In particular, a 2017 paper develops a framework for computing the coarsest simulation preorder included in an initial preorder, using maximal transitions, \(\mathscr{U}\)-stability, and partition-relation refinement. Its main complexity theorem gives time
\[
O(|P_{\mathrm{sim}}|.|\rightarrow|)
\]
and bit-space
\[
O\bigl(|P_{\mathrm{sim}}|^2.\log(|P_{\mathrm{sim}}|) + |Q|.\log(|Q|)\bigr).
\]
The paper is explicit that it does not define or compute nested simulation, including \(2\)-nested simulation [1709.01826].

What this establishes directly is a strong algorithmic base for ordinary simulation inside an arbitrary initial preorder. A plausible implication is that such a procedure can serve as a primitive in staged constructions of \(2\)-nested simulation, because the 2017 framework is parameterized by the initial preorder and organized as a decreasing sequence of stable preorders. The cited note states this relevance carefully: the paper does not explicitly mention nested simulation, does not prove that repeated use computes \(2\)-nested simulation, and does not provide complexity analyses for nested or \(k\)-nested simulation. Its significance for \(2S\) is therefore foundational rather than definitive.

Taken together, the cited sources present the 2-nested-simulation preorder as the second level of a strict behavioral hierarchy, logically characterized by \(\mathcal{L}_{2S}\), semantically tied to upward closures and kernel classes under \(\preceq_{2S}\), and computationally distinguished by a small-model property and a polynomial-time maximal-lower-bound construction. Those features make \(2S\) the threshold level at which nested simulation becomes substantially richer than ordinary simulation while still remaining below the PSPACE behavior of the higher nested levels [2505.22277].

Source: https://www.emergentmind.com/topics/2-nested-simulation-preorder