Witnesses for Fixpoint Games on Lattices

Abstract: We construct witnesses that can be used to derive strategies in fixpoint games and provide proof that the least fixpoint of a function is either above or not below some given bound. We rely on a lattice-theoretical approach, including a Galois connection that connects a lattice representing the "logic universe", where the witness lives, with another lattice representing the "behaviour universe", over which the function is defined. In fact we consider two types of games -- primal and dual games -- and in both cases show how to derive winning strategies in the game from witnesses and construct witnesses from strategies. The two games differ wrt. their rules and the choice of basis of the lattice. The theory can be instantiated to well-known examples: in particular we compare with the construction of distinguishing formulas in standard bisimilarity and behavioural metrics for probabilistic systems. As a new case study we consider witnesses for certifying lower bounds for the termination probability for Markov chains.

• The paper establishes a constructive correspondence between lattice witnesses and finitary winning strategies in primal and dual fixpoint games, with ordinal degree measures guaranteeing termination.
• Witness existence precisely characterizes strict lower bounds and non-upper bounds for least fixpoints, under continuity, co-continuity, irreducibility, and properness assumptions.
• The framework generates certificates for bisimilarity, Kantorovich behavioural metrics, and Markov-chain termination probabilities, including formula-, linear-programming-, and tree-based witnesses.

Overview

The paper develops a lattice-theoretic framework for constructing witnesses — abstract counterparts of distinguishing formulas — that certify lower bounds for least fixpoints of monotone functions, or certify that a given element is not an upper bound. The setting is a Galois connection $\alpha \dashv \gamma$ between a "logic universe" $\mathbb{L}$ (where witnesses live) and a "behaviour universe" $\mathbb{B}$ (where the behaviour function $\mathsf{beh}$ is defined), with the compatibility condition $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$, which guarantees $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$ and preservation of all Kleene iteration stages (2603.11908). This generalizes the Galois-connection account of the Hennessy-Milner theorem from prior work by the same group.

The central contribution is a bidirectional correspondence between witnesses and winning strategies in two fixpoint games on continuous lattices: a primal way-below game (where the witness guides the existential player $\exists$) and a dual game (obtained by dualizing the greatest-fixpoint game, where the witness guides the universal player $\forall$). Both directions are constructive: witnesses yield finitary strategies, and strategies yield witnesses via inductive definitions whose termination is controlled by ordinal-valued degree measures.

Witnesses and degrees

A primal witness for $b \in \mathbb{B}$ is an element $a \in \mathbb{L}$ with $\mathbb{L}$0 and $\mathbb{L}$1; a dual witness satisfies $\mathbb{L}$2. Existence is characterized exactly: assuming $\mathbb{L}$3 continuous with a join basis of irreducibles, a primal witness for basis element $\mathbb{L}$4 exists iff $\mathbb{L}$5, and a dual witness exists iff $\mathbb{L}$6. The proof reduces to the fact that the left adjoint $\mathbb{L}$7 preserves directed suprema, so $\mathbb{L}$8 can be witnessed on a basis element below $\mathbb{L}$9.

Two ordinal-valued measures organize the constructions: the degree $\mathbb{B}$0, the least $\mathbb{B}$1 with $\mathbb{B}$2, and the co-degree $\mathbb{B}$3, the least $\mathbb{B}$4 with $\mathbb{B}$5. Key properties include additivity of degree over finite joins ($\mathbb{B}$6) and the bound $\mathbb{B}$7 relating a witness to the co-degree of its target. These ensure that recursive witness construction terminates.

Fixpoint games and finitary strategies

The paper introduces a new variant of the fixpoint games of Baldan, König, Mika-Michalski, and Padoan: the primal way-below game, in which $\mathbb{B}$8 must play $\mathbb{B}$9 with $\mathsf{beh}$0 and $\mathsf{beh}$1 answers with $\mathsf{beh}$2. Its soundness and completeness rest on continuity of $\mathsf{beh}$3: $\mathsf{beh}$4 has a winning strategy from $\mathsf{beh}$5 iff $\mathsf{beh}$6. Notably, when $\mathsf{beh}$7 is Scott-continuous, $\mathsf{beh}$8 has a finitary winning strategy — each move is a finite join of basis elements of strictly smaller degree — so the game terminates within $\mathsf{beh}$9 steps. The argument uses compactness of sets of the form $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$0 and openness of $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$1.

Continuity of the lattice is essential, not incidental: the paper gives a counterexample on a non-continuous lattice where $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$2 wins although $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$3. For such settings the dual game applies instead, requiring only co-continuity plus co-properness of $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$4 (inverse images of downward closures of basis elements are co-compact). Under these hypotheses, $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$5 possesses a finitary winning strategy that is uniform over $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$6's moves: a single finite set $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$7 of basis elements covers all replies, each decreasing the co-degree, so $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$8 wins within $$\alpha \circ \mathsf{log} = \mathsf{beh} \circ \alpha$$9 steps. In the bisimilarity instantiation, the dual game coincides with a coupling game, and the primal game recovers the classical Stirling-style bisimulation game.

Witness–strategy correspondence

The technical core is a pair of translations along the Galois connection, mediated by three auxiliary choice functions ($\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$0, $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$1, $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$2) whose existence follows from the basis/irreducibility lemmas:

• Witnesses to strategies: given a primal witness $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$3 for $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$4 and a finitary strategy for $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$5 in the way-below game on $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$6, one obtains a valid move $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$7 in the game on $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$8; every reply $\alpha(\mu\,\mathsf{log}) = \mu\,\mathsf{beh}$9 admits a sub-witness of strictly smaller degree. Dually, a dual witness yields a move for $\exists$0 in the dual game via $\exists$1.
• Strategies to witnesses: given a finitary winning strategy $\exists$2, the witness is defined inductively as $\exists$3, with $\exists$4; analogously for the dual game using $\exists$5.

An immediate consequence is that these are effective procedures: computing a distinguishing formula reduces to solving the local inequalities defining the strategy at each step.

Case studies

Three instantiations validate the framework. First, bisimilarity on finitely branching transition systems reproduces distinguishing formulas à la Cleaveland: the witness is a Hennessy-Milner formula built as $\exists$6 over recursively computed sub-witnesses, and the primal game mirrors the standard bisimulation game.

Second, behavioural metrics for labelled Markov chains under the Kantorovich lifting. Here $\exists$7, the logic side consists of sets of random variables closed under $\exists$8, truncated subtraction, and max, and $\exists$9 maps a set of functions to the sup-norm distance it induces. The instantiation of probabilistic systems into this Galois-connection framework is itself new relative to earlier work. Strategy computation reduces to linear programming over coupling polytopes: since optima are attained at vertices, one solves inequalities $\forall$0 subject to $\forall$1, yielding basis elements $\forall$2. A practical benefit noted by the authors: because the framework certifies strict lower bounds rather than exact distances, it sidesteps the known impossibility of witnessing exact behavioural distance by a single formula.

Third, a new case study on termination probabilities in Markov chains, where witnesses are trees: a tree rooted at $\forall$3 with children of distinct roots under-approximates the termination probability via the functional $\forall$4, and $\forall$5 maps a set of trees to the pointwise supremum of their values. The authors prove both that $\forall$6 under-approximates $\forall$7 and that $\forall$8, so the framework applies verbatim; the resulting certificates are explicit tree-shaped proofs of lower bounds on termination probability.

Limitations and open questions

Several assumptions carry real weight. The primal game's completeness fails without continuity of $\forall$9, as the counterexample shows, and the dual route requires co-properness of $b \in \mathbb{B}$0 — a condition whose verification is nontrivial even in the metric case, where it rests on a compactness argument the authors themselves flag for rechecking. Finitary strategies additionally require Scott-continuity of the behaviour function, which excludes systems where fixpoints are reached only at transfinite stages. The auxiliary functions $b \in \mathbb{B}$1, $b \in \mathbb{B}$2, $b \in \mathbb{B}$3 are defined by existence guarantees rather than algorithms, so concrete instantiations must supply effective choices. Open questions raised include the treatment of characteristic formulas, the use of the dual game on the logic side, connections to the codensity game (which would involve simultaneous play on both lattices), and links to recent proof-system-based approaches to apartness and lower-bound witnesses.

Conclusion

The paper provides a uniform, constructive account of witnesses for least-fixpoint statements over lattices, establishing exact correspondences between witnesses and finitary winning strategies in a primal and a dual fixpoint game, with termination guaranteed by ordinal degree measures. It subsumes known constructions for bisimilarity and probabilistic behavioural metrics and delivers a new certification method for termination probabilities in Markov chains, while making precise which domain-theoretic hypotheses (continuity, co-continuity, properness) each guarantee requires.