Abstract
Simulation and bisimulation relations define pre-orders on processes which serve as the basis for approximation based verification techniques, and have been extended towards the design of continuous and hybrid systems with complex logic specifications. We study pre-orders between hybrid systems which preserve stability properties with respect to input. We show that these properties are not bisimulation invariant, and hence propose stronger notions which strengthen simulation and bisimulation relations with uniform continuity constraints. We show that uniform continuity is necessary on the relations corresponding to both the state-space and the input-space, and continuity itself does not suffice. Finally, we demonstrate the satisfiability of our definitions by casting the well-known Lyapunov function based techniques for stability analysis as constructing a simple one-dimensional system which is stable and uniformly continuously simulates the original system.
Similar content being viewed by others
References
Angeli D (2002) A Lyapunov approach to incremental stability properties. IEEE Trans Autom Control 47(3):410–421
Aubin JP, Frankowska H (1990) Set-valued analysis. Birkhauser, Boston
Cai C, Teel AR (2009) Characterizations of input-to-state stability for hybrid systems. Syst Control Lett 58(1):47–53
Caspi P, Benveniste A (2002) Toward an approximation theory for computerised control. In: EMSOFT, pp 294–304
Clarke JrE M, Grumberg O, Peled DA (1999) Model checking. MIT Press, Cambridge
Cuijpers PJL (2007) On bicontinuous bisimulation and the preservation of stability. In: HSCC, pp 676– 679
Davoren JM (2009) Epsilon-tubes and generalized skorokhod metrics for hybrid paths spaces. In: HSCC, pp 135–149
Girard A, Pola G, Tabuada P (2010) Approximately bisimilar symbolic models for incrementally stable switched systems. IEEE Trans Autom Control 55(1):116–126
Goebel R, Sanfelice R, Teel A (2009) Hybrid dynamical systems. IEEE Control Systems Magazine 29:28–93
Heemels WPMH, Weiland S (2008) Input-to-state stability and interconnections of discontinuous dynamical systems. Automatica 44(12):3079–3086
Henzinger TA (1996) The theory of hybrid automata. In: LICS, pp 278–292
Hespanha JP, Morse AS (1999) Stability of switched systems with average dwell-time. In: IEEE CDC, pp 2655–2660
Hespanha JP, Liberzon D, Teel AR (2008) Lyapunov conditions for input-to-state stability of impulsive systems. Automatica 44(11):2735–2744
Kaynar DK, Lynch NA, Segala R, Vaandrager FW (2003) Timed I/O automata: a mathematical framework for modeling and analyzing real-time systems. In: IEEE RTSS, pp 166–177
Khalil HK (1996) Nonlinear systems. Prentice-Hall Inc, Upper Saddle River
Lee D, Yannakakis M (1992) Online minimization of transition systems (extended abstract). In: STOC, pp 264–274
Li Y, Sanfelice RG (2015) On necessary and sufficient conditions for incremental stability of hybrid systems using the graphical distance between solutions. In: IEEE CDC, pp 5575–5580
Liu J, Liu X, Xie WC (2012) Class-kl estimates and input-to-state stability analysis of impulsive switched systems. Syst. Control Lett. 61:738–746
Milner R (1989) Communication and concurrency. Prentice-Hall, Inc, Upper Saddle River
Müller MA, Liberzon D (2012) Input/output-to-state stability and state-norm estimators for switched nonlinear systems. Automatica 48(9):2029–2039
Papachristodoulou A, Prajna S (2002) On the construction of Lyapunov functions using the sum of squares decomposition. In: IEEE CDC, pp 3482–3487
Parrilo PA (2000) Structured semidefinite programs and semialgebraic geometry methods in robustness and optimization. Doctoral dissertation, California Institute of Technology
Postoyan R, Biemond J, Heemels W, Van de Wouw N (2015) Definitions of incremental stability for hybrid systems. In: IEEE CDC, pp 5544–5549
Prabhakar P, Soto MG (2013) Abstraction based model-checking of stability of hybrid systems. In: CAV, pp 280–295
Prabhakar P, Dullerud GE, Viswanathan M (2012) Pre-orders for reasoning about stability. In: HSCC, pp 197–206
Prabhakar P, Liu J, Murray RM (2013) Pre-orders for reasoning about stability properties with respect to input of hybrid systems. In: EMSOFT, pp 1–10
Sontag E, Wang Y (1995) On characterizations of the input-to-state stability property. Syst. Control Lett. 24(5):351–359
Sontag ED (2006) Input to state stability: Basic concepts and results. In: Nonlinear and Optimal control Theory. Springer, Berlin, pp 163–220
Tabuada P, Balkan A, Caliskan SY, Shoukry Y, Majumdar R (2012) Input-output robustness for discrete systems. In: EMSOFT, pp 217–226
Vu L, Chatterjee D, Liberzon D (2007) Input-to-state stability of switched systems and switching adaptive control. Automatica 43(4):639–646
Acknowledgements
This work was partially supported by NSF CAREER award no. 1552668 to Pavithra Prabhakar and NSERC Canada Discovery Grant no. RGPIN-2016-04139 and the Canada Research Chairs (CRC) Program to Jun Liu.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Proof of the super-position Theorem 1
Proof
δ ISS ⇒ (C1) − (C3): It is straightforwardto check that δ ISS implies conditions (C1) − (C3).In fact, choosing u 1 = u 2 = u and δ such that β(δ, 0) < 𝜖 in Eq. 3implies
provided that |ζ 1 − ζ 2| < δ. This shows (C1)is true. Moreover,since
as t →∞, for anygiven 𝜖 and ζ 1,ζ 2, we can choose T independent of u suchthat
for all t > T. This shows (C2)is true. Finally,choosing ζ 1 = ζ 2 = ζ and δ such that γ(δ) < 𝜖 in Eq. 3implies
providedthat | |u 1 −u 2| | ∞ < δ. Thisshows (C3)is true.
(C1) − (C3) ⇒ δ ISS: The prooffor the opposite implication essentially follows from the proof of Lemma 4.5 in Khalil (1996). Therefore, thedetailed argument is omitted and the following is an outline of the proof. First, by (C3), we can prove thereexists a \({\mathcal {K}_{\infty }}\)function γ such that
holds for all initial states ζ 1,ζ 2.Second, conditions \((C1)\)and (C2)imply thatthere exist a \({\mathcal {K}\mathcal {L}}\)function β such that
holds for all input trajectory u. Now given any pair of initial states ζ 1, ζ 2and any pair ofinput trajectories u 1, u 2,it follows from Eqs. 14and 15that
This completes the proof. □
Appendix B: Proof of Theorem 4
Proof
Let us assume \(\mathcal {H}_{2}\)is hISS with respect to \({\mathcal {T}}_{2}\).We need to show that \(\mathcal {H}_{1}\)is hISS with respect to \({\mathcal {T}}_{1}\).We will show that \(\mathcal {H}_{1}\)satisfies conditions (D1) − (D3).Proof of satisfaction of Condition (D1) Let us fix an 𝜖 1 > 0.We need to find a δ 1 > 0such that Condition (D1)holds in \(\mathcal {H}_{1}\)and \({\mathcal {T}}_{1}\).Let 𝜖 2be the uniformity constant of \(R^{-1}_{1}\)corresponding to 𝜖 1.Let δ 2be the constant satisfying Condition (D1)for \(\mathcal {H}_{2}\)corresponding to 𝜖 2.Set δ 1to be the uniformity constant of R 1corresponding to δ 2.
Let us fix an input signal σ u 1and states \(\eta \in {\textit {First}({{{\textit {Exec}}(\mathcal {H}_{1})}|_{{\sigma ^{u}}_{1}}})}\)and \(\hat {\eta }_{1} \in {\textit {First}({{{\mathcal {T}}_{1}}|_{{\sigma ^{u}}_{1}}})}\)such that \({d^{s}}(\hat {\eta }_{1}, \eta _{1}) < \delta _{1}\).Let σ s 1be such that \(({\sigma ^{u}}_{1}, {\sigma ^{s}}_{1}) \in {\textit {Exec}}(\mathcal {H}_{1})\)and First(σ s 1) = η 1.We need to find \({\hat {\sigma }^{s}}_{1}\)such that \(({\sigma ^{u}}_{1}, \hat {\sigma }_{1}^{s}) \in {\mathcal {T}}\),\({\textit {First}({\hat {\sigma }^{s}}_{1})} = \hat {\eta }_{1}\)and \({d^{s}}({\sigma ^{s}}_{1}, \hat {\sigma }_{1}^{s}) < \epsilon _{1}\).
First, we establish that R 2(σ u 1)is a singleton set. This follows from (A3).Let R 2(σ u 1) = {σ u 2}.Hence, from (A1),we have that there exists \(\hat {\eta }_{2} \in {\textit {First}({{{\mathcal {T}}_{2}}|_{{\sigma ^{u}}_{2}}})}\)such that \(R_{1}(\hat {\eta }_{1}, \hat {\eta }_{2})\).From uniform continuity, we have \(R_{1}(B_{\delta _{1}}(\eta _{1})) \subseteq B_{\delta _{1}}(R_{1}(\eta _{1}))\).Hence, \(\hat {\eta }_{2} \in B_{\delta _{2}}(\eta _{2})\)for some η 2 ∈ R 1(η 1).From the definition of input simulation and uniqueness of the choice of σ u 2,we have that there exists σ s 2such that First(σ s 2) = η 2,\(({\sigma ^{u}}_{2}, {\sigma ^{s}}_{2}) \in {\textit {Exec}}(\mathcal {H})\)and R 1(σ s 1,σ s 2).Then, from the hISS of \(\mathcal {H}_{2}\)with respect to \({\mathcal {T}}_{2}\),we obtain that there exists \(({\sigma ^{u}}, \hat {\sigma }_{2}^{s}) \in {\mathcal {T}}_{2}\)such that \({\textit {First}({\hat {\sigma }^{s}}_{2})} = \hat {\eta }_{2}\)and \({d^{s}}(\hat {\sigma }_{2}^{s}, {\sigma ^{s}}_{2}) < \epsilon _{2}\).From (A2),there exists \(\hat {\sigma }_{1}^{s}\)such that \({\textit {First}(\hat {\sigma }_{1}^{s})} = \hat {\eta }_{1}\),\(({\sigma ^{u}}_{1}, \hat {\sigma }_{1}^{s}) \in {\mathcal {T}}_{1}\)and \(R_{1}({\hat {\sigma }^{s}}_{1}, {\hat {\sigma }^{s}}_{2})\).
It remains to show that \({d^{s}}({\sigma ^{s}}_{1},\hat {\sigma }_{1}^{s}) < \epsilon _{1}\).Note that \({d^{s}}({\sigma ^{s}}_{2}, \hat {\sigma }_{2}^{s}) < \epsilon _{2}\)and \(R_{1}^{-1}(\hat {\sigma }_{2}^{s}) = \{\hat {\sigma }_{1}^{s}\}\)(a singleton, from (A4)).To show that \({d^{s}}({\sigma ^{s}}_{1},{\hat {\sigma }^{s}}_{1}) < \epsilon _{1}\),we need to show that \({d^{s}}({\mathit {gr}}({\sigma ^{s}}_{1}),{\mathit {gr}}(\hat {\sigma }_{1}^{s})) < \epsilon _{1}\).Consider (t 1,i 1,x 1) ∈g r(σ s 1),we need to find \((t^{\prime }_{1}, i^{\prime }_{1}, x^{\prime }_{1}) \in {\mathit {gr}}(\hat {\sigma }_{1}^{s})\)such that d s((t 1,i 1,x 1), (t1′,i1′,x1′)) < 𝜖 1.Since R 1(σ s 1,σ s 2),there exists (t 2,i 2,x 2) ∈g r(σ s 2),such that R 1((t 1,i 1,x 1), (t 2,i 1,x 2)).Note, in fact, that \(t_{1} = t_{2}\)and i 1 = i 2,and R 1(x 1,x 2).Since, \({d^{s}}({\sigma ^{s}}_{2}, \hat {\sigma }_{2}^{s}) < \epsilon _{2}\),there exists \((t^{\prime }_{2}, i^{\prime }_{2}, x^{\prime }_{2}) \in {\mathit {gr}}({\hat {\sigma }^{s}}_{2})\)such that d s((t 2,i 2,x 2), (t2′,i2′,x2′)) < 𝜖 2.Then, from the continuity of \(R_{1}^{-1}\), (t 2,i 2,x 2)is within 𝜖 1of some element in \(R_{1}^{-1}(t^{\prime }_{2}, i^{\prime }_{2}, x^{\prime }_{2})\).But, since, the latter is a singleton set (from (A4)),namely, (t1′,i1′,x1′),we obtain that d s((t 1,i 1,x 1), (t2′,i2′,x2′)) < 𝜖 1.Note that \(t^{\prime }_{1} = t^{\prime }_{2}\),and \(i^{\prime }_{1} = i^{\prime }_{2}\).The other part where given \((t^{\prime }_{1}, i^{\prime }_{1}, x^{\prime }_{1}) \in {\mathit {gr}}(\hat {\sigma }_{1}^{s})\),we need to find (t 1,i 1,x 1) ∈g r(σ s 1)such that d s((t 1,i 1,x 1), (t1′,i1′,x1′)) < 𝜖 1is similar.Proof of satisfaction of Condition (D2) Let us fix an \(\delta _{1}, \epsilon _{1} > 0\)and T 1 ≥ 0,such that Condition (D2)holds in \(\mathcal {H}_{1}\)and \({\mathcal {T}}_{1}\).Let 𝜖 2be the uniformity constant of \(R^{-1}_{1}\)corresponding to 𝜖 1.Let δ 2be the constant satisfying Condition (D1)for \(\mathcal {H}_{2}\)corresponding to 𝜖 2.Set δ 1to be the uniformity constant of R 1corresponding to δ 2.Let T 2 = T 1.We will show that Condition (D2)holds in \(\mathcal {H}_{2}\)and \({\mathcal {T}}_{2}\)for \(\delta _{1}, \epsilon _{2}\)and T 2.
The proof is similar to that of Condition (D1).Here we need to show that \({d^{s}}({{{\sigma ^{s}}_{1}}|_{T_{1}}},{\hat {\sigma }_{1}^{s}|_{T_{1}}}) < \epsilon _{1}\)instead of \({d^{s}}({\sigma ^{s}}_{1},\hat {\sigma }_{1}^{s}) < \epsilon _{1}\).Note that in the previous proof the choice of the times whenwe move from one system to the other are the same. That is,\(t_{1} = t_{2}\)and t1′ = t2′.We mainly need to check if the triples being chosen at different stepsbelong to the time restricted signals rather than the complete signalsas required in the previous proof. Hence, if we consider the triple\((t_{1}, i_{1}, x_{1}) \in {\mathit {gr}}({{{\sigma ^{s}}_{1}}|_{T_{1}}})\),the corresponding triple (t 2,i 2,x 2)will belong to \({\mathit {gr}}({{{\sigma ^{s}}_{1}}|_{T_{1}}})\),since, T 1 = T 2and the time stamp t 2 ≥ T 2.Here, we can choose the triple \((t^{\prime }_{2}, i^{\prime }_{2}, x^{\prime }_{2})\)such that its time stamp \(t^{\prime }_{2} \geq T_{2}\),because we have \({d^{s}}({{{\sigma ^{s}}_{2}}|_{T_{2}}},{\hat {\sigma }_{2}^{s}|_{T_{2}}}) < \epsilon _{2}\)instead of \({d^{s}}({\sigma ^{s}}_{2},\hat {\sigma }_{2}^{s}) < \epsilon _{2}\).Finally, the triple (t1′,i1′,x1′)will have time stamp \(t^{\prime }_{1} = t^{\prime }_{2} \geq T_{2} = T_{1}\).Hence, \((t^{\prime }_{1}, i^{\prime }_{1}, x^{\prime }_{1}) \in {\mathit {gr}}({\hat {\sigma }_{1}^{s}|_{T_{1}}})\).Proof of satisfaction of Condition (D3) Let us fix an 𝜖 1 > 0.We need to find a δ 1 > 0such that Condition (D3)holds in \(\mathcal {H}_{1}\)and \({\mathcal {T}}_{1}\).Let 𝜖 2be the uniformity constant of \(R^{-1}_{1}\)corresponding to 𝜖 1.Let δ 2be the constant satisfying Condition (D3)for \(\mathcal {H}_{2}\)corresponding to 𝜖 2.Set δ 1to be the uniformity constant of R 2corresponding to δ 2.
Let us fix input signals \({\sigma ^{u}}_{1}, \hat {\sigma }_{1}^{u}\),state \(\eta _{1} \in {{{\mathcal {T}}_{1}}|_{{\hat {\sigma }^{u}}}}\).Let \(d({\sigma ^{u}}_{1}, \hat {\sigma }_{1}^{u}) < \delta _{1}\)and \(({\sigma ^{u}}_{1}, {\sigma ^{s}}_{1}) \in {\textit {Exec}}(\mathcal {H})\)with First(σ s 1) = η 1.We need to find \(\hat {\sigma }_{1}^{s}\)such that \((\hat {\sigma }_{1}^{u}, \hat {\sigma }_{1}^{s}) \in {\mathcal {T}}_{1}\),\({\textit {First}(\hat {\sigma }_{1}^{s})} = \eta _{1}\)and \({d^{s}}({\sigma ^{s}}_{1}, \hat {\sigma }_{1}^{s}) < \epsilon _{1}\).
Since \(\eta _{1} \in {{{\mathcal {T}}_{1}}|_{{\hat {\sigma }^{u}}_{1}}}\),we obtain from (A1)that there exists \(\hat {\sigma }_{2}^{u}\)and \(\eta _{2} \in {{{\mathcal {T}}_{2}}|_{{\hat {\sigma }^{u}}_{2}}}\)such that \(R_{1}(\eta _{1}, \eta _{2})\)and \(R_{2}(\hat {\sigma }_{1}^{u}, \hat {\sigma }_{2}^{u})\).Again, since, R 1(η 1,η 2),from the definition of simulation, there exists \(({\sigma ^{u}}_{2}, {\sigma ^{s}}_{2}) \in {\textit {Exec}}(\mathcal {H}_{2})\)such that First(σ s 2) = η 2, R 1(σ s 1,σ s 2)and R 2(σ u 1,σ u 2).From the choice of δ 1,we have that \({\sigma ^{u}}_{2} \in B_{\delta _{1}}(R_{2}(\hat {\sigma }_{1}^{u}))\).However, since, \(R_{2}(\hat {\sigma }_{1}^{u})\)is unique, namely, \(\hat {\sigma }_{2}^{u}\),we obtain that \({d^{u}}(\hat {\sigma }_{2}^{u}, {\sigma ^{u}}_{2}) < \delta _{2}\).Therefore, from the hISS of \(\mathcal {H}_{2}\)with respect to \({\mathcal {T}}_{2}\),there exists \(\hat {\sigma }_{2}^{s}\)such that \((\hat {\sigma }_{2}^{u}, \hat {\sigma }_{2}^{s}) \in {\mathcal {T}}_{2}\),\({\textit {First}(\hat {\sigma }_{2}^{s})} = \eta _{2}\)and \({d^{s}}({\sigma ^{s}}_{2}, \hat {\sigma }_{2}^{s}) < \epsilon _{2}\).Finally, from (A2),there exists \({\hat {\sigma }^{s}}_{1}\)such that \({\textit {First}(\hat {\sigma }_{1}^{s})} = \eta _{1}\),\(R_{1}(\hat {\sigma }_{1}^{s}, \hat {\sigma }_{2}^{s})\)and \((\hat {\sigma }_{1}^{u}, \hat {\sigma }_{1}^{s}) \in {\mathcal {T}}_{1}\).
It remains to show that \({d^{s}}({\sigma ^{s}}_{1}, {\hat {\sigma }^{s}}_{1}) < \epsilon \).The argument is the same as that in the proof of part D1. □
Rights and permissions
About this article
Cite this article
Prabhakar, P., Liu, J. & Murray, R.M. Simulations and bisimulations for analysis of stability with respect to inputs of hybrid systems. Discrete Event Dyn Syst 28, 349–374 (2018). https://doi.org/10.1007/s10626-017-0262-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10626-017-0262-9