Abstract
Causal Graph Dynamics extend Cellular Automata to arbitrary, bounded-degree, time-varying graphs. The whole graph evolves in discrete time steps, and this global evolution is required to have a number of physics-like symmetries: shift-invariance (it acts everywhere the same) and causality (information has a bounded speed of propagation). We study a further physics-like symmetry, namely reversibility. We extend a fundamental result on reversible cellular automata by proving that the inverse of a causal graph dynamics is a causal graph dynamics. We also address the question of the evolution of the structure of the graphs under reversible causal graph dynamics, showing that any reversible causal graph dynamics preserves the size of all but a finite number of graphs.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Arrighi, P., Dowek, G.: Causal graph dynamics. In: Czumaj, A., Mehlhorn, K., Pitts, A., Wattenhofer, R. (eds.) ICALP 2012, Part II. LNCS, vol. 7392, pp. 54–66. Springer, Heidelberg (2012)
Arrighi, P., Martiel, S., Nesme, V., Cayley, G.: Graphs, cellular automata over them submitted (long version) (2013). Pre-print arXiv:1212.0027
Arrighi, P., Nesme, V., Werner, R.: Unitarity plus causality implies localizability. J. Comput. Syst. Sci. 77, 372–378 (2010). QIP 2010 (long talk)
Arrighi, P., Martiel, S., Perdrix, S.: Block representation of reversible causal graph dynamics. In: Kosowski, A., Walukiewicz, I. (eds.) FCT 2015. LNCS, vol. 9210, pp. 351–363. Springer, Heidelberg (2015)
Boehm, P., Fonio, H.R., Habel, A.: Amalgamation of graph transformations: a synchronization mechanism. J. Comput. Syst. Sci. 34(2–3), 377–408 (1987)
Danos, V., Laneve, C.: Formal molecular biology. Theoret. Comput. Sci. 325(1), 69–110 (2004). Computational Systems Biology
Durand-Lose, J.O.: Representing reversible cellular automata with reversible block cellular automata. Discret. Math. Theoret. Comput. Sci. 145, 154 (2001)
Ehrig, H., Lowe, M.: Parallel and distributed derivations in the single-pushout approach. Theoret. Comput. Sci. 109(1–2), 123–143 (1993)
Ferrari, G.-L., Hirsch, D., Lanese, I., Montanari, U., Tuosto, E.: Synchronised hyperedge replacement as a model for service oriented computing. In: Boer, F.S., Bonsangue, M.M., Graf, S., Roever, W.-P. (eds.) FMCO 2005. LNCS, vol. 4111, pp. 22–43. Springer, Heidelberg (2006)
Gromov, M.: Endomorphisms of symbolic algebraic varieties. J. Eur. Math. Soc. 1(2), 109–197 (1999)
Hasslacher, B., Meyer, D.A.: Modelling dynamical geometry with lattice gas automata. In: Expanded Version of a Talk Presented at the Seventh International Conference on the Discrete Simulation of Fluids Held at the University of Oxford, June 1998
Hedlund, G.A.: Endomorphisms and automorphisms of the shift dynamical system. Math. Syst. Theor. 3, 320–375 (1969)
Kari, J.: Reversibility of 2D cellular automata is undecidable. In: Cellular Automata: Theory and Experiment, vol. 45, pp. 379–385. MIT Press (1991)
Kari, J.: Representation of reversible cellular automata with block permutations. Theor. Comput. Syst. 29(1), 47–61 (1996)
Kari, J.: On the circuit depth of structurally reversible cellular automata. Fundamenta Informaticae 38(1–2), 93–107 (1999)
Klales, A., Cianci, D., Needell, Z., Meyer, D.A., Love, P.J.: Lattice gas simulations of dynamical geometry in two dimensions. Phys. Rev. E. 82(4), 046705 (2010)
Konopka, T., Markopoulou, F., Smolin, L.: Quantum graphity. Arxiv preprint arXiv:hep-th/0611197 (2006)
Morita, K.: Reversible simulation of one-dimensional irreversible cellular automata. Theoret. Comput. Sci. 148(1), 157–163 (1995)
Sorkin, R.: Time-evolution problem in Regge calculus. Phys. Rev. D. 12(2), 385–396 (1975)
Taentzer, G.: Parallel and distributed graph transformation: formal description and application to communication-based systems. Ph.D. thesis, Technische Universitat Berlin (1996)
Taentzer, G.: Parallel high-level replacement systems. Theoret. Comput. Sci. 186(1–2), 43–81 (1997)
Tomita, K., Kurokawa, H., Murata, S.: Graph automata: natural expression of self-reproduction. Physica D: Nonlinear Phenom. 171(4), 197–210 (2002)
Acknowledgements
This work has been funded by the ANR-12-BS02-007-01 TARMAC grant, the ANR-10-JCJC-0208 CausaQ grant, and the John Templeton Foundation, grant ID 15619. The authors acknowledge enlightening discussions with Bruno Martin and Emmanuel Jeandel. This work has been partially done when PA was delegated at Inria Nancy Grand Est, in the project team Carte.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
A Proofs of Sections 4 and 5
A Proofs of Sections 4 and 5
1.1 A.1 Proofs of Section 4
Lemma 1
(Shift-equivalence classes isometry). Let \(X\in \mathcal{X}_{\varSigma ,\varDelta ,\pi }\) be a graph. If \(C_1\subseteq {V(X)}\) and \(C_2\subseteq {V(X)}\) are two shift-equivalence classes of X, then \(|C_1|=|C_2|\).
Proof
Consider two shift-equivalent and distinct vertices u and v in X. Consider a path w. The vertices u.w and v.w are shift-equivalent and distinct. More generally, if we have n shift-equivalent and distinct vertices \(v_1,...,v_n\), any vertex \(u=v_1.w\) will be shift-equivalent to \(v_2.w,...,v_n.w\) and distinct from all of them, hence the equivalence classes are all of the same size.\(\Box \)
Lemma 2
(Invertible CGD preserves shift-equivalence classes). Let \((F,R_\bullet )\) be a shift-invariant dynamics over \(\mathcal {X}_{\varSigma ,\varDelta ,\pi }\), such that F is a bijection. Then for any X and any \(u,v\in X\), \(u \approx v \) if and only if \(R_X(u) \approx R_X(v)\).
Proof
\(u\approx v\) expresses \(X_u=X_v\), which by bijectivity of F is equivalent to \(F(X_u)=F(X_v)\) and hence \(F(X)_{R_X(u)}=F(X)_{R_X(v)}\). This in turn is expressed by \(R_X(u) \approx R_X(v)\).\(\Box \)
Lemma 3
(Properties of primal extensions). Any primal extension \({}^\Box X\) is asymmetric.
Proof
As \({}^\Box X\) has a prime number of vertices, by Lemma 1, its has either one single equivalence class of maximal size or only trivial equivalence classes. As the primal extension adds at least two vertices and that these vertices have different degree (1 for the last vertex on the line, and 2 for its only neighbor), \({}^\Box X\) contains at least two non equivalent vertices, hence the first result.\(\Box \)
Theorem 1
(Invertible implies almost-vertex-preserving). Let \((F,R_\bullet )\) be a CGD over \(\mathcal {X}_{\varSigma ,\varDelta ,\pi }\), such that F is a bijection. Then there exists a bound p, such that for any graph X, if \(|X| > p\) then \(R_X\) is bijective.
Proof
When \(|\pi |\le 1\), \(\mathcal {X}_{\varSigma ,\varDelta ,\pi }\) is finite so the theorem is trivial. So we assume in the rest of the proof that \(|\pi |>1\).
[Finite graphs] First we prove the result for any finite graph. By contradiction, assume that there exists a sequence of finite graphs \((X(n))_{n\in \mathbb {N}}\) such that |X(n)| diverges and such that for all n, \(R_{X(n)}\) is not bijective. As this sequence is infinite, we have that one of the two following cases is verified an infinite number of n:
-
\(R_{X(n)}\) is not surjective,
-
\(R_{X(n)}\) is not injective.
\(\bullet \) \([R_{X(n)}\) not surjective]. There exists a vertex \(v' \notin {\text {im}}R_{X(n)}\). Without loss of generality, we can assume that \(|v'|<b\) where b is the bound from the boundedness property of F. We will now consider a particular primal extension of F(X(n)), \({}^\Box F(X(n))\), where the chosen vertex in F(X(n)) is the furthest away from the pointed vertex \(\varepsilon \). Indeed, if F(X(n)) is large enough, a vertex lying at maximal distance of \(\varepsilon \) in F(X(n)) either has a free port or is part of a cycle, and thus is a valid vertex to perform the primal extension. Indeed, if this vertex has no free port, then any of its edge can be removed without splitting the graph, as it would contradict its maximality – therefore it is in a cycle. Now, consider the graph \(Y(n)=F^{-1}({}^\Box F(X(n)))\). Using uniform continuity of \(F^{-1}\) and \(R_\bullet \), and the fact that |X(n)| is as big as we want, we have that there exists an index n and a radius r such that \(Y(n)^r=X(n)^r\) and \(R^b_{Y(n)^r}=R^b_{X(n)^r}\). As F(Y(n)) is asymmetric by construction, \(v'\in {\text {im}}R_{Y(n)^r}^b\) which contradicts \(v' \notin {\text {im}}R_{X(n)}\).
\(\bullet \) \([R_{X(n)}\) not injective]. There exist two vertices \(u,v \in X(n)\) such that \(R_{X(n)}(u)=R_{X(n)}(v)\) and \(u\ne v\). Without loss of generality, we can assume that \(u =\varepsilon \) as F is shift-invariant. According to Lemma 2, we have that \(\varepsilon \approx v\). Moreover, using the uniform continuity of \(R_\bullet \), we have that, as \(R_{X(n)}(v)=R_{X(n)}(\varepsilon )=\varepsilon \), there exists a radius l, which does not depend on n, such that \(|v| <l\) . Let us consider a primal extension of X(n), \({}^\Box X(n)\), where the primal extension has been performed at maximal distance from \(\varepsilon \), by the same argument as in the previous \(\bullet \). In this graph, \(\varepsilon \) and v are not shift-equivalent and thus, \(R_{{}^\Box X(n)}(\varepsilon )\ne R_{{}^\Box X(n)}(v)\) . By continuity of \(R_\bullet \), we have that there exists a radius \(r>l\) such that \(R_{{}^\Box X(n)^r}^0=R_{ X(n)^r}^0\) for a large enough n, hence \(R_{{}^\Box X(n)^r}^0(v)=R_{ X(n)^r}^0(v)=\varepsilon \), which contradicts \(R_{{}^\Box X(n)}(\varepsilon )\ne R_{{}^\Box X(n)}(v)\).
[Infinite graphs] Now we show that the result on finite graphs can be extended to infinite graphs, proving that for any infinite graph \(R_X\) is bijective:
\(\bullet \) \([R_X\) injective ]. By contradiction. Take X infinite such that there is \(u\ne v\) and \(R_X(u)=R_X(v)\). Without loss of generality we can take \(u=\varepsilon \), i.e. \(v\ne \varepsilon \) and \(R_X(v)=\varepsilon \). By continuity of \(R_\bullet \), there exists a radius r, which we can take larger than |v| and p, such that \(R_{X}=R_{X^r}\). Then \(R_{X^r}(v)=R_X(v)=\varepsilon \), thus \(R_{X^r}\) is not injective in spite of \(X^r\) being finite and larger than p, leading to a contradiction.
\(\bullet \) \([R_X\) surjective ]. By contradiction. Take X infinite such that there is \(v'\) in F(X) and \(v'\notin {\text {im}}R_X\). By boundedness, there exists \(u'\in F(X)\) such that \(u'\) lies at distance less than b of \(v'\). Using shift-invariance, we can assume without loss of generality that \(u'=\varepsilon \), hence, \(|v'|<b\). By continuity of \(R_\bullet \), there exists a radius r, which we can take larger than p, such that the images of \(R_{X}\) and \(R_{X^r}\) coincide over the disk of radius b. Then, \(v'\notin {\text {im}}R_X\) implies \(v'\notin {\text {im}}R_{X^r}\), thus \(R_{X^r}\) is not surjective in spite of \(X^r\) being finite and larger than p, leading to a contradiction.
1.2 A.2 Proofs of Section 5
Lemma 4
If \((F,R_\bullet )\) is an invertible, shift-invariant dynamics such that for all X, \(R_X\) is a bijection, then \((F^{-1},S_\bullet )\) is a shift-invariant dynamics, with \(S_{Y}=(R_{F^{-1}(Y)})^{-1}\).
Proof
Consider Y and \(u'.v'\in Y\). Take X and \(u.v\in X\) such that \(F(X)=Y\), \(R_X(u)=u'\) and \(R_X(u.v)=u'.v'\). We have: \({F^{-1}(Y_{u'})} ={F^{-1}(F(X)_{R_X(u)})}={F^{-1}(F(X_u))}={X_{(R_X)^{-1}(u')}}={F^{-1}(Y)_{S_Y(u')}}\). Moreover, take \(v\in X_u\) such that \(R_X(u.v)=R_X(u).R_{X_u}(v)=u'.v'\). We have: \(S_Y(u'.v') =(R_X)^{-1}(R_X(u.v)) =u.v =(R_{X})^{-1}(u').(R_{X_u})^{-1}(v')= S_Y(u').S_{Y_{u'}}(v')\). Â Â Â \(\square \)
Theorem 2
(Invertible implies reversible). If \((F,R_\bullet )\) is an invertible CGD, then \((F,R_\bullet )\) is reversible.
Proof
Continuity of \(F^{-1}\) is directly given by the continuity of F together with the compactness of \(\mathcal {X}_{\varSigma ,\varDelta ,\pi }\). Its boundedness derives either from the bijectivity of \(R_X\) for \(|X|>p\) or from the finiteness of X when \(|X|>p\).
We must construct \(S_\bullet \). For \(|F(X)|=|X|>p\), we know that \(R_X\) is bijective and we let \(S_{F(X)}=R_X^{-1}\). For \(|X|\le p\), we will proceed in two steps. First, we will construct an appropriate \(S_{F(X)}\) for X. Second, we will make consistent choices for \(S_{F(X)_{u'}}\) so that \(S_\bullet \) is shift invariant.
We write \(\tilde{u}\) for the shift-equivalence class of u in X. For all \(v'\in F(X)\), we make the arbitrary choice \(S_{F(X)}(\tilde{v'})=v\), where v is such that its image \(R_X(v)\) is shift equivalent to \(v'\) in F(X), i.e. \(R_X(v)\approx v'\). For this X, we have enforced \(\approx \)-compatibility. Then we make consistent choices for \(S_{F(X)_{u'}}\). This is obtained by demanding that \(S_{F(X)_{u'}}(\widetilde{\overline{u'}.v'})=\overline{u}.v\). Indeed, this accomplishes shift-invariance because \(S_{F(X)_{u'}}(v')=S_{F(X)_{u'}}(\overline{u'}.u'.v')=\varepsilon .v'=v'\) implying the equality: \(S_{F(X)}(u'.v')=u.v=S_{F(X)}(u').S_{F(X)_{u'}}(v')\). Moreover, \(S_{F(X)_{u'}}\) is itself shift-invariant because: \(S_{F(X)_{u'.v'}}(w')=S_{F(X)_{u'.v'}}(\overline{u'.v'}.u'.v'.w')=\overline{u.v}.u.v.w=w\), and \(S_{F(X)_{u'}}(v')=v\) implying that \(S_{F(X)_{u'}}(v'.w')=v.w=S_{F(X)_{u'}}(v').S_{F(X)_{u'.v'}}(w')\) , and \(\approx \)-compatible because \(v'\approx w'\) implies \(S_{F(X)_{u'}}(v')=S_{F(X)_{u'}}(w')\), and thus \(S_{F(X)_{u'}}(v')\approx S_{F(X)_{u'}}(w')\).
Continuity of the constructed \(S_\bullet \) is due to the continuity of \(R_\bullet \) and the finiteness of p.
Shift-invariance of \((F^{-1},S_\bullet )\) follows from \(\approx \)-compatibility of \(S_\bullet \) and shift-invariance of \((F,R_\bullet )\), because \(F^{-1}(F(X)_u')=X_v\) where v is such that \(R_X(v)\approx u'\), hence \(F^{-1}(F(X)_u')=X_{S_{F(X)}(u')}\). Â Â Â \(\square \)
Rights and permissions
Copyright information
© 2016 Springer International Publishing Switzerland
About this paper
Cite this paper
Arrighi, P., Martiel, S., Perdrix, S. (2016). Reversible Causal Graph Dynamics. In: Devitt, S., Lanese, I. (eds) Reversible Computation. RC 2016. Lecture Notes in Computer Science(), vol 9720. Springer, Cham. https://doi.org/10.1007/978-3-319-40578-0_5
Download citation
DOI: https://doi.org/10.1007/978-3-319-40578-0_5
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-40577-3
Online ISBN: 978-3-319-40578-0
eBook Packages: Computer ScienceComputer Science (R0)