From: AAAI Technical Report FS-92-01. Copyright © 1992, AAAI (www.aaai.org). All rights reserved. DISTRIBUTED Dipartimento AI MODELS FOR PERCOLATION S.Bandini, G.Cattaneo, G.Tarantello di Scienze dell’Informazione - Universit£ email: bandini@hermes.mc.dsi.unimi.it Abstract The challenge of modelling complex physical situations by a qualitative approach represents a good platform for testing alternative or integrative methods to quantitative approaches. Weselected a well-knowntheoretical and physical problem as a field for studying classical solutions and their limits, proposing and testing a qualitative distributed approach based on molecular ontology. The problem is percolation and the physical situation we selected is coffee percolation phenomenon. Introduction Percolation is a complex phenomenon which has been studied from several theoretical, applicative and computational standpoints. Several formal tools have been tested or developed as possible solution to the problem of modeling and simulating it. Molecular ontology theory [1],[7], developed to obtain qualitative and intuitive computational models for simulating the behaviour of substances (fluids, electrons, etc.), presents some advantages in simulation tasks where a pure numerical approach presents intrinsic difficulties in terms of computational size or collecting boundary conditions. Then, we are studying percolation for testing our theory and comparing it with classical approaches. Percolation: the picture Percolation is the simplest model of disordered system whosetheoretical problems are not exactly solved [17]. It is an example of phase transition phenomenonand many real problems, like materials conductivity and magnetic properties, involve similar processes. According to a classical definition, given a square bidimensionai grid, percolation problem is that of determining the lowest concentration of filled places pc such that an unique infinite path of filled places can exist. Such a path is called percolation cluster and pc is called percolation threshold. Formulating the problem in this way *Supported by CNR(Prog. Fin. Sistemi Informatici Calcolo Paxallelo, sott. 7,OB3)and by Iliycaff$ S.p.A. di Milano we obtain site percolation; another possibility is bond percolation where the links, rather than the places, are filled. Besides these classic percolation models, several different models have been defined in order to represent particular situations. For instance, directed percolation, where a direction is assigned to the lattice links, is a good model for a fire spreading in a forest under a strong wind in one direction, or for a random network of resistors that can be crossed in only one way. In this case percolation threshold separates insulation from conduction [6]. Migration and sedimentation phenomena, like electrolysis, can be modeled by diffusion limited aggregation mechanisms [16] where a migrating particle is represented by a random walker jumping on the lattice from one site to another one (as for the Brownian movement), and adhering to the aggregate whenis adjacent to it. Finally, invasion percolation [13], finds its best application in the problem of washing out the petroleum rocks. Here, water spreads throughout the rock and petroleum comes out from the opposite side. The porous rock is modeled by a lattice where the sites are pores connected by thin channels (the links) and a random number, assigned to each link, represents channel resistance to water flowing: at each instant, water flows along the lowest resistance path, removing petroleum. Rest: a cup of coffee A more circumscribed, but no less complex percolation problem is "espresso" coffee percolation. A coffee producing company offered to us the very interesting instance of studying this process in terms of a scientific problem: several difficulties in espresso cooking simulation by classic numerical models suggested, as alternative, Artificial Intelligence qualitative methods. The beverage (the final espresso we drink) is obtained by a suitable percolation device where high temperature and hydrostatic pressure water flows throughout a given quantity of ground and rosted coffee particles (the "cake"). Each coffee particle is a piece of porous material containing air and small oil drops, so, our phenomenonis similar in some aspects to the petroleum case. Nevertheless, it is more complex owing to pro- From: AAAI Technical Report FS-92-01. Copyright © 1992, AAAI (www.aaai.org). All rights reserved. ceases like oil emulsion, glucide dissolution, the formation of air bubbles and particles compacting. Percolation: the tools Percolation as a static phenomenonis studied by numerical analysis in order to determine the exact value of percolation threshold: exact results have been obtained only in some particular cases like monodimensional systems or the Bethe lattice [17]. However, if we focus our attention on simulating real processes, as espresso percolation, we must adopt a dynamic point of view. One possibility in this direction comes from fractal theory [4],[14]. In fact, percolation clusters are often autosimilar on variable scales as are deterministic fractals (for example,Sierpinskji triangle [4]), thus, somebodystudies percolation using iterated function sets and considering fractal dimension [6]. On the other hand, a suitable non completely quantitative approach for modeling percolation in porous media [15],[9],[12], is obtained by cellular automata for fluid dynamics: the HPP[10] and FHP[8] Lattice Gas models respecting Navier-Stokes equation. Weare expanding these models to represent water flow throughout porous coffee particles with its chemical interactions. However, even if computational simulation of these models by means of CAM[18] showed the possibility of representing some aspects of a coffee cake, it was not possible to represent, in a qualitative way, important features of the knowledge about this phenomenon. Thus, we captured the suggestion of employing a local and distributed approach but integrating it by means of methods of knowledge- based simulation [5]. Percolation: lular automata for fluid dynamics suggested of adopting a local and distributed model and experimenting molecular ontology tool. an alternative Molecular ontology model [1],[7], developed in Qualitative Physics context [11], is defined in terms of a set of individual entities ("molecules"), lying on the sites of an adjacency graph, and changing their position as they react to some external qualitative fields [2]. They can communicate by messages exchange and link together to form aggregates [3]. Their actions are prescribed by a suitable set of local behaviour rules wich can embody, in a qualitative way, the knowledge of domain experts about espresso percolation. So, if a distributed set of molecules is used in order to model coffee particles and active water clusters, cake geometrical modification effect, determined essentially by high pressure presence and consisting of the local movements of coffee particles, can be obtained by a suitable communication between the molecules allowing them to knowif there are free neighbour sites, (containing air, for instance), to be occupied. Conclusions Weare investigating an AI qualitative approach in simulation of percolation. The advantages offered by cel- References [1] Bandini S., Cattaneo G., Stofella P., 1988. A Theory for Molecule Sructures: the Molecular Ontology Theory, Proc. of 2nd Workshop on Qualitative Physics, Paris. [2] Bandini S., Cattaneo G., Cordioli S., Vian G., 1989. A Qualitative Description of Field Theory in terms of Molecular Ontology, Proc. 1st National Conference of Italian Association of Artificial Intelligence, Trento. [3] Bandini S., Cattaneo G., Tarantello G., 1992. Messages as Qualitative Fields, Proc. of llth European Meeting on Cybernetics and Systems Research, Vienna. [4] Barnsley M., 1988. Fractals Everywhere, Academic Press. [5] Barr A.,1990. Knowledge Based Simulation, in A.Barr (ed) "The Handbookof Artificial Intelligence", Vol. IV, Addison Wesley, Readings, MA. [6] Bunde A., Havlin S. (Eds.), 1991. Fractals and Disordered Systems, Springer Verlag, Berlin. [7] Collins J., Forbus K.D., 1977. Reasoning about Fluids via Molecular Collections, Proc. AAAI-87, Seattle, WA. [8] Frish U., Hasslacher B., Pomeau Y., 1986. Lattice-Gas Automata for the Navier-Stokes Equation, Physical Review Letters, 56, 14. [9] Gunstensen K., Rothman H., 1991. A LatticeGas Model for 3 Immiscible Fluids, Physica D, 47. [10] Hardy J., Pomean Y., De Pazzis O., 1973. Time Evolution of a Two- Dimensional Model System. Invariant States and Time Correlation Functions, J. Math. Phys., 14, 12. [11] Hobbs J.R., Moore R.C., 1985. Formal Theories of the CommonsenseWorld, Ablex Pu.Co., NJ. [12] Janecky D., Wells J., ~l~ravis B., 1991. A Lattice Gas Automata Model for Heterogeneous Chemical Reactions at Mineral Surfaces and in Pore Networks, Physica D, 47. [13] Wilkinson D., Willemsen J., 1983. Invasion Percolation: a NewForm of Percolation Theory, J. Phys. A: Math. Gen. 16. [14] Mandelbrot B., 1983. The Fractal Geometry of Nature, W. H. Freeman and Co., NewYork. [15] Rothman H., 1988. Cellular-Automaton fluids: a Model for Flow in Porous Media, Geophysics, 53, 4. [16] Sander L., Witten T., 1981. Diffusion-Limited Aggregation, a Kinetic Critical Phenomena, Physical Review Letters, 47. [17] Stauffer D., 1985. Introduction to Percolation Theory, Taylor and Francis Inc., London. [18] Toffoli T., Margolus N., 1987. Cellular Automata Machines: a New Environment for Modeling, MIT Press, Cambridge, Massachussets.