Supported by EPSRC,
the University of Nottingham
has set up initial links with Imperial
College London with regards to research collaborations between
the two institutes.
In particular, Dr
Li Bai at the School
of Computer Science & IT, University of Nottingham, will
work with Professor
Henrik Jensen at the Department
of Mathematics, Imperial College on "Understanding Emergent
Behaviours in Complex Systems".
The project starts April,
2005.
Simulation of Complex Systems
- Swarm/PDE/Level Set models
We have implemented a termite
bound building simulation as an example of emergence in complex
systems. We have also implemented a PDE integrator and used Bonabeau's
PDE termite model as a test (with a lot of help from Matt at Imperial
College). The graphs show active materials at 60000 time steps in
the pillar formation experiment. The simulation takes quite a long
time to run.
In
order to simulate and analyse the construction process of natural
social agents, such as termites, there are a number of approaches
available, each with specific strengths and weaknesses. At the
top-end is a fully embodied simulation of all physical interactions
and thought processes. Naturally, for a complex system such as
a termite colony, this is infeasible with current technology.
Individual-based models are related
to cellular automata and feature simplified agents, who move in
a simplified (generally discretised) approximation of their real
environment. These models allow simulation of individual agents'
decision making processes; but because of the need to make the
simulation tractable the agent and environment are often over-simplified,
with parameters set for convenience over realism, making the model
only valid as a proof-of-concept at some high level of description.
In order to more accurately represent
spatial/temporal scales and natural processes, such as diffusion,
whilst still maintaining a high enough level of description to
make a simulation tractable, we can turn to partial differential
equation (PDE) models. PDE simulation models operate in a continuum,
with the distribution of a mass of substance modelled, as opposed
to the individual particles/items making up the mass. This means
that individual agent behaviour is not modelled; rather the overall
behaviour of the agents in an area. Also, physicality is hard
to represent, especially in such a complex system as termite construction,
so enforcing physical constraints is tricky.However PDE models
can, as stated, represent scale much more easily and can include
vector/scalar field-based physical processes,such as diffusion,
naturally.
As a first step towards a PDE simulation
of termite construction we turn to the system of equations presented
in (Bonabeau et al., 1998) for investigating the initiation of
basic structures involved in termite mound construction.Three
'substances'are modelled: cement pheromone, loaded termites and
active material.
For the cement pheromone at position x at time t,
H(x,t):
(1)
Where: k2, k4 and DH
are constants and P is the amount of ‘active’ (cement carrying)
material.
Only loaded termites (termites carrying soil) are modelled, denoted
by C(x,t):
(2)
Where: Φ, k1, DC and
γ are constants.
Finally, we only model the active material, P(x,t):
(3)
In
order to implement this system of equations we must numerically
integrate with respect to time over a given area of space. To
do this we first discretise the space by partitioning an area
of continuous space into a discrete lattice of a given resolution,
with each cell containing the 'smoothed-out' mass of substance
within that area. Then we asynchronously update each cell using
the equations above multiplied by some (small) time-step. This
is the basic methodology of numerical integration. Several expanded
algorithms exist to improve accuracy, but all at the cost of simulation
speed, which is already at a premium in a system such as this.
The below screenshots
show the state of active material in a 2D environment at t=0 and
t=100. The simulation was run on a 180x180 lattice with a cell
spacing of 0.01 and a time-step of 0.0003. All values were initialised
to 0, except for P, which was initialised to a random value such
that 0 < P(x,0) < 3. Other parameters were set as follows:
As can be seen from the graphs, the initial random distribution of
material becomes a series of smooth ‘piles’ as statistical fluctuations
mean certain sites attract more termites and are positively reinforced,
whereas others decay. There is a more-or-less regular distribution
of these proto-pillars, which is an emergent property of the system
created by the competition between neighbouring sites.
This system works in 2D, uses a minimal set of substances, does not
account for physical constraints or take into consideration many
facets of termite construction that may have a serious impact
on the system. However, the result is still encouraging, and with
little modification the equations can be used to explain the initiation
of many other base structures in termite construction (see Bonabeau
et al., 1998). Further development into a 3D system with more
factors taken into account could potentially be very rewarding,
though it would likely need the use of a great deal of computing
power. The possibility also exists of improving system efficiency
algorithmically and hybridizing the system with other simulation
methodologies (e.g. particle systems) to expand the capabilities
of the approach.
Simulation of Complex Systems
- Co-evolution of the Termites and Fungi
Note: click 'graph results', then 'population',
or'resources', etc.
Observations of the Simulation
The results for Population
behave as expected, the population will grow (or shrink if it starts
above a certain size) until it hits the ceiling value, it then fluctuates
around that level, generally as the population of one type of agent
decreases, the other increases. Though they will always rebalance
due to the limited resources.
The distribution of reproduction
probability starts off Gaussian distributed, but mutations lead
to a slow upward trend. This is due to the fact that those with
higher probabilities are more likely to reproduce, this trait is
passed onto the young, and so the overall probability increases.
It is important to note that the average probability for the system
approaches 1.
The distribution of resource
production probability fluctuates around the same point that is
starts. There is no advantage or disadvantage to individual agents
if this value increases or decreases. So no real evolution takes
place. It is important to note that because this is random it can
end up well above or below its starting value, purely by chance.
You can see in the graph for
resources that the resources grow exponentially to begin with. But
because of the imposed limit (due to waste) they fluctuate very
wildly around the maximum. The graph is actually very similar to
the graph for population but growth rate is faster. This causes
the fluctuations to be bigger and closer together.
A final important observation
is that after some time systems lose stability and the population
falls to zero. The reason for this seems to be that reproduction
probability gets very high, and resource production probability
gets very low. We have already seen that reproduction probability
grows, so this is not surprising. But that the production probability
decreases is surprising. The best explanation is that because the
resource production probability fluctuates randomly it will sometimes
fall much lower than its average. If there are enough agents in
the system then they will consume all the stockpiled resources and
die out before the probability has a chance to stabilise. This is
supported by the fact that lower mutation rates cause this to happen
less quickly on average. and that resource levels fall very low
just before the agents die out.
Activities
11/4/2005 - Meeting at Imperila
College
19-20/4/2005 - Disciplinary
Bridging in Nottingham
27/4/2005 - Dr Bai's Seminar
at Imperial College
15/6/2005 - Prof. Jensen's
siminar in Nottingham
We present a review of the
attempt within the Tangled Nature model to understand the effect
of evolution and interaction of ecological and evolutionary observables.
We report on the relation between the interaction structure in genotype
space and the resulting Species Abundance Distribution (SAD). Ecological
relevant SADs are only obtained if the genotype space allow for
a potential high connectivity between species. We also study the
relation between the degree of genotype interaction and species
diversity. Furthermore we include spatial degrees of freedom to
investigate the Species Area Relation from an evolutionary perspective.
Comparison with observed SARs is favourable and suggests that evolution
may be a fundamental factor in understanding the observed power
law-like SARs.
4/5/2005 - Vera:
Renormalization and Haar wavelets
in upscaling.
11/5/2005 - Matthew:
The random walker as a mean-field
SOC sandpile - the
fun you can have with analytical solutions.
17/5/2005 - Ole Peter:
Self-organized critical (SOC)
sandpile models are close relatives of fixed energy sandpiles displaying
absorbing-state phase transitions (AS) at a given energy density.
A recipe for designing equivalent AS models for given SOC models
was furnished by Dickman and coworkers in 1998. However, finite
size effects are not taken into account in their work. Including
these opens up a new series of questions about the exact nature
of the link between AS and SOC. I will present the AS mechanism
for attaining SOC and explain the puzzling questions raised by including
finite-size effects. Preliminary results from recent numerical work
addressing these questions will also be presented.
25/5/2005 - David:
Swarm agent construction.
......
Workshop
Stochasticity in multi-agent
systems (SMAS), Siguenza, Spain, June 2-4, 2005
SMAS
is a three days mini-workshop intended to strengthen the collaboration
between people working on topics closely related to the goals of
the STOCHDYN programme.
The main issue to be addressed
in the mini-workshop will be agent-based models with applications
in different fields (Biology, Economy, Social Sciences,...) with
a special focus on the role played by fluctuations in those systems
and the proper way to study it analytically, specifically how evolutionary
game theory can be extended to include fluctuation effects.
Conference
The European Conference on
Mathematical and Theoretical Biology (ECMTB), Dresden, Germany,
July 18-22, 2005
The ECMTB
conference aims to communicate the best of recent advances in mathematical
modelling approaches in the life sciences and to identify key targets
of future work.
Seminars
Isaac Newton Institute for
Mathematical Sciences - Pattern Formation in Large Domains
1 - 5 August: Training Course:
Pattern Formation in Large Domains (David Feltell received financial
support)
8 - 12 August: Developments
in Experimental Pattern Formation
19 - 23 September: Theoretical
Aspects of Pattern Formation
26 - 30 September: Theory and
Applications of Coupled Cell Networks
12 - 16 December: Pattern Formation
in Fluid Mechanics
Proposals
EPSRC Proposal on pattern formation
EPSRC Proposal on pattern recognition
as patternformation
EU proposal on simulating emergence
in complex systems
Papers
Swarm intelligence paper.
AI2005 paper.
BIOSIGNAL 2008 paper.
IJMIC 2008 paper.
|