The realization that relevant and impactful problems in science are in the interface of various disciplines gives rise to the science of complexity. In contrast with systems that can be understood by traditional reductionist approach, the premise of a complex system is that “the whole becomes not merely more, but very different from the sum of its parts (Anderson 1972)”. The basic features of a complex system are: 1) it is adaptive and constantly evolving and hence rarely approaches static equilibrium; 2) it has many components or agents that are strongly coupled and they are interacting in a non-linear and non-reductionist manner; 3) the interaction among units gives rise to emergent phenomena (self-organization) that operates in various spatial and temporal scales. Complex systems researches always attempt to pinpoint the underlying simplicity that gives rise to the seemingly complex dynamics.- C.P. Monterola
Yes. In fact, the ability of complex system modelling in deriving the cause end effect relationship is better then statistical and data modelling. However, if one is looking for simple causes to explain certain phenomena, then one may not be looking at a complex system. In fact, complex systems research differs from the traditional scientific research in that we do not hold the reductionist view of single cause per effect. Hence, there are likely to be many causes in order to build up to a single effect; for example, a stock market crash. Complex system modelling can find these multiple causes, but may not have the ability to reduce it to a simple form.
The emergence of phenomena that defies common sense is the key to complex systems, and often do not have a simple elegant cause and effect relationship. The goal of complex system model is to make these multiple causes apparent and the human experts still need to provide the interpretations and respond appropriately. - S. Kuo
We do not know unless similar phenomena has been observed in the physical world previously.
This question is often asked in relation to the prediction of tipping points; i.e. when the model predicts a mode of failure that was never seem before. In this scenario, it should be taken that there exists this mode of failure, but the exact condition for it to occur will not be accurately predicted by the model as it is not being calibrated by historical precedence. - S. Kuo
AnyLogic is a simulation tool for studying complex system, but it does not provide customer specific models except very generic ones. At A*STAR, we are looking into new modelling techniques, and the theory behind emergence due to complex interactions, which can be facilitated by using tools such as AnyLogic, but not limited by it.
Within IHPC, we are also focusing on achieving “high performance” in computation and simulation speed. Often, this cannot be achieved through commercial software. Hence what you see in our examples tend to be custom made simulations in general programming languages. - S. Kuo
No. In the early days, people tended to used ODE and state machine based models. These are often referred to as Dynamical Systems model. These models can be solved using Continuous methods such as Euler's methods, or approximated using discrete event simulations.
As the system under consideration becomes less homogeneous, the state space (number of variables in the equation) grows to the extent that writing down the state transitions or the ODE equations, becomes too cumbersome and no longer provide any insights. This is when agent based models provide a more direct way to model the system by simply allowing the researcher to specify the agent behaviour and to observe the phenomena without needing to formulate the system into a large set of equations.
A more specialised, and less used technique is system dynamics (not to be confused with Dynamical System) modelling. This method is often used to describe information flow or some abstract phenomena and is commonly used by those with an process engineering background. Specialised simulation tools are used for this purpose such as Simulink.
In many cases, it is necessary to use the above models together to form a larger simulation. It could be that the output of one type of simulation drives the parameter of another. For example, using an agent based model to estimate a hidden parameter to be used in a dynamical system model. It could also be a hierarchical relationship, such as an agent based model in which the agents represents a specific community whose behaviour is modelled by the system dynamics description. - S. Kuo
Yes transportation dynamics is a complex system. For purpose of illustration let us use the Singapore's MRT system and identify its complex system features (link to: Q: What is a complex system and how it differs from traditional science?)
Singapore's MRT system has many interacting components and agents, it has about 1.25 M passengers daily utilizing 122 stations. Agents are strongly coupled with about 2 million journeys/day using multiple routes. It is highly nonlinear since travel delays are correlated and crowd dynamics play a role to one's travel decisions. There are many emergent phenomena as the system is shown to approach tipping points such that travel delays after certain critical loads increases exponentially. The Singapore's MRT also constantly evolves since new lines are added and the agents behavior adapt and evolve dependent of these new developments.