Complex Systems Analysis and Modelling

Complex systems connect microscopic interactions to macroscopic patterns. Probability distributions describe fluctuations and extreme events, network models explain how connections form, and graph metrics quantify the resulting structure. The key tools are the central limit theorem, stable distributions, power laws, preferential attachment, and network algorithms.

A formula is only useful with its conditions: identify the random variable, graph type, and normalization convention before substituting numbers.

On smaller screens, wide equations and tables can be scrolled horizontally.

Topics

Topic	Key concepts
Introduction	Interactions, emergence, and modelling approaches
Central limit theorem	Probability and moments, MGFs, proof, and random walks
Scale-free distributions	Stable distributions, Cauchy, power laws, Pareto, earthquakes, and extreme events
Scale-free networks	BA continuum derivation, discrete master equation, ER model, and Poisson limit
Network fundamentals	Graph types, matrices, projections, and connectivity
Network metrics and algorithms	Centrality, PageRank, clustering, communities, and complexity
Working with NetworkX	Graph construction, attributes, function selection, and reproducible examples
Comprehensive practice	Eight problems, worked answers, and common mistakes
Helpsheet	Six groups of core formulas and conditions

Notation

Symbol	Meaning and possible ambiguity
$S_n=\sum_{i=1}^{n}X_i$, $\bar X_n=S_n/n$	Sum of random variables and sample mean
$\mu$, $\sigma^2$	Population mean and variance; the second parameter in $N(\mu,\sigma^2)$ is the variance
$M_X(t)$, $\varphi_X(q)$	Moment-generating function (MGF) and characteristic function (CF)
$\gamma$, $\alpha$	Power-law PDF and CCDF exponents satisfy $\gamma=1+\alpha$; the stable-family parameter $\alpha=2$ is the Gaussian exception. Some stable-distribution formulas instead use $\gamma$ for a positive scale parameter, as defined locally
$n$, $M$, $m$	Number of nodes, total number of edges, and edges introduced by each new BA node; distinguish the edge count $M$ from $M_X(t)$ by context
$A_{ij}$	A directed edge from $i$ to $j$: row sums give out-degrees and column sums give in-degrees
$p(x)$, $p_k$, $\boldsymbol p$	A continuous density, a discrete degree probability, and a PageRank probability vector, respectively

Introduction: Why Study Complex Systems?

From Individuals to the Whole: Where Complexity Arises

The defining feature of a complex system is more than a large number of parts: interactions among many individuals produce structures and behaviors that an isolated individual does not possess. Four useful characteristics are many-body, autonomous, interaction, and emergence. These refer to many constituent units, decentralized autonomous activity, interactions that often involve nonlinear effects, and the resulting patterns or collective behavior. They provide a framework for recognizing complex systems rather than mathematical axioms that every such system must satisfy exactly.

Autonomy means that overall order need not be designed step by step by a central controller. Emergence means that new observable regularities appear at the collective level; knowing the properties of an individual constituent is not enough to read off the behavior of the whole.

Example	Local mechanism or individual level	Collective phenomenon illustrated
Army ants and ant rafts	Individuals coordinate through local information, including contact with nearby individuals	Self-organization, resource efficiency, and collective capabilities
Bird flocks	Individuals adjust their motion and influence one another	Large-scale synchronization and collective behavior
Fireflies and applause	Individual oscillations or rhythms are influenced by others	Spontaneous synchronization and pattern formation
Stadium waves	Responses to nearby participants and the number of people participating	A critical number of participants and a collective response resembling a phase transition
Traffic, cities, and epidemics	Networks of people, roads, contacts, and feedback	Congestion, spatial organization, and spreading

The Uses and Limits of Reductionism

Reductionism seeks to understand the whole through its components and their interactions. It provides a starting point for microscopic models. Complex systems, however, involve correlations, feedback, and behavior across scales, so simply adding the results for isolated individuals linearly does not explain the whole.

For example, knowing how a single bird flies does not directly predict how a flock turns; we must also specify how each bird responds to other birds. Physical approaches seek common regularities across systems whose details differ, a property known as universality. This usually requires translating the question into measurable macroscopic quantities, such as distribution tails, the degree of synchronization, network connectivity, or clustering.

How to structure a short answer: Identify the individuals, explain their interactions, identify the collective phenomenon, and explain why the properties of an isolated individual are insufficient to determine that phenomenon. Saying only that “the system is large and complicated” is not enough.

Modelling Tools and Their Connections

• Agent-based models: Specify individual states and rules for actions or interactions, then observe the aggregate behavior. A cellular automaton usually places states on discrete lattice sites and updates them according to neighborhood rules.

• Complex networks: Represent individuals as nodes and interactions as edges. Distributions describe what the data look like, while networks express who interacts with whom. The same local rules can produce different macroscopic outcomes on different connection structures.

• Nonlinear dynamics and chaos: Nonlinearity allows a response to be disproportionate to its input. Chaos concerns motion in deterministic systems that is highly sensitive to initial conditions. Complex systems may be ordered or chaotic; “complex” does not mean “random” or “chaotic.”

• Disorder, heterogeneity, and pattern formation: Real systems differ in their connections, individual properties, or coupling strengths. The question is how these differences affect the emergence of order.

• Phase transitions and criticality: Changing a control parameter can qualitatively change a system’s macroscopic behavior. Statistical physics provides tools for studying macroscopic phases through microscopic interactions; complex-systems research extends such methods to irregular, heterogeneous structures.

• Self-organized criticality (SOC): Some driven systems may approach a critical state through their own dynamics. A sandpile provides an intuitive example. Observing a power law does not by itself prove that a system exhibits SOC.

• Power laws, heavy tails, and scale-free behavior: Large events can be more common than a normal model predicts, suggesting that the assumptions behind a normal approximation may fail. A power law is a testable statistical regularity; it does not automatically identify a unique microscopic mechanism.

A Unifying Line of Reasoning

If many contributions can reasonably be treated as an additive combination of independent, identically distributed variables with finite variance, the CLT supplies a normal baseline prediction. When data depart substantially from that prediction, examine independence, the finiteness of the variance, whether the variable is actually a sum, and whether sampling mixes different mechanisms. Stable distributions and power laws offer another possible class of statistical structures, while growth and preferential attachment provide a specific mechanism that can generate a power law in a network. Network metrics then turn a complicated connection pattern into quantities that can be calculated and compared.

Concept check: If a group of people synchronizes while watching the same external metronome, has self-organization been demonstrated?

Answer: No. Synchronization may result from a shared external drive. To establish self-organization, we need to examine how interactions within the group produce order. This distinction also shows that the same macroscopic phenomenon can have more than one mechanism.

---

Central limit theorem →