1. The idea in one sentence
Entropy measures how many ways a system can be arranged, or how uncertain its description is. Order, instead, tries to understand how those arrangements are organized in space.
Key phrase: two systems can have similar entropy, yet very different shapes. One may be random, while another may contain clusters, symmetries, perturbed lattices or regular minimum distances.
It is a bit like saying that knowing how many words there are in a book is not enough to know whether it is a novel, a dictionary, a poem or a shopping list. The number matters, but the structure matters just as much.
2. The minimum vocabulary: state, probability, entropy
Imagine a system made of many particles, for example dots on a plane. A configuration is one possible arrangement of those dots. In physics one often speaks of microstates: each microstate is one precise way in which the system can be made.
2.1 Boltzmann entropy: counting possible ways
The classical formula is:
Where:
- S is entropy;
- W is the number of possible microstates;
- ln is the natural logarithm;
- kB is a physical constant.
At high-school level, we can read it like this: if W increases, entropy increases. The logarithm makes the number grow more gently.
2.2 Shannon entropy: measuring uncertainty
When the states do not all have the same probability, a similar formula is used:
Here pi is the probability of case number i. The symbol Σ means “sum”.
Simple interpretation: if everything is very predictable, entropy is low. If many possibilities are open and none clearly dominates, entropy is high.
3. The parable of the library
Imagine two libraries with the same number of books.
In the first one, the books are placed on the shelves at random: novels, manuals, poems and atlases are all mixed together.
In the second one, the books are divided by subject, although each section still contains many different titles.
If we look only at the number of books and the number of possible swaps, the two libraries may appear similar. But for someone walking in and looking for information, they are completely different: the second library has an internal structure.
This is the point: entropy is very powerful, but it does not always see hidden architecture. To see architecture, we need additional tools.
4. Why order is not simply “low entropy”
In everyday language we often say: “more entropy = more disorder”. This is a useful first intuition, but it becomes too poor as soon as the system has geometry.
Common mistake
Thinking that there is one single number capable of saying how ordered a system is.
Better idea
Order is a family of properties: clusters, symmetries, minimum distances, periodicity, orientations and domains.
It is therefore better to represent order not as a single value, but as a profile:
It is like describing a person not only by “height = 1.80 m”, but also by weight, age, language, profession, character, health and experience. One number is not enough.
5. From statistics to geometry
Suppose we have many points on a plane. To begin, the analytic geometry taught in high school is enough.
If two points have coordinates:
the distance between them is:
This formula alone opens the door to many questions:
- are the points close together or spread out?
- do they form groups?
- are they almost equally spaced?
- do they have preferred angles, for example 90° or 60°?
6. Turning particles into a network
To study the shape of a set of points, we can build a graph. A graph is a network: the points become nodes and the links between points become edges.
The link can be strong if two points are close, weak if they are far apart.
This formula looks more difficult than it really is. It simply says:
- if the distance d(i,j) is small, the weight wij is large;
- if the distance is large, the weight becomes small;
- σ chooses the scale of the lens through which we observe the system.
Imagine a city seen from above. Houses are the points. Roads are the links.
If two houses are close, it is easy to move from one to the other: strong link. If they are far away or separated by a river, the link is weak.
The structure of the city is not only how many houses exist, but how they are connected: districts, bridges, squares, suburbs and main roads.
6.1 The Laplacian, without fear
In the technical version, the graph Laplacian appears. For a first reading, there is no need to calculate it by hand. It is enough to know that it is a matrix describing the relationship between:
- how much a node is connected to the others;
- how much it differs from its neighbours;
- how easily information, heat or diffusion would propagate through the network.
The compact notation is:
Where W contains the link weights and D contains, for each node, the sum of the links starting from that node.
7. The spectrum: listening to shape
When the spectrum of the Laplacian is computed, one obtains a sequence of numbers:
These numbers are called eigenvalues. They are usually a university topic, but the intuitive idea is accessible: they describe the natural ways in which the network can “vibrate”, “relax” or “separate”.
Think of a guitar. You do not need to know all the equations of strings to understand that a string has natural modes of vibration: fundamental tone, harmonics and resonances.
In the same way, a network of particles has natural modes. The low modes, associated with small eigenvalues, describe slow and collective motions: the skeleton of the shape.
7.1 Low modes and global order
If a configuration is truly organized, a large part of the important information is often concentrated in a few low modes. Very schematically:
If Ck is already large with only a few modes, the shape has a simple collective structure. If many modes are needed, the structure is more fragmented or noisier.
7.2 The spectral gap
Another useful signal is the jump between two consecutive eigenvalues:
A large jump may indicate that the first modes describe something important and separated from the rest: for example clusters, domains or a global structure.
8. A taxonomy of order
The operational heart of the article is this: there is not just one type of order. It is better to use multiple channels, each sensitive to a different property.
| Channel | What it looks for | Intuitive example |
|---|---|---|
| Om modular / spectral order |
Clusters, domains, global organization, dominant low modes. | A city divided into recognizable districts. |
| Oo4 4-fold orientational order |
Alignments with square symmetry, angles close to 90°. | A grid of orthogonal streets. |
| Oo6 6-fold orientational order |
Alignments with hexagonal symmetry, angles typical of triangular or hexagonal lattices. | Honeycomb cells or regular packing. |
| Oe exclusion order |
Regular minimum distance: points do not come too close to one another. | People in a room keeping a minimum distance. |
8.1 Orientational order through angles
To understand 4-fold or 6-fold order, just think of the angles between a point and its neighbours.
- In a square lattice, preferred directions appear: right, left, up and down. A 4-fold channel is natural.
- In a hexagonal lattice, six privileged directions appear. A 6-fold channel is natural.
The technical formula uses complex numbers:
But for an introductory reading we can translate it like this:
Simple translation: I look at the angles of the neighbours. If the angles respect a regular symmetry, the value grows. If the angles are random, the value remains low.
8.2 Exclusion order through average and standard deviation
For each point, we measure the distance to its nearest neighbour. We obtain many numbers:
We compute the average:
and the dispersion, that is, how much those values fluctuate around the average. A simple index is the coefficient of variation:
If CV is low, the minimum distances are fairly regular: there is a form of order even if there is no perfect lattice.
9. Six families of configurations
The original article compares several families of systems. The interesting point is that each family “switches on” different channels.
| Family | What it looks like | Channel that tends to respond | Physical reading |
|---|---|---|---|
| Uniform random | Points distributed almost at random. | Low values almost everywhere. | Strong correlations are absent. |
| Gaussian blobs | Soft condensations, similar to clouds. | High Om. | Clustering and modularity. |
| Hex clustered | Clusters with a hexagonal tendency. | Oo6 and Om. | Local hexagonal order plus collective organization. |
| Square clustered | Domains or grids with orthogonal directions. | Oo4 and Om. | Square symmetry and domains. |
| Poisson disk | Points not too close to each other, but without a lattice. | High Oe. | Short-range exclusion. |
| Density gradient | One denser region and one less dense region. | Mixed or weak signals. | Density heterogeneity, not necessarily true geometric order. |
Imagine a square full of people.
If everyone moves at random, there is not much order. If small groups form, modular order appears. If everyone arranges themselves in rows and columns, square order appears. If they maintain a minimum distance without forming rows, exclusion order appears.
In all cases there are people in the square, but the social geometry is different.
10. Scale: looking from near or far
A delicate point is the scale of observation. A system may appear disordered from nearby but ordered from far away, or the opposite.
For this reason it is useful to measure the indices at several scales:
The expression “weighted average” means that some scales can count more than others:
with weights w chosen so that their sum is 1.
Example: looking at a city from a window, from a drone or from a satellite does not produce the same information. Districts, roads and single houses emerge at different scales.
11. Connections with physics
This taxonomy does not come from nowhere. It connects with classical ideas in statistical physics.
Radial correlation
Measures how likely it is to find a particle at distance r from another. It is useful for seeing exclusion, shells and typical distances.
Structure factor
In the Fourier domain, sharp peaks indicate periodicity or quasi-periodicity.
Order parameters
In phase transitions, a quantity can grow or decrease when temperature, density, noise or interaction changes.
The proposal is therefore compatible with a very physical idea: it is not enough to say “more or less disordered”. We must ask which type of structure is emerging.
12. Why this distinction is useful
This way of reading structures can become useful whenever complex patterns are studied:
- images textures, patterns, microstructures, defects;
- materials porosity, grains, phases, aggregates;
- biology cell colonies, spatial distributions, networks;
- sensors measurement networks, event distributions;
- simulations particle systems, agents, automata, fields.
In all these cases the question is not only: “how random is it?”. The richer question is: “which form of organization is appearing?”.
13. Operational conclusion
Entropy tells us how many ways, or how much uncertainty, are present in a system.
The taxonomy of order instead tells us what kind of structure is hidden inside those ways: clusters, symmetries, exclusion, periodicity, domains and multi-scale organization.
The most honest synthesis is therefore not a single number, but a profile:
This is the strongest conceptual point: order is not one single quantity, but a family of phenomena.
14. Mini-exercises to fix the idea
- Library: two libraries have the same number of books. One is random, the other is organized by subject. Is a purely “numerical” entropy enough to distinguish them?
- Distances: if in a configuration all points have a nearest neighbour at a similar distance, which order channel probably grows?
- Lattice: if the points almost form a square grid, do you expect Oo4 or Oo6 to be higher?
- Clusters: if the points form three well-separated groups, which channel should switch on?
- Scale: can an image look noisy close up but have a global structure from far away? Which idea in the taxonomy is useful here?